Conference Agenda

AMSTer (SAR & InSAR Automated Mass processing Software for Multidimensional Time series) is an open-source toolbox that provides a complete, automated workflow from SAR data download to multidimensional deformation time series and web-based dissemination of results.

The AMSTer is a continuous development of three scientific components:

• The AMSTerEngine: A command line InSAR processor allowing to perform all required

SAR interferometric processing. The AMSTerEngine is a set of command line routines written in C.

• The AMSTer toolbox scripts: A set of bash and python scripts meant to automate the AMSTerEn-

gine and manage time series.

• The MSBAS (Multidimensional Small Baseline Subset) software

AMSTer aims at processing a large number of interferometric pairs to feed and run the

MSBAS processor to obtain the desired 2D or 3D deformation maps and time series products.

We present here new developments integrating Extended Time Annotation Dataset (ETAD) management and fine coherence tracking approach within the AMSTerEngine, and so, within the AMSTer toolbox.

ETAD aims at correcting Sentinel1 SLC data range and azimuth times for inaccuracies in Synthetic Aperture Radar (SAR) focusing and for geophysical effects using external models such as the operational Integrated Forecasting System (IFS) tropospheric models of ECMWF, Digital Elevation Models (DEM) or land tides models.

These corrections being different from an image to another, time inaccuracies differences are still present in interferometric processing (InSAR) leading to sub-pixels local misregistration with respect to an ideal co-registration corresponding to a perfectly stationary situation. Therefore ETAD management allows improving the georeferencing and geoprojection of InSAR products and mitigating non-turbulent atmospheric phase component.

On it side, fine coherence tracking performs a series of co-registrations, shifted a fraction of a pixel away from the stationary solution. This sequence of co-registrations allows computing a stack of interferograms and coherence images across which the range and azimuth shifts leading to the optimum coherence can be found on a pixel-by-pixel basis. This approach leads to a pixel-wise 2D mapping of coherence with respect to imposed shifts. It is shown that subtle local misregistration of pixels can be measured with precision through 2D curve fitting. This coherence tracking method being purely data-driven, it is suited to perform a cross validation with differential ETAD shifts that are mainly model-based.

In addition to providing local range and azimuth shifts, side products of the methodology are the optimised coherence itself and the tracked optimised interferogram that can in turn be used in time series processing.

(AMSTer Toolbox). AMSTer software is freely available under the terms of the GNU Affero General Public (AGPL) License.

Interferometric Synthetic Aperture Radar (InSAR) is a powerful tool for mapping surface movements, but tropospheric delays complicate deformation interpretation. Tropospheric errors are influenced by various spatiotemporal factors, including water vapor, temperature and pressure and all these factors are related to satellite orbit configurations. This means that although tropospheric errors are independent of signal wavelengths, different satellites may encounter completely different tropospheric effects. However, while previous studies focus on physical properties of the troposphere, orbit-specific tropospheric features remain underexplored. In this paper, we investigate the spatiotemporal characteristics of tropospheric effects using nine years of image pairs globally derived from Sentinel-1A/B’s orbit constellation configuration (acquisition intervals, dates and time of day) and the Generic Atmospheric Correction Online Service for InSAR (GACOS). Our findings quantify pronounced spatial heterogeneity and temporal variability in tropospheric errors, with globally variable linearity, seasonality and randomness in image pair time series. Linear constrained time series inversions (e.g., image pair stacking) demonstrate the effectiveness of long-temporal-baseline image pairs in enhancing accuracy, but such improvement is not continuously growing, highlighting the need to balance the number of image pairs with achievable accuracy. Obtaining seasonal deformation faces greater challenges due to dominant tropospheric seasonality, especially in cases with delayed seasonal responses driven by processes like groundwater extraction or water erosion. These findings offer a framework for understanding tropospheric effects and practical recommendations for improving deformation inversion accuracy, providing valuable insights that can serve as indicators for orbit parameter design and optimization of future SAR missions.

Interferometric Synthetic Aperture Radar (InSAR) measures phase differences between multiple SAR images and allows observation of a variety of signals, including surface deformation from seismicity, volcanic unrest , and anthropogenic activities (e.g., wastewater injection or groundwater withdrawal). The precision of InSAR-derived vertical displacement rates can approach a millimeter per year over spatial scales of a few kilometers or less, particularly when a large number of observations are available spanning multiple years. However, the quality of individual interferograms may be affected by a variety of noise sources, including noise introduced by water vapor. Characterization of noise sources is an essential part of the development of high-quality constraints on subjects of interest (e.g., ground deformation or glacial dynamics).

As microwaves pass through the troposphere, they are refracted, introducing errors of up to tens of centimeters or more. Many InSAR studies attempt to reduce the impact of atmospheric noise by using many images, under the assumption that these effects will average to zero. Since these “stacking” approaches rely on multiple acquisitions, small transient deformation signals spanning only a few acquisitions may still be completely overwhelmed by tropospheric noise. Therefore, other methods dedicated to the correction of tropospheric delays have been devised. This includes empirical models, where the relationship between water vapor delays and elevation is utilized to estimate tropospheric effects. Other approaches for mitigating the effects of the troposphere on InSAR observations use humidity, pressure, and temperature derived from weather model products (e.g., ERA-5) to predict and remove models of tropospheric delay. These weather model-based corrections tend to have spatial resolutions >10 km and temporal resolutions on the order of hours.

In most cases, the spatiotemporal resolutions of these models are far too coarse to resolve the fine structures of fast-moving storm systems, where cumulative phase delays vary horizontally within 100s of meters. This has major implications for monitoring ground deformation in study regions such as ours in Oklahoma, where the presence of storm systems in SAR images is commonplace. Using a novel high-pass phase-based approach for filtering and pixel selection, we demonstrate the limitations of current InSAR tropospheric models when attempting to correct for storm signatures. Furthermore, we propose an empirical correction technique based on independent ground-based weather radar. We validate our model through numerical weather simulations in the Weather Research and Forecasting Model. Finally, we quantify the errors that storms may have on geophysical parameter estimation.

Using the InSAR-related NSBAS chain processing (Doin et. al 2011, Thollard et. al. 2021) and the 2016-2025 Sentinel-1 radar data, we aim at obtaining a velocity map over the European Alps with a target uncertainty of a mm/yr at the scale of the massif. Mountainous areas present inherent challenges for InSAR study such as geometric distortions, decorrelation due to vegetation or snow cover changes, and tropospheric delays. To achieve millimeter-per-year precision, particular care must be taken when mitigating the atmosphere contribution in interferometric processing. Tropospheric contributions have spatial and temporal variations that do not cancel when forming interferograms (Massonnet and Feigl, 1994) due to pressure, temperature and relative humidity changes. Tropospheric delays are still a major factor limiting the accuracy of InSAR measurements in slowly deforming areas, as a change of 20% of relative humidity between two acquisitions is responsible for a 10 cm error in deformation interpretation (Zebker and Rosen, 1997). Additionally, tropospheric delays are not easily separable from the rest of the phase contributions and can be misinterpreted. Therefore, its impact has been widely studied and several methods have been proposed to mitigate tropospheric noise (Bekaert et. al. 2015). Among them, we focus on two main approaches: mitigation by trying to separate stochastic noise from ground motion signal (for example, exploiting empirical relationship between elevation and troposphere) and by the usage of external data. For the latter, two main sources of data are available. We can benefit from data from global atmosphere models (GAM) (Doin et al., 2009; Jolivet et. al. 2011) (particularly the ERA5 reanalysis of the ECMWF) or from zenithal troposphere delays (ZTD) obtained during the GNSS processing (e.g. Onn and Zebker, 2006; Albino et. al., 2025).

In mountainous areas such as the Alps, the ability of the these three methods to correct tropospheric delays is crucial but limited by the complexity of orogenic features and associated atmospheric flows (e.g. Baines, 1998 or Sandu et. al., 2019). Empirical models, when their complexity is increased to adjust to the lateral variations in stratified delays in separate valleys, may remove elevation-dependent deformation. GNSS sparse network forces interpolation. GAM is limited by its spatial resolution (30x30 km for ERA5), especially considering that delay correction in deep valleys are obtained from an extrapolation of temperature, humidity and pressure below the surface of the model. For example, a previous InSAR processing in the Alps (Mathey et. al. 2020) has shown that atmospheric-related contribution remained in deep Alpine valleys because of the coarse resolution of ERA5.

We explore here the usage of GNSS Alps data processed using the GipsyX software at EPOS-UGA to mitigate tropospheric delays in interferograms, but also of the new CERRA reanalysis which has a finer spatial resolution (5x5 km). As we aim to reach the mm/yr accuracy in the measurement of the Alps uplift, we must carefully examine potential trends in these data sets over the period 2015-2025. We thus first compare dispersion, bias and bias trend in ZTD time series of the EPOS-UGA to other solutions namely the SPOTGINS GNSS solution from the CNES and the CERRA on common locations.

Then, to compute atmospheric phase screens from GNSS ZTD, we take into account the vertical dependent term as well as the lateral variations due to atmospheric turbulence using Iterative Tropospheric Decomposition (ITD) (Yu et. al. 2018). As the Alps GNSS stations are sparse, we also test a combination of a GAM with GNSS results. We present the method, the results obtained and a comparison against the widely used ERA5 model.

Finally, we assess methods’ performance using the correlation of phase delay residuals with elevation because atmosphere stratified component is the predominant source of troposphere errors. Moreover, since atmospheric noise is relatively high compared with deformation signal, we study the standard deviation reduction in unwrapped interferograms when we apply or not our corrections.

We focus our conclusion by the analysis of the deep Alpine valleys and foothills where tropospheric delays are the harder to mitigate.

Albino, F. et al. (2025) ‘Benefits of GNSS Local Observations Compared to Global Weather-Based Models for InSAR Tropospheric Corrections Over Tropical Volcanoes: Case Studies of Piton De La Fournaise and Merapi’, Journal of Geophysical Research: Solid Earth, 130(4), p. e2024JB028898.

Baines, P. G. (1998) ‘Topographic Effects in Stratified Flows’, Cambridge University Press.

Bekaert, D.P.S. et al. (2015) ‘Statistical comparison of InSAR tropospheric correction techniques’, Remote Sensing of Environment, 170, pp. 40–47.

Doin, M.-P. et al. (2009) ‘Corrections of stratified tropospheric delays in SAR interferometry: Validation with global atmospheric models’, Journal of Applied Geophysics, 69(1), pp. 35–50.

Doin, M.-P. et al. (2011) ‘Presentation of the small baseline NSBAS processing chain on a case example: the Etna deformation monitoring from 2003 to 2010 using ENVISAT data’. In Proc. of the Fringe 2011 Workshop “Advances in the Science and Applications of SAR Interferometry” (Vol. ESA SP-697). Frascati, Italy : ESA.

Jolivet, R. et al. (2011) ‘Systematic InSAR tropospheric phase delay corrections from global meteorological reanalysis data’, Geophysical Research Letters, 38(17).

Massonnet, D. and Feigl, K.L. (1998) ‘Radar interferometry and its application to changes in the Earth’s surface’, Reviews of Geophysics, 36(4), pp. 441–500.

Mathey, M. et al. (2022) ‘Spatial Heterogeneity of Uplift Pattern in the Western European Alps Revealed by InSAR Time‐Series Analysis’, Geophysical Research Letters, 49(1), p. e2021GL095744.

Onn, F. and Zebker, H.A. (2006) ‘Correction for interferometric synthetic aperture radar atmospheric phase artifacts using time series of zenith wet delay observations from a GPS network’, Journal of Geophysical Research: Solid Earth, 111(B9).

Sandu, I. et al. (2019) ‘Impacts of orography on large-scale atmospheric circulation’, npj Climate and Atmospheric Science, 2(1), p. 10.

Thollard, F., et al. (2021). Flatsim: The form@ ter large‐scale multi‐temporal sentinel‐1 interferometry service. Remote Sensing, 13(18), 3734.

Yu, C., Li, Z. and Penna, N.T. (2018) ‘Interferometric synthetic aperture radar atmospheric correction using a GPS-based iterative tropospheric decomposition model’, Remote Sensing of Environment 204, pp. 109–121.

Zebker, H.A., Rosen, P.A. and Hensley, S. (1997) ‘Atmospheric effects in interferometric synthetic aperture radar surface deformation and topographic maps’, Journal of Geophysical Research: Solid Earth, 102(B4), pp. 7547–7563.

Interferometric synthetic aperture radar (InSAR) phase estimations are typically interpreted using physically motivated forward models. However, the estimated phase is a superposition of multiple contributions, the separation of which is often model dependent. In this study, we present a unified approach for identifying and extracting theoretically expected phase components using independent component analysis (ICA), without imposing theoretical models during the separation stage. The key idea is to first perform model-free source estimation and separation, and then validate the extracted sources using theory-consistent tests based on correlation structures and expected coefficient dependence. We present two complementary applications: (1) estimation of ionospheric delay components from a single interferogram using split-spectrum processing and (2) estimation of residual DEM error-related components from time-series InSAR results.

In the case of ionospheric delay, we start with a single interferogram and use split-spectrum processing to generate multiple band-limited interferograms. ICA is then performed on these derived images to separate independent sources embedded in the interferometric phase, aiming to recover a component attributable to ionospheric delay. Because the estimation is driven by statistical independence rather than a prescribed ionospheric model, this approach provides a data-driven test of whether an ionospheric-like component is detectable. Candidate components are evaluated using correlation analyses between spatial patterns and frequency-dependent coefficients, and by comparing ICA-estimated coefficients with the dependence expected from standard split-spectrum theory.

In the case of DEM error, ICA is applied to the cumulative phase time series to separate statistically independent sources. Candidate DEM error-related components are evaluated using quantitative consistency tests that focus on whether their temporal coefficients exhibit acquisition-geometry dependence expected from DEM error propagation. We compute correlation coefficients in both spatial and temporal domains, and compare ICA-derived mixing coefficients and spatial patterns with values predicted by the conventional geometry-based model equation. This provides an objective assessment of whether a DEM error contribution is present in the time series and whether ICA can recover it in a physically interpretable form, despite not assuming a DEM error model in the decomposition process.

Across both applications, the main contribution is a consistent validation framework that links model-free ICA decomposition to theory-consistent interpretation. By combining ICA with correlation-based comparisons in space–frequency (for ionospheric delay) and space–time (for DEM error), we demonstrate how components described by standard InSAR theory can be identified and extracted without incorporating model assumptions into the separation process itself. This approach provides an assumption-light pathway for identifying nuisance contributions such as ionospheric delay and residual DEM error, with direct implications for improving the interpretability and reliability of deformation estimates from InSAR data.

Tropospheric delay remains one of the dominant sources of error affecting the accuracy of Interferometric Synthetic Aperture Radar (InSAR) time-series analysis. Atmospheric stratification produces phase signals that are strongly correlated with topography, particularly in mountainous regions where large elevation gradients interact with complex atmospheric circulation. These effects can degrade the quality of deformation measurements derived from InSAR and complicate the interpretation of centimeter-scale surface motion. Improving the mitigation of atmospheric artifacts is therefore essential for reliable deformation monitoring in high-relief environments.

The Eastern Cordillera of the Southern Central Andes in northwestern Argentina provides an ideal natural laboratory to investigate atmospheric correction strategies under complex environmental conditions. The study region is characterized by steep climatic and topographic gradients, where low-elevation eastern foreland areas with dense vegetation transition to semi-arid and arid high-elevation terrain with sparse vegetation cover. Elevation differences exceeding several kilometers occur over relatively short horizontal distances, and the region is influenced by strong seasonal climatic variability associated with the South American Summer Monsoon. Moisture transport across the orogen generates pronounced spatial variability in atmospheric water vapor, producing strong elevation-correlated tropospheric delay signals that significantly influence interferometric phase observations and complicate deformation measurements.

In this study we analyze multi-year Sentinel-1 C-band SAR observations acquired between 2014 and 2019 over the Eastern Cordillera of northwestern Argentina. The interferometric time series is generated using a small-baseline interferometric network, enabling consistent estimation of surface deformation over multi-year observation periods. Our InSAR time-series analysis reveals numerous slow-moving landslides across the region, which typically deform at rates of approximately 5–10 cm yr⁻¹. We present a comprehensive comparison of stratified tropospheric delay correction approaches based on atmospheric reanalysis products and spatial statistical methods.

Tropospheric delays are estimated through ray-tracing calculations using atmospheric profiles derived from global weather reanalysis datasets including ERA5 (European Centre for Medium-Range Weather Forecasts Reanalysis version 5), ERA5T (the near-real-time extension of ERA5), and MERRA-2 (Modern-Era Retrospective Analysis for Research and Applications, version 2). These datasets provide vertically resolved atmospheric parameters such as temperature, pressure, and water vapor, allowing the propagation of radar signals through the troposphere to be modeled along the radar line of sight.

Because ray-tracing corrections depend on assumptions about the vertical extent of the atmospheric column used in the delay calculation, we investigate the sensitivity of tropospheric delay estimates to variations in the maximum atmospheric integration height used during the ray-tracing process. This parameter determines the vertical portion of the atmosphere included in the modeled delay and may influence how effectively elevation-dependent atmospheric signals are captured in regions characterized by large topographic relief.

To further address atmospheric variability that may not be fully captured by large-scale atmospheric models, we introduce a robust sliding-window atmospheric correction approach that estimates localized elevation–phase relationships within moving spatial windows. This spatially localized framework allows stratified atmospheric delay to be modeled at regional scales while reducing the influence of deformation signals and outliers. By capturing spatially varying atmospheric gradients across steep terrain, the method provides an adaptive strategy for mitigating residual elevation-correlated atmospheric artifacts.

The performance of the proposed approach is evaluated using both real InSAR observations and synthetic simulations in which stratified atmospheric delays and deformation signals representative of slow-moving landslides are generated. These simulations allow the capability of the sliding-window method to separate atmospheric artifacts from deformation signals to be assessed under controlled conditions.

Our results indicate that the selected maximum atmospheric integration height influences the magnitude and spatial distribution of modeled tropospheric delay in steep mountainous terrain. When the integration height is insufficient, stratified atmospheric delay may be underestimated, leaving residual elevation-dependent phase signals in corrected interferograms. Increasing the integration height improves atmospheric delay estimation up to a threshold, beyond which excessively large integration heights may introduce artifacts in the modeled delay field.

Both synthetic experiments and real InSAR time-series analysis demonstrate that the proposed robust sliding-window atmospheric correction effectively reduces elevation-correlated phase artifacts and improves the stability of deformation estimates in areas affected by complex atmospheric variability. These findings highlight the importance of appropriate parameterization of atmospheric corrections and demonstrate the potential of localized atmospheric modeling approaches for improving InSAR time-series analysis in mountainous environments.

Radar interferometry is a technique capable of measuring precise surface displacement by analyzing the interferometric phase between two images acquired at different times. Spaceborne SAR is effective for large-scale studies but remains limited by long revisit cycles. Furthermore, geometric distortions such as layover, foreshortening, and shadowing frequently result in significant blind spots in the field of view. To overcome these limitations, ground-based radar (GBR) systems have been increasingly deployed, offering high temporal resolution and flexible imaging geometries tailored to specific monitoring objectives. In contrast, GBR offers flexible control over acquisition time, location, and antenna geometry to specific monitoring objectives. GBR is particularly effective for observing rapid and localized movements in slopes, landslides, and critical infrastructure. Despite these advantages, ground-based radar is inherently sensitive to atmospheric phase delays caused by temporal and spatial variations in temperature, humidity, and barometric pressure. Temporal and spatial variations in temperature, humidity, and pressure alter the atmospheric refractivity. This fluctuation modifies atmospheric refractivity, altering radar wave propagation paths and inducing phase components unrelated to actual surface displacement, with the effect becoming more pronounced at higher frequencies such as the Ku-band.

This study aims to quantitatively evaluate the impact of atmospheric interference on GBR persistent scatterer interferometry (PSI) and proposes an atmospheric correction methodology using synchronous meteorological data. The experiment was conducted using the Gamma Portable Radar Interferometer-II (GPRI-II) system, which operates in the Ku-band (17.1–17.3 GHz). A 33-hour continuous dataset comprising 397 Single-Look Complex (SLC) images was collected at a levee site in South Korea at a 5-minute acquisition interval. Synchronous temperature and humidity data were recorded on-site using hygrometers near the corner reflectors (CRs). Barometric pressure data were obtained from a meteorological station located 6 km away to refine calibration model accuracy. The dataset was partitioned into eleven 3-hour segments for detailed correlation analysis. This analysis revealed a strong relationship between the calculated atmospheric refractivity and the interferometric phase, yielding a maximum coefficient of determination (R²) of 0.9 between 20:00 and 23:00.

An initial atmospheric phase model for calibration was developed using the calculated refractivity. This model was then refined by analyzing the interferometric phase observed at the corner reflectors. Finally, 396 atmospheric-phase images generated from the refined model were applied to the SLC stack to compensate. The efficacy of the proposed correction was validated using two installed corner reflectors, assuming zero actual displacement. Experimental results demonstrated a substantial improvement in phase stability. The standard deviation of the interferometric phase decreased from 0.26 to 0.12 radians for the first CR and from 0.36 to 0.14 radians for the second CR. Furthermore, the estimated line-of-sight displacement rate error was significantly mitigated, dropping from 0.5–0.7 cm/hour to 0.0–0.1 cm/hour after correction. This study concludes that integrating meteorological-based atmospheric modeling effectively enhances the precision of GBR PSI, providing a reliable framework for monitoring rapid geotechnical and structural deformations.

The vast amount of Sentinel-1 data over the last decade and the high accuracy of related data such as precise orbit ephemerides (3-D RMSE below 10 mm) allows for the exploration of measurements of the solid Earth surface dynamics in a global reference frame. This is applicable in regions lacking geodetic instrumentation, and particularly useful at tectonic transform boundaries where plate motion models may not be accurate. On the other hand, model residuals allow us to evaluate non-tectonic signals, including due to the atmosphere or orbit inaccuracies.

We updated our previously published methodology (2023) investigating precise along-track (azimuth) coregistration offsets from spectral diversity of burst overlaps, averaged over Sentinel-1 frames consisting of ~13 bursts per swath w.r.t. satellite position. This was leading to decomposed absolute velocities at coarse resolution (~250x250 km^2 pixel size) estimated from 6-year time series over the Alpine-Himalayan Belt. Median 2-sigma errors were ~4 mm/year northwards and ~20 mm/year eastwards. We have further extended our dataset, partly reprocessed using updated orbits in a modified processing chain, e.g. avoiding intensity cross-correlation in azimuth to prevent ionosphere-related errors. We improved ionospheric correction by scaling JPL high resolution global ionosphere maps (GIM) vertical total electron content (VTEC) using geometry estimated from IRI2020 ionospheric profiles. Finally, we added time series of range pixel offsets and corrected them for solid Earth tides, ionosphere and tropospheric delay.

We observe and further investigate and report on unmapped consistent signal over humid/highly vegetated regions, particularly in range, and constant offsets between Sentinel-1A and Sentinel-1B in both azimuth and range directions. By decomposing overlapping ascending and descending data from both azimuth and range in the same framework covering 2016-2022 (avoiding peaking solar cycles) to absolute velocities, we estimate median 2-sigma accuracy of [3.5, 6.7, 5.2] mm/year for [eastwards, northwards, vertical] directions, respectively. Furthermore, we fully remove eastward bias w.r.t. GNSS data from the median 9 mm/year reported in the original work.

Velocities estimated from the overall range and azimuth coregistration offsets can be directly combined with the standard mean-centered line-of-sight and burst-overlap InSAR velocities, respectively, in order to leverage them into the global reference frame measurements. Due to the existing noise that is higher in humid regions and during increased solar activity, we evaluate the best-case scenario over the arid/semi-arid region of the Dead Sea transform zone, comparing to velocities from GNSS and reporting on achievable accuracy.

Deep learning has established a new paradigm in Synthetic Aperture Radar (SAR) image despeckling, shifting the focus from traditional spatial filters to sophisticated neural architectures. Among these, the MERLIN (coMplex sElf-supervised despeckLINg) framework has emerged as a state-of-the-art solution by exploiting the statistical independence of the real and imaginary components of Single-Look Complex (SLC) images for self-supervised training. However, ensuring the necessary orthogonality of these components is technically demanding in practice. It requires mandatory, specialized preprocessing—ranging from spectral recentering in Stripmap data to complex deramping for Sentinel-1 TOPSAR—to correct for asymmetrical spectra and Doppler offsets that would otherwise introduce significant artifacts. These preprocessing requirements and the strict necessity for full complex SLC data create a significant operational bottleneck for large-scale industrial Interferometric SAR (InSAR) and monitoring production chains. Conversely, an amplitude-only input would bypass these SLC-dependent constraints, facilitating the seamless integration of advanced deep-learning despeckling into legacy processing pipelines and low-latency services.

Previous attempts to adapt the MERLIN framework for amplitude-only inference, as noted in the original literature, have proven suboptimal due to the spatial characteristics of speckle. Generating a synthetic phase field for intensity-only images introduces high-frequency texture artifacts and residual noise stemming from the correlation mismatch between the actual intensity and the random phase. While these effects can be partially mitigated through spatial subsampling to whiten the speckle, such preprocessing unavoidably degrades high-frequency structural details, such as thin lines. Consequently, the operative understanding has been that full, correctly preprocessed SLC data remains a non-negotiable prerequisite for high-fidelity self-supervised despeckling.

In this paper, we present an operational breakthrough that bridges the architectural gap between complex-domain self-supervision and amplitude-domain inference. We propose a Teacher-Student Knowledge Distillation (TSKD) framework that decouples the learning of speckle statistics from the constraints of the inference data format. Our approach distills the "collective intelligence" of multiple sensor-specific "Teacher" models—each optimized for X-, C-, and L-band SLC data—into a single, robust "Student" network. By training the Student to regress the Teachers' denoised outputs using only amplitude inputs, the model effectively internalizes diverse spectral shapes and spatial correlation patterns. This allows the Student to produce high-fidelity, speckle-free imagery without requiring the complex-domain preprocessing (deramping or demodulation) typically needed to model these statistics.

Experimental results across X-, C-, and L-band sensors demonstrate that the Student model replicates the Teachers’ performance with near-lossless fidelity, achieving an average PSNR of 37.81 dB and an SSIM of 0.978 on the test dataset. The primary advantage of this framework is the total elimination of complex-domain preprocessing during the inference phase, enabling truly sensor-agnostic deployment. While the original MERLIN framework requires networks tailored to specific sensors and acquisition modes, our approach yields a single, universal model. By operating strictly on amplitude data, it bypasses metadata-dependent steps, providing high-quality products across varying systems. Furthermore, by processing a single amplitude channel instead of dual complex components, the model achieves a significant increase in inference speed. This efficiency is crucial for wide-area-processing services, transforming deep-learning despecklers into scalable, on-demand solutions for global SAR monitoring.

Interferometric synthetic aperture radar (InSAR) is a common remote sensing method used to measure ground deformation for a variety of applications. For a given stack of SAR images, InSAR produces 1D displacement estimates along the satellite line-of-sight (LOS). While 3D decomposition is theoretically possible using three or more LOS observations with different viewing geometries, the near-polar orbits of current-generation SAR satellites mean that angular diversity is limited, and the 3D inverse problem is ill-posed. Capella Space’s constellation of high-resolution X-band SAR satellites, occupying both sun-synchronous and inclined orbits, have the potential to enable native 3D InSAR measurements. Capella Space is working towards an operational interferometric capability, and the data used here were collected during experimental testing in June-August 2024.
In this study, we present high-resolution 3D InSAR displacement time series estimates using a combination of Capella-14 imagery (45° orbital inclination, 2.95 day revisit) and TerraSAR-X/PAZ data (97° orbital inclination, alternating 4 and 7 day revisit). LOS time series were analysed using a small baseline subset approach, and the 3D solution was calculated using a generalized least squares formulation. The test site used is an active open-pit mine, with fast displacement and varied motion geometries.
We compare the 3D results to a conventional 2D (East-West/Up-Down decomposition) scenario using only TSX/PAZ data, and use combinations of three to five datasets to examine the impact of different observation geometries on the uncertainty budget of the 3D solution. Results indicate that the 3D solution provides sensitivity to previously underestimated North-South (NS) displacement, and captures the expected motion direction well. Overall uncertainty increases when solving in 3D due to the additional degree of freedom and the less well constrained NS component. Including additional viewing geometries, however, leads to substantial reductions in both NS and total uncertainty. We also show that the 3D solution removes biases that arise in the EW/UD 2D approach due to unmodeled NS displacement.
These results highlight the significant benefits of SAR satellites in mid-inclination orbits for InSAR. High-resolution 3D InSAR would allow observations of slopes and displacement orientations that current satellites have limited sensitivity to, enabling end users to extract more complete information and develop more robust interpretations.

As a well-established method for high-precision ground motion monitoring, SAR Interferometry (InSAR) enables the detection of slow deformations – typically due to subsidence, landslides, earthquakes, and volcanic phenomena, also affecting buildings or infrastructures – with millimetric precision and sub-metric spatial detail [1]-[3]. The cornerstone of this technique lies in the identification of points that exhibit consistent interferometric phase coherence over a sequence of acquisitions: over time, several approaches have been brought to an operational level, including the Small BAseline Subset (SBAS), the Persistent Scatterer (PS), and concept of Distributed Scatterer (DS) appraoches [4]-[19]. All the mentioned techniques identify the coherent points, that typically are in correspondence of man-made structures or natural terrains scarcely vegetated and non-cultivated, through the usage of different statistics based from the image amplitudes (e.g. amplitude dispersion, signal-to-clutter ratio) and/or phases (e.g. temporal coherence).

In this context, we recently proposed an effective, efficient, and elegant algorithm for the selection of the measurement points, called Point Coherence Estimation (PCE) [20]. The main novelty and relevance of the PCE approach is that it is based only on a commonly accepted assumption about the statistical independence between the SAR signals in neighboring pixels. Based on this, we derive an algorithm capable to determine the coherence of all points in the interferometric image stack without critical approximations and at full resolution (without any pixel averaging operation).

In this work, we provide a rigorous mathematical formulation of the PCE algorithm. Furthermore, we present a comprehensive quantitative validation of our processing chain across multiple radar wavelenghts (using COSMO-SkyMed, Sentinel-1 and SAOCOM data for X, C, L bands, respectively), analyzing a wide range of deformations (including landslides, subsidences, volcanic activities, bradyseism) in diverse geomorphological scenarios (urban areas, flat or mountain regions, coastal areas, and in general all types of areas with coherent point-like or distributed scatterers).

The PCE methodology begins by considering the phase differences between neighbouring points (typically within a few hundred meters). In these differences, spatially correlated components, such as atmospheric delays, orbital artifacts, and large-scale motions, cancel out. The remaining signal allows for the estimation of the differences in elevation and velocity by maximizing the temporal coherence of these arcs. The temporal coherence of each pair-of-points depends primarily on the localized phase noise (thermal, temporal, and geometric decorrelations, for instance) of the two constituent points.

Assuming that the phase noise is statistically independent between the neighbouring points of each pair, the expected value of the temporal coherence for an arc is equal to the product of the expected coherence values of the two individual points. By taking the logarithm of these relationships, we derive an overdetermined system of linear equations linking the arc coherences with the individual point ones. This system can be solved with good computational efficiently using solvers minimizing residuals via L1 or L2 norms. This yields a reliable, consistent estimate of the temporal coherence for every single point, facilitating the final identification of measurement points.

Crucially, this method requires no prior assumptions regarding the probability distribution of the phase noise. However, under a Gaussian hypothesis, the system elegantly simplifies to the principle that the noise variance of a phase difference is the sum of the individual noise variances.

By operating as a selection filter that avoids spatial "smearing," the PCE ensures that the original phase information remains unaltered, providing a clean, high-integrity input for the subsequent InSAR processing steps, including phase unwrapping.

To assess the performance of the PCE-based workflow, results are benchmarked against current state-of-the-art PS and DS methodologies, and existent monitoring services like the European Ground Motion Service (EGMS) [3]. Metrics such as measurement density and coverage are considered. In addition, the cleanliness (noise and outliers) of the measurements signal is evaluated by statistical spatio-temporal analyses.

Results show that the PCE method makes it possible to obtain more measurement points with respect to standard PS techniques, both in terms of density and of coverage of areas. Among the other things, it is worth noting that the PCE method makes it possible to obtain a significant number of interferometric measurements not only corresponding to strong scatterers but also in areas characterized by low and distributed backscattering (such as bare soil and scarcely vegetated non-cultivated terrains), where standard PS techniques are not able to detect signals.

Regarding DS techniques, they are very effective in obtaining interferometric measurements in these types of areas, but effects due to changes in the dielectric constant (e.g., soil moisture and atmospheric humidity) can accumulate over time in the phases obtained by DS techniques, leading to signals that can be erroneously interpreted as ground deformations [21]. On the contrary, the PCE method exploits direct measurements at large spatial and temporal baselines and is not affected by these problems; moreover, the PCE technique preserves the signal at full resolution, without the need of performing phase averages and adaptive multilooking. Tests against current DS approaches show that PCE method can provide a good trade-off between point coverage and preservation of original phase values. However, it is worth noting that the PCE algorithm can be applied to any type of SAR images stack: to maximize the points coverage, a DS-filtered images stack could be set as input for the algorithm.

The quantitative assessment performed in this study highlights the PCE algorithm, part of our innovative processing chain, as a robust and efficient alternative to traditional methods for the measurement point selection in InSAR. The validation campaign shows a superior balance in terms of accuracy, density, and coverage of ground deformations measurements, compared to current state-of-the-art InSAR methods.

REFERENCES

[1] M. Crosetto, O. Monserrat, M. Cuevas-González, N. Devanthéry, B. Crippa, "Persistent Scatterer Interferometry: A review," ISPRS Journal of Photogrammetry and Remote Sensing, vol. 115, pp. 78-89, 2016.

[2] D. HO TONG MINH, R. Hanssen, F. Rocca, “Radar Interferometry: 20 Years of Development in Time Series Techniques and Future Perspectives,” Remote Sens., vol. 12, pp. 1364, 2020, https://doi.org/10.3390/rs12091364

[3] M. Costantini et al., "EGMS: Europe-Wide Ground Motion Monitoring based on Full Resolution InSAR Processing of All Sentinel-1 Acquisitions," IEEE International Geoscience and Remote Sensing Symposium - IGARSS, Kuala Lumpur, Malaysia, pp. 5093-5096, 2022, doi: 10.1109/IGARSS46834.2022.9884966.

[4] P. Berardino, G. Fornaro, R. Lanari, and E. Sansosti, “A new algorithm for surface deformation monitoring based on small baseline differential SAR interferograms,” IEEE Transactions on Geoscience and Remote Sensing, vol. 40, no. 11, 2002.

[5] A. Ferretti, C. Prati and F. Rocca, "Nonlinear subsidence rate estimation using permanent scatterers in differential SAR interferometry," IEEE Transactions on Geoscience and Remote Sensing, vol. 38, no. 5, pp. 2202-2212, Sept. 2000. 

[6] N. Adam et al. “Wide area persistent scatterer interferometry: Algorithms and examples.” Proc. of ESA Fringe, 2011.

[7] K. Goel, R. Shau, N. Adam, “Single-network wide-area persistent scatterer interferometry: Algorithms with application to Sentinel-1 inSAR data.” American Geophysical Union (AGU) Fall Meeting, 14-18 Dec. 2015, San Francisco, US

[8] R. Lanari, O. Mora, M. Manunta, J. J. Mallorquì, P. Berardino, and E. Sansosti, “A small-baseline approach for investigating deformations on full-resolution differential SAR interferograms,” IEEE Transactions on Geoscience and Remote Sensing, vol. 42, no. 7, pp. 1377–1386, 2004.

[9] A. Hooper, H. Zebker, P. Segall, and B. Kampes, “A new method for measuring deformation on volcanoes and other natural terrains using InSAR persistent scatterers,” Geophysical research letters, vol. 31, no. 23, 2004.

[10] A. Hooper, P. Segall, and H. Zebker, “Persistent scatterer interferometric synthetic aperture radar for crustal deformation analysis, with application to volcan Alcedo, Galapagos,” Journal of Geophysical Research: Solid Earth, vol. 112, no. B7, 2007.

[11] B.M. Kampes, “Radar Interferometry – Persistent Scatterer Technique.” Springer. 2006, ISBN-10 1-4020-4576-X (HB).

[12] P. Blanco-Sanchez, J. J. Mallorquì, S. Duque, and D. Monells, “The coherent pixels technique (CPT): An advanced DInSAR technique for nonlinear deformation monitoring,” Pure and Applied Geophysics, vol. 165, no. 6, pp. 1167–1193, 2008.

[13] F. Zhao and J. J. Mallorqui, "A Temporal Phase Coherence Estimation Algorithm and Its Application on DInSAR Pixel Selection," IEEE Transactions on Geoscience and Remote Sensing, vol. 57, no. 11, pp. 8350-8361, Nov. 2019.

[14] M. Costantini, S. Falco, F. Malvarosa, F. Minati, F. Trillo, and F. Vecchioli, “Persistent Scatterer Pair Interferometry: Approach and Application to COSMO-SkyMed SAR Data.” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2014, manuscript ID JSTARS-2014-00117.

[15] A. Ferretti, A. Fumagalli, F. Novali, C. Prati, F. Rocca and A. Rucci, “A New Algorithm for Processing Interferometric Data-Stacks: SqueeSAR,” IEEE Transactions on Geoscience and Remote Sensing, vol. 49, no. 9, pp. 3460-3470, Sept. 2011.

[16] E.A. Hetland, P. Musé, M. Simons, Y.N. Lin, P.S. Agram, C.J. DiCaprio, “Multiscale InSAR Time Series (MInTS) Analysis of Surface Deformation.” J. Geophys. Res. Solid Earth, vol. 117, pp. 8731, 2012.

[17] K. Goel, N. Adam, “A Distributed Scatterer Interferometry Approach for Precision Monitoring of Known Surface Deformation Phenomena.” IEEE Transactions on Geoscience and Remote Sensing, vol. 52, pp. 5454–5468, 2014

[18] G. Fornaro, S. Verde, D. Reale and A. Pauciullo, “CAESAR: An Approach Based on Covariance Matrix Decomposition to Improve Multibaseline–Multitemporal Interferometric SAR Processing,” IEEE Transactions on Geoscience and Remote Sensing, vol. 53, no. 4, pp. 2050-2065, April 2015.

[19] H. Ansari, F. De Zan, R. Bamler, “Efficient Phase Estimation for Interferogram Stacks,” IEEE Transactions on Geoscience and Remote Sensing, vol. 56, pp. 4109–4125, 2018.

[20] Vecchioli, F., Costantini, M., Minati, F., & Zavagli, M. (2023). A Novel Algorithm for Point Coherence Estimation in SAR Interferometry. IGARSS 2023 - 2023 IEEE International Geoscience and Remote Sensing Symposium, 7868-7871.

[21] Samiei Esfahany, S., Lopez Dekker, P., & Hanssen, R. (2017). On the Effect of Soil Moisture Phase Inconsistencies on Phase Estimators from Distributed Scatterers in InSAR Stacks. 44-45. Abstract from Fringe 10th International Workshop on , Helsinki, Finland.

The Sentinel User Preparation- Synthetic Aperture Radar (SUPSAR) programme of the European Space Agency (ESA) aims to prepare the scientific community for the synergistic exploitation of the next-generation SAR missions, such as Sentinel-1 Next Generation (NG) and ROSE-L, as part of a broader System-of-Systems. These future missions will offer complementary C- and L-band capabilities, increase temporal density and enable advanced retrieval of forest parameters and dynamics. Yet temporal gaps and irregular sampling in SAR time series will remain limiting factors, particularly over forested regions where temporal decorrelation and acquisition constraints can compromise interferometric analyses. To fully exploit multi-mission synergies and identify optimal baseline configurations for multi-frequency SAR integration, reliable data simulation methodologies are required to reconstruct or interpolate missing SAR observations and ensure consistent temporal coverage for both scientific and operational applications.

In this work, we propose a data-driven simulation framework designed to predict missing single-look complex (SLC) or coherence images in multi-temporal SAR datasets. The approach is developed from perspective that each SAR acquisition represents a unique spatial–temporal “view” of a scene. Given a stack of L available SAR acquisitions over a targeted region, a network is designed and trained to reconstruct one missing observation using the remaining L–1 images as input. The task is formulated as an interpolation problem, where the model learns spatial and temporal dependencies directly from the data. Our current investigation explores the combination of Convolutional Neural Networks (CNNs) and Transformer architectures to jointly capture the spatial structure and temporal evolution of SAR coherence. The CNN component encodes local spatial relationships, while the transformer module models long-range temporal correlations between acquisitions. This hybrid design aims to improve the model’s capacity to predict realistic SAR image or coherence for a given spatial (bₙ) and temporal (tₙ) baseline pair. By controlling the interferometric baselines within the simulation framework, the methodology can be adapted to both near-zero baseline configurations, such as those of ROSE-L or Sentinel-1, and non-zero baseline configurations, such as BIOMASS data. The results obtained from representative applications under each scenario, for example forest tomography in the case of non-zero baselines and vegetation temporal indices in near-zero baseline conditions, will be reported and analyzed to evaluate model performance. Ultimately, this work contributes to the SUPSAR objective of developing innovative approaches that maximize the scientific return of upcoming Copernicus Sentinel Expansion missions and enhance readiness for multi-frequency SAR synergy and data-driven simulation in Earth observation.

In this work, we study some of the building blocks of the processing chain that underpins deformation monitoring using PSInSAR (Persistent Scatterer Interferometric Synthetic Aperture Radar). While a complete understanding of the entire chain is necessary to grasp the technique, an exhaustive treatment would be extremely time-consuming. As a consequence, we focus on two of the main techniques often used in this chain: phase unwrapping and Kriging interpolation. We re-examine the methods in depth and conduct experiments to test their performance. We also provide an open-source implementation and online demo for phase unwrapping.

***Introduction***

In both methods, we study maps recorded at different dates containing a sparse scatter plot of points at which a signal of interest remains coherent over the time series. We shall call these points PSs or Persistent Scatterers [5]. In particular, the scatter plot is not a regular grid, so phase unwrapping has to be adapted for this situation, and Kriging interpolation can be used to obtain an estimate of atmospheric signals on the regular grid.

We shall begin with phase unwrapping, which deals with the removal of the ambiguity mod 2π of the phase differences observed between two dates. Doing so, we follow the steps of Costantini, who first proposed a formulation of the problem as a minimum-cost flow (MCF) optimization over a grid [1], and as a more general linear programming (LP) problem in the case of nonplanar graphs, for instance, those exhibiting a degree of edge redundancy [2]. In both cases, the author seeks to minimize the L1 norm of the correction to be added to the wrapped, observed gradient, ensuring the consistency of the solution across all the cycles of the graph. The solutions are guaranteed to be multiples of 2π, making the methods both elegant and efficient.

-We examined and implemented Costantini’s work from a grid to a general planar graph along with a way to find a minimum cycle basis on which to carry out the MCF algorithm. In particular, we focused on Delaunay triangulations, which are planar graphs like the grid.

-In the case of LP, we tested Costantini’s idea on Delaunay graphs with redundant edges, rendering them nonplanar. We noted that adding redundant edges is rarely useful for unwrapping unless the phase signal is accompanied by severe atmospheric noise. Also, since LP is less efficient than MCF and redundancy adds many edges, redundancy might not only introduce errors but also make the work very computationally intensive.

-We published in the IPOL journal a detailed article with an online demonstration open to the public and an open source implementation [6].

Next, we have work in progress regarding Kriging for atmospheric signal interpolation, mainly based on the literature [3], [4]. In the literature, Kriging has been used for atmospheric delay interpolation [5]. We tested Kriging both on real and simulated data.

-We found that the performance of Kriging interpolation for atmospheric signals is relatively dependent on input parameters, to such an extent that it might be outmatched by parameterless methods such as Natural-neighbor interpolation when the input strays too far from the ground truth. For real atmospheric data, good-quality Kriging typically implies fitting a function such as a covariance or a variogram very precisely. These input parameters (covariance, variogram…) are well-documented and modeled in the literature [3], but we also witnessed significant deviations from the models when using real data, which might be problematic for Kriging with a high level of precision.

-While we also noticed that Kriging is relatively slow, we noted that Kriging is approximately a local weighted nearest neighbors for atmospheric signals, that is, the interpolation at a point depends essentially on the known values of the first few closest sample points. This means Kriging can be achieved and greatly accelerated in a typical divide-and-conquer setting using overlapping patchwork methods without significant loss in precision. The individually Kriged patches are then sewn back together on the overlapping areas using a continuous weighting scheme, yielding a seamless rendering.

-As in the case of phase unwrapping, we are planning to publish an online demo showcasing Kriging interpolation along with an associated, detailed article.

***Method and experiments***

Phase unwrapping relies on the smoothness of the signal to recover its gradient on a graph and, through integration, to recover the full signal up to a global constant. To do so, we seek to remove the residues of the wrapped observed gradient on the graph cycles. Under the smoothness hypothesis, we expect the wrapped observed gradient to be an exact estimate very often; otherwise unwrapping becomes intractable. If errors are made, the estimate is almost surely not a global gradient, i.e., it will display nonzero residues. To recover the signal through integration, we must fix this estimated gradient by adding the least correction in L1 norm to render it conservative by cancelling the residues. When the graph is planar, we can exploit the dual graph to formulate an MCF problem, yielding a solution that cancels out the residues. When the graph is not planar, we switch to the more general but slower LP method, cancelling the residues by writing down the necessary equations over a subset of cycles of minimal cardinality called a cycle basis.

We carried out experiments on both simulated and real terrain profiles [6], and in almost all the cases, we noted a sharp transition from a perfect unwrapping to a very flawed, unexploitable one at a threshold in percentage of wrongly estimated gradients. This was done by gradually contracting or dilating the terrain by manipulating the unit distance, so the wrongly estimated gradients were not distributed uniformly. However, we saw that the transitions were induced by local errors in unwrapping salient features becoming too steep. For instance, dilating a peak too much induces a ring of wrongly estimated gradients around it, ultimately resulting in a local error in a whole region (the interior of the ring, i.e., the peak itself), whose estimation was either bumped up or down by a multiple of 2π. Because those regions represent a fraction of the whole image, this translates into an error curve which is almost zero below the threshold, and then exhibits a sudden, very steep increase, looking like a staircase. Each step or jump corresponds to a spurious region, whose size depends on the area and the magnitude of the error. The only case in which such a sharp transition did not happen was the case of images containing no salient features, e.g., only noise on a flat terrain or a gentle ramp. Also, in almost all the cases, lower redundancy allowed to unwrap steeper terrains. Higher redundancy was shown to be marginally helpful only in the case of severe noise.

Kriging is an interpolation method which can be interpreted as a weakened form of the conditional expectation. More specifically, it is the conditional expectation restricted to linear functions. This restriction makes the computations tractable, relying only on first and second-order moments (for Simple Kriging), or even only on second-order moments (for Ordinary Kriging). It is required to be unbiased, making it the best unbiased linear estimator. It is extensively discussed in the literature [4] and can be used in the context of PSInSAR [5].

Given the values of some sample points (the PS) and a covariance function or a variogram, it is possible to estimate the value at an unknown point by minimizing the square error made when assuming it is a linear combination of the values at the known sample points. If the mean of the signal is unknown, unbiasedness is achieved by adding a condition over the coefficients, which must sum up to one and adds a Lagrange multiplier to the square error to minimize.

It then suffices to solve for the coefficients and infer from them the estimated value from the sample points. If N is the number of points and the number of PS is O(N), then Kriging has time complexity O(N^3). Nevertheless, experiments show that good approximations can be obtained using only the nearest neighbors. Therefore, we propose a method that divides the Kriging process into overlapping patches, that are then sewn back together using a continuous weighting scheme. The result is obtained very quickly with almost no loss in precision.

The only downside is that Kriging is heavily dependent on input parameters like the covariance function or the variogram. If a fit can be achieved with enough precision, Kriging works quite well, otherwise, standard parameterless techniques such as Natural nearest neighbors weighting can outperform it. The problem is that having to fit a variogram can make reliable unsupervised Kriging challenging.

In practice, it is often observed that the power spectral density (PSD) of the atmospheric signal in frequency space follows an inverse power law of exponent alpha, i.e. PSD(k) ~ 1/k^alpha [3]. Assuming that the signal is second-order stationary, that is, its first and second-order moments exist and are invariant under translation, we can use the Wiener-Khinchin theorem to extract the autocorrelation or covariance function of the signal by computing the inverse Fourier transform of the PSD, when possible. However, the PSD might not be integrable as the exponent alpha can be greater than two, leading to a divergence at zero. In this case, the signal isn’t second-order stationary anymore, the Wiener-Khinchin theorem does not hold and the covariance function does not exist. Nevertheless, it can be shown that when 2 < alpha < 4, the variogram gamma behaves in the space domain as gamma(h) ~ h^(alpha-2) .

We experimented Kriging interpolation on both simulated and real data. For simulated data, we used an inverse power-law PSD, typically with exponent 8/3, and we scattered a percentage of PS points uniformly across the image. We show that Kriging works best using a power-law variogram with an exponent close to alpha-2, otherwise significant deviation leads to errors of a magnitude such that even a parameterless method like Natural interpolation might provide better results.

Empirically, for real data, we show that PSD(k) ~ 1/k^(8/3) and gamma(h) ~ h^(2/3) on average, but that some deviation from this model also occurs naturally (~10% in relative value) and in this case the exponent alpha needs to be fitted properly to avoid Kriging errors. As discussed in the previous paragraph, accounting for this phenomenon can introduce significant errors in Kriging. Therefore, good quality Kriging depends on the precision of the fit of the variogram which has to be made for each image.

***Conclusion***

We have described two of the main methods intervening in the building blocks of the PS chain, along with the experimental results that they yielded. It remains to interlock these methods in the right order to remove unwanted components in the original, raw signal and retrieve only the signal related to deformation, which is the goal of this present work.

***Bibliography***

[1] Costantini, M. (1998). A novel phase unwrapping method based on network programming. IEEE Transactions on Geoscience and Remote Sensing, 36(3), 813–821. https://doi.org/10.1109/36.673674

[2] Costantini, M., Malvarosa, F., & Minati, F. (2012). A general formulation for redundant integration of finite differences and phase unwrapping on a sparse multidimensional domain. IEEE Transactions on Geoscience and Remote Sensing, 50(3), 758–768. https://doi.org/10.1109/TGRS.2011.2162630

[3] Hanssen, R. (2002). Radar Interferometry, Data Interpretation and Error Analysis. Kluwer Academic Publishers.https://doi.org/10.1007/0-306-47633-9

[4] Wackernagel, H. (2003). Multivariate Geostatistics (3rd edition). https://doi.org/10.1007/978-3-662-05294-5

[5] Ferretti, A., Prati, C., & Rocca, F. (2001). Permanent scatterers in SAR interferometry. IEEE Transactions on Geoscience and Remote Sensing, 39(1), 8–20. https://doi.org/https://doi.org/10.1109/36.898661

[6] Alexandre Achard-de Lustrac, Roland Akiki, Axel Davy, and Jean-Michel Morel, L1-Norm Redundant Delaunay Phase Unwrapping and Gradient Correction, Image Processing On Line, 15 (2025), pp. 108–162. https://doi.org/10.5201/ipol.2025.583

Homogeneous pixel selection (HPS) is a key step in time-series distributed scatterer InSAR (DS-InSAR) processing, and its quality directly affects the stability and accuracy of local covariance estimation, phase linking, and subsequent deformation time-series inversion. Existing HPS methods can be broadly categorized into nonparametric test-driven approaches and parametric assumption-driven approaches. The former typically determine homogeneity by applying distribution-consistency tests to amplitude (or intensity) time series, such as the Kolmogorov–Smirnov (KS) and Anderson–Darling (AD) tests. Although these methods are statistically interpretable and computationally simple, they often suffer from limited statistical power when the number of acquisitions is small, and are sensitive to distribution shifts under complex scattering mechanisms or pronounced nonstationarity, which leads to insufficient detection of homogeneous pixels, threshold sensitivity, and unstable neighborhood sizes. The latter aim at fast screening by introducing simplified amplitude/intensity statistical models and deriving decision rules based on confidence intervals, with representative methods such as FaSHPS. While being efficient and interpretable, their effectiveness relies on specific modeling assumptions and parameter estimation; model mismatch in real scenes can cause over-selection or under-selection, and the resulting uncertainty may further propagate to covariance estimation and phase linking, ultimately increasing phase residuals and degrading inversion stability. In recent years, deep learning has been introduced into HPS to learn local scattering structures and spatial patterns, but existing strategies typically depend on manually labeled samples or use statistical-test outputs as pixel-wise pseudo-labels for all pixels within a window. Such designs are vulnerable to pseudo-label noise and severe class imbalance, causing conservative predictions, an insufficient number of homogeneous pixels, and unstable spatial patterns. To address these issues, we propose a prior-constrained and consistency-learning DS-InSAR homogeneous pixel selection method, termed DLHPS. DLHPS constructs a statistical prior by fusing voting results from KS, AD, and ttest with respect to the window-center reference pixel, and further extracts high confidence homogeneous and high confidence non-homogeneous sample sets. By replacing dense hard supervision over all window pixels with sparse high-confidence constraints, DLHPS alleviates imbalance-induced degradation and reduces the adverse impact of pseudo-label noise. In addition, DLHPS incorporates amplitude-perturbation-based data augmentation with a dual-view consistency constraint, together with a lightweight spatial coherence regularization, to improve robustness and spatial continuity. Experimental results demonstrate that DLHPS achieves a 90.55% increase in mean coherence and a 71.89% reduction in phase residuals, providing more reliable homogeneous neighborhoods for subsequent DS-InSAR phase linking.

With the high quality, regularly acquired SAR time series that are now available from constellations such as Sentinel-1, ALOS, and others, routine processing of InSAR displacement histories is becoming much more common, with these products reaching a steadily growing community. In recent years, it has also been widely recognized that contributions to the interferometric phase from changes in surface characteristics (e.g., soil moisture and vegetation water content) and shallow surface change (e.g., small-scale deformation at the pixel-to-pixel scale) can introduce biases to interferogram time series beyond simply the bias associated with the process itself. Spatial averaging/ filtering of the complex-valued phase introduces a bias when the phase values within the averaging window have a non-zero skewness that has some correlation over time - this is true even when the "unwrapped", real-valued phase values within that window have a zero-mean. Such biases can significantly degrade the utility of InSAR time series products, particularly for those aimed at relatively new members of the community.

Here we examine methods for assessing whether this bias is "correctable": We explore techniques including the widely-used approach of incorporating redundant interferograms (i.e., not only the nearest-neighbor pairs), as well as an empirical model for the sensitivity of the phase of a given pixel to variations in soil moisture. We have previously used this model to show that the phase changes associated with a large tropical cyclone that impacted the southern Arabian peninsula can be reduced through corrections derived from phase changes associated with two unrelated storms. Each full-resolution pixel exhibits a similar phase change relative to its neighbors for every event. Some pixels are associated with a larger effect than others, to a degree that roughly follows an exponential distribution. A major limitation of this model is that it works best when there are few other sources of temporal decorrelation, such as that introduced by vegetation change. It also implicitly assumes that phase changes are due to soil moisture variability rather than to actual ground displacement. We will show examples from the western United States where these assumptions break down, and explore methods for incorporating independent metrics for ground displacement and soil moisture into our workflow. Examples include the use of co- and cross-polarized data (rather than just co-polarized), and the use of SAR-based soil moisture models (e.g., from the new NASA-ISRO SAR platform, NISAR), or the direct use of the SAR backscatter values that inform such models.

Synthetic Aperture Radar (SAR) enables high-resolution Earth observation, which is less sensitive to weather conditions and day-night cycles, and is therefore widely used for urban infrastructure monitoring and disaster response. In dense urban environments, however, high-rise buildings, bridges, and complex man-made structures frequently cause layover, in which scatterers at different elevations are superimposed within the same resolution cell. This superposition increases ambiguity in interpretation because phase contributions from multiple scattering mechanisms are mixed within a single resolution cell. Because conventional line-of-sight SAR imaging is inherently two-dimensional, it is fundamentally difficult to geometrically separate mixed scatterers. Consequently, 3-D reconstruction with explicit elevation resolution is required to determine the vertical distribution of urban scattering components.

SAR tomography (TomoSAR) is a 3-D imaging technique that exploits a multi-baseline SAR stack acquired over the same area with varying baselines to estimate the reflectivity distribution along the elevation dimension, in addition to the slant-range and azimuth dimensions. In multi-baseline observations, baseline-dependent phase variations can be interpreted as sampling of the elevation spatial frequency, enabling the separation of multiple scatterers within a single resolution cell. TomoSAR can unmix superimposed returns from building facades, roofs, and the ground, and reconstruct the 3-D scattering distribution in urban scenes where layover is prevalent.

In this study, we use an X-band COSMO-SkyMed multi-baseline dataset acquired over urban Pohang, Republic of Korea, to (i) quantitatively detect layover-affected areas and (ii) reconstruct 3-D scattering distributions of man-made structures via elevation-wise scatterer separation. To mitigate sidelobe leakage and false multi-peak responses in the elevation spectrum, we perform precise coregistration and apply phase calibration steps to reduce residual orbital errors and long-wavelength phase components.

For tomographic inversion, we evaluate three representative spectral reconstruction approaches under consistent conditions: conventional beamforming, Capon-based high-resolution spectral estimation, and compressed sensing (CS)-based reconstruction. Beamforming is computationally efficient and straightforward, but may exhibit limited sidelobe suppression. Capon methods employ adaptive weighting to improve effective resolution and sidelobe control. CS-based reconstruction can provide super-resolution under the assumption of elevation sparsity, but it is computationally demanding and sensitive to regularization and model parameters. We compare these methods in terms of effective elevation resolution relative to the theoretical Rayleigh resolution, peak sidelobe ratio, multi-scatterer separability, and computational cost.

Layover detection is formulated as a hypothesis-testing problem using a generalized likelihood ratio test (GLRT). We consider the single-scatterer versus double-scatterer hypotheses within each resolution cell. Based on the tomographic reconstruction, we estimate the maximum likelihood under each hypothesis and compute the GLRT statistic using the likelihood ratio. Pixels for which the double scatterer hypothesis is significantly favored under a false-alarm-controlled threshold are labeled as layover candidates. This approach is intended to reduce false alarms relative to decisions based solely on the presence of multiple peaks, and to produce layover masks and 3-D scatterer point candidates that are directly usable for structural interpretation.

Finally, we extract elevation profiles over representative targets in urban areas (e.g., high-rise apartment complexes, bridges, and major infrastructures) and assess the consistency and error bounds of reconstructed elevations in areas where reference height information is available, such as public elevation datasets (e.g., DSM/topographic products) or optical/visual height cues. By integrating GLRT-based quantitative layover detection with a comparative analysis of tomographic spectral estimation methods, this work aims to provide an optimized framework for urban 3-D reconstruction and quality control, ultimately supporting more reliable deformation analysis and structure-level monitoring.

The revisit cycle of a single synthetic aperture radar (SAR) satellite system directly determines the temporal resolution of interferometric SAR (InSAR) observation, which is a critical factor for accurate surface deformation monitoring. Although both ALOS-4 PALSAR-3 and ALOS-2 PALSAR-2 provide a nominal 14-day revisit cycle, actual acquisition opportunities are often limited by their global observation scenarios, constraining the achievable temporal sampling density. To overcome these limitations, this study investigates the feasibility of cross-sensor InSAR between ALOS-4 PALSAR-3 and ALOS-2 PALSAR-2 under specific acquisition conditions. In general, InSAR processing using different satellite systems is feasible when the sensors operate at the same center frequency, share nearly identical orbital geometry, and have similar incidence angle. ALOS-4 PALSAR-3 and ALOS-2 PALSAR-2 satisfy these conditions, as both operate at L-band and follow the same orbital path with a 14-day revisit cycle. Therefore, their combined use has the potential to reduce temporal gaps and enhance effective temporal resolution. In this study, cross-sensor interferometry was evaluated in the Stripmap (SM) and ScanSAR wide-swath (WD) modes to assess phase continuity across different sensor–mode combinations.

To assess feasibility across varying land-surface cover, datasets were collected in Jinju (South Korea), Hokkaido (Japan), and Chittagong (Bangladesh). This study reports the results from Bangladesh, where the temporal baselines between acquisitions were shortest, minimizing temporal decorrelation effects and allowing a clearer evaluation of coherence behavior. The acquisition dates were: ALOS-4 SM (25 March 2025), ALOS-4 WD (8 April 2025), ALOS-2 SM (14 April 2025), and ALOS-2 WD (28 April 2025). All single look complex (SLC) images were resampled to match the pixel spacing of the ALOS-4 SM reference by compensating for differences in range and azimuth sampling caused by analog-to-digital converter rate and pulse repetition frequency variations, followed by precise co-registration. Differential InSAR processing was applied to remove flat-earth and topographic phase components and generate interferograms for all possible pairs. Common-band and adaptive filtering were used to reduce phase noise. Residual phase ramps were subtracted using a quadratic model of unwrapped interferometric phases.

The highest mean coherence (~0.93) was observed for the 2SM-4SM pair, followed by 2SM-4WD (~0.92), 4SM-4WD (~0.86), 2SM-2WD (~0.71), and 4SM-2WD (~0.55). Interestingly, the 2WD-4WD pair, which is generally considered incompatible for interferometric processing due to significant spectral and acquisition differences, exhibited low mean coherence (~0.40). However, locally coherent interferometric phase signals were still observed in portions of the interferogram, indicating that interferometric phase formation is not entirely precluded even in this challenging configuration. Similar coherence trends were observed in Jinju and Hokkaido, suggesting that the cross-sensor and cross-mode interferometry using ALOS-4 PALSAR-3 and ALOS-2 PALSAR-2 were achievable.

To further interpret variations in coherence, spectral analysis was conducted in both the range and azimuth dimensions. Spectral overlap was more diverse in the azimuth direction than in the range direction. The pairs showing the highest and lowest coherence (2SM–4SM and 2WD–4WD) corresponded to the maximum and minimum azimuth spectral overlap. The other pairs did not consistently exhibit coherence levels proportional to their spectral overlap, suggesting that additional geometric factors beyond spectral overlap influence coherence. Overall, despite variations in coherence values across interferometric pairs, interferogram generation was generally achievable under compatible acquisition conditions. These findings demonstrate the technical feasibility of cross-sensor interferometry between ALOS-2 and ALOS-4 when orbital geometry and incidence angles are sufficiently compatible. The combined use of both satellites can enhance acquisition density and improve temporal resolution for long-term InSAR time-series analysis, thereby strengthening deformation-monitoring capabilities.

Abstract

Due to their flexible acquisition geometry and revisit times, airborne platforms allow for very high-resolution imaging and timely monitoring of areas of interest by means of SAR imaging and repeat-pass SAR interferometry. A potential error source which might compromise the quality of the acquired SAR data (and thus the quality of derived geo/biophysical parameters) are errors in the sensor positioning. Such positioning errors are due to the limited accuracy of the employed GNSS-aided inertial navigation systems (INS) and may lead to not optimally focused SAR images as well as phase undulations in the single-pass and repeat-pass interferograms [1-2].

Previous work showed that residual positioning errors of agile platforms can be reduced by placing an ad-hoc GNSS reference station in the vicinity of the SAR sensor’s flight path, compared to the case in which only remote GNSS reference stations of a network of permanent GNSS receivers are employed [3].

In the present work, the impact of residual positioning errors in repeat-pass L and S-band InSAR data acquired from UAV platforms [4,5] is analyzed. The data were acquired using Gamma L-band and S-band SAR sensors, while the navigation setup consisted of a Honeywell Guide n580/n500 GNSS-aided INS, with a GNSS ad-hoc reference station located nearby.

To assess the impact of potential residual errors in the repeat-pass interferograms, a subaperture-based approach was employed. Specifically, the original full aperture of the acquisitions composing each interferometric pair was split into multiple (possibly overlapping) subapertures of given length. Each subaperture was focused independently for each of the acquisitions. Image focusing was performed by means of a time-domain backprojection algorithm [6], which can accurately process data acquired even from significantly non-linear trajectories, as can happen in the case of UAV platforms. For each subaperture, interferograms were computed from the pair of SAR images thus obtained. Finally, the presence of residual unknown positioning error was assessed by checking for phase trends in the differential interferograms between interferograms of different subapertures.

First results obtained from the analysis of a UAV-borne L-band dataset do not show significant phase undulations as a result of potential residual positioning. This is proven also by the high quality of the full-aperture interferogram, which has does not show any substantial phase variations, also due to the very short temporal baseline. The lack of significant residual positioning errors can be attributed to several factors, first of all the high accuracy of the navigation data (in the presence of a GNSS station nearby), combined with the relatively long wavelength of the L-band acquisitions. In our contribution, we will expand our analysis to further UAV-borne SAR data sets including different wavelengths.

Bibliography

[1] G. Fornaro, G. Franceschetti and S. Perna, "Motion compensation errors: effects on the accuracy of airborne SAR images," in IEEE Transactions on Aerospace and Electronic Systems, vol. 41, no. 4, pp. 1338-1352, Oct. 2005, doi: 10.1109/TAES.2005.1561879.

[2] N. Cao et al., "Estimation of Residual Motion Errors in Airborne SAR Interferometry Based on Time-Domain Backprojection and Multi-squint Techniques," in IEEE Transactions on Geoscience and Remote Sensing, vol. 56, no. 4, pp. 2397-2407, April 2018, doi: 10.1109/TGRS.2017.2779852.

[3] R. Coscione, I. Hajnsek, C. Werner and O. Frey, "Assessing the Impact of Positioning Errors in Car-Borne Repeat-Pass SAR Interferometry With a Controlled Rail-Based Experiment," in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 15, pp. 8402-8415, 2022, doi: 10.1109/JSTARS.2022.3193053.

[4] O. Frey, C. L. Werner and R. Coscione, "Car-borne and UAV-borne mobile mapping of surface displacements with a compact repeat-pass interferometric SAR system at L-band," Proc. IEEE Int. Geosci. Remote Sens. Symp., Yokohama, Japan, 2019, pp. 274-277, doi: 10.1109/IGARSS.2019.8897827.

[5] O. Frey, C. Werner, S. Leinss, T. Batt, R. Caduff, T. Dixon, T. Sadeghi Chorsi, R. Van Alphen, M. Schmitt, M. Eitel, F. Sica, E.J. Deeb, A.L. LeWinter, D.L. Filiano, C.J. Wagner, Z. Hoppinen, "Multicopter-UAV- and car-borne repeat-pass SAR interferometry and SAR tomography with the compact Gamma SAR systems: first examples and use cases at S- and L-band", In Proc. IEEE Int. Geosci. Remote Sens. Symp., Brisbane, Australia, Aug. 2025, pp. 1374-1377. IEEE. doi:10.1109/IGARSS55030.2025.11243477.

[6] O. Frey, C. Magnard, M. Ruegg and E. Meier, "Focusing of Airborne Synthetic Aperture Radar Data From Highly Nonlinear Flight Tracks," in IEEE Transactions on Geoscience and Remote Sensing, vol. 47, no. 6, pp. 1844-1858, June 2009, doi: 10.1109/TGRS.2008.2007591.

Accurate measurement of north-south surface deformation at sub-centimetre scale is essential for monitoring tectonic processes such as interseismic strain accumulation and postseismic relaxation along strike-slip fault systems. However, conventional Interferometric Synthetic Aperture Radar (InSAR) observations are primarily sensitive to vertical and east-west motion due to the side-looking radar acquisition geometry, leaving the north-south component poorly constrained. Future SAR missions aim to address these limitations by providing additional viewing geometries and improved azimuth displacement measurements, including the NASA-ISRO Synthetic Aperture Radar (NISAR) mission, ESA’s ROSE-L mission planned for launch in 2028, and ESA’s Harmony mission expected in 2030, which will operate alongside Sentinel-1 with companion satellites to improve line-of-sight diversity and enable more complete three-dimensional deformation measurements.

Here we address this limitation by integrating Subswath and Burst Overlap Interferometry (SBOI) derived from Sentinel-1 TOPS acquisitions within the COMET-LiCS processing infrastructure. The objective is to generate operational along-track interferometric measurements that complement conventional line-of-sight InSAR products and improve the characterization of north-south deformation using the already more than decade-long Sentinel-1 observation record since end of 2014.

The developed workflow extends the COMET LiCSAR processing system (Lazecký et al., 2020) to automatically generate SBOI interferograms from Sentinel-1 data. To improve the reliability of the along-track signal, several correction terms are implemented, including mitigation of ionospheric gradient effects, and removal of non-tectonic contributions such as solid Earth tides and plate motion projected into the along-track direction. The resulting datasets are analysed using COMET LiCSBAS time-series processing (Morishita et al., 2020) and evaluated across several north-south-oriented strike-slip fault systems worldwide, including the Dead Sea Fault, Chaman Fault, and East Anatolian Fault.

The SBOI interferograms are generated from the same Sentinel-1 acquisitions as the conventional line-of-sight interferograms, ensuring temporal consistency between datasets while providing an additional measurement geometry. After applying the correction framework, the resulting time series reveal localized north-south tectonic deformation with millimetre-level sensitivity along major strike-slip fault systems.

The integration of SBOI processing into the COMET-LiCS portal provides open and accessible along-track interferometric products for the geodetic community. When combined with conventional InSAR and GNSS observations (Elliott et al., 2025), these datasets improve the capability to resolve three-dimensional tectonic deformation and prepare the community for future SAR missions with enhanced viewing geometries.

Elliott, J. R., Fang, J., Lazecký, M., Maghsoudi, Y., Ou, Q., Payne, J. A., Rollins, C., Wang, D., & Hooper, A. (2025). Deformation, Strains and Velocities for the Alpine Himalayan Belt from trans-continental Sentinel-1 InSAR & GNSS.

Lazecký, M., Spaans, K., González, P. J., Maghsoudi, Y., Morishita, Y., Albino, F., Elliott, J., Greenall, N., Hatton, E., Hooper, A., Juncu, D., McDougall, A., Walters, R. J., Watson, C. S., Weiss, J. R., & Wright, T. J. (2020). LiCSAR: An Automatic InSAR Tool for Measuring and Monitoring Tectonic and Volcanic Activity. Remote Sensing, 12(15), 2430. https://doi.org/10.3390/rs12152430

Morishita, Y., Lazecky, M., Wright, T. J., Weiss, J. R., Elliott, J. R., & Hooper, A. (2020). LiCSBAS: An Open-Source InSAR Time Series Analysis Package Integrated with the LiCSAR Automated Sentinel-1 InSAR Processor. Remote Sensing, 12(3), 424. https://doi.org/10.3390/rs12030424

Monitoring ground deformation in urban environments is essential for infrastructure stability, hazard mitigation, and long-term urban planning (Kumar et al., 2021; Ramirez et al., 2020). The continuous acquisition of Sentinel-1 SAR imagery enables large-scale deformation monitoring using time-series interferometric techniques, with Persistent Scatterer Interferometry (PSI) widely applied to retrieve long-term displacement signals from stable radar targets (Crosetto et al., 2016). The StaMPS (Stanford Method for Persistent Scatterers) framework provides a well-established methodology for PSI time-series analysis and is widely adopted within the InSAR community (Hooper, 2008; Hooper et al., 2010). However, the original StaMPS implementation relies on a MATLAB-based environment and a multi-stage workflow that requires several external dependencies and manual configuration steps, which can limit accessibility and integration within modern data-processing pipelines. To address these limitations, this work presents PyStamps, a Python-based open-source implementation of the StaMPS PSI workflow developed within the UrbanSAR framework (Girohi et al., 2025), enabling flexible and scalable PSI processing.

PyStamps preserves the core methodological structure of StaMPS, including temporal coherence estimation, persistent scatterer selection, phase unwrapping, and spatially correlated look-angle (SCLA) correction (Hooper et al., 2010), while adapting the workflow to a fully open-source Python architecture. The implementation leverages widely used scientific computing libraries to efficiently perform interferometric phase analysis and deformation estimation (Harris et al., 2020; Virtanen et al., 2020). The framework incorporates scalable data management and optimized computational strategies to support large interferometric stacks and improve computational efficiency. These developments maintain compatibility with the established StaMPS methodology while enabling greater flexibility for integration into automated and large-scale analysis systems.

The performance and reliability of PyStamps were evaluated using Sentinel-1 InSAR time-series data over the Houston metropolitan region in the United States, an area characterized by well-documented subsidence associated with groundwater extraction and urban development. Two regions of interest were defined to assess scalability, including a compact study region and a larger regional extent within the same geographic setting. Results indicate that PyStamps produces deformation estimates consistent with those obtained using the original MATLAB-based StaMPS workflow. Velocity distributions from both implementations exhibit nearly identical behavior, with persistent scatterer velocities agreeing at sub-millimeter precision and more than 99.99% spatial overlap in the compact test area. Performance comparison further indicates that PyStamps operates with lower computational overhead, demonstrating reduced CPU usage while maintaining stable processing performance across different regions of interest and spatial extents.

Overall, PyStamps provides a robust and permissively licensed implementation of the StaMPS workflow, enabling scalable PSI processing within a modern open-source InSAR framework. The approach supports efficient large-scale deformation monitoring and facilitates integration of PSI analysis into contemporary scientific computing and operational geospatial environments.

Instantaneous State InSAR (IS-InSAR) was recently introduced as an alternative to traditional history-based parameterizations for deformation monitoring, enabling recursive near-real-time estimation of displacement through a state-space formulation. Central to this framework is the smoothness doublet, which governs the stochastic behavior of the instantaneous velocity via an Ornstein–Uhlenbeck process. The smoothness doublet consists of two physically interpretable parameters: (1) the variance of the instantaneous velocity and (2) its correlation time. Together, these parameters control the degree to which temporal variations in motion are permitted, which influences both the predicted state evolution and the relative weighting of new observations in the recursive update.

In this contribution, we present a systematic analysis of the smoothness doublet—which needs to be chosen based on external contextual information—and its impact on IS-InSAR performance. We investigate how variations in the velocity variance determine the amplitude of allowable short-term deviations from the mean motion, while changes in the correlation time regulate the temporal persistence of these deviations. Adjusting these parameters modifies the estimated instantaneous position and velocity components of the state vector, and propagates into the stochastic model underlying the phase observations. As a consequence, the estimation and validation of integer phase ambiguities—an essential component of high-precision InSAR—are significantly affected.

We demonstrate that small velocity variances and long correlation times approximate the traditional assumption of near-infinite smoothness, leading to conservative updates and reduced responsiveness to behavioral change. Conversely, larger variances and shorter correlation times increase adaptivity but may amplify noise sensitivity and ambiguity instability. The smoothness doublet therefore acts as a critical tuning mechanism that balances stability and responsiveness in near-real-time deformation monitoring.

To quantify these effects, we introduce a set of evaluation metrics that assess (i) state prediction consistency, (ii) innovation statistics, i.e., the difference between observations and predictions, (iii) ambiguity resolution reliability, and (iv) temporal responsiveness to behavioral change. Through simulated and experimental scenarios, we illustrate how different configurations of the smoothness doublet influence displacement reconstruction quality and ambiguity robustness.

Our results provide practical guidance for selecting and tuning smoothness parameters in application-aware (triple-A) IS-InSAR projects. By explicitly analyzing the stochastic structure of the instantaneous velocity process, this study establishes a rigorous framework for optimizing recursive deformation monitoring under varying dynamic conditions, thereby enhancing the reliability and interpretability of near-real-time InSAR products.

Reliable characterization of uncertainty in Interferometric Synthetic Aperture Radar (InSAR) deformation time series is essential for the correct interpretation of geophysical signals. Although substantial progress has been made in modelling individual noise sources in InSAR data (such as decorrelation effects, atmospheric delays), most studies focus on the statistical properties of noise at the interferogram level time-series. However, InSAR deformation products are derived through complex time-series processing chains that include spatial and temporal filtering, atmospheric phase screen mitigation, and inversion procedures. These processing steps fundamentally modify the statistical characteristics of the noise and introduce correlations across both space and time, which are rarely accounted for in existing uncertainty descriptions.

Although error propagation through the InSAR processing chain has been investigated, most existing approaches rely on the assumption that the stochastic properties of interferometric observations are known a priori. In practice, however, these properties are rarely known with sufficient accuracy, as they depend on factors such as acquisition geometry, surface characteristics, and noise contributions that may vary across scenes and acquisitions. Consequently, uncertainty modelling based solely on predefined stochastic assumptions may not adequately represent the true noise characteristics of InSAR observations, limiting the reliability of the resulting deformation estimates. Moreover, the processing parameters and filtering strategies vary significantly between studies, making it difficult to define a generic stochastic model for InSAR deformation products.

In this study, we present an integrated framework for numerical assessment of the spatio-temporal noise structure in InSAR deformation time series by combining rigorous error propagation with Variance Component Estimation (VCE). The proposed method propagates uncertainty from interferometric observations to the final deformation estimates while simultaneously estimating the variance components associated with dominant noise sources. This integrated strategy removes the need for prior assumptions about the exact noise structure of the interferograms and allows the stochastic model to be inferred directly from the data within the processing framework.

Particular emphasis is placed on the computational efficiency and scalability of the approach to enable its application to large InSAR datasets. An iterative estimation strategy is adopted, in which the noise structure is first characterised in the temporal domain and subsequently refined in the spatial domain. The framework focuses primarily on two dominant noise components in InSAR time series: atmospheric disturbances and decorrelation noise. In addition, the method accounts for model-induced noise arising from deviations between the true deformation signal and the functional models implicitly or explicitly assumed during spatio-temporal filtering and inversion. For error propagation, a flexible and general framework is employed, capable of accommodating different processing settings and spatio-temporal assumptions within the applied processing chain.

The methodology is demonstrated using Sentinel-1 data over several subsidence regions in Iran, illustrating its applicability for large-scale InSAR time-series analyses. The results show that InSAR deformation time series commonly exhibit strongly correlated noise patterns in both spatial and temporal domains. The proposed framework quantifies how these correlations depend on the initial measurement uncertainties as well as on the applied processing settings. Such correlated noise can occasionally mimic real deformation signals, leading to potential misinterpretations. By explicitly estimating and propagating the noise structure through the processing chain, the proposed method enables a more reliable assessment of deformation signals.

High-resolution wide-swath (HRWS) SAR missions often use staggered acquisitions, varying the pulse repetition frequency (PRF) along azimuth to enlarge the swath while maintaining high resolution. A key side effect is spatially variant azimuth ambiguities that appear in multiple regions and degrade image quality. Existing countermeasures (e.g., Doppler shifting and coherence-based detection) can leave residual artifacts, while subaperture averaging reduces azimuth resolution.

We propose an azimuth-ambiguity removal framework based on partially-missing-band synthetic-aperture processing. By repeatedly forming images while selectively excluding narrow azimuth-frequency (Doppler) bands, we isolate sidelobe-dominated components and preserve mainlobe returns. Contaminated bands are detected using a per-iteration mean-centered azimuth-time distance metric, which suppresses band-dependent biases and highlights sidelobe outliers, and are then suppressed while maintaining the effective imaging bandwidth required for high azimuth resolution.

Experiments on ALOS-4 PALSAR-3 Level-1.2 variable-PRF data show reduced ambiguity over both ocean and land scenes and improved interferometric phase quality, indicating benefits for InSAR applications. Because the method targets ambiguity behavior induced by synthetic-aperture processing rather than sensor-specific tuning, it is expected to generalize to other variable-PRF missions such as NISAR and ROSE-L.

Ground-based frequency-modulated continuous-wave (FMCW) radar systems provide a versatile platform for high-resolution imaging of moving targets under well-controlled observational geometries. Compared with airborne or spaceborne systems, ground-based configurations enable repeated measurements, flexible parameter selection, and detailed experimental analysis of imaging characteristics. In this study, full-polarimetric inverse synthetic aperture radar (ISAR) imaging of a ship target is experimentally characterized using the GPRI-II ground-based Ku-band radar system. The objective is to evaluate polarization-dependent ISAR image formation and to examine the practical implementation of coherent full-polarimetric processing in a maritime observation scenario.

The GPRI-II radar operates with FMCW waveforms and supports fully polarimetric data acquisition. The experiment was conducted in a coastal environment, where the radar antenna remained stationary in a staring configuration while the ship's motion generated the synthetic aperture required for cross-range resolution. In this geometry, the target motion induces Doppler modulation over the coherent processing interval, enabling two-dimensional ISAR image formation without physical antenna movement. Fully polarimetric measurements were recorded in the HH, HV, VH, and VV channels, allowing comprehensive characterization of scattering mechanisms across polarization states.

Experiments were conducted using multiple chirp durations (0.5, 2, 4, and 16 ms) to investigate the influence of chirp duration on ISAR image formation. For the 2 ms acquisition, corresponding to a pulse repetition frequency of approximately 125 Hz, the effective coherent interval contributing to azimuth focusing is approximately 19 s for the ship moving at a nominal speed of 16.7 km/h. Under these conditions, with the ship observed at a slant range of about 430 m and predominantly cross-track motion, the available slow-time aperture provides sufficient Doppler bandwidth for high-resolution azimuth focusing, yielding cross-range resolution in the centimeter range under stable motion conditions. Varying the chirp duration modifies the PRF and consequently, the Doppler sampling characteristics, directly influencing focusing behavior in the azimuth direction and the achievable cross-range resolution. A unified signal-processing workflow was applied consistently across all channels to ensure comparability. The processing sequence included FMCW range compression through matched filtering, selection of a region of interest containing the ship target, and slow-time phase error estimation using a phase-gradient autofocus algorithm applied to the dominant co-polarized channel. To preserve inter-channel phase coherence, the phase correction estimated from the reference co-polarized channel was applied identically to the remaining channels prior to the Fourier transform in the azimuth direction. This strategy enables the coherent reconstruction of full-polarimetric ISAR images while minimizing relative phase inconsistencies arising from independent channel processing. Radiometric normalization and dynamic range adjustment were further applied to facilitate visual and quantitative comparison among polarization channels.

The reconstructed ISAR images exhibit pronounced polarization-dependent scattering. The co-polarized channels (HH and VV) exhibit strong, spatially concentrated scattering centers corresponding to dominant structural components of the ship, including the hull, deck boundaries, masts, and superstructure elements. Specular reflections from metallic surfaces and double-bounce interactions between vertical structures and the water surface are clearly observed in these channels. In contrast, the cross-polarized channels (HV and VH) present comparatively lower overall intensity but reveal complementary structural details. Cross-polarized returns emphasize depolarization mechanisms associated with geometrically complex features, tilted surfaces, and multiple-scattering interactions. These differences highlight the sensitivity of polarimetric ISAR imaging to target geometry and orientation.

To further interpret scattering behavior, Pauli-based polarimetric combinations were generated to visualize the relative contributions of different scattering mechanisms. Such representations enhance the discrimination between dominant structural scatterers and more diffuse or anisotropic responses. The spatial distribution of polarimetric signatures provides additional insight into structural heterogeneity across the ship body. Comparative analysis across channels confirms that full-polarimetric acquisition significantly improves interpretability relative to single-polarization imaging, particularly in distinguishing coherent structural reflections from depolarized components.

The experimental results demonstrate the practical feasibility of full-polarimetric ISAR imaging using a ground-based FMCW radar platform and validate coherent multi-channel processing in a realistic maritime environment. The study provides a systematic assessment of polarization-dependent ISAR image characteristics and establishes an experimental foundation for further investigations into polarimetric target characterization and classification. These findings support the broader application of ground-based full-polarimetric ISAR systems in maritime monitoring, structural analysis, and radar-based target interpretation.

Phase bias in multilooked short-baseline interferograms introduces systematic distortions in InSAR time series, particularly in regions of low long-term coherence. While phase linking (PL) techniques can mitigate these effects, their applicability remains limited in densely vegetated or seasonally dynamic environments.

We previously developed a mitigation strategy based on short-term loop-closure analysis and bias inversion from wrapped interferograms [1, 2]. The approach estimates bias terms from closure residuals and stabilises the solution using temporal regularisation. Validation over selected test sites demonstrated significant reduction of spurious deformation signals in vegetated regions.

Here, we extend this framework to a nationwide-scale implementation over the Italian peninsula using seven ascending Sentinel-1 COMET-LiCSAR frames covering the entire region. The dataset spans multiple climatic zones and land cover classes, providing a comprehensive test bed for large-scale assessment. All interferograms required for closure analysis were generated using LiCSAR processing, and the inversion was performed under optimised regularisation weights derived from misfit–smoothness trade-off analysis.

To further reduce unstable bias estimates, we implemented a two-stage quality-control framework. First, bias unknowns insufficiently constrained by loop-closure observations were identified and nulled based on per-unknown support metrics. Second, a temporal consistency criterion was applied to the estimated bias time series using a circular coherence measure within a moving temporal window to suppress oscillatory or poorly constrained solutions. This strategy significantly improved inversion stability in low-coherence and seasonally decorrelated regions.

The large-scale processing enables systematic evaluation of phase bias behaviour across forests, agricultural areas, mountainous terrain, and urban regions. We quantify residual loop closures before and after correction and analyse the seasonal variability of bias amplitude. Results show that phase bias is strongly correlated with vegetation dynamics and moisture conditions, leading to false subsidence or uplift patterns in uncorrected velocity fields. The proposed correction significantly reduces closure residuals and achieves strong agreement with PL-derived velocities in coherent areas, while preserving meaningful deformation signals in decorrelated regions where PL fails.

This work demonstrates the scalability and operational robustness of the proposed phase bias correction framework and provides new insights into the large-scale spatial and temporal characteristics of InSAR phase bias.

[1] Y. Maghsoudi, A. J. Hooper, T. J. Wright, M. Lazecky, and M. Pinheiro, "Advances in mitigating InSAR non-closure phase bias: A refined processing approach," Science of Remote Sensing, vol. 12, p. 100304, 2025/12/01/ 2025.

[2] Y. Maghsoudi, A. J. Hooper, T. J. Wright, H. Ansari, and M. Lazecky, "Characterizing and Correcting Phase Biases in Short-Term, Multilooked Interferograms," Remote Sensing of Enironment (in review), preprint available at EarthArXiv, 2021.

Interferometric Performance Evaluation of Sentinel-1C and Sentinel-1D: Results from the Independent Calibration Campaign of the Commissioning Phase

Matteo Nannini¹, Andrea Pulella¹, Pau Prats-Iraola¹, Patrick Klenk¹, Dirk Geudtner²

¹ Microwaves and Radar Institute, German Aerospace Center (DLR), Germany

Münchener Str. 20, 82234 Weßling, Germany

Email: matteo.nannini@dlr.de

²European Space Agency (ESA-ESTEC)

This contribution summarizes the interferometric investigations carried out during the in-orbit commissioning phase of both Sentinel-1C and Sentinel-1D satellites. The presented investigations were conducted by the DLR-HR Institute during the commissioning phase of the two spacecrafts as part of the independent calibration campaign for the Sentinel-1C and Sentinel-1D SAR systems on behalf of ESA [1]. The analyses were performed on Level-1 data processed by the IPF processor of ESA. Sentinel-1C was launched on 5 December 2024 to replace Sentinel-1B (decommissioned after suffering non-recoverable on-board failures at the end of 2021). Sentinel-1D was launched on 4 November 2025 to eventually replace Sentinel-1A.

This contribution focuses on reporting the Sentinel-1 InSAR-related performance in terms of the main interferometric parameters: perpendicular baseline, Doppler centroid, common Doppler bandwidth, and burst mis-synchronization. These figures of merit are particularly critical for the TOPS acquisition mode [2] adopted by the Copernicus Sentinel-1 constellation as default mode for acquisitions over land [3], as this mode imposes stringent interferometric requirements [4]. For example, assuming most of the above-mentioned parameters as nominal, the synchronization among bursts must remain within ±5ms to enable high-quality interferometry. These parameters have been evaluated through a statistical analysis of hundreds of InSAR data pairs acquired by the sensors during their respective commissioning phases, hence providing reliable statistics to assess the suitability of the system for InSAR applications. The annotation data, which accompany the SAR data, were the primary source of information for this analysis. It is worth noting, that after the retirement of Sentinel-1B, Sentinel-1A has operated concurrently with Sentinel-1C to ensure observational continuity. Consequently, some of the results presented in this work were obtained involving Sentinel-1A data, still demonstrating the usefulness of Sentinel-1A for InSAR. The analyses presented in this contribution cover both the co-sensor case, which involves data acquired by the same spacecraft (e.g., Sentinel-1C), to assess the individual sensor performance, and the cross-sensor case, which involves generating cross-interferograms between the different sensors. This latter analysis is crucial due to the synergistic design of the Copernicus constellation, since the cross-sensor configuration enables the possibility of halving the repeat-pass time. In addition to that, the commissioning phase provided a unique opportunity to begin analyzing data before the final operational orbit configuration with a 180° orbit phasing (i.e., the nominal six-day repeat) was implemented. Specifically, the first repeat cycles after the launch of a new Sentinel-1 sensor (C or D) were operated in a 30° orbit phasing, allowing for a one-day exact repeat with respect to the corresponding predecessor. This unique orbit phasing for the initial cycles enabled the generation of one-day repeat-pass cross-interferograms.

To additionally evaluate the stability of the system in terms of the aforementioned parameters, long data take analyses were performed, demonstrating exceptional performance of both sensors and ensuring high-quality interferometric products.

In addition to the quantitative assessment of the Sentinel-1C/D spacecrafts performance through interferometric statistical analysis, this contribution also presents a set of interferometric results to qualitatively demonstrate the impressive achievable interferometric performance. Stationary scenarios, such as the Atacama Desert [4], monitoring of ground deformation in Mexico City, and non-stationary scenarios, including the Mt. Fentale, Zachariæ Isstrøm Glacier, and the 2025 Myanmar earthquake [5], are exemplary outcomes that were obtained during the commissioning phase of Sentinel-1C. Furthermore, the performance for the different SAR acquisition modes, such as the Interferometric Wide (IW), Extra Wide (EW), and Stripmap modes will be reported. These results highlight the system's inherent operational readiness. The contribution provides insights into some of the analyses concerning interferometry that were conducted during the commissioning phase, underlying the importance of this phase in performing final system tuning before providing data to the end users.

This comprehensive interferometric analysis confirms the readiness of the Sentinel-1C and Sentinel-1D sensors for operational use, ensuring high-quality interferometric products for a wide range of applications, from environmental monitoring to disaster response.

Literature

[1] Klenk, P., Giez, J., Schmidt, K., Nannini, M., & Schwerdt, M. "Independent calibration of the Sentinel-1C SAR system." EUSAR 2024; 15th European Conference on Synthetic Aperture Radar. VDE, 2024.

[2] F. De Zan and A. Monti Guarnieri, TOPSAR: Terrain Observation by Progressive Scans, IEEE Transactions on Geoscience and Remote Sensing, Vol. 44, No. 9, September 2006, pp 2352-2360.

[3] R. Torres, R., et al., “GMES Sentinel-1 Mission”, Special Issue of Journal of Remote Sensing of Environment “The Sentinel Missions – New Opportunities for Science”, Vol. 120, pp. 9-24, May 2012.

[4] N. Yague-Martinez, Prats, P., Gonzalez, F, R., Brcic, R., Shau, R., Eineder, M., Geudtner, D. and Bamler, R., “Interferometric Processing of Sentinel-1 TOPS Data”, IEEE Trans. Geoscience and Remote Sensing, Vol. 54, No. 4, pp. 2220-2234, 2016.

[5] ESA - Sentinel-1C demonstrates power to map land deformation

[6] ESA - Sentinel-1 captures ground shift from Myanmar earthquake

The accessibility of Synthetic Aperture Radar (SAR) data, particularly from missions such as Sentinel-1, has transformed remote sensing and fueled rapid growth in both scientific research and operational applications. However, converting raw interferometric phase measurements into reliable, actionable displacement products remains technically complex and poses a significant barrier for non-specialists. To bridge this gap, the Observational Products for End-Users from Remote Sensing Analysis (OPERA) project spearheaded by the NASA Jet Propulsion Laboratory, California Institute of Technology, leverages cloud-based processing to deliver standardized, continental-scale InSAR Analysis-Ready Data (ARD).

This presentation focuses on the OPERA Surface Displacement (DISP) product suite, which systematically generates displacement time series across North America. Notably, DISP is the first-ever CEOS-ARD compliant InSAR product, setting a new international standard for interoperable InSAR time series data. OPERA is currently processing the Sentinel-1 archive from 2016, providing a continuous record of ground motion through the present. We will also highlight the NASA-ISRO SAR (NISAR) mission and the plan to integrate NISAR to the OPERA DISP suite (DISP-NI). This will complement the DISP-S1 products by significantly expanding spatial coverage, increasing temporal revisit times, and improving coherence in vegetated areas over North America. The DISP suite enables investigation of natural and anthropogenic processes, including but are not limited to volcanic unrest, tectonic processes, landslides, and land subsidence caused by hydrocarbon or groundwater extraction, by providing standardized displacement time series that reduces the computational burden of large-scale InSAR time series processing.

These products also serve as the foundation for the forthcoming Vertical Land Motion (VLM) suite, a higher-level derivative that integrates InSAR displacement time series with geodetic reference frames and GNSS observations. The VLM suite is designed to provide consistent vertical displacement time series, and where feasible, horizontal components, enabling more robust and direct assessments of long-term subsidence, uplift, and broader crustal deformation processes.

Beyond ground motion, OPERA also provides a comprehensive portfolio of Earth observation Level-3 products derived from SAR (Sentinel-1, NISAR) and optical sensors (Sentinel-2, and Landsat 8/9) on a near-global scope. These include the Dynamic Surface Water Extent (DSWx) and Surface Disturbance (DIST) suites for monitoring hydrological variations and land disturbance change. In this presentation, we will also showcase OPERA’s intermediate products: Radiometric Terrain Corrected (RTC) backscatter and Coregistered Single-Look Complex (CSLC) products. While these datasets serve as the primary inputs for the Level-3 suites, they also enable advanced users to generate customized products and perform specialized geophysical analyses. All OPERA products are freely available through the NASA Distributed Active Archive Centers and Earthdata Search, supporting easy access and rapid use in both research and operational contexts.

Differential SAR interferometry (DInSAR) and Multi-Temporal DInSAR (MTInSAR) are largely exploited for measuring slope stabilities. Several datasets are currently available at different wavelengths, spatial resolutions, and revisit times, spanning national or continental areas, and collectively covering long time periods (even more than 20 years). A reliable monitoring of ground instabilities and related early warning signals requires a detailed analysis of both spatial patterns and time series of displacement measurements derived from DInSAR/MTInSAR for assessing their availability, reliability, and significance with respect to the application on hand. End users, indeed, should be able to check accurately the availability of DInSAR/MTInSAR-based information over the area of interest, to model the sensitivity of interferometric measurements to the ground displacements, and to recognize different signal components and possible artifacts affecting the MTInSAR products, such as, for instance, those related to atmospheric artifacts or phase unwrapping errors. Consequently, end users need specialized skills and, possibly, tools, which may support a reliable exploitation of DInSAR/MTInSAR products covering wide areas and long time periods and consisting of a huge number of coherent targets (up to millions) (Bovenga, 2024).

This is particularly pertinent when dealing with applications, such as slope instabilities, occurring in critical environmental settings that negatively impact DInSAR/MTInSAR products (Wasowski and Bovenga, 2014). First, the steep topography may lead to unfavourable illuminating conditions in terms of either unfeasible detection over layover and shadow areas or low sensitivity to the ground displacement. Second, the presence of dense vegetation and changeable cover conditions causes DInSAR signal decorrelation and a low density of MTInSAR coherent targets (CTs). Third, displacement kinematics are characterised by nonlinear components and high displacement rates, leading to measurements corrupted by aliasing. All these critical issues negatively impact the applicability and interpretation of this well-established technology.

We developed a QGIS plugin based on the PyQGIS library (Bovenga and Piccolino, 2025), which, starting from standard DInSAR/MTInSAR products and a few ancillary layers, derives additional products useful for supporting the end-user interpretation and ground instability assessment over the area under investigation.

First, the tool estimates the visibility of the area of interest (AOI) with respect to the satellite line of sight (LOS). It combines the satellite acquisition geometry and the ground geomorphic information to derive an index of visibility, which allows end-users to check the applicability of DInSAR analysis over the AOI just based on geometrical factors and before performing DInSAR processing.

Moreover, the reliability of DInSAR products may also depend on the orientation of the local slope within the AOI. For instance, when dealing with landslides, for slopes facing north or south, the downslope movement is basically perpendicular to the LOS direction, thus leading to unfeasible DInSAR-based estimation of displacements. Hence, the tool estimates the percentage of downslope movement captured from the DInSAR geometry along the LOS and, for each CT, computes the downslope mean displacement rate corresponding to the LOS component measured by MTInSAR.

These outputs of the tool may be combined with other layers such as NDVI, DInSAR coherence, and landslide inventory for performing a feasibility analysis before DInSAR/MTInSAR processing for both checking the reliability and supporting the interpretation of DInSAR/MTInSAR products for ground instabilities.

Moreover, the tool automatically computes the percentage of the AOI surface covered by coherent targets (CTs). This allows end users to estimate how significant the information derivable from MTInSAR within the AOI is and to decide whether more complementary information is needed for assessing the instability of the area or not (Bovenga et al., 2023).

Finally, the tool investigates the spatial homogeneity of the CTs' distribution within the AOI. Indeed, a relatively high percentage of surface covered by CTs does not necessarily imply a uniform coverage of the entire AOI, which may include regions lacking CTs. Hence, the tool, by using the inhomogeneous Ripley’s K-function (Dixon, 2002), detects the presence of voids in the CT spatial distribution, resulting in a lack of kinematic information needed for a reliable assessment of the ground instability.

Since the tool was designed to deal with generic DInSAR/MTInSAR products, we tried to reduce the parameters required in the inputs associated with SAR/DInSAR processing, which may be inaccessible to the end user. To this aim, we developed an approximated formula for calculating the local heading angle, defined as the angle between the satellite ground track and a meridian for an arbitrary point on the ground track, by using just the pixel latitude, incident angle, the orbit elevation, and the number of satellite revolutions per day. Moreover, we developed an approximated relationship between look angle and incident angle (without the need for precise geocoding computation), in case the incident angle is not available. We assessed the errors associated with both the local heading angle and the look angle computed through the approximate formulas by comparing them with the values computed through the correct geometrical formulation involving orbital state vectors, carried out through a standard interferometric processing chain. We performed this analysis by exploiting four Sentinel-1 IW products acquired from both ascending and descending passes over two areas with large height variations located in both the Northern and Southern hemispheres and involving the three IW sub-swaths. The local heading angle error is below 0.7° with a mean of 0.53° and a standard deviation below 0.01°.

Finally, the tool performs a displacement time series analysis based on automated procedures recently developed for identifying CTs with nonlinear signals and based on fuzzy entropy and Fisher statistics (Bovenga et al., 2022). This allows end users to focus their investigations on a smaller set of CTs affected by nonlinear displacements (including warning signals) and potentially deserving sophisticated geophysical or geotechnical analysis.

The work introduces the methodologies and functionalities of the tool and provides examples of its application on DInSAR/MTInSAR products derived by processing Sentinel-1 data over mountainous areas, where slow mass movements with diverse mechanisms and with different deformation rates and patterns occur.

References

Bovenga, F., Argentiero, I., Refice, A., Nutricato, R., Nitti, D.O., Pasquariello, G., Spilotro, G., 2022. Assessing the Potential of Long, Multi-Temporal SAR Interferometry Time Series for Slope Instability Monitoring: Two Case Studies in Southern Italy. Remote Sensing, 2022, 14(7): 1677. DOI: 10.3390/rs14071677.

Bovenga, F., Argentiero, I.. Belmonte, A., Refice, A., Cuozzo, G., Heredia, M. S., Callegari, M., Notarnicola, C., Nutricato, R., Nitti, D.O., 2023. Assessing Rock Glacier Activity In Val Senales By Exploiting Multiband SAR Data Through Differential SAR Interferometry And Offset Tracking in 12th International Workshop on "Advances in the Science and Applications of SAR Interferometry and Sentinel-1 InSAR" - FRINGE 2023, University of Leeds, Leeds, UK, 11-15/09/2023

Bovenga F., 2024. Cloud-based and On-premises Tools for Earth Observation Data Processing in Disaster Management Activities. 2024 IEEE International Humanitarian Technologies Conference (IHTC), Bari, Italy, 2024: 1–7. DOI: 10.1109/IHTC61819.2024.10855063.

Bovenga, F., Piccolino, F., 2025. InSAR Product Analysis (IPA): a QGIS tool for slope instability assessment based on SAR interferometry. EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17224. DOI: 10.5194/egusphere-egu25-17224.

Dixon, P. M., 2002. Ripley’s K-function. In Encyclopedia of Environmetrics, 1796–1803. Wiley 2024, 1-7. DOI: 10.1109/IHTC61819.2024.10855063.

Wasowski, J., Bovenga, F., 2014. Investigating landslides and unstable slopes with satellite Multi Temporal Interferometry: Current issues and future perspectives. Engineering Geology 174: 103–138. DOI: 10.1016/j.enggeo.2014.03.003.

Acknowledgment

This work was supported in part by Regione Puglia (Italy) under project “Utilizzo di intelligenza artificiale e dati satellitari per il monitoraggio dell’instabilità del territorio”, POC PUGLIA FESR-FSE 2014 / 2020 - Programma Regionale RIPARTI, grant agreement 01975b92; and in part by the European Union - Next Generation EU, Mission 4, Component 2, CUP H53D23001660006 (PRIN22 Project "MIRAGE: Mass movement Investigation and prediction through geomorphology, Remote sensing and Artificial intelligence").

The proliferation of massive InSAR processing services, such as the European Ground Motion Service (EGMS), has provided an unprecedented volume of data regarding Earth’s surface dynamics. However, translating these complex displacement time-series into actionable insights remains a significant challenge for non-expert end-users in critical sectors. This communication presents the MoMo (Motion Modeling) service, an advanced post-processing framework designed to bridge the gap between Interferometric SAR outputs and decision-making in civil engineering and urban management.

At the core of MoMo is a robust statistical strategy based on Multiple Hypothesis Testing (MHT). Following the Occam’s razor principle and the B-method for statistical testing, the algorithm automatically identifies the most parsimonious model that explains the displacement behavior of each measurement point. This approach enables the precise detection of break-points and seasonally driven deformation patterns through the integration of exogenous variables (e.g., temperature, groundwater levels, or precipitation), allowing the underlying drivers to be characterized with millimetric accuracy.

The scientific value of the MHT strategy lies in its ability to filter noise and provide a reliable physical interpretation of the data. However, its true impact is demonstrated in its application to large-scale infrastructure management. We present case studies showcasing the monitoring of terrestrial transport networks, large-scale port facilities, and dam stability. Special emphasis is placed on the detection of "hotspots" in entire cities, where the tool transforms large volumes of EGMS-like data points into prioritized risk maps.

The core innovation of MoMo lies in the democratization of advanced analytics for non-InSAR expert domains such as infrastructure management and urban planning. The service transforms millimeter-level displacement time series into actionable causality for asset managers, enabling them to move beyond “dots on a map” toward a clear understanding of deformation drivers. By systematically integrating climatic and environmental variables (such as temperature and precipitation), MoMo allows users to distinguish between seasonal structural responses and genuine geotechnical hazards. Furthermore, its intuitive detection of break-points and trend shifts provides infrastructure managers with precise temporal markers to assess the impact of extreme events (e.g., major storms or flooding episodes) on long-term structural behavior. By operationalizing global datasets such as EGMS within a user-friendly analytical framework, MoMo represents a concrete step toward data-driven infrastructure resilience within a digital twin paradigm.

By automating the interpretation of InSAR time-series, MoMo facilitates the dissemination and adoption of radar interferometry in sectors that traditionally struggle with data complexity. This tool empowers stakeholders to move from reactive maintenance to proactive risk mitigation strategies.

The service is available on the DestinE Platform.

Destination Earth (DestinE) is a flagship initiative of the European Commission aimed at developing a high-precision digital representation of the Earth system. By combining large volumes of environmental data with advanced modelling capabilities, DestinE supports the monitoring and simulation of natural and human-driven processes, helping policymakers, researchers, and stakeholders make more informed decisions for climate adaptation and disaster risk management.

Destination Earth (DestinE) is a European Union funded initiative, with the aim to build a digital replica of the Earth system. The initiative is being jointly implemented by three entrusted entities: the European Space Agency (ESA), responsible for building the ‘Core Service Platform’, the European Centre for Medium-Range Weather Forecasts (ECMWF), responsible for the creation of the first two ‘digital twins’ and the ‘Digital Twin Engine’, and the European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT), responsible for the creation of the ‘Data Lake’. Access Destination Earth at destine.platform.eu.

Towards a national-scale high-resolution InSAR Ground Motion Service for the Netherlands

Suresh Krishnan Palanisamy Vadivel¹, Manon Verberne¹, Ece Ozer², Thibault Candela¹, Kay Koster¹, Megan Wouters¹

¹TNO Geological Survey of the Netherlands

² TNO, Department of Reliable Structures

Abstract

Ground motion in the Netherlands, including subsidence, is often human induced, resulting from salt and hydrocarbon extraction, groundwater management, anthropogenic loading, and post-mining effects. Existing services like the European Ground Motion Service provides continental-scale products, however, higher-resolution InSAR data are required for the Netherlands-specific applications. This project develops a national, high-resolution ground motion service using RADARSAT-2 XF SAR data. TNO Geological Survey of the Netherlands implements an operational workflow combining persistent and distributed scatterer time-series InSAR to generate Level 2a line-of-sight (LOS) displacements. Products are calibrated with the GNSS-based A-EPND model to produce Level 2b data referenced to ETRF2014, ensuring geodetic consistency. A rigorous validation framework assesses the accuracy of deformation estimates, geolocation, and height estimates against in-situ observations. An interactive viewer integrates subsurface geological information to support interpretation. Initial results for South Limburg demonstrate the capability to provide insights into post-mining deformation and support the development of a nationwide service.

1. Introduction

Subsidence in the Netherlands is mainly driven by human-induced processes, such as progressive lowering of groundwater levels leading to peat oxidation and soil consolidation, as well as the extraction of salt and hydrocarbon, resulting in cavern and reservoir compaction [1]. Reliable deformation monitoring is essential for infrastructure management, hazard assessment, and spatial planning. Satellite-based InSAR enables millimetre-scale detection of surface displacement over large areas. When integrated with GNSS observations, it provides a robust framework for geodetic referencing and validation. Although the European Ground Motion Service (EGMS) [2] delivers continental-scale products, Dutch stakeholders often require higher resolution products, especially when monitoring relatively small assets such as roads, railways, or dikes. To overcome this, this project aims to establish a national-scale, high-resolution ground motion service using RADARSAT-2 XF SAR data, delivering calibrated, validated, and publicly accessible deformation products integrated with subsurface information.

2. Objectives and Services Concept

The primary objective of our national-scale InSAR ground motion service is to deliver publicly accessible, high-resolution ground deformation data using RADARSAT-2 XF SAR images. The scope includes nationwide coverage of the whole Netherlands and aims to significantly increase the density of persistent and distributed scatterers compared to medium-resolution Sentinel-1 data, enabling asset-level deformation assessment. TNO Geological Survey of the Netherlands (GDN) will implement an optimum strategy that integrate targeted services such as national-scale InSAR products and subsurface knowledge into a single platform.

3. Data and Processing Methodology

3.1. Data and Methods

Netherlands ground motion products are generated from C-band SAR data acquired by RADARSAT-2 XF mode, selected for its long-term mission continuity from 2015 to 2025, and high-spatial resolution of 5m x 5m. Both ascending and descending tracks are processed to allow decomposition of line-of-sight (LOS) deformation into vertical and east-west components. Figure 1 illustrates the ascending and descending track geometries providing full national coverage.

The processing chain presented in Figure 2 follows a time-series Persistent Scatterer (PS) and Distributed Scatterer (DS) InSAR methodology optimized for the Netherlands terrain conditions, where low deformation gradients and rural landscapes require high sensitivity and point density. Pre-processing includes precise orbit correction, radiometric calibration, sub-pixel co-registration, interferogram generation using a small-baseline network, and removal of topographic phase contributions using a high-resolution LiDAR-based national scale digital elevation model (AHN). Candidate scatterers are identified using amplitude dispersion and temporal coherence indices for persistent targets [3] and statistical homogeneity criteria for distributed targets estimated through phase-linking using coherence matrix generated from SHPs (Statistically Homogeneous Pixels) [4-6], maximizing coverage in both urban and agricultural areas. Atmospheric phase screen effects are estimated and mitigated through spatial-temporal filtering, and 2D phase unwrapping is performed prior to least-squares inversion for time-series estimation. The resulting deformation histories include both linear velocities and non-linear displacement components. Products are referenced to stable areas and integrated with GNSS observations to ensure consistency with the Earth-centred reference frame.

3.2. Products

The Netherlands ground motion service delivers a hierarchical product structure designed to serve both expert and policy-oriented users. Level 2a (L2a) products consist of point-based deformation measurements in radar geometry, including LOS velocities, full displacement time series, coherence values and residual statistics for ascending and descending tracks separately. These products represent minimally interpreted geophysical observations and are primarily intended for advanced analysis and research. Level 2b (L2b) products are geocoded into the national coordinate system and include decomposed vertical and east-west velocity components derived from combined ascending and descending measurements, along with propagated uncertainty estimates. L2b products are optimized for GIS integration and practical application in infrastructure monitoring, urban planning, and water management. These products translate technical InSAR measurements into actionable information for policymakers and asset managers.

3.2. Validation

Given the small magnitude of typical deformation signals in the Netherlands, the Netherlands ground motion services places strong emphasis on quality assurance and uncertainty quantification. Each measurement point is assigned quality indicators including temporal coherence, velocity standard deviation, residual phase statistics, geolocation and height estimates. Internal consistency checks are complemented by external validation against GNSS stations, corner reflectors (CRs), levelling benchmarks, EGMS datasets, and geological models. Level 2a and 2b products are evaluated at selected test sites including South Limburg mainly for deformation estimates, height errors and geo-localisations. This multi-layered quality control framework provides confidence in millimetre-scale velocity estimates and supports responsible interpretation in engineering and policy contexts.

3.3. Data Dissemination and viewer

The Netherlands ground motion service promotes open data access and societal awareness through a publicly accessible web-based viewer designed for both expert and non-specialist users. The viewer allows visualization of velocity maps, inspection of individual deformation time series, and comparison of ascending and descending tracks. Users can query point attributes and download datasets. By providing intuitive visualization alongside transparent uncertainty information, the viewer enhances accessibility and fosters trust in satellite-based deformation monitoring. Figure 3 shows the interactive InSAR viewer displaying Level 2a LOS displacement results for the test site.

4. Case Study: South Limburg

The South Limburg area, located in the southeasternmost part of the Netherlands, is characterized by structurally complex geology, past underground coal mining, and heterogeneous urban and rural land use. The region experienced historical subsidence related to past mining activities and deformation linked to anthropogenic loading and post-mining effects such as natural recovery of groundwater levels. A total of 120 RADARSAT-2 XF single-look complex (SLC) images covering the period 2015–2024 were collected for the South Limburg area. The dataset comprises ascending acquisitions from 11 June 2015 to 24 April 2024 and descending acquisitions from 4 December 2016 to 14 April 2024. The PSDS InSAR processing algorithm mentioned in Section 3.1 was applied to the test site to evaluate the performance, spatial consistency and sensitivity of the Level 2a products at regional scale. Within this test site, Level 2a LOS velocities derived from the processing workflow were analysed to assess spatial and temporal deformation patterns.

External validation was performed through systematic comparison with continuous GNSS observations available within the South Limburg area. The InSAR-GNSS comparison was conducted using a double-difference approach to reduce reference frame inconsistencies [7]. GNSS data for the stations SEL2, VOER, MSTR, and AACH were obtained from Nevada geodetic Laboratory [8]. GNSS observations were projected from ENU to LOS to assess agreement in InSAR LOS displacements. Figure 5 presents the agreement between InSAR and GNSS LOS displacement time-series.

5. Outlook and National Implementation

Future work focuses on the national implementation of high-resolution InSAR data production and dissemination using RADARSAT-2 XF data for operational ground motion services in the Netherlands. The next steps include the developments on nationwide processing, integration with the GNSS reference frame, and automated quality control to ensure millimetre-scale reliability. This service framework is designed for seamless dissemination of high-resolution InSAR data from 2015-2025 to public and stakeholders with annual updates until 2028.

References

1. Fokker, P. A. et al. (2025). Subsidence. In: Ten Veen, J.H., Vis, G.-J., De Jager, J. & Wong, Th.E. (eds): Geology of the Netherlands, second edition. Amsterdam University Press (Amsterdam), pp. 825-847. DOI: 10.5117/9789463728362_ch23 .
2. Costantini, M., Minati, F., Trillo, F., Ferretti, A., Novali, F., Passera, E., ... & Andersen, H. S. (2021). European ground motion service (EGMS). 2021 IEEE international geoscience and remote sensing symposium IGARSS, IEEE, 3293-3296. DOI: 10.1109/igarss47720.2021.9553562.
3. Ferretti, C. Prati and F. Rocca, "Permanent scatterers in SAR interferometry," in IEEE Transactions on Geoscience and Remote Sensing, vol. 39, no. 1, pp. 8-20, Jan. 2001, https://doi:10.1109/36.898661.
4. Shirzaei, M. (2012). A wavelet-based multitemporal DInSAR algorithm for monitoring ground surface motion. IEEE Geoscience and Remote Sensing Letters, 10(3), 456-460, DOI: 10.1109/LGRS.2012.2208935.
5. Ansari, H., De Zan, F., & Bamler, R. (2018). Efficient phase estimation for interferogram stacks. IEEE Transactions on Geoscience and Remote Sensing, 56(7), 4109-4125, https://doi:10.1109/TGRS.2018.2826045.
6. Wang, C. et al., (2022). A new likelihood function for consistent phase series estimation in distributed scatterer interferometry. IEEE Transactions on Geoscience and Remote Sensing, 60, 1-14, https://doi:10.1109/TGRS.2022.3170567.
7. Hanssen, R. F. (2001). Radar interferometry: data interpretation and error analysis. Dordrecht: Springer Netherlands.
8. Blewitt, G., Hammond, W., & Kreemer, C. (2018). Harnessing the GPS data explosion for interdisciplinary science. Eos, 99(2), https://doi.org/10.1029/2018EO104623.

GDM-SAR-In (Ground Deformation Monitoring from SAR data using InSAR) is one of the on-demand services offered by the French research infrastructure Data Terra, developed and operated by its solid Earth hub, FormaTerre, in collaboration with ISTerre / OSUG , IPGP, CNES and the National Observation Service ISDeform of INSU / CNRS. This service has been opened to French users in June 2024 and is dedicated for processing InSAR products from Sentinel-1 radar imagery. A wider opening of the service with a quota dedicated to EPOS users is planned by the end of 2026. The service web page https://en.poleterresolide.fr/gdm-sar-in-service/ gives information about the access and the use of the services.

The service is deployed on the CNES computing center. Based on the NSBAS processing chain using a small baseline approach, GDM-SAR-In allows an automated computation of single interferogram or a network of interferograms with its associated unwrapped phase time series giving access to measurement of ground deformations worldwide and with a revisit time down up to 6 days. This service allows non-expert users to run processing with simple option choices without having to worry about setting up and maintaining a complex processing chain (including downloading Sentinel-1 images, precise orbit data, digital terrain model and atmospheric model data, and the demanding TOPSAR mode processing of Sentinel-1 acquisitions) on a computing cluster.

It also offers expert users a simple and fast way to explore a new area or a specific phenomenon such as a volcanic or seismic crisis, while keeping a certain flexibility in the choice of processing parameters. The availability of intermediate products and processing information allow expert users to reprocess, by their own, parts of the processing if necessary.

Users access the service through a web interface specifically designed for radar interferometry usage. The interface allows the user to interactively choose the study area and the Sentinel-1 data suitable for InSAR processing and to follow the progress of the processing. The generated products are available for download for a limited period of time (a few weeks). A preview of the products is possible directly on the interface.

Most of the products are provided in both radar and ground geometry (in geotiff format), interferograms are available in different versions (wrapped/unwrapped, filtered/unfiltered, with/without atmospheric correction from global model) allowing for user-customized post-processing. A time series of the unwrapped phase can also be generated as well as many other auxiliary products allowing advanced analysis of ground displacements by the user. Products are compatible with the catalog and data formats of FormaTerre and of the Thematic Core Service Satellite Data of the European research infrastructure EPOS. Examples of applications and products are available from the service website covering multidisciplinary applications (e.g. volcanic eruptions, earthquakes, landslides, forest fire, ...). These examples are showing how the service can be used as a function of the scientific goal, taking into account the potential limits of an automatic InSAR processing, and they aim to encourage the use of the service in a broad range of applications.

The systematic monitoring of surface deformation, infrastructure stability, cryospheric dynamics, and tectonic activity increasingly relies on large-scale interferometric processing of Synthetic Aperture Radar (SAR) data. The Copernicus Sentinel-1 mission provides a global, high-revisit C-band archive that has become the backbone of operational InSAR services. However, transforming this continuously growing archive into actionable information requires more than algorithmic expertise. It demands scalable cloud infrastructure, efficient data formats, interoperable metadata standards, and AI-ready data representations.

This contribution presents the capabilities of CREODIAS as an EO-cloud environment designed to support large-scale InSAR processing and advanced SAR analytics. The platform integrates co-located object storage and elastic compute resources, enabling direct access to the full Sentinel-1 archive without the need for local data replication. This architecture significantly reduces data transfer overhead and supports parallelized interferometric processing chains at regional to continental scales.

A key component of scalable InSAR processing is burst-level handling of Sentinel-1 TOPS acquisitions. The platform supports extraction and indexing of individual bursts, enabling precise stack construction and optimized interferogram generation. This burst-oriented strategy reduces computational redundancy and facilitates localized deformation monitoring with improved geometric consistency. Containerized processing environments allow integration of standard and custom InSAR toolchains, ensuring reproducibility and flexibility across scientific and operational applications.

Beyond raw and interferometric products, the generation of Analysis-Ready Data (ARD), including Radiometrically Terrain Corrected (RTC) backscatter and burst-aligned stacks, plays a central role in enabling downstream analytics. To support high-performance time-series analysis, datasets can be transformed into cloud-optimized formats such as chunked Zarr-based data cubes. These multidimensional structures allow efficient parallel access to spatio-temporal subsets, making them particularly suitable for distributed computing environments and machine learning workflows.

Interoperability and discoverability are ensured through implementation of the SpatioTemporal Asset Catalog (STAC) standard. STAC-based indexing provides structured metadata for SAR-specific attributes such as polarization, orbit direction, acquisition geometry, and burst identifiers. Importantly, the catalog does not only reference raw Sentinel-1 products but also derived assets, including interferograms, deformation maps, and machine learning outputs. This approach guarantees compliance with FAIR principles and facilitates integration with widely used geospatial libraries and APIs.

The platform further supports AI-ready data representations through large-scale feature extraction and embedding generation from SAR backscatter and time-series products. Self-supervised learning approaches enable the derivation of compact numerical embeddings that capture spatial and temporal radar signal characteristics without reliance on extensive labeled datasets. These representations can be used for similarity search, clustering, and anomaly detection in deformation fields or land surface dynamics. Embedding-based analytics complement classical InSAR pipelines by enabling data-driven identification of unusual spatio-temporal patterns, thereby enhancing early warning and monitoring capabilities.

The integration of scalable compute resources, cloud-native storage, standardized metadata, and machine learning workflows demonstrates that modern InSAR services are fundamentally infrastructure-driven. Processing at global volume requires not only efficient algorithms but also carefully designed data architectures that ensure accessibility, interoperability, and reproducibility.

In summary, CREODIAS provides a comprehensive EO-cloud ecosystem for burst-level InSAR processing, SAR data cube generation, STAC-compliant cataloging, and AI-enhanced analytics. Such integrated environments represent a critical step toward operational, scalable, and intelligent Earth Observation services capable of supporting scientific research, environmental monitoring, and decision-making processes at continental and global scales.

In recent years, the number of Synthetic Aperture Radar (SAR) satellites in low Earth orbit has increased significantly, largely due to the reduction in launch costs for commercial and scientific payloads. In particular, commercial X-Band imagery acquired by ICEYE, Capella space and Umbra space provide very high resolution (VHR) and relatively short revisit times, thereby offering unique opportunities when compared to traditional open-access missions. To achieve such revisit over an area, the satellites illuminate the target from highly varying geometries (ascending and descending orbits, with right or left look) and incidence angles.

Here we present a DEM reconstruction pipeline applied to pairs of VHR SAR based on an existing open-source stereo reconstruction tool developed for optical satellite imagery, i.e., the Ames Stereo Pipeline developed by NASA AMES [1]. The Rational Polynomial Coefficients (RPCs) are used to model the sensor geometry. Two images are then map-projected, after tie points are detected and matched. Epipolar rectification is performed using the RPC model, and dense stereo matching is applied to generate a disparity map. The disparity field is then converted into a 3D point cloud and interpolated to produce a DEM.

We tested our approach on challenging alpine terrain in the Aletsch Glacier region (Switzerland), characterised by strong relief with elevation ranging from ~700 to over 4’000 m a.s.l. We considered ICEYE Extended Spotlight datasets: one acquired between August and September 2022 (32 acquisitions), and a second acquired between 22 and 25 March 2024 (7 acquisitions).

The most challenging task for the AMES tool has been finding robust image matching across acquisitions with differing viewing geometries. The best results were obtained using opposite-orbit, same side-looking acquisitions (e.g. ascending-left combined with descending-right acquisitions). These geometries entail long perpendicular baselines (up to ~100 km), which lead to high convergence angle (over 20°), without compromising the matching capabilities.

Resulting DEM co-registration and error propagation analyses were performed using xDEM, following established methodologies [3]. Elevation differences were evaluated over stable, non-glacier terrain against the Swiss national lidar-derived DEM, swissALTI3D. For one pair (descending-right acquisition from 15 August 2022 and ascending-left acquisition from 5 September 2022), we obtained a mean vertical error of 1.72 m over stable terrain.
Ongoing refinements aim to better adapt the pipeline to SAR image characteristics. This will open the opportunity of creating multiple DEMs over the same target area for surface change monitoring, as well as averaging multiple DEMs to further improve elevation accuracy.

[1] Shean, D. E. et al. An automated, open-source pipeline for mass production of digital elevation models (DEMs) from very-high-resolution commercial stereo satellite imagery. ISPRS Journal of Photogrammetry and Remote Sensing 116, 101–117 (2016).
[2] xDEM contributors. (2024). xDEM (v0.1.0). Zenodo. https://doi.org/10.5281/zenodo.11492983
[3] Hugonnet, R. et al. Uncertainty Analysis of Digital Elevation Models by Spatial Inference From Stable Terrain. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 15, 6456–6472 (2022).

Insarviz is a tool designed to visualize and interactively explore the spatiotemporal datacubes derived from InSAR data processing chains.
For several reasons (data availability, non-intuitive radar image geometry, complexity of the processing, etc.), InSAR has long remained a niche technology and few free open-source tools have been dedicated to it compared to the widely-used, multi-purpose optical imagery. Most existing tools are focused on data processing. Generic remote-sensing or Geographic Information System (GIS) softwares are limited when used to visualize InSAR data, due to their unusual geometry and formats. A few visualization tools with dedicated InSAR functionalities exist, that were designed to visualize a single radar image or interferogram.
However, recent spatial missions, like the Sentinel-1 mission of the European program COPERNICUS, with a systematic background acquisition strategy and an open data policy, provide unprecedented access to massive SAR datasets. From these new datasets, a network of thousands of interferograms can be generated over a single area. The consecutive step is a time-series analysis which produces a spatiotemporal data cube: a layer of this data cube is a 2D map that contains the displacement of each pixel of an image relative to the same pixel in the reference date image. A typical data cube size is 4000x6000x200, where 4000x6000 are the spatial dimensions (pixels) and 200 is a typical number of images taken since the beginning of the spatial mission.
The aforementioned tools are not suited to manage such large and multifaceted datasets. In particular, fluid and interactive data visualization of large, multidimensional datasets is non-trivial. If data cube visualization is a more generic problem and an active research topic in EO and beyond, some specifics of InSAR (radar geometry, wrapped phase, relative measurement in space and in time, multiple types of products needed for interpretation…) call for a new, dedicated visualization tool.
To that end, InsarViz aims to provide a user-friendly interface for visualization and analysis of InSAR products, with in the current version a particular focus on large data cubes visualisation. It boasts an array of useful features, such as :

• handling of most usual data formats (Tiff, R4, HDF, depl_cumul, ...) thanks to the ever-useful GDAL library
• handling of large multi-band data sets (> 10G) with ease
• easily combining multiple layers onto a data cube (e.g. a mean-velocity layer, or a DEM)
• precisely managing each layer's look through interactive histograms and a variety of color maps
• loading data in radar or ground coordinates, showing background maps when possible
• loading a LUT (lookup table) to visualize projections of radar data onto the ground
• interactively defining features (points and profiles) on the map, and plotting spatial and temporal measurements from those features
• real-time configurable model fitting for temporal plots, handling seasonal and semi-seasonal deformation, and post-seismic (logarithmic) relaxation

Online documentation : https://deformvis.gricad-pages.univ-grenoble-alpes.fr/insarviz/latest/index.html
Quick demo (video) : https://formater.osug.fr/demos/insarViz/insarViz-demo.mp4
We will present the latest InsarViz version, and perform a live demo on a large data-cube over the region of Mexico. Afterwards, we will discuss further advancements into InSAR visualization, notably featuring interactive visualisation of interferogram networks, and radar-geo correspondence.

1 in 6 properties in England are at risk of flooding from rivers, sea and surface water. Recent UK-wide flood events following the succession of Storms Claudia, Bram, Goretti and Chandra from November 2025 to January 2026 highlight the current threat from flooding. Flood risk is only set to increase in the future as a result of climate change driven sea-level rise and increases in the frequency and magnitude of rainfall events.

Lack of information about current and historical floods was highlighted as a major limitation to improving flood resilience and evaluating warnings by flood responders and forecasters at a workshop organised by the Environment Agency (EA) in March 2023. Traditional methods of flood mapping, such as aerial surveillance and visual, in-situ observations, are often cost and time-intensive, unacceptably dangerous and may not provide a consistent record of past floods.

Earth observation satellites offer an alternative method of monitoring floods and are uniquely capable of systematically observing large geographical areas at a resolution of 20 m or less. The Copernicus Global Flood Monitoring (GFM) program was developed in 2021 and provides a near real-time service by automatically processing new Sentinel-1 acquisitions. However, the approach has been generalised to meet the demands of global monitoring and provides inconsistent analysis products over the UK. Meanwhile, products using commercial satellites are only capable of observing small regions on a case-by-case basis and would require considerable funding to provide an ongoing service across the UK.

Through a collaboration with the Met Office and the University of Leeds, SatSense has designed an algorithm (FloodMap) to meet the needs of a cost-effective UK-wide flood monitoring service. FloodMap combines open-access Sentinel-1 data with UK-wide EA LiDAR data in a probabilistic time-series approach. The result is a product that accounts for uncertainties inherent to SAR data while significantly reducing the number of misclassifications from such noise through the application of physically realistic constraints on the flood surface. We demonstrate this capability by producing flood maps for three different case studies of recent large-scale UK flooding. We also perform a validation against independent observations from the Sentinel-2 optical satellite and compare this to GFM to demonstrate the improved performance of FloodMap.

The safety and resilience of urban infrastructure are of great concern in the context of rapid urbanization. Human activities, such as construction, excavation, and groundwater extraction, may cause ground subsidence and threaten infrastructure, including buildings and roads. Interferometric Synthetic Aperture Radar (InSAR) is an advanced SAR technique used to identify and measure surface deformation with sub-centimeter precision.

Recent advances in InSAR techniques, the availability of long time series of SAR data, and cloud-based processing facilitate long-term monitoring of surface deformation within urban environments. However, the use of InSAR in infrastructure safety assessment presents challenges and requires careful data processing, analysis, and interpretation due to heterogeneity of urban areas. For example, tall buildings may cause geometric distortion and variable coherence over time, which can complicate time series inversion and interpretation.

Within the project Actionable Data Space for Urban Climate Adaptation and related socio-ecological, local Transformation (ADUCAT), we aim to develop an automated pipeline tailored to urban applications to enable stakeholders on demand InSAR processing using freely available Sentinel-1 time series data. The development of the methodology is twofold: (1) automatization of surface deformation measurements, including (semi-)automated SAR data query and download, interferogram stack generation, surface deformation and velocity estimation in vertical and horizontal directions, and (2) localization of surface deformation based on infrastructure types, such as buildings and roads, by including coherence information and high-resolution digital elevation model data. Precise in-situ measurements and user feedback will be used to assess the accuracy and applicability of the results and the proposed pipeline.

The proposed pipeline supports on-demand InSAR analysis using existing SAR data in an open-source environment. Tailored for urban areas, the pipeline facilitates systematic deformation monitoring and facilitates early warning by detecting and mapping areas exhibiting surface displacement, which may indicate areas susceptible to structural damage and infrastructure instability.

Persistent Scatterer Interferometry (PSI) is a remote sensing technique well suited for regularly monitoring ground surface deformation. The launch of the Sentinel‑1 (S1) satellites has provided a continuous stream of SAR images, prompting a shift in recent PSI developments from analyzing fixed time periods to continuously updating deformation datasets as new data become available. However, frequently processing an ever‑growing dataset raises questions regarding persistent scatterer (PS) density, processing efficiency, and the comparability of results across successive updates.

In previous studies, we evaluated three strategies for processing a steady stream of SAR images. The first strategy involved processing all available SAR images at each update, resulting in a continuously expanding dataset. The second strategy processed the incoming SAR images in consecutive, non‑overlapping subsets. The third strategy used overlapping consecutive subsets, similar to a sliding window. Comparing the results of these three strategies—specifically in terms of PS density, processing efficiency, and the consistency of results across updates—showed that the third strategy is best suited for long‑term monitoring of specific objects or areas of interest.

To fully exploit the potential for reusing intermediate results within this third strategy, we designed and implemented the PSISlider processing chain. Its main deviation from a conventional PSI workflow is the use of two different reference images: one for co‑registration and another for interferogram formation. All secondary images are co‑registered to the same reference image, regardless of which image is used as the local reference for interferogram formation. The reference image used for co‑registration is termed the co‑registration reference image, while the reference image used for interferogram formation is termed the local reference image.

Using two different reference images has several implications:

1. For new subsets, only the newly acquired SAR images require interferometric pre‑processing.
2. Spatial baselines between secondary images and the local reference image must be calculated.
3. Differential interferograms between secondary images and the local reference image must be generated.

However, employing a consistent co-registration reference image across all subsets enables the reuse of several intermediate products from previous processing steps, including:

1. The amplitude dispersion index for selecting PS candidates (PSC).
2. The probability of a PSC being a PS, for PS selection.
3. The spatially uncorrelated look‑angle error.

In this presentation, we outline the PSISlider processing chain and demonstrate its application using a case study from the coastal city of Patras, Greece.

Persistent Scatterer Interferometry (PSI) is widely used for long-term deformation monitoring with Sentinel-1 time-series data. In operational PSI processing, phase stability forms the basis of displacement estimation, whereas intensity images are primarily used for visualization and scatterer selection. Nevertheless, intensity products remain essential for interpreting structural geometry and contextualizing persistent scatterer distributions, particularly in infrastructure monitoring applications such as bridge analysis.

This study presents a practical integration of conventional intensity-domain super-resolution into PSI workflows. The objective is to enhance the spatial interpretability of Sentinel-1 intensity images without modifying the interferometric processing chain. The proposed framework operates exclusively in the intensity domain and does not alter interferograms, coherence estimation, or displacement retrieval. Consequently, deformation analysis remains strictly unaffected, and the enhancement can be introduced as an optional module within existing PSI environments.

The super-resolution method applies a conventional multi-frame super-resolution approach based on Iterative Back-Projection (IBP) to log-intensity images derived from Sentinel-1 IW-mode SLC amplitudes. Multiple co-registered acquisitions from the same relative orbit are used to exploit natural sub-pixel spatial diversity arising from slight orbital variations and sampling offsets. A key aspect of the integration is the use of sub-pixel co-registration parameters estimated during standard PSI interferometric processing. Rather than performing independent alignment, the IBP reconstruction relies on the same geometric solution used for interferogram generation. This shared co-registration strategy ensures strict geometric consistency between the enhanced intensity products and the interferometric stack while avoiding redundant processing and additional alignment uncertainties.

Reconstruction follows a conventional IBP scheme assuming a Gaussian point spread function as an approximation of the Sentinel-1 IW-mode impulse response. Starting from an upsampled estimate, simulated low-resolution observations are generated and iteratively compared with measured intensity images, with residuals back-projected to refine the solution. Processing in the log-intensity domain converts multiplicative speckle into an additive component, thereby improving statistical stability. The procedure is computationally lightweight and fully model-based, requiring no training data or sensor-specific tuning beyond the assumed point spread function.

Although recent SAR super-resolution studies increasingly employ data-driven or deep learning approaches, the present framework deliberately adopts a transparent and reproducible conventional method. For infrastructure monitoring within operational InSAR contexts, methodological interpretability and deterministic behavior remain essential, particularly when analysis results may support engineering decisions.

The integration was evaluated using Sentinel-1 IW-mode SLC data acquired over coastal built environments containing bridge structures and adjacent water surfaces. These scenes are characterized by strong amplitude contrasts, double-bounce scattering, and sparse but dominant persistent scatterers aligned with structural elements. The IBP-enhanced intensity products demonstrate improved continuity of linear scatterers, clearer delineation of structural edges, and enhanced geometric readability compared to conventional multi-look intensity images. Importantly, interferograms and PSI-derived deformation time series remain unchanged, confirming that the enhancement preserves interferometric integrity.

The method does not eliminate speckle and cannot exceed intrinsic system bandwidth limits; excessive iteration may amplify noise in low-signal regions. However, the objective is not to generate artificial detail but to exploit existing spatial diversity in a geometrically consistent manner. By maintaining strict alignment with the PSI co-registration framework, the enhancement supports improved interpretation of infrastructure geometry without altering deformation estimation.

In conclusion, conventional IBP-based multi-frame super-resolution can be structurally integrated into PSI workflows through direct use of interferometric co-registration parameters. By operating solely in the intensity domain and preserving phase-based processing integrity, the approach provides a transparent and operationally compatible enhancement mechanism. The results indicate that physically interpretable super-resolution techniques can improve the spatial clarity of infrastructure-related scatterers in Sentinel-1 imagery while remaining fully consistent with PSI-based deformation monitoring.

Interferometric Synthetic Aperture Radar (InSAR) is a powerful remote sensing technique used for high-resolution topographic mapping and surface deformation monitoring. However, the quality of InSAR measurements is often degraded by interferometric decorrelation, which leads to gaps and loss of phase information in the resulting interferograms. Decorrelation can be caused by various factors, including dense vegetation, significant changes in land cover over time, and large deformation gradients, such as those associated with earthquakes. The presence of these decorrelated areas poses a significant challenge to the subsequent analysis of interferograms, as the continuity of the wrapped phase field is a fundamental prerequisite for downstream processing steps like phase unwrapping and inversion. Therefore, the ability to accurately reconstruct the InSAR phase in these decorrelated regions is of paramount importance for a wide range of geophysical applications, including the monitoring of ground subsidence, the analysis of seismic and volcanic deformation, and the generation of accurate digital elevation models. The paper introduces a novel two-stage adversarial joint learning model designed to address the challenge of reconstructing the InSAR phase in decorrelated areas. The proposed methodology is based on a Generative Adversarial Network (GAN) framework and is divided into two main stages: an Edge Mapping Stage (EMS) and a Phase Predicting Stage (PPS). The Edge Mapping Stage employs an Edge Connector Network (ECN), which is a gradient-based edge detector built upon the VGG19 model. The primary function of the ECN is to identify and reconstruct the fringe lines, which represent phase discontinuities, within the interferogram. The ECN is an encoder-decoder network that utilizes dilated convolutions and is trained using a combination of adversarial and feature-matching losses. The Phase Predicting Stage then utilizes a Phase Predictor Network (PPN) to predict the phase values within the decorrelated regions. The PPN, which shares the same generator and discriminator architecture as the ECN, uses the reconstructed fringe lines from the EMS as a guide for the phase prediction. This stage employs an image-to-image translation technique and adversarial self-supervised learning. The training of the PPN is performed using a composite edge map that combines the background region edges with the generated edges from the ECN. The training of the entire model is conducted in two phases: individual stage training followed by joint learning. The joint learning phase utilizes a patch discriminator with both pixel-space and feature-space losses to ensure a more robust and accurate reconstruction. The model was trained on a large dataset of 100,000 simulated interferograms, which were generated using the forward path of the Okada model and included various noise sources to mimic real-world conditions. The proposed two-stage adversarial joint learning model demonstrates a significant improvement in the reconstruction of InSAR phase in decorrelated areas. The model's performance was evaluated using a variety of metrics, including accuracy, precision, recall, Mean Absolute Error (MAE), Mean Squared Error (MSE), Structural Similarity Index (SSIM), and Peak Signal-to-Noise Ratio (PSNR). In the Edge Mapping Stage, the VGG-based method significantly outperformed the traditional Canny edge detector, achieving improvements of 0.8%, 3.5%, and 3.4% in accuracy, precision, and recall, respectively. The joint learning of the ECN and PPN led to substantial improvements in the phase prediction, with MAE and MSE improving by as much as 700% and 852.1%, respectively, in noisy interferograms. The overall performance of the model in fringe reconstruction achieved an accuracy of 84% and an SSIM of 96%. The model's effectiveness was further validated on real-world data from two seismic events: the M 6.5 Tonopah, Nevada earthquake of May 15, 2020, and the M 6.3 Western Xizang earthquake of July 22, 2020. The model successfully reconstructed the co-seismic deformation interferograms in these challenging cases, demonstrating its robustness and practical applicability. The cross-correlation between the reconstructed and original interferograms in the Greater Bay Area (GBA) dataset ranged from 0.72 to 0.87, further highlighting the model's ability to produce high-quality reconstructions. In conclusion, the novel two-stage adversarial joint learning model presented in this paper offers a powerful and effective solution for reconstructing the InSAR phase in decorrelated areas. The model's superior performance, validated on both simulated and real-world data, underscores its potential to significantly enhance the utility of InSAR for a wide range of scientific and engineering applications.

The European Ground Motion Service (EGMS) provides continental-scale Persistent Scatterer Interferometry (PSI) time series derived from Sentinel-1 acquisitions, representing one of the most comprehensive operational ground-motion datasets currently available. EGMS distributes two Level 2 products: a basic product (L2a) and a calibrated product (L2b), which are referenced to a deformation model derived from GNSS data. However, the L2b correction exhibits step-like temporal discontinuities that, if left unaddressed, can propagate into downstream analyses. In addition, temporal variations in the dispersion of the time series may affect the stability of multi-geometry deformation reconstruction. This study presents a systematic GNSS-assisted pipeline for post-processing EGMS Level-2 products prior to geometry decomposition and evaluates the resulting solutions against EGMS ORTHO products at locations collocated with GNSS stations.

The processing pipeline consists of three sequential calibration steps. In the first step, we extract the L2b correction by computing the pointwise difference between the L2b and L2a time series. This correction is computed step-wise, introducing small temporal discontinuities. We apply a smoothing procedure to reduce them and reconstruct from the L2a product a compatible L2b product.

In the second step, we calculate a daily dispersion diagnostic for the deformation time series at each PSI measurement point (MP). We first detrend each MP's time series to isolate short-term variability, then compute a robust estimate of the dispersion across MPs at each epoch. The resulting daily robust dispersion estimates are normalized to obtain a scaling factor that serves as an empirical temporal uncertainty proxy, providing a day-dependent σ that captures the collective variability across MPs.

In the third step, the calibrated time series from ascending and descending orbits are integrated with GNSS observations to reconstruct three-dimensional deformation. GNSS data are sourced from Nevada Geodetic Laboratory PPP solutions and independently corrected for network common-mode effects prior to integration. GNSS East, North, and Up displacements are projected into each PSI line-of-sight (LOS) direction with full three-dimensional error propagation, using its incidence and heading angles, ensuring geometry-consistent integration at the PSI level. Integration is performed through weighted least squares, combining ascending LOS, descending LOS, and GNSS constraints within a consistent uncertainty model.

The integration framework explicitly accounts for InSAR’s intrinsic limited sensitivity to the North component, which is primarily informed by GNSS, and quantifies the information loss incurred when geometry-only decomposition is applied without geodetic constraints. Solutions are computed for ascending-only, descending-only, and combined ascending-descending configurations, with and without GNSS integration, enabling a controlled assessment of each data source’s contribution to component-reconstruction stability.

Final ENU deformation solutions are compared with EGMS ORTHO products around GNSS-collocated locations. ORTHO products represent the officially distributed geometry-combined solution, providing East–West and vertical motion components with deformation velocities. Therefore, it serves as a natural benchmark for evaluating the structural consistency of the proposed pipeline. This comparison is designed to characterize how the temporal variability at the L2 stage relates to ORTHO behaviour, whether such variability propagates into geometry-combined solutions or is inherently absorbed during ORTHO generation. This provides operational users with a quantitative basis for deciding when explicit L2 calibration adds value over directly adopting the distributed combined product.

The framework is demonstrated on a regional case study and is designed to be computationally lightweight, reproducible, and transferable across EGMS coverage areas. Looking ahead, the approach is intended to scale toward high-density GNSS environments, such as local monitoring networks equipped with cost-effective Galileo-enabled receivers, where multiple closely spaced geodetic constraints can further improve component reconstruction and support applications in structural monitoring, infrastructure assessment, and local geohazard characterization.

Accurate and scalable monitoring of ground surface displacement is fundamental to understanding geophysical processes and mitigating the risks associated with natural hazards such as landslides, earthquakes, volcanic activity, and subsidence. Spaceborne and airbone Synthetic Aperture Radar (SAR) has become one of the most powerful tools for this purpose due to its all-weather, day–night imaging capability and sensitivity to surface motion. However, conventional SAR-based techniques—particularly interferometric SAR (InSAR) and amplitude-based pixel offset tracking (POT)—face persistent challenges in complex environments. These include temporal decorrelation, heterogeneous land cover, steep terrain, and rapidly evolving surface conditions, all of which degrade measurement reliability and spatial resolution. As a result, some of the most hazardous and dynamic regions remain difficult to monitor effectively.

This study introduces DeepOT, a deep learning–based offset tracking framework designed to estimate pixel-level ground surface displacement directly from SAR amplitude image pairs. The proposed approach complements traditional methods by leveraging data-driven learning to infer displacement fields without relying on explicit cross-correlation or phase-based measurements. A central innovation of DeepOT lies in its synthetic-to-real training strategy, which addresses one of the most significant bottlenecks in supervised learning for geophysical applications: the scarcity of reliable ground-truth displacement data. Instead of depending on observational labels or outputs from conventional algorithms—which may propagate biases and limitations—we embed geophysically plausible synthetic displacement fields directly into real SAR amplitude imagery. This process generates large-scale, pixel-accurate training datasets while preserving the statistical and radiometric properties of real SAR observations.

The synthetic displacement generation framework is designed to capture the diversity and complexity of real-world deformation patterns. Multiple displacement components are probabilistically combined to produce composite fields that mimic natural geophysical behavior. These include anisotropic localized deformation representing landslides or subsidence zones; fractional Brownian motion fields that reproduce spatially correlated rough displacement patterns or atmospheric artifacts; constant displacement fields simulating rigid-body motion; topography-correlated displacement reflecting terrain-driven effects; and transitional or step-like displacement patterns analogous to fault slip or shear zones. Randomized parameter sampling ensures that the training dataset spans a broad range of spatial scales, magnitudes, and deformation styles, promoting robust generalization to unseen scenarios.

DeepOT is implemented as a modular and extensible framework that supports multiple deep learning architectures, including FlowNet2, U-Net, U-Net++, and Siamese-network-based designs (CC-ResSiamNet), as well as a newly introduced deformable convolution-based model (DisplaceDCN) tailored for spatially heterogeneous displacement fields. This flexibility allows the framework to accommodate varying levels of model complexity and adapt to different application contexts. In addition, the workflow is designed with scalability in mind, incorporating efficient data management, high-throughput training pipelines, and compatibility with large-scale SAR datasets.

We evaluate the performance of DeepOT using both synthetic validation experiments and real-world case studies. Two contrasting landslide environments are used for quantitative and qualitative assessment: the Slumgullion landslide in Colorado and the Barry Arm landslide in Alaska. At Slumgullion, independent extensometer measurements provide ground-based validation of displacement time series, enabling direct comparison with model predictions. Results demonstrate that DeepOT successfully reconstructs spatially coherent displacement fields and captures temporal evolution with strong agreement to in-situ observations. Notably, the framework maintains performance in areas where InSAR coherence is low and where conventional offset tracking struggles due to surface heterogeneity or noise.

In the Barry Arm case study, characterized by complex terrain and rapid deformation, DeepOT produces continuous and detailed displacement maps that reveal fine-scale spatial patterns often obscured by traditional methods. The model effectively resolves localized deformation zones and sharp gradients without the smoothing artifacts commonly associated with window-based cross-correlation techniques. Furthermore, qualitative analyses of earthquake-induced deformation indicate that DeepOT generalizes beyond landslide scenarios, demonstrating its applicability to large-scale, high-gradient displacement fields.

Compared to classical POT approaches, DeepOT offers several key advantages. First, it eliminates the need for predefined correlation windows, enabling higher spatial resolution and improved sensitivity to localized motion. Second, it significantly reduces computational costs during inference, allowing rapid generation of displacement maps over large areas. Third, by avoiding reliance on POT-derived labels, it mitigates the risk of inheriting systematic biases and artifacts from traditional methods. These improvements collectively enhance the reliability and scalability of SAR-based displacement monitoring.

Despite its strengths, the proposed framework also highlights important considerations for future research. While synthetic training provides a powerful solution to data scarcity, the realism and diversity of simulated displacement fields remain critical factors influencing model performance. Continued refinement of synthetic generation strategies—potentially incorporating physics-based models or hybrid approaches—could further improve generalization. Additionally, integration with multi-sensor data, such as optical imagery or GNSS measurements, may enhance robustness and enable more comprehensive monitoring systems.

In summary, DeepOT represents a significant step forward in the application of artificial intelligence to geophysical remote sensing. By combining synthetic training data with advanced deep learning architectures, the framework enables direct, pixel-level estimation of ground displacement from SAR amplitude imagery in challenging environments where traditional methods often fail. Its modular design, scalability, and demonstrated performance across diverse case studies position it as a promising tool for next-generation hazard monitoring and early warning systems. The approach also establishes a broader paradigm for leveraging synthetic data to overcome fundamental limitations in Earth observation, opening new avenues for data-driven analysis of complex geophysical processes.

Reference: Kim, J.-W., Jung, H., Lu, Z., DeepOT: A Deep Learning Framework for Pixel-Level Ground Surface Displacement Estimation from SAR Amplitude Imagery. ESS Open Archive . January 16, 2026. DOI: 10.22541/essoar.176853814.47876330/v1

The continued decline in the Dead Sea water level in recent decades has resulted in sinkhole collapses along its western shores, posing a serious hazard on agricalture, industry, infrastructure, and daily life in the area. The sinkhole collapses are accompanied by gradual ground subsidence before, during, and after the sinkhole’s collapse.

The Geological Survey of Israel (GSI) monitors the sinkhole-related land subsidence using InSAR measurements every 11 days from TerraSAR-X ascending and descending tracks, and annual LiDAR measurements, regularly since 2019. The current mapping of sinkhole-related subsidence relies on manual interpretation of wrapped phase data, a time-consuming and human-error-prone procedure. Deep Learning provides an opportunity to reduce processing time, increase precision and scalability, and support real-time decision-making. Encoder–decoder–based semantic segmentation models, such as UNet, Attention UNet, SAM, TransUNet, and SegFormer have shown effectiveness in learning geospatial deformation patterns in InSAR and related remote sensing data.

We provide an evaluation of a Deep Learning UNet segmentation model applied to InSAR data for annotating land subsidence areas that occur as part of the sinkhole-formation process along the western shores of the Dead Sea. We use the manually delineated subsidence of the GSI’s operational sinkhole monitoring system between 2019 and 2023 as the ground truth in the supervised learning process. The wrapped phase data include atmospheric signals, compaction subsidence along the Dead Sea shores, decorrelation, and other signal and noise sources, as well as the sinkhole-related subsidence. This unique data poses challenges for annotation, learning, and interpretability, making the dataset both non-trivial and valuable for advancing research in applied remote sensing and its application in the Dead Sea.

The model is trained across three partition schemes–random tiles partitioning, temporal partitioning, and geospatial partitioning–each representing a different type and level of generalization. We use object-level metrics to assess the subsidence-area detection ability while accommodating human-induced annotation variability and uncertainty. Our data include 400 fully and partly annotated interferograms with ~47,000 sinkhole-related subsidence polygons.

We demonstrate the model’s ability to effectively identify and generalize subsidence areas in InSAR data across different setups and temporal conditions. The model shows promising potential for geographical generalization in previously unseen areas. Finally, we infer full interferogram subsidence areas by reconstructing smaller-scale patches and evaluate them for different confidence thresholds.

Urban building subsidence monitoring is essential for risk mitigation and asset management, particularly in Chinese megacities where numerous existing buildings and dense populations are exposed to long-term ground deformation driven by groundwater dynamics, urban loading, and intensive underground construction [1]. Interferometric Synthetic Aperture Radar (InSAR) time-series observations provide wide-area, cost-effective deformation measurements and therefore offer a practical basis for city-scale building monitoring [2-4]. Nevertheless, interpreting InSAR scatterer measurements to provide reliable building-level assessments remains challenging. First, the correspondence between InSAR scatterers and buildings is frequently ambiguous. Deterministic spatial assignment strategies based on simple geometric rules (e.g., nearest-footprint or footprint-intersection criteria) are sensitive to geocoding uncertainty, footprint boundary effects, radar layover/shadow, and environmental scatterers from roads and other urban objects. These factors can contaminate building-level deformation time series and make it difficult to quantify the reliability of building-level results. Second, urban building deformation is inherently complex, and different modes such as overall settlement, progressive tilting, differential settlement, and episodic accelerations may co-occur. As a result, a single indicator (e.g., mean velocity) can lead to both false alarms and missed detections.

To address these issues, we propose an AI-assisted urban building subsidence monitoring framework using Sentinel-1 InSAR time-series data in Chinese megacities (e.g. Shanghai, Beijing, Shenzhen, and Guangzhou). We first develop a probabilistic scatterer-to-building attribution model that integrates geometric likelihood, radar observability constraints, and temporal consistency, producing posterior probabilities between scatterers and candidate buildings, together with building-level measurability and confidence indicators. These posteriors act as attention-like weights to suppress ambiguous or non-building scatterers and to yield more reliable building observations. Based on the attributed observations, we then derive three complementary building-level deformation sequences. They are (i) a settlement time series obtained by confidence-weighted aggregation of line-of-sight deformation, (ii) a tilting time series estimated via robust plane fitting, and (iii) a differential settlement time series quantified using robust within-building dispersion measures. Finally, we introduce a multivariate deep time-series anomaly detection model that learns joint patterns across settlement, tilt, and differential-settlement sequences. It identifies buildings with abnormal evolution and outputs anomaly evidence by localizing the most contributing sequences and time windows, together with a confidence score to support risk assessment.

Preliminary experiments in Chinese megacities demonstrate the feasibility of the proposed pipeline. The probabilistic attribution step produces more coherent building-level time series than deterministic geometric assignment, particularly near footprint boundaries and in complex urban environments, and the resulting confidence indicators effectively flag buildings with insufficient measurability. Moreover, joint anomaly detection across the three building-level sequences highlights candidate buildings with consistent signatures of accelerated settlement and increasing differential deformation, providing interpretable cues for subsequent inspection. These preliminary results suggest that the proposed framework can serve as a generalizable and confidence-aware strategy for building-level InSAR monitoring and early risk identification in large metropolitan regions.

References:

[1] Drougkas A, Verstrynge E, Van Balen K, et al. 2021. Country-scale InSAR monitoring for settlement and uplift damage calculation in architectural heritage structures. Structural Health Monitoring, 20(5): 2317-2336.

[2] Ferretti A, Prati C, Rocca F. 2000. Nonlinear subsidence rate estimation using permanent scatterers in differential SAR interferometry. IEEE Transactions on Geoscience and Remote Sensing, 38(5): 2202-2212.

[3] Ferretti A, Prati C, Rocca F. 2001. Permanent scatterers in SAR interferometry. IEEE Transactions on Geoscience and Remote Sensing, 39(1): 8-20.

[4] Hooper A, Bekaert D, Spaans K, et al. 2012. Recent advances in SAR interferometry time series analysis for measuring crustal deformation. Tectonophysics, 514: 1-13.

Ground motion represents a persistent and often underestimated hazard for transport infrastructure, particularly in high-latitude and hydro-climatically sensitive environments. Landslides, subsidence, freeze–thaw dynamics, and underground construction can progressively weaken railways and roads, making early detection of displacement essential for risk mitigation. Interferometric Synthetic Aperture Radar (InSAR) offers a unique capability to monitor such processes with millimetre-scale precision over large spatial extents, yet the choice of InSAR time-series methodology strongly influences measurement density and interpretability along narrow infrastructure corridors.

This study presents a systematic comparison of two widely used InSAR approaches for infrastructure monitoring in Sweden: (1) Persistent Scatterer Interferometry (PSI) as implemented in the European Ground Motion Service (EGMS, PS-only calibrated LOS product), and (2) a modern phase-linked Persistent- and Distributed-Scatterer (PS+DS) workflow. Five transport corridors were analysed, spanning railways, highways, and underground tunnels across diverse geomorphological and climatic settings, including landslide-prone terrain, freeze–thaw-affected regions, and zones influenced by large-scale excavation.

Using identical quality thresholds for amplitude dispersion and temporal coherence, we evaluated point density, displacement time-series residuals (RMSE), short-range velocity variability, land-cover sensitivity, and the spatial relationship between measurement points and mapped road and railway centrelines. The PS+DS workflow consistently produced substantially higher point densities across all corridors, particularly in natural and mixed land-cover environments, while maintaining displacement RMSE values comparable to PSI, typically below 6 mm in the satellite line-of-sight direction. Although EGMS exhibits narrower RMSE distributions, reflecting its conservative PSI strategy, median error levels and velocity patterns are broadly similar between the two approaches. Both datasets capture coherent displacement signals along known unstable or climatically sensitive segments, and distance-to-centreline analyses reveal no systematic spatial bias favouring either method.

Overall, the results demonstrate that phase-linked PS+DS processing significantly enhances spatial coverage for infrastructure-scale monitoring without compromising displacement accuracy. These findings have direct implications for national ground-motion services and operational risk assessment in Sweden, particularly where early warning and corridor-wide situational awareness are critical.

The comparison also highlights practical trade-offs between conservative continental-scale products and more flexible, corridor-focused processing strategies for operational infrastructure surveillance. These differences are particularly relevant for agencies seeking scalable yet corridor-sensitive ground motion monitoring frameworks to support prioritised maintenance and early-warning decision workflows.

Tectonic Subsidence and Coastal Erosion: A PSI-InSAR Study in the Southern Caribbean Coast

José Francisco Valverde Calderón

Affiliation: School of Surveying, Cadastre and Geodesy, National University, 86-3000, Heredia, Costa Rica

e-mail: jose.valverde.calderon@una.cr

Gustavo Barrantes Castillo

Affiliation: School of Geographic Sciences, National University, 86-3000, Heredia, Costa Rica

e-mail: gbarrantes@una.ac.cr

Name: Diana Ninette Paniagua Jiménez

Affiliation: School of Surveying, Cadastre and Geodesy, National University, 86-3000, Heredia, Costa Rica

e-mail: diana.paniagua.jimenez@una.cr

Name: Matías Andrés Poch Clavero

Affiliation: Independent researcher, Santiago, Chile

e-mail: pochclavero@gmail.com

The Southern Caribbean coast of Costa Rica represents a complex and geodynamically active region. This area is characterized by the North Panama Deformed Belt (NPDB), a broad deformation zone featuring reverse faults and folds that extends westward into the Costa Rican territory. In recent years, the most notable seismic events occurred on April 22, 1991. The major shock had an intensity Mw 7.7, which caused substantial coseismic uplift along the coast. In the decades following this event, the interseismic behavior of the crust—specifically whether the land is rising or sinking—has been less understood due to a lack of continuous geodetic data available for the zone.

Understanding Vertical Land Motion (VLM) in this region is critical because any subsidence of the land, together the effects of sea-level rise, contributes directly accelerating coastal erosion and increases vulnerability of coastal communities. Recent observations indicate that sectors such as Cieneguita and Cahuita are suffering from severe erosion, beach shortening, and vegetation loss, prompting the need for a comprehensive assessment of the variables causing these effects.

To address the lack of spatial data on vertical ground movements, a study was done applying Persistent Scatterer Interferometry (PSI) to estimate VLM velocity across the Southern Caribbean of Costa Rica, covering key areas such as the city of Limón, Cahuita, Puerto Viejo, and Manzanillo.

The research utilized Single Look Complex (SLC) images from the Sentinel-1 mission, acquiring data from both ascending and descending orbits to resolve the vertical component of the displacement. A total of 33 ascending images (spanning January 2019 to February 2021) and 32 descending images (spanning January 2019 to February 2021) were processed. The VV polarization was selected for processing due to its superior coherence stability over the region’s land cover, which includes urban areas, rock outcrops, and dense vegetation.

The image processing begins with preprocessing in the SNAP software to coregister images, correct orbital errors using precise Copernicus orbits, formation of the interferograms and remove the topographical phase. Later, the Stanford Method for Persistent Scatterers (StaMPS) was employed to identify persistent scatterers (PS). Atmospheric phase contribution was corrected using a linear phase-based tropospheric model available in the TRAIN toolbox. Finally, LOS velocities from both geometries ascending and descending were combined to obtain vertical velocity.

As results, average RMSE for the LOS velocities was ±1.67 mm/yr for the ascending orbit and ±1.21 mm/yr for the descending orbit, indicating a high level of precision in the interferometric results. The estimated vertical velocities for the study area averaged -7.04 mm/yr, with values ranging from a minimum of -20.4 mm/yr to a maximum of -1.54 mm/yr. Specific analysis of the city of Limón showed vertical velocities between -3.5 mm/yr and -6.0 mm/yr, particularly in the eastern sector bordering the coast.

The reliability of these satellite-derived measurements was validated by comparing them with data from the continuous GNSS station "LIMN," located in Limón. The InSAR vertical velocity of this station aligns well with the station's geodetic record showing a vertical velocity trend of -4.32 mm/yr. Furthermore, another nearby station, VRAI, reported a vertical velocity of -5.88 mm/yr, providing further independent corroboration of the subsidence magnitude and direction identified by the PSI analysis.

These findings have implications for the tectonic study of the region. The observed subsidence is consistent with an interseismic phase of a subduction cycle. Following the coseismic uplift of the 1991 earthquake, the plates have likely re-coupled. In this locked state, the accumulation of stress drags the upper plate (the Panama Microplate) downward, resulting in the observed elastic subsidence. This geophysical behavior mirrors the well-documented cycle of the Nicoya Peninsula on Costa Rica's Pacific coast, where the land subsides during the interseismic period due to the strong coupling between the Cocos and Caribbean plates, only to rise suddenly during major earthquakes.

Beyond tectonics, the study highlights the critical impact of Vertical Land Motion on coastal dynamics. The relative sea-level rise experienced by a coastal community is the sum of the absolute sea-level rise (climate-induced) and the vertical movement of the land. Previous analyses of the Limón tide gauge suggested an accelerated relative sea-level rise rate of over 8 mm/yr in recent years; however, when corrected for the tectonic subsidence identified in this study (approx. -4 to -7 mm/yr), the absolute sea-level rise aligns more closely with global averages, yet the relative threat remains critically high.

This rapid relative rise provides a physical explanation for the erosive processes reported in recent years, such as the destruction of infrastructure and beach retreat in Cieneguita and Cahuita. Consequently, this research emphasizes that coastal risk management and adaptation strategies in the Southern Caribbean must account for this tectonic subsidence. Ignoring the vertical land motion component would lead to a underestimation of future sea-level impacts and coastal vulnerability.

Finally, the study demonstrates the value of interferometric techniques as a complementary tool to GNSS networks, providing high-density spatial data essential for monitoring crustal deformation and informing decision-making in coastal zones.

Santiago de Puriscal, Costa Rica, is a city located approximately 40 km southwest of the country's capital, San José, with a population of about 12,500. The urban area is characterized by the active presence of a landslide covering between 4–5 km². According to the National Emergency Commission of Costa Rica, the Puriscal canton faces both hydrometeorological and geological hazards. The Santiago de Puriscal landslide falls into the latter category; it is the largest in the country and is situated directly beneath the most densely populated sector. The landslide's impact is intensified by periods of local seismic activity and the rainy season.

Studies conducted in the 1990s showed displacement velocities between 5 and 15 cm/yr toward the northwest. Consequently, local infrastructure has been severely impacted: several buildings have been demolished, and others have sustained significant cracking. The most notable example is the old Catholic church, which was declared uninhabitable due to the risk of structural collapse.

The present project used Single Look Complex (SLC) images from the Sentinel-1 mission, acquiring data from both ascending and descending orbits. A total of 92 ascending images (spanning January 2018 to December 2023) and 90 descending images (also spanning January 2018 to December 2023) were processed. The VV polarization was selected for processing. To minimize the possible effects of temporal decorrelation, the processing period was split into 2 segments: the first one from January 2018 to November 2020 and the second from December 2020 to December 2023.

The image processing begins with preprocessing in the SNAP software to coregistered images, correct orbital errors using precise Copernicus orbits, formation of the interferograms and remove the topographical phase. Later, the Stanford Method for Persistent Scatterers (StaMPS) was employed to identify persistent scatterers (PS) and its velocities. Finally, LOS velocities from both geometries ascending and descending were combined to obtain vertical velocity.

As results, for the first period of processing, the average LOS velocity is -10.18 mm/yr and +7.58 mm/yr for both ascending and descending orbit. RMSE for the LOS velocities is ±1.54 mm/yr for the ascending orbit and ±1.31 mm/yr for the descending orbit, indicating a high level of precision in the interferometric results. The estimated vertical velocities for the study area averaged -2.40 mm/yr, with values ranging from a minimum of -14.84 mm/yr to a maximum of +5.01 mm/yr. The projection of the horizontal velocity (vHald) has an average value of 13.41 mm/yr, with a maximum value of +24.32 mm/yr and a minimum value of -7.42 mm/yr.

With respect to the second period of processing, the average LOS velocity is -6.02 mm/yr and +7.64 mm/yr for both ascending and descending orbit. RMSE for the LOS velocities is ±1.55 mm/yr for the ascending orbit and ±1.31 mm/yr for the descending orbit, indicating a high level of precision in the interferometric results. The estimated vertical velocities for the study area averaged +0.47 mm/yr, with values ranging from a minimum of -8.54 mm/yr to a maximum of +9.78 mm/yr. The projection of the horizontal velocity (vHald) has an average value of +10.51 mm/yr, with a maximum value of +28.95 mm/yr and a minimum value of -2.99 mm/yr.

This study evaluated the potential of applying InSAR techniques in Costa Rica for landslide monitoring, specifically the PSI-InSAR method, using the Santiago de Puriscal landslide as a case study. This landslide has been active for several decades and has caused damage to public and private infrastructure. The results show that the landslide is still active, making it necessary to continue its ongoing monitoring.

Finally, the study demonstrates the value of interferometric techniques as a complementary tool to GNSS networks or other geodetic techniques, providing high-density spatial data essential for monitoring active landslides.

Western Carpathians region is due to its complex lithology and humid climate is highly prone to landslides. Most of them are rather stable. Since catastrophic mega-landslide in Handlová in 1960 a lot of effort has been put into mapping, monitoring and remediation (cite) also of other locations than Handlová town. For decades the overall situation was really stable, only solitary locations were activated. Until it came “year of landslides”, 2010. Exceptionally heavy precipitation during May and June 2010 triggered 577 newly active landslides, and many previously stable sites were activated. Surveying was done mainly by the geologists from SGIDS. Probably the worst case was a night catastrophic landslide directly in the center of village Nižná Myšľa on 4. June 2010 when overnight 40 houses were damaged and as a consequence 144 inhabitants had to be evacuated. Landslide 1 500 x 500 m with slip surface about 14 m deep appeared from evening till morning. The reason was high water saturation in deep clay and tuffite sediments, Neogene in age..

Remediation work hand in hand with monitoring commenced immediately in the area. Horizontal and vertical boreholes were drilled, ground water was drained, water flow was monitored, network of devices like inclinometers and ground water level measuring indicators were installed. Since year 2018 the location has been observed also by PS InSAR technique with Sentinel-1 data. Natural scatterers as well as artificial corner reflectors are in use.

In this study we combine long term Earth Observation data deriving precipitation, soil moisture, evapotranspiration and water saturation, with ground-based measurements about surface and undersurface displacement, and ground water regime.

Mainly free accessible ESA Copernicus L1, L2 data serve as input, yet combined with other higher level products – e.g. TU Wien ACTIVE/PASSIVE soil moisture, ECMWF climate indicators. There is automatic climatological station in the settlement, thus also these data together with high resolution data from the meteorological radar are also compared.

First we evaluate data quality and assess relationships among the datasets. Next step is to apply time series multivariate statistical methods in order to quantify dependency between PS InSAR displacements, underground slip surface deformations, and atmospheric and surface climatological circumstances.

Using 20 years of historical data, different model cases are described by means of exact correlation coefficients. These shall be applied later on in raster-based spatio-temporal monitoring model yet on different places without in-situ ground-based measurements.

The Flemish Institute for Technological Research (VITO) provides centralized access to remote sensing datasets and scalable processing capabilities via Terrascope, with a strong focus on science-enabling and decision-ready data products. In collaboration with NASA’s JPL Advanced Rapid Imaging and Analysis (ARIA) team and the Alaska Satellite Facility (ASF) Distributed Active Archive Center (DAAC), the TerraScope portfolio has been expanded to re-host over Belgium Geocoded Unwrapped Interferograms (GUNW) and Radiometrically Terrain Corrected (RTC) products derived from Sentinel-1 and NISAR. These datasets are fully integrated within the TerraScope ecosystem, lowering barriers for local and regional user communities to engage in advanced SAR and InSAR analysis. This provides a straightforward entry point for multi-frequency investigations within a scalable and extensible framework that can readily expand to other regions.

Example applications include change detection, disturbance mapping, surface water extent monitoring, and displacement analysis. The uniform accessibility of these SAR datasets along side optical and hyperspectral products further facilitates the development of advanced analytics and machine learning approaches built upon consistent, multi-frequency sensor data streams. We will present two features namely the TerraScope viewer and the openEO framework in the context of Sentinel-1 and NISAR data access and processing.

The TerraScope interactive viewer enables discovery, query, and elementary analysis of optical, hyperspectral, and SAR observations across a broad portfolio of low- to higher-level Analysis Ready Data (ARD), spanning domains from ecosystems to solid Earth monitoring. Its backbone relies on a STAC-compliant architecture, allowing users to seamlessly access and integrate datasets for cross-domain and multi-sensor investigations. A new TerraScope QGIS plugin and Leafmap compatibility allow for streamlined discovery and visualization.

The OpenEO framework enables scalable downstream processing, where users request processing through an awarded credit system and execute either their own workflows or existing ones (e.g., within the CDSE ecosystem). We introduce a new openEO workflow for the generation of customizable GUNW products from Sentinel-1, enabling users to process interferometric data over user-defined areas with tailored processing parameters.

Interferometric Synthetic Aperture Radar (InSAR) has become a key technique for millimetre-level monitoring of ground deformation over large spatial scales. While the method is inherently relative, many geodetic and infrastructure-monitoring applications require transformation of these relative measurements into a stable and traceable terrestrial reference frame. Achieving this objective demands well-characterised fiducial radar targets and robust integration with established national geodetic infrastructure.

To address this need, Lantmäteriet (The Swedish mapping, cadastral and land registration authority) has established a national SAR fiducial network consisting of active electronic radar transponders and passive corner reflectors distributed across Sweden. Since 2020, three active transponders and twenty passive reflectors have been installed at carefully selected locations that ensure long-term stability and optimal radar visibility. The network has been designed with emphasis on geometric configuration, environmental stability, and compatibility with Copernicus Sentinel-1 acquisition geometry.

All radar targets are co-located with permanent GNSS stations and, where available, absolute gravity benchmarks. These multi-technique reference sites strengthen the national geodetic infrastructure by providing observations from independent measurement techniques. The configuration supports reference frame maintenance, enhances traceability, and enables rigorous comparison between InSAR-derived deformation and other geodetic time series. The sites also provide opportunities for calibration and validation (Cal/Val) of SAR-derived products, including independent assessment of results from the European Ground Motion Service (EGMS). Furthermore, the infrastructure facilitates linking relative InSAR-derived ground motion to the national geodetic reference frame.

This contribution presents the recent progress in the installation and characterisation of both active and passive radar targets is described. Performance analyses have been carried out for selected sites, with particular focus on radar cross section (RCS) stability and signal-to-clutter behaviour. Special attention is given to seasonal effects during snow-covered periods, which are of particular relevance in high-latitude environments. Initial characterisation results are presented and discussed in relation to their implications for long-term stability assessment and future Cal/Val applications.

The establishment of this national SAR fiducial network represents an important step toward strengthening the geodetic integration of InSAR in Sweden and contributes to ongoing European efforts to ensure consistent and traceable ground motion products.

1 Objectives, Study Area, and Data

This study aims to improve the monitoring and understanding of earthquake-induced landslides as well as associated ground deformations, thru the application of advanced satellite remote sensing techniques. The main objectives are as follows:

• Detect and quantify surface displacements triggered by seismic activity,

• Identify areas vulnerable to landslides in order to refine hazard assessment,

• Implement Synthetic Aperture Radar (SAR) methodologies, particularly Differential InSAR (DInSAR), Coherence Change Detection (CCD), and InSAR time-series analysis via the LiCSBAS software, to comprehensively monitor the spatiotemporal evolution of ground movements,

• Validate deformation measurements derived from SAR by integrating high-resolution optical images from Sentinel-2 satellites and complementary field inspections.

The research focuses on the Mila region, located in northeastern Algeria, within the seismically active Mediterranean zone, characterized by the convergence between the African and Eurasian plates. On August 7, 2020, a moderate earthquake with a moment magnitude (Mw) of 5.0 occurred near Mila (epicenter coordinates: 36.550°N, 6.271°E, depth ~10 km), causing significant landslides and damage in the towns of Kherba and Grarem.

Fig. 1. shows a location map of Mila, orbit footprints, and earthquake epicenters. 3D representations of Kherba and Grarem AoIs (QGIS, ESRI basemap), with boundary of change identified by InSAR.

In order to capture a robust temporal and spatial representation of ground deformations before and after the seismic event, the study relies on a dataset covering the period from April 2015 to October 2020. This includes 35 Sentinel-1 C-band SAR images (ascending and descending orbits) used for interferometric analysis, allowing for the application of DInSAR and CCD for the assessment of displacements and coherence. Complementary Sentinel-2 multispectral optical images are used to validate the surface changes detected by SAR methods. This multi-sensor and multi-temporal approach allows for a detailed analysis of the dynamics of landslides triggered by the earthquake and contributes to the identification of potential precursor movements or persistent post-seismic deformation.

2 Methodology

The study employs several remote sensing approaches focused on Interferometry Synthetic Aperture Radar (InSAR) to analyze earthquake-induced landslides and associated ground deformations. The synthetic aperture radar (SAR) is an active microwave imaging system capable of acquiring high-resolution data, regardless of weather conditions and solar illumination, thus offering robust spatial and temporal monitoring capabilities.

Principles of Synthetic Aperture Radar Interferometry (InSAR)

InSAR techniques exploit the phase difference between two or more SAR images acquired over the same area at different times, in order to detect subtle surface changes with millimeter-level precision. The interferometric phase includes contributions related to surface displacement, topographic variations, atmospheric effects, orbital errors, and noise. A rigorous data processing is therefore necessary to identify the displacement component relevant to geotechnical phenomena.

Differential Interferometric Synthetic Aperture Radar (DInSAR)

DInSAR specifically targets phase variations associated with surface deformation by subtracting topographic contributions and other static phase components. It allows for the effective measurement of coherent ground movements, such as subsidence or uplift, that have occurred between two successive SAR acquisitions. However, DInSAR has limitations in cases of rapid or large-scale landslides, where the temporal decorrelation of the radar signal disrupts phase coherence and makes displacement estimates unreliable.

Coherence Change Detection (CCD)

To overcome the limitations of phase-based methods in highly dynamic landslide areas, CCD uses the coherence metric, which quantifies the similarity of radar signals between image pairs, in order to identify areas of significant surface disturbance. A decrease in coherence indicates changes such as ground cracking, vegetation modifications, or soil disturbances. CCD thus serves as a complementary tool for effectively mapping landslides marked by decorrelation, expanding detection capabilities beyond coherent deformations measurable by DInSAR.

LiCSBAS Time-Series Analysis

Advancing beyond pairwise interferograms, time-series analysis, applied using the open-source software LiCSBAS, processes large sets of Sentinel-1 SAR images to generate displacement velocity maps and temporal deformation profiles. This approach improves noise reduction, mitigates atmospheric artifacts, and reveals slow or precursor ground movements over long periods. It proves essential for distinguishing trends before, during, and after seismic events, thereby facilitating a better understanding of the triggering and evolution of landslides.

Processing optical images for validation

In addition to SAR data, multispectral optical images from Sentinel-2 satellites are used to validate and contextualize the detected surface changes. Optical images offer an intuitive visualization of surface features, such as landslide cracks and vegetation changes, although their usefulness may be limited by cloud cover and lighting conditions. True-color compositions, processed via QGIS software, allow for spatial correlation with deformation maps derived from SAR, thereby enhancing the reliability of remote sensing interpretations and supporting field validation.

3 Results and Discussion

The application of SAR interferometric techniques, complemented by optical imagery and field surveys, allowed for a comprehensive detection and characterization of landslides and ground deformations induced by the earthquake in the Mila region of Algeria. Two main areas of interest, Grarem and Kherba, exhibited significant deformation patterns related to the Mw 5 magnitude earthquake that occurred on August 7, 2020.

Fig. 3. Ground cracks due to landslides in Kherba, Mila, 2:5m offset towards the north. (a) Drone aerial photo from LNHC (2021). (b, c) Lateral displacements (photos: courtesy M. Yacoub Ali, University of Setif, Algeria).

Landslide and Ground Deformation Detection

In the Grarem area, differential interferograms revealed small but distinct deformation fringe patterns, corresponding to ground displacement increments on the order of half a wavelength of the Sentinel-1 radar (approximately 2.77 cm per fringe). These coherent deformations were spatially confined to an area of approximately 3.94 km², validated by coherence losses at the fringe boundaries and by field observations highlighting surface cracks. The Time-series analysis further demonstrated that the displacements were co-seismic with the event, without any indication of precursor deformation in the preceding years.

Fig. 4. Detected fringes in interferograms 3, 17, and 22, with images focused on the Grarem zone.

Conversely, the Kherba area exhibited more complex ground behavior, including large landslides causing significant decorrelation and inconsistent radar returns, limiting the effectiveness of DInSAR phase analysis alone. In this context, coherence change detection (CCD) proved indispensable, allowing for the mapping of two distinct landslide toes and the detection of coherence reductions reaching 23%. The CCD time series showed that landslide activity persisted and evolved dynamically in the weeks following the earthquake, although data quality subsequently deteriorated later due to environmental and anthropogenic noise.

Fig. 5. Coherence time series maps of the Kherba landslide, Sentinel-1.

Fig. 6. Poor coherence in the acquisition on 3 August was due to unfavorable weather conditions, while other acquisition showed better coherence.

Displacement Velocities and Time-Series Insights

The long-term analysis conducted using LiCSBAS allowed for the establishment of velocity maps, revealing a significant post-seismic subsidence reaching approximately 110 mm yr⁻¹ at the back of the slope of the Kherba landslide, likely triggered by a mass redistribution following the earthquake. On the other hand, the displacements in the Grarem area remained relatively stable during the post-event monitoring period. These spatiotemporal displacement patterns highlight the critical value of extended time-series analyzes to distinguish co-seismic transient deformation from ongoing or pre-existing ground movements.

Validation and Integration with Optical and Field Data

The Sentinel-2 optical images proved to be an effective validation tool, confirming the boundaries of landslides as well as surface cracks identified by SAR methods. However, optical images did not detect certain subtle or initial features of landslides visible thru coherence changes in SAR data, emphasizing the superior sensitivity of radar monitoring in vegetated regions or those subject to frequently cloud cover. Field inspections confirmed the presence of fractures and slope instabilities in the mapped landslide areas, reinforcing the reliability of remote sensing observations.

Fig. 7. Sentinel-2 optic images: (a) of 30 July 2020 and (b) dated 9 August 2020, co-event coherence ratio, Sentinel-1. The green box indicates RoI and red spots represent significant changes of coherence in the landslide region.

Implications and Limitations

The study demonstrates the complementary strengths of DInSAR, CCD, and LiCSBAS time-series analysis for multi-temporal and multi-scale monitoring of earthquake-induced landslides. While DInSAR proves effective for detecting coherent and relatively slow ground movements, CCD is essential for identifying areas affected by rapid or incoherent changes associated with large landslides. Time-series analysis enriches the depth of observation by revealing displacement trends and potential precursor activities, although no such signal was identified before this event.

However, challenges remain in accurately quantifying the horizontal and vertical components of displacement, as well as in reducing noise related to atmospheric disturbances, vegetation changes, and weather conditions affecting coherence. Moreover, the limitations of phase unwrapping for high displacement gradients restrict the accuracy of deformation estimates in heavily damaged areas.

4 Conclusions

This study demonstrates that InSAR techniques offer an effective and precise means of monitoring earthquake-induced landslides, enabling the detection and quantification of ground deformations across large and complex terrains. Three major landslides were successfully identified and characterized in the Mila region, with detailed measurements revealing substantial ground shifts, most notably the up to 2.5-meter displacement in the Kherba area. The complementary application of Differential InSAR (DInSAR) and Coherence Change Detection (CCD) highlights their respective strengths, DInSAR is well suited for detecting coherent, relatively slow ground movements, while CCD excels in capturing rapid, large-scale landslides marked by strong signal decorrelation. The findings underscore the importance of integrating satellite-based observations with in-situ monitoring tools such as GPS and inclinometers to enhance both spatial and temporal resolution. Additionally, Persistent Scatterer InSAR (PS-InSAR) is identified as a promising approach for future high-precision, long-term monitoring of slope dynamics and stability in critical zones.

In the context of displacement monitoring using satellite SAR data, measurable points are confined to persistent scatterers; however, PSInSAR analysis is highly effective because of its capacity for estimating high-precision displacement. Osaka Bay in Japan contains numerous reclaimed land areas, including Kansai International Airport and Kobe Airport, with certain locations experiencing subsidence.

The annual average subsidence at Kansai International Airport was recorded as 6 cm for Phase 1 Island and 24 cm for Phase 2 Island, respectively.

This study analyzed ground surface displacement around Osaka Bay using PSInSAR with C-band Sentinel-1 and L-band ALOS-2 SAR data acquired over approximately eight years, from January 2017 to March 2025. The Sentinel-1 data utilized for the analysis comprised 245 scenes at 12-day intervals, whereas the ALOS-2 SAR data comprised 33 scenes covering the northern area, including Kobe Airport, and 32 scenes covering the southern area, including Osaka International Airport.

In particular, within reclaimed coastal areas where significant land subsidence has been observed, the influence of the SAR sensor type, characteristics of the observed structures and ground conditions, analysis period, number of SAR scenes, and amplitude dispersion index (ADI) threshold used for PS point extraction was quantitatively evaluated in terms of measurable displacement coverage and estimation accuracy. Furthermore, validation using ground-surveyed subsidence data from Kansai International Airport demonstrated that Sentinel-1 enabled millimeter-level accuracy in displacement estimation.

The comprehensive PSInSAR findings are as follows:

For PS points selected using Sentinel-1, the analysis results for Kobe Airport and Port Island over 1, 2, 4, and 8 years revealed that for buildings, roads, and runway lights, where surface changes over time were estimated to be minimal, the number of PS points decreased with longer analysis periods and more scenes, albeit only by approximately 20%. In contrast, for structures such as seawalls, where changes over time are estimated to be larger than those for buildings, the number of PS points decreased by approximately 40%. For runways, where changes over time are estimated to be even larger, the number of PS points decreased by approximately 10%.

When comparing the number of PS points using ALOS-2 with Sentinel-1, both employing an amplitude dispersion index threshold of 0.4 for PS point selection, the number of PS points for buildings and similar features was approximately ten times higher for ALOS-2. Regarding the effect of the amplitude dispersion index threshold on the PS points, thresholds of 0.3 and 0.1 resulted in reductions of approximately 40% and 10%, respectively, compared with the threshold of 0.4. Areas with grass growth near the runways were more readily detected by ALOS-2 than by Sentinel-1, and PS points were obtained at some locations even during the eight-year analysis period.

The accuracy of measuring PS points using Sentinel-1 SAR data, as determined from the analysis results for 8, 4, 2 years, and 1 year, indicated that for buildings and roads, the standard deviation of displacement in 50m×50m blocks decreases with longer analysis periods and an increased number of scenes. Specifically, for 8-, 4, and 2 years periods, it is approximately 1 mm/year or less, whereas for 1-year periods, it is approximately 4 mm/year. Notably, for periods of two years or longer, augmenting the number of scenes does not significantly enhance the measurement accuracy.

At Osaka International Airport, the measurement accuracy of the PS points using Sentinel-1, adjusted by establishing reference points on each island, revealed an RMSE of approximately 3 and 4 mm for the first-phase island and approximately 7 and 9 mm for the second-phase island over the 2 years and 1 year analysis periods, respectively. Despite the large annual ground subsidence of up to 28 cm and the measurement values being in the centimeter range, it was confirmed that displacement measurements can be achieved with high accuracy in the millimeter range.

The measurement accuracy of ALOS-2 was compared with the analysis results from Sentinel-1 for Kobe Airport and Port Island. The standard deviation of vertical displacement in 50m×50m grids at amplitude dispersion indices of 0.4, 0.3, and 0.1 was observed to be 2.0 mm/year, 1.6 mm/year, and 1.2 mm/year, respectively, for structures such as buildings. This illustrates the influence of the amplitude dispersion index on measurement accuracy.

Deformation time series derived from Synthetic Aperture Radar (SAR) interferometry are routinely generated through established processing chains and operational services. While the generation of these products is well established, structured environments for their systematic downstream analysis remain limited. Interpretation of spatial patterns, temporal evolution, and acquisition geometry is often implemented through study-specific analytical workflows rather than through structured and reusable frameworks. This makes it challenging to reproduce analyses consistently and to apply comparable methods across different datasets.

We present ITAS (InSAR Time-Series Analysis), an open-source Python framework for structured downstream analysis of InSAR-derived deformation products. ITAS operates on deformation time series generated by external InSAR processing chains and integrates spatial context, temporal behaviour, and acquisition geometry in a modular framework. The framework is structured around a user-defined Area of Interest (AOI) and implements a reproducible project structure with transparent data handling and modular analytical components.

ITAS is organized into three complementary analytical domains: Spatial Data Analysis (SDA), Temporal Data Analysis (TDA), and Spatio-temporal Data Analysis (STDA). SDA addresses the spatial characteristics of deformation fields, including geometric relationships to terrain and acquisition geometry, spatial calibration, directional projection, and component decomposition. This module establishes a consistent spatial reference framework for multi-geometry integration and supports geometry-aware transformation of line-of-sight measurements into interpretable deformation components. TDA focuses on the temporal behaviour of deformation time series at individual locations, including trends, variability, and time-dependent changes in deformation characteristics. The module also supports integration of complementary time series, such as meteorological observations, to facilitate exploration of potential relationships between external forcing and deformation response. STDA integrates spatial and temporal perspectives to examine how deformation patterns evolve coherently across space and time. Together, these domains provide a structured analytical framework for consistent interpretation of InSAR-derived deformation products across different study contexts.

ITAS has been developed with slope instability research as a primary application domain, while remaining adaptable to other deformation contexts including subsidence, infrastructure monitoring, and cryospheric processes. By providing a modular and reproducible analytical environment, the framework supports consistent interpretation of InSAR-derived deformation products across different study areas and datasets. The framework is implemented in Python and designed for interactive and script-based workflows, facilitating transparent and transferable analytical practice.

The monitoring of ground deformations in mining areas is essential for ensuring urban safety. The study investigates land subsidence caused by underground hard coal mining at the former “Pokój” mine located in the urbanized center of Ruda Śląska, within the Upper Silesian Coal Basin (USCB), Poland. The objective is to determine the duration required for subsidence troughs to stabilize, assessed on an annual basis covering the years 2019 – 2025. The project compares radar images from before and after the closure of the mine, taking into account that the definitive cessation of the extraction took place in early 2021.

The investigated area covers 11.2 km² and includes significant urban infrastructure, such as key transport routes and various public buildings. The analysis employs a satellite radar interferometry (InSAR) approach to detect deformations using Sentinel-1 imagery. The dataset consists of C-band, SLC images acquired in interferometric wide (IW) swath mode. For each year, a pair of images from the first quarter were processed using the differential satellite interferometric synthetic aperture radar (DInSAR) method. The time interval between each image in a pair was 36 days. All these factors allowed to obtain satisfying coherence and identify deformations in further steps.

The final results of the research present displacement maps with defined boundaries of the subsidence basins. The findings demonstrate the presence of active subsidence troughs in 2019 and 2020, which was the period of active extraction. They occurred in the Wirek and Bielszowice districts, with displacements reaching up to -8 cm within the analyzed 36-day windows. Comparing 2019 to 2020, spatial analysis reveals an increase in troughs extent and their migration following the advancing mining front.

During the analyzed periods after the cessation of extraction, in 2021 – 2025, no new displacements were recorded. These results illustrate that the mine closure process does not necessarily lead to long-term subsidence lasting for several years post-decommissioning. The rock mass can return to a stable state shortly after the cessation of exploitation, depending on factors such as mining intensity and depth. In the case of the "Pokój" mine, the relatively shallow exploitation levels (320 m, 600 m, and 790 m) favored rapid terrain stabilization.

Consequently, the “Pokój” mine represents an example of a mining area where the formation of subsidence troughs ceased almost immediately after the closure of the extraction and the terrain remained stable in the following years.

The “Democratization of Earth Observation” (DEM-EO) project, funded by the Italian Space Agency (ASI), aims to reduce key barriers that still limit the operational uptake of Earth Observation for ground-motion monitoring - namely fragmented access to data and services, complex processing and interpretation workflows, and the scarcity of specialized expertise within end-user organizations - by developing an innovative, web-based platform that makes advanced EO analytics more accessible. Despite the growing availability of EO data and services, many potential users still face fragmented access to datasets, complex processing chains, and significant interpretation challenges, which often require highly specialized skills and lead to reliance on outsourcing. DEM-EO addresses this gap through an integrated environment that streamlines the path from data discovery and acquisition to the generation, visualization, and export of information products, with a specific focus on monitoring terrain instability and the behavior of critical infrastructures. The platform integrates complementary methodologies to support robust interpretation across a range of scenarios: Advanced Differential InSAR (A-DInSAR) for retrieving the temporal evolution of ground deformation from multi-temporal SAR interferometry, and PhotoMonitoring™ techniques to derive high-detail displacement and change information from optical imagery, including Digital Image Correlation (DIC) with sub-pixel sensitivity and Change Detection (CD) for mapping radiometric and geometric variations relevant to hazard evolution and asset integrity. A multi-mission approach (e.g., Sentinel-1, COSMO-SkyMed, SAOCOM) is considered to enhance spatial and temporal coverage and strengthen the understanding of deformation processes by combining different acquisition geometries, revisit times and wavelengths.

A central element of DEM-EO is the emphasis on usability and operational transfer: the platform is intended to guide users through coherent, repeatable workflows, reducing manual steps and minimizing the need to master multiple disconnected tools. This is particularly relevant in organizational contexts where EO competences are scarce, where procurement cycles are long, and where the need for rapid, defensible information can arise both in routine monitoring and in emergency conditions. In addition to technology integration, DEM-EO explicitly embeds user enablement mechanisms—guidance, optional support and validation pathways, and a dedicated training plan—to foster uptake beyond expert communities and help organizations build internal capability for continuous monitoring and timely decision-making. The training activities are designed to transfer not only “button-level” platform skills, but also the conceptual background required to interpret deformation and change signals correctly, understand uncertainty sources, and recognize typical pitfalls in EO-based monitoring (e.g., the role of acquisition geometry, coherence loss, atmospheric artefacts, seasonal effects, or illumination differences in optical change analyses). In this way, DEM-EO supports a progressive pathway from assisted use to greater user autonomy, enabling stakeholders to integrate EO products into their internal procedures and reporting chains.

Planned validation activities play a key role in ensuring that the platform’s outputs are credible and actionable. For ground-motion products, A-DInSAR results are expected to be compared - where available - with independent in-situ observations as well as with ancillary information that can support interpretation (e.g., known instability inventories, infrastructure maintenance records, or documented events). Complementary usability and performance assessments will verify that the platform remains effective for both expert and non-expert users, and that it can support different operational rhythms, from periodic surveillance to rapid screening after triggering events. The integration of SAR-based ground-motion analytics with optical change and displacement mapping is specifically intended to improve interpretability: deformation time series can be complemented by optical evidence of surface changes, damage proxies, or localized displacement patterns, supporting a more robust understanding of the processes at work and helping users prioritize field inspections and mitigation actions.

Ultimately, DEM-EO seeks to accelerate the transition from EO data availability to operational use by pairing reliable processing capabilities with practical knowledge transfer and adoption support. By reducing barriers to access and interpretation, expanding the pool of EO-enabled stakeholders, and shortening decision latency through more autonomous use, the project aims to strengthen the EO downstream ecosystem for infrastructures and geo-environmental risk management. In this perspective, DEM-EO contributes to making EO-based ground-motion monitoring not only technically feasible, but also organizationally sustainable, supporting routine risk-informed management and more timely responses when critical conditions emerge.

Operators of critical rail infrastructure face a continuous challenge in monitoring their networks for ongoing maintenance, and preventing potentially catastrophic accidents. In recent years, Earth observation (EO) technology has greatly enhanced our ability to undertake remote, large-area monitoring, yet many challenges remain unresolved. The upcoming Copernicus Expansion Missions will provide new sources of data and opportunities to address these challenges. Working in synergy, the Radar Observing System for Europe in L-band (ROSE-L) and Copernicus Hyperspectral Imaging Mission for the Environment (CHIME) missions will provide particularly useful data for rail infrastructure monitoring purposes, by bringing together routine, high precision L-band SAR and hyperspectral information for the first time.

In this context, ESA has funded the Rail-Adjacent Vegetation monitoring with Copernicus Expansions (RAVE) project under the auspices of the Sentinel User Preparation initiative, which is currently being delivered by Airbus Defence & Space UK in collaboration with Champion Users Network Rail and the Société Nationale des Chemins de fer Français Réseau (SNCF Réseau). These entities are the primary rail operators in the UK and France, respectively.

Here, we showcase the potential of the Copernicus Expansion Missions for a range of Champion User-defined critical infrastructure monitoring applications, with a particular emphasis on new insights that will be enabled by ROSE-L. Specifically, we present new backscatter- and InSAR-enabled workflows for the retrieval of rail-adjacent Tree Height, Tree Windthrow and Flood Extent monitoring, designed and tested using emulated ROSE-L imagery produced from existing L-band SAR data sources (ALOS, SAOCOM-1A/1B) over a selection of UK- and France-based test locations. As part of the wider SUP initiative, these emulated datasets will be made publicly available via ESA’s Project Results Repository and the Network of Resources to support other Copernicus Expansion Mission preparation activities in the future. Further information about the RAVE project, including links to early outputs, can be found on the project website (https://rave.apex.esa.int/en).

Tropical peatlands are major global carbon stores, yet widespread drainage and land conversion have transformed many into persistent carbon sources through peat oxidation. In Indonesia, Sumatran peatlands are estimated to contribute to two-thirds of the country’s peat carbon emissions, but emissions remain poorly constrained due to limited observations. Peat subsidence provides a measurable indicator of peat oxidation and carbon loss, but large-scale, spatially consistent monitoring has been hindered by dense vegetation, cloud cover, and sparse field measurements.

Here, we present an eight-year (2016 to 2024) L-band Interferometric Synthetic Aperture Radar (InSAR) analysis of vertical land motion across 6.81 million hectares of peatlands in Sumatra. The L-band wavelength allows detection past dense tropical vegetation. By integrating ALOS-2 InSAR with Global Navigation Satellite System (GNSS) data and isolating tectonic signals, we derive spatially continuous peat motions at 180 m resolution, enabling regional-scale interpretation of peat dynamics independent of tectonic deformation.

We find that 93% of Sumatran peatlands are subsiding, with median rates of -2.0 cm yr^-1 (5^th / 95^th percentiles: -4.3 / 0.3 cm yr^-1). Subsidence is widespread across all provinces and not confined to intensively drained land uses. Acacia plantations – the most intensely drained land use – show the fastest mean subsidence of -3.3 ± 1.2 cm yr^-1 as expected. However, peat swamp forests – areas classified as pristine and without drainage – also exhibit subsidence, with a mean rate of -1.5 ± 1.0 cm yr^-1, indicating ongoing carbon loss, and challenging assumptions that forests typically function as carbon sinks. Some degraded peatlands including Acacia plantations, oil palm plantations, and smallholder agriculture areas also subside more slowly than previously reported. Variability of peat motions within the same land use and drainage classes is high, demonstrating that categorical proxies alone cannot reliably represent peat carbon loss.

The high variability in InSAR-derived peat motions within the same land use classes indicates that Intergovernmental Panel on Climate Change (IPCC) Tier 1 accounting, which assigns fixed emission factors to each land use class, can introduce substantial bias. To address this, we convert pixel-level subsidence rates into carbon emissions to implement a Tier 3-equivalent approach, which entails site-specific, locally informed estimations at fine spatial resolution. This yields total peat carbon emissions of approximately 61 million t CO₂-C yr^-1 across Sumatra. By capturing spatial heterogeneity that Tier 1 emission factors cannot resolve, the InSAR-based method provides an observation-driven framework to refine greenhouse gas inventories and support more accurate climate mitigation strategies.

RASTOOL-DoS focuses on the development of a prototype ground-motion analysis service designed to be integrated within the Copernicus Emergency Management Service (CEMS) On-Demand Mapping framework. The methodology builds upon previous European initiatives on geohazard monitoring and prevention, evolving from periodic deformation mapping toward semi-automated detection and interpretation of active ground motion. Central to this approach is the identification of Active Deformation Areas (ADAs), obtained by clustering Persistent Scatterer time-series derived from multitemporal InSAR datasets. These ADAs enable the extraction of deformation hotspots and the derivation of higher-level products, including preliminary impact indicators based on the spatial intersection with exposed infrastructure and population datasets.

The proposed workflow emphasizes reproducibility and rapid deployment during emergency scenarios. Standardized processing chains and predefined analytical steps allow users with limited InSAR expertise to exploit ground-motion information within operational timelines. This represents a shift from purely scientific analysis toward service-oriented products aligned with the needs of emergency response actors.

The applicability of this framework is illustrated through several recent emergency cases in Spain. Following the catastrophic 2024 Valencia DANA hydrometeorological event, ADA-derived information from EGMS data was integrated with cadastral datasets to identify buildings with pre-existing deformation patterns and to assess potential cascading effects associated with unstable slopes near reservoirs. Complementary SAR products, including high-resolution flood mapping datasets provided by commercial constellations, supported hydrodynamic modelling efforts and improved situational awareness. During the intense rainfall episodes in southern Spain (January–February 2026), EGMS time-series were analysed to identify pre-existing ground deformation in mountainous sectors. These areas experienced significant impacts, including the full evacuation of one village and the partial evacuation of another due to landslide hazards and concerns over possible karstic collapse. Exploratory comparisons between the temporal evolution of precipitation and deformation signals were conducted to investigate possible terrain responses to previous rainfall events. Additionally, Sentinel-1 interferograms were generated to evaluate localized post-event changes, revealing the limitations of regional-resolution InSAR products for small-scale instabilities and emphasizing the need for rapid access to high-resolution SAR acquisitions.

These real-world applications highlight key challenges for operational InSAR services, including data availability constraints, the need for workflows prepared prior to emergencies, and the importance of translating deformation measurements into prioritized actions for field teams. The RASTOOL-DoS initiative contributes to addressing these gaps by advancing toward scalable ground-motion services capable of supporting both preparedness and response phases within the Copernicus ecosystem.

Overall, this work demonstrates how EGMS-derived information, when integrated with automated ADA detection and standardized analytical pipelines, can support the transition from satellite-based deformation monitoring to operational ground-motion services tailored to emergency decision making.

Infrastructure monitoring is essential for assessing structural performance, detecting and mitigating potential damage before it evolves into hazardous conditions, and supporting early response strategies. Interferometric Synthetic Aperture Radar (InSAR) has become a widely applied tool for this purpose due to its spatial coverage, medium temporal sampling rate, and cost-effective operational capabilities. However, accurate interpretation of structural deformation requires appropriate physical modeling of phase contributions.

In this study, we investigate the Shezi Bridge in Taipei, Taiwan, an asymmetric cable-stayed structure connecting Shezi Island with the Beitou district across the Keelung River. We analyze ascending and descending Sentinel-1 acquisitions spanning 2014–2025, complemented by 60 high-resolution TerraSAR-X descending images covering 2017–2021.

Time series analysis was conducted using the single-look InSAR methodology implemented in SARvey. The standard phase model in SARvey retrieves DEM error and linear velocity from arc-based phase observations. We extend this model by incorporating a thermal expansion component, which is expected to be significant given the region’s high temperature variability and humidity.
Results reveal substantial differences between the original and extended phase models. The conventional model struggles to retrieve coherent points along the bridge deck, frequently classifying them as low coherent. In contrast, the extended model successfully captures thermally induced deformation patterns across the bridge structure.

Using the extended model, we have found an annual seasonal signal with amplitudes of around 10 mm. Results from ascending and descending orbits show similar amplitudes, however this seasonal signal features different phase shifts resulting in a signal with peaks that match summer time in the descending orbit and peaks matching winter time in ascending orbit. Descending orbit from high resolution TerraSAR-X dataset features the same patterns as Sentinel-1 descending orbit.
The discrepancy between ascending and descending orbits seems to be anchored by the bridge pylon and suggest a longitudinal deformation with respect to it.

These findings demonstrate that incorporating thermal expansion into InSAR phase modeling significantly improves deformation retrieval in large bridge infrastructures. Moreover, multi-geometry time-series analysis enables the resolution of subtle, temperature-induced structural responses with high sensitivity and reliability, highlighting the potential of advanced InSAR modeling for structural health monitoring applications.

Land subsidence has become a major environmental concern in Golestan Province, northern Iran, where intensive agricultural development and long-term groundwater extraction have altered subsurface hydrological conditions. The Gorgan Plain, dominated by cultivated land, has experienced progressive ground deformation over the past decade. However, assessment of long-term deformation dynamics is limited by sparse historical SAR acquisitions before 2014 and low coherence over agricultural surfaces. In this study, we investigate the spatial pattern and temporal evolution of land subsidence in Golestan from 2014 to 2024 using Sentinel-1 time series InSAR and examine its relationship with hydro climatic variability and available groundwater observations.

By combining ascending and descending Sentinel-1 observations, we derived vertical subsidence time series for the study area. The results reveal a broad, elliptical subsidence bowl centered within the Gorgan Plain. Maximum vertical subsidence rates reach approximately 15 cm/year, indicating substantial and sustained ground deformation during the observation period. The spatial continuity and persistence of deformation suggest a basin scale response consistent with regional groundwater withdrawal rather than localized structural instability.

To analysis temporal variability, the mean vertical displacement time series over the main subsiding area was decomposed using Seasonal Trend decomposition based on Loess (STL). This approach separates the signal into trend, seasonal, and residual components, allowing the long-term deformation tendency to be isolated from seasonal fluctuations. The temporal derivative of the extracted trend component was then computed to characterize changes in subsidence rate over time. This derivative based metric was compared with the Standardized Precipitation-Evapotranspiration Index (SPEI), which reflects hydro climatic conditions. The analysis yields a maximum correlation coefficient of approximately 0.6, indicating a meaningful relationship between climate variability and temporal modulation of subsidence rates. These results suggest that while long-term subsidence is primarily driven by sustained groundwater extraction, short-term to mid-term acceleration or deceleration of deformation is partially influenced by climate related recharge variability.

Groundwater level records from available monitoring wells were also examined. Although the number of wells is limited and some time series contain data gaps. Some wells show declining trends that are consistent with the subsidence patterns derived from InSAR. In contrast, several wells show relatively stable or locally rising water levels. Such heterogeneity may be related to differences in well depth, aquifer properties, or localized groundwater management practices. These observations indicate spatial variability in aquifer response across the plain.

By integrating decadal Sentinel-1 deformation time series with STL based temporal analysis, hydro climatic indicators, and in situ groundwater observations, this study provides a consistent interpretation of subsidence dynamics in a data limited agricultural basin. The results demonstrate the capability of C-band time series InSAR to quantify large magnitude subsidence and to resolve its temporal variability in semi-arid environments, supporting improved groundwater management and hazard assessment in the Golestan Province.

The identification and monitoring of surface deformation over hydrocarbon fields are essential for ensuring that the benefits of hydrocarbon exploration and production are achieved in harmony with environmental sustainability. In hydrocarbon-producing regions, ground deformation - particularly land subsidence - can pose significant risks to infrastructure, ecosystems, and local communities. Therefore, understanding the magnitude, spatial extent, and temporal evolution of surface deformation within production zones is crucial for effective resource management, environmental protection, and risk mitigation.

In this study, surface deformations around the hydrocarbon fields of the Niger Delta, Nigeria, are analyzed using a combination of Interferometric Synthetic Aperture Radar (InSAR) and Geographic Information System (GIS) techniques. The methodological framework consists of three main components. First, regional-scale deformation detection is carried out to generate a comprehensive deformation map across the Niger Delta oil-producing region. This step enables a detailed assessment of deformation patterns and rates across a broad spatial extent. Second, active deformation areas (ADAs) are identified to pinpoint zones most susceptible to subsidence and ground instability. Third, inventory integration and localized analysis are performed by combining a spatial database of known hydrocarbon fields with the derived deformation map within a GIS environment. This integration facilitates the spatial correlation of deformation patterns with specific production sites and operational activities.

To ensure the robustness and reliability of the results, the methodology will be validated by applying the same processing workflow to descending orbit geometry data. The comparison between ascending and descending geometries will allow for cross-verification of deformation signals and improve confidence in the interpretation of vertical and horizontal displacement components. This validation step enhances the methodological rigor and supports the consistency of the findings.

The results are expected to provide critical insights into the relationship between subsidence and hydrocarbon production activities in the Niger Delta. Such insights are vital for sustainable resource management, infrastructure planning, and the mitigation of environmental impacts in hydrocarbon-rich regions. Overall, the proposed methodology offers a reliable and transferable framework for policy-makers, regulators, environmental authorities, and industry stakeholders seeking to ensure that oil and gas exploration and production proceed in a manner that minimizes environmental and societal risks.

Interferometric Synthetic Aperture Radar (InSAR) is a remote sensing technique that enables precise large-scale displacement monitoring with accuracy comparable to terrestrial geodetic methods. However, the relative character of InSAR displacement measurements in the Line-of-Sight (LOS) direction poses challenges for achieving absolute geodetic accuracy. To address this limitation, collocation with complementary geodetic techniques, such as Global Navigation Satellite Systems (GNSS), is essential to tie InSAR observations to a globally recognized terrestrial reference frame.

The Slovak InSAR Corner Reflector Network (SKICRN) is an innovative geodetic infrastructure developed to integrate InSAR and GNSS measurements and to serve as an experimental framework. The SKICRN includes strategically deployed corner reflectors co-located with GNSS stations (part of the Slovak real-time positioning service (SKPOS) network) across Slovakia (21 sites), enabling detailed analysis of radar backscattering properties and spatial alignment.

The experimental setup also involved precise determination of the spatial vectors between GNSS and InSAR reference points, achieved through high-precision geodetic surveying techniques, including static GNSS observations and total station measurements. This collocation strategy not only enhances the reliability of displacement monitoring but also establishes a scalable framework for integrating InSAR with existing geodetic networks.

All stations were analyzed for their radar characteristics (SCR, RCS) and InSAR displacement time series. Selected stations were compared with the GNSS displacement time series. Processing of the GNSS data from the SKPOS network was performed using Bernese 5.4 software, following the latest Guidelines for EPN Analysis Centres. The SKPOS network solution is aligned with the EPN reference frame solution using a minimum-constraints condition at 12 EPN reference stations of class C0–C2. The results (cleaned daily XYZ coordinates) are estimated in IGS20/IGb20 and subsequently transformed to ETRS89 (ETRF2000) at the mid-epoch of the day.

A geodetic approach implemented in the GECORIS software package was used to process Sentinel-1 data. For each station, all available tracks (covering the period from CR installation to April 2026) were processed within the immediate vicinity (10 × 10 km) using a free-network solution, including tropospheric effects. All stations demonstrated robust backscatter properties, with SCR values exceeding the threshold required for reliable InSAR processing.

The results indicate that the SKICRN improves the geodetic precision of InSAR measurements by transforming relative LOS displacements into a terrestrial reference frame. This research demonstrates the viability of the SKICRN as a foundation for geodetic innovation, highlighting its potential for regional- and national-scale implementation in Slovakia and beyond.

During the last decade, wide area monitoring of infrastructure with InSAR has seen tremendous progress. A large number of case studies, proposals for regional monitoring systems for diverse tasks and first operational monitoring services have appeared. In particular, the monitoring of linear infrastructure (roads and train tracks) is of high relevance. With regard to the coverage of roads with InSAR measurements, the joint use of PS and DS has been proven to be significantly superior to only using PS. Here, we will show that in X-band and to a lesser degree in C-band there is room for distinct qualitative improvements in estimating DS signals on asphalt surfaces by using Temporary Coherent DS (TCDS).

Complementary to this submission, a second submission to Fringe 2026 has been made (A study on Temporary Coherent Distributed Scatterers on roads and runways on the level of single scatterers, A. Fingerle, M. Even, A. Seidel, H. Kutterer), that studies Temporary Coherent Distributed Scatterers on roads, runways and train tracks on the scatterer level. It contrasts DSC appearances on asphalt and train tracks regarding e.g. bandwidth, grouping criterion and surface properties and discusses the impact of these factors on grouping and several quality numbers.

Our approach for TCDS is based on a new closure phase-based matrix and the phase triangulation matrix. Beside adding another type of information, the closure phase-based matrix improves the discernability between acquisitions for which the DS in question has low-quality signal and acquisitions with high quality. The phase triangulation matrix measures how well the estimated DS phase history is able to reproduce the phases of the coherence matrix. Specifically in the case of asphalt surfaces, a simple approach for discerning quality of acquisitions is successful. It is based on the observation, that asphalt surfaces in X-band posses low to medium but persistent coherence values. Only for certain acquisitions, coherences and values of are close to zero, which presumably is caused by altered backscattering properties because of water or snow. In order to detect such acquisitions, we derive two types of quality numbers from the columns of above matrices. Simple thresholding on these two numbers allows to discard acquisitions (specific for each DS candidate) with low quality signal.

In order to assess our approach, beside the quality numbers phase triangulation coherence and a measure of phase closure consistency, the number of discarded acquisitions and the temporal coherences of point pairs are investigated. We are confident that this approach can be refined for application in cases, where has a block-structure (often the case on train tracks).

During the last decade, wide area monitoring of infrastructure with InSAR has seen tremendous progress. A large number of case studies, proposals for regional monitoring systems for diverse tasks and first operational monitoring services have appeared. In particular, the monitoring of linear infrastructure (roads and train tracks) is of high relevance. With regard to the coverage of roads with InSAR measurements, the joint use of PS and DS has been proven to be significantly superior to only using PS. Here, we will present preparatory investigations to identify Temporary Coherent DS (TCDS) in X-band and in C-band on asphalt surfaces.

Based on our findings, we will show in a second submission to Fringe 2026 (Even et al. Temporary Coherent Distributed Scatterers on roads and runways) that TCDS in X-band and to a lesser degree in C-band allow for distinct qualitative improvements for displacement analysis on asphalt surfaces.

In order to develop an approach for TCDS, we studied the coherence matrix, a new closure phase-based matrix, the residual phase angle matrix and the phase triangulation matrix. Beside adding another type of information, the new closure phase-based matrix improves the discernability between acquisitions for which the DS in question has low-quality signal and acquisitions with high quality. The residual phase angle matrix and the phase triangulation matrix measure how well the estimated DS phase history is able to reproduce the phases of the coherence matrix. Specifically in the case of asphalt surfaces, a simple approach for discerning quality of acquisitions is successful. It is based on the observation, that asphalt surfaces in X-band posses low to medium but persistent coherence values. Only for certain acquisitions, coherences and values of entries of the closure phase-based matrix are close to zero, which presumably is caused by altered backscattering properties because of water or snow. In order to detect such acquisitions, we derive quality numbers for each acquisition from the columns of the named matrices. The simplest approach to identify acquisitions (specific for each DS candidate) with low quality signal is thresholding on these numbers. We investigate their suitability for this purpose for many examples of asphalt surfaces of different width and variability in mean amplitude with X- and C-band. Setting thresholds based on e.g. closure phase consistencies proves to be easier than using coherence because they do not exhibit the pronounced dependence on magnitude of coherence and neighborhood size. In addition, we study the influence of the grouping method on the selection of the statistically homogenous pixels and on the quality numbers.

Our findings show that the proposed approach works well on asphalt surfaces. Beyond that, we are confident that this method can be refined for application in cases, where the new closure phase-based matrix has a block-structure (often the case on train tracks).

Multi-track Interferometric Synthetic Aperture Radar (InSAR) methods hold significant potential for three-dimensional (3-D) deformation monitoring. However, both right-looking-only and joint right–left-looking combinations remain susceptible to noise amplification, particularly in the weakly constrained North-South (N–S) component.While regularization is an effective tool to suppress noise, traditional empirically tuned or single-parameter schemes often struggle to balance bias and variance across all three components simultaneously.To address this, this study proposes a spectrum-block-based regularization parameter-selection method. The proposed method analyzes the spectrum of the 3-D normal equations to identify the dominant noise-amplification mode for a given combination of observations. It then applies spectrum-block-based regularization with separate strengths for the strong subspace and the weak component, stabilizing the weak component while limiting bias in the others. Mogi simulations show that, compare with least squares (LS) and the single-parameter L-curve method, the proposed method substantially reduces component-wise RMSE under ill-conditioned observation combinations. Furthermore, for the eruption of the Hayli Gubbi volcano in November 2025, the joint results of the 3-D decomposition of NISAR - Sentinel-1 in the right-left look keep the components E-W and U consistent with LS while effectively suppressing high-frequency noise in the N–S component. Thereby highlighting that the proposed method improves the robustness of 3-D deformation reconstruction under ill-conditioned geometries and weak-constraint scenarios.

Limited attention is typically paid to the cause-effect relationship between land subsidence due to aquifers overexploitation in expanding metropolises and urban growth models and patterns. This paper implements an integrated urban and satellite Interferometric Synthetic Aperture Radar (InSAR) approach to investigate subsidence, multi-decadal urban growth and peopling trends in the Metropolitan Area of Morelia (ZMM) in the Mexican state of Michoacán. Stacking of JRC’s Global Human Settlement Layer, DLR’s World Settlement Footprint and INEGI’s National Geostatistical Framework datasets revealed a predominant edge-expansion growth model, with urban densification in 1975–2020 and some sprawling in 1990–2000. Population of the ZMM doubled in the last 30 years, reaching over 1 million inhabitants. The ENVISAT and Sentinel-1 InSAR analysis confirms that subsidence is structurally-controlled by the main normal faults within the Cuitzeo half-graben. Differential sinking and ground discontinuities are aligned with buried tectonic faults and contrasting compressible sediment thickness. Non-linearly deforming subsidence bowls develop at extraction wells in both old and newly urbanized sectors of the ZMM. Maximum vertical displacement velocities increased from –2.5 (2003–2010) to –9.0 (2014–2021) cm/year, with subsidence migrating towards recently urbanized zones. More than 250 new groundwater wells were added to the public registry since 2000, many of which within new urban sectors. Time-lapse InSAR reveals a 4 km² rapidly subsiding bowl that formed at the largest social housing neighbourhood of Villas del Pedregal, as building lots were progressively completed and sold, and new wells registered. With angular distortions due to the differential subsidence reaching 0.12% in 2014–2021, new buildings and roads are exposed to fracturing and surface faulting risk of comparable level as the city historic building blocks located along the main faults. By providing useful insights into the relationship between urban growth and land subsidence in the ZMM, the approach proves valuable for application to other metropolises worldwide.

Full paper: Cigna F., Tapete D. 2022. Urban growth and land subsidence: Multi-decadal investigation using human settlement data and satellite InSAR in Morelia, Mexico. Science of The Total Environment, 811, id.152211, doi:10.1016/j.scitotenv.2021.152211

Over recent decades, glacier retreat in the European Alps has accelerated dramatically due to climate change, exposing extensive glacier forefields that are undergoing rapid geomorphological transformation. These newly deglaciated areas are characterised by unstable sediments, degrading dead-ice bodies, reworked moraines, and increased slope activity. As a result, the spatial distribution and intensity of geomorphological processes are changing, posing emerging risks to alpine infrastructure such as mountain huts, hiking trails, and access routes. Continuous and spatially comprehensive monitoring of these dynamic environments is therefore essential, particularly under accelerating climate warming and increasing geomorphic instability worldwide today.

Synthetic aperture radar (SAR), and particularly interferometric SAR (InSAR), offers unique advantages for monitoring high-mountain terrain, as it enables weather-independent, large-scale, and repeatable measurements of surface displacement. In this contribution, we present a radar-based investigation of the glacier forefield of Taschachferner in the Ötztal Alps (Austria), integrating multi-temporal Sentinel-1 InSAR time series analysis with detailed geomorphological mapping.

Surface deformation rates are derived from Sentinel-1 C-band data using multi-temporal interferometric processing. The resulting velocity rates are analysed in relation to geomorphological units identified from optical imagery and field-based interpretation, including lateral and terminal moraines, debris cones, rock glaciers, and debris-covered glacier remnants. Attention is paid to spatial variations in deformation across different geomorphological features.

The InSAR results reveal heterogeneous deformation patterns within the glacier forefield, including zones of enhanced subsidence likely related to dead-ice melt, as well as localised displacement signals associated with sediment redistribution and slope processes. Stable sectors, in contrast, show minimal displacement and persistent coherence. By comparing radar-derived kinematic information with geomorphological interpretation, we demonstrate how InSAR can refine the delineation of debris-covered dead-ice areas and identify sectors with anomalous deformation rates that may require intensified hazard monitoring.

The study highlights the added value of combining geomorphological expertise with radar-based surface motion analysis to better understand landscape dynamics in rapidly evolving alpine terrain. Beyond scientific insights, our approach provides a transferable framework for monitoring of recently deglaciated areas. Sentinel-1 time series analysis thus represents a powerful tool for supporting hazard assessment and adaptive management strategies in high-mountain environments increasingly affected by climate-driven change.

Ensuring the structural integrity of dams is a critical challenge in civil engineering, particularly as these infrastructures age and face increasing stress from environmental and anthropogenic factors. While traditional monitoring methodologies, such as in-situ sensors and geodetic surveys, offer 8high sensitivity, they are often limited by high costs, sparse spatial coverage, and labor-intensive operations, which struggle to capture large-scale, complex deformation patterns. To address these limitations, the DARTWIN project proposes a transformative approach by integrating satellite-based remote sensing, physics-based numerical modeling, and artificial intelligence (AI) within a Hybrid Digital Twin (HDT) framework.

The core of the DARTWIN methodology relies on the refinement of Multi-Temporal InSAR (MT-InSAR) techniques to provide continuous, millimeter-scale deformation monitoring. A significant challenge in dam surveillance is the detection of rapid or non-linear deformations, such as those associated with primary consolidation, rapid reservoir drawdown/filling, or episodic slope instabilities. Current regional services, such as the European Ground Motion Service (EGMS), often struggle to capture these fast-moving or complex behaviors due to constraints in temporal resolution and observation windows. DARTWIN overcomes this by leveraging a synergistic use of multi-mission SAR data, combining high-resolution X-band data from the PAZ and TerraSAR-X missions with extensive C-band archives from Copernicus Sentinel-1. This multi-frequency approach, processed through advanced tools, allows for more reliable monitoring in diverse environments, including data-scarce scenarios.

A key innovation of the project is the development of an AI-driven layer designed to automatically detect and classify anomalous movements within MT-InSAR time series. By employing machine learning algorithms such as Long Short-Term Memory (LSTM) networks and Random Forests, the system can distinguish between expected seasonal oscillations and accelerating structural anomalies, providing a predictive rather than reactive monitoring strategy.

Furthermore, to bridge the gap between high-fidelity simulations and real-time monitoring, DARTWIN implements Reduced-Order Models (ROMs) derived from Finite Element Method (FEM) simulations. While traditional hydro-mechanical FEM models are computationally intensive, ROMs allow for simplified yet accurate simulations that can be updated in near real-time using observational data. These models are integrated into the HDT, creating a continuously updating digital replica of the dam's behavior. Finally, all results are delivered through a geospatial viewer, an interactive platform that facilitates data visualization and decision support for dam operators and Water Authorities, ensuring the long-term resilience of critical water cycle infrastructures.

Groundwater has been over-pumped and excessive use during the past decades due to the lack of sufficient surface water caused by rapid economic developments and growing population in the central Taiwan. The alluvial fan of the Cho-Shui River in Western Taiwan suffers the most serious land subsidence hazard with a maximum subsidence rate in excess of 3 cm/yr which affect the transportation infrastructures across the land subsidence area. The long-term spatial land subsidence variation from 1995 to 2024 reveals that the center of land subsidence area changed significantly with time from the coastal area to inland area. The decreasing of land subsidence could obviously be detected by the velocity profile along the Taiwan High Speed Rail during different time periods from geodetic measurements. Not only the vertical displacement increases the risk on potential damage on transportation infrastructures across the land subsidence bowls but also the induced additional horizontal displacement owing to the vertical subsidence could result in the unexpected risk for the infrastructure. In this study, we used the multi-temporal InSAR to calculate the vertical deformation and the east-west deformation combined with the velocity field of the ascending and descending orbits. Three large-scale subsidence bowls are detected and accompanied by maximum additional horizontal deformation of ~8 mm/yr than that predicted by tectonic movement outside of the subsidence bowl. This additional E-W displacement is the major risk concerns of the N-S trending Taiwan High Speed Rail. In addition, we also use hydro-geological data from several monitoring wells to construct numerical modeling to discuss the deformation patterns in different aquifer and aquitard system for the land subsidence induced horizontal deformation.

Hydrotechnical infrastructure, including flood protection dams, plays a huge role within critical infrastructure systems, climate change adaptation strategies, and the management of extreme precipitation events. The long-term operation of these structures, cyclical hydrological loads, and increasing flood frequency require systematic, precise monitoring of displacements. Satellite radar interferometry enables continuous monitoring by acquiring spatially information on deformation. It is an ideal complement to in-situ measurements, which provide more precise point-based monitoring. Importantly, these methods are non-invasive and do not interfere with the structural integrity of hydrotechnical facilities. However, monitoring earthen dams and reservoir embankmentsremains challenging due to signal decorrelation caused by vegetation cover and temporal variability in soil moisture conditions.

This study presents a Persistent Scatterer InSAR (PSInSAR) analysis of selected retention reservoirs in Poland located in the Odra River basin (an area affected by the 2024 flood event)—including reservoirs: Racibórz Dolny, Otmuchów, Nysa, and Kozielno, Stronie Śląskie as well as Świnna Poręba in the Upper Vistula basin. The analysis was conducted using C-band SAR data from the Sentinel-1 mission covering the period 2022–2025. An integrated processing scheme combining Differential InSAR (DInSAR) and PSInSAR was applied to enhance deformation detectability across heterogeneous structural components. Particular attention was given to coherence threshold optimization and to the separation of seasonal signals in partially vegetated areas.

A three-year InSAR time series enabled the estimation of linear displacement velocities and the identification of seasonal components. For the majority of the analyzed structures, no statistically significant long-term deformation trends were observed. The estimated displacement amplitudes were approximately 5 mm, which is consistent with the expected measurement uncertainty of the C-band PSInSAR approach.
In contrast, for the Racibórz Dolny Reservoir, localized clusters of persistent scatterers exhibiting cumulative vertical displacements of up to –30 mm (approximately –9 mm/year) were identified within the northern sector of the embankment.Time-series decomposition indicates predominantly linear behaviour with minor seasonal modulation related to hydrological loading cycles rather than structural instability.

The results are intended for integration into the National Satellite Information System (NSIS – https://nsisplatforma.polsa.gov.pl/?lang=en), operated by the Polish Space Agency (POLSA), enabling the operational dissemination of InSAR products to stakeholders responsible for hydrotechnical infrastructure monitoring and safety.

The results confirm that Sentinel-1–based PSInSAR enables millimeter-level deformation detection in large hydrotechnical systems, supporting scalable satellite-based early warning frameworks and enhancing infrastructure resilience under changing climatic conditions.

The product was developed on behalf of the Polish Space Agency (POLSA) by the Institute of Geodesy and Geoinformatics at the Wrocław University of Environmental and Life Sciences (UPWr).

As climate change, urban expansion, aging infrastructure, and resource development intensify global geotechnical challenges, the demand for scalable and cost-effective monitoring solutions continues to grow. Interferometric Synthetic Aperture Radar (InSAR) offers a non-invasive, globally applicable approach with significant potential for risk reduction and resilience enhancement.

InSAR as a method, is used for at least two decades now for monitoring ground deformation and assessing infrastructure stability from space, using radar satellite data, and has applicability that can vary from regional to global scales. The millimeter-scale precision detection of surface displacements combined with the unparalleled spatial coverage, long-term historical analysis, and cost-effective monitoring compared to conventional ground-based techniques, make it an ideal solution for covering the deformation monitoring needs in several sectors. The rapid expansion of public and commercial SAR satellite missions has generated extensive multi-decadal archives, allowing both retrospective and near-real-time deformation analyses almost anywhere on Earth.

The large spatial footprint of SAR imagery enables continuous monitoring over hundreds of square kilometers, offering a distinct advantage over localized ground instruments. This broad coverage supports the detection of both localized instabilities and regional deformation trends. Historical satellite archives allow further long-term structural integrity assessments of critical infrastructure. As satellite revisit frequencies increase, the potential for operational and near-real-time monitoring continues to grow. enable reliable deformation monitoring across diverse landscapes. InSAR achieves high-precision deformation measurements even in complex environments.

In 2024, a dedicated and scalable commercial InSAR service, Inio, was launched as a result of the collaboration between Kongsberg Satellite Services (KSAT) and the Norwegian Geotechnical Institute (NGI). This partnership combines NGI’s geotechnical engineering expertise with KSAT’s global satellite data access and ground station infrastructure, delivering an integrated service that transforms satellite-derived deformation measurements into actionable engineering insight. Inio is designed to enhance ground stability assessments, mitigate operational risks, and support informed decision-making across multiple sectors worldwide.

Inio applies these capabilities across transportation, water management, mining, energy, and urban development sectors. Bridges, roads, railways, tunnels, hydroelectric dams, and tailings storage facilities can be systematically monitored for subtle displacement that may indicate emerging instability. In urban environments, InSAR enables large-scale subsidence mapping, supporting municipal planning and infrastructure resilience. In mining and industrial contexts, deformation monitoring enhances risk management while reducing the need for continuous field-based instrumentation.

Inio will be presenting examples of various cases worldwide, using both open source and commercial data, of different resolutions, to prove the InSAR’s effectiveness in environmental and geohazards monitoring, and assess instabilities and deformation, combined with geotechnical analysis. For cases where geotechnical and engineering monitoring to detect unstable ground was critical in construction and infrastructure, in order to manage and assess stability and identify hazardous movements early, subtle surface movements over large areas have been detected. Additionally, timely and actionable information was provided, supporting disaster preparedness.

With multi-temporal InSAR analysis for mapping deformation patterns across the affected areas, we are able to identify active zones of displacement. The spatial continuity of satellite observations is providing comprehensive coverage beyond what ground-based measurements alone could achieve, demonstrating the technology’s value in remote and hazard-prone regions.

Beyond traditional geotechnical applications, Inio supports emerging energy-sector needs, including Enhanced Oil Recovery (EOR) and Carbon Capture, Utilization, and Storage (CCUS). Fluid injection and extraction can induce measurable surface deformation due to subsurface pressure changes. Continuous InSAR monitoring allows operators to track subsidence or uplift, verify reservoir behavior, and detect anomalies that may signal geomechanical instability. Integrating deformation data with geological and reservoir models enhances environmental safety, regulatory compliance, and operational resilience.

The defining strength of Inio lies in its multidisciplinary framework. Satellite-derived deformation data are interpreted alongside geological information, digital elevation models, engineering parameters, and historical records. This integration ensures that displacement measurements are contextualized within a robust geotechnical framework, reducing uncertainty and supporting practical mitigation strategies. Rather than delivering raw remote sensing outputs alone, Inio provides comprehensive assessments of ground stability and associated risks. The launch of Inio marks a significant step in operationalizing advanced InSAR methodologies within a commercial, engineering-oriented service model.

InSAR has evolved into an indispensable global technology for ground stability assessment and geohazard mitigation. Through the combined expertise of KSAT and NGI, Inio maximizes the value of satellite radar data by integrating technical precision with engineering interpretation. This global service enhances safety, safeguards investments, and supports sustainable infrastructure and energy development in an increasingly complex and dynamic world.

Monitoring and understanding surface deformation associated with subsurface energy storage is becoming increasingly important as underground gas, CO₂, and hydrogen storage expand in scale and strategic relevance. Traditional monitoring methods these activities struggle to correlate surface deformations with deep subsurface changes, which are critical for ensuring safety, managing pressure changes, and preventing incidents in subsurface storage facilities Synthetic Aperture Radar Interferometry (InSAR) offers spatially dense measurements of surface displacement, but interpreting these observations in terms of subsurface processes requires physically based modelling. This project demonstrates a combined geomechanical - InSAR workflow using open data from the Yela Underground Gas Storage (UGS) facility in Spain as a test case, aiming to assess the value of integrated deformation monitoring for operational storage sites.

We processed 190 Sentinel‑1 scenes (2020–2025) using Enhanced Persistent Scatterer (EPS) and Enhanced SBAS (ESBAS) workflows, comparing their performance in an area dominated by farmlands and sparse shrublands. The EPS workflow provided the most reliable results, generating stable deformation time series with low noise outside agricultural zones. Despite displacement magnitudes being small (within ±5 mm), the time series above the Yela reservoir revealed a clear seasonal signal corresponding to the annual gas injection and withdrawal cycle.

A 2D axisymmetric geomechanical, numerical model was developed in Abaqus to simulate vertical and horizontal deformation associated with seasonal reservoir pressure changes. Sensitivity analyses showed that reservoir and underburden stiffness, together with pressure variation, exert the strongest control on surface deformation. Model results reproduced the observed annual uplift–subsidence pattern but overestimated the amplitude, reflecting uncertainties in geomechanical input data, reservoir geometry, and operational timing. This underscores the potential for InSAR to refine model calibration when accurate injection schedules and geomechanical information are available.

By comparing modelled and observed deformation at varying distances from the facility, the study demonstrates how InSAR can be used not only for detection of seasonal deformation, but also for identifying anomalous behaviour, informing parameter calibration, and improving confidence in subsurface monitoring. The project highlights the need for integrated workflows that translate InSAR displacement information into mechanical Earth model (MEM) inputs—a critical missing step for future operational monitoring of CO₂ and H₂ storage.

This work provides a foundation for developing automated, scalable monitoring systems that combine satellite observations, computational analytics, and geomechanics.

Acknowledgement: This work has been funded by the Norwegian Space Agency under grant JOR25032.

Mining-induced ground subsidence is one of the most severe long-term geohazards in Poland, posing a direct and growing threat to public safety, critical infrastructure, and sustainable land use. In regions affected by intensive underground mining, annual subsidence rates locally exceed one meter, resulting in progressive deformation of the ground surface. Such rapid and large-magnitude ground movements lead to damage and failure of linear infrastructure such as roads, railways, pipelines, and power lines, as well as to structural degradation of residential, industrial, and public buildings. The cumulative effects of long-term mining activity significantly increase the vulnerability of densely populated and industrialized areas, making reliable subsidence monitoring a key component of risk management and spatial planning.

This study presents an integrated approach combining satellite-based Interferometric Synthetic Aperture Radar (InSAR) measurements from the European Ground Motion Service (EGMS) with differential digital terrain models derived from airborne LiDAR data to improve the assessment of mining-related deformation hazards. The research focuses on three Polish regions experiencing the most significant and persistent subsidence: the Legnica–Głogów Copper District (LGCD), the Upper Silesian Coal Basin (USCB), and the Lublin Coal Basin (LCB). These regions represent different geological settings and mining histories but share a common challenge of intense and long-lasting surface deformation.

In areas characterized by extremely high deformation rates, EGMS InSAR observations primarily delineate the outer boundaries of mining influence zones, where ground motion remains within the physical limits of the interferometric method. The most hazardous deformations, typically located in the central parts of subsidence basins, are often not detected due to phase decorrelation and velocity saturation. As a result, the zones with the highest rates of surface lowering—frequently exceeding one meter per year and posing the greatest risk to infrastructure and buildings—are insufficiently represented in satellite-based deformation products.

To overcome these limitations and to fully exploit freely available geospatial data, the Polish Geological Survey applies differential airborne LiDAR-derived terrain models to quantify large-magnitude surface subsidence. By comparing high-resolution digital terrain models acquired at different time intervals, it is possible to directly measure rapid and high-amplitude vertical displacements that are invisible to InSAR techniques. These LiDAR-based differential models provide detailed information on the geometry and depth of subsidence troughs and allow for accurate identification of zones of maximum deformation.

The work presents a methodology for integrating EGMS InSAR measurements with LiDAR-derived differential models to reconstruct complete and continuous deformation fields. The combined dataset captures both small-magnitude, spatially extensive ground motion detectable by InSAR and extreme subsidence occurring in the centers of mining-induced depressions. This integrated approach significantly improves the reliability and completeness of deformation mapping and provides a more realistic representation of mining-related geohazards.

The results demonstrate that data fusion is essential for effective geohazard identification, risk assessment, and land-use planning in mining-affected regions. The research is conducted as part of statutory activities of the Polish Geological Survey and supports national geohazard monitoring, crisis management, and decision-making processes at local and regional administrative levels, with the ultimate goal of enhancing public safety in areas impacted by long-term mining activity.

The Świnna Poręba (Mucharskie) Reservoir in southern Poland, constructed between 1986 and 2017 on the Skawa River, is one of the longest-running hydrotechnical investments in Europe. Geologically, the reservoir is located within the Silesian and Magura nappes of the Outer Carpathian flysch belt, characterized by steep relief, low-permeability lithologies, complex tectonics, and interbedded sandstones and shales. These conditions, combined with shoreline undercutting, reservoir level oscillations, and wave erosion, predispose the area to slope instability and recurrent mass movements. Previous investigations under the Polish Landslide Counteracting System (SOPO) have primarily emphasized geological and hydrogeological controls, while anthropogenic drivers remain comparatively underexplored.

This study proposes a multiproxy framework to disentangle natural and anthropogenic factors influencing slope stability around the reservoir. The research integrates:

• Multitemporal InSAR time-series analysis (Sentinel-1 interferometry data cubes) to detect millimetric ground deformation trends,
• Correlation of displacement signals with reservoir water level fluctuations and hydro-meteorological records,
• Optical and active remote sensing for land-use, water-level fluctuations and vegetation change detection,
• UAV-based photogrammetric models (high-resolution DEMs and digital outcrop models) for selected landslides and rockfall sites,
• Geological mapping and structural analysis,
• Geophysical investigations (ERT and conductivity profiling),
• Analysis of archival geotechnical documentation and landslides documentation.

The core objective is to determine whether slope activation is primarily controlled by litho-structural conditions and hydrological forcing, or whether anthropogenic factors, such as reservoir-induced water level oscillations, historical excavation, shoreline modifications, and land-use changes - play a dominant role in reactivation dynamics.

The use of multitemporal InSAR (also transformed to data cubes) enables continuous deformation monitoring over multi-annual timescales and allows synchronization of displacement acceleration phases with rapid drawdown events, seasonal hydrological variability and extreme precipitation. This approach supports quantitative assessment of reservoir-induced slope response mechanisms, including delayed pore-pressure dissipation and cyclic weakening.

All the collected data and results may one day serve as a foundation for a Digital Twin of the reservoir slope system, integrating deformation monitoring with hydrogeological model and geological structure. Such a digital framework could support hazard assessment, spatial planning decisions, and infrastructure management along the reservoir shoreline.

The proposed methodology aligns with ESA’s Earth Observation objectives by demonstrating how Copernicus radar time-series data, integrated with multidisciplinary datasets, can enhance understanding of complex landslide processes in flysch environments affected by anthropopressure and large hydrotechnical infrastructure.

The “Romanian Ground Motion Pilot Service for Sustainable Infrastructure” (RO-GMS) addresses the Earth Observation (EO) market in Romania with the primary objective of facilitating the operational uptake of Persistent Scatterer Interferometry (PSInSAR) by public authorities. By leveraging the Copernicus European Ground Motion Service (EGMS), the project delivers tailored, actionable monitoring services for critical local infrastructure.

A central component of this initiative is a customized, user-friendly web platform (https://pstool.terrasigna.com/) engineered to lower the barrier to entry for users with limited remote sensing expertise. The environment hosts specialized analytical tools, including an interactive transect tool that enables PSInSAR results analysis and animations across target areas. Additionally, an on-the-fly analysis module provides immediate stability assessments at the scale of individual infrastructure elements uploaded by the user.

To enhance data accuracy, the project introduces advanced methodological improvements. Firstly, the standard PSInSAR processing algorithm was refined to successfully detect and map quick, highly non-linear ground dynamics. Secondly, we integrated a novel thermal layer, utilizing thermal sensitivity models calculated from EGMS data to separate temperature-correlated displacements from actual structural motion.

The efficacy of these advancements is demonstrated through two distinct case studies. The first case study highlights a bridge subjected to large-scale dynamic shifts related to thermal motion. The thermal models estimated from long term EGMS data and local weather information can be used to improve phase unwrapping in short time monitoring projects on areas with large thermal motion.

The second case study examines a water dam that underwent rapid subsidence triggered by a significant drop in the reservoir's water level. The Sentinel-1 derived PSInSAR results for the water dam were rigorously validated against in-situ terrestrial measurements spanning from 2017 to 2024. This ground-truth validation relied on a dedicated geodetic monitoring system, comprising a crest subnetwork and markers installed to measure quarterly the behavior of the surrounding rocks, using high-precision total stations and leveling instruments. The strong correlation between these geodetic field campaigns and the satellite observations underscores the reliability of the RO-GMS platform as a robust, operational tool for sustainable infrastructure management.

Part of the presented results were obtained within the ESA project “Verification of innovative applications integrating national InSAR capabilities and the European Ground Motion Service” (2025 – 2026).

Distributed crustal deformation across the Philippines, given its plate boundary setting, indicates fast fault slip rates that translate to high seismic hazard coming from several possible earthquake sources. Further, ongoing rapid urbanization introduces anthropogenic ground deformation that exacerbates impacts of existing hazards. Advancement of knowledge on the behavior of individual earthquake generators and regional tectonics has been hampered, however, by limited observations from ground instrumentation and a tropical setting that can be challenging for InSAR analysis. Recognizing the gap, the Philippine national government funded the Leveraging InSAR for Observation and modeling of Earthquake generators (LInOG; lee-nog) Project to move towards a systematic ground deformation mapping system with InSAR for the Philippines. In this work, we use available L-band data to identify the seismogenic potential of major branches of the Philippine Fault in Luzon and reveal other sources of ground deformation. We process ALOS PALSAR Fine Beam resolution L1.1 (SLC) data from 2007–2011 using ISCE2 (from coregistration to interferogram formation) and use MintPy for time-series analysis (assuming linear velocity function). Initially covering ~20,000 sq. km of the island of Luzon, we find evidence of significant localized subsidence in developing urban centers, with more than ~5 cm/yr of subsidence. An east-west long-wavelength ramp can be seen in the InSAR velocity field that may correspond to regional shortening due to compression of the Philippine Mobile Belt between the Philippine Sea Plate and Eurasia; however, the gradient exceeds the expected ~1.4 cm/yr E-W shortening measured from GNSS. This indicates that careful ramp removal should be considered in InSAR data processing in a rapidly deforming tectonic environment. Better coherence is found compared to C-band InSAR results, though thickly vegetated mountainous areas do exhibit low temporal coherence that should be masked. Distinguishable ionospheric signals are also present in specific dates of the time-series. We assess alignment of the InSAR velocity data with the local reference frame, as well as the effectiveness of higher-resolution digital elevation models for topography-based corrections. The insights from this work will inform upcoming efforts that aim to incorporate newer L-band SAR datasets (NISAR and ALOS-4) that could potentially offer higher temporal resolution, alongside with local tropospheric and ionospheric datasets. Further work is also expected to be done to translate the results to be useful for localized understanding and operational use.

The closure of deep underground mines induces a gradual rise in groundwater levels, triggering a series of geomechanical processes observable at the surface. These processes commonly manifest as continuous deformations, including ground uplift, and discontinuous deformations, notably sinkholes, which typically develop in areas of historical shallow mining. Discontinuous deformations present a considerable risk to infrastructure and to the safety of communities residing in post-mining areas. In recent years, particularly across Europe, the rate of underground mine closures has increased, highlighting the need for effective surface deformation monitoring, improved understanding of post-mining dynamics, and informed strategies for preventive planning and sustainable land management.

This study examines the Olkusz–Pomorzany zinc and lead mining district in southern Poland, where extraction ceased in 2021. Since mine closure, numerous sinkholes, as well as zones of surface uplift and localized subsidence, have been documented, impacting forests, agricultural lands, and infrastructure. The site presents a particularly complex case due to its long mining history, dating back to the 13th century, encompassing diverse extraction methods, including shallow and partially undocumented workings. Further complexity arises from the geological framework, comprising fractured and locally karstified carbonate rocks overlain by unconsolidated Quaternary deposits of highly variable thickness.

The study aimed to identify potential precursors of sinkhole formation using satellite-based observations. In the first stage of research, variations in the C-band backscatter coefficient, the Moisture Index (MI), and the Normalized Difference Vegetation Index (NDVI) were analyzed. Breakpoint analysis of a 26-month pre-event time series revealed a common structural change in mid-2021, approximately six months prior to the first recorded sinkhole, coinciding with a rapid rise in groundwater levels. Chow tests confirmed statistically significant differences in regression coefficients across the identified breakpoint. In the second stage of research, satellite radar interferometry (InSAR) was employed to analyze a one-year observation period in 2024. This analysis delineated zones of potential displacement indicative of emerging deformation fields associated with sinkhole initiation. Comparison of these zones with discontinuous deformations observed in 2025 demonstrated the high efficacy of InSAR in detecting areas prone to sinkhole-related surface movements. This study enhances our understanding of aquifer system deformation mechanisms in post-mining areas. The results also allow to identify potential precursors associated with sinkhole formation.

The highly developed Dinaric karst systems are emblematic of a particularly well-expressed karst morphology. They feature complex hydrological networks in which water flows primarily underground through fractures towards springs (e.g. Milanović, 2015). Understanding these systems is crucial for hydroelectric engineering, on which the countries of the Adriatic coast strongly depend, but also for water supply to cities and agriculture in these heterogeneously populated regions that must accommodate high seasonal tourist demand.

Water transfer through karst depressions and fractures towards discharge areas may induce measurable deformation of the Earth’s surface, as suggested by e.g. Silveri et al. (2019) or Lesparre et al. (2016). During periods of substantial recharge, on the one hand, overpressure within fractures, porous or clay-rich formations generates uplift and outward horizontal deformation, while, on the other hand, water loading induces subsidence and horizontal motion towards the added load. The relative contribution of these two water-driven deformation mechanisms depends on the local hydrological setting and mechanical properties of the subsurface. In this context, the development of hydrogeodetic approaches based on InSAR, offering dense spatial coverage, spatial continuity and multi-year monitoring capability (e.g. Chaussard et al., 2017), provides a powerful tool to study karst hydrogeology. Though, it remains challenging for InSAR to measure low amplitude deformations related to hydrology, whose spatio-temporal patterns remain poorly constrained, particularly in regions with high topographic gradients and where ground-based independent data for validation and comparison are scarce.

In this study, we use Sentinel-1 InSAR time series processed using the CNES-FormaTerre FLATSIM service (Thollard et al., 2021) over the eastern Adriatic (Croatia, Bosnia, Montenegro, part of Albania, Slovenia, and Serbia), over the 2014-2021 period, from 2 descending and 2 ascending tracks. We perform a parametric decomposition of each time series in the Line of Sight (LOS), separating linear trends, coseismic signals (displacement steps for the largest magnitude earthquakes), and seasonal variations (sum of a sine and a cosine terms). Focusing on the seasonal component, we first reference the sine and cosine amplitude to a null-seasonal-deformation polynomial surface. This enables to extract, in a regionally and from track to track consistent way, areas where seasonal transients are significant. We then analyse the characteristics of the amplitude and timing (phase) of these seasonal displacements, and investigate potential hydrology-driven controlling mechanisms.

The seasonal signal reveals spatial correlations with known hydrological features. Seasonal amplitudes reach several centimeters, with maxima mainly located in karstic depressions (“poljes”) filled with Tertiary sediments. Phase analysis reveals a bimodal behavior: one group exhibits maximum motion towards the satellite at the end of the winter (coeval with peak groundwater level), in anti-phase with the other group. Decomposition into vertical and horizontal components reveals specific spatial patterns of horizontal extension in several poljes, associated with significant uplift, suggesting that the horizontal deformation provides complementary and potentially discriminating constraints on the main driving mechanisms of seasonal deformation measured by InSAR.

We further compare the observed seasonal displacement with forward models of poroelastic strain and elastic loading (Larochelle et al. 2022) at two representative sites : the Nevesinjsko polje in Bosnia and Herzegovina and the Skadar lake between Montenegro and Albania. The comparison suggests that overpressure-related mechanisms (from poroelastic deformation or fracture opening) dominate at Nevesinjsko polje, whereas elastic loading prevails in the Lake Skadar catchment area. The bimodality of the phase observed at the large regional scale reflects these two end-member mechanisms of deformation. Although local complexities in subsurface properties and hydro-climatic forcings, and anthropogenic water usage may limit the applicability of both simple parametrization and simple models, our results demonstrate the significant potential of multi-track Sentinel-1 time series to constrain seasonal hydro-mechanical processes in complex karst environments using horizontal and vertical deformation patterns.

References :

Chaussard et al. (2017). doi : 10.1002/2017JB014676

Larochelle et al. (2022) doi : 10.1029/2021JB023097

Lesparre et al. (2016). doi : 10.1093/gji/ggw446

Milanović, Petar. (2015). doi : 10.1007/s12665-014-3923-0

Silverii et al. (2019). doi : 10.1016/j.epsl.2018.10.019

Thollard et al. (2021) doi : 10.3390/rs13183734

FLATSIM Data Products. CNES. (Dataset). doi: https://doi.org/10.24400/253171/FLATSIM2020

The recent availability of high-resolution, open-access MT-InSAR data, alongside free tools for data interpretation such as ADAtools, has enabled the development of wide-area, value-added geospatial products. This study demonstrates the integration of data delivered by the European Ground Motion Service with open-access ancillary datasets, ADAtools and machine learning tools to efficiently identify, map, and classify ground deformation phenomena across the countries covered by EGMS between 2015 and 2021.

The first step of our work consists of identifying clusters of measurement points into polygons of Active Deformation Areas (ADAs), thus reducing data complexity and simplifying the interpretation by focusing the analysis and classification on significant deformation areas. Hence, two alternative techniques to classify the ADAs into different classes of ground deformation processes have been developed and compared.

The first technique, ADAclassifier, part of ADAtools, is based on a scoring system and multiple parallel decision trees to assess each deformation category. The algorithm provides default values of thresholds and scores, although expert users can customize them to their specific needs. ADAclassifier classifies ADAs into five different deformation classes, i.e. landslides, subsidence, uplift, construction settlement and sinkholes.

The second technique, ground motion classifier (GMC), is based on a supervised classification of ADAs into four deformation classes, landslide, deep-seated gravitational slope deformation subsidence and uplift, using machine learning. The training dataset for this classifier is obtained by matching the European ADA map with ground truth/labelling data, including the Italian National Landslide Inventory, the subsidence map of Emilia-Romagna region (Italy), and clusters of known uplift areas across Europe, e.g. the dewatering areas in the United Kingdom. The supervised ground motion classifier is implemented through the Extreme Gradient Boosting (XGB) technique. XGB belongs to the ensemble learning family and is used in various applications due to its good performance, versatility, and capability to cope with missing values. In this work, the Catboost implementation of XGB was chosen due to its better performance. The XGB classifier employs spatio-temporal features extracted, for each ADA polygon, from different data sources, i.e. the EGMS-PSInSAR data (e.g. mean velocity, acceleration, seasonality, temporal coherence of the measurement point displacement values), Corine Land Cover map, Digital Elevation Model (DEM) and its derived terrain attributes (local slope and aspect).

Being based on the availability of specialized datasets, such as landslides or sinkhole inventories, the ADAclassifier was implemented in the territory of Spain, whereas its expansion over the whole European territory is an ongoing work. On the other hand, GMC is based on pan-European datasets, prioritizing scalability, and providing a dataset of classified ADAs over the whole European territory, however with a reduced number of classes with respect to ADAclassifier.

A central component of this study is the validation of the ADA classification results, which is performed over the ADAs contained in the Spanish territory. Expert user validation addressed discrepancies and ensured the accuracy of the classifications, particularly in complex scenarios. In addition, the ADAclassifier and GMC output were compared, showing a high degree of consistency, which reinforces their reliability. The classification results highlighted subsidence and landslides as prevalent phenomena, aligning with known geohazard distributions. While the ADAclassifier effectively identified subsidence, landslides, and uplifts, it faced challenges in distinguishing construction settlement and sinkholes, indicating a need for further refinement.

Future work should focus on refining decision-tree methodologies, integrating time-series data, and enhancing classification accuracy for overlapping deformation types. The proposed methodologies are a first step towards systematic geohazard monitoring, supporting risk assessment and the development of targeted mitigation strategies.

References

Barra, A., Solari, L., Béjar-Pizarro, M., Monserrat, O., Bianchini, S., Herrera, G., Crosetto, M., Sarro, R., González-Alonso, E., Mateos, R. M., Ligüerzana, S., López, C., Moretti, S., 2017. A Methodology to Detect and Update Active Deformation Areas Based on Sentinel-1 SAR Images. Remote Sensing, 9(10), 1002. https://doi.org/10.3390/rs9101002

Crosetto, M., Crippa, B., Mróz, M., Cuevas-González, M., Shahbazi, S., 2025. Applications based on EGMS products: A review. Remote [42] Sensing Applications: Society and Environment, 37. https://doi.org/10.1016/j.rsase.2025.101452.

Palamà, R., Barra, A., Cuevas-González, M., Monserrat, O. Crosetto, M., 2024. Ground Motion Classification from European Ground Motion Service Data Using Extreme Gradient Boosting. IGARSS 2024 - 2024 IEEE International Geoscience and Remote Sensing Symposium. https://doi.org/10736-10739. 10.1109/IGARSS53475.2024.10640768.

Cuevas-González, M., Barra, A., Palamà, R., Ezquerro, P., Monserrat, O., Crosetto, M., Navarro, J. A., Rivera-Rivera, J., Béjar-Pizarro, M., García-Davalillo, J. C., Galve, J. P., Beyond the European Ground Motion Service: identification and classification of unstable areas at the national level, submitted to IEEE Journal of Selected Journal of Selected Topics in Applied Earth Observations and Remote Sensing

Continuous Global Navigation Satellite System (GNSS) position time series represent a fundamental geodetic observable for the quantitative characterisation of long-term crustal deformation processes. Their millimetric accuracy in three-dimensional positioning enables the estimation of velocity fields associated with regional tectonics, postglacial rebound, and large-scale geodynamic processes. In this study, a new European long-term surface displacement model derived from the analysis of 5,980 continuous GNSS stations distributed across the European continent and surrounding regions, is presented. The dataset was retrieved from the archive of the Nevada Geodetic Laboratory (UNR-NGL) and referred to the IGS14 reference frame, aligned with ITRF2014 and consistent with the Eurasian plate-fixed realization.

A fully automated and reproducible processing chain was developed to isolate the low-spatial-frequency tectonic component from the high-spatial-frequency one mainly due to localized displacement phenomena. The methodological framework is based on a robust trend estimation of the GNSS stations and on a spatial correlation analysis of the velocity components, with the primary objective of excluding possible perturbations in the long-term signal (e.g., antenna modifications, co-seismic offsets, anthropogenic disturbances, undocumented modifications). In particular, the initial stage of the workflow consists of a rigorous screening of GNSS time series, in order to ensure the reliability of velocity components estimation. The GNSS time-series that in the corresponding metadata exhibit discontinuities or gaps are subject to automatic modifications to exclude possible bias in the long-term velocity estimation. Moreover, after the screening procedure only the GNSS stations that provide a minimum temporal coverage of two years, will be considered in the analysis. This step reduces the initial dataset of GNSS to a subset characterised by temporal stability and linear long-term trends.

Subsequently, a spatial similarity analysis is performed to discriminate GNSS stations influenced by local-scale deformation phenomena. For each GNSS station, a similarity index is computed separately for the East–West, North–South, and vertical mean velocity components, within a 1° × 1° spatial window. The similarity metric is a quantifiable measure of the coherence between the velocity vector of a target GNSS station and those of neighbouring GNSS stations. It is noteworthy that only GNSS stations that satisfy a high coherence for all the three components are considered reliable for the model reconstruction. After this procedure a subset of 4,545 GNSS stations was considered representative of regional tectonic signals.

The implemented procedure continues by applying an inverse distance weighting (IDW) interpolation on GNSS velocity components, with the aim of creating a denser grid of points with a spacing of 0.5° in both latitude and longitude, to ensuring uniform spatial sampling and enhancing the stability of subsequent interpolation procedure. More precisely, the final continuous displacement field was obtained through the utilisation of a moving-window kriging interpolation. The final grid resolution was set to 10 arc-seconds, enabling high-resolution representation of low-spatial-frequency deformation patterns. It is important to highlight that no a priori fault geometry or kinematic constraints were imposed during interpolation.

The resulting European long-term surface displacement model is provided as 1° × 1° GeoTIFF tiles referenced to WGS84 and consistent with the naming convention adopted for SRTM products. The model well fits the principal geodynamic domains of Europe, encompassing the south-westward motion of the Aegean–Anatolian region (horizontal velocity of up to 2 cm/yr), the uplift of the Central Alps (vertical velocity of less than 0.2 cm/yr), the radial pattern of Fennoscandian postglacial rebound (vertical velocity of up to 1 cm/yr), and the spreading regime of Iceland along the Mid-Atlantic Ridge (horizontal velocity of up to 2 cm/yr). A validation process comparing both the exploited screened GNSS velocity dataset and the EGMS-GNSS long-term model, was performed. Such a validation indicates that discrepancies are predominantly below 0.3 cm/yr for horizontal components. However, larger residuals in the vertical component are observed, which are indicative of a higher sensitivity to localized vertical deformation. Furthermore, discrepancies are primarily concentrated along major active faults, thereby highlighting the effect of not using a fault model in the implemented procedure.

In conclusion, the principal enhancements in the usability of GNSS velocity data are twofold. First, the data are now provided as spatially continuous distributed in a tiled structure and made accessible in GeoTIFF format. This significantly improves data handling, visualization, and interoperability with standard GIS and remote sensing software. Second, the spatial resolution has been substantially increased, from the 50 kilometers grid spacing of the EGMS-GNSS product to a resolution of 10 arc-seconds (approximately 300 meters) in our datasets.

The Bangkok Metropolitan Region (BMR), Thailand’s economic and administrative centre, has undergone significant land subsidence over the past several decades. Subsidence is primarily attributed to excessive groundwater extraction and the consolidation of thick, highly compressible Quaternary Bangkok clay deposits underlying this low-lying deltaic plain. The BMR includes Bangkok and the surrounding provinces of Samut Sakhon, Samut Prakan, Nakhon Pathom, Nonthaburi, and Pathum Thani, forming one of Southeast Asia’s most densely urbanised and industrialised regions. Ongoing vertical ground deformation threatens critical infrastructure, increases flood susceptibility, and amplifies the impacts of relative sea-level rise in a region already vulnerable to climate-driven hydrological extremes. This study quantifies the spatial and temporal evolution of land subsidence across the BMR using Sentinel-1 Synthetic Aperture Radar (SAR) time-series interferometry. We generated interferometric products using the UK Centre for Observation and Modelling of Earthquakes, Volcanoes and Tectonics (COMET)-LiCSAR automated processing system, and displacement time-series analysis was performed with LiCSBAS. A total of 368 Sentinel-1 acquisitions spanning December 2014 to December 2024 were processed to derive line-of-sight (LOS) velocity fields and cumulative displacement time series. The applied workflow enables systematic, large-area deformation monitoring with millimetre-scale precision and temporal consistency. Results reveal spatially heterogeneous yet regionally persistent subsidence patterns. Maximum LOS subsidence rates reach approximately 30 mm/year in Samut Sakhon, with substantial ongoing deformation also identified in parts of Samut Prakan, Pathum Thani, and western Bangkok. High subsidence rates correlate spatially with zones of intensive groundwater extraction and areas underlain by thick compressible clay sequences. Temporal analysis indicates sustained deformation in industrial estates and peri-urban districts, whereas selected sectors exhibit attenuated subsidence trends, potentially reflecting the long-term effects of groundwater regulation policies. These findings demonstrate the robustness of the LiCSAR–LiCSBAS Sentinel-1 framework for regional-scale geodetic monitoring in complex megacity environments. The derived deformation fields provide quantitative constraints for hydrogeological assessment, subsidence hazard evaluation, and integrated urban resilience planning. Continuous satellite-based InSAR monitoring is essential for resolving evolving deformation dynamics and mitigating compounded risks associated with anthropogenic subsidence and accelerating relative sea-level rise in the Bangkok Metropolitan Region.

Keywords:

Land subsidence; Sentinel-1 InSAR; LiCSAR; LiCSBAS; Time-series deformation; Bangkok, Thailand

Vegetated, deep‑seated landslides in the Polish Carpathians pose a significant challenge for satellite synthetic aperture radar (SAR) monitoring because dense vegetation and limited infrastructure provide few persistent natural radar scatterers. As a result, the engineering applicability of conventional Interferometric SAR (InSAR) techniques is often constrained by low coherence, temporal decorrelation, and unfavorable viewing geometry. To overcome these limitations, we installed six artificial corner reflectors (CRs) designed to act as stable, high backscatter radar targets. The network consists of two off‑slide reference reflectors and four reflectors within the active Kłodne landslide, including two sites colocated with piezometer/inclinometer installations to enable cross‑validation with in‑situ deformation measurements.

We combined Sentinel‑1 SAR data acquired in both ascending and descending orbits with multi‑temporal InSAR processing to construct a long, high quality line‑of‑sight (LOS) displacement time series covering the period 2018–2024. The dual geometry SAR configuration is critical in this setting, as it allows partial separation of vertical (Up) and horizontal (East) displacement components, which cannot be resolved from a single viewing direction. To isolate geophysically meaningful deformation from SAR‑specific artefacts, the LOS time series was decomposed using discrete wavelet analysis into long‑term trends, seasonal or thermally driven components, and short‑term anomalies. Epochs affected by snow cover,identified as periods with degraded radar backscatter and phase stability, were explicitly masked to avoid bias related to snow‑induced decorrelation and phase delays.

The combined ascending and descending SAR observations indicate consistent eastward motion of approximately 2–4 mm yr⁻¹ at few reflector locations. Seasonal signals are most pronounced in the Up component, highlighting the sensitivity of SAR measurements to vertical motion and thermally or hydrologically driven surface processes. By integrating SAR‑derived displacement with rainfall and groundwater observations, we observe modest but physically consistent hydro kinematic relationships. In particular, rainfall to groundwater response lags of approximately 4–5 days are evident, while rainfall to displacement lags vary spatially from 0 to 86 days, reflecting heterogeneity in subsurface structure and landslide kinematics.

Validation against nearby inclinometer records confirms that the eastward displacement inferred from SAR and corner reflector measurements is consistent with independently observed subsurface deformation trends. A comprehensive error budget is presented, accounting for SAR measurement noise, geometry‑related uncertainties, and time‑series decomposition effects. Based on this analysis, deformation detection thresholds of approximately ≥ 0.5–1.0 mm yr⁻¹ (95% confidence interval) are achieved, demonstrating that SAR monitoring supported by artificial reflectors can resolve slow landslide motion at engineering relevant scales.

Rwanda is a densely populated, mountainous country located next to the East African Rift System, where steep topography, intense seasonal rainfall, land‑use pressure, and long‑term deforestation contribute to a high susceptibility to landslides and other ground‑instability hazards. Past rainfall triggered landslides have resulted in significant loss of life and displacement, particularly in the northern and western regions of the country. Despite the societal impact of these hazards, information on landslide activity and ground deformation in Rwanda remains fragmented across disparate studies and reports, limiting its operational use for hazard assessment and early warning.

This feasibility study outlines the conceptual framework for a Rwanda Ground Motion Service based on satellite synthetic aperture radar (SAR) interferometry, developed within the PanAfGeo+ Country Window Rwanda Project.

The proposed service aims to systematically map and monitor ground deformation associated with active landslides, mining and post‑mining areas, and tectonically influenced zones. Owing to Rwanda’s favorable geographic position near the African Rift Valley System, the country benefits from dense temporal and spatial coverage of Sentinel‑1 SAR data, including both ascending and descending acquisition geometries. A single Sentinel‑1 acquisition covers nearly the entire national territory, significantly simplifying large‑scale InSAR processing and national level product generation.

The Ground Motion Service will employ multi‑temporal SAR interferometric techniques, including Small Baseline Subset (SBAS) and Persistent Scatterer Interferometry (PSI), to derive ground motion maps and displacement time series. The availability of dual‑orbit Sentinel‑1 data enables improved characterization of deformation patterns and enhances the robustness of detected signals. Climatic conditions, however, impose important constraints on SAR interferometry in Rwanda. The climate is characterized by two main rainy seasons (September–December and March–May), a short less‑rainy period (January–February), and a dry season (June–August), all of which influence surface coherence and must be explicitly considered during InSAR processing and interpretation.

By integrating Sentinel‑1 SAR interferometry with a coordinated institutional framework for monitoring and early warning, the proposed Ground Motion Service represents a scalable and cost‑effective approach to hazard assessment in Rwanda.

The increasing frequency and severity of natural hazards pose significant challenges for disaster managers, emergency responders, and scientists studying long-term Earth surface change. Such hazards often affect large and/or remote regions, complicating ground-based delineation of affected areas and populations and delaying the deployment of personnel and equipment. Spaceborne synthetic aperture radar (SAR) and optical imagery provides timely, broad-scale, and repeatable observations that enable rapid and comprehensive characterization of disaster-impacted regions. SAR-based products provide all-weather, day-night observations of surface deformation, flooding, and land cover change, supporting rapid hazard assessment preceding, during, and following disasters. Complementary optical products enable mapping of surface water extent, vegetation impacts, burn severity, and geomorphic change where cloud-free observations are available. Integration of these data facilitates rapid emergency response while the repeat acquisition of observations enables investigation of the longer-term impacts of natural hazards. However, SAR and optical satellite imagery is often not immediately interpretable for practical applications (e.g., mapping flood, wildfire, or landslide extents), creating a clear need for standardized, analysis-ready products that can be readily accessed and used by both scientists and emergency responders.

The Observational Products for End-Users from Remote Sensing Analysis (OPERA) project at NASA’s Jet Propulsion Laboratory (JPL), in response to the needs identified by the Satellite Needs Working Group (SNWG), is delivering a suite of open-source and analysis-ready products derived from SAR and optical satellite imagery to address a wide range of scientific disciplines and disaster response needs. These products are free and available for download from NASA's Distributed Active Archive Centers (DAACs; see links below). By leveraging systematic observations provided by the Copernicus Sentinel 1/2 and NASA/USGS Landsat 8/9 constellations, as well as data from the recently-launched NISAR mission, OPERA produces standardized, validated datasets designed to lower barriers to use across scientific and applied communities. In this presentation we will highlight examples of disaster applications enabled by OPERA’s SAR- and optically-derived science-level products. Example applications include: (1) hurricane-induced coastal flood inundation mapping with the OPERA Dynamic Surface Water Extent from Sentinel-1 and Harmonized Landsat and Sentinel-2 (DSWx-S1/DSWx-HLS) products; (2) wildfire delineation and burn progression tracking with the OPERA Land Surface Disturbance from Sentinel-1 and Harmonized Landsat and Sentinel-2 (DIST-S1/DIST-HLS) products; (3) landslide detection with the OPERA Radiometrically Corrected (RTC) radar backscatter product; and (4) synergistic applications using multiple OPERA products. Additionally, we will discuss a collection of in-development tools (see link below) designed to automate notification, discovery, processing, and visualization of OPERA products over disaster-impacted areas and demonstrate how these open-source tools and products can be integrated into operational and research workflows to more rapidly quantify hazard impacts across large regions and support more comprehensive situational awareness for emergency responders. We seek and welcome feedback from those working across the academic, disaster management, and emergency response community who have applied OPERA products in their own work, as well as those looking to integrate these data in their own operational and/or research workflows.

--------------------------------------------------------------------------------------------------------------

More information about OPERA’s mission, funding, and the Satellite Needs Working Group (SNWG): https://science.nasa.gov/science-research/earth-science/opera-addressing-societal-needs-with-satellite-data/; https://www.earthdata.nasa.gov/data/projects/nsite

More information about the OPERA project and data: https://www.jpl.nasa.gov/go/opera/; https://www.jpl.nasa.gov/go/opera/products/

More information about OPERA data access: https://www.earthdata.nasa.gov/data/projects/opera

More information about open-source OPERA disaster response tools: https://github.com/OPERA-Cal-Val/disasters

Port infrastructure monitoring requires distinguishing between benign consolidation and critical structural anomalies from InSAR displacement time series. This study presents a Multiple Hypothesis Testing (MHT) framework that tests competing physical models to detect regime shifts while controlling false alarm rates. The approach employs Baarda's B-method for dimension-fair model selection, comparing linear velocity, thermal expansion, Heaviside step functions, and wave-climate interaction models against observed deformation patterns.

To bridge the gap between point-level measurements and operational decision-making, we define Vulnerable Areas of Interest (VAIs) as standardized spatial units representing distinct structural assets. Within each VAI, we derive four monitoring metrics: State (cumulative displacement), Rate (instantaneous velocity), Rate of Change (velocity increments at breakpoints), and Offset (modeled discontinuities). Spatial coherence is assessed via Moran's Index to distinguish systematic structural responses from scatterer noise.

The framework is applied to three Spanish ports using five years of EGMS Sentinel-1 data (2018--2023). At Castellón, the analysis reveals step-function offsets temporally aligned with Storm Gloria (January 2020) and spatially concentrated on the outer East Breakwater face, suggesting localized structural response to extreme wave loading (H_s > 6 m). The Closure Breakwater exhibits gradual differential settlement in mid-reach sectors, indicating ongoing consolidation processes. At Huelva, the Muelle Ingeniero Juan Gonzalo shows expected consolidation curves with seasonal thermal components, while the Paseo de la Ría displays differential settlement patterns correlating with documented remediation efforts. Dique Juan Carlos I reveals offset detections in early 2018 coinciding with energetic wave events from SIMAR records. At Algeciras, multi-geometry analysis of the detached breakwater shows divergent ascending and descending track trends in the northern sector, suggesting horizontal displacement components that single-geometry analysis would not resolve.

SAR interferometry (InSAR) is currently the primary source of information for surface motion measurements across a wide range of spatial scales, particularly when historical deformation analysis is required. The Copernicus Sentinel-1 mission represents a game changer, as its systematic acquisition strategy and open data policy have democratized access to satellite observations, allowing both the reconstruction of displacement histories over areas of interest worldwide and the development of operational surface motion monitoring chains. The emergence of cloud-based platform solutions is further accelerating the uptake of satellite data for mapping and monitoring surface motion by ensuring seamless access to data archives, large storage capacity, computational resources, and the execution of advanced processing chains in a straightforward manner. This significantly lowers the technical requirements for thematic domain professionals seeking to utilize these measurements within their areas of expertise.

In this context, the mining sector can greatly benefit from satellite-based surface motion monitoring for assessing the stability of mine walls, pit slopes, tailings facilities, and above-ground infrastructure within operational mining sites. However, one of the intrinsic characteristics of open-pit mines is the continuously changing surface topography, which introduces challenges for most multi-temporal SAR processors. Rapid surface changes may disrupt the continuity of phase stability, which is a fundamental requirement for Persistent Scatterer Interferometry (PSI), leading to reduced spatial coverage of PS targets, especially when long observation periods are considered.

To mitigate this limitation, we propose a sequential processing strategy based on the division of the entire observation period into distinct automatic or user-defined sub-periods. Independent interferometric processing is performed for each sub-period, followed by a concatenation procedure to reconstruct a continuous displacement time series. The linkage between independent time series accounts for potential offsets arising from different starting dates, the need to extrapolate over temporal gaps where linear behavior can be reasonably assumed, and trend adjustments across overlapping intervals.

Such an approach enhances measurement density while allowing the ingestion of updated Digital Elevation Models (DEMs) representing contemporary topography. This is particularly critical in mining environments where significant topographic changes occur due to excavation activities. Since updated DEMs are often derived from drone-based surveys and may be spatially limited to areas of active elevation change, a dedicated assimilation strategy is implemented to integrate these high-resolution local datasets with global height products. Assumptions still remain, primarily concerning the exact spatial correspondence of measurement points between independent processing intervals. In the current implementation, this is addressed through sampling individual datasets over a common grid prior to the concatenation process, effectively transitioning from distributed point targets to regularly spaced grid points. This ensures spatial consistency and allows the generation of continuous time series over the full observation period.

The developed workflow has been implemented as an upgrade to the existing SNAPPING PSI service of the Geohazards Exploitation Platform (GEP), offering to mining operators the benefit of utilizing an automated online service. In addition to the standard service capabilities, users can define temporal breakpoints and optionally upload updated external height datasets for each processing interval. Results from the analysis of multiple mining sites demonstrate a significant enhancement in measurement density and spatial coverage, allowing comprehensive surface motion monitoring without restricting the analysis to shorter time spans.

Acknowledgements

This work was carried out within the framework of the MASTERMINE project (Grant Agreement No. 101091895), funded by the European Union’s Horizon Europe research and innovation programme.

InSAR is becoming more widely recognised as a valuable tool for monitoring ground and structural movement at various scales. At one end of the scale, Sentinel-1 can provide a cost-effective way of monitoring movement at the scale of regions, countries or even continents. SatSense have processed Sentinel-1 data to produce nationwide maps of the United Kingdom and New Zealand as well as large areas affected by extractive industries such as the Permian Basin in the USA. These large scale InSAR datasets can be kept up to date as new Sentinel-1 images are acquired and the data are made available through a dedicated online visualisation platform. At the other end of the scale, high-resolution InSAR data from satellites such as TerraSAR-X (TSX) and COSMO-SkyMed (CSK) can be used for detailed analysis of individual structures and to provide a more granular picture of movement across smaller regions. Here, we will show examples of how InSAR data have been used across multiple scales, industries and locations.

In the UK, we will show how nationwide Sentinel-1 data can be used to monitor entire infrastructure networks for customers such as Network Rail and others. InSAR offers several valuable opportunities to infrastructure owners:

• Remote monitoring can reduce the number of in-person inspections, reducing risk to personnel, costs and carbon footprint

• InSAR can provide a consistent, network-wide picture of movement risk

• Areas of concern can be flagged for further inspection, enabling prioritisation of resources

We will also show examples of how nationwide InSAR can also be used to monitor the ongoing risk from historic mining across large areas of the UK. These data can be used by regulatory authorities, insurance companies and geotechnical engineers to understand not just the risk of ground movement in a given area, but whether the ground has actually moved in the last 10 years.

In addition to our UK-wide InSAR dataset, SatSense have partnered with Earth Sciences New Zealand (previously GNS Science) to provide a nationwide Sentinel-1 InSAR dataset for New Zealand. New Zealand is exposed to a range of geohazards and InSAR provides valuable insight about how these geohazards are evolving at scale. We will show examples of how this InSAR data is used by academics, government agencies and industries throughout New Zealand to monitor landslides, geothermal fields, fault movements and more.

In the Permian Basin, USA, we will show examples of InSAR imaging complex movements associated with oil and gas extraction, and accompanying wastewater pumping. These movements can vary in space and time as fluids move beneath the surface and are well imaged using wide-area InSAR. InSAR offers cost-effective, wide-area context and opportunities to monitor ongoing, dynamic movements which can be indicative of subsurface operations.

Throughout the world, some applications require higher resolution InSAR data. TSX and CSK data can be used to provide measurements at the property level or for high-value assets such as tunnels, bridges, dams, airports and more. The higher spatial resolution and improved height estimation from their baseline variation allows for more precise positioning of the radar reflectors in 3D. These precise 3D positions, coupled with higher sensitivity to small movements due to a shorter X-band wavelength make these satellites well suited to monitoring how distinct parts of structures move over time. We will show examples of how these higher resolution satellite datasets can be used to monitor a range of properties and structures for different kinds of movement.

The combination of wide-area Sentinel-1 data and focused high-resolution InSAR datasets means InSAR is becoming more widely recognised as valuable for a range of sectors. The recent launch of NISAR and the expected launch of further InSAR capable satellites over the coming years will lead to even more potential for InSAR monitoring.

Triggered by the systematic availability of SAR imagery acquired by the ESA ERS satellites, InSAR time-series methodologies began to be developed in the late 1990s. One example is the Delft Implementation of Persistent Scatterer Interferometry (DePSI) (van Leijen, 2014), the extension to TU Delft’s DORIS InSAR package (Kampes et al, 2003) . DePSI is characterized by the use of geodetic estimation and hypothesis-testing techniques to rigorously assess parameter estimates and identify and remove incorrectly estimated parameters. By implementing DePSI in MATLAB—the primary programming language used in the educational program at Delft University of Technology—students were able to work with and further develop the code base.

Since then, the paradigm has shifted toward open-source software solutions, and students are now primarily trained in Python. At the same time, the rapidly increasing volume and resolution of SAR acquisitions from modern satellite missions pose significant challenges in terms of scalability and extensibility. To ensure the future development of DePSI, we have reimplemented the software in Python. The resulting open-source package is designed to efficiently handle large InSAR datasets while adhering to modern Python software engineering standards. It builds on the original MATLAB implementation (van Leijen, 2014) and extends it with a scalable, modular, and community-oriented architecture.

To address the challenges of intensive InSAR processing, DePSI is built on Xarray and Dask, enabling efficient manipulation of multi-dimensional datasets and seamless scalability from local laptops to High-Performance Computing (HPC) environments. This design allows DePSI to process large SAR stacks while maintaining memory efficiency and parallel performance.

DePSI adopts a functional programming–oriented design, facilitating the integration of new PSI algorithms alongside existing conventional methods. Comprehensive user and developer documentation, including example Jupyter notebooks, is provided to lower the barrier for adoption and extension. Modern software quality practices—such as unit testing, continuous integration, and version control—are fully implemented, ensuring robustness and long-term maintainability and fostering community-driven development.

DePSI aims to provide a scalable, extensible, and high-quality open-source platform for next-generation PS-InSAR research and applications. In our contribution we will present the software design and use, together with example use cases. The DePSI repository can be accessed via https://github.com/TUDelftGeodesy/DePSI.

Kampes, B. M., Hanssen, R. F., and Perski, Z., 2003. Radar interferometry with public domain tools. In Third International Workshop on ERS SAR Interferometry, ‘FRINGE03’, Frascati, Italy.

van Leijen, F., 2014. Persistent Scatterer Interferometry Based on Geodetic Estimation Theory, Ph.D. dissertation, Delft University of Technology, Delft, The Netherlands.

Land subsidence, predominantly driven by excessive groundwater extraction, poses significant risks to urban infrastructure and public safety in Iran. Previous national-scale studies have highlighted Iran as one of the countries most affected by subsidence, with accelerating trends over recent decades. However, detailed, high-resolution assessments of subsidence hazards within the country’s largest urban centers remain limited. This study leverages ten years of Sentinel-1 Interferometric Synthetic Aperture Radar (InSAR) data to conduct a comprehensive analysis of land subsidence across the most populous megacities of Iran with populations exceeding 1.5 million, including Tehran, Mashhad, Shiraz, Tabriz, Karaj, and Isfahan.

Our investigation focuses on multiple facets of subsidence dynamics, emphasizing both vertical and horizontal ground deformations and their implications for urban systems. First, we provide an updated overview of general subsidence patterns in these cities, identifying hotspots of accelerated deformation and regions where subsidence trends are intensifying over time. Beyond this general assessment, the study presents a precise evaluation of differential subsidence hazards, with particular attention to densely built urban areas and critical infrastructures, including transportation networks, utilities, and historic sites.

A key contribution of this work is the spatio-temporal characterization of subsidence within each city. High-resolution InSAR time-series analysis enables the detection of localized deformation variations, revealing how subsidence evolves differently across urban neighborhoods and industrial zones. Additionally, we implement a novel methodology to decompose line-of-sight (LOS) deformation into three-dimensional displacement components, allowing for the first detailed mapping of horizontal ground strains induced by subsidence. This approach highlights areas where horizontal deformation may compromise structural integrity, complementing differential vertical deformation assessments and providing a more complete hazard evaluation framework.

Results indicate that all six megacities exhibit significant subsidence, with pronounced spatio-temporal variability and heterogeneous strain distributions. Cities such as Tehran and Isfahan show persistent high subsidence rates correlated with intensive groundwater withdrawal and urban expansion. The integration of vertical and horizontal deformation patterns provides actionable insights for urban planners, policymakers, and disaster risk managers, enabling targeted interventions to mitigate infrastructure damage and enhance urban resilience.

This study underscores the critical need for continuous, high-resolution monitoring of subsidence in rapidly growing urban centers and demonstrates the utility of Sentinel-1 InSAR data for multi-dimensional hazard assessment in megacities prone to anthropogenic land deformation.

Miami’s barrier islands have experienced a rapid increase in high-rise development over the past decade, with construction ongoing along much of the coastline. In addition, the tragic collapse of the Champlain Towers South in 2021 raises the possibility that some deformations may be related to construction activities and subsurface conditions. This study uses Interferometric Synthetic Aperture Radar (InSAR) time-series analysis, processed with the open-source SARvey software framework, which builds on Mintpy and MiaplPy, to analyze subsidence occurring along Miami’s barrier islands between 2017 and 2025. Using Sentinel-1 and TerraSAR-X data, the study documents radar line-of-sight (LOS) displacements up to 15 cm affecting highrises and their surroundings in multiple sub-regions (Sunny Isles, Surfside, North Beach, and South Beach). In Sunny Isles, which appears to have experienced more building development than other areas over the past decade, we documented deformation with an average of 10 mm/yr LOS displacement velocities monitored in nearly one third of the buildings over 100 meters in height. The detected subsidences are presumed to be construction related, considering the construction activities taking place nearby. We also correlate the measured displacements with geological cross-sections obtained from geotechnical reports, and we consider the possibility that these subsidences may have been caused by the dynamic settlement and creep of sandy layers within the limestone. This study examines that construction of highrise buildings can lead to creep deformation in the Miami limestone area, and highlights the capability of InSAR technology to observe settlement and integrity of structures.

Satellite-based interferometric synthetic aperture radar (InSAR) is attracting increasing attention as a tool for monitoring deformation in road and railway infrastructure. Despite this growing interest, its practical application is still limited. To estimate the usability and applicability for a routine monitoring, the vertical settlements along a 30 km railway line were estimated using Sentinel-1 data and compared with existing in-situ surveys. The suitability of pre-processed data from the European Ground Motion Service (EGMS) and custom-processed Sentinel-1 data was evaluated.

The comparison was made between Sentinel-1-derived settlements and conventional railway measurements for five selected bridges, including the adjacent free-track sections. Additionally, deformation trends were assessed along the entire corridor.

Persistent scatterer (PS) points were aggregated along the line using spatial clustering, which reduces sensitivity to unreliable individual scatterers and accounts for geolocation uncertainty at the metre level. A median-based statistic is then applied within the clusters and the line of sight (LOS) deformation is converted into vertical and horizontal settlement components, enabling comparison with geodetic results.

There was very good agreement between the InSAR-derived settlements and the local measurement data for the investigated section. Unlike previous experience, custom processing did not substantially improve the accuracy of the results compared to those obtained using EGMS products. However, the recently constructed section of the corridor must be analysed, as it is not yet included in the EGMS database.

The evaluation highlights two key challenges in interpretation. Firstly, rapid surface changes due to track ballast tampering can occur within a revisit interval of six days, exceeding the reliable deformation increment of approximately one quarter of the radar wavelength (1.4 cm). This increases the risk of phase unwrapping ambiguities. Therefore, integrating tampering records is essential, as this enables sudden surface changes to be successfully interpreted that would otherwise be poorly represented in the InSAR time series. A simplified method of accounting for this has been developed for demonstration purposes only; a more adequate method will follow. Secondly, transitions between areas with different settlement behaviour (e.g. bridge structures versus the surrounding ground) cannot always be resolved precisely due to limitations in spatial sampling. In practice, an effective along-track spacing of 20 metres can only be achieved under optimal conditions; 50 metres is a more robust alternative.

To strengthen future monitoring, corner reflectors were introduced as stable reference targets. Three reflectors have been installed and their detectability was verified to provide deformation measurements representative of the behaviour of the bridge rather than the track alone. Overall, the study shows that Sentinel-1 EGMS products can support netwide settlement assessment, and that local interpretation benefits greatly from additional maintenance data and dedicated reference targets.

Taiwan, situated at a convergent plate boundary, is characterized by steep terrain and fractured geology. Under the influence of extreme climate events, frequent typhoons and heavy rainfall often induce large-scale landslides and debris flows. During severe weather, conventional optical remote sensing is frequently limited by extensive cloud cover, creating a critical information gap that hinders immediate disaster response and spatial situational awareness. To address this challenge, this study establishes a wide-area rapid screening mechanism based on Synthetic Aperture Radar (SAR) imagery, leveraging its all-weather, cloud-penetrating capabilities to identify and assess large-scale landslides and landslide-dammed lakes (barrier lakes).

We utilized Sentinel-1 and ALOS-2 satellite imagery, employing the Log Ratio Method to calculate differences in backscatter intensity before and after disaster events. By analyzing the statistical distribution of pixel values and applying 95th or 99th percentile thresholds, we effectively extracted hotspots of surface change. To enhance interpretation accuracy, a noise filtering algorithm was implemented to exclude artifacts caused by riverbed sediment transport, flat terrain, and geometric distortions, retaining only significant deformation areas larger than 1 to 2 hectares for subsequent hazard assessment.

Following major meteorological events in 2025, specifically Typhoons Danas (July 4–9) and Wipha (July 17–21), we conducted a wide-area rapid assessment. Using multi-temporal Sentinel-1 ascending and descending orbit imagery, we successfully identified critical disaster hotspots, including dammed lakes in the Qingshui River (Central Taiwan) and the upstream Mataian River (Eastern Taiwan). The Mataian landslide-dammed lake, validated via Planet optical imagery with an area of approximately 146,383 m², served as a primary case study. By integrating digital terrain models (DTM) and the dimensionless blockage index (DBI), we estimated the short-term stability of the dam. These results were promptly disseminated to relevant disaster prevention authorities, demonstrating the rapid response capability of radar satellites in monitoring remote mountainous areas under cloud cover.

Furthermore, following the heavy rainfall event from July 28 to August 1, 2025, an island-wide screening using integrated Sentinel-1 and ALOS-2 imagery revealed surface change hotspots concentrated in the Gaoping River basin (Southern Taiwan) and the Beinan, Hualien, and Xiuguluan River basins (Eastern Taiwan). Notably, two large-scale landslides (totaling ~14 hectares) were identified at a tributary confluence in the middle reaches of the Baolai River, posing a high risk of river blockage. ALOS-2 imagery also indicated potential instability in the Lakusi River basin, where six landslide areas (>0.3 ha) were detected.

To validate the timeliness and accuracy of the radar-based detection and to investigate the landslide mechanisms, we incorporated data from the Broadband Seismic Network. Through seismic signal inversion, we pinpointed the collapse times of the Baolai River landslides to 06:26:10 and 06:26:40 on August 1, 2025, with estimated volumes of 1.75 and 2.14 million m³, respectively. These high-precision temporal and volumetric parameters not only corroborated the spatial changes detected by SAR but also provided critical physical constraints for slope stability analysis, compensating for the temporal resolution limitations of satellite imagery.

Finally, the study conducted a detailed disaster simulation for the high-risk Mataian dammed lake. Using DTM-based spatial intersection analysis, we determined the dam geometry and reservoir capacity. Hydraulic dam-break simulations were performed total dam failure and partial failure scenarios with breach durations of 0.5, 1.0, and 3.0 hours. Numerical results indicate that in the worst-case scenario (total dam failure, 0.5-hour duration), the peak discharge at the downstream Mataian Bridge would reach 10,546 CMS with a water level of 8 meters. Even under a partial breach scenario, the peak discharge (8,424 CMS) significantly exceeds the 100-year return period design flow (4,043 CMS). These findings highlight the catastrophic potential of a rapid breach and provide a scientific basis for emergency evacuation planning.

In conclusion, this study presents a comprehensive disaster assessment framework integrating wide-area SAR screening (36,000 km²), seismic signal validation, and hydraulic simulation. Proven by the 2025 typhoon and rainfall events, this workflow effectively overcomes weather limitations to rapidly identify high-risk zones, filling the information void during early disaster stages and significantly enhancing monitoring and early warning capabilities for compound disasters in extreme environments.

ePISAV (Enhanced Permanent Interferometric Scatter Analysis & Visualization) is an advanced tool for integrated analysis of PS-InSAR displacement data. The tool is designed for operational monitoring tasks, including slope stability assessment, subsidence evolution, and long-term ground-motion analysis. Developed on the ENVI geospatial platform, the system consolidates data import, visualization and modelling into a single workflow, enabling the use of PS-InSAR measurements and complementary reference datasets such as GPS data. The tool combines different functions for PS (Permanent Scatters) analysis, in particular: advanced time-series analysis, frequency analysis by FFT (Fast Fourier Transformation) transformation, statistical analysis, supervised and unsupervised point classification, data transformation from point clouds to raster surfaces and data calibration between InSAR and GPS data. These functions allow rapid comparison of displacement behaviors in different areas and in different periods, highlighting the anomalies and characterizing the deformations with higher interpretability. ePISAV enhances the interpretability of PS data through a set of modules: i.-visualization module the color-coded intervals can be assigned to any PS attribute, enabling fast mapping of sectors with similar velocities or accelerations and revealing coherent deformation domains in the same area; ii-3D module converts PS distributions into continuous raster or raster-series surfaces, making spatial gradients and evolving deformation fronts immediately visible through 3D visualization and temporal animation; iii-calibration module aligns PS-derived displacement with GPS trends, reducing systematic offsets and improving reliability in multi-sensor comparison scenarios; iv-classification module supervised and unsupervised classification algorithms (Minimum Distance, SAM, Binary Encoding, SID) group PS based on the similarity of their displacement signals, delineating areas with distinct kinematic behaviors useful to better understand landslides, subsidence and calibrate geomechanical modelling for structural monitoring. The aim of ePISAV is to simplify InSAR analysis into a unified environment that reduces processing time and promotes consistent methodologies across diverse case studies, from local monitoring tasks to regional-scale ground-motion analyses. By integrating all analytical steps, the tool provides an operational framework that can be directly integrated with external thematic datasets, such as geological information, land-use maps, hydrological layers and geomechanical models, enabling efficient transition from data to insights, ensuring consistent, interpretable and robust analysis capabilities for a wide range of deformation-monitoring applications.

Coal mining in South Limburg (the Netherlands), active from the late 19th century until the 1970s, has left a lasting imprint on a now densely populated mining region of ~234 km². Decades after mine closure, subsurface processes continue to induce surface deformation and sinkhole hazards, posing risks to infrastructure and the built environment.

Here, we present a multidisciplinary approach that combines satellite geodesy (InSAR), hydrogeological observations, geological data, and historical mining records through geomechanical models to better understand and manage post-mining hazards. From the previously identified potential after-effects we focus on sinkholes and other surface deformation due to their strong potential for damaging infrastructure and buildings.

Central to this framework is the analysis of ~28 years of InSAR data acquired from five satellite missions (ERS, Envisat, Radarsat, TerraSAR-X, and Sentinel), enabling the characterisation of long-term surface deformation patterns. These data reveal a persistent regional uplift signal (~5 mm/yr), first detected in early ESA missions and still on-going and confirmed by more recent observations. By integrating piezometer data, we relate this uplift to mine water rebound and aim to predict coupled groundwater and surface displacement dynamics.

At the local scale, we combine deformation data with geological, mining information and geomechanical models from a known sinkhole location to identify areas with similar subsurface conditions, significantly reducing the search space for potential sinkhole occurrence. These zones are further linked to exposure datasets (buildings and infrastructure) to assess vulnerability.

The results have first been integrated into an Atlas as a building block for supporting probabilistic hazard assessment. Here, we present the status of ongoing development of a framework to assess vulnerability, current results, and prospects to accommodate for annual updates of InSAR data.

This study highlights the value of long-term InSAR observations as a key component in a holistic framework for post-mining aftercare.

The European Ground Motion Service (EGMS), managed by the European Environment Agency, is part of the Copernicus Land Monitoring Service portfolio. It provides freely accessible, continental-scale land deformation data across Europe based on Sentinel-1 SAR imagery. The objective of EGMS is to deliver a consistent and long-term monitoring tool to support the understanding and management of land dynamics, including those driven by climate change.

Given the large spatial coverage of EGMS and the complexity of InSAR data for non-expert users, a systematic and continuous validation framework is essential. In parallel with each EGMS release, we have developed the first continental-scale validation procedure, enabling both quantitative and qualitative evaluation of successive product updates.

Validation activities carried out across Europe assess the agreement between EGMS products and independent reference datasets (in-situ and Earth Observation), based on two main criteria: (1) Consistency & Accuracy and (2) Applicability & Usability. Consistency and Accuracy are evaluated through comparisons of measurement point (MP) velocities and time series with GNSS and in-situ observations, together with assessments of geolocation accuracy and temporal behaviour using corner reflectors. Applicability and Usability are analysed by examining the coherence of EGMS outputs with other Ground Motion Services, geospatial datasets, and inventories, supported by the identification of Active Deformation Areas (ADAs) derived from EGMS MPs.

This work is carried out by a multidisciplinary consortium spanning six European countries, bringing together geological surveys, a research institute, and an industry partner, reflecting the diversity of expertise required for continental-scale validation.

In this study, we present the validation framework, methodologies, and results for the first three EGMS releases: 2015–2021, 2018–2022, and 2019–2023 developed within the contract No 3506/R0-COPERNCA/EEA.59565 valid between 2021 and 2025.

Petroleum Development Oman (PDO) has developed an integrated geospatial and geomechanical monitoring framework combining nationwide Interferometric Synthetic Aperture Radar (InSAR), continuous GPS stations, microseismic networks, production data, and stratigraphic analysis to support both Carbon Capture, Utilization, and Storage (CCUS) site selection and sustained hydrocarbon production.

Leveraging Sentinel-1 SAR imagery from 2016 to May 2022, PDO with its business partner SkyGeo generated the first millimetre-precision deformation map of Oman, revealing that over 90% of the country exhibits minimal vertical displacement (<5 mm/yr), indicating broad geological stability ideal for long-term CO₂ storage. Notable deformation regimes include localized uplift/subsidence in active hydrocarbon fields directly correlated with fluid injection and production activities, and shallow, transient displacements over dunes and sabkhas, which collectively inform CCUS site screening and the establishment of pre-injection deformation baselines.

Beyond carbon storage, InSAR is operationally embedded in PDO’s Exploration & Production (E&P) activities. It is used to map steam/heat distribution in Enhanced Oil Recovery (EOR) projects, monitor surface subsidence to correlate fault movements with microseismicity, calibrate reservoir compaction models, and assess topographic and drainage changes impacting surface facility integrity and upgrade planning. To enhance the temporal resolution and absolute accuracy of InSAR-derived deformation, PDO has integrated a network of continuous GPS monitoring stations across key fields and infrastructure zones. These GPS stations provide real-time, point-based ground truthing, anchor InSAR time-series to a stable reference frame, and validate the representativeness of deformation signals, particularly critical in areas with complex or shallow deformation.

Critically, PDO is advancing an early warning system that combines InSAR and GPS deformation data with microseismic event catalogs, real-time production rates, and high-resolution stratigraphic models. This multi-parameter integration enables proactive identification of anomalous deformation patterns, such as unexpected uplift/subsidence coinciding with injection rate changes or microseismic clusters, allowing operators to flag potential geomechanical risks before they escalate. Stratigraphic context ensures that observed surface movements are interpreted within the framework of subsurface layering, fault architecture, and reservoir compartmentalization, reducing false positives and improving diagnostic accuracy. This system not only supports reservoir integrity management and facility safety but also provides a scalable template for MMV (Measurement, Monitoring, and Verification) during CCUS operations, capable of detecting subtle, early-stage deformation signals during injection and containment phases.

This work demonstrates how advanced interferometric monitoring, when strategically combined with complementary geospatial, geophysical, and operational data, enables proactive reservoir and facility management, reducing operational risk and supporting sustained production.

Synthetic Aperture Radar Interferometry (InSAR) enables precise observation of surface deformation. Wetland InSAR provides a unique capability to map water-level fluctuations across vegetated wetlands at high spatial resolution by leveraging double-bounce scattering between the water surface and emergent herbaceous vegetation. The Ciénaga Grande de Santa Marta (CGSM) in northern Colombia is a vast deltaic wetland system (~4,931 km²). Designated as a Ramsar site and a UNESCO Biosphere Reserve, it functions as an important carbon sink. The CGSM is experiencing a severe environmental crisis, in which extensive dikes and road infrastructure have disrupted hydrological connectivity, resulting in hypersalinization and widespread mangrove degradation across tens of thousands of hectares. Climate-driven variability, including El Niño-Southern Oscillation effects on precipitation and freshwater inflow, further compounds salinity and hydraulic stress. Although periodic monitoring is essential, in situ gauge observations are sparse and inherently point-based, and the limited CGSM monitoring network makes a remote-sensing-based approach indispensable.

This study uses ALOS-4 PALSAR-3 to retrieve high-resolution water-level changes in the CGSM and addresses two practical limitations of wetland InSAR: spatial discontinuities and temporal gaps. Spatial discontinuities frequently arise near wetland boundaries where scattering mechanisms transition abruptly, making mosaicking across neighboring tracks and acquisition dates challenging and reducing spatial continuity. Leveraging ALOS-4’s high-resolution, wide-area observations, we establish continuous large-area coverage by mosaicking nine Stripmap acquisitions collected over adjacent tracks and evaluate the feasibility of generating a spatially continuous, high-resolution interferogram. We further employ a cross-mode InSAR strategy that integrates Stripmap and ScanSAR data to compensate for temporal gaps arising from acquisition constraints and to assess performance at longer temporal baselines. Despite the enhanced observation capability enabled by ALOS-4’s digital beamforming, the current reliance on fixed pulse repetition frequency (PRF) operation introduces blind areas (~10–20% of the scene) and may limit acquisition opportunities, potentially increasing temporal baselines. Such constraints are particularly critical in wetlands, where rapid water-level variability and vegetation-driven changes in scattering characteristics accelerate decorrelation, underscoring the need for strategies that preserve both spatial and temporal monitoring capability.

We generated interferograms from mosaicked Stripmap acquisitions on 17 June and 26 August 2025 and produced a cross-mode interferogram by incorporating a ScanSAR acquisition on 2 December 2025. Processing was performed using GAMMA: oversampling was applied to harmonize azimuth pixel spacing across acquisitions, multilooking with a 4 × 12 (range × azimuth) was used, and the topographic phase was removed using the Copernicus 30 m DEM. Adaptive filtering was then applied to suppress phase noise. The mosaicked Stripmap interferogram achieved a high mean coherence of 0.76 and revealed clear fringe patterns consistent with water-level changes in the CGSM. Importantly, fringes were continuous across mosaic boundaries, indicating that the proposed mosaicking strategy can produce spatially consistent interferograms over large wetland areas. The Stripmap-ScanSAR interferogram yielded a mean coherence of ~0.40, demonstrating the feasibility of cross-mode interferogram generation and suggesting that mode integration may help mitigate decorrelation as temporal baselines increase. Overall, these results support large-scale wetland monitoring with ALOS-4 and provide a foundation for developing precise time-series datasets for the CGSM.

Vertical land motion (VLM), defined as the upward or downward movement of the Earth's surface over time, is a critical component of coastal change assessments. VLM can vary substantially across local to regional scales and over decadal to multi-year timescales due to differences in local geology, natural processes, and anthropogenic influences. These spatial and temporal variations introduce significant complexity into relative sea-level rise (RSLR) assessments. Therefore, accurate characterization of both components is essential for reliable RSLR projections, particularly along low-lying coastlines.

Coastal Massachusetts (MA) is a low-lying coastal area encompassing geologically diverse environments from densely urbanized shorelines in Boston Harbor to environmentally sensitive wetlands along Cape Cod. In this study, we characterize near-decadal (2016–2025) spatial and temporal variability in VLM across coastal MA by integrating continuous Global Navigation Satellite System (GNSS) time series with high-resolution (~90 m) Sentinel-1 Interferometric Synthetic Aperture Radar (InSAR) observations. GNSS analysis reveals significant temporal variability in VLM, including accelerating subsidence across the Cape Cod Peninsula, decelerating trends along the South Coast, and near-stable conditions in Massachusetts Bay. Building on the GNSS-based insights into temporal variability in VLM, we examine the spatial variability of present-day deformation across coastal MA by referencing InSAR observations to temporally representative GNSS-derived VLM rates. The resulting deformation map reveals a north–south gradient in present-day vertical motion, ranging from near-stable conditions along the North Shore and Massachusetts Bay to subsidence rates exceeding ~2–3 mm yr⁻¹ toward the Cape Cod Peninsula and Outer Islands. We further assessed implications for coastal flooding by incorporating InSAR-derived VLM rates into mid-century (2050) RSLR projections under the SSP2-4.5 scenario at tide stations and other coastal locations. These locally resolved VLM estimates yield spatially variable RSLR projections ranging from approximately 0.48 m at Cape Ann to over 0.59 m in the Outer Islands, differing from those based on regionally averaged Intergovernmental Panel on Climate Change (IPCC) VLM estimates.

This study demonstrates that integrating temporally representative GNSS-derived rates reduces inconsistencies arising from time-dependent VLM and improves characterization of present-day coastal deformation. With increasing availability of Synthetic Aperture Radar (SAR) observations and longer GNSS records, integrating both datasets will be valuable for tracking the evolution of deformation patterns and updating coastal vulnerability assessments. Overall, this study contributes essential insight into the complexity of VLM in coastal MA and provides a critical foundation for adaptive coastal management strategies in an era of accelerating sea-level rise (SLR).

There is a clear link between global warming and the increase in glacier melting, leading to the expansion of glacial lakes, often dammed by fragile moraines. Triggers such as heavy rainfall, earthquakes, landslides, avalanches, glacier breakoffs, or thawing permafrost can cause glacial lake outburst floods (GLOFs). These events result in moraine breaches, releasing flood waves of mud and debris that can cause significant damage and endanger populations. On August 16, 2024, a GLOF from the Thyanbo glacial lakes affected the village Thame, Nepal. This flood caused destruction of the local infrastructure, buildings and agricultural land, and displaced over 135 inhabitants. According to first investigations, it seems that an initial trigger originated from the upper glacial lake and overtopped its terminal moraine. This flood wave further ran into the lower glacial lake, which overtopped the terminal moraine and caused it to breach. All mentioned cascading incidents triggered the GLOF running downstream. As no in situ data is available, we used high-resolution optical as well as Synthetic Aperture Radar (SAR) remote sensing data to map the lakes dynamics and measured the ground deformations at the terminal moraines. To date, such analyses have been applied only to glacial lakes and terminal moraines without documented GLOF events, but not to systems affected by a previously occurred GLOF.
High-resolution PlanetScope multispectral images from 2019 to 2024 showed an expansion of the upper lake by 213.3 % before the event, followed by a loss of 25.9 %, while the lower lake just increased slightly by 2.8 % over the timeseries, but lost over 74.3 % of its area during the GLOF. The analysis showed that the upper terminal moraine has not eroded at all or only very slightly, whereas the lower moraine has largely eroded. Consequently, while the lower lake no longer represents a future hazard, the upper lake continues to pose a high risk.
SAR Sentinel-1 images from 2020 to 2024 were used to perform a Persistent Scatterer Interferometry (PSI) with the Stanford Method of Persistent Scatterers (StaMPS). By combining ascending and descending orbits, the vertical and horizontal movements of the resulting scatterers were deconstructed. As the GLOF was likely triggered by an external factor, no abrupt movements were detected in advance by the PSI. Nevertheless, significantly stronger vertical and horizontal subsidence was observed at the lower terminal moraine, reflecting its greater exposure to the GLOF relative to the upper moraine. The analysis demonstrated that, with certain improvements, remote-sensing data combined with PSI can be used to assess the overall stability of terminal moraines and enable meaningful comparisons between them. Nevertheless, numerous limitations and restrictions have arisen along the workflow, and improvements identified should be taken into account and incorporated in future studies.
Based on this case study, the methodology will be transferred to a regional approach in the Himalayas in a future study to contribute to a more comprehensive inventory of potentially dangerous glacial lakes by adding the parameter of terminal moraine stability, which has not yet been considered in depth.

Introduction

Monitoring infrastructure has become a key application of InSAR time-series analysis since the introduction of Persistent Scatterer (PS) processing [1], later extended to include less reflective Distributed Scatterers (DS) [2]. As traditional in situ monitoring methods for primary road infrastructure have grown increasingly costly for Rijkswaterstaat, InSAR has become an attractive alternative for deformation monitoring of road infrastructure.

In some cases, applying a PS and DS processing chain without incorporating contextual information produces useful results. However, in many situations the processing approach must be adapted to obtain reliable InSAR deformation time series. Here, we present a collaboration between Rijkswaterstaat and SkyGeo demonstrating how the integration of object-specific information can substantially improve the quality of the final deformation product, using the sunken access roads of three tunnels in the Netherlands as a case study.

These sunken roads are currently being monitored for uplift, after a U-shaped concrete road element of a similar sunken access road was forced upward by groundwater pressure, following the failure of its ground anchors [3]. To avoid prolonged road closures caused by potential uplift of road elements at other locations, a monitoring system based on X-band InSAR time-series data was implemented. However, achieving the required vertical displacement accuracy of 1 mm proved impossible without substantial modifications to the standard PS and DS processing chain.

Unwrapping network

An initial analysis revealed that phase unwrapping errors are a common issue in these types of structures. This is caused by annual horizontal displacements of several centimetres resulting from thermal ratcheting of the entire construction. Along the sunken access road, adjacent U-shaped concrete road elements exert longitudinal forces on each other due to thermal expansion, leading to cumulative horizontal movement along the length of the road. The smallest displacements occur near the tunnel entrance, whereas the largest are observed at the beginning of the access road.

Incorporating this structural behaviour into the unwrapping network resolves the ambiguity issues, as the differential movement between adjacent elements is much smaller than their absolute displacement. By densifying the unwrapping network accordingly, the phase unwrapping errors can therefore be mitigated.

Similar challenges may arise in other segmented structures, such as bridges, viaducts, and flyovers [4], which are a key element of nationwide infrastructure monitoring programs [5].

Road Maintenance

Another common challenge is the decorrelation of time series caused by road maintenance activities. Such interventions often divide a continuous InSAR time series into two or more independent coherent time series. Although several processing strategies exist to automatically detect coherence loss and split time series accordingly [6], these approaches may incorrectly identify the timing of maintenance, or leave it undetected, resulting in phase jumps in the final time series.

Moreover, once a time series is split, the separate segments must still be reconnected to reconstruct a consistent deformation history. In our approach, we use precise information on the timing of maintenance activities to segment the time series. Because no significant modifications are made to the physical structure of the construction during maintenance, we can assume that the radar reflections originate from the same physical location, meaning the scatterer coordinates remain unchanged.

Therefore, we incorporate a step function at the precise dates of the maintenance into the temporal displacement model. This allows us to use the entire time series to estimate the model parameters, including the phase jumps during maintenance. This circumvents the need to split and stitch time series and therefore minimizes the loss in precision due to maintenance.

Assignment of DS to road elements

Finally, the correct assignment of individual PS and DS points to specific road elements is essential for a reliable decomposition in horizontal and vertical time series. In standard DS processing, the search for “brotherhood” pixels is performed within a predefined neighbourhood in the radar grid [7, 8]. In this case, however, such an approach would allow DS pixels from different tunnel elements to be grouped together, thereby mixing the deformation signals of adjacent elements and contaminating the results.

This issue can be avoided by restricting the selection of brotherhood pixels to those located within the same road element. To achieve this, the pixel locations must be known prior to the brotherhood selection step, whereas in conventional workflows this assignment is typically performed afterwards based on the phase data. Because the exact geometry of the structure and its surroundings is available, we can pre-assign pixels to individual tunnel elements using a high-resolution 0.5-m DEM. This enables a processing approach that does not mix radar reflections of different road elements in the final deformation product.

Conclusion

Combining the presented methods, we could improve the quality of the final InSAR time series in such a way that they now meet the quality requirements needed to monitor the tunnel access ramps. This would not have been possible without incorporating contextual information about the structure, its maintenance history, and its expected deformation behaviour. These results demonstrate that integrating technical knowledge of the monitored object can substantially enhance both the development of InSAR processing strategies and the reliability and usability of the resulting deformation products.

References:

[1] A. Ferretti, C. Prati, and F. Rocca, “Permanent Scatters in SAR Interferometry,” IEEE Transactions on Geoscience and Remote Sensing, vol. 39,no. 1, pp. 8–20, 2001.

[2] S. Samiei Esfahany, “Exploitation of distributed scatterers in synthetic aperture radar interferometry,” Ph.D. dissertation, Delft University of Technology, Delft, 2017. [Online]. Available: https://doi.org/10.4233/uuid:22d46f1e-9061-46b0-9726-760c41404b6f

[3] M. Harbers, https://zoek.officielebekendmakingen.nl/kst-29296-50.odt, pp. 1–4, 9 2023.

[4] X. Song, Z. Lei, and Z. Lu, “Retrieval of Discontinious Deformation Induced by Thermal Expansion and Contraction of Bridges with Adaptive MTInSAR,” in IGARSS 2024 - 2024 IEEE International Geoscience and Remote Sensing Symposium. IEEE, 2024, pp. 11 825–11 828.

[5] N. Dore, V. Belloni, A. Mazzoni, and M. Crespi, “Safe Bridge: Geomatic Monitoring Services for Safe Bridges,” in IGARSS 2024 - 2024 IEEE International Geoscience and Remote Sensing Symposium. IEEE, 2024, pp. 1607–1610.

[6] F. Lattari, A. Rucci, and M. Matteucci, “A Deep Learning Approach for Change Points Detection in InSAR Time Series,” IEEE Transactions on Geoscience and Remote Sensing, vol. 60, 2022.

[7] K. Spaans and A. Hooper, “InSAR processing for volcano monitoring and other near-real time applications,” Journal of Geophysical Research: Solid Earth, vol. 121, no. 4, pp. 2947–2960, 4 2016.

[8] A. Parizzi and R. Brcic, “Adaptive InSAR stack multilooking exploiting amplitude statistics: A comparison between different techniques and practical results,” IEEE Geoscience and Remote Sensing Letters, vol. 8, no. 3, pp. 441–445, 5 2011.

The continuous production and injection of hydrocarbons induce changes in reservoir pore pressure that modify the effective stress within the rock framework, resulting in compaction or expansion that manifests at the surface as subsidence and uplift. The Karamay oilfield in China provides a clear example of such deformation driven by evolving reservoir dynamics.

In this study, we assess the ground deformation across the Karamay oilfield using Sentinel-1 InSAR observations from 2017 to 2025 combined with geotechnical modeling. The InSAR time series reveals pronounced uplift and subsidence during the 2017–2018 period, with peak rates reaching approximately 118 mm/yr and –38 mm/yr, respectively. After this initial phase, deformation magnitudes declined substantially, with an approximately 80% reduction in peak uplift and ~50% reduction in peak subsidence by 2025, indicating progressive reservoir pressure equilibration.

Three-dimensional surface motion was reconstructed by integrating ascending and descending line-of-sight velocities, while the north–south component was estimated using a tilt-based approach derived from spatial gradients of the vertical deformation field observed during 2017–2018. The resulting mean deformation pattern is characterized by dominant vertical uplift accompanied by secondary horizontal divergence, consistent with reservoir inflation.

Geomechanical modeling was conducted using ascending LOS deformation during the 2017–2018 uplift phase to constrain reservoir geometry. These structural parameters were then held fixed to evaluate temporal pressure evolution in subsequent periods. The modeling results indicate strong reservoir pressurization during 2017–2018, reduced pressurization between 2019 and 2020, and a transition to pressure depletion and elastic compaction from 2020 to 2025, consistent with the observed LOS.

This study provides a detailed assessment of surface deformation in the Karamay Oilfield using multi-temporal Sentinel-1 SBAS InSAR observations. The deformation time series indicates a progressive transition from pronounced uplift to peak and spatially complex deformation, followed by persistent subsidence. These stages correspond to shifts in reservoir conditions from pressurization to stabilization and subsequent depletion, offering insight into the long-term geomechanical behavior of the reservoir. The agreement between the observed line-of-sight deformation patterns and the modeled pressure variations indicates that surface displacement is primarily governed by subsurface pressure evolution.

These findings improve understanding of the long-term mechanical response of the Karamay reservoir and demonstrate the value of combining multi-geometry InSAR observations with tilt-based 3D reconstruction and geomechanical modeling for monitoring deformation associated with hydrocarbon production.

Multi-Temporal Interferometric Synthetic Aperture Radar (MT-InSAR) is a consolidated approach for ground deformation investigations. However, the potentialities of the technique for structural health monitoring, especially by integrated methods based on multi-resolution and multi-frequency approaches, are still scarcely explored.

In this framework, the proposed approach employs the MT-InSAR technique to detect ground deformation phenomena and to monitor displacements and deformations across an asset of infrastructures. In particular, an integrated methodology combining MT-InSAR products, conventional monitoring techniques and Building Information Modeling (BIM), is presented. This integrated approach represents a pivotal component in the assessment of infrastructure safety, in the generation and definition of the digital twins of the built environment and, more in general, in the digital transition in civil engineering.

The methodology adopts a dual-scale analysis and employs multi-frequency satellite radar data from the European Space Agency's (ESA) Copernicus project and the Italian Space Agency (ASI). First, wide-area screening is performed using European Ground Motion Service (EGMS) calibrated products, processed from ESA Sentinel data, with a spatial resolution of 20 m x 5 m; this provides the ground deformations across the entire study area (the municipality of Modena, Italy). Second, high resolution (3 m x 3 m) X-band data from the ASI COSMO-SkyMed constellation (1st and 2nd generation) are processed via MT-InSAR to characterize displacements at the scale of single structures.

The novelty of this work lies in the procedure adopted to ensemble displacement data from MT-InSAR with BIM and structural analysis. This integration simplifies the classification and management of risk, as exemplified by a specific case study related to a bridge. This predictive maintenance paradigm allows for early intervention before significant damage occurs paving the way to the effective long-term monitoring of infrastructures with assessment of structural safety.

In the near future, the presented approach will benefit from the availability of recent and future SAR constellations, particularly those operating in L-band, such as the Argentinian SAOCOM (Satellite Argentino di Osservazione COn Microonde), NISAR (NASA-ISRO Synthetic Aperture Radar) and ROSE-L (Radar Observing System for Europe in L-band) by ESA. The availability of SAR data at different frequencies could enhance the structural analysis of infrastructures, facilitating the disentangling of ground deformation phenomena from structural behavior and supporting the safety assessment.

This work was supported by the Università di Modena e Reggio Emilia – Fondazione di Modena Project “Ensembling SATellite monitoring and BIM methods in the SAFety assEssment of road infrastructure (SAT‐SAFE)”, FAR 2024 - Bando per il finanziamento di progetti di ricerca interdisciplinari.

Underground natural gas storage systems constitute a fundamental component of modern energy infrastructure, ensuring supply security and market stability in response to seasonal demand fluctuations. In many countries, natural gas consumption increases significantly during winter months due to residential heating and industrial needs, leading to substantial withdrawal from storage facilities. Conversely, during warmer periods, excess supply is injected back into underground reservoirs to balance the system. These cyclic injection–withdrawal operations induce pressure variations within the storage formation, potentially altering the mechanical equilibrium of the reservoir and overlying strata.
Such pressure-driven changes may result in measurable surface deformation, particularly in geological settings where storage occurs within porous or cavernous formations. Although these deformations are often subtle, their cumulative or cyclic nature can have implications for infrastructure integrity, environmental safety, and long-term reservoir performance. Therefore, monitoring surface responses to underground gas storage operations is critical for distinguishing between natural hydrogeological trends and storage-induced mechanical effects, and for supporting sustainable energy management practices.
This study was conducted at the Tuz Gölü Underground Natural Gas Storage Facility in Türkiye and is based on a seven-year InSAR time series analysis covering the period from 2018 to 2025. Sentinel-1 SAR datasets were processed using the Small Baseline Subset (SBAS) approach. Atmospheric phase delays were corrected using ERA5 reanalysis data and ascending-descending orbit data were jointly analyzed to decompose surface displacements into vertical and horizontal components.
The results reveal that vertical surface deformation approaching approximately 6 cm per year occurs in the vicinity of the storage facility. Time series analyses demonstrate a clear temporal consistency between observed surface deformation patterns and gas injection–withdrawal cycles. Noticeable deformation changes correspond to increased gas withdrawal during winter months, while different deformation behavior is observed during reinjection periods in summer. These findings indicate that reservoir pressure variations associated with storage operations exert a direct influence on surface deformation patterns.
In addition to storage data, groundwater level variations in the region were evaluated to assess their potential contribution to surface movements. Long-term groundwater observations were statistically compared with deformation time series. Although initial correlation analyses suggested some degree of association, detrended analyses-performed to eliminate the influence of long-term trends-demonstrated that no strong or direct causal relationship exists between groundwater level changes and the observed surface deformation.
Overall, the integration of multi-temporal InSAR analysis, atmospheric corrections, component decomposition, and statistical evaluation indicates that the detected surface deformations are predominantly linked to natural gas storage operations. This study provides a robust methodological framework for monitoring surface-reservoir interactions in underground gas storage fields and contributes to the sustainable management of critical energy infrastructure.

Land subsidence is a widespread geohazard threatening infrastructure and water resources sustainability in many alluvial plains worldwide. Accurate characterization of its spatiotemporal dynamics requires integrating multiple observation techniques due to limitations inherent to individual sensors. This study proposes a multi-sensor data assimilation framework that combines Interferometric Synthetic Aperture Radar (InSAR) time series, Global Navigation Satellite System (GNSS) measurements and hydro-meteorological data to analyze and predict land subsidence in the Choushui River Alluvial Fan, Taiwan.

InSAR-derived deformation fields were calibrated using GNSS observations to generate a high-resolution deformation dataset. A Long Short-Term Memory (LSTM) neural network was then employed to model nonlinear temporal relationships between ground deformation and hydrological drivers, including groundwater level fluctuations and precipitation.

Results demonstrate that the proposed framework improves deformation accuracy and captures subsidence dynamics effectively, enabling short-term forecasting of cumulative subsidence and the generation of susceptibility maps. The methodology provides a transferable approach for large-scale subsidence monitoring and hazard mitigation in sedimentary basins worldwide.

InSAR without context is useful only for users with a high tolerance for errors. There’s probably no better demonstration of this statement than configuring InSAR displacement monitoring of manmade offshore structures, such as artificial islands or fixed oil & gas platforms.

Many processing challenges await the InSAR practitioner here: offshore sites are often far beyond the phase decorrelation distance from land, meaning reliable spatial reference points are absent. Moreover, continuous structural changes on offshore sites often invalidate traditional coherent scatterer pre-selection techniques, and the common assumption of long-term scatterer coherence no longer holds. Rapid, non-linear settlement, often exceeding the phase ambiguity, makes reliable phase unwrapping extremely challenging.

Under these conditions, InSAR can only deliver accurate and actionable results when supported by strong contextual constraints. With SkyGeo’s contextual InSAR approach, we show that we can still derive decision-critical insights for our customers even under these highly incoherent offshore conditions.

We present an ongoing project monitoring settlement during the construction of an artificial offshore island comprising 28 individual caissons deployed over two years.

Continuous monitoring was configured using both ascending and descending TerraSAR-X orbits. On-site context strictly drives the coherent point selection and phase unwrapping, as caissons are deployed with a multi-day lag, initially undergoing primary settlement of >10 cm within days, followed by secondary settlement due to installation of neighbouring caissons. To utilise acquisitions as soon as individual caissons become coherent, we apply the Temporary Coherent Scatterers approach [1]. The primary settlement is constrained using in-situ tachymetry and GNSS observations, while InSAR is used to estimate the secondary and long-term settlement. The InSAR time series functional model for unwrapping is guided by predicted caisson settlements from the structural simulation model.

Based on the structural design norms, each caisson has a strictly defined monitoring segment. Therefore, we use precise, sub-pixel geolocation of scatterers and careful aggregation on the individual monitoring segments. Because structural engineers work in 3D, not satellite line-of-sight or 2D projection, we estimate full 3D displacement vectors using strap-down decomposition [2], resolving seaside-landside tilt and vertical settlement.

Our contextual InSAR monitoring approach satisfies the settlement monitoring standards required by the structural engineering team. For most caissons, estimated values match the predicted settlements. During the winter construction pause, we also detected a significant (> 5cm) horizontal tilt of the southern wave-break wall, likely caused by temporary exposure of the inner caisson wall, not designed to bear the load of the predominant ocean currents. This finding provided valuable input for potential remedial actions.

Remote sensing always helps in boots-on-the-ground costs - but here we show how it is crucial for continuous monitoring of this island, as the construction site remains inaccessible for large portions of the year.

[1] Hu, F., Wu, J., Chang, L., & Hanssen, R. (2019). Incorporating Temporary Coherent Scatterers in Multi-Temporal InSAR Using Adaptive Temporal Subsets. IEEE Transactions on Geoscience and Remote Sensing, 57(10), 7658-7670. Article 8756305. https://doi.org/10.1109/TGRS.2019.2915658

[2] Brouwer, W. S., & Hanssen, R. F. (2024). Estimating three-dimensional displacements with InSAR: The strapdown approach. Journal of Geodesy, 98(12), Article 110. https://doi.org/10.1007/s00190-024-01918-2

Within the Italian Government's guidelines on space and aerospace matters, "Telecommunications, Earth Observation and Navigation" satellite services and applications (so-called "downstream") will be exploited by citizens and valorized by Institutions under an integrated application perspective. The Italian Space Agency (ASI) is committed to contribute to the downstream development, in order to support national policies related to many global challenges, including mitigation of weather-climatic events and the effects of global warming. Following the ASI’s roadmap for scientific downstream applications (Tapete & Coletta, 2022), ASI runs the “Innovation for Downstream Preparation for Science” (I4DP_SCIENCE) program devoted to the Scientific User Community, i.e. Italian Universities and Public Research Bodies.

I4DP_SCIENCE is composed of joint projects with ASI demonstrating the usefulness of either novel or consolidated methods and algorithms to support applications of users’ interest falling within topics of national relevance, e.g. defined by the National Copernicus User Forum, and/or falling within international agendas, e.g. the UN Sustainable Development Goals (SDGs). All the demonstrations are carried out jointly with the reference users who are actively engaged since the initial user requirement consolidation and, throughout the project, via capacity building and training activities towards the user uptake.

Interferometric Synthetic Aperture Radar (InSAR) techniques are nowadays very well established approaches for hazard assessment and have achieved a level of operational maturity that, in Italy, they are already exploited to support institutions and public administrations, e.g. for civil protection and structural monitoring purposes. This outcome has also been facilitated by the availability of long-term regular SAR image collections over the national territory not only via Sentinel-1 constellation, but also via the COSMO-SkyMed MapItaly Plan since 2011 and the systematic L-band SAOCOM acquisition plan, both coordinated by ASI. The combination of these observation capabilities provides an extraordinary wealth of multi-sensor SAR datasets over Italy that can enable a multitude of operational services and downstream applications.

However, further efforts are required in order to expand the portfolio of institutional users that can benefit from InSAR-based services, for example to better inform urban planning and land management. The present paper will showcase examples from the demonstration projects undertaken in the framework of ASI’s I4DP_SCIENCE programme.

An example is the recently completed GEORES project (Agreement ASI – UNIBA n. 2023-42- HH.0 – CUP F93C23000240005), funded by the Italian Space Agency (ASI) with University of Bari and the Institute for Electromagnetic Sensing of the Environment (IREA) of the National Research Council of Italy (CNR) (Lafortezza et al., 2024). Multi-mission InSAR deformation data have been used to address the Land Displacement module inputting into a multi-risk assessment framework to identify “hot-spots” of urban and peri-urban territory in Apulia region, southern Italy, at high risk from the point of view of land degradation caused by phenomena of hydrogeological instability, sediment flow or vegetation fires.

The paper will also discuss how visualisation and analytical tools such as WebGIS platforms are effective ways to disseminate InSAR data and associated thematic products, and enable final users to utilise these geospatial layers to support the decision making process.

References

Lafortezza R. et al. (2024) The GEORES Project: Geospatial Application in Support of Environmental Sustainability and Resilience to Climate Changes in Urban Areas. IGARSS 2024 - 2024 IEEE International Geoscience and Remote Sensing Symposium, Athens, Greece, pp. 1384-1387, https://doi.org/10.1109/IGARSS53475.2024.10642728

Tapete, D. and Coletta, A. (2022) ASI’s roadmap towards scientific downstream applications of satellite data, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-5643, https://doi.org/10.5194/egusphere-egu22-5643

Accurate three-dimensional (3D) displacement estimation is crucial for interpreting surface changes driven by natural and anthropogenic processes and for improving hazard assessment in areas affected by complex deformation. Interferometric Synthetic Aperture Radar (InSAR) is widely used for displacement monitoring due to its high spatial resolution and precision. However, InSAR measurements are limited to the satellite line-of-sight (LOS) direction. The decomposition of LOS signals into complete 3D displacement: north–south (dN), east–west (dE), and vertical (dU) remains challenging, particularly for the NS component, which usually exhibits the lowest accuracy.

To address this limitation, this study proposes an integration framework that combines LOS displacement measurements with independent 3D displacement data derived from laser scanning techniques, including airborne laser scanning (ALS), UAV-borne laser scanning (ULS), or UAV photogrammetry. For both laser scanning and photogrammetric datasets, 3D displacements were estimated from multi-temporal Digital Surface Models (DSMs) using Least Squares Matching (LSM) method implemented in the OPALS software. The joint decomposition is solved using a Weighted Least Squares (WLS) approach, enabling robust decomposition and improved estimation of all displacement components.

The approach was tested in a challenging mining area affected by underground exploitation, located near the Marcel Mine in the Upper Silesian Coal Basin (southern Poland). This type of environment is particularly demanding for InSAR due to decorrelation over rural areas and the potential for advanced time-series approaches to underestimating rapid or large deformations when their temporal models cannot capture the deformation dynamics. Therefore, we used a classical DInSAR workflow to generate cumulative LOS displacement maps. Sentinel-1 data from three viewing geometries (one ascending and two descending orbits) were used to maximize LOS diversity. ALS data were obtained from the national geodetic repository (geoportal.gov.pl), while ULS and UAV photogrammetry campaigns were carried out during one of our projects. For validation, a network of stabilized control points was measured using GNSS-RTN for horizontal displacements and precise leveling for vertical displacements, providing independent reference observations for accuracy assessment.

We evaluated several weighting strategies in WLS, including variants based on accuracy estimated from in situ reference data, quality indicators such as InSAR coherence, DSM roughness and their combinations, as well as equal weights. Integration with ULS achieved RMSE values of 0.040 m (dN), 0.031 m (dE), and 0.031 m (dU), while ALS integration yielded 0.102 m, 0.075 m, and 0.066 m, respectively. Photogrammetry-based integration provided RMSE values of 0.054 m (dN), 0.022 m (dE), and 0.016 m (dU). The most pronounced improvement was obtained for dN, with an accuracy gain of up to 90%. The best performance was achieved when observation weights were derived from reference-based accuracy estimates. Nevertheless, the equal-weight solution remained a practical alternative when reference data were unavailable.

The results confirm that multi-sensor integration within a WLS framework substantially enhances LOS decomposition and enables reliable retrieval of all 3D displacement components in demanding deformation settings. The approach is intentionally generic, as it is not restricted to a particular InSAR processing product (DInSAR or multi-temporal methods), and can incorporate additional observations whenever they are available and trustworthy. Likewise, the surface-based displacement input can be derived from different point-cloud sources, including laser scanning and photogrammetric dense image matching, making the methodology broadly applicable to monitoring tasks beyond mining subsidence such as landslides, open-pit mining, infrastructure deformation, and other high-risk environments.

This study assesses the structural vulnerability of built cultural heritage in the Historic Centre of Mexico City (CH-CDMX), a UNESCO World Heritage Site affected by long-term land subsidence. The research integrates satellite radar interferometry (InSAR) and geospatial data science to quantify, at multiple spatial and temporal scales, the impact of ground deformation on individual heritage structures.

The main objective is to develop an integrated methodological framework for subsidence monitoring based on structure-specific parameters. The workflow comprises: (1) integration and validation of heterogeneous geospatial datasets to generate a unified geodatabase of heritage assets (building footprints, addresses, and architectural plans); (2) application of high-resolution SAR interferometry to characterize regional subsidence dynamics across the CH-CDMX; (3) building-scale parameterization of four deformation patterns—differential settlement, apparent subsidence/emergence, and directional tilting; (4) implementation of a multi-criteria exposure index to identify the most affected structures; and (5) benchmarking the capability of openly available SAR data for cost-effective, continuous monitoring.

A total of 34 X-band SAR scenes acquired between 2011 and 2013 by TerraSAR-X in StripMap mode (3 × 3 m spatial resolution) were processed over multi-temporal windows up to two years, generating more than 430,000 line-of-sight deformation velocity measurements. Unlike conventional regional-scale assessments, deformation statistics were computed individually for each heritage building, enabling structure-specific evaluation of displacement gradients and tilt vectors. Operational thresholds were defined to classify structural behavior, identifying 46 buildings exhibiting critical deformation patterns. Building–environment interaction analysis revealed spatially coherent tilting trends predominantly oriented toward the southeast and northeast, as well as significant intra-structural differential gradients in monumental constructions.

Additionally, C-band SAR data from Sentinel-1 (~5 × 20 m spatial resolution) were processed to evaluate the performance of moderate-resolution, openly accessible imagery for monitoring structural deformation in dense historic urban environments. The comparative analysis demonstrates the trade-offs between spatial resolution and monitoring scalability, highlighting the potential of Sentinel-1 for sustained, low-cost surveillance of subsidence-prone heritage districts.

The five buildings exhibiting the highest structural vulnerability indices were Palacio Nacional, Antiguo Convento de la Encarnación, Palacio de Bellas Artes, Museo Nacional de Arte, and Museo Franz Mayer.

This research delivers the first integrated InSAR-based framework for structural vulnerability assessment of heritage buildings in CH-CDMX. The proposed methodology supports preventive conservation prioritization, demonstrates transferability to other subsiding megacities, and advances the application of SAR interferometry as an operational tool for cultural heritage risk monitoring.

The European Ground Motion Service (EGMS), as part of the Copernicus Land Monitoring Service, represents a paradigm shift in regional and continental geohazard monitoring, providing standardized InSAR displacement maps across Europe. However, the operational exploitation of these massive datasets remains a significant challenge for many end-users. The primary obstacles include the management of extremely large files in Text File Document (.txt) format, which often leads to software instability in standard Geographic Information Systems (GIS), and the high computational demand required for spatial subsetting and format conversion on local workstations. To address these bottlenecks, we present EGMStream, an innovative, high-performance web ecosystem designed to streamline the downstream processing of EGMS products. The technical core of EGMStream has evolved significantly from its original R-based implementation into a robust, server-side architecture. The current web-based version is built within a Python environment and encapsulated using Docker containers on a dedicated server infrastructure at the University of Florence. This containerized approach ensures maximum portability, eliminates library dependency conflicts, and guarantees consistent performance regardless of the host operating system. The most significant innovation is the implementation of a parallelized conversion engine. By distributing the processing load across multiple CPU cores, EGMStream can handle the ingestion and transformation of millions of persistent scatterers (PS) in a fraction of the time required by traditional single-threaded desktop tools, making the process virtually independent of the user's local hardware limitations. The user interface is meticulously organized to facilitate complex data management through a streamlined, two-panel interactive layout. The left panel serves as the primary control hub where users input EGMS download hyperlinks and define advanced processing parameters. A key feature is the ability to handle InSAR time series optionally, allowing users to reduce file size when only average velocity is required. The spatial management has been revolutionized with the introduction of an advanced cropping module. Users can define their Area of Interest (AoI) by uploading custom vector files in Shapefile, .KML, or .KMZ formats. This allows for precise data extraction tailored to specific administrative or geological boundaries. A fundamental pillar of EGMStream is its asynchronous operational logic. Recognizing that converting continental-scale datasets can be time-consuming, the application leverages a server-side queuing system. Once the process is initiated, the user is free to close the web browser or shut down their device, as the server handles the entire task autonomously. Upon completion, an automated routine generates a secure download link and notifies the user via email. Supporting multiple output formats such as GeoPackage, GeoJSON, and Shapefile, EGMStream ensures seamless interoperability with modern geospatial software. Accessible at https://egmstream.unifi.it/, this tool represents a critical contribution to the Open Science movement, democratizing access to European Ground Motion Service data and empowering researchers and practitioners to focus on geohazard interpretation rather than data pre-processing.

Mértola is a town in the south of Portugal, near the border with Spain, whose earliest evidence of human occupation dates back to the Iron Age. Over the centuries, Mértola has been witness to several civilisations, such as Roman and Islamic, all of them leaving marks of their cultural identities. One of the most distinctive features of Mértola is that many of the ancient structures have reached the present days in a good conservation state and functionality, providing invaluable information about the past occupation of the territory. The most distinctive structures are, nowadays, the castle and the wall that surrounds the ancient area of the town. However, the town is often subjected to severe meteorological events, such as the recent floods that inundated the port area and almost reached the town's defensive wall. The structural safety of the wall poses a special concern, as many infrastructures – municipal buildings, roads, residential properties, markets, museums – are located in its vicinity and will be highly affected in case of failure.

In the current context of climate change, monitoring of Mértola defensive wall is of the utmost importance to ensure the safety of people and property. ARTEMIS project – Applying Reactive Twins to Enhance Monument Information Systems – funded by Horizon Europe program, envisages to develop a reactive digital twin of Mértola wall. This twin will integrate data from several sensors to enable continuous, timely, and remote monitoring of the structure, capable of issuing warnings if any signs of anomalous behaviour are detected.

In this study, the potential of InSAR data to inform the reactive digital twin is evaluated. Data from the European Ground Motion Service (EGMS) was collected and used to evaluate displacements in Mértola historic centre, including the wall. Data from different satellite orbits were combined to achieve vertical and east-west components of displacements, and spatiotemporal patterns were analysed. Dissimilarities between displacement time series were evaluated and used to form clusters of measurement points with similar behaviour. There were no spatial constraints imposed on the clusters, as they were formed based on the similarity of the displacement time series alone. The identification of spatial clusters of points exhibiting anomalies in their displacement time series may indicate potential structural instability. In the case of Mértola historic centre, a method highly sensitive to the presence of outliers was employed to compare the displacement time series. This strategy enabled the automatic aggregation of measurement points according to their displacement magnitude, movement direction and the presence of trends or discontinuities in their displacement time series. Recent structural occurrences suggest some degree of deterioration of the wall, which should be interpreted considering the trend of the measured displacements. Additionally, while providing insights into the evolution over the past few years, the InSAR analysis results also supported decision-making regarding the high-risk areas for the installation of the monitoring system.

The results show that InSAR is a promising method for inclusion in digital twins of cultural heritage sites, as it is a non-destructive monitoring method and each measurement point can be used as an individual sensor contributing to the model. This preliminary study used EGMS data from 2019 to 2023, but the findings indicate that it is worth investing in a near-real-time InSAR monitoring system and applying systematically the proposed method.

Since their publication in mid-2020, the European Ground Motion Service (EGMS) products have been seen as a breakthrough in InSAR applications for the analysis and monitoring of natural and man-made hazards on Earth’s surface. The EGMS measurements reliable InSAR measurements of ground deformations and include three types of products: (i) Basic (Level 2A) which provides line-of-sight velocity maps in ascending and descending orbits referenced to a local point; (ii) Calibrated (Level 2B) provides GNSS-calibrated full-resolution velocity and displacement time series for the ascending and descending orbits; (iii) Ortho (Level 3) calculated displacement vectors in the vertical and E-W directions, resampled to a 100 x 100 m grid. These datasets are available for five-year periods, except for the 2015-2020 period, which constitutes the EGMS baseline, followed by the 2018-2022 and 2019-2023 periods, and will continue to be updated until 2027, according to the official reports, providing continuity to existing datasets.

Their versatility makes them useful for a wide range of analyses and investigations of many geological and geomorphological processes, including slope-related processes, land subsidence, sinkholes, volcanic activity, and more. Also, their availability enables the assessment and monitoring of structural and infrastructure displacements, aiming to mitigate potential hazards that could affect society. Although the products clearly increased the use of InSAR across many studies, their full potential remains to be unlocked. Their underuse is mainly limited by the interpretability of the ascending and descending orbit measurements, which are not always straightforward, whereas the Ortho products average deformation velocities, making them unfit for local-scale investigations such as landslide and building monitoring.

In this work, we aim to minimise this shortcoming by developing a command-line workflow in the R environment that computes displacement vectors in the vertical and east-west directions at the measurement-point level. Basically, every single calibrated ascending or descending InSAR measurement point (MP) is considered for the computation of displacement components by searching for its nearest neighbour from the other orbit. For the identified pair of points, the middle point is computed, and for that point, the mean velocity and displacement time series for the vertical and East-West components are calculated. The resulting synthetic points, along with time series and mean velocity for the vertical and east-west directions, will significantly improve understanding of local deformations. We tested our approach on sinkholes, slope-related deformations, and mining activity in different environments, delivering results that would have provided important information prior to the event.

Interferometric Synthetic Aperture Radar (InSAR) has become a central tool in geoscience for detecting ground deformation with high spatial detail across broad regions. Its value has been repeatedly demonstrated for monitoring landslides, subsidence, volcanic processes, and other surface changes. However, despite the method’s scientific maturity, many potential users still face substantial challenges in carrying out complete time-series analyses. The processing chain typically involves numerous software dependencies, command‑line operations, data‑handling steps, and parameter choices that can discourage newcomers and slow down non‑specialist workflows. As a result, there remains a demand for accessible, end‑to‑end environments that simplify routine tasks while still allowing scientists to make informed methodological decisions.

To address this gap, we introduce a new open‑source graphical interface built in Python that guides users through the full workflow for generating Sentinel‑1 deformation time series. The platform wraps established processing components from GMTSAR and organizes them into a structured, step‑by‑step framework that reduces the need for command‑line scripting. Instead of replacing expert judgement with automated defaults, the environment emphasizes transparency and user interaction by exposing intermediate outputs and critical parameters at each processing stage.

The interface covers the complete processing sequence, beginning with software preparation and project initialization. It supports automatic searching, downloading, and preparing of Sentinel‑1 scenes, as well as the retrieval of orbit files needed for accurate co‑registration. Baseline calculations, interferogram network design, and subsequent interferogram generation are handled within guided modules that summarize relevant information in an intuitive layout. Users can inspect spatial baselines, refine the connectivity of the interferometric network, and evaluate alternative choices before proceeding.

Further utilities guide users through alignment and interferograms generation for time‑series analysis. A flexible and optional masking module allows users to define exclusion zones based on correlation thresholds, manually drawn polygons, or both, helping to remove unreliable areas for unwrapping, thereby enhancing the robustness, quality and efficiency of the process. Optional atmospheric corrections are supported through direct interaction with the Generic Atmospheric Correction Online Service, allowing users to incorporate external models before the inversion process.

The environment also includes interactive visualization capabilities, enabling users to explore deformation patterns, assess temporal behaviour, and export results for further analysis. These tools are intended to help users interpret processing choices and understand the consequences of different parameter configurations throughout the workflow.

Surface deformation observed by multi-temporal InSAR contains information on subsurface hydrological processes through poroelastic coupling; however, conventional groundwater inference typically relies on load and poroelastic Green’s function approaches, sometimes constrained by gravity variations at coarse scales, and the resulting inversion remains nonlinear, ill-conditioned, and dependent on strong assumptions about deformation sources. We present a physics-aware Swin3D U-Net framework that learns the spatiotemporal mapping between InSAR-derived deformation and groundwater storage anomalies, allowing high-resolution InSAR data to be exploited for aquifer-level groundwater inference rather than being constrained by coarse-scale analytical inversion schemes.

The model ingests multi-temporal InSAR deformation stacks, optionally combined with auxiliary hydrological variables, and represents them as 3D spatiotemporal tokens processed through hierarchical shifted-window self-attention. This architecture captures local and long-range dependencies across space and time while preserving multi-scale structure. A U-Net–style encoder–decoder with skip connections reconstructs full-resolution groundwater storage fields. Physical consistency is encouraged through constraints linked to poroelastic response and regional water-mass balance.

Training and evaluation are performed on aligned deformation and hydrological model datasets covering the period 2015–2025 over the Emilia-Romagna region, Italy, an intensively exploited aquifer basin exhibiting pronounced subsidence. The network reproduces coherent groundwater storage dynamics from surface motion signals, resolving basin-scale variability while retaining sensitivity to aquifer-level spatial patterns.

From a computational perspective, GPU implementation provides a 7.5× increase in inference speed relative to convolutional baselines, enabling regional-scale monitoring.

Keywords: Groundwater Storage Retrieval, Swin3D-UNet, Spatiotemporal Inversion, Poroelastic Lag, Multimodal Learning

Accurate deformation monitoring using Interferometric Synthetic Aperture Radar (InSAR) remains challenging in densely vegetated and geomorphically active environments where temporal decorrelation significantly limits the availability of persistent radar scatterers. This limitation is particularly critical along transportation corridors in the southeastern United States, where highway infrastructure frequently traverses forested terrain prone to slow-moving landslides and slope instabilities. Conventional InSAR time-series approaches, including Persistent Scatterer Interferometric Synthetic Aperture Radar (PS-InSAR) technique, often struggle to maintain signal coherence in such environments, resulting in sparse measurement coverage and increased uncertainty in deformation estimates.

This study investigates the deployment of artificial trihedral corner reflectors (CRs) as stable radar targets to enhance InSAR-based deformation monitoring along highway corridors. A network of three CRs was installed in January 2026 along the Littleville corridor (U.S. Highway 43) in Alabama, a site identified through multi-year satellite observations as exhibiting active slope deformation. Historical observations from the Sentinel‑1 satellite mission indicate cumulative line-of-sight displacements exceeding approximately 60 mm between 2017 and 2025. However, the dense forest canopy in this region significantly limits the presence of natural coherent scatterers that are typically required for reliable InSAR time-series analysis. The deployed CR network therefore provides controlled, high-coherence radar targets that can support both calibration of satellite-derived displacement fields and validation of deformation measurements in areas where conventional approaches are challenged.

Each reflector was carefully oriented to match the acquisition geometry of Sentinel-1 ascending passes, ensuring optimal radar return and consistent detectability across acquisitions. Prior to installation at the Littleville site, a calibration reflector was deployed at the National Center for Asphalt Technology test facility to verify installation procedures and evaluate satellite detectability under known conditions. The reflectors were successfully detected in the first Sentinel-1 acquisition following installation, demonstrating strong coherent radar backscatter responses and confirming appropriate geometric alignment with the satellite look direction.

A key objective of this work is to evaluate how the presence of CR targets improves the reliability and calibration of different InSAR time-series processing methodologies. At the reflector locations, deformation time series will be derived using both the Persistent Scatterer Interferometric Synthetic Aperture Radar approach implemented within the Generic Mapping Tools Synthetic Aperture Radar processing framework (GMTSAR) and the New Small Baseline Subset time-series approach implemented through the LiCSBAS processing framework, which applies small-baseline interferometric time-series analysis to Sentinel-1 interferometric stacks. The CR targets provide well-defined, phase-stable reference points that enable direct comparison between PS-InSAR and small-baseline time-series displacement estimates, allowing evaluation of consistency, noise characteristics, and potential biases between the two processing strategies.

The reflector network is further integrated with ground-based monitoring data to support validation of satellite-derived deformation measurements. In previous investigations conducted along other Alabama highway corridors, subsurface inclinometer measurements installed by the Alabama Department of Transportation were used to evaluate the performance of InSAR-derived deformation estimates in densely vegetated terrain. These comparisons demonstrated that while satellite-based InSAR techniques can successfully detect long-term deformation trends, the absence of stable radar scatterers often limits spatial coverage and introduces uncertainty in reference selection. The deployment of dedicated CR targets addresses this limitation by providing stable phase anchors that improve the robustness of time-series analysis and facilitate direct comparison with in-situ measurements.

Beyond calibration and validation, the CR network provides an experimental testbed for evaluating the performance of multiple satellite missions in vegetated landslide environments. As additional Sentinel-1 acquisitions accumulate, the reflectors will enable millimeter-scale assessment of deformation time series derived from both persistent scatterer and small-baseline processing workflows. Furthermore, the established reflector infrastructure is well positioned to support future cross-mission comparisons with the upcoming NASA‑ISRO Synthetic Aperture Radar (NISAR) mission. The longer L-band radar wavelength of NISAR is expected to improve coherence in forested environments, providing an opportunity to evaluate the complementary capabilities of C-band and L-band radar observations for monitoring landslides affecting critical transportation infrastructure.

The results demonstrate that strategically deployed corner reflectors can substantially improve InSAR observability in densely vegetated terrain by introducing bright, phase-stable artificial scatterers at controlled locations. This approach enables improved calibration of satellite-derived deformation measurements and supports rigorous validation of multiple InSAR time-series techniques, ultimately advancing the use of satellite remote sensing for operational monitoring of landslides along highway infrastructure corridors.

With rapid global urbanization, Interferometric Synthetic Aperture Radar (InSAR) has become a crucial tool for urban infrastructure safety monitoring due to its all-weather capability, wide-area coverage, and millimeter-level sensitivity to deformation (Wu et al., 2023). However, in dense urban high-rise environments, the side-looking SAR imaging geometry produces severe geometric distortions (e.g., layover and shadow), which strongly affect the visibility of individual buildings and the reliability of deformation measurements, introducing a “pixel-to-building” gap in urban InSAR practice. Consequently, before reliable deformation analysis can be conducted, two prerequisite challenges should be addressed. At the pixel level, scatterer attribution is critical to reliable deformation interpretation. Establishing an accurate link between SAR scatterers and real-world objects remains a key challenge (Yang et al., 2019). At the building level, owing to the impact of geometric distortions, it is equally essential to determine whether a building is suitable for InSAR monitoring under a given acquisition geometry. To enhance the engineering practicality of urban building InSAR monitoring, this study proposes an integrated supporting framework for urban building InSAR applications. The framework provides systematic solutions to two key prerequisite problems, including pixel-level 3D structure recovery and building-level monitoring-target selection.

First, to establish an accurate link between scatterers and real-world objects, 3D InSAR point cloud reconstruction provides an effective solution by projecting SAR imagery into the 3D spatial domain. However, existing methods exhibit obvious limitations. SAR Tomography (TomoSAR) (Zhao et al., 2025; Zhu et al., 2016) demands dense spatial baselines and large data stacks, which limits its applicability over large regions. Persistent Scatterer Interferometry (PSI) (Ferretti et al., 2001) relies solely on highly coherent scatterers, leading to incomplete InSAR point cloud structures. Traditional Look-Up-Table (LUT)-based backward geocoding also suffers from positional ambiguity in layover areas, making it difficult to reconstruct the vertical structure of buildings accurately. To address these issues, we propose a Digital Surface Model (DSM)-assisted 3D InSAR point cloud reconstruction method, incorporating an Approximate Iso-Doppler (AID) forward geocoding approach and a dominant scatterer identification strategy. The AID method constructs an approximate iso-Doppler plane that intersects the DSM to derive terrain profiles. By searching for intersections using tangent lines to iso-range contours, it retrieves the 3D positions of potential scatterers within each SAR pixel while simultaneously detecting shadow and layover. To further mitigate layover-induced location ambiguity, dominant scatterers are identified via local fringe-frequency estimation from a single interferogram (Rossi and Eineder, 2015), enabling each layover pixel to be assigned to its most likely scattering source. The proposed InSAR point cloud reconstruction method was validated over the Wuhan University campus and surrounding area (~3 km²), using two TerraSAR-X Stripmap-mode images together with a 1 m LiDAR-derived DSM. Experimental results show that the AID geocoding method achieves a residual accuracy better than 10^-3 pixels relative to the analytical solution of the rigorous Range-Doppler (R-D) model and is also over six times faster than traditional backward geocoding. The proposed approach can also identify typical slope clusters corresponding to roofs/ground and façades, providing a more complete 3D InSAR point-cloud foundation for subsequent deformation monitoring.

Second, after obtaining 3D InSAR point clouds, another key question in engineering practice must be answered. In dense urban environments, which buildings are suitable for InSAR monitoring? Currently, there is a lack of applicability assessment methods specifically targeted at the urban building level. Previous studies have mostly focused solely on visibility analysis (Del Soldato et al., 2021), overlooking the completeness of the deformation information and the interpretability of the results. To address this, we propose a new building-oriented applicability assessment method within the upper module of the supporting framework. This framework introduces three complementary quantitative evaluation indices: (1) the Visibility Index (VID), which evaluates whether a building is observable under a given SAR viewing geometry; (2) the Deformation Completeness Index (DCI), which evaluates whether the visible building structures can provide sufficiently complete and unbiased deformation information , thereby avoiding biases dominated by a single component (such as the roof); and (3) the Deformation Interpretability Index (DII), which evaluates whether the extracted deformation signal can be clearly attributed to a specific building, thereby suppressing signal confusion caused by layover from neighboring buildings. All the aforementioned indices are calculated based on 3D InSAR point clouds and building footprint data.

The proposed framework is validated over the main urban area of Shenzhen using high-resolution TerraSAR-X data, an open-source DSM (Zhu et al., 2025), and building footprint data. By combining the geometric characteristics of the 3D InSAR point cloud with the applicability assessment results, the buildings in the study area are classified into four categories: invisible (33.26%), visible but with incomplete deformation information (4.59%), visible but with deformation that is difficult to interpret (5.11%), and well-suited for monitoring (57.04%). The results indicate that more than half of the buildings are suitable for InSAR monitoring under the current viewing geometry, while layover and shadow remain the main limiting factors in dense high-rise areas.

In conclusion, by integrating pixel-level 3D structural recovery with building-level applicability evaluation, the proposed supporting framework mitigates the inherent interpretation ambiguities caused by geometric distortions in dense urban areas. It provides a key theoretical and practical basis for future large-scale, standardized, and automated urban InSAR monitoring systems.

References

Del Soldato, M., Solari, L., Novellino, A., Monserrat, O., Raspini, F., 2021. A New Set of Tools for the Generation of InSAR Visibility Maps over Wide Areas. Geosciences 11, 229. https://doi.org/10.3390/geosciences11060229

Ferretti, A., Prati, C., Rocca, F., 2001. Permanent scatterers in SAR interferometry. IEEE Trans. Geosci. Remote Sensing 39, 8–20. https://doi.org/10.1109/36.898661

Rossi, C., Eineder, M., 2015. High-Resolution InSAR Building Layovers Detection and Exploitation. IEEE Trans. Geosci. Remote Sensing 53, 6457–6468. https://doi.org/10.1109/TGRS.2015.2440913

Wu, S., Zhang, B., Ding, X., Zhang, L., Zhang, Zhijie, Zhang, Zeyu, 2023. Radar Interferometry for Urban Infrastructure Stability Monitoring: From Techniques to Applications. Sustainability 15, 14654. https://doi.org/10.3390/su151914654

Yang, M., López-Dekker, P., Dheenathayalan, P., Liao, M., Hanssen, R.F., 2019. On the value of corner reflectors and surface models in InSAR precise point positioning. ISPRS Journal of Photogrammetry and Remote Sensing 158, 113–122. https://doi.org/10.1016/j.isprsjprs.2019.10.006

Zhao, X., Dong, J., Yu, Y., Liao, M., Zhang, L., Gong, J., 2025. A review of SAR tomography. Geo-spatial Information Science 1–44. https://doi.org/10.1080/10095020.2025.2510365

Zhu, X.X., Chen, S., Zhang, F., Shi, Y., Wang, Y., 2025. GlobalBuildingAtlas: an open global and complete dataset of building polygons, heights and LoD1 3D models. Earth Syst. Sci. Data 17, 6647–6668. https://doi.org/10.5194/essd-17-6647-2025

Zhu, X.X., Montazeri, S., Gisinger, C., Hanssen, R.F., Bamler, R., 2016. Geodetic SAR Tomography. IEEE Trans. Geosci. Remote Sensing 54, 18–35. https://doi.org/10.1109/tgrs.2015.2448686

Coastal environments worldwide are facing compounded threats from climate-driven Sea Level Rise (SLR) and Vertical Land Motion (VLM) (Shizaei et al., 2021; Nicholls et al., 2021). Accurate quantification of Relative Sea Level Rise (RSLR), particularly by resolving the spatial heterogeneity of VLM, over large spatiotemporal scales is critical for effective risk management (Tay et al., 2022). However, this remains technically challenging due to the complexities of generating seamless wide-area InSAR deformation fields and integrating them into dynamic hydrological models.

As detailed in our recently published works (Gong et al., 2025; Gong et al., 2026), we present a comprehensive framework that bridges advanced multi-frame InSAR processing with hydrodynamic inundation modeling, using the Bohai Rim, a densely populated and geologically complex coastal zone in China, as a case study. In this framework, the ESA Sentinel-1 mission plays an indispensable role. Its systematic observation strategy, vast spatial coverage, and highly reliable C-band acquisitions provide the critical data foundation required for continuous, continent-scale deformation monitoring.

First, to address the limitations of conventional InSAR merging, specifically inter-frame inconsistencies and systematic biases arising from varying incidence angles and atmospheric artifacts, we developed a novel adaptive gridded adjustment model. This method employs a quadtree-based decomposition algorithm that dynamically optimizes grid sizes based on deformation gradients, integrated with sparse GNSS constraints to remove long-wavelength orbital and atmospheric errors. Applied to 2,078 Sentinel-1 images from 2018 to 2022 across six frames, this approach reduced inter-frame discrepancies by 38%, and achieved a root-mean-square error (RMSE) of 0.4 mm/yr against GNSS validation. The resulting seamless deformation map reveals significant spatial heterogeneity in VLM, ranging from -141 mm/yr to +40 mm/yr, driven by diverse factors including groundwater depletion, hydrocarbon extraction, and sediment compaction.

Building on this high-precision geodetic baseline, we demonstrate the critical role of VLM in RSLR assessment through dynamic inundation simulations under multiple climate scenarios. we assessed future inundation risks under IPCC AR6 scenarios (SSP1-2.6, SSP3-7.0, and SSP5-8.5) by 2100 using a "Flow-tub" dynamic inundation model. This approach advances beyond static "bathtub" methods by incorporating hydrological connectivity and path-based water level attenuation. Crucially, our results identify VLM as the dominant driver of local RSLR variability. Under the high-emission scenario (RSLR-SSP5-8.5), incorporating of InSAR-derived VLM expands the projected inundation extent to 17,756 km², representing a 33-50% increase compared to SLR-only projections (13,314 km²), and potentially exposing 10.4 million residents to flood risks.

In conclusion, this study combines advanced land deformation monitoring with flood modeling. We demonstrate that ignoring VLM significantly underestimates coastal risks. The proposed InSAR method offers an effective solution for large-scale monitoring. Furthermore, our findings provide practical guidance for coastal protection, emphasizing the urgent need to focus on areas with severe subsidence.

Building upon the success of this regional framework, we are currently expanding this integrated approach to the entire coastline of mainland China. This ongoing scale-up initiative aims to bridge existing research gaps in national scale VLM quantification and deliver a comprehensive, high-resolution coastal hazard assessment to support resilient coastal planning.

Keywords: Vertical Land Motion (VLM); Multi-frame InSAR; Adaptive Gridded Adjustment; Relative Sea Level Rise (RSLR); Dynamic inundation Modeling;

References:

[1] Gong, Z., Liao, M., Dong, J., Lan, Q., Wang, R., Lai, S. (2025). Wide-area coastal deformation extraction using multi-path/frame InSAR: A case study of the Bohai Rim. Remote Sensing of Environment, 114988.

[2] Gong, Z., Wu, J., Dong, J., Lan, Q., Lai, S., Lin, J., Liao, M. (2026). Relative sea-level rise and inundation risks in the Bohai Rim: Dominant role of vertical land motion. International Journal of Applied Earth Observation and Geoinformation, 105115.

[3] Shirzaei, M., Freymueller, J., Törnqvist, T. E., Galloway, D.L., Dura, T. (2021). Measuring, modelling and projecting coastal land subsidence. Nature Reviews Earth & Environment, 2(1), 40-58.

[4] Nicholls, R. J., Lincke, D., Hinkel, J., Brown, S., Vafeidis, A.T., Meyssignac, B., Hanson, S.E., Merkens, J.-L., Fang, J. (2021). A global analysis of subsidence, relative sea-level change and coastal flood exposure. Nature Climate Change, 11(4), 338–342.

[5] Tay, C., Lindsey, E.O., Chin, S.T., McCaughey, J.W., Bekaert, D., Nguyen, M., Hua, H., Manipon, G., Karim, M., Horton, B.P., Li, T., Hill, E.M. (2022). Sea-level rise from land subsidence in major coastal cities. Nat. Sustain. 5, 1049–1057.

Subsidence due to pressure decline from extensive groundwater extraction is one of the most common surface deformations detected by PSInSAR. This is exactly what is occurring at our case study site in Mezőtárkány, Hungary, where groundwater extraction driven by increased demand for drinking water, as well as industrial use has caused significant subsidence in the settlement's southeastern part.
We processed both ascending and descending Sentinel-1A data from 2014 to 2024 using the IPTA method and a dual master approach. To achieve this, we combined results from two independent 5-year single-reference processing runs, selecting two reference epochs in 2017 and 2021. This ensured high-quality interferograms for displacement determination despite the long study period, thereby determining displacements in a total of 274 epochs in the descending orbit and 260 epochs in the ascending orbit.
Time series confirm that the detected anomaly persists across the entire ~10-year study interval. The surface has subsided at a nearly uniform rate, exceeding 1 cm/year in the central area at the location of the well, resulting in total subsidence of over 10 cm during the period. In contrast, the western–southwestern side of the settlement, farther from the well, experienced less than 0.5 cm of displacement over the entire decade. These data indicate that the subsidence is truly local and most likely linked to the groundwater extraction.
Surface deformation reflects subsurface hydrological processes, where substantial fluid withdrawal causes pressure decline in the rock matrix, leading to its compaction. We investigated these subsurface processes using groundwater flow modeling based on geological and production data provided by the water utility, using the MODFLOW modeling package. We compared the temporal patterns of production and modeled compaction, including the effects of delayed-response layers, the spatial shape and extent of the subsidence bowl, and the impacts of varying model and layer parameters on these features. Ultimately, we developed a model that best fits the satellite measurement data, assuming homogeneous layers. This model reveals the impacts of groundwater extraction, predicts future subsidence from continued overproduction, and identifies unaffected areas. The latter indicates aquifers uninvolved in overexploitation, whose location may be ideal for establishing new production wells in the future.

Ground deformation in urban environments can pose risks to high‑value infrastructure, yet dedicated structural monitoring is often limited or short‑lived. Interferometric Synthetic Aperture Radar (InSAR) provides a long‑term, independent record of displacement, but interpreting InSAR measurements on tall buildings requires approaches that account for complex motion and structure geometry. This work presents an InSAR‑based methodology developed by 3vGeomatics to reconstruct multi‑year displacement histories of high‑rise buildings, demonstrated using the Millennium Tower in San Francisco as well as several additional nearby skyscrapers.

For Millennium Tower—a 58‑story (197 m) residential tower reported in 2016 to be experiencing settlement and tilt—we processed 225 descending and 107 ascending TerraSAR‑X StripMap scenes collected between 2009 and 2018 at 3 m resolution. From these data, we generated more than 2000 persistent scatterer measurement points distributed across all four façades and the roof. By modeling the building as a rigid body capable of vertical motion and tilting around a pivot near the base, we reduce thousands of InSAR measurement points into a small number of interpretable parameters: vertical displacement, tilt magnitude, and tilt direction. The approach is designed to work even with a single satellite viewing geometry. This flexibility is especially important in dense city centers, where certain viewing geometries may be partially obstructed by neighboring buildings. Analyzing ascending and descending data independently provides two self‑consistent solutions, while a combined fit yields the most precise displacement history.

Across the nine‑year record, Millennium Tower was found to have undergone approximately 20 cm of vertical settlement and 45 cm of lateral movement at the rooftop, with both the rate and direction of tilt evolving over time. To quantify parameter uncertainties, the same modeling procedure was applied to multiple nearby skyscrapers with suitable data. Assuming these comparison buildings to be stable, the variance in their fitted parameters provided empirical uncertainty estimates: 2 cm for vertical displacement and 5 cm for tilt (95% confidence).

Extending this analysis to other high‑rise structures in the surrounding district, we demonstrate the broader applicability of this method for detecting subtle, long‑term building deformation. These results highlight how multi‑year InSAR archives, combined with appropriate structural modeling, can support urban risk assessment, infrastructure management, and engineering investigations for large construction projects.

Interferometric Synthetic Aperture Radar (InSAR) reanalysis studies often mis-represent the true capabilities of InSAR for Near Real-Time (NRT) monitoring. While post-event retrospective analysis, benefiting from complete datasets and hindsight, reliably identifies precursory displacement, this approach fails to address the critical question for operational utility: when would a real-time decision-maker identify a threat and be able to take action?

Retrospective analysis usually presents InSAR timeseries processed for all the available data up to the point of failure. However, the sources of error are rarely consistent in time and the lack of bounding data results in greater uncertainties at the start and end of the timeseries. Thus the data available to decision makers at any point is not simply the truncated version of the full timeseries that would tend to over-represent the power of InSAR to resolve early signs of e.g. acceleration. This paper investigates this divergence through comparative analyses of a prominent failure event where definite precursory displacement was observed. We generate the successive results that would be available to the end-user at each stage and compare them to the eventual retrospective-analysis product. At each stage we assess what other zones within the dataset exceed similar thresholds. Our case study demonstrates that while significant precursory displacement exists, its utility for raising an alarm in NRT is dependent on the inherent spatial and temporal variability of InSAR sensitivity (e.g., residual atmospheric phase, coherence drop).

To further generalize, we extend the case-study results with simulated displacement patterns injected into multiple real SAR datasets complete with the various operational constraints (e.g. typical acquisition schedules, atmosphere, coherence degradation). We argue that the primary challenge for NRT monitoring is not just estimating displacement, but establishing reliable measures of significance across the entire monitoring area. To achieve this, quantitative assessment of the sources of error (variance and covariance) is essential. We advocate for a shift in priority from localized displacement accuracy to the ability to detect regions of displacement and/or acceleration with characterization of false-positive and false-negative rates. Ultimately, for InSAR to transition from a retrospective analysis tool to an effective NRT monitoring data stream, a framework for quantifying significance is mandatory for managing the high-consequence risks associated with catastrophic events.

Abstract

Large-scale land reclamation over thick, compressible marine deposits presents significant geotechnical challenges, particularly in understanding post-construction consolidation behavior. This study presents a comprehensive framework for four-dimensional (space and time) mapping of soil consolidation at the Hong Kong International Airport’s (HKIA) three-runway system expansion, where extensive reclamation employed two key ground improvement techniques: Prefabricated Vertical Drains (PVD) and Deep Cement Mixing (DCM)—the latter being first applied in Hong Kong for this project. We applied multitemporal Synthetic Aperture Radar Interferometry (MT-InSAR) using multiple SAR data stacks to investigate the spatiotemporal characteristics of land deformation following runway pavement completion. A combined persistent scatterer (PS) and distributed scatterer (DS) strategy was implemented to address low radar coherence at the site. Our observations reveal varying degrees of land subsidence, with maximum sinking rates reaching ~150 mm/year during September 2021–October 2023 and subsequently evolving to ~80 mm/year by September 2025. Three-dimensional deformation fields further detected apparent horizontal displacement in the taxiway southwest of the runway.

To disentangle the complex consolidation mechanisms, we employed Independent Component Analysis (ICA) to identify underlying sources contributing to the measured deformation. Three distinct signals were unveiled: (1) an exponential decay signal representing rapid compaction of surficial materials, (2) a linear signal indicating continuous subsidence from marine deposits—predominantly located in areas with prefabricated vertical drains, and (3) a periodic signal associated with thermal effects on structures. Integrating InSAR observations with Terzaghi consolidation theory and 3D finite element modelling enabled us to extend surface displacement monitoring to subsurface characterization. This approach allowed retrieval of critical geotechnical parameters, including ultimate primary settlement, compression index (C_c), the coefficient of consolidation (C_v) and the degree of consolidation at unprecedented spatial detail. Mapping these derived parameters revealed distinct spatial heterogeneities that strongly correlated with the different treated foundations. This coupled approach captured the time-dependent consolidation behavior and quantified the distinct stability performance of PVD and DCM zones. Our quantitative analysis demonstrates that DCM achieves geological stability more rapidly than PVD, with a time advantage of approximately 0.08–1.39 years, while effectively controlling primary settlement to 29%–83% of that observed with PVD. This integrated framework advances traditional InSAR applications by transforming surface deformation measurements into actionable insights on underground consolidation processes. The findings provide a crucial quantitative basis for evaluating residual settlement, planning targeted reinforcement measures, and informing future reclamation practices—particularly those employing innovative DCM techniques in similar geological settings.

Acknowledgements: National Natural Science Foundation of China (Grant No. 42304052), the Research Grants Council of Hong Kong (Grant No. 15229523, No. 25202125) and the Otto Poon Research Institute for Climate-Resilient Infrastructure (Grant No. P0055919).

Water cycle infrastructures are critical for adapting society to current and future climate scenarios. Among them, dams play a key role in water supply and regulation, while their safety frameworks must evolve to address emerging climatic and operational challenges. At the same time, infrastructure monitoring is rapidly advancing due to progress in remote sensing, modelling and machine learning. Dam managers increasingly face a context with multiple monitoring data sources—such as traditional instrumentation, remote sensing observations and environmental variables—that are often not fully integrated, which can hinder the operational adoption of new technologies.

Recent advances in Synthetic Aperture Radar interferometry (InSAR) enable quasi-real-time monitoring of dams and surrounding infrastructures with high spatial coverage and millimetric sensitivity. The integration of multi-temporal InSAR (MT-InSAR) with classical instrumentation and contextual information provides new opportunities to improve dam safety assessment. Standardized databases and integrated analysis frameworks facilitate consistent data management and enable joint analysis of deformation measurements and external drivers.

In this context, this work proposes a modelling framework based on model that integrates interpretable linear trend models, breakpoint detection and the estimation of linear dependence on external factors such as temperature and reservoir level. The approach enables assessing whether dam deformation follows the expected behaviour or evolves anomalously, while breakpoint detection supports a posteriori analysis of potential links between deformation changes and extreme events.

By modelling deformation behaviour as a function of external variables that are easily measurable or routinely available, the proposed framework contributes to improving the interpretability and operational value of satellite-based monitoring. Such models support early detection of anomalous behaviour and strengthen the integration of remote sensing into dam safety practices, ultimately enhancing the safety and resilience of water infrastructures in the context of ongoing global change.

Differential interferometric synthetic aperture radar (DInSAR) has become a powerful technique for estimating precise surface deformation on Earth's dynamic surface. With the increasing availability of spaceborne synthetic aperture radar (SAR) observations, multi-temporal InSAR (MT-InSAR) enables the retrieval of long-term displacement time series. However, time-series displacement signals are often contaminated by various components, including coupled deformation, atmospheric delays, noise, and seasonal effects, despite numerous approaches to mitigate these artifacts. Such components can lead to misinterpretation of actual deformation patterns or their driving mechanisms. To isolate displacement-related signals from mixed time-series data, several statistical methods, such as seasonal trend decomposition using loess (STL), principal component analysis (PCA), and independent component analysis (ICA), have been widely applied to identify different deformation trends or reduce residual atmospheric delays and noise. Among these methods, ICA offers a distinct advantage by effectively separating statistically independent source signals from large, mixed datasets without requiring prior assumptions. Unlike PCA, which maximizes variance rather than physical independence, ICA is particularly suitable for separating deformation processes driven by different physical mechanisms.

In this study, we applied ICA to extract spatially coherent deformation patterns that are statistically independent of MT-InSAR time-series displacements derived using the small baseline subset (SBAS) algorithm in Gimhae City, South Korea. The study area is characterized by soft ground conditions and thick sedimentary layers, which increase its susceptibility to persistent surface deformation. Ongoing urban development for residential and industrial purposes further underscores the importance of continuous surface-displacement monitoring for mitigating urban geohazards. We analyzed two descending-orbit SAR datasets: COSMO-SkyMed acquisitions from January 2013 to April 2019 and Sentinel-1 acquisitions from January 2016 to October 2021. MT-InSAR time-series displacements were generated for each dataset, and ICA was subsequently applied to the results. By combining these datasets, we reconstructed surface deformation from 2013 to 2021 and identified three independent components (ICs). These components correspond to linear, quadratic (acceleration-related), and seasonal trends, and their associated spatiotemporal IC score maps were used to identify clusters of similar deformation behavior.

To evaluate the representativeness of the ICs, correlation analyses were conducted between the ICs and time-series parameters, including velocity and acceleration. In the COSMO-SkyMed results, IC1 and IC3 show strong correlations with velocity and acceleration, with coefficients of determination (R²) of 0.76 and 0.90, respectively. In the Sentinel-1 results, IC1 and IC2 are correlated with velocity and acceleration, with R² values of 0.88 and 0.60, respectively. The remaining components, IC2 for COSMO-SkyMed and IC3 for Sentinel-1, are interpreted as representing seasonal effects with additional noise contributions. To reveal the deformation associated with each component and to identify deformation patterns in the hotspots, ICA was applied to the time-series displacements. Spatial analysis of the ICs reveals a clear uplift in the central urban area, with cumulative displacements of approximately 5.8 cm for COSMO-SkyMed and 2.3 cm for Sentinel-1. Acceleration-related ICs exhibit nonlinear temporal patterns, suggesting gradual stabilization of deformation, while seasonal effects are relatively minor. Comparison with groundwater-level observations reported in our previous study indicates that the groundwater recharge over the past 10 years is the primary driver of the observed uplift. In contrast, subsidence hotspots identified in the eastern part of the study area exhibit cumulative displacements of approximately -4.2 cm for COSMO-SkyMed and -7.4 cm for Sentinel-1. The COSMO-SkyMed IC3 time-series indicates accelerated deformation, whereas the Sentinel-1 IC2 time-series reflects deformation deceleration during the later observation period.

Overall, this study demonstrates that applying ICA to MT-InSAR-derived deformation time series effectively separates deformation signals from seasonal effects and noise, enabling robust identification of uplift and subsidence hotspots and estimation of deformation velocity and acceleration. The proposed framework provides valuable insights into urban deformation processes and offers a practical tool for assessing infrastructure vulnerability in rapidly developing urban environments.

The Permian basin is one of the most prolific and developed onshore oil & gas basins in the world, and one of the most dynamically deforming regions driven by anthropogenic activity. Sustained saltwater injection associated with unconventional hydrocarbon production has been linked to induced seismicity and to the development of two distinct, superimposed deformation patterns. At the regional scale (∼1,000 km²), subsidence and uplift signals are driven by variations in the stress regime and differences in injection practices across distinct state regulations. Superimposed on the regional background, localized (∼1 km²) blisters represent rapidly developing, near-catastrophic uplift or subsidence events triggered by pressure changes within compartmentalized shallow formations (e.g., the San Andres Formation in the Midland Basin and the Delaware Mountain Group in the Delaware Basin), where saltwater disposal occurs.

This study demonstrates: (1) how to overcome the challenges associated with near-real-time Interferometric Synthetic Aperture Radar (InSAR) monitoring in one of the most dynamic, human-impacted regions globally, the Permian Basin (∼90,900 km²); and (2) how rapidly evolving small-scale (∼1 km²) blisters can be systematically detected and quantitatively characterized.

To monitor these surface changes, a Small Baseline Subset (SBAS) workflow was developed using two ascending and one descending Sentinel-1 regional-scale stacks. One of the main challenges of near-real time InSAR monitoring at a regional scale is atmospheric signal delay, which can impose meters of apparent displacement that must be corrected to identify the relevant mm-cm scale displacement patterns. We solve this challenge by modeling the atmospheric signal delay for each new acquisition based on the expected spatiotemporal patterns derived from over ten years of Sentinel-1 acquisitions. Temporal filtering enables separation of displacement signals from atmospheric phase delays, which is particularly useful to monitor slowly-evolving displacement; however, unfiltered time series must be examined to identify rapidly evolving, kilometer-scale blisters that may otherwise be attenuated by filtering.

By clustering, the distinct temporal behaviors observed in the SBAS time series across the Permian Basin (i.e., blisters) can be isolated from the regional background signal. Recurrent comparison of the detected blisters, combined with expert interpretation, enables characterization of their temporal evolution and qualitative assessment of their behavior. Effective separation of atmospheric signal delay from true displacement, combined with the comparison between temporally filtered and unfiltered time series, is essential for reliably detecting blisters.

This workflow demonstrates that rapidly evolving kilometer-scale deformation anomalies can be resolved at near-real-time frequency across a basin-wide region. It provides a critical tool for understanding saltwater disposal behavior in the subsurface, supporting geohazard mitigation, and informing regulatory decision-making in one of the most intensively developed hydrocarbon basins worldwide.

Deformation of solar evaporation pond perimeter dikes needs to be effectively monitored and managed to prevent overflow and stability issues that can lead to failure and major damage. This case study describes how Interferometric Synthetic Aperture Radar (InSAR), complemented by terrestrial survey methods, identified a new settlement anomaly in 2023 along a perimeter dike that presented immediate operational risk and raised concerns about a horizontal displacement component driven by one-sided hydraulic loading. Close collaboration between InSAR analysts and on-site teams turned the deformation measurements into actionable results framed around operational needs.

We processed TerraSAR-X (descending) and Sentinel-1A (ascending and descending) data spanning January 2023 to November 2025 using the Small Baseline Subset (SBAS) approach. Since a stable reference area is difficult to guarantee in this environment, we first reference the results to the most stable and reliable scatterers within the area of interest and then correct the bias/rates based on ground control point time series. The TerraSAR-X and Sentinel-1A descending datasets show good agreement, strengthening confidence in the detected acceleration from July 2023 onward and in its subsequent evolution.

Differences between ascending and descending Line-of-Sight (LOS) time series during the anomaly indicate that deformation is not purely vertical. Standard decomposition of dual-orbit InSAR datasets are often used to separate displacement into vertical and east–west motion. However, the orientation of the dike structures implies that the expected horizontal displacement may occur in a different direction. We employ the strapdown decomposition approach [1] and assume that horizontal displacement is predominantly perpendicular to the dike axis (transversal) and that the longitudinal (along-dike) component is negligible. We explicitly calculate and communicate the increase in uncertainty for dike segments where the tangent approached the north–south axis.

We validate the InSAR results with existing in-situ survey monitoring. Crest points, spaced approximately 250 m apart, are monitored using static Global Navigational Satellite System (GNSS) receivers and closed-loop digital leveling. Comparisons at representative chainages confirm the July 2023 acceleration and show good agreement where the local dike orientation provides sufficient sensitivity with respect to the satellite LOS. The observations are consistent with the broader geohydrological setting, including sinkhole activity and brine discharge features observed near the shoreline. These findings support mitigation planning and prioritization, including accelerated crest raising and strengthening measures in the most critical sectors.

[1] Brouwer, W. S., & Hanssen, R. F. (2024). Estimating three-dimensional displacements with InSAR: The strapdown approach. Journal of Geodesy, 98(12), Article 110. https://doi.org/10.1007/s00190-024-01918-2

Linear infrastructure monitoring with MTInSAR benefits from the wide swath of Sentinel-1 images as very long segments of the linear infrastructures are depicted in one frame. However, it comes with computationally expensive processing when using conventional MTInSAR for large areas, because the majority of the processed pixels are not directly relevant to the linear infrastructure, rather, they are required to aid phase unwrapping and are used to remove the atmospheric phase contribution. To address this limitation, we propose a new method that estimates the spatial displacement gradient solely along the linear infrastructure, thereby direclty providing the differential displacement and being computationally light as the displacement time series is not derived. While estimating local displacement signals at linear infrastructures from Sentinel-1 images is challenging due to the medium spatial resolution of the sensor, estimating displacement signals on a regional scale is reasonable from Sentinel-1 mission. The presented method is not intended to replace MTInSAR, but rather to provide a rapid screening of displacement signals at linear infrastructures on a large-scale, thus highlighting hazardous areas for which detailed MTInSAR processing is recommended.
The proposed method takes a co-registered stack of SLC images and selects coherent pixels in single-look interferograms based on the location of the linear infrastructure which is known beforehand from GIS data, e.g. from OpenStreetMap. Coherent pixels within a buffer around the linear infrastructure are selected and grouped into grid cells of regular size. For each grid cell, the interferometric phases are averaged per interferogram. Then, the differential displacement velocity is estimated from the wrapped phases of two neighbouring grid cells by maximing the temporal coherence as it is commonly applied during temporal unwrapping. The result of the method comprises the differential displacement velocity between neighbouring grid cells which is used to assess the hazard related to the displacement. The proposed approach directly estimates the spatial displacement gradient and has the advantage that error propagation from pixels outside the linear infrastructure is avoided.
We conducted an experiment with a study area in the Rhenish Coalfields in North-Rhine Westphelia in Germany exhibiting subsidence of up to 7.5cm/year. The area includes multiple railway tracks and highways for which the displacement gradients are assessed with our proposed method. We coregistered 58 Sentinel-1 images from descending track with GAMMA for the period 04-01-2021 to 30-12-2021. To validate the results of our method, we compare the displacement results with results from 29 high spatial resolution TerraSAR-X Stripmap images from descending track for the same study area and period. Moreover, we use the state-of-the-art MTInSAR method implemented in the open-source research software SARvey to estimate the displacement velocities and time series from the two datasets for comparison. The results indicate good agreement between the displacement velocities estimated from the TerraSAR-X and Sentinel-1 datasets for both the proposed gradient approach and the conventional MTInSAR method. The RMSE between the spatial displacement gradients from the proposed approach and gradients derived from the MTInSAR results are decreasing with increasing grid size from 0.35cm/year down to 0.05cm/year for a grid size of 50m up to 500m, respectively. The results show that the chosen grid size has a low-pass filter effect which preserves the main regional displacement signal along the linear infrastructure. Retrieving displacement signals with a higher spatial frequency is more likely with a smaller grid cell size, but depends on the availability of coherent pixels and comes at the cost of introducing higher sensitivity to single decorrelated pixels. The proposed method enables a fast assessment of displacement gradients for monitoring of linear infrastructure by exploiting the potential of the Sentinel-1 mission due to its wide area coverage and regular acquisition plan.

This study explores advanced InSAR post-processing techniques aimed at enhancing structural health monitoring for dam safety, focusing on two critical challenges: the mitigation of complex residual periodic signals and the interpretation of displacements in non-ideally oriented structures.

1. Mitigation of Residual Periodicity

Standard InSAR software typically compensates for primary environmental factors, such as thermal expansion—a common practice in bridge or high-rise monitoring. However, dam monitoring often requires multi-factor compensation, including water level-induced (hydrostatic) displacements. Our analysis revealed that on an Earth-filled dam, even after accounting for thermal and hydrostatic effects, significant residual periodicity remained, particularly near the spillways. Initial visual inspection suggested a correlation with water temperature (approximate one-year cycle with minimum around March and maximum around November); however, a detailed parametric estimation of frequency, amplitude, and phase showed a near-uniform distribution of phase of the residual periods across the year. This contradicts the assumption of a single driving parameter, suggesting a more complex superposition of inseparable periodic signals.

Furthermore, the presence of seasonal vegetation on the downstream face introduces noise and temporal decorrelation. To maximize the density of Persistent Scatterer (PS) points, our processing chain prioritizes point retention even at the cost of temporal discontinuities in the time series. We present a refined post-processing approach where thermal effects are iteratively re-estimated and subtracted. Following this procedure, the phase of the residual periods exhibits a bimodal distribution, confirming the presence of two superimposed signals that must be accounted for to accurately separate reversible movements from potentially risky permanent deformations.

2. Displacement Decomposition for East-West Oriented Dams

Interpreting InSAR data for dams oriented in the East-West direction poses a significant geometrical challenge, as North-South displacements are often neglected in standard vertical and East-West decompositions. In such cases, Northward displacement typically inflates the vertical subsidence component by approximately 20%. To address this, we implemented a "lateral decomposition" targeting the vertical and dam-perpendicular directions. Despite the inherent noise and higher standard deviations (in the order of mm/y), the resulting displacement maps showed high spatial smoothness, likely due to high temporal coherence on the non-vegetated sections.

Unfortunately, this “lateral decomposition” is highly unreliable in case there are real displacements in the longitudinal, i.e. East-West direction: such possible displacements biases the lateral and vertical component by 300% and 60% of the real longitudinal displacement, respectively.

Displacement in the longitudinal direction, on the other hand, can be easily and reliably estimated.

Conclusion

By refining the compensation of periodic residuals and adapting decomposition geometries to the dam’s orientation, we provide a more reliable assessment of structural integrity, enabling a clearer distinction between cyclical environmental responses and critical geohazards.

Interferometric Synthetic Aperture Radar (InSAR) has emerged as a powerful geodetic tool for monitoring land deformation with high spatial resolution and temporal continuity, providing critical insights into deformation processes in rapidly urbanizing regions. This study presents updated spatiotemporal deformation data from October 2014 to December 2021 across the Greater Manila Area, derived from Sentinel-1 InSAR time-series analysis using UK COMET’s LiCSAR products and validated against global navigation satellite system (GNSS) measurements. The InSAR-derived deformation fields reveal spatially coherent subsiding zones in the Pampanga Delta near Manila Bay (with the highest rate observed at -95.2 mm/yr, sustained for more than seven years), western Rizal, northern Metro Manila, and portions of Cavite and Laguna. Horizontal compressive motion detected by InSAR further indicates lateral convergence toward areas of maximum subsidence, forming subsidence or settlement funnels.

Correlation with lithological and hydrogeological datasets shows that deformation is most pronounced in areas underlain by Quaternary Alluvium and clay-rich Oligocene to Pliocene-Quaternary formations, which exhibit low permeability and high compressibility. Thick clay layers contribute significantly to subsidence in these zones. In contrast, regions underlain by Pleistocene tuffaceous formations with robust aquifer systems display relative stability. Similarly, sandy unconfined aquifers exhibit reduced deformation, while active recharge zones in western Metro Manila appear to mitigate subsidence.

InSAR time-series analysis also reveals vertical velocities ranging from -36.7 mm/yr to 0.7 mm/yr, averaging -4.3 mm/yr within the Marikina Valley, a graben bounded by segments of the dominantly dextral Valley Fault System (VFS). Maximum subsidence occurs in its southernmost portion, where human-induced factors likely contributed to the larger displacement. Further south, near Biñan (Laguna) and Carmona (Cavite), InSAR deformation maps highlight elongated northeast–southwest subsidence zones that align with left-stepping en echelon fault structures. These features mark a releasing stepover segment of the VFS, where pre-existing extensional fractures and subsidiary splays accommodate differential settlement intensified by groundwater overextraction.

This work underscores the effectiveness of InSAR as a monitoring framework for urban deformation, enabling the detection of subtle, spatially variable patterns that traditional ground-based methods may overlook. The findings emphasize the need to integrate InSAR-derived geospatial data with geological and hydrological information to support evidence-based land-use planning, hazard assessment, and subsidence mitigation strategies in the Greater Manila Area and similar urban environments worldwide.

Over the past two decades, Differential SAR Interferometry (DInSAR) and Persistent Scatterer Interferometry (PSI) techniques have undergone major methodological and processing advances, supported by the continuous growth in spaceborne SAR data acquisition capabilities. The Sentinel-1 constellation, under the European Copernicus Programme, enables systematic monitoring of ground deformation at continental scale, combining short revisit times, high spatial resolution, and an open data policy. In parallel, the increasing availability of SAR missions with different acquisition geometries, revisit times, and spatial resolutions has created a multi-resolution Earth Observation landscape, requiring flexible methodologies capable of adapting to heterogeneous inputs.

A major milestone in operational ground motion monitoring is the European Ground Motion Service (EGMS), which provides harmonized and standardized displacement information across Europe, covering both natural and anthropogenic processes. Updated annually and offering millimetric precision, EGMS products include mean annual velocities, displacement time series from 2015 onwards, ascending and descending line-of-sight measurements, and derived vertical and horizontal components. While EGMS is adopted in this work as the primary data source due to its continental coverage and consistency, the proposed methodology is sensor-agnostic and can be applied to InSAR products derived from different missions and processing chains, allowing adaptation to local resolution requirements and data availability.

Despite the availability of wide-area displacement datasets, their direct operational exploitation remains challenging due to volume, density, and interpretation complexity. This work presents a methodology designed to systematically exploit displacement maps in an automated and scalable manner, with the objective of identifying buildings and urban structures potentially exposed to damage. The framework focuses on spatial gradients of displacement (differential deformation), a key indicator in built environments where structural damage is often associated with high deformation gradients rather than absolute displacement values.

Two complementary analysis scales are implemented. The first approach is based on the automatic extraction of Active Deformation Areas (ADA), enabling area-based assessment using the full deformation information within each cluster. This provides systematic coverage over all deforming zones, delivering a low-to-medium level screening of structural susceptibility. The second approach operates at single-building scale, analysing only the displacement measurements directly associated with each asset. Although this method is applicable only where measurement density is sufficient, it provides more detailed, asset-specific outputs.

Differential deformation values are used as intensity indicators to classify potential damage levels at both ADA and building scales.

The methodology is currently being advanced within the RASTOOL-DoS project as a candidate service for integration into the Copernicus Emergency Management Service (CEMS) Risk and Recovery Mapping portfolio.

Methodological details, application examples, and preliminary evaluation results will be presented.

The progressive expansion of spaceborne Synthetic Aperture Radar (SAR) missions, and in particular the operational continuity of the Sentinel-1 constellation within the Copernicus Programme, has enabled the generation of national-scale Differential SAR Interferometry (DInSAR) products. These datasets, derived through advanced multi-temporal DInSAR algorithms such as the Parallel-Small BAseline Subset (P-SBAS) approach, consist of millions of Measurement Points (MPs) characterized by displacement time series spanning several years and the retrieved mean annual velocity. While such massive datasets represent an unprecedented opportunity for large-scale ground deformation monitoring, they also present a critical methodological challenge: the necessity for scalable, automated, and statistically robust tools capable of exploiting both the temporal and spatial domains of DInSAR information.

In this work, an integrated analysis framework specifically designed for national-scale applications, the TS-DInSAR tool, is presented. The tool is developed and tested over the entire Italian territory using Sentinel-1 SAR data acquired between June 2016 and November 2023 and processed via the P-SBAS algorithm implemented by the Institute for Electromagnetic Sensing of the Environment (IREA). The processing chain generates more than 64 million MPs, geocoded onto a standard 1 arc-second SRTM grid, providing both vertical and horizontal displacement components derived from ascending and descending Line-of-Sight (LOS) measurements.

The TS-DInSAR workflow is structured into four main analytical steps. The initial step pertains to the reduction of data dimensionality. Given the original dataset size of approximately 250 GB, the MPs were filtered by applying a velocity threshold of ±1 cm/year (representing 3σ) in at least one displacement component. This threshold reduces the dataset to approximately 600,000 MPs characterized by significant kinematic behaviour, while preserving the full paired time series (horizontal and vertical) for each selected MP. The subsequent step involves the unsupervised temporal characterisation of deformation patterns. Principal Component Analysis (PCA) is applied independently to both horizontal and vertical time series datasets after standardisation, enabling dimensionality reduction while maximizing retained variance. The extracted Principal Components (PCs) represent dominant temporal behaviours within the dataset. Subsequently, a K-means clustering algorithm is implemented with the number of clusters automatically determined according to the optimal number of retained PCs. This procedure yields a categorical label for each MP in both components (EW_cluster and UP_cluster), summarizing the temporal pattern of the DInSAR time series. The third step involves the integration of geomorphological information through the incorporation of a high-resolution Digital Elevation Model (DEM), such as the terrain slope. Additionally, a kinematic ratio parameter (K_VH), defined as the absolute ratio between vertical (V_V) and horizontal (V_H) mean velocities, is introduced to quantify the relative predominance of displacement components. The fourth step of the process is the implementation of a supervised classification, that integrates the parameters defined in the preceding steps into a decision-tree framework. The model assigns each MP to one of seven deformation-triggering categories: (i) landslide, (ii) potential landslide, (iii) subsidence, (iv) uplift, (v) uplift related to volcanic/tectonic activity, (vi) soil erosion, and (vii) undefined. It is noteworthy that the classification procedure has been designed to operate without the use of pre-existing thematic inventories (e.g. national landslide inventories), thereby ensuring methodological reproducibility across different geographic contexts.

The application of TS-DInSAR at the national scale revealed that slope-related instabilities (i.e. landslide and potential landslide combined) represent the most frequent deformation class (~53% of selected MPs), followed by subsidence (~18%), soil erosion (~15%), and volcanic/tectonic uplift (~11%). A number of case studies have been selected for illustrative purposes, including the Berceto deep-seated landslide, subsidence at the airport of Fiumicino (Rome), gully erosion in the Atri Natural Reserve, and bradyseismic uplift in the Campi Flegrei caldera. These case studies demonstrate the capability of the TS-DInSAR tool to capture distinct kinematic signatures and correctly associate them with geomorphological and geodynamic processes.

In conclusion, the TS-DInSAR tool constitutes an automated, scalable, and statistically robust framework that integrates both temporal and spatial domains of DInSAR analysis within a unified methodological procedure. The pivotal enhancement introduced by the tool lies in its capability to systematically manage and interpret massive DInSAR datasets at the national scale. A key methodological strength of the proposed approach is its independence from pre-existing thematic inventories, such as landslide inventories. This inventory-independent architecture enables the implementation of the tool in regions where ancillary datasets are incomplete, outdated, or unavailable, while preserving an objective, data classification of deformation-triggering phenomena.

Large-scale urban monitoring calls for fine-grained, high-density measurements to capture the spatially heterogeneous evolution of cities and to support infrastructure risk management. In this context, dense and reliable retrieval of both 3-D structural parameters (e.g., facade/roof heights) and surface deformation time series is essential for safety assessment and sustainable city management. Owing to the side-looking SAR imaging geometry, dense built-up areas are severely affected by layover, where multiple scatterers with identical slant-range positions are superimposed within a single resolution cell. SAR tomography-based PSInSAR (Tomo-PSInSAR) alleviates this limitation by extending time-series InSAR to the elevation dimension, enabling the separation of layovered scatterers and the joint estimation of their elevation and deformation parameters.

Despite its potential, practical Tomo-PSInSAR performance for metropolitan-scale monitoring remains constrained by the finite tomographic elevation resolution imposed by the baseline aperture. In particular, two closely spaced scatterers may not be reliably separated, leading to missed detections and consequently limiting the achievable monitoring-point density in layover-dominated areas. Although super-resolution strategies have been introduced to mitigate this limitation, greedy compressive sensing solvers (e.g., OMP and CoSaMP) can still degrade when the tomographic dictionary exhibits strong atom correlation. In such cases, two nearby scatterers may be merged into a single reconstructed component or represented by atoms from the same neighborhood, weakening effective separation.

To address this issue, we propose an enhanced two-tier Tomo-PSInSAR framework for fine-grained urban monitoring. The first tier constructs a PSC-based spatial network and refines its topology using an elevation Rayleigh-resolution constraint, thereby pruning unreliable arcs and improving network reliability. The second tier integrates our Adaptive Correlation Filtered Orthogonal Matching Pursuit (ACF-OMP) for arc-wise parameter estimation, which enhances the resolvability of closely spaced double scatterers. Specifically, ACF-OMP adaptively suppresses highly correlated atoms by evaluating the similarity between the inner products of selected atoms and the residual, thereby reducing duplicated false detections and improving the separability of closely spaced scatterers. This capability substantially increases the density of reliable monitoring points, which represents the primary improvement of our approach, while also enhancing the accuracy of height and deformation retrieval.

Experiments on TerraSAR-X time series over Changsha demonstrate that, compared with a conventional Tomo-PSInSAR baseline using beamforming for arc-wise parameter estimation, the proposed method increases the number of effective monitoring points by more than 80% and reduces the deformation RMSE by approximately 22%, particularly in dense building blocks with severe layover.

Keywords: urban monitoring, Tomo-PSInSAR, layover separation, deformation time series, height estimation

Interferometric Synthetic Aperture Radar (InSAR), which measures surface displacement from repeat-pass phase differences, provides millimeter-level, wide-area deformation measurements and, with expanding satellite archives, multi-temporal InSAR has gradually evolved into a routine monitoring tool [1], [2]. In urban environments, the presence of abundant man-made stable scatterers and the need to estimate residual height errors make Persistent Scatterer InSAR (PS-InSAR) a particularly suitable framework: it jointly estimates deformation, residual height error, and atmospheric delay over pixels that remain coherent over time, effectively exploiting stable reflectors associated with built structures [3], [4]. Nevertheless, robust and reliable phase-quality estimation remains challenging. Some methods suppress phase noise and facilitate phase-quality estimation through spatial averaging, but they are not well suited to urban scenes with strong height variations [5], [6]. PSP-based approaches provide a simple arc-wise phase quality measure using double-differenced observations, yet their high computational cost often forces restrictive arc selection in practice, compromising the full exploitation of available information in exchange for efficiency [7]. To address these challenges, we propose Exhaustive Pairwise InSAR (EP-InSAR), an arc-based PS framework aimed at densely recovering pixels with reliable phase histories in complex urban areas exhibiting localized and non-linear deformation.

By replacing the conventional single-master network with a daisy-chain baseline linkage, where interferograms are formed only between adjacent acquisitions, EP-InSAR emphasizes short-interval phase evolution, tolerates more unmodeled phase variations, and strengthens the ability to capture non-linear deformation. Because deformation phase accumulates only weakly over such short intervals, the temporal deformation term in the double-differenced phase series has minimal impact on arc assessment, allowing each arc to be solved primarily for height difference and reducing the demanding two-dimensional parameter search to a one-dimensional problem. Leveraging this simplification, EP-InSAR provides a fully vectorized implementation that efficiently batch-processes millions of arcs, enabling substantially denser arc exploration.

Building on the efficient arc-evaluation capability, this study further proposes a flexible arc search framework that aims to approximate exhaustive exploitation of high-quality arcs while controlling computation. Instead of relying on a single conservative point-selection criterion, we distinguish anchor points and usable points using separate thresholds defined by the quality, number, and spatial extent of their associated arcs. Starting from an initial anchor set, we iteratively expand the network via triangulation-based k-ring expansion by linking each candidate point to nearby anchors, promoting new anchors, and thereby discovering additional usable points. This iterative expansion reduces reliance on conventional phase-quality priors and empirically approaches near-exhaustive coverage in urban windows while examining only a small fraction of all possible pixel pairs.

From the resulting set of high-quality arcs, we provide a complete deformation time-series workflow for deformation field reconstruction. Arc-wise observations of differential residual height error and thermal expansion coefficient are first integrated into global parameter fields using Huber-type IRLS to down-weight outliers, with initial weights determined by arc coherence. The integrated fields are then used to compensate interferograms, effectively reducing phase spatial frequency and easing phase unwrapping. At this stage, baseline combinations are no longer restricted to the daisy-chain configuration. Based on the compensated interferograms, we apply spatiotemporal filtering to derive the final deformation time series. Specifically, atmospheric delays are estimated by applying spatial low-pass filtering to the high-frequency temporal components, yielding localized long-term deformation after removing correlated artifacts.

A large-scale experiment over eastern Shanghai using TerraSAR-X validates the framework and demonstrates its ability to reveal localized deformation signals. The dataset comprises 54 descending scenes acquired between 2020 and 2024, covering the Yangtze River estuary and the surrounding islands. Due to Python’s inefficient inter-process data sharing in multiprocessing, the scene is divided into dozens of patches, processed independently, and then mosaicked to produce an overview deformation map. As reported in many previous studies, pronounced subsidence is observed along northeastern boundary embankments associated with large-scale land reclamation. Beyond these broad patterns, the proposed method also captures a large number of highly localized yet reliable signals, including deformation around road junctions and individual buildings. By comparison with historical satellite imagery, we interpret representative patterns consistent with typical urban geotechnical processes. For a concise reliability check, we further compare results over Pudong International Airport using different sensors and algorithms. Applying the proposed method to a descending Sentinel-1 stack yields deformation zones broadly consistent with the TerraSAR-X results, whereas StaMPS, run on a TerraSAR-X subset with a standard SBAS network, struggles to recover localized deformation.

The discussion notes that EP-InSAR differs from existing approaches in its spatial differencing pattern, baseline combination strategy, and relaxed prior assumptions. By allowing unmodeled components to retain signal content or exhibit abrupt transitions, the framework seeks to exploit more of the available phase information rather than treating it as noise. However, in the presence of phase wrapping, different baseline combinations cannot be made statistically equivalent simply by reweighting, and the conventional single-master strategy can retain practical advantages in stability. This motivates the use of baseline-dependent statistical thresholds. In addition, we evaluate the arc expansion strategy against a brute-force reference in a 500×500 pixel urban window. As the k-ring degree increases from 0 to 5, only a small fraction of arcs needs to be examined, while the recovered point set quickly approaches the brute-force upper bound. Degree 0 recovers 62% of all points with 4.53M arcs, and degree 5 reaches 83% (21,960 points) with 44.7M arcs. By contrast, a 150 m radius-dense scheme examines 1.20B arcs (about 27× more) yet yields slightly fewer points (21,908). The remaining gap is mainly confined to window corners where a fully enclosed observation network cannot form.

In summary, EP-InSAR presents an arc-based PS InSAR framework aimed at dense recovery of reliable phase histories in complex urban environments. By combining daisy-chain interferogram formation, batched one-dimensional periodogram search, near-exhaustive arc exploration, robust global integration, and spatiotemporal filtering, the method improves spatial coverage and reveals more spatially and temporally compact deformation signals with stable and interpretable time series, supporting large-scale urban deformation monitoring. Several limitations remain. The daisy-chain linkage reduces computational burden and is more tolerant to unmodeled phase variations, but it can increase arc-level phase noise and blur arc-quality thresholds. Omitting explicit estimation of linear deformation velocity may also cause points with extreme linear deformation to be missed. Finally, relaxing model constraints can introduce artifacts, motivating careful verification and more robust baseline-dependent reweighting.

This work was supported by the National Key Research and Development Program of China (Grant No. 2025YFC3215200 and 2025YFC3215200-01).

Keywords: EP-InSAR; urban deformation monitoring; time-series InSAR; scientific computing; parameter estimation

[1] D. Perissin and T. Wang, “Time-Series InSAR Applications Over Urban Areas in China,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 4, no. 1, pp. 92–100, Mar. 2011, doi: 10.1109/JSTARS.2010.2046883.

[2] P. Ma et al., “Toward Fine Surveillance: A review of multitemporal interferometric synthetic aperture radar for infrastructure health monitoring,” IEEE Geosci. Remote Sens. Mag., vol. 10, no. 1, pp. 207–230, Mar. 2022, doi: 10.1109/MGRS.2021.3098182.

[3] A. Ferretti, C. Prati, and F. Rocca, “Permanent scatterers in SAR interferometry,” IEEE Trans. Geosci. Remote Sensing, vol. 39, no. 1, pp. 8–20, Jan. 2001, doi: 10.1109/36.898661.

[4] A. Ferretti, C. Prati, and F. Rocca, “Nonlinear subsidence rate estimation using permanent scatterers in differential SAR interferometry,” IEEE Trans. Geosci. Remote Sensing, vol. 38, no. 5, pp. 2202–2212, Sep. 2000, doi: 10.1109/36.868878.

[5] A. Hooper, P. Segall, and H. Zebker, “Persistent scatterer interferometric synthetic aperture radar for crustal deformation analysis, with application to Volcán Alcedo, Galápagos,” J. Geophys. Res., vol. 112, no. B7, p. B07407, Jul. 2007, doi: 10.1029/2006JB004763.

[6] A. Hooper, D. Bekaert, K. Spaans, and M. Arıkan, “Recent advances in SAR interferometry time series analysis for measuring crustal deformation,” Tectonophysics, vol. 514–517, pp. 1–13, Jan. 2012, doi: 10.1016/j.tecto.2011.10.013.

[7] M. Costantini, S. Falco, F. Malvarosa, F. Minati, F. Trillo, and F. Vecchioli, “Persistent Scatterer Pair Interferometry: Approach and Application to COSMO-SkyMed SAR Data,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 7, no. 7, pp. 2869–2879, Jul. 2014, doi: 10.1109/JSTARS.2014.2343915.

Sinkhole hazards remain a major concern in former mining areas due to the long-term instability of abandoned underground workings. In many European regions, these former mining areas are now densely urbanized, increasing the need for reliable and spatially extensive ground stability monitoring (e.g. Strozik et al., 2015; Delnauy et al., 2025). For instance, in France, monitoring mainly relies on localized in situ instrumentation deployed in identified high-risk areas, resulting in limited spatial coverage. In this context, Interferometric Synthetic Aperture Radar (InSAR) has recently emerged as a complementary approach, providing spatially continuous measurements of millimetric ground deformation over large, urbanized mining areas (e.g. Raucoules et al., 2007; Radutu and Vlad Sandru, 2023).

In this study, in the frame of the EU funded SIRIMA project (Sinkhole hazard and risk management in post-mining areas), we performed advanced InSAR time-series analyses to monitor two French post-mining regions: Thil in the Lorraine basin and Saint-Étienne in the Massif Central. In Saint-Étienne, a sinkhole occurred in 2021 under an industrial building, whereas Thil former shallow galleries are under in-situ monitoring for sinkhole risk, following the recent underground collapses of 2018-2020. To monitor those areas with InSAR, the entire Sentinel-1 archive (2015–2025) was processed using IPTA (Interferometric Point Target Analysis), i.e., the Persistent Scatterer InSAR (PS-InSAR) techniques implemented in the GAMMA processing toolbox (Wegmüller & Werner, 1998).

Velocity maps obtained over both mining sites reveal several localized deformation zones, reaching up to 8 mm/yr, primarily associated with former mining activities. In Thil, the InSAR time series over some high-risk areas that were only recently placed under supervision show the onset of ground deformation, providing complementary information to the field instrumentation. At the 2020 Saint-Étienne sinkhole site, time-series analysis over the damaged building highlights an acceleration of subsidence during the three months preceding the collapse, with a total of ~9 mm of cumulative movement. However, the medium spatial resolution of Sentinel-1 limits the direct characterization of small features, such as the ~10 m diameter sinkhole, emphasizing both the strengths and limitations of medium-resolution SAR data, such as Sentinel-1, for operational monitoring of sinkhole hazards in urban post-mining environments.

Acknowledgment

This work is conducted in the frame of the SIRIMA project (Sinkhole hazard and risk management in post-mining areas), funded by the EU RFCS 2013, Grant agreement 101157400.

References

Delaunay, T, Lefebvre, O, Vuidart, I & Bigarré, P 2025, 'Sinkhole post-mining risks: the French methodology', in S Knutsson, AB Fourie & M Tibbett (eds), Mine Closure 2025: Proceedings of the 18th International Conference on Mine Closure, Australian Centre for Geomechanics, Perth, https://doi.org/10.36487/ACG_repo/2515_47

Radutu, A. and Vlad Sandru, M.I. (2023). Review on the Use of Satellite-Based Radar Interferometry for Monitoring Mining Subsidence in Urban Areas and Demographic Indicators Assessment. Mining Revue / Revista Minelor. 29. 42-62. 10.2478/minrv-2023-0004.

Raucoules D., Colesanti C. and Carnec C, (2007), Use of SAR interferometry for detecting and assessing ground subsidence, Compte Rendus Geosciences, vol 339, n°5, p 289

Strozik G, Jendruś R, Manowska A, Popczyk M. Mine Subsidence as a Post-Mining Effect in the Upper Silesia Coal Basin. Polish Journal of Environmental Studies. 2016;25(2):777–785. doi:10.15244/pjoes/61117.

Wegmüller, U., & Werner, C. L. (1998). SAR processing, interferometry, differential interferometry and geocoding software. InEuropean conference on Synthetic Aperture Radar, EUSAR98, Friedrichshafen, Germany, 25-27 May 1998.

As an island setting, Taiwan is highly vulnerable to coastal environmental change. The sandy coasts of western Taiwan, particularly from Taichung to Pingtung, have experienced persistent land subsidence and coastal inundation over the past five decades. Accelerated subsidence has significantly amplified flood hazards, posing complex challenges to coastal resilience. Although terrestrial geodetic measurements and InSAR analyses have been applied to selected local areas, comprehensive large-scale assessments integrating land subsidence with sea-level rise scenarios remain limited. In this study, we quantify decadal-scale coastal subsidence using multitemporal InSAR observations and evaluate associated flood risks under combined effects of land subsidence, sea-level rise, and extreme weather events. SAR datasets acquired from ERS-1/2, Envisat, and Sentinel-1 spanning three distinct time periods were processed using ISCE and MintPy to generate line-of-sight deformation time series. Constrained by horizontal GNSS velocities, we derived two vertical land motion models assuming uniform and nonlinear subsidence rates across western Taiwan. These models were further used to project coastal elevation changes over the coming decade. By integrating projected land subsidence with sea-level rise scenarios under global warming and extreme weather conditions, our results indicate that approximately 4–5% of Taiwan’s land area could be exposed to coastal inundation. Furthermore, hydrological simulations incorporating climate change–driven extreme precipitation were conducted for the Choushui River basin, the largest river system in Taiwan. The simulations suggest that the combined effects of subsidence, rising sea levels, and intensified rainfall could lead to widespread flooding and substantially elevate risks along major river corridors. These findings provide critical quantitative constraints for future coastal resilience planning and digital twin–based hazard assessment frameworks.

Multi-temporal (MT) Differential SAR Interferometry (DInSAR) techniques are widely used in Earth Observation applications, due to their capability to detect and monitor ground displacements associated with natural and anthropogenic hazard scenarios with sub-centimetric accuracy [1-5]. Among several MT-DInSAR techniques, the Parallel Small BAseline Subset (P-SBAS) approach [6-7] represents a consolidated and computationally optimized implementation of the original SBAS algorithm [8], able to retrieve spatially and, whenever possible, temporally dense deformation time series and the corresponding mean deformation velocity maps of an area of interest. In particular, the P-SBAS processing chain leverages both High-Performance Computing (HPC) architectures and multi-node/multi-thread parallel computing strategies to efficiently and automatically manage massive multi-look SAR interferometric datasets, thus enabling the extraction of medium-resolution LOS-projected displacement measurements, for regional, national- and continental-scale surface deformation analyses.

Recent advancements have led to the Full-Resolution P-SBAS (FR P-SBAS) approach [9], which further enhances the detection and mapping of a wide range of displacements by enabling the estimation of displacement time series at the native spatial resolution of the exploited SAR data. This evolution is highly supported by advanced HPC environments and Graphics Processing Unit (GPU)-based parallelization strategies, which allow us to process large DInSAR datasets in significantly short time frames. The FR P-SBAS approach allows the monitoring of a wide range of deformation phenomena: indeed, it is particularly suited for investigating localized displacements associated with extended built-up environments, as those affecting critical infrastructures and individual buildings, while maintaining the capability to effectively perform advanced DInSAR analyses across multiple spatial resolution scales (for both regional and local scale investigations). With this respect, the exploitation of SAR images characterized by high spatial resolution represents a valuable solution to maximize the MT-DInSAR mapping capabilities and retrieve accurate deformation signals, which are essential for assessing building and infrastructure vulnerabilities and supporting risk mitigation strategies. In this context, the SAR sensors onboard the Italian COSMO-SkyMed constellation of first (CSK) and second (CSG) generation of the Italian Space Agency (ASI), provide a unique source of X-band (~3 cm wavelength) SAR data, characterized by high spatial resolution (less than 3 m in Stripmap acquisition mode), relatively short revisit intervals (on the order of a few weeks), and long-term temporal continuity since 2009. Owing to these characteristics, the integrated CSK/CSG datasets are particularly well suited for FR P-SBAS-based deformation monitoring applications related to bridges, dams, transport infrastructures, pipelines, and urban buildings, where it may be crucial to identify localized deformation signals, differential displacements at the scale of single buildings or specific structural elements of extended built-up environments, thus requiring high spatial resolution and consistent temporal coverage for reliable detection.

In this study, we present a nationwide MT-DInSAR assessment analysis based on the FR P-SBAS approach, exploiting more than a decade of Stripmap SLC data collected since 2011 by the SAR sensors of the CSK and CSG constellation, as part of the MapItaly program [10]. This huge SAR dataset is suitable for investigating spatial and temporal variations of very localized displacements associated with anthropogenic hazard scenarios, as well as for assessing the structural conditions of critical infrastructure assets across the Italian built-up environment. The presented results, achieved by performing a FR P-SBAS analysis over selected Italian cities, including Roma, Napoli, and Bologna, highlight the potential of the combined FR P-SBAS and CSK/CSG framework to support infrastructure resilience and risk mitigation strategies through DInSAR-based mapping.

A key aspect of the FR P-SBAS approach is the capability to perform multi-scale DInSAR analyses [11] by means of the decomposition of displacement time series into spatially low-pass and high-pass components. This feature simplifies the identification of localized differential displacements through the discrimination between soil-driven regional deformation (low-pass component) and localized structural responses (high-pass component), which is crucial in areas characterized by dominant regional-scale deformation phenomena that can mask localized displacements affecting single buildings and/or infrastructures.

To enhance robustness and spatial completeness of our displacement retrieval assessment, we also present the FR P-SBAS results obtained by processing L-band SAR data (~10 × 5 m spatial resolution in the Stripmap mode) acquired through the twin satellites of the Argentinian SAOCOM-1 constellation, operated by CONAE. Owing to the longer wavelength of L-band data, these acquisitions are less affected by temporal decorrelation and phase unwrapping errors. These intrinsic properties enhance their mapping capability and increase the density of coherent points, providing a valuable complement to the X-band analysis [12].

References

[1] Ferretti, A., C. Prati, and F. Rocca, “Nonlinear subsidence rate estimation using permanent scatterers in differential SAR interferometry,” IEEE Trans. Geosci. Remote Sens., vol. 38, no. 5 I, 2000, doi: 10.1109/36.868878.

[2] Werner, C., U. Wegmüller, T. Strozzi, and A. Wiesmann, “Interferometric Point Target Analysis for Deformation Mapping,” in International Geoscience and Remote Sensing Symposium (IGARSS), 2003. doi: 10.1109/igarss.2003.1295516.

[3] Mora, O., J. J. Mallorquí, and A. Broquetas, “Linear and nonlinear terrain deformation maps from a reduced set of interferometric SAR images”, IEEE Trans. Geosci. Remote Sens., vol. 41, no. 10, pp. 2243–2253, 2003, doi: 10.1109/TGRS.2003.814657

[4] Lanari, R., Mora, O., Manunta, M., Mallorquí, J.J., Berardino, P., and Sansosti, E., ”A small baseline approach for investigating deformations on full resolution differential SAR interferograms”. IEEE Trans. Geosci. Remote Sens., 42, 1377-1386, 2004.

[5] Hooper A. J., “A multi-temporal InSAR method incorporating both persistent scatterer and small baseline approaches,” Geophys Res Lett, vol. 35, no. 16, 2008, doi: 10.1029/2008GL034654.

[6] F. Casu et al., “SBAS-DInSAR parallel processing for deformation time-series computation,” IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens., vol. 7, no. 8, 2014.

[7] M. Manunta et al., “The Parallel SBAS Approach for Sentinel-1 Interferometric Wide Swath Deformation Time-Series Generation: Algorithm Description and Products Quality Assessment,” IEEE Trans. Geosci. Remote Sens., vol. 57, no. 9, pp. 6259–6281, 2019.

[8] Berardino, P., Fornaro, G., Lanari, R., and Sansosti, E., “A new Algorithm for Surface Deformation Monitoring based on Small Baseline Differential SAR Interferograms”. IEEE Trans. Geosci. Remote Sens, 40, pp.2375-2383, 2002.

[9] M. Bonano et al., "New Advances of the P-SBAS Approach for an Efficient Parallel Processing of Large Volumes of Full-Resolution Multitemporal DInSAR Interferograms," in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 18, pp. 2317-2341, 2025, doi: 10.1109/JSTARS.2024.3507542.

[10] ASI, Italian Space Agency Upgrades Access To MAPITALY Data, Accessed: Dec. 1, 2023. [Online]. Available: https://www.asi.it/en/2023/12/asi-italian-space-agency-upgrades-access-to-mapitaly-data

[11] Manunta M., M. Marsella, G. Zeni, M. Sciotti, S. Atzori, and R. Lanari, “Two‐scale surface deformation analysis using the SBAS‐DInSAR technique: a case study of the city of Rome, Italy,” Int. J. Remote Sens., vol. 29, no. 6, pp. 1665–1684, Mar. 2008.

[12] De Luca, C, et al. "SAOCOM-1 L-band DInSAR Time Series generation through the P-SBAS approach: algorithm extension and products analysis." IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 18 (2025): 2680-2703.

Satellite-based Synthetic Aperture Radar Interferometry (InSAR) deformation monitoring has reached a high level of scientific maturity; however, its systematic integration into disaster risk reduction (DRR) and emergency management frameworks remains limited. This contribution demonstrates how operational InSAR products can be embedded within national and local disaster management processes to support evidence-based decision making. Long-term SAR observations acquired from Sentinel-1 and complementary high-resolution missions were processed to generate deformation maps and time-series products for earthquake, landslide and subsidence hazards across selected regions of Türkiye, including the Konya Closed Basin, East Marmara (İzmit Bay) region and western Anatolian landslide-prone zones. In these areas, vertical deformation rates ranging between –5 mm/yr and –35 mm/yr were detected, with localized subsidence exceeding 40 mm cumulative displacement over multi-year observation periods (2014–2023). Co-seismic and post-seismic deformation patterns associated with recent moderate-to-strong earthquakes were resolved with sub-centimetric accuracy. The derived deformation products were directly integrated into Local Disaster Risk Reduction Plans (İRAP) and the Turkey Disaster Response Plan (TAMP) workflows, enabling spatial prioritization of high-risk zones, identification of vulnerable infrastructure corridors and support for preparedness planning. Institutional challenges such as data latency, product interpretability and inter-agency coordination are discussed, and practical solutions including standardized deformation indicators, GIS-based dissemination tools and targeted capacity-building mechanisms are proposed. The results demonstrate that InSAR-based deformation monitoring significantly enhances situational awareness, supports preventive planning and strengthens preparedness strategies when aligned with existing disaster governance structures. This contribution highlights the critical role of Earth Observation in bridging the science–policy gap and strengthening disaster resilience at regional and national scales.

Regions of underground mining activities impose significant challenges for satellite-based terrain monitoring. Source modelling of mining-induced is particularly complex due to the co-appearance of multiple triggering mechanisms acting simultaneously. On one hand, shallow mining activities generate small-range displacement patches forming subsidence bowls with highly dynamic spatial and temporal evolution. Their appearance is strongly related with the depth and thickness of the exploited seams, the extraction sequence, and the applied mining technology. In addition, regulated underground explosions with higher power (often triggering tremors of up to M4) can influence the displacement pattern. Accurate delineation of the deformation source geometry has to properly consider the atmospheric phase screen affecting to the displacement signal. Moreover, temporal decorrelation typical of rural areas can influence the quality of the unwrapping process. Besides that, soil moisture variability should be properly assessed for avoiding a bias in the interpretation of the surface displacement.

In this study, an eight-year InSAR monitoring analysis over an intensively exploited coal mining area in Southern Poland, covering the Upper Silesian Coal Basin (USCB) is presented. The Enhanced Small BAseline Subset (E-SBAS) method, implemented in SARscape software, is applied to multi-temporal dataset of Sentinel-1 imagery. The observed surface displacements rates range between 0.5 and 1.5 m/year and exhibit non-linear spatial and temporal behaviour. The E-SBAS approach integrates Permanent Scatterers (PS) and Distributed Scatterers (DS) pixels, increasing spatial density and improving robustness in areas affected by decorrelation.

The displacement time series are classified into regions of behaviour based on temporal evolution patterns, supporting the delineation of deformation zones and associated risk levels. Intensity time-series analysis is incorporated to assess correlations with soil moisture variations and crop condition changes. This multi-parameter evaluation supports discrimination between deformation-driven and moisture-driven signal components.

Multi-temporal forward modelling is applied to delineate the multi-source deformation events. Monthly stacked displacement fields are modelled using the Okada elastic half-space. Model parameter optimisation is conducted through iterative fitting between modelled and observed displacement fields, allowing the assessment of model residuals and uncertainty ranges.

The proposed integrated strategy aims to complement conventional modelling and prediction procedures in mining operations, which are typically based on in situ levelling measurements, typically acquired at monthly to six-month intervals. By providing higher temporal resolution and spatial coverage, the methodology improves the understanding of complex mining-induced deformation processes and supports risk management in intensively exploited coal mining regions.

Soil moisture refers to the quantity of water present within the unsaturated zone of the soil. It is a key indicator of many Earth’s surface processes, governing plant water uptake, crop yields, groundwater recharge, and the exchange of heat and carbon between the land and atmosphere. Accurate and timely knowledge of its spatial and temporal distribution is therefore indispensable for many applications such as precision agriculture, drought early‑warning systems, and flood forecasting.

Remote sensing sensors represent a unique possibility for measuring soil moisture in a frequent and non-invasive manner from large up to global scale. Well-known microwave radiometer missions such as NASA’s Soil Moisture Active Passive (SMAP) and ESA’s Soil Moisture and Ocean Salinity (SMOS) have provided daily global acquisitions at rough resolutions from some kilometers. Current Synthetic Aperture Radar (SAR) systems such as the Sentinel-1 constellation, which is capable of acquiring data at a resolution of 10 m, have attracted the attention of the scientific community to improve the quality and the resolution of soil moisture products.

In our study, we investigate a novel deep learning-based (DL) solution for accurate and time-tagged soil moisture retrieval by combining, for the first time, backscatter and repeat-pass interferometric information derived from Sentinel-1 multi-temporal data. To overcome the challenge posed by the scarcity of high-quality reference data required for fully-supervised training, we propose a two-step method: a weakly-supervised pre-training of the model on a larger amount of data with lower accuracy, followed by a fully-supervised fine-tuning, from on-ground high-reliable measurements of soil moisture from 0 to 5 cm depth.

The selected initial DL model architecture is based on a state-of-the-art fully convolutional neural network (CNN), chosen for its proven ability to extract multi-scale spatial features from radar data for regression tasks (Carcereri et al. 2023). It follows a modular structure consisting of an Input Block, five Hidden Blocks, and an Output Block. The architecture comprises approximately 760,000 trainable hyperparameters, ensuring sufficient model capacity to learn complex nonlinear relationships between radar observables and soil moisture dynamics while maintaining computational efficiency.

The input feature set includes SAR backscatter intensity, InSAR coherence at different temporal baselines, and interferometric phase triplets, i.e. residual phase closure components computed from three consecutively acquired interferograms (De Zan et al. 2014).. All these input features provide together sensitivity to both surface scattering properties and structural changes related to soil moisture dynamics. By processing 5 Sentinel-1 acquisitions, 5 input SAR channels are obtained and 10 InSAR channels for each InSAR feature are derived. Overall, considering interferometric coherence and phase triplets, up to 20 input InSAR channels are used in the deep learning investigations. The target spatial resolution for the first implementation is set at 1 km, ensuring consistency with existing products. However, the framework is designed with scalability in mind, with the potential to refine the spatial resolution down to 100–200 m, leveraging the full information content of Sentinel-1 observations and improved calibration from high-resolution reference datasets.

The experimental setup is based on five distinct Regions of Interest (ROIs) selected to represent diverse environmental and climatic conditions across the study area in central Europe (Belgium, Netherlands, Central Germany, East Germany, and Denmark). For each ROI, a three-year time span (2018–2020) of Sentinel-1 acquisitions is considered, providing a consistent temporal coverage for soil moisture retrieval and model evaluation. Each ROI comprises approximately 180 time-series samples, capturing seasonal variations and a wide range of soil and vegetation states. This multi-regional, multi-temporal configuration ensures that the model is trained and validated under heterogeneous conditions, thereby enhancing its robustness and generalization capability across different land cover types and climatic regimes. A geographic subdivision of the ROIs is performed for splitting the dataset into training, validation, and testing regions. This avoids data leakage and ensures independency among the different subsets. To avoid the estimation of soil moisture over forested areas, the ESA WorldCover 2020 is used to generate a land-cover land-use (LCLU) mask. Tree cover, built-up areas (mainly cities), water bodies, and permanent snow, are filtered out. Moreover, the generation of a LCLU mask allows us to investigate the soil moisture retrieval over different land cover types.

For the first investigations on weakly-supervised learning of the proposed DL model, we define a common setup, which consists on dividing the defined ROIs in training (Denmark, Central Germany, and East Germany), validation (Netherlands) and testing (Belgium). The separation of the ROIs attends to have a similar distribution of the used data set in all cases. As reference map we use the Copernicus Global Land Operations - Surface Soil Moisture (CGLS-SSM) product, which provides daily observations at 1 km spatial resolution of soil moisture saturation level in percentage. This dataset is generated from Sentinel-1 backscatter using a radiative transfer model developed by TU Wien (Bauer-Marschallinger et al. 2019). As input channels for the DL model, we limited the set of Sentinel-1 input features to: SAR backscatter, InSAR phase triplets, and local incidence angle (LIA), to account for the acquisition’s geometry.

As first results on the testing dataset, we have obtained that the combination of Backscatter and LIA achieves a Mean Error (ME) of 6.63%, a Mean Absolute Error (MAE) of 10.01%, a Root Mean Squared Error (RMSE) of 13.20% and a correlation coefficient (R²) of 0.83. These results improve to ME=2.63%, MAE=8.71%, RMSE=11.37%, and R²=0.87, when adding the phase triplets. On the other hand, the combination of phase triplets and LIA alone achieves poor results, with a MAE and a RMSE around 20% and a R²=0.4%. Looking at the influence of the land cover type, we obtain in general better results for cropland than when considering grassland/schrubland/bareland. As example, when considering all three input channels: ME=3.19%, MAE=7.94%, RMSE=10.36%, and R²=0.90, are obtained for cropland, while for the other considered land cover types, ME=1.20%, MAE=10.72%, RMSE=13.64% and R²=0.79. First analysis on the temporal generalization capability of the model shows a good agreement with the reference data. For the months between January and May, there is a quite uniform distribution of the reference data with values between 0% and 95%, and the results of the proposed DL model on the testing dataset achieves R² between 0.8 and 0.9, depending on the considered land cover class. During summer (June – August), the reference soil moisture data values are concentrated between 30% and 50%. The obtained R² during these months is 0.5 – 0.6. The effects of crops, with significant variations in vegetation growth during such a period, affects the generalization capacity of the DL model investigated.

These preliminary results show in general a promising performance for the pretext, weakly-supervised task. Only few combinations of input channels have been investigated up to now, but they already reveal the potential of the Sentinel-1 short time series and DL for soil moisture retrieval. It appears that the combination of backscatter and LIA shows the most informative content, which is expected, since the reference map has been generated by considering Sentinel-1 backscatter data. When using the InSAR feature alone (Phase triplets and LIA), the results are poor. This might be caused by either an inconsistency between the reference data and the information provided by the phase triplets or by the high level of noise that might affect the phase triplets in presence of complex terrain characterizing the selected ROIs in central Europe. Moreover, the phase triplets show anyway to help the DL model to better understand some cases, since the results for the combination of all three Sentinel-1 input features obtain the best results.

(1) D. Carcereri, P. Rizzoli, D. Ienco and L. Bruzzone, "A Deep Learning Framework for the Estimation of Forest Height from Bistatic TanDEM-X Data," in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 16, pp. 8334-8352, 2023.

(2) F. De Zan, A. Parizzi, P. Prats-Iraola and P. López-Dekker, "A SAR Interferometric Model for Soil Moisture," in IEEE Transactions on Geoscience and Remote Sensing, vol. 52, no. 1, pp. 418-425, Jan. 2014.

(3) B. Bauer-Marschallinger et al., "Toward Global Soil Moisture Monitoring with Sentinel-1: Harnessing Assets and Overcoming Obstacles," in IEEE Transactions on Geoscience and Remote Sensing, vol. 57, no. 1, pp. 520-539, Jan. 2019.

The closure phase, constructed by a circular summation of three interferometric phases, each obtained from multilooking a SAR interferogram, consists of a geophysical component and phase noise, often exhibits non-zero values. These non-conservative closure phases challenge the validity of the implicit phase consistency assumption in SAR interferometry. This assumption relies on a geometric interpretation of the interferometric phase, where the expected values of the three interferometric phases are redundant, given that their sum, the closure phase, is equal to zero. While non-zero closure phases have been studied at local scales, a comprehensive and systematic statistical characterization of closure phases across broader spatial and temporal scales remains lacking. This study aims to fill this gap by systematically analyzing closure phases and exploring their statistical properties, with emphasis on assessing the extent to which the observed closure phases can be explained by current geophysical understanding.

Beyond phase noise, the existing literature identifies three contributors to geophysical closure phases: (i) changes in the dielectric properties of the medium, including soil moisture variations, wet biomass accumulation, and snow metamorphism; (ii) volume scattering in combination with perpendicular baselines; and (iii) differential skewed motions [1, 2, 3]. Analytical models have been developed for two mechanisms: one involves the variation in propagation properties of dielectric media due to soil moisture changes, and the other relates to volume scattering in the presence of perpendicular baselines [1, 4].

Large systematic closure phases of geophysical origin often appear in low-coherence scenarios with high phase noise levels. Estimating closure phases under these conditions requires maintaining a certain level of coherence or applying extensive spatial averaging (multilooking) to reduce noise. To minimize coherence loss due to temporal decorrelation, we use continuous 6-day revisit Sentinel-1 acquisitions to construct closure phases. For multilooking, we employ an adaptive strategy incorporating a statistical test on amplitudes to select suitable samples for averaging, ensuring they are realizations of the same random distribution. In addition, since we are interested in covering a wide area, we studied closure phases with a kilometer resolution across the extensive Iberian Peninsula, which encompasses various land cover types [5] and spans multiple climate zones [6].

Using the multilooked products, we analyze the statistics of closure phases for sub-regions categorized by various land cover and climate conditions. We start with estimating the standard deviation of the closure phases using a null hypothesis that they arise from phase noise, and then assess the statistical significance of the geophysical closure phases. Next, we evaluate the mean values and percentiles of closure phases and discuss their spatiotemporal patterns. Subsequently, we apply existing interferometric models for soil moisture variation and volume scattering mechanisms [1, 4, 7] to quantify the contribution of each component to the closure phases. In this study, the coefficient of determination ($R^{2}$) is used to estimate the proportion of explainable variance attributable to the known mechanisms. For cases with high unexplained variance, we visualize and discuss time-series examples to investigate unexplained signals that are not yet captured by current models.

Our results confirm the widespread presence of geophysical closure phases, characterize closure phase signatures in areas with varying geophysical conditions, highlight distinct attributes that reveal the mechanisms underlying the geophysical components, suggest new opportunities for using closure phases to detect Earth surface variations, and emphasize the importance of understanding the origin of non-zero closure phases in accurate deformation estimation.

References
[1] Francesco De Zan, Mariantonietta Zonno, and Paco Lopez-Dekker. Phase inconsistencies and multiple scattering in sar interferometry. IEEE Transactions on Geoscience and Remote Sensing, 53(12):6608–6616, 2015.
[2] Simon Zwieback, Xingyu Liu, Sofia Antonova, Birgit Heim, Annett Bartsch, Julia Boike, and Irena Hajnsek. A statistical test of phase closure to detect influences on DInSAR deformation estimates besides displacements and decorrelation noise: Two case studies in high-latitude regions. IEEE Transactions on Geoscience and Remote Sensing, 54(9):5588–5601, 2016.
[3] Francesco De Zan, Paloma Saporta, and Giorgio Gomba. Spatiotemporal analysis of C-band interferometric phase anomalies over Sicily. In EUSAR 2022; 14th European Conference on Synthetic Aperture Radar, pages 1–4. VDE, 2022.
[4] Francesco De Zan, Alessandro Parizzi, Pau Prats-Iraola, and Paco López-Dekker. A sar interferometric model for soil moisture. IEEE Transactions on Geoscience and Remote Sensing, 52(1):418–425, 2013.
[5] Marcel Buchhorn, Bruno Smets, Luc Bertels, Bert De Roo, Myroslava Lesiv, Nandin-Erdene Tsendbazar, Linlin Li, and AJ Tarko. Copernicus global land service: Land cover 100m: Version 3 globe 2015-2019: Product user manual. 2020.

[6] Murray C Peel, Brian L Finlayson, and Thomas A McMahon. Updated world map of the köppen-geiger climate classification. Hydrology and earth system sciences, 11(5):1633–1644, 2007.

[7] Francesco De Zan and Giorgio Gomba. Vegetation and soil moisture inversion from sar closure phases: First experiments and results. Remote sensing of environment, 217:562–572, 2018.

We present a masking methodology based on interferogram phase that exploits non-zero phase closure residuals as a proxy for decorrelation noise.

InSAR time series products are limited by decorrelation-driven phase noise, which propagates through filtering and phase unwrapping and is typically mitigated by masking “low quality” pixels. The most common quality metric, spatial coherence, is estimated over a window and may therefore mix spatially heterogeneous scattering behaviours: isolated high-quality pixels can be rejected if surrounded by decorrelated neighbours, while low-quality pixels can be retained inside generally coherent areas. This motivates a need for alternative quality indicators that operate closer to pixel scale.

For a closed loop of three acquisitions A, B, C, with interferograms AB, BC and CA, the wrapped phase closure should cancel to zero. In practice, non-linear interferogram operations such as multilooking and spatial filtering, which are applied independently to each interferogram, break strict phase consistency and generate a non-zero closure residual. These residuals are commonly attributed to volume scattering effects often associated with spatial and temporal variations in vegetation and soil moisture.

We compute the per-pixel wrapped closure residual, and use its absolute value as an indicator of pixel quality. Intuitively, pixels dominated by a single strong scatterer (PS-like behaviour) or by a homogeneous distributed field (DS-like behaviour) tend to yield small residuals after complex averaging. Heterogeneous or rapidly changing scattering within the multilook window or filter produces larger residuals.

Individual closure residual maps are noisy due to stochastic variations, but in general agree well with the level of decorrelation noise within the constituent interferograms. We construct an interferogram network and extract linearly independent closures to calculate an average closure residual map calculated over a temporal window. This yields a stable, high-resolution pixel quality layer derived purely from phase information, without requiring amplitude statistics, deformation smoothness assumptions, or a particular time-series model.

We demonstrate the method on a large Sentinel‑1 time series across an active mine, where spatial and temporal variations in decorrelation behaviour pose a challenge for conventional coherence-based masking. A phase closure residual can be combined with per-interferogram coherence constraints to handle short-term temporal variability, and long-timespan pairs where decorrelation is systematically higher.

The resulting average residual map preserves fine spatial texture while clearly separating stable from decorrelated areas. Compared to coherence-only masking, closure-based selection increases the number of unwrapped pixels at similar uncertainty levels (assessed via the standard deviation of linear fit to displacement time series).

Beyond masking, we discuss how the spatial distribution and temporal evolution of closure residuals can further inform filtering operations and change detection, offering a lightweight diagnostic tool for adaptive processing in large-scale InSAR pipelines.

Landslides can occur on any hillslope, making them a far-reaching and impactful hazard. Various phenomena, such as heavy precipitation, debuttressing, and earthquake loading, can lead to hillslope acceleration and possible failure. In the last few decades, interferometric synthetic aperture radar (InSAR) satellite data have been used to monitor slow-moving landslides and conduct research into the physical processes that govern the triggering conditions of slope failure. The ubiquitous nature and widespread potential hazard of landslides motivate monitoring hillslope movement on a regional or national scale, and the near-global availability of SAR imagery makes it an ideal monitoring tool. There are several challenges to address with InSAR-based landslide monitoring, however. Snow cover, rapid ground displacement, vegetation, soil moisture, and steep topography can all impact the interpretability of an interferogram. Fast-moving landslides can become incoherent. Interferograms with short temporal baselines maintain higher spatial coherence, but time series derived from such a network can have phase bias, leading to inaccurate displacement measurements. Some of these issues can be addressed using persistent scatterer techniques, but persistent scatterers are often rare on natural hillslopes. Here we investigate optimal InSAR network design to capture displacement measurements across an entire landslide for diverse environments and compare processed InSAR data to in-situ displacement observations.

We analyze C- and L-band InSAR from Sentinel-1, ALOS-2 and UAVSAR, and in-situ data from GNSS (Global Navigation Satellite Systems) stations, extensometers, and ground-based SAR in a variety of landslide environments in the USA, including the Hooskanaden earthflow in southwestern Oregon, the Slumgullion landslide in Colorado, and the Barry Arm landslide in the Prince William Sound region of southcentral Alaska. These landslides were chosen for the various problems they pose for InSAR processing, including vegetation cover, rapid landslide movement, and significant snow cover, in addition to the availability of in-situ data. We process multi-year networks of interferograms using the InSAR Scientific Computing Environment (ISCE) and vary their spatial resolution (100 m, 30 m, and 20 m) and network connectivity (connect-1, connect-3, connect-5, all connections, where more connections increase redundancy). We use MintPy to process the time series and velocities and compare them to available in-situ data. We incorporate data-driven network modification using multiple thresholds of spatial coherence to evaluate if there is an improvement in displacement measurement accuracy. Increasing network connectivity without data-driven network modification resulted in poor results in all cases. When landslide movement remained coherent in interferograms with longer temporal baselines, higher network connectivity outperformed the connect-1 scenarios. In some cases, the lower spatial resolution (100 m) results more accurately captured landslide movement compared to higher resolution (20 m), likely due to better noise suppression in highly vegetated areas. Smaller landslides, however, could be missed at 100 m spatial resolution. Finally, we compare the new NASA product DISP-S1 from the OPERA team to the in-situ data and evaluate how successful it is. DISP-S1 combines persistent scatterer and distributed scatterer techniques to map land surface displacement using Sentinel-1 data over large regions of North America, including our study locations. Preliminary results indicate that the DISP-S1 product performs well in regions with high spatial coherence, but more work is needed to understand its accuracy in noisier regions. Our results in these differing environments will inform future work towards landslide monitoring on a regional scale.

Glacial mélange, which is commonly observed in the pro-glacial region of marine-terminating glaciers, is a complex granular material characterised as a mixture of sea ice, distintegrated calved ice, and large tabular icebergs. This material has been observed to interact with the terminus region of glaciers by exerting a backstress or “buttressing” varying in strength with the mélange thickness, which affects calving behaviour and flow velocities at the ice front. Intensity offset tracking produces an estimate of the magnitude of ice speed through a normalised cross correlation peak-fitting of Synthetic Aperture Radar (SAR) Single Look Complex (SLC) image pairs acquired on 6- and 12- day repeat passes of Sentinel-1. The application of offset tracking to glacial mélange is informative because it facilitates the interpretation of transient ice dynamics in the fjord over short (sub-weekly to weekly) timescales, and at critical transitional stages such as mélange formation and break-up.

The Northeast Greenland Ice Stream presents a compelling case study for the investigation of the impact of mélange on ice frontal dynamics, as coupling between terminus and mélange speeds can be evaluated at high spatial and temporal resolution over two neighbouring fjords. Time series at Zachariae Isstrøm and Nioghalvfjerdsfjorden (79N) Glacier over the Sentinel-1 observational record show the temporal evolution of ice speed across both glaciers and their associated mélange, and speed trend maps indicate areas of localised acceleration and deceleration; these datasets are used here to evaluate the interconnectedness of extreme mélange conditions and glacier dynamics in this region within a decadal timeframe (2015-2025). Through a classification of ice velocity data obtained with the intensity offset tracking method, a proxy record for seasonal mélange cover is additionally generated within both fjords. The existing velocity processing chain for land ice is adapted to improve the accuracy of the tracking retrieval over mélange, and these results are compared with the terminus region of Zachariae Isstrøm for a period of a) exceptionally slow speeds at the grounding line during summer 2018 and b) a large observed deceleration at the grounding line at the end of the annual melt season in 2023. Paired with a novel sea ice thickness algorithm optimised for mélange, this methodology examines observational evidence for mélange-glacier connectivity both through time and concurrent with sub-seasonal ice speed extremes.

Azimuth phase perturbations caused by high-latitude ionospheric variability, particularly scintillations, introduce along-track image shifts in spaceborne SAR data. These shifts directly propagate into ice velocity products, generated measuring mutual shifts between images, producing errors of several meters and significantly degrading measurements accuracy. This effect is especially critical for precise glacier flow monitoring in Greenland.

C-band data, such as those acquired by Sentinel-1, are moderately affected, and yearly averaging of all available shift estimates reduces the overall effect in the annual velocity measurements. However, short-term velocity changes remain difficult to monitor due to the scintillation-induced variability. Moreover, radars operating at lower frequencies, such as L- or P-band, are more sensitive to ionospheric disturbances, with the effect scaling by approximately one to two orders of magnitude, making precise displacement measurements not possible.

High-latitude ionospheric variability exhibits a spatial correlation structure that can be described using anisotropic Matérn covariance functions or equivalently by power-law spectral density models. Climatological models, such as WBMOD, provide parameters for these correlation descriptions, including the integrated turbulence strength CkL. These parameters can be used to statistically evaluate expected azimuth shift errors or to generate realistic simulations of ionospheric phase screens.

In addition to modeling approaches, SAR observations themselves provide an empirical source of information on ionospheric variability. We present a multi-year statistical analysis (2019–2025) of azimuth shift variability derived from hundreds of ice velocity products over Greenland generated from Sentinel-1 data. For each scene, azimuth-direction profiles were extracted and the yearly averaged glacier flow signal was removed. The residual along-track displacements were interpreted as ionospheric azimuth shift errors. The standard deviation of these residuals provides a measure of the scintillation variability impacting velocity retrieval.

The resulting time series reveals a clear solar cycle dependence as well as a pronounced seasonal modulation, with peak scintillation-induced errors during winter months under enhanced auroral activity. These observations provide quantitative constraints on expected azimuth accuracy degradation at C-band and allow extrapolation toward L-band systems, including ROSE-L and NISAR. In addition to benefiting from the presented statistics, upcoming L-band missions may themselves serve as valuable data sources for extending such analyses.

The derived empirical error distributions support performance prediction for SAR-based ice velocity retrieval, validation of ionospheric phase screen simulations driven by climatological parameters (e.g., CkL), and assessment of mitigation algorithm robustness under realistic high-latitude disturbance conditions. This work provides an observation-based benchmark for quantifying ionospheric impacts on azimuth geolocation and ice velocity accuracy in present and future SAR missions.

Interferometric synthetic aperture radar (InSAR)–derived ice velocities are commonly interpreted
as measurements of surface glacier flow. However, radar wavelength fundamentally controls the
depth over which ice motion is sampled through penetration into snow and firn. In this study,
I investigate wavelength-dependent velocity sensing at Pine Island Glacier, West Antarctica,
by comparing Sentinel-1 C-band ice velocity products with multi-sensor MEaSUREs InSAR velocitiesthat incorporate longer-wavelength observations, including L-band SAR. After spatial
harmonization and physically motivated masking, we analyze velocity differences and longitudinal
profiles along the glacier trunk. Systematic velocity offsets are observed near the grounding
zone and along shear margins, where C-band velocities exceed longer-wavelength estimates by
several hundred meters per year. These differences persist across seasons and between independentSentinel-1 monthly products, indicating a robust physical signal rather than seasonal
or processing artefacts. I interpret the observed offsets as a consequence of depth-weighted velocity sampling, where longer radar wavelengths integrate motion over a greater fraction of the
deforming ice column. Extrapolating this framework suggests that future low-frequency SAR
missions operating at P-band may provide enhanced sensitivity to internal glacier deformation.
My results highlight the importance of penetration-aware interpretation of InSAR velocities for
ice-dynamical studies and future mission design.

Tenerife is the largest island of the Canary archipelago. Its complex volcanic landscape is the result of overlapping shield volcanoes, Las Cañadas caldera growth and collapse cycles, Teide-Pico Viejo evolved central stratovolcanoes and persistent rift activity. While eruptions are historically scarce, the island remains volcanically active, a fact proven by the historical records of the 1704-1705 Fasnia-Arafo-Siete Fuentes, 1706 Garachico, 1798 Chahorra and 1909 Chinyero eruptions and the seismic-volcanic crisis during 2004-2005. Following over ten years of relative stability, the volcanic monitoring system of the Instituto Geográfico Nacional (IGN, Spain) identified the onset of persistent seismic and geochemical unrest in 2016. This was followed by the detection of sustained, gradual ground deformation in 2023, a trend that remains ongoing.

Synthetic aperture radar interferometry (InSAR) is considered one of the most relevant techniques for surface deformation monitoring. Implemented as part of the IGN volcanic monitoring operations, this method uses C-band Sentinel-1 and X-band PAZ radar data. The workflow follows a two-fold approach: producing displacement maps via traditional two-pass interferometry and generating long-term time series and velocities.

In this study, we present time series and velocity maps of Tenerife, derived from Sentinel-1 and PAZ sensors from 2017 to the present. These datasets were generated using Gamma software through a SBAS approach suited to Tenerife’s specific characteristics (oceanic island, prominent topography and a strong atmospheric component). The results from both datasets exhibit similar surface deformation patterns starting in 2023. An east-west trend is observed on both sides of Teide stratovolcano, located in the center of Tenerife island. This pattern is compatible with the radial extension detected by the permanent cGNSS station network of the IGN volcanic monitoring system, with velocities of a few milimetres per year. On the other hand, the vertical component exhibits a more challenging behavior, as no clear uplift is observed with either cGNSS or InSAR, except for a small area located on the northern flank of Teide edifice near the Icod valley, which also shows low velocities in the range of several milimetres per year.

In this work, the obtained results will be modeled and analyzed, with the aim of identifying a source-model able to explain the origin of the deformations, ensuring compatibility with the rest of seismic and geochemical observables recorded by the IGN volcanic monitoring system.

The rapid increase in the availability of Synthetic Aperture Radar (SAR) data has enabled the generation of large-scale deformation maps at unprecedented spatial and temporal resolutions. Services such as the European Ground Motion Service (EGMS) and numerous national initiatives now provide extensive time-series datasets covering continental and regional scales. However, despite this abundance of data, there remains a significant gap in the automated analysis and interpretation of deformation patterns. Existing approaches are often limited by closed-set assumptions, where all observations are forced into predefined classes, reducing their ability to capture the inherent complexity and variability of real-world geophysical processes.
This paper proposes a novel framework for the classification and analysis of SAR-derived deformation time series, combining deep learning–based embeddings with open-set recognition. The approach leverages a time series foundation model to extract discriminative features and integrates supervised classification with a probabilistic open-set detection mechanism.
The methodology follows a structured workflow. First, real-world deformation data are analyzed to define the target classes of interest. Second, a synthetic dataset is generated to simulate representative deformation behaviors, including linear, piece-wise linear, step-like, thermal (harmonic), and non-linear profiles. The non-linear class is treated as an unknown category to support open-set calibration. Controlled noise is introduced into the simulated data to mimic real-world measurement uncertainties and to increase classification difficulty. Third, the deep learning model is fine-tuned as a closed-set classifier on the simulated dataset. Fourth, the open-set recognition framework is calibrated using a combination of distance-based and probabilistic features. Finally, the trained model is applied to real-world deformation data from the Praid area in Romania, a region affected by salt mining activities and flooding-induced ground instability.
Experimental results on the simulated dataset demonstrate strong overall performance, achieving an accuracy of approximately 0.81 and balanced F1-scores across most classes. High classification performance is observed for well-defined deformation patterns, while more complex or overlapping classes show slightly reduced accuracy. The detection of unknown patterns remains more challenging, reflecting the intrinsic difficulty of open-set recognition. Initial empirical analysis on real-world data indicates that the model is consistent with the trends observed in the simulated experiments.