Conference Agenda

Constructing a fault slip (deficit) model is helpful for understanding earthquake mechanism. The Nankai subduction zone, where the Philippine Sea plate subducts beneath the Eurasian (or Amur) plate in southwest Japan, has the potential to cause earthquakes greater than M8. Many geodetic researches have estimated slip deficit distribution on the Nankai subduction zone (e.g. Ochi 2015; Yokota et al. 2016; Nishimura et al. 2018). Even though Japan has dense GNSS observation network (GEONET), its spatial spacing is limited to about 20 km. InSAR can observe surface displacements with high spatial resolution, and some researches revealed InSAR improved the resolution of slip deficit distribution even at deeper area of subduction zones (Maubant et al. 2023; Kinoshita and Furuta 2024). In this study, we focused on the impact of InSAR for slip deficit estimation on the Nankai subduction zone.

We conducted GNSS and InSAR time-series analyses to detect interseismic displacements. In GNSS analysis, we first corrected postseismic displacements of the 2011 Tohoku-oki earthquake based on Tobita (2016), then we fitted a function composed of linear, trigonometric, and offset terms to GEONET and GPS-A coordinates (Takamatsu et al. 2023; Yokota et al. 2016) from September 2014 to March 2025. In InSAR processing, we first made interferograms for all possible pairs using ALOS-2 Stripmap mode SAR data in 7 frames within Shikoku island. All 7 frames were observed from descending tracks. We corrected ionospheric and neutral atmospheric noises (Wegmüller et al. 2018; Kinoshita 2022), solid earth tide effect (Yunjun et al. 2022), and postseismic effect of the 2011 Tohoku-oki earthquake (Tobita 2016) from each interferogram. We then obtained LOS velocity fields through NSBAS-based time-series analysis (Ló pez-Quiroz et al. 2009). Root mean square errors (RMSEs) between InSAR velocities and GNSS velocities projected to LOS directions were between 1.3 mm/year and 2.1 mm/year, which suggested that InSAR velocity fields were consistent with those of GNSS. All LOS velocity fields showed long-wavelength phase variations reflecting interseismic deformation at the Nankai subduction zone. InSAR velocities showed a maximum of 27 mm/year at Cape Muroto and gradually decreased with increasing distance from the Nankai Trough. Obvious velocity gradient changes around inland faults such as the Median Tectonic Line were not recognized, which was consistent with a previous study (Shimotsuma et al. 2026). Therefore, we did not consider inland fault effects in the following fault slip deficit estimation.

Next, we estimated slip deficit rates on the Nankai subduction zone through fault slip inversion analysis. We used 115 GEONET stations mainly located in the Shikoku or Chugoku districts, 9 GPS-A stations in the Nankai offshore area, and InSAR velocities in Shikoku. Before the inversion, we reduced the number of InSAR pixels by performing a uniform downsampling procedure to approximately 4 × 4 km pixel interval. The subducted plate interface was divided into 25 × 25 km subfaults based on the plate geometry model (Iwasaki et al. 2015). Displacements due to subfaults’ slip were calculated using Okada dislocation model (Okada 1992) assuming a homogeneous elastic half-space. We imposed the smoothing constraint for slip deficit distribution as prior information, whose strength was determined based on Akaike Bayesian Information Criterion (ABIC) (Yabuki and Matsu'ura 1992). We iterated this inversion analysis by changing the relative data weight ratio between GNSS and InSAR, and we adopted the optimal ratio which minimized the weighted residual sum of squares (Maubant et al. 2023). Compared with a slip deficit rate distribution estimated from only GNSS, the distribution from GNSS and InSAR showed weaker smoothing. Considering the ABIC definition, this was attributed to the increasing amount of data, which reduced the penalty for violating the smooth slip. By adding InSAR data, slip deficit rates at the depth of about 40 km showed higher slip deficit rates, and slip deficit rates around the central Shikoku showed lower values.

What is the most suitable fault model for slip inversions, given a distribution of InSAR-measured displacements and its corresponding spatial correlation information? To answer this question, we develop a quantitative method to determine the optimal smoothing and weighting for correlated InSAR data as well as the optimal fault discretization by adopting an Akaike Bayesian Information Criterion (ABIC)-minimizing approach, and apply it to multiple tracks of InSAR data covering the creeping central San Andreas fault.

Although InSAR offers surface displacement measurements with fine spatial resolution (40 m or less), since both the geophysical signals and the atmospheric noise in the data are highly spatially correlated, the measurements are almost certainly not independent of each other. Inversions for fault slip from InSAR data can be strongly affected by spatially-correlated atmospheric noise. Without properly taking into account such correlations when formulating the inverse problem, for example by weighting the data based on their expected covariances, such models can be biased and poorly resolved. Additionally, since the data is correlated, the actual information content the data offers is significantly smaller than the number of data points. Often, one downsamples the data to reduce the redundancy of data points and has to regularize the inversion (e.g. by applying a smoothing constraint) due to the inversion being ill-posed. The common way to downsample the data is through quadtree downsampling, which is based on the variance of the measurements in each downsampled block; however to correctly account for spatially correlated noise, the downsampling process should also account for its correlation lengthscale.

Beyond the method of downsampling data, several factors and choices made in designing a slip inversion can affect the final result. Traditional methods for choosing the strength of smoothing, like the L-curve, are somewhat subjective, relying on human interpretation. Another key factor is how the fault is discretized. Often this is done with resolution in mind, which is also related to the spatial distribution of the data. For slip models constrained with surface observations, we typically find resolution is good for shallower portions of the fault, motivating the use of smaller shallow fault elements, but degrades appreciably as depth increases, motivating the use of larger deep fault elements. Therefore, together with the consideration of data covariance and its spatial distribution, there ought to be an optimal fault discretization for the information content the data contains. Thus, we build into our approach a method for testing multiple candidate discretizations.

In this study, we seek to quantify the effects that data distribution and covariance has on slip inversion, and explore the optimal fault model that they permit. Our target is to model the creep distribution on the central San Andreas fault, using data from UAVSAR (14 tracks) and Sentinel-1 (3 tracks). We estimate the covariance structure of the noise in each InSAR dataset by first removing a preliminary estimate of the tectonic signal and calculating the autocorrelation function of the residuals, and its radial average. We approximate the covariance as a function of distance by fitting an exponential function to this averaged autocorrelation. Second, we downsample the InSAR measurements and generate perturbed data by adding spatially correlated noise with the same covariance. Next, we randomly generate thousands of fault geometries, with the goal of determining the “best” one. We fix the strike and dip angle, and the total length and width of the whole fault. For each geometry realization, we first randomly draw the number of layers and the number of patches. Second, we draw the width of the first layer within a bound and enforce the width of each layer to grow exponentially that will eventually be constrained by the total width and the number of layers. The randomness of the generation should capture all combinations of the given constraint considering a sufficient sample size. After generating the fault geometries, we calculate the ABIC value for each fault model. The ABIC formulation takes into account of 1) the relative weighting of the smoothing with respect to the data, 2) number of independent model parameters and 3) the effective number of data points. The model with the smallest ABIC value represents the optimal geometry, smoothing and slip. Lastly, we invert for fault slip based on the optimal fault model geometry.

Our results show significant strike-slip, consistent with the fast creeping behavior of the central San Andreas fault. Creep peaks around Bitterwater at rates of around 3.2 cm/yr which tapers down to both the north (~1.6 cm/yr, partitioned onto the Calaveras fault) and south (~0.5 cm/yr, with partial locking at Parkfield). However, we find that modeling only the strike-slip component of fault slip is insufficient to explain the data – we identify significant fault-perpendicular as well as vertical surface deformation signals that require modeling both the tensile opening/closing and the dip-slip components of motion. The dip-slip component is mostly seen around the Parkfield segment, which could result from accommodating the partial locking at depth. For the tensile deformation, we find a general closing rate around 0.5 cm/yr along the fault except for the Bitterwater section, which shows tensile opening at a similar rate. Possible mechanisms for such tensile deformation could be the interplay between regional stress orientation and the fault geometry, where the fault is clamped by stress acting on a higher angle and relaxed when the angle is more gentle, or could be the response of changes in pore fluid pressure at depth. Detailed investigation is needed to address the actual physical mechanism acting on the fault.

The Pamir and Tianshan orogens are two major tectonic units in western China and Central Asia, both of which have experienced direct or far-field effects from the India–Eurasia collision during the Cenozoic. Although both regions are actively deforming, they exhibit significant differences in their deformation mechanisms. Deformation in the Pamir Plateau is likely driven by a dual intracontinental subduction system, whereas the Tianshan orogen deforms primarily through crustal shortening and thickening. Existing GNSS stations, though sparsely distributed, reveal substantial strain accumulation across the orogens; however, they provide limited information on vertical motion due to the scarcity of observations. InSAR offers a more effective approach for mapping surface deformation in this challenging environment, providing improved spatiotemporal sampling. Therefore, accurate extraction of surface deformation from SAR data is essential for understanding the contrasting deformation mechanisms of these two tectonic units and their potential interactions.

By exploiting the extensive archive of Sentinel-1 SAR data over the Pamir–Tianshan orogenic system, we are able to map tectonic deformation at an unprecedented resolution of ~200 m—a scale unattainable using GNSS alone in such a tectonically active region. To mitigate decorrelation effects inherent in C-band SAR time series spanning approximately ten years, especially over rugged topography with few man-made targets, we apply a persistent scatterer InSAR approach (modified version of the StaMPS PS-InSAR workflow) to process the data while preserving long time-series coherence. We integrate ascending and descending SAR data acquired from 28 orbital tracks to ensure dense PS coverage across the entire region, particularly over the Pamir Plateau, where even horizontal GNSS measurements are scarce. The PS-InSAR method offers advantages over conventional small-baseline (SBAS) algorithms, which rely on multilooked, filtered images and retain only the most coherent signals over short intervals. In contrast, the PS approach utilizes more complete time-series records, yielding higher accuracy for both horizontal and vertical deformation estimates.

Given that the PS method operates on full-resolution SAR images with extended time-series acquisitions and concatenated frames, we parallelize the entire workflow on a cluster to efficiently process both conventional single-look interferograms and StaMPS-derived PS targets. Cluster-based processing enables us to analyze surface deformation with efficiency comparable to SBAS methods, but with a more complete spatial-temporal coverage. To better constrain Line-of-Sight (LOS) deformation, we incorporate established error correction techniques originally developed for SBAS into our PS processing framework. These include corrections for solid Earth tides (SET), tropospheric delays based on ECMWF ERA5 models, and ionospheric delays using JPL’s GIM models. Residual errors are further mitigated through the common scene stacking (CSS) technique, allowing the final time-series to support both average LOS velocity estimation and temporal displacement analysis—essential for vertical deformation studies. Our results demonstrate promising PS coverage across the entire region, including the Pamir Plateau.

Due to the absence of GNSS vertical deformation data over such a vast area, we develop a simple fusion method to derive east and vertical surface deformation by combining PS-derived LOS deformation with interpolated horizontal GNSS velocities. GNSS data are interpolated onto the PS network to estimate north–south deformation, yielding a high-resolution (~200 m) horizontal deformation field across the study area. This high-resolution field is subsequently averaged to resolutions of ~1000–5000 m for horizontal strain mapping.

Our results highlight the potential of PS-InSAR for analyzing deformation in large orogens, maximizing the value of SAR data for both tectonic and non-tectonic applications (e.g., aquifer and groundwater studies). While conventional SBAS methods efficiently capture first-order deformation patterns across broad spatial scales—such as across tectonic units—the PS approach offers enhanced capability for vertical deformation mapping, especially across active deforming zones and their boundaries. We employ horizontal PS velocities for 2D kinematic modeling of permanent strains and elastic deformation across the Pamir and Tianshan orogens. Our findings align broadly with previous fault slip rates and locking depths, while revealing greater detail along a refined fault network. Although the results over the Tianshan region are well explained by 2D models, the Pamir Plateau presents a more complex scenario, where intracontinental subduction is inferred from multiple geophysical imaging studies. Accordingly, we are now applying a 3D model integrating both horizontal and vertical PS velocities to better understand Pamir’s deformation dynamics. Detailed results and interpretations will be presented at the workshop.

The 14 August 2021 Mw 7.2 Nippes, Haiti earthquake ruptured a major portion of the Enriquillo–Plantain Garden Fault (EPGF) system along the southern peninsula of Haiti, producing widespread damage and complex patterns of coseismic and postseismic deformation. Post-earthquake geodetic investigations using L‑band ALOS‑2 and C‑band Sentinel‑1 InSAR revealed several instances of triggered shallow creep on both the primary EPGF strands and nearby thrust faults located within, or immediately adjacent to, the rupture zone. These early observations suggested that fault creep may have played a role in accommodating part of the coseismic strain release, but the spatial extent, duration, and kinematics of this creeping behavior remained incompletely characterized.

Using a comprehensive Sentinel‑1 time‑series analysis, we demonstrate that the triggered creep was significantly more persistent and spatially extensive than previously understood. Several EPGF segments continued to creep for up to ~300 days after the mainshock, indicating prolonged shallow fault afterslip and a sustained redistribution of stress along the fault system. The creeping zones extend westward from the main rupture area deep into the Miragoâne pull‑apart basin, and their distribution exhibits pronounced along‑strike variability, revealing clear evidence of structural segmentation within the EPGF. This segmentation likely reflects variations in lithology, fault frictional properties, and near-surface structural complexity, all of which influence the extent to which individual segments undergo stable sliding versus stick‑slip rupture.

To assess the relationship between creep and seismicity, we compare the spatiotemporal evolution of the observed postseismic slip with the relocated aftershock catalog of Douilly et al. (2023) and with the sequence of M 5+ earthquakes that occurred on 24 January 2022, approximately four months after the mainshock. This comparison shows that the triggered creep was not entirely aseismic; instead, the slip on several shallow, off‑fault structures appears to have initiated the January 2022 earthquake sequence. That sequence, in turn, appears to have reactivated creep on the main EPGF strand. This cascade of creep–seismicity interactions strongly indicates velocity‑weakening behavior on portions of the EPGF and highlights the potential for triggered slip to modulate loading rates on adjacent fault segments. These findings carry important implications for understanding postseismic strain redistribution, assessing the seismogenic potential of off‑fault structures, and evaluating the frictional and mechanical properties governing the broader EPGF fault system.

Active deformation in the Eastern Central Andes is caused by east-west shortening. Geologic and instrumental records indicate active compressional structures that potentially host large earthquakes, as indicated by significant geodetic strain. Our study area covers a latitudinal extent of approximately 1,000 km, from southwestern Bolivia (~16°S) to northwestern Argentina (~26°S). Our aim is to quantify and localize thin-skinned (Bolivia) vs. thick-skinned (Argentina) shortening patterns across the region.

We analyzed ten years (2015-2025) of Interferometric Synthetic Aperture Radar (InSAR) time-series to measure surface deformation rates. We rely on descending ALOS-2 radar imagery (L-band) acquired in wide-swath (~350 km) ScanSAR mode. We used the "alos2stack" workflow in the ISCE-2 software and substantially downsampled the interferograms to suppress noise, resulting in a ground-range pixel spacing of ~136 m. We generated deformation time-series with the MintPy software and applied corrections for topography, Solid Earth tides, plate motion, and stratified tropospheric signal delay using ERA5 weather models. We further applied a split-spectrum method to suppress the ionospheric phase contribution. The resulting rate maps are complemented by published, pointwise displacement rates from accurate positioning (GNSS) projected into the satellite line-of-sight (LOS). We then compare the LOS rates along multiple cross-orogen transects with geology, fault databases, seismicity, and topography.

The deformation rates exhibit kinematics from the Puna Plateau through the Eastern Cordillera to the highly vegetated Subandes. They reveal a variety of active ongoing processes in the eastern Central Andes, including inflation at Cerro Overo volcano (~1.5 cm/yr LOS). The dataset also reveals the dynamics of salars, with LOS motion toward the satellite at Salar de Arizaro (~0.7 cm/yr LOS) and LOS motion away from the satellite at Salar de Olaroz (~-2.6 cm/yr LOS). Significant motion away from the satellite (interpreted as subsidence) occurs in agricultural areas of Bolivia, including Punata (~-2.3 cm/yr LOS) and Cochabamba (~-1.2 cm/yr LOS). The deformation rates also exhibit ~8 cm of coseismic displacement associated with the 2020 magnitude Mw 5.8 Humahuaca normal-faulting earthquake in the Argentine Andes. Finally, we identify several landslide-related deformation signals, predominantly in Bolivia, concentrated in steep, deeply incised valleys along the eastern flank of the Eastern Cordillera and in the Interandean Zone.

Tierra del Fuego island is crossed by the South America-Scotia plate boundary, with a relative motion of ~5-6 mm/year between them. The boundary is mainly represented by the Magallanes-Fagnano Fault (MFF), which held a historic M ~7.5 earthquake doublet in 1949. While publication of new data in the present century has improved our understanding of the MFF, its seismogenic potential and possible segmentation remain poorly constrained. We present new geodetic evidence that allows for the first estimation of strain accumulation along the entire inland portion MFF from Seno Almirantazgo to the Atlantic ocean shore. Data from two GNSS stations deployed ~8 km north and south of the fault trace, as well as ascending and descending velocity fields derived from Sentinel-1 SAR data, record interseismic velocity of ~6 mm/year along the entire inland portion of the plate boundary, in agreement with previous estimations. However, Bayesian back-slip dislocation models at different longitudes along the fault show along-strike variations of the locking-depth, suggesting the MFF is indeed segmented. We compare our results with current geodetic and geomorphic evidence discussing potential segmentation from Holocene until the present, incluiding potential discrepancies on the previously proposed rupture estension of the 1949 events.

On 2021 January 18, a blind mid-crustal Mw ∼6.4 earthquake occurred near San Juan, Argentina. The observation of associated ground deformation with single interferograms is obscured by strong tropospheric signals. We apply appropriate corrections to the data and reconstruct the deformation field associated to the event through InSAR time-series approach. We show it is possible to retrieve this signal to invert the fault parameters. The observed ground deformation is consistent with a high angle NW-dipping fault plane at a centroid depth of ∼19 km. The geometry of this fault supports the reactivation of pre-existing structures within the Cuyania Terrane, suggesting a direct structural connection and strain transfer to the actively deforming, east-vergent Precordillera front. We analyse our findings to deduce a static friction coefficient ≤ 0.3 for mid-crustal faults of the region.

The 2016 Mw 7.8 Kaikōura earthquake was characterized by exceptionally complex rupture behavior involving multiple fault segments and multidirectional displacement over a total rupture length of approximately 170 km. We build on previous studies of the three-dimensional (3D) coseismic displacement associated with this event by reconstructing a refined and comprehensive 3D deformation field using an extensive suite of ALOS-2 and Sentinel-1 synthetic aperture radar (SAR) observations. We derive both interferometric phase and pixel-offset measurements from carefully screened, artifact-free datasets, and apply additional atmospheric and ionospheric corrections. The SAR processing is carried out using modern Python-based workflows that incorporate atmospheric corrections, selective dataset exclusion, coherence-based masking, and optimized spatial downsampling strategies to reduce noise while preserving deformation signals. Six independent satellite tracks from ascending and descending viewing geometries are jointly analyzed to revisit the coseismic deformation nearly a decade after the earthquake, providing substantially improved spatial coverage and robustness relative to earlier reconstructions. Line-of-sight and azimuth displacement measurements from all tracks are combined using a weighted least-squares inversion to resolve displacement in the east, north, and vertical directions, followed by the removal of planar ramps and residual topography-correlated signals to mitigate long-wavelength orbital and systematic artifacts. This integrated approach significantly improves the signal-to-noise ratio of the 3D offset measurements relative to previous studies and enhances internal consistency across sensors and viewing geometries. The resulting 3D displacement field reveals pronounced spatial variability consistent with the highly segmented, multi-fault rupture geometry of the Kaikōura earthquake, with clearer expression of individual fault strands and improved constraints in remote and poorly accessible regions. Updated displacement estimates are obtained for key fault segments, highlighting variations in slip magnitude and direction that reflect complex rupture kinematics. The derived displacement field is validated against independent field observations, photogrammetry-derived 3D point-cloud data, and GNSS measurements, demonstrating strong agreement across all displacement components. Overall, this study demonstrates that integrating multi-sensor, multi-track InSAR data with advanced preprocessing, masking, and inversion strategies substantially improves 3D coseismic displacement estimation and provides a robust foundation for future earthquake source characterization and subsurface fault slip analysis.

Earthquakes originating near Earth’s surface pose significant hazards to human safety and infrastructure, as their associated surface deformation can result in widespread structural damage and loss of life. Accurate characterization of coseismic displacement, the ground-surface motion caused by earthquakes, can provide critical insight into fault geometry and surface rupture processes as well as actionable information for disaster response and mitigation. Large-magnitude, shallow earthquakes often affect spatially extensive and remote regions, where both the scale of impact and limited accessibility hinder comprehensive ground-based characterization. Synthetic aperture radar (SAR) interferometry (InSAR), as well as pixel offset tracking of both SAR and optical imagery, can provide detailed measures of coseismic displacement within days of an earthquake and over broad regions, enabling more rapid characterization of the impacted region and associated emergency response. Despite the rapidly growing availability of SAR and optical satellite products, InSAR and pixel offset tracking processing routines present computational and data storage challenges to global-scale earthquake analyses. Additionally, displacement estimates derived from instruments with different operating frequencies complicates their integration and interoperability in surface deformation analyses and downstream modeling workflows. These barriers highlight a critical need for standardized, analysis-ready products that can be seamlessly analyzed within GIS platforms by end-users spanning a range of disciplines and technical backgrounds.

The Advanced Rapid Imaging and Analysis (ARIA) project at the NASA Jet Propulsion Laboratory has developed a global archive of accessible, standardized, and analysis-ready coseismic displacement products derived from spaceborne SAR (Sentinel-1A/B/C) and optical (Sentinel-2A/B/C) imagery to facilitate more comprehensive studies of earthquake rupture processes and improve estimates for downstream rapid response efforts. The archive currently includes coseismic displacement products for all large (M 6.0 and greater) and shallow (30 km or less) onshore and near-shore earthquakes from October 2014 to present, and is continually updated with products corresponding to new events meeting these significance criteria. This dataset will soon be expanded to include coseismic displacement products derived from ALOS-2 PALSAR-2 ScanSAR data to provide coseismic displacement estimates in densely vegetated regions where shorter-wavelength sensors often lose coherence. Our product archive is unique from existing coseismic displacement product databases in terms of the data provided, data format, and data accessibility. First, our 30-meter resolution products are sensor-agnostic and provided in standardized units; this allows for the simultaneous use of multiple data sources and enables more rapid integration into GIS platforms and modeling workflows. Second, our product generation workflow integrates multiple scenes along the satellite track to capture the full rupture zone, eliminating the need for downstream mosaicking. Third, the availability of complementary SAR and optical coseismic displacement estimates for each earthquake provides increased sensitivity to surface displacement via pixel offset tracking. Finally, correction layers for solid-earth tides and ionospheric propagation path delays are embedded directly within the analysis-ready products for the end-user. Our workflow leverages the existing ARIA-HyP3 framework and capabilities to cost-effectively generate coseismic products for hundreds of earthquakes in the cloud. For future events meeting our significance criteria, our workflow implements an auto-trigger mechanism to generate coseismic displacement products at 90-meter resolution as soon as post-seismic data are available, delivering results with low latency (<24 hours after source data availability) to provide critical information regarding surface deformation and damage extents. In this presentation, we will demonstrate the product generation workflow and capabilities, as well as examples of earthquake science use-case and disaster response applications that showcase the advantages of our automated, standardized, and sensor-agnostic coseismic displacement products.

Detecting millimeter-order surface displacements by InSAR-related techniques like SAR time series analysis is still one of challenges due to several noises such as the atmospheric propagation delay effect, errors in phase unwrapping, and spatial and temporal decorrelation. Because of recent increase of SAR earth observation satellites and improvement of the recurrence period of each SAR satellite, now we can utilize over 100 SAR images for the time series analysis and thus can achieve millimeter-order displacement velocity detection, even though most of such events were limited to having the permanently displacing nature. However, SAR time series now struggles to detect transient, short time scale displacement like a slow slip event (SSE), which is one of urgent issues to be tackled with. Here we are trying to develop an anomaly detection model that is optimized for the InSAR time series analysis for the purpose of automatically detecting unknown SSEs.
Our proposed anomaly detection model uses the sliding window k-nearest neighbor method to detect surface deformation signal objectively. The sliding window k-nearest neighbor method is one of machine learning methods for a general time series data. In this method, one time series data was grouped together with multiple adjacent observations and represented as a collection of multiple partial time series as vector. Then, the multiple partial time series before SSE were used as a reference. The minimum distance between the partial time series for anomaly detection and reference partial time series were defined as the anomaly level.
To evaluate the performance of the proposed model, we prepared a pseudo time series dataset by simulating InSAR observations. As for a pseudo InSAR image generation, we simulated a displacement signal and a turbulent delay signal caused by the neutral atmosphere. Because state-of-the-art atmospheric delay correction methods exhibited effective removal for the stratified delay effect and the ionospheric phase disturbance, we did not consider them in this experiment. The turbulent component of the atmospheric delay was produced by applying a two-dimensional Gaussian random field. The displacement signal was simulated using the Okada rectangular dislocation model with a reverse-type slip, which is a typical situation of the SSE. We prepared several scenarios with different maximum displacement amplitudes: from 0.0 cm to 3.0 cm with an interval of 0.1 cm. In the InSAR time series simulation, we considered that SAR observations were conducted 50 times every 14 days by referring ALOS-2/-4, and that the SSE continued for 2 months with the constant slip velocity. To make the pseudo InSAR time series data more realistic, we applied spatial and temporal gaussian filtering to suppress the noise effect before anomaly calculation. To assess the detection ability statistically, we prepared 100 InSAR time series datasets for each displacement scenario.
The sliding window k-nearest neighbor method successfully detected anomaly where the signal to noise ratio of over 0.4 (ratio of the maximum displacement amplitude to the maximum noise amplitude). Looking at the transition of anomaly in one time series, during the period when the transient displacement was occurring, the anomaly level also showed increase associated with increasing the accumulated displacement. After finishing the displacement, the anomaly level kept high because the displacement was a secular one. To suppress the residual delay noise effect in the estimated anomalies, we applied the stacking procedure in the time dimension to the pseudo InSAR time series. After the stacking was applied to the simulation data, the high anomaly area associated with the displacement was enhanced. Visually, we clearly detected high anomaly value when the maximum displacement amplitude over 1.3 cm. By statistical assessment with a parametric and non-parametric tests (Welch's t-test and Mann-Whitney U test), our proposed anomaly detection model could significantly detect displacements with 0.9 cm.

Damage Proxy Mapping (DPM) is a powerful remote sensing tool to effectively deploy emergency resources (technical, personal, or financial) in the aftermath of natural disasters. Organizations such as the United Nations Satellite Center (UNOSAT) or the European Union's Copernicus Emergency Management Service (Copernicus EMS) produce such damage maps by comparing pre- and post-event optical and radar imagery acquired from air- or spaceborne sensors. The maps are then distributed to governments and first responders of the affected regions. Due to the urgent nature of the product, the focus in generating DPMs is on speed and reliability, using the often-limited datasets at hand rather than producing the highest-quality output, which could take months or even years to generate.
Typically, DPMs are created through manual mapping from high resolution optical imagery, which offers easy visual interpretation. However, machine learning approaches are progressively being adopted to further automate and significantly accelerate production and response workflows. A key limitation of optical imagery, however, is its reliance on clear skies. To address this, methods based on Synthetic Aperture Radar (SAR) have been proposed as SAR provides an all-weather and day-and-night data acquisition capability. Yet, SAR-based damage analysis is hindered by its lower resolution (at least regarding spaceborne sensors), higher computational demands, and the complexity of interpreting radar data.
In recent years, deep-learning based DPMs have shown great potential. These methods handle the large volume of radar datasets effectively and produce interpretable results rapidly once trained. In particular, Stephenson et al. (2022) used a Recurrent Neural Network (RNN) on transformed coherence timeseries to produce DPMs for various study regions affected by earthquakes. A key factor for their successful performance was the use of the entire coherence timeseries available up until the event date, which could span several years. However, this required the preprocessing of large data collections for each single study area, including the coregistration of stacks of single look complex (SLC) imagery and the computation of a coherence timeseries in the highest possible spatial and temporal resolution. The significant storage and computational demands of these steps likely contributed to the method never being widely adopted by operational DPM providers (to the knowledge of the authors).
In this study, we present initial steps to adapt the Stephenson et al. (2022) method to the terrabyte High Performance Computing (HPC) cluster, a cooperation by the Leibniz Supercomputing Center (LRZ) and the German Aerospace Center (DLR). The goal of terrabyte is to enable both operational Earth observation data processing as well as research and development. This infrastructure is also used by DLR’s Center for Satellite Based Crisis Information (ZKI), which produces DPMs and other disaster-related products from SAR imagery (though not yet based on RNN-techniques). Our study aims to facilitate future integration of an RNN-based workflow into cluster-based operational disaster response processors. Among the improvements we present are data-parallel training, the integration of coherence timeseries generation into operational SAR processors, the preprocessing of high seismic risk areas in preparation of potential future earthquake events, and the incorporation of ancillary datasets (e.g., precipitation, topography, and land use) into the training algorithm to assess their effect on the assessment accuracy.

Understanding low-magnitude crustal deformation in intraplate regions remains a major scientific challenge, particularly where tectonic signals are subtle yet seismically significant. We present an integrated GNSS–InSAR framework to investigate active deformation along the Delhi–Aravalli Ridge (DAR), a prominent Proterozoic orogenic belt within the northwestern Indian Shield. Despite its classification as part of a stable continental interior, the DAR experiences recurrent low-to-moderate seismicity, especially beneath the densely urbanized Delhi–National Capital Region (NCR). The coexistence of ancient structural inheritance, ongoing Himalayan convergence, and intense anthropogenic modification makes this region ideal for multi-sensor geodetic investigation.

Continuous and campaign-mode GNSS observations across the DAR reveal low horizontal deformation rates, generally below 2 mm/yr. Although small in magnitude, these velocities are spatially coherent and geodetically robust, indicating measurable strain accumulation along inherited Proterozoic shear zones and fault systems. The orientation of velocity vectors suggests differential motion between crustal blocks bordering the ridge system, consistent with stress transmission from the Himalayan collision zone to the north. Rather than behaving as a rigid block, the DAR appears to act as a mechanically weak corridor that accommodates distributed intraplate strain. These findings reinforce the concept that stable continental regions may host slow but persistent deformation along pre-existing zones of weakness.

While GNSS provides precise three-dimensional velocities at discrete points, it is limited in spatial density. To overcome this limitation, we employ multi-temporal InSAR time-series analysis to derive high-resolution surface displacement fields across the Delhi–NCR and adjoining sectors of the DAR. InSAR observations reveal pronounced vertical deformation patterns, with localized subsidence reaching approximately 15 mm/yr in certain urban and peri-urban zones. Much of this deformation correlates with groundwater extraction, sediment compaction, and infrastructure loading. However, deformation gradients frequently align with mapped basement faults and lithological boundaries, suggesting structural control over the spatial distribution of subsidence.

The synergy between GNSS and InSAR is central to resolving the origin of observed deformation. GNSS establishes a stable regional reference frame and constrains long-wavelength horizontal strain accumulation. InSAR, in contrast, resolves short-wavelength vertical displacements at meter-scale spatial resolution. By integrating GNSS-derived velocities with InSAR line-of-sight time series, we separate horizontal tectonic strain from vertical subsidence signals and minimize ambiguities related to orbital errors and atmospheric artifacts. This joint inversion approach enhances the reliability of deformation estimates in a region where tectonic signals are subtle and easily masked by anthropogenic effects.

The combined dataset reveals that deformation in the DAR is neither purely anthropogenic nor purely tectonic. Instead, surface displacement reflects the superposition of long-term strain accumulation along reactivated shear zones and short-term hydrological and infrastructural loading. GNSS data indicate slow accumulation of elastic strain across structural corridors, while InSAR detects localized compaction within sedimentary basins bounded by faults. In several locations, subsidence compartments appear structurally confined, implying that ancient fault geometries influence present-day groundwater-driven deformation. Such structural modulation may alter local stress fields, potentially influencing seismic hazard in densely populated areas.

From a tectonic perspective, the deformation pattern supports a model involving differential motion between adjacent cratonic domains and reactivation of Proterozoic shear systems within the DAR. Far-field compressional stresses associated with ongoing Himalayan convergence propagate southward into the Indian Shield, exploiting mechanically weak zones. Although strain rates are low, the cumulative effect over decades to centuries may contribute to intraplate seismicity. Seasonal hydrological loading further modulates stress conditions, introducing temporal variability that may influence earthquake triggering in critically stressed faults.

For the built environment, the implications are significant. The Delhi–NCR hosts critical infrastructure, high-rise development, and lifeline systems that are sensitive to millimeter-scale ground motion. InSAR-derived deformation maps enable identification of subsidence hotspots and infrastructure vulnerability zones, while GNSS ensures long-term stability monitoring. Together, they provide a scientifically rigorous framework for hazard assessment, urban planning, and risk mitigation in slowly deforming continental interiors.

This study demonstrates that the integration of GNSS and InSAR is essential for detecting and interpreting subtle intraplate deformation. In regions like the Delhi–Aravalli Ridge, where tectonic signals are small yet societally important, multi-sensor geodesy bridges the gap between deep crustal processes and surface impacts. The proposed framework advances the science and applications of SAR interferometry in earthquake studies, seismic hazard evaluation, and infrastructure monitoring, offering a transferable methodology for stable continental regions worldwide.

Understanding how deformation is accommodated within the accretionary prism is important for accurately characterizing the seismic potential during large megathrust events. Observations from large subduction zone earthquakes, such as the 2011 Tohoku earthquake, have shown that faults within the accretionary prism can rupture together during the coseismic event. In addition, some accretionary prisms may accommodate deformation inelastically through folding, as seen in the Zagros fold-and-thrust belt in Iran. Identifying the contribution of the different deformation mechanisms is important as they influence how strain accumulates and is released with time, changing the seismic and tsunami hazard in subduction zones.

However, studying deformation within the accretionary prism in most subduction zones can be challenging, given the distance from land-based GNSS and the sparse distribution of GNSS-A. In this study, we use observations from ALOS-2 wide-swath from 2015 to 2022 to analyze the deformation on the overriding plate of the Rakhine-Bangladesh megathrust, a unique subaerial accretionary prism, but which has proven difficult to study with InSAR due to dense vegetation and strong ionospheric influence. We produced interferometric time-series from 49 descending ALOS-2 scenes using ISCE and Mintpy, incorporating ionospheric correction using the split-spectrum method. We developed an enhanced workflow to improve the removal of the ionospheric signal by reducing the number of looks by half and doubling the Goldstein filtering patch size.

We successfully identified localized ground deformation signals, including anticlines with interseismic uplift signals up to 2 mm/yr. We were unable to fit the observations using fault-based models without having the fault slip rate exceed the total convergence rate in the region, which is physically implausible. Instead, we propose that these interseismically uplifting anticlines could be explained by a process of active aseismic folding within the accretionary prism above a shallow decollement, and show that the anticlinal spacing is consistent with this ductile mechanism given the depth to the decollement. These results suggest that the strain energy available to drive coseismic rupture could be lower than predicted by purely elastic models, highlighting how accurately constraining the potential for ductile deformation in the overriding plate is an important consideration for improving future seismic hazard assessments.

Rapid and reliable post-seismic damage assessment is critical for effective emergency response, situational awareness, and resource prioritization following large earthquakes. Synthetic Aperture Radar (SAR) imagery offers significant advantages for such applications due to its ability to acquire data under all weather and illumination conditions. This study investigates the use of C-band Sentinel-1 SAR backscatter change detection for rapid delineation of urban damage following the 6 February 2023 Mw 7.8 Türkiye earthquake. Two severely affected urban areas, Hatay and Kahramanmaras, were analyzed using ascending (Track 14) and descending (Track 21) Sentinel-1 GRD acquisitions. Field surveys were conducted to identify completely collapsed buildings within the selected study areas. These data were used for quantitative validation, including 6,401 collapsed buildings in Hatay and 1,205 in Kahramanmaras. A multi-temporal pre-event baseline was constructed from five images acquired before the earthquake. Processing included the application of precise orbit files, thermal and border noise removal, radiometric calibration, speckle filtering using a Lee Sigma filter, and geometric correction based on the Copernicus Digital Elevation Model. Backscatter coefficients were converted to a decibel scale and organized into pre- and post-event image stacks. Post-event changes were calculated by subtracting the mean pre-event reference from two post-event acquisitions: the first available image immediately after the earthquake (February 2023) and a later acquisition approximately two months afterward (April 2023). Significant negative anomalies were identified using an adaptive threshold based on the global mean and standard deviation of the change image.

The results indicate clear temporal variations in detection performance. For the first post-event acquisition (February), Building Detection Rates (BDR) ranged from 4.9% to 16.1%, while for the April acquisition, they ranged from 9.5% to 16.1%, demonstrating improved performance in most analyzed scenarios. In Hatay, BDR increased from 4.9% to 12.8% in ascending geometry and from 7.1% to 16.1% in descending geometry. Descending observations consistently showed superior performance, suggesting an influence of urban orientation relative to radar viewing direction.

To mitigate effects associated with spatial resolution and geometric uncertainties, a 15 m buffer was applied during the spatial intersection analysis. The application of the buffer significantly improved area-based precision, reaching up to 58.4%, showing the sensitivity of spatial evaluation to sensor resolution.

Although individual building detection remains constrained by the spatial resolution of Sentinel-1 products, the results confirm that SAR backscatter change detection provides a rapid, automated, and scalable approach for large-area post-seismic screening, with potential for further improvement through the integration of higher spatial and temporal resolution datasets.

These findings highlight the operational value of Sentinel-1 SAR data for rapid, automated, and scalable large-area post-seismic damage screening, particularly in the critical early response phase.

Co-seismic surface deformation mapped with interferometric Synthetic Aperture Radar (InSAR) may provide valuable insights into earthquake mechanics, revealing information about the rupture in areas where traditional seismological methods are not efficient due to lack of suitable seismic network geometry. Conventional geodetic approaches utilize complex, non-linear inversions of InSAR images to estimate geometric (e.g., azimuth, width, dip) and kinematic (e.g., dip-slip, strike-slip throws) quantities. However, these methods may prove computationally expensive to apply, while manual parameter bounding introduces biases. Herein, we seek to explore the potential of a deep-learning approach to reframe this problem as a faster computer vision classification and regression task.

We aim to train ESPI-ResNet (Zhao et al., 2021) on a synthetic dataset to evaluate its applicability and performance. ESPI-ResNet is a deep-learning framework designed to extract earthquake fault parameters from InSAR imagery. It utilizes a deep residual convolutional network, employing skip connections to retain interferometric phase gradients, which splits in two branches (one per sub-task). First, it aims to solve a classification problem, i.e., identify whether the image represents a normal, reverse right-lateral or left-lateral strike-slip fault type. This is achieved by the deep convolutional neural network which extracts spatial deformation features from the interferogram. Second, the regression branch of the framework processes the extracted features through a sequence of dense layers to estimate the geometric parameters of the rupture. To produce the synthetic dataset, we used the Okada model (Briole, 2017), as real images that depict the sought-after surface deformation are too scarce to compose a very large data volume, necessary for training, validation and testing. A uniform slip distribution was assumed, with parameters configured according to seismological data of medium-to-strong earthquakes. High-resolution 1024-by-1024 images were produced. Even though, as explained next, the network expects a coarser 224-by-224 resolution, we decided to maintain initially high resolutions for posteriority and dataset completeness. The final synthetic dataset comprised ~47k interferograms of the four fault types, covering 50-by-50 km areas.

Before training, we followed an augmentation scheme to add diversity to the dataset and avoid a completely idealized input. Therefore, our augmentation was constrained to normalization, geometric, i.e., rotations, flipping and mirroring of the images, and pixel-level, i.e., brightness, contrast and color changes, operations. Through this, we sought to teach the model spatial invariance and make it resistant to changes in coherence levels. For training, we used a split with 70% of data for training, 10% for validation and 20% for testing. The model was optimized using the Adam optimizer, with an exponentially decaying learning rate. To prevent overfitting on the synthetic data and better allocate computing resources, we implemented an early stoppage mechanism which monitored validation loss (with a patience value of 5 epochs). This enabled training to stop early (before the 100 epochs) if the loss metric plateaued. A batch size of 128 was used.

Preliminary evaluation on the 20% test set demonstrated the network’s capacity for classifying fault types and identifying their respective geometric parameters. Our augmentation pipeline may have reduced classification accuracy and increased geometric errors but introduces variability in the learning procedure and prepares the model for application in imperfect conditions. In the future, we aim to incorporate a complex noise augmentation step, representing the expected noise sources in real InSAR imagery.

References

Briole, P., 2017. Modelling of earthquake slip by inversion of GPS and InSAR data assuming homogenous elastic medium. https://doi.org/10.5281/ZENODO.1098399

Zhao, X., Wang, C., Zhang, H., Tang, Y., Zhang, B., Li, L., 2021. Inversion of seismic source parameters from satellite InSAR data based on deep learning. Tectonophysics 821, 229140. https://doi.org/10.1016/j.tecto.2021.229140

Quantifying high-resolution interseismic strain rates is a key input for assessing fault-related seismic hazard and is commonly derived from geodetic observations, particularly Interferometric Synthetic Aperture Radar (InSAR) velocity fields. However, this process is complicated by noise inherent to both SAR data and subsequent InSAR processing workflows, necessitating the spatial filtering of the raw velocity field. Existing filtering approaches, such as Gaussian and median filters, often struggle to balance noise suppression with the preservation of sharp strain gradients. Gaussian filtering tends to introduce excessive spatial blurring ("blobbiness"), leading to systematic underestimation of peak on-fault strain magnitudes. In contrast, median filtering can cause spatial shifts in the location of concentrated fault strain in the presence of data gaps (e.g., data edges, decorrelation zones, or masked non-tectonic deformation). It may also generate Gibbs-ringing artifacts—oscillating sub-parallel signals adjacent to primary tectonic gradients—which can be misinterpreted as distributed deformation or spurious strain across multiple faults.

Here, we present a new nested bilateral filtering framework designed to more faithfully recover on-fault strain magnitudes while effectively suppressing off-fault noise. The method employs a bilateral filter that jointly leverages spatial-domain and value-domain weighting. To enhance numerical stability and reduce parameter sensitivity, a localized small-window spatial kernel is embedded within the value-domain weighting scheme, forming a nested structure that stabilizes edge preservation while maintaining robust noise attenuation. We first evaluate its performance using synthetic datasets based on a 2D screw dislocation model, demonstrating clear improvements over commonly used Gaussian and median filters. A systematic grid-search analysis further provides practical guidance for parameter selection. We then apply the method to recently compiled trans-continental-scale geodetic velocity fields across the Alpine–Himalayan Belt (AHB), encompassing diverse tectonic settings including the Pamirs, Tibet, Anatolia, Makran, and the Hindu Kush. The results more effectively capture both the magnitude and spatial localization of strain along major faults and reveal previously unresolved strain features in several tectonically active regions.

Finally, we discuss the implications of these improvements for geodetically informed seismic hazard assessment and continental deformation studies. These include: (1) more reliable identification of creeping fault segments; (2) improved quantification of slip-deficit rates and seismic moment accumulation rates; and (3) clearer characterization of on-fault versus off-fault strain partitioning across different tectonic domains.

On March 28^th, 2025, a Mw 7.7 earthquake struck central Myanmar, approximately 16 km west of Mandalay, along a segment of the Sagaing Fault, a major right-lateral strike-slip fault system. The shallow rupture generated severe ground shaking, resulting in more than 4,900 fatalities and approximately 6,000 injuries. Seismic effects were reported at distances exceeding 1,000 km from the epicentral area, with the collapse of a high-rise building in Bangkok (Thailand) potentially attributable to site-specific amplification associated with predominantly loose, unconsolidated sedimentary deposits (Shahzada et al., 2025). This study aims to identify the coseismic ground displacement field induced by the earthquake and to constrain the geometry and slip distribution of the seismogenic source. To this end, we integrated satellite-based Synthetic Aperture Radar (SAR) techniques, including Pixel Offset Tracking (POT), Multiple Aperture Interferometry (MAI), and Interferometric SAR (InSAR). These techniques were applied to a dataset of 12 C-band Sentinel-1 SAR images acquired in Interferometric Wide Swath (IWS) mode, covering the broad region affected by the seismic event. The resulting displacement fields were subsequently inverted through a two-step analytical modeling approach to retrieve the fault geometry and key source parameters (Atzori et al., 2009; Atzori and Antonioli, 2011; Atzori et al., 2019). The use of multiple Sentinel-1 datasets ensured full spatial coverage of the rupture zone. In fact, the results revealed a rupture extending approximately 490 km across three distinct fault segments, characterized by a nearly vertical, north–south orientation and predominantly right-lateral strike-slip mechanism, consistent with the regional tectonic setting. The maximum coseismic slip identified by the seismic source modeling reached about 5 m in the central portion of the fault, corresponding to the segment previously identified by Hurukawa and Maung Maung (2011) as the seismic gap. This study highlights that the integration of multiple SAR-based techniques—each characterized by distinct sensitivity and accuracy levels—proved effective in capturing the full deformation field over a wide area, with strong consistency observed across the different techniques. Furthermore, this approach also enabled the retrieval of both horizontal (east–west and north–south) and vertical displacement components, thereby increasing the robustness of the source model inversion, including its slip distribution. Overall, this comprehensive analysis provides important insights into the rupture extent and kinematic behavior of the earthquake, contributing to a more robust characterization of the seismogenic source and improving our understanding of rupture processes and seismic hazard in tectonically active regions.

REFERENCES

Atzori, S., Hunstad, I., Chini, M., Salvi, S., Tolomei, C., Bignami, C., Stramondo, S., Trasatti, E., Antonioli, A., & Boschi, E. (2009). Finite fault inversion of DInSAR coseismic displacement of the 2009 L’Aquila earthquake (central Italy). Geophysical Research Letters, 36(15). https://doi.org/10.1029/2009GL039293

Atzori, S., & Antonioli, A. (2011). Optimal fault resolution in geodetic inversion of coseismic data. Geophysical Journal International, 185(1), 529–538. https://doi.org/10.1111/j.1365-246X.2011.04955.x

Atzori, S., Antonioli, A., Tolomei, C., De Novellis, V., De Luca, C., & Monterroso, F. (2019). InSAR full-resolution analysis of the 2017–2018 M>6 earthquakes in Mexico. Remote Sensing of Environment, 234, 111461. https://doi.org/10.1016/j.rse.2019.111461

Shahzada, K., Noor, U. A., & Xu, Z.-D. (2025). In the wake of the March 28, 2025 Myanmar earthquake: A detailed examination. Journal of Dynamic Disasters, 1(2), 100017. https://doi.org/10.1016/j.jdd.2025.100017

The Northern Anatolian Fault Zone (NAFZ) is the major strike-slip boundary between the Anatolian and Eurasian plates, extending for approximately 1500 km and accommodating dextral motion at an average rate of 20–30 mm/yr. The western portion of the NAFZ, near the city of Izmit, was the site of a Mw 7.6 earthquake in August 1999 that produced over 120 km of surface rupture across several fault segments. This devastating event was one of the most recent in a mostly westward-migrating sequence of large earthquakes along the NAFZ since 1939. If this westward progression continues, the next significant earthquake could occur near Istanbul and other densely populated centers. In order to better estimate earthquake potential and seismic risk in this region, it is critical to develop a robust understanding of the seismic cycle and fault dynamics specific to the western NAFZ. This master’s thesis aims to quantify fault slip, model fault geometries, and analyze locking depths within this region of the NAFZ, using a combination of geodetic and seismological datasets. The study will employ joint inversion methods and time-series analysis to determine slip deficits and improve understanding of the coseismic and interseismic phases associated with the Izmit earthquake. Building on the work of Delouis et al. (2000, 2002), this research will use tools such as GMTSAR and Pyrocko to reanalyze the Izmit earthquake and provide new insight into uncertainty estimates for fault slip and slip deficits from the time of the event until 2025. Early results of the ERS data reanalysis show that SAR pixel offset measurements provide a complementary dataset that can help fill the gaps of SAR interferometry where coherence is low and improve data quality. Initial earthquake source analysis results are presented here, showing slip distributions and fault geometries. This thesis anticipates producing an updated kinematic model of the Izmit event derived from several data sources and analyzing time series data to estimate slip deficit along the western portion of the NAFZ. Ultimately, this research will provide additional uncertainty constraints on the fault slip experienced in this region and underscore the importance of integrating diverse datasets to enhance seismic-cycle and earthquake source analyses.

The Mw 6.5 Norcia earthquake, which struck Central Italy on October 30, 2016, represents the climactic and most destructive event of the recent Apennine seismic sequence. Nucleating within the complex, extensional Mt. Vettore-Bove Fault System (MVBFS), this mainshock ruptured a critical structural gap between the historical 1997-1998 Colfiorito and 2009 L’Aquila earthquake sequences. Despite extensive geodetic monitoring over the past years, precisely resolving the three-dimensional surface deformation field and interpreting the intricate subsurface source geometry in this topographically rugged region remains a significant challenge for both geological engineering and geophysics.

To accurately capture the coseismic displacement, we utilized multi-orbit Synthetic Aperture Radar (SAR) datasets acquired by the C-band Sentinel-1 and L-band ALOS-2 satellite missions. Traditional Differential Interferometric SAR (DInSAR) processing often struggles with severe decorrelation in high-gradient epicentral zones and relies heavily on subjective, empirical weighting when fusing heterogeneous multi-source data. To overcome these limitations, we implemented the advanced Strain-Model Variance Component Estimation (SM-VCE) framework. This technique incorporates a sophisticated spatial strain model to mathematically characterize the physical deformation correlations between adjacent ground pixels. Concurrently, it employs the VCE algorithm to iteratively and objectively determine the optimal variance components and contribution weights for each respective dataset based on their stochastic properties. This fusion strategy effectively mitigated atmospheric artifacts and successfully preserved critical near-fault deformation signals, yielding highly reliable two-dimensional (East-West and Vertical) coseismic displacement fields.

The SM-VCE derived surface deformation maps reveal a pronounced and highly asymmetric kinematic pattern. In the horizontal plane, the fault zone accommodated a net East-West extension of approximately 60 cm. The vertical displacement field is characterized by a massive subsidence trough localized in the hanging wall, reaching maximum downward displacements of 70 to 80 cm. This strongly contrasts with the minor uplift of only 10 to 14 cm observed in the footwall block. Based on these high-precision measurements, we conducted a rigorous 3D volumetric integration. The calculations exposed an extreme volumetric unbalance, demonstrating that the subsided rock volume is approximately 14 times larger than the uplifted volume. This severe mass deficit poses a direct challenge to standard elastic rebound paradigms and indicates complex crustal interactions.

To demystify the mechanical origins of this profound asymmetry and volume deficit, we applied the Defsour® (Free-geometry Multi-Source 3D Inversion) algorithm. Unlike conventional kinematic inversions that artificially constrain slip onto predefined, idealized planar faults, Defsour® adopts a purely data-driven, free-geometry strategy. It performs a global optimization across a dense 3D subsurface grid to simultaneously adjust arbitrary pressure and dislocation sources without a priori geometric assumptions. This autonomous inversion successfully reconstructed the primary slip distribution along the main southwest-dipping normal fault while independently identifying a distinct, east-northeast-dipping antithetic fractured zone. The incorporation of this antithetic structure significantly improved the consistency between the simulated and observed data, effectively resolving the misfit commonly encountered in single-fault models.

The 2023 Mw 7.8 and Mw 7.6 Kahramanmaraş earthquake doublet in southeastern Turkey generated exceptionally complex regional deformation patterns, providing a rare natural laboratory for investigating how large strike-slip earthquakes influence fault activity outside the main rupture zone. In this study, we employ time-series Interferometric Synthetic Aperture Radar (TS-InSAR) to quantitatively invert coseismic and postseismic surface deformation in the epicentral area and surrounding regions, with particular emphasis on independent fault structures that were previously widely regarded as inactive or only weakly active. Leveraging high-temporal-resolution InSAR time series, we not only capture the spatial continuity of deformation but also track its temporal evolution from months to years after the mainshocks, thereby distinguishing persistent postseismic deformation signals from short-lived responses and residual coseismic effects.

Our results show that, beyond the primary rupture zones, multiple independent faults distributed around the broader earthquake-affected region exhibit clear postseismic deformation. In these areas, deformation continues to accumulate after the mainshocks and deformation rates increase markedly, supporting an interpretation of earthquake-triggered deformation on these faults rather than delayed release or residual effects of coseismic displacement. The observed postseismic deformation is heterogeneous: some independent faults display rapidly decaying early postseismic transients, whereas others exhibit sustained, quasi-steady deformation that persists for months and in some cases extends to multi-year timescales.

To further investigate the physical mechanisms driving the earthquake-triggered deformation of these independent faults, we compare the deformation patterns inverted from InSAR with modeled static Coulomb stress changes, as well as with the spatial extent of dynamic stress perturbations induced by strong ground shaking. The results indicate that independently triggered faults show a pronounced spatial correspondence with regions of positive Coulomb stress increase and are also commonly located within areas experiencing strong dynamic shaking. This correspondence suggests that static stress transfer and dynamic shaking likely acted synergistically: static stress loading may have brought favorably oriented faults closer to failure, while dynamic perturbations may have accelerated nucleation processes or reduced effective strength, thereby facilitating triggering and prolonging postseismic deformation.

Overall, this study emphasizes that independent faults outside the main rupture zone can accommodate a non-negligible fraction of postseismic strain release and play an important role in regional deformation partitioning. This insight has direct implications for seismic hazard assessment, as conventional fault–earthquake frameworks centered on major seismogenic faults often underestimate the potential contribution of such independent faults. By integrating TS-InSAR time-series observations with fault modeling, our results highlight the need to explicitly consider distributed, earthquake-triggered deformation on faults outside the main rupture zone when investigating postseismic processes and evaluating regional seismic hazard in tectonically complex settings.

Understanding how crustal strain is accumulated and released between seismic and aseismic processes is essential for a better assessment of seismic hazard. Monitoring the surface deformation allows us to infer processes occurring in tectonically-active areas during the different phases of the seismic cycle. With the incipient arc-continent collision between the Philippine Sea Plate and the Eurasian Plate, southwestern Taiwan is becoming an important area for observing such fault movements. Southwestern Taiwan, characterized by a complex fold-and-thrust belt, exhibits high strain rates (> 1 strain/yr) despite low seismic activity in the shallow crust (< 10km). This apparent discrepancy raises key questions regarding the proportion of deformation accommodated elastically, with important implications for future large earthquakes. Due to the lack of shallow seismic activity in this area, most of the deformation appears to be aseismic. Geodetic observations represent the primary means of quantifying surface deformation over the past decades. To obtain high-resolution surface deformation in space and time, and to achieve three-dimensional decomposition, we plan to integrate complementary datasets from multiple satellite missions using InSAR and GNSS observations. We processed Sentinel-1 and ALOS2 dataset using a small baseline approach. the Sentinel-1 dataset (2014-2025) using the NSBAS processing chain, including the atmospheric and unwrapping error corrections to ensure signal integrity in Taiwan’s challenging subtropical environment. Furthermore, to overcome temporal decorrelation over the densely vegetated foothills, we used the ALOS-2 dataset covering from 2015 to 2025, processed with ISCE2 and Mintpy software to address some sources of uncertainty and time series inversion. Preliminary results demonstrate the complementarity of Sentinel-1 and ALOS-2 in terms of spatial coverage across the study area. While the Sentinel-1 dataset provides spatial detail over the coastal and Pingtung plains, revealing new fine details in the active tectonic structures, the most active structures exhibiting the largest deformation are located east of the coastal plain densely vegetated regions. In these challenging areas, the L-band ALOS-2 data proved to be critical to retrieve the deformation.

The detailed deformation field will facilitate more reliable observations for understanding the creeping behaviors, identifying the process causing deformation, and separating tectonic from non-tectonic signals (e.g., hydrological signals from significant seasonal groundwater fluctuations). These deformation products will help distinguish between elastic and non-elastic deformation components in further analysis, providing an opportunity to evaluate the contribution of aseismic processes to the observed deformation field.

Advances in space geodesy have revealed slow deformation transients and complex interactions between aseismic and seismic slip along major strike-slip faults. The Haiyuan fault (northeast China) provides a documented example of this dual behavior, combining sections that ruptured during Mw ~8 earthquakes with a persistent creeping section [1, 2]. We investigate this ~35 km-long creeping section located between the western end of the rupture of the 1920, Mw 7.9, Haiyuan earthquake and the eastern end of the so-called « Tianzhu seismic gap » revisiting both its spatial distribution and temporal evolution using ERS, Envisat, and Sentinel-1 InSAR data.

We analyze in more detail Sentinel-1 displacement time series processed with the FormaTerre FLATSIM service operated by CNES [3], spanning the period 2014–2020, for one ascending and two descending tracks. Linear and seasonal components are estimated pixel-by-pixel from the LOS time series. The linear term is then decomposed into horizontal, fault-parallel, and vertical velocity components using a multi-track inversion. Creep signature is visible in all LOS and decomposed horizontal velocity fields. Surface creep reaches values up to ~5 mm yr⁻¹ in the fault-parallel direction, with significant variations along fault strike. Localized subsidence of ~8 mm yr⁻¹ is observed within the Jingtai pull-apart, a major extensional relay zone between the fault section ruptured in 1920 and the seismic gap.

To assess the long-term persistence of the aseismic slip, we compare fault-perpendicular profiles from ERS (1993–1998), Envisat (2003–2009), and Sentinel-1 (2014–2020) fault-parallel velocity maps. These profiles show that the localization of the steep velocity gradient at the mapped fault trace is maintained over nearly three decades, despite differences in data temporal sampling, with a significant gain in signal to noise ratio with Sentinel-1.

We invert the InSAR velocity fields using the CSI software [4], jointly constrained by GNSS data, to model the distribution of shallow slip along the seismogenic zone.

An Independent Component Analysis reveals the slip partitioning within the various fault strands around the Jingtai pull-apart basin and suggest the existence of episodes of slip acceleration along the creeping section. To further investigate such temporal variability, we construct (from the Sentinel-1 dataset) cumulative fault-parallel and vertical displacement time series from combined LOS data on common spatial and temporal grids. Series of fault-perpendicular profiles along fault and through time are used to estimate slip rates between consecutive acquisitions and to build time–longitude representations of slip evolution. The results indicate that creep is not steady but intermittent, evolving through alternating phases of increased and reduced fault-parallel slip rate. Transient episodes are spatially confined to the creeping section and involve both horizontal creep accelerations and subsidence-rate increases within the relay zone.

These observations constrain the partitioning of deformation within the creeping step-over and document time-dependent variations in shallow slip, the underlying mechanical processes of which we discuss.

References:
[1] Gaudemer et al., 1995, doi.org/10.1111/j.1365-246X.1995.tb01842.x

[2] Jolivet et al., 2012, doi.org/10.1029/2011JB008732

[3] Thollard et al., 2021,and FLATSIM Data Products. CNES. (Dataset) doi.org/10.3390/rs13183734

[4] Jolivet et al., 2020, doi.org/10.1029/2019GL085377

Generating large-scale, high-accuracy velocity fields from Sentinel-1 Interferometric Synthetic Aperture Radar (InSAR) data remains a challenge due to long-wavelength orbital errors, atmospheric delays, and frame inconsistencies between adjacent tracks. In this work, we present a method for calibrating and combining Line-of-Sight (LOS) velocities across multiple processed LiCSAR products using a spatially weighted surface fitting approach integrated with 3D GNSS measurements. Our study area encompasses complex tectonic regimes in Türkiye, including the North Anatolian Fault Zone, East Anatolian Fault Zone, and Central Anatolian Block.

To model and correct for long-wavelength InSAR deviations, we utilize a Continuously Operating Reference Station (CORS) network (~200 stations) for model training and a much denser campaign-based GNSS network (~1000 stations) dedicated to independent validation. Before initiating calibration, we first project the ENU (East, North, Up) GNSS velocity components onto a specific InSAR line-of-sight (LOS) geometry. This projection is obtained by using the average incidence and heading angles derived from coherent InSAR pixels located within a 1 km radius of each GNSS station, providing a geometric match between two different datasets.

Next, we calculate the velocity difference between the GNSS-projected LOS and the raw InSAR LOS to detect the deviation at each station location. We propagate this bias surface across the frame using a Weighted Least Squares (WLS) approach based on GNSS and InSAR uncertainty measures. By incorporating the observation uncertainties of the GNSS stations into the algorithm, we tested interpolation models such as Quadratic Surface Fitting and Thin Plate Spline (TPS).

To merge different frames with overlapping areas, we first reference each InSAR frame separately to the same GNSS network. By doing this, the velocity differences present in the overlapping regions of adjacent tracks are reduced before the merging process. Then, a pixel-based point mapping is applied to these overlapping regions. Adjacent tracks are merged by calculating the average velocity differences of the overlapping pixels. In other words remaining discrepancies are minimized by using a weighted average approach, thus linking the frames together.

The performance of the tested interpolation algorithms was evaluated using an independent campaign-based GNSS network. Both standard Root Mean Square Error (RMSE) and Weighted RMSE (WRMSE) metrics were calculated for this evaluation. In conclusion, this study provides a basis for extending the calibration process to align geometries of multiple overlapping frames in both ascending and descending directions.

The 2019 Mw 6.4 Durrës earthquake struck Albania on 26 November, causing several fatalities and severe economic loss. This earthquake, which concluded a longer seismic sequence that began in September 2019, reminds us that Albania faces one of the highest seismic hazards in Europe, despite low interseismic strain accumulation, with rates of only 10-20 nstrain/yr measured along the Adriatic-European plate boundary (e.g. Jouanne et al., 2012; D’Agostino et al. 2020; Métois et al., 2025).

The earthquake occurred in the Peri-Adriatic foredeep of the Albanides belt, where compression related to subduction dominates. Recent geodetic studies (e.g. Piña Valdés et al., 2022; Meridi et al., in review) indicate that this compressional deformation spans from the coastline to the external-internal Albanides boundary, marking the transition to an extensional regime further east. This depression is characterized in the literature by a fold-and-thrust belt system involving crustal basement, active SW-verging reverse faults and conjugate back-thrusts, some of which are currently active (Teloni et al., 2021). To the west, deeper crustal SE-dipping faults, blind and oriented parallel to the inferred subduction zone, concentrate instrumental seismogenic activity at depth, accommodating the NNE motion of the Apulia microplate (4 mm/yr) relative to stable Eurasia.

Despite several seismological and geodetic studies, the fault geometry of the 2019 Mw6.4 earthquake remains debated, with two candidates for the fault plane : (i) a steep SW-dipping backthrust (71°) corresponding to the Vorë fault, or (ii) a low-angle NE-dipping (15°) subduction interface plane (Govorčin et al., 2021). Moreover, transient signals, including potential Slow Slip Events (SSEs), seem to have been observed several months after the main shock near the epicentral area, that remain to be confirmed (Matraku et al. 2024). Understanding the timing and distribution of slip during this complex sequence is of paramount importance to better assess seismic hazard in the most populated region of Albania.

In this study, we take advantage of several continuous datasets: (i) 2014-2021 InSAR time series (so up to 14 months after the earthquake) processed by the FLATSIM CNES/Formaterre service (Thollard et al., 2021) over the Balkans (240 m resolution, 14 ascending and descending tracks), (ii) permanent and campaign GNSS time series from Matraku et al. (2024) covering the pre and post-earthquake period, and (iii) catalog of relocated aftershocks starting 15 days after the mainshock from Schurr et al. (2026) to reconstruct the surface deformation and slip history associated with this earthquake.

The postprocessing conducted on the FLATSIM time-series allows for extracting a clean coseismic pattern, and for exploring the existence of afterslip or other postseismic processes. First, we use parametric decomposition to extract the coseismic displacement in the LOS that is then inverted jointly with GNSS, strong motion, and teleseismic data to constrain the fault geometry and slip distribution .

Our postseismic analysis combines three complementary approaches. We first map cumulative postseismic deformation (after removing the pre-seismic trend) to reveal its spatial footprint, then examine time series along profiles perpendicular to potential source faults to track temporal evolution. Preliminary observations suggest shallow slip on a secondary fault, possibly linked to shallow aftershocks on the Vorë fault. We therefore perform a moment budget analysis comparing aftershock moments with InSAR-derived slip, to discriminate between afterslip on the main fault or triggered slip on the Vorë fault.

In addition, we apply careful regional referencing of seasonal terms to a null-deformation polynomial surface. This allows us to analyse their spatial expression, with amplitudes reaching up to 20 mm/yr locally in the Tirana and Dürres basins, and to evaluate whether they might be misinterpreted as tectonic transients. Finally, we compare InSAR and GNSS time series at stations potentially impacted by SSEs, providing an independent test for the presence of transient deformation.

References :

Jouanne et al. (2012), 10.1016/j.tecto.2012.06.008

D’agostino et al. (2020), 10.1016/j.epsl.2020.116246

Govorčin et al. (2021), 10.1029/2020GL088990

Teloni et al. (2021), 10.1093/gji/ggaa582

Thollard et al. (2021), 10.3390/rs13183734 / Dataset FLATSIM, 10.24400/253171/FLATSIM2020

Piña-Valdés et al. (2022), 10.1029/2021JB023451

Matraku et al. (2024), 10.1093/gji/ggad101

Métois et al. (2025), 10.55575/tektonika2025.3.1.99

Schurr et al., (2026) 10.5880/GFZ.FPXO.2026.001

Meridi et al., (2026) (in review)

We communicate the development of a fully automated workflow, by the Aristotle University of Thessaloniki (AUTh) for the generation of SAR interferometric products from the Copernicus Sentinel-1 mission following strong earthquakes in Greece. Differential SAR Interferometry (DInSAR) has long been recognized as an effective technique for mapping ground deformation associated with earthquakes. In recent years, the systematic availability of data from the Copernicus programme has allowed several institutions to develop services for automatic response to co-seismic deformation.

We present a fully unsupervised, event-driven DInSAR processing chain triggered by recordings from the AUTh Seismological Station (https://seismo.auth.gr) for the rapid generation of co-seismic deformation products. The service is integrated into the AUTh HPC infrastructure, ensuring high performance and scalability while enabling the efficient execution of the automated processing workflow.

When an earthquake exceeding a predefined magnitude is detected, the workflow is automatically activated and registered users are notified of the expected dates of upcoming post-event satellite acquisitions. At this stage the system already identifies suitable pre-event Sentinel-1 SAR acquisitions covering the affected region. The area of interest is defined by a radius of several kilometers around the earthquake epicenter and considers all possible Sentinel-1 tracks intersecting this region. While remaining on hold for one earthquake, the system remains responsive to new earthquakes as independent triggers. Once post-event imagery is disseminated through dedicated portals, the download and ingestion process is initiated. Upon completion of processing, notification messages are sent including links for accessing the results. The outputs include wrapped differential interferograms, interferometric coherence levels, and displacement values, all geocoded to the GGRS87 (EPSG 2100) map projection.

Particular emphasis is placed on operational readiness and FAIR-by-design principles, ensuring that the resulting deformation products are not only timely, but also transparent, interoperable, and reusable. All generated products are distributed in widely adopted open formats accompanied by structured metadata and provenance documenting processing parameters is systematically recorded to ensure full reproducibility. The final datasets are intended to be published in FAIR-enabling repositories, ensuring long-term discoverability, accessibility, and reuse. The workflow is demonstrated through a representative earthquake case, illustrating its potential to support both scientific analysis and rapid-response applications. Future developments will focus on automated fault modelling as well as enhancing user interaction and accessibility through the integration of the generated products within an interactive platform for intuitive visualization and exploration.

Acknowledgements

The authors acknowledge the OSCARS project, which has received funding from the European Commission’s Horizon Europe Research and Innovation programme under grant agreement No. 10112975.

Abstract :

This study examines the seismotectonic characteristics of the 2023 Al Haouz earthquake (Mw 6.8) in the Western High Atlas, Morocco. The research combines seismic catalog analysis with Differential Interferometric Synthetic Aperture Radar (DInSAR) observations to investigate the deformation pattern and identify the associated seismogenic structures. Sentinel- 1 SAR data were processed using the SNAP software to generate ground displacement maps related to the earthquake. The results reveal clear surface deformation patterns consistent with the regional tectonic framework and provide new insights into the geometry of the fault responsible for the event. The analysis suggests that the Tizi n’Test fault was the main tectonic structure reactivated during this earthquake. These findings contribute to a better understanding of earthquake mechanisms in the High Atlas and demonstrate the value of integrating seismic data with satellite geodesy for seismic hazard assessment in mountainous regions.

Earthquakes are among the most destructive natural disasters. Approximately 1 million deaths have been caused by earthquakes between 2000 and 2020. It is still difficult to forecast earthquake events spatially and temporally. Post-event damage assessment aims to mitigate the human health risk of earthquakes. Rapid assessment of building damage is critical to understand the scale of civil protection response required, for example to allocate funds and set up emergency shelter. Moreover, rapid damage assessment is crucial for post-disaster rescue. 75% of the casualties in disasters are caused by building damage, and the maximum survival time of people trapped in collapsed buildings rarely exceeds 4-6 days.
Ground surveys provide reliable building-level damage severity assessment, but may take weeks or months to complete. Optical and Synthetic Aperture Radar (SAR) remote sensing imagery are commonly used for large-scale earthquake damage assessment on short timescales. Post-event SAR imagery is often available within one to two days resulting from recent satellite launches (e.g., ALOS-4, NISAR, Sentinel-1 and commercial constellations such as Capella Space and ICEYE). In contrast to the more commonly applied optical data, SAR can provide rapid disaster response data regardless of light or cloud conditions. However, SAR models often overestimate damage due to temporal decorrelation and are typically outperformed by optical models in comparative studies [1,2]. Furthermore, distinguishing low and moderate damage severity with remote sensing data is still challenging.
In our work, we present a Sentinel-1 SAR coherence timeseries model for rapid earthquake damage assessment and investigate multimodal fusion of SAR, shaking intensity and building taxonomy data. We apply our methods to the Central Italy 2016–2017 sequence (Mw>6) and Zagreb March 2020 earthquake (Mw 5.4) to test model effectiveness. We validate results against ground survey data and compare performance to manual assessment of high-resolution optical data by the Copernicus Emergency Management Service.
Our SAR change detection model shows similar agreement with ground survey data as the manual interpretation of optical satellite data. Both medium-resolution SAR and sub-meter resolution optical data are found to be most valuable for identifying clusters of collapsed buildings. In contrast, empirical fragility curves and feature importance analysis of a fusion model reveal that building construction period and construction type data are valuable for identifying low to moderate damage grades, complementing remote sensing data assessments. Our findings have implications for rapid damage assessment data acquisition strategies, fusion model implementation and model validation approaches.
[1] Stramondo, S., Bignami, C., Chini, M., Pierdicca, N., & Tertulliani, A. (2006). Satellite radar and optical remote sensing for earthquake damage detection: results from different case studies. International Journal of Remote Sensing, 27(20), 4433-4447.
[2] Sun, Y., Wang, Y., & Eineder, M. (2024). QuickQuakeBuildings: Post-earthquake SAR-Optical Dataset for Quick Damaged-building Detection. IEEE Geoscience and Remote Sensing Letters, 21, 1-5.

A M 6.0 earthquake occurred on 31 August 2025 in northeastern Afghanistan in the Pamir-Hindu Kush region. The Pamir-Hindu Kush is among one of the most seismically active regions of the western Himalayan syntaxis, formed and driven by the ongoing continental collision between the Indian and Eurasian plates.This complex continental collision has produced compression and crustal shortening, resulting in continental crust thickening, uplift, and the development of complex fault systems. Afghanistan has limited local seismic network data and field access is limited, meaning that remote sensing, especially InSAR, is the best method for studying earthquakes in this area. This moderate-sized earthquake had a significant impact – it resulted in ~2200 fatalities and economic losses of over $100 million. Estimates of the source time function from teleseismic data and also the InSAR deformation pattern as we describe below, indicate that this was a complex rupture with multiple subevents.

In this study, we use InSAR data from four Sentinel-1 tracks, processed using the ISCE software, to constrain and model the complex multi-segment rupture. The InSAR data shows two regions of significant displacement – a broad region with moderate decrease of range with a smaller, higher amplitude area of range decrease superimposed at the southern edge of the first (Figure 1). These are seen in data from both ascending and descending tracks, indicating that these are two regions of uplift, and our interpretation is that each of them could be associated with a reverse fault. To improve the coseismic signal and reduce unwrapping errors, we construct a preliminary two-fault model based on hand-digitized data to flatten the interferogram prior to unwrapping. By examining the connected components mask we identify isolated areas of the unwrapped interferogram that have phase jumps at their edges and then manually add or subtract multiples of 2pi to those areas until no phase jumps remain. Once corrected, we downsample each interferogram using a quadtree decomposition and model the fault geometry (e.g. fault length, fault width, slip, fault depth, strike, dip, rake) using rectangular dislocations in an elastic half space (Okada, 1985), determining the best-fitting parameters through nonlinear optimization (a Powell algorithm with multiple Monte Carlo restarts).

Our preliminary results suggest that the rupture involved near-simultaneous rupture of two faults – a primary ENE-striking south dipping structure in the north and a secondary WSW-striking north dipping structure, fringing the area of additional uplift located ~9 km to the southeast. Both faults exhibit inward-dipping oblique thrust faulting mechanisms with minor right-lateral strike-slip components, suggesting that the earthquake is accommodating regional tectonic compression. To explore model fit to the observed data, we model the event with one, two and three fault segments. The reference figure shows results from Sentinel-1 descending track 005, including the data, model, and residuals of each of these models. We use a statistical test, the Bayesian Information Criterion (BIC), that evaluates whether adding more complexity (i.e. more fault segments) to a model results in improvements in fit to the data that are better than expected given the larger number of free parameters. In our preliminary analysis, the two-fault model has the lowest BIC, indicating that it is the most robust, and our preferred model. The northern fault segment has a fault length of ~4.7 km and a downdip width of ~7.0, with a uniform slip of ~1.0 m, centered at a depth of ~ 5.5 km. The secondary fault segment has a fault length of ~3 km and a downdip width of ~3.5, with a uniform slip of ~0.8 m centered at a depth of ~2 km. These inward dipping structures do not connect at depth as the secondary fault is shallower by ~3.5 km, however it may be a back thrust. The moment magnitude estimated from the preferred model is 6.02, consistent with seismic estimates.

The Mw 8.8 Kamchatka megathrust earthquake of 29 July 2025 ruptured a structurally complex segment of the Kuril-Kruzenstern subduction zone and propagated over more than 300 km along strike. The event initiated within a rough portion of the plate interface associated with the subduction of the Kruzenstern fracture zone and subsequently expanded southward into a comparatively smoother and more strongly locked megathrust domain. The mainshock was preceded by an Mw 7.0 earthquake in August 2024 and an Mw 7.5 event nine days earlier, and followed by two major aftershocks of Mw 7.4 and Mw 7.8 on 13 and 18 September 2025, respectively. Notably, all Mw ≥ 7 earthquakes in this sequence nucleated within the fracture zone. This unusual spatiotemporal evolution challenges the prevailing view that structurally rough megathrust segments primarily act as rupture barriers and rarely host the initiation of giant earthquakes.

Understanding the generation environments of great earthquakes in global subduction zones is fundamental to earthquake physics and hazard assessment. Great megathrust events have traditionally been linked to relatively smooth plate interfaces, often overlain by thick trench sediments or lacking pronounced structural heterogeneity. Classic examples include the 2011 Mw 9.0 Tohoku-oki, the 1960 Mw 9.5 Chile, and the 2004 Mw 9.2 Sumatra earthquakes, all of which ruptured extensive and comparatively smooth megathrust segments. In contrast, rough fault domains, characterized by subducted seamounts, fracture zones, or strong bathymetric and density contrasts, are widely interpreted as segmentation boundaries that promote small to moderate earthquakes, host seismic swarms or slow-slip events, and impede the propagation of neighboring large ruptures. Geodetic and seismic studies from subduction systems including Sumatra, Peru, Chile, and northern Mexico have documented rupture termination or locking segmentation associated with fracture-zone subduction. Reports of giant earthquakes initiating within such rough segments are rare, rendering the 2025 Kamchatka event a compelling story.

The fracture zone in Kamchatka is marked by strong along-strike variations in isostatic anomaly and persistent shallow seismic swarms that have previously been proposed to inhibit large coseismic rupture. Nevertheless, the Mw 8.8 mainshock evolved into a giant event. Rapid source analyses indicate that rupture propagated predominantly southward beyond the fracture zone into a smoother segment characterized by thicker sediment cover and reduced large-scale roughness. This southern domain previously hosted the 1952 Mw 9.0 earthquake and appears to be locked to a higher degree, or over a broader area, than the northern segment. The 2024–2025 earthquake sequence therefore provides a unique opportunity to examine whether a structurally rough segment can act not only as a barrier, but also as a gateway that facilitates cascading rupture into an adjacent smooth and highly stress-accumulated patch.

Here we reconstruct the rupture processes of the 2024–2025 Kamchatka earthquake sequence by integrating comprehensive inland geodetic observations with teleseismic waveform modeling. Our joint inversions resolve a mosaic of discrete slip patches within the fracture zone that collectively bridged to a large, high-slip region in the smoother southern megathrust. The results demonstrate that rupture nucleation within a rough structural domain can dynamically connect to and trigger extensive failure on an adjacent smooth segment. These findings provide the first well-constrained evidence of a giant earthquake initiating on a rough megathrust patch and subsequently evolving into a large-scale cascading rupture. The observed gateway behavior calls for a systematic reassessment of structural controls on megathrust segmentation worldwide and offers new insights into earthquake nucleation, rupture dynamics, and integrated seismic and tsunami hazard assessment.

The Mw 6.5 Norcia earthquake, which struck Central Italy on October 30, 2016, represents the climactic and most destructive event of the recent Apennine seismic sequence. Nucleating within the complex, extensional Mt. Vettore-Bove Fault System (MVBFS), this mainshock ruptured a critical structural gap between the historical 1997-1998 Colfiorito and 2009 L’Aquila earthquake sequences. Despite extensive geodetic monitoring over the past years, precisely resolving the three-dimensional surface deformation field and interpreting the intricate subsurface source geometry in this topographically rugged region remains a significant challenge for both geological engineering and geophysics.

To accurately capture the coseismic displacement, we utilized multi-orbit Synthetic Aperture Radar (SAR) datasets acquired by the C-band Sentinel-1 and L-band ALOS-2 satellite missions. Traditional Differential Interferometric SAR (DInSAR) processing often struggles with severe decorrelation in high-gradient epicentral zones and relies heavily on subjective, empirical weighting when fusing heterogeneous multi-source data. To overcome these limitations, we implemented the advanced Strain-Model Variance Component Estimation (SM-VCE) framework. This technique incorporates a sophisticated spatial strain model to mathematically characterize the physical deformation correlations between adjacent ground pixels. Concurrently, it employs the VCE algorithm to iteratively and objectively determine the optimal variance components and contribution weights for each respective dataset based on their stochastic properties. This fusion strategy effectively mitigated atmospheric artifacts and successfully preserved critical near-fault deformation signals, yielding highly reliable two-dimensional (East-West and Vertical) coseismic displacement fields.

The SM-VCE derived surface deformation maps reveal a pronounced and highly asymmetric kinematic pattern. In the horizontal plane, the fault zone accommodated a net East-West extension of approximately 60 cm. The vertical displacement field is characterized by a massive subsidence trough localized in the hanging wall, reaching maximum downward displacements of 70 to 80 cm. This strongly contrasts with the minor uplift of only 10 to 14 cm observed in the footwall block. Based on these high-precision measurements, we conducted a rigorous 3D volumetric integration. The calculations exposed an extreme volumetric unbalance, demonstrating that the subsided rock volume is approximately 14 times larger than the uplifted volume. This severe mass deficit poses a direct challenge to standard elastic rebound paradigms and indicates complex crustal interactions.

To demystify the mechanical origins of this profound asymmetry and volume deficit, we applied the Defsour® (Free-geometry Multi-Source 3D Inversion) algorithm. Unlike conventional kinematic inversions that artificially constrain slip onto predefined, idealized planar faults, Defsour® adopts a purely data-driven, free-geometry strategy. It performs a global optimization across a dense 3D subsurface grid to simultaneously adjust arbitrary pressure and dislocation sources without a priori geometric assumptions. This autonomous inversion successfully reconstructed the primary slip distribution along the main southwest-dipping normal fault while independently identifying a distinct, east-northeast-dipping antithetic fractured zone. The incorporation of this antithetic structure significantly improved the consistency between the simulated and observed data, effectively resolving the misfit commonly encountered in single-fault models.

Spaceborne Interferometric Synthetic Aperture Radar (InSAR) is increasingly demonstrating transformative value beyond the research domain, enabling actionable, real-world solutions to pressing societal and environmental challenges. We present two complementary success stories that illustrate the breadth and maturity of InSAR technology: rapid disaster damage assessment following a catastrophic earthquake sequence, and satellite-based carbon emissions monitoring over tropical peatlands.

Success Story 1: Earthquake Damage Mapping after the 2023 Kahramanmaraş, Türkiye Earthquakes

Following a disaster, responders need to rapidly assess the extent of the damage. The prevailing view is that very-high-resolution optical satellite images can provide more accurate estimates of building damage than lower-resolution synthetic aperture radar (SAR). However, we demonstrate that SAR- based damage proxy maps we produced after the 2023 Kahramanmaraş Türkiye earthquakes outperformed comparable maps derived from optical imagery. This finding held true for both a state- of-the art machine learning method and for visual interpretation. The SAR-based maps achieved an F1 performance score approximately twice as high as the optical-based maps (0.47 vs 0.24, 0.23 and 0.15). Additionally, SAR offers established advantages in both coverage and timeliness: SAR was able to image the entire affected area within ten days, whereas the very-high-resolution optical dataset covered only 5.4% of the SAR-covered area during the same timeframe. Using the largest ground dataset that we know of for any event, we show that the damage distribution captured by our SAR- based maps strongly correlates with the ground observations over a wide range of spatial scales from neighbourhoods (~1 km2) to provinces (~10,000 km2). Based on this experience, we argue that any reliable remote sensing-based damage assessment system should incorporate radar to complement other techniques. (Ainscoe et al., 2025)

Success Story 2: Satellite Radar for Carbon Emissions Accountability over Tropical Peatlands

Carbon markets face growing criticism over unreliable measurements of carbon credits. Tropical peatlands, which contain some of Earth’s most concentrated carbon, represent a huge, untapped opportunity for emission reductions, but remain excluded from the market due to challenges in measuring emissions due to degradation. Here, we demonstrate satellite L-band Interferometric Synthetic Aperture Radar as a solution to estimate carbon dioxide emissions by accurately measuring peat subsidence. Our framework accounts for major radar noise sources in tropical environments that were previously unaddressed, and is validated against high-rate ground-measured peat motion in both space and time. The radar results capture episodic peat motion linked to dry-wet cycles across different land uses, and long-term rates accurate up to 0.6 mm yr−1, equivalent to 0.97 t CO2 ha−1 yr−1 in emissions. This scalable, cost-effective approach provides a robust tool for Monitoring, Reporting, and Verification, benefitting carbon markets, local regulation, and global climate mitigation efforts. (Tay et al., 2025)

Outlook

Together, these two applications – disaster response and climate accountability – demonstrate the maturity of spaceborne InSAR as a tool for societally critical monitoring at scale. Both studies highlight the unique advantages of SAR: all-weather, day-and-night imaging, broad spatial coverage, and high sensitivity to subtle surface changes. With the combination of L-band and C-band global coverage (e.g., NISAR, Sentinel-1, ALOS-2/4), the temporal and spatial sampling capacity for both applications will improve substantially, reinforcing the role of InSAR as an indispensable component of global Earth observation infrastructure.

References

Ainscoe, E.A., Swaminathan, R., Way, L., Modugno, S., Chin, S., Panta, N., Crevoisier, T., Yun, S., “Earthquake damage mapped more comprehensively and accurately by radar satellites”, Communications Earth & Environment, DOI: 10.1038/s43247-025-02623-4, 2025.

Tay, C., Jovani-Sancho, A. J., Yulianti, L., Evans, C., Callaghan, N., Jaya, A., Salman, R., Zheng. Y., Susilo, S., Dohong, S., Yun, S. H., “Satellite radar advances carbon emissions accountability over tropical peat”, Communications Earth & Environment, 10.1038/s43247-025-02926-6, 2025.

Dike intrusions accommodate most of the plate extension in magmatic rifts. However, as the crust extends by diking, normal faulting also occurs, forming morphologically clear graben. Here, we analyze Sentinel-1 DInSAR coseismic products available in the EPOSAR earthquake catalog [1] to investigate deformation patterns associated with the seismic crisis and the dike intrusion that occurred between September and November 2024, within the Fentale–Dofen segment of the Main Ethiopian Rift.

We compute spatial gradients of interferograms and LOS displacement maps using a geodetically consistent approach that accounts for latitude-dependent pixel spacing, enabling quantitative analysis of deformation gradients [2–3]. The resulting gradient field and its magnitude reveal sharp spatial variations along the edges of the deformation field, delineating the fault planes that slipped more clearly than displacement data alone. Application of this method allows the identification of discrete fault segments that progressively develop and lengthen during dike propagation. We plan to derive the vertical and horizontal displacements along the identified faults and compare the patterns to the dyke emplacement and the seismic moment release.

This study demonstrates that quantitative analysis of LOS displacement gradients provides a robust framework for identifying deformation boundaries, improving interpretation of geodetic observations in magmatic rift settings.

[1] EPOS, European Plate Observing System, [Online]. Available at https://www.epos-ip.org /tcs/satellite-data.

[2] Hofmann, B., Magee, C., & Wright, T. J. (2025). Throw distribution across the Dabbahu–Manda Hararo dike-induced fault array: Implications for rifting and faulting. Geology, 53(2), 161–165. https://doi.org/10.1130/G52665.1

[3] Argo Galih, S., Atriyon, J. InSAR-derived surface displacement gradients unveil subseismic faults of the 2022 Cianjur earthquake. Model. Earth Syst. Environ. 11, 338 (2025). https://doi.org/10.1007/s40808-025-02502-z

Four Mw 6.5 earthquakes struck northwest of Herat, Afghanistan, in a complex sequence of thrust faulting between 7 and 15 October 2023. This event raises important questions about the conditions that favor single-fault earthquakes, complex multi-fault ruptures, as well as tectonic stress conditions and structural factors that controlled the progression of this sequence. We investigated these events using Sentinel-1 and ALOS-2 InSAR, burst overlap interferometry (BOI) and pixel offset line-of-sight (LOS) and pixel offset along-track surface deformation data. Our slip distribution models indicate that two events on 7 October were generated by two WNW–ESE-trending, NNE-dipping blind thrust faults that are overlap by ~1 km. Their geometric overlap produced a static stress shadow on the more westerly segment, potentially contributing to the delayed triggering.

In contrast, the 11 and 15 October events were generated by NE–SW-trending, NW-dipping blind thrust faults that overlap by ~9 km. The overlap between these faults produced a static stress shadow along the edges of the 15 October fault, which may also have contributed to its delayed triggering. In addition, the intersection between the more easterly 7 October fault and the 11 October fault, which differ in strike of ~43^o, generated a static stress shadow that may have contributed to the delayed triggering of the 11 October event.

We also modeled an E–W–trending, north-dipping thrust fault that slipped aseismically from shallow depths up to the surface following the 11 October event. Postseismic slip persisted for at least one year after the earthquake. Static Coulomb stress change calculations suggest that the 7–11 October events may have promoted slip on this fault.

Our quasi-dynamic 3D numerical simulations using aging-type rate-and-state friction laws suggest that delayed triggering occurs more frequently between faults with orientation similar to more easterly 7 October and 11 October faults, which differ in strike of ~43^o, than between faults whose interaction is primarily controlled by geometric overlap. These results highlight an important implication for earthquake sequence studies: each faulting event in the sequence may contribute to the delayed triggering of subsequent events, and changes in strike angle may be more susceptible to multiplet formation than faults interacting mainly through geometric overlap.

The rapid growth of high-resolution commercial SAR constellations has created new opportunities for stereo radargrammetry beyond the traditional constraints of interferometric processing. We propose a robust pyramidal stereo radargrammetric framework for accurate Digital Surface Model (DSM) generation from multi-acquisition SAR datasets characterized by significant diversity in range and azimuth viewing geometries.
The method is designed for challenging environments where interferometric coherence is degraded and optical stereo approaches are limited by illumination, weather, or surface conditions. Rather than performing computationally expensive all-versus-all matching across highly heterogeneous acquisitions, the proposed strategy relies on pairwise close-angle matching. Only geometrically compatible image pairs are matched, improving triangulation stability and reducing ambiguity while still exploiting the angular diversity provided by modern acquisition campaigns.
A coarse-to-fine pyramidal estimation scheme ensures computational efficiency and robustness. At each pyramid level, regularization constraints are introduced to suppress outliers, stabilize disparity estimation, and prevent error propagation across scales. The framework naturally supports cross-sensor integration among SAR platforms with comparable spatial resolution. Geometry-aware weighting and radiometric normalization allow heterogeneous commercial datasets to be jointly processed within a unified estimation model.
To address layover and geometric distortions inherent to SAR imaging, two complementary operating modes are considered: (i) distributed-target dense reconstruction for DSM generation, and (ii) a bright-target configuration focusing on isolated high-reflectivity scatterers. The latter enables highly accurate 3D point cloud extraction suitable for ground control point generation in a Fan-SAR configuration, where relaxed angular separation constraints can be applied due to improved localization precision.
The proposed approach maximizes the geometric information content of multi-angle commercial SAR acquisitions, providing a scalable and robust solution for precise 3D reconstruction in scenarios where conventional interferometric and optical methods are unreliable.

In this study we evaluate the potential of the SARvey open-source InSAR time series analysis software for performing InSAR time-series analysis on high-resolution SAR data acquired by the commercial Capella Space constellation. The objective is to assess the capability of SARvey to process Capella data and to investigate deformation signals over a rapidly subsiding urban environment using high-resolution commercial data.

We analyze a dataset composed of 18 single look complex (SLC) images acquired by Capella over Mexico City between 26 June 2024 and 15 August 2024, corresponding to a temporal span of approximately 1.5 months. The data has a high spatial resolution with approximately 60 by 100 cm spacing in range and azimuth, respectively. The study area is well known for significant land subsidence driven exceeding 30 cm/yr at locations primarily due to groundwater extraction. This making it an ideal test site for evaluating InSAR processing strategies.

We perform the preprocessing of the data using the GAMMA Software. All acquisitions are coregistered to a common reference scene and resampled to generate a stack of SLC images. The Copernicus digital elevation model is used to remove the topographic phase contribution and for geocoding.

The resulting coregistered stack is subsequently processed with SARvey for interferometric time-series analysis. Different processing configurations are tested to evaluate the robustness of the workflow for Capella data. These include multi-looked and full-resolution processing as well as different interferometric network strategies, specifically a star network and a small-baseline network. Furthermore, multiple phase unwrapping approaches are investigated in order to assess their influence on the stability and quality of the deformation estimates.

Initial results suggest a deformation signal of up to approximately 4 cm in the radar line-of-sight direction over the observation period. The spatial pattern of the detected deformation is consistent with the well-documented subsidence processes affecting Mexico City. Despite the relatively short observation window, the time-series analysis demonstrates that Capella SAR data can capture ground displacement at high spatial resolution.

These results indicate on one hand that SARvey provides a viable framework for processing commercial high-resolution SAR data and extracting meaningful deformation signals. On the other hand, the study highlights the potential of Capella observations for monitoring rapid urban subsidence when combined with flexible InSAR time-series processing approaches.

Spaceborne Multi-Temporal InSAR (MT-InSAR) has evolved into a well-established operational tool for mapping ground deformation, although its performance remains strongly sensitive to sensor resolution, radar wavelength, acquisition geometry, and land cover characteristics [1,2]. C-band ESA Sentinel-1 offers near-global coverage, but its effectiveness is limited by temporal decorrelation in densely vegetated areas. This limitation becomes even more pronounced at higher frequencies, such as those used by the X-band Italian Space Agency (ASI) COSMO-SkyMed constellation, which, despite providing considerably finer spatial resolution, is particularly susceptible to decorrelation over vegetated terrain. In contrast, L-band sensors — including SIASGE SAOCOM-1, JAXA ALOS-2, the recently launched NASA/ISRO NISAR and the upcoming ESA ROSE-L missions — operate at longer wavelengths (λ ≈ 23 cm) that are inherently less affected by volumetric decorrelation. This allows them to penetrate vegetation canopies more effectively, preserving coherence over extended spatial and temporal baselines in both vegetated and non-urbanised environments. Although L-band MT-InSAR has been successfully demonstrated using ALOS data [e.g., 3], the development and adaptation of algorithms specifically tailored to the operational exploitation of SAOCOM-1 for MT-InSAR applications remains relatively unexplored [4].

In the framework of the SIASGE Earth Observation programme, the Argentine SAOCOM‑1 L‑band constellation is currently being tasked by ASI to collect systematic Stripmap acquisitions over Italy, enabling country‑wide MT-InSAR analyses. We present the adaptation of the existing MT-InSAR algorithm implemented in the SPINUA chain [5] to process interferometric stacks of SAOCOM-1 SAR data, to the aim of providing insights into the capabilities that L-band MT-InSAR can enable for natural and anthropogenic hazards.

We assess MTInSAR performance on an InSAR dataset obtained from the processing of an interferometric stack of 38 SAOCOM‑1 images (2020‑2024) covering a test site in the Gargano Promontory (Puglia, Italy), within the ASI-funded GEORES project framework [6]. We benchmark the results against C‑band Sentinel‑1 and X‑band COSMO‑SkyMed/COSMO-SkyMed Second Generation (CSK/CSG) stacks. We report on persistent scatterer (PS) spatial densities with respect to land cover and geomorphic parameters, and provide examples illustrating geocoding accuracy and displacement time‑series consistency, highlighting the complementary value of L‑band observations for multi‑hazard ground motion monitoring in southern Italy.

The test case confirms theoretical expectations that longer wavelengths maintain coherence in vegetated environments, and may mitigate ambiguity problems in displacement estimation. The gain in PS density directly translates into improved sampling of non‑urban hazards (coastal cliff retreat, karst sinkholes). While CSK‑CSG sensors still provides superior PS densities due mainly to their higher spatial resolution, L-band consistently outperforms C-band data over most terrain classes. Strategically, then, the addition of a regular plan with SAOCOM‑1 strengthens the multi-frequency SAR data sources over the Italian territory and thus enables the retrieval of surface deformation estimates in different land cover/land use conditions. Remaining challenges with SAOCOM-1 data include radio frequency interference affecting some frames and the coarser azimuth sampling impacting small‑scale feature detection.

Acknowledgements

Research performed in the framework of the GEORES project - Agreement ASI – UNIBA n. 2023-42-HH.0 – CUP F93C23000240005, funded by the Italian Space Agency (ASI) in the framework of the “Innovation for Downstream Preparation for Science” (I4DP_SCIENCE) programme.

References

[1] Zebker, H.A., Villasenor, J. 1992. Decorrelation in interferometric radar echoes. IEEE Trans. Geosci. Remote Sens., 30(5), 950–959.

[2] Sabater, J.R., Duro, J., Arnaud, A., Albiol, D., Koudogbo, F.N., 2011. Comparative analyses of multifrequency PSI ground deformation measurements, in: ESA FRINGE Conference on ERS SAR Interferometry, ESA SP-697. p. 81790M. https://doi.org/10.1117/12.898916

[3] Wegmüller, U., Magnard, C., Strozzi, T., Caduff, R., & Jones, N. 2024. Landslide velocity mapping using ALOS-2 PALSAR-2 ScanSAR data. Procedia Computer Science, 239, 2278–2285. https://doi.org/10.1016/J.PROCS.2024.06.419.

[4] De Luca, C. et al., 2025. 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, vol. 18, pp. 2680-2703, 2025, https://doi.org/ 10.1109/JSTARS.2024.3507554.

[5] Bovenga, F., Nutricato, R., Refice, A., Wasowski, J. 2004. SPINUA: a flexible processing chain for ERS/ENVISAT long‑term interferometry. Proc. ESA‑ENVISAT Symposium, Salzburg, Austria, 6‑10 Sep 2004.

[6] Lafortezza, R., Nutricato, R., Refice, A., Capolongo, D., Sacco, P., Tapete, D. 2024. The GEORES Project: Geospatial Application in Support of Environmental Sustainability and Resilience to Climate Changes in Urban Areas. IGARSS 2024, Athens, Greece, pp. 1384–1387. https://doi.org/10.1109/IGARSS53475.2024.10642728

Ground-based interferometric radar systems such as the GAMMA Portable Radar Interferometer (GPRI-II) are commonly used for slope and infrastructure monitoring. However, their stationary configuration leads to range-dependent degradation of cross-range resolution due to the limited real aperture length. In contrast, satellite-based synthetic aperture radar (SAR) provides wide-area deformation monitoring but remains constrained by revisit intervals and fixed acquisition geometry, limiting its flexibility for localized, rapid-response measurements. To address these limitations, we explore a car-borne Ku-band SAR configuration aimed at flexible, high-resolution terrestrial deformation monitoring under variable acquisition geometries.

By synthesizing the aperture along the vehicle trajectory, the system achieves a theoretical azimuth resolution of approximately 0.12 m, largely independent of range, representing a substantial improvement over conventional rotating GPRI systems, whose cross-range resolution deteriorates with increasing distance. A roof-mounted Ku-band (≈17.25 GHz) frequency-modulated continuous-wave (FMCW) radar, integrated with INS/GNSS units, was deployed to acquire data along arbitrary paths and curved road segments. Additionally, vibration-isolating rubber mounts were installed between the vehicle and the radar mounting frame to reduce vehicle-induced vibration and enhance phase stability during motion. Compact patch-type antennas were adopted to ensure a lightweight, low-profile design suitable for mobile operation while maintaining sufficient radiometric performance. Motion parameters derived from post-processed navigation data were incorporated into a Time-Domain Back-Projection (TDBP) scheme, enabling precise slant-range calculation for each pulse. Geocoded multi-look intensity images were generated directly on a DEM grid, facilitating geometrically consistent map-projected products without requiring intermediate reprojection from slant-range coordinates.

High-resolution imaging with extended along-track coverage was successfully achieved, enabling continuous mapping beyond the spatial footprint of stationary installations. Repeat-pass interferometric processing produced stable phase measurements, although coherence was reduced under the present experimental conditions, likely due to the short Ku-band wavelength and trajectory irregularities. Despite these limitations, the results suggest that car-borne Ku-band SAR offers considerable promise for deformation monitoring in roadside and suburban settings. Ongoing work focuses on characterizing variability in coherence under different motion and acquisition geometries to improve interferometric robustness. With further refinement, the approach may provide a practical alternative to fixed ground-based installations and support mobile radar interferometry across a broader range of applications.

Keywords : Car-Borne, SAR, motion compensation, TDBP

The grounding line – the interface where a floating ice shelf detaches from its bedrock – is a critical marker of ice sheet dynamics and a key indicator of climate‑induced sea‑level rise. It has been classified as an essential climate variable (ECV) within ESA's Antarctic Climate Change Initiative program (AIS-cci). Grounding Line Location (GLL) products derived from differential interferometric synthetic aperture radar (DInSAR) have enabled continuous monitoring of this feature for major Antarctic ice streams and Ice Shelves since the ERS era through to the current Sentinel‑1 mission.

Two temporal processes affect the location of the grounding line. Changes in ocean tide move the grounding line back and forth on sub‑hour to monthly timescales, and long‑term, climatologically driven migration (typically retreat) is driven by ice thinning. Distinguishing these regimes is essential for accurate interpretation of ice sheet stability, yet current GLL products often delineate only single lines making it impossible to separate short‑term variations from long‑term trends.

In an AIS‑cci optional activity a Grounding Line Migration (GLM) product will be defined, which explicitly separates the short‑term temporal variability within the grounding zone from the long‑term migration that relocates this zone.

A core challenge is to define a distance metric between two spatially complex GLLs that captures both local changes and global shift of the grounding zone. Three candidate metrics will be evaluated for a DInSAR grounding line time-series over the Getz Ice Shelf. The approaches include the area‑averaged displacement calculated by the Box‑Method (Moon & Joughin, 2009), individual line-to-line distances by the Point‑to‑Line (PoLiS) method (Avbelj et al., 2014) and a retreat along glacier centerlines/flowlines. We will address the methods’ abilities to capture representative retreat rates for an entire ice shelf vs. localized regions prone to grounding line retreat and discuss possible representations of the derived retreat rates in a future AIS-cci GLM product.

Moon, T., & Joughin, I. R. (2008). Changes in ice front position on Greenland’s outlet glaciers from 1992 to 2007. Journal of Geophysical Research: Earth Surface, 113(F2). https://doi.org/10.1029/2007JF000927

Avbelj, J., Müller, R., & Bamler, R. (2015). A Metric for Polygon Comparison and Building Extraction Evaluation. IEEE Geoscience and Remote Sensing Letters, 12(1), 170–174. IEEE Geoscience and Remote Sensing Letters. https://doi.org/10.1109/LGRS.2014.2330695

The grounding line location (GLL) is a geophysical product of the Antarctic Ice Sheet Climate Change Initiative (AIS_cci) ESA project. It has been derived for major ice streams and glaciers around the continent through the InSAR technique, covering the period 1994 – 2025 from ERS-1/2 era to Sentinel-1. The AIS_cci GLL product’s metadata annotations include information about model-based ocean tide levels and air pressure at satellite acquisition times for meaningful and interpretable comparison of GLLs.

The position of the grounding line shifts in two distinct ways: (a) over short timescales, it experiences temporary migration caused by vertical uplift of ice shelves due to ocean tides, and (b) over longer timescales, it undergoes more consistent migration, typically retreating landward as a result of ice thinning.

Recent grounding line products have acknowledged the short-term variation of the grounding line position, annotating a grounding zone instead of single grounding lines (Rignot et al., 2023). In line with the need to quantify the grounding line movements, we designed an additional parameter for the AIS_cci GLL, the Grounding Line Migration (GLM). The AIS_cci GLM product aims to provide short-term temporal variations within the grounding zone and separate this short-term position change from a long-term climatic-induced relocation of the grounding zone.

Time stamped grounding lines appear as fragmented segments with various lengths depending on the data coverage and SAR interferometric coherence preservation. We use the time-annotated AIS_cci GLLs to derive an average grounding line for a certain period (e.g. one year) which will be further used to calculate migration. Our custom procedure fills the gaps with grounding lines from manual and machine-learning delineations (Ramanath et al., 2025) of temporally close Sentinel-1 DInSAR interferograms and external datasets. For GLM generation various metrics have been investigated to quantify the distance between two or more GLLs acquired in different periods. Here we plan to show GLM products over relevant sites prone to grounding line retreat with a dense time series of GLLs and outline the specifications and contents of the new GLM product.

References:

Rignot, E., Mouginot, J. & Scheuchl, B. (2023). MEaSUREs Grounding Zone of the Antarctic Ice Sheet. (NSIDC-0778, Version 1). Boulder, Colorado USA. NASA National Snow and Ice Data Center Distributed Active Archive Center. https://doi.org/10.5067/HGLT8XB480E4. Date Accessed 03-02-2026.

Ramanath, S., Krieger, L., Floricioiu, D., Diaconu, C.-A., and Heidler, K.: Automatic grounding line delineation of DInSAR interferograms using deep learning, The Cryosphere, 19, 2431–2455, https://doi.org/10.5194/tc-19-2431-2025, 2025.

The lack of regular, spatially detailed observations of snow mass (snow water equivalent, SWE) is a main gap in cryosphere monitoring. Spaceborne SAR systems offer various options for resolving this deficiency, as these sensors provide high spatial resolution and the signals penetrate snow as long as the snowpack is dry. A promising method for mapping SWE is differential radar interferometry (DInSAR), measuring the path delay of a radar signal propagating through a snow layer. C-band and L-band radar frequencies are well suitable for this approach, as the observed phase delay is only marginally affected by snow microstructure and density of dry snow. Critical issues for DInSAR SWE retrievals are the temporal decorrelation of the interferometric signal and the 2p phase ambiguity. We studied methods and performance for spatially distributed SWE retrievals in mountainous terrain from C-band and L-band interferometric data of airborne and spaceborne C-band and L-band SAR sensors.

The airborne campaign was conducted over a test site in high-Alpine terrain, where multiple polarimetric and interferometric repeat pass data in C- and L-band were acquired with the F-SAR system of DLR, spanning snowfall events of different intensity. Coherence, interferometric phase and computed snow accumulation (delta SWE) images were computed for the snowfall events and were evaluated in connection with comprehensive field measurements on snow properties. The two frequencies exhibit differences in the susceptibility to phase ambiguities, in the exposure to temporal decorrelation caused by snowfall and in the phase sensitivity in respect to SWE. L-band has lower phase sensitivity than C-band which is, at least partly, compensated by higher coherence. C-band is affected by lower coherence and 2p phase ambiguities but has higher phase sensitivity. These characteristics confirm the complementarity of interferometric data of the two frequencies for setting up a reliable and robust SWE monitoring system.

In order to further the synergistic use of C-and L-band data for SWE monitoring, we analysed interferometric Sentinel-1 C-band 6-day and ALOS PALSAR L-band 14-day repeat data over Alpine test sites. The interferometric processing workflow includes coherence estimation, removal of the atmospheric phase screen (e.g., using ETAD products), and topographic phase flattening based on the Copernicus DEM with 10 m spatial resolution. As reference points, targets with known changes in snow accumulation are used, such as in situ meteorological stations or dedicates snow stations located within contiguous areas.

Six-day repeat-pass Sentinel-1 C-band SAR data show sufficient coherence for detecting moderate snow mass accumulation in open terrain. In forested area, the C-band signal tends to decorrelate. These regions are therefore masked out in the analysis. Short-repeat cycles of C-band data with sub-daily repeat would be provided by the proposed Earth Explorer 12 Candidate Mission Hydroterra+, ameliorating the decorrelation problem at least in open terrain. The L-band phase signal is less affected by temporal changes in surface properties. Even in forested areas, phase coherence is largely preserved, providing spatial continuity between open and forested areas. However, correction for the attenuation of the surface signal within the forest canopy is required. With shorter revisit intervals, such as the six-day repeat cycle expected from the upcoming ROSE-L mission, further improvements for SWE retrievals are to be expected.

We present results from ongoing studies on methods and performance of SWE retrievals using Sentinel-1 C-band and ALOS PALSAR L-band data over Alpine sites, as well as a brief overview on the airborne campaign . Finally, open challenges and remaining steps toward an operational SWE monitoring service are discussed.

Mountain hazards such as avalanches, extreme snowfall, and landslides threaten people and infrastructure in alpine regions worldwide throughout the year. These hazards can damage lifelines, houses, agricultural areas, and limit sustainable community development. Due to the remoteness of many mountain locations and the large spatial extent of hazardous processes, ground-based monitoring and early warning systems are often difficult to implement for practical and economic reasons. Moreover, mountain hazards typically affect wide areas, requiring spatially distributed monitoring capabilities.

To address these challenges, we present the activities carried out within the ESA-funded project SUMMIT (Satellites-based Utility for Monitoring Mountains Integrated Transitions). The project integrates multiple data sources, including Sentinel-1 SAR, Sentinel-2 optical imagery, Galileo GNSS technology, and Automatic Weather Stations (AWSs) records, to enable near real-time monitoring and the historical characterization of multiple and cascading mountain hazards.

The system is demonstrated in Livigno, an alpine village located in the central Italian Alps. This area is characterized by intense winter and summer tourism and is therefore particularly exposed to natural hazard impacts.

Beginning in autumn, snow accumulation over mountain slopes plays a critical role in avalanche forecasting and hydrological management. To address this, SUMMIT integrates automatic detection of avalanche deposits, snowpack depth mapping, and wet snow extent monitoring during the melt season. Furthermore, climate change is significantly affecting ground deformation in periglacial environments during the summer season, which requires timely and continuous monitoring to detect early signs of instability and support risk mitigation strategies.

Snow avalanches represent one of the primary winter hazards in Europe, causing more than 100 fatalities annually and damage to infrastructures worldwide. Avalanche detection is achieved through a novel deep learning algorithm capable of automatically identifying avalanche deposits from both wet and dry snow events using Sentinel-1 SAR backscatter imagery in a change detection framework. The system is demonstrated for both near real-time monitoring and historical reconstruction of spatial and temporal avalanche activity distribution. The system’s performance is evaluated against observed avalanche records available for the study area, including an assessment of the main limitations and external interferences affecting SAR-based avalanche detection.

Snowpack depth mapping is enabled by a novel mathematical model based on a dual-polarimetric SAR index variation, achieving a spatial resolution of 50 m with an RMSE of 22.4 cm and MAE of 18.1 cm. The model accounts for the influence of the Local Incidence Angle (LIA) on SAR signal depolarization, a factor shown to significantly affect retrieval accuracy.

During the melting season, we exploit Sentinel-1 backscatter changes to map and quantify the spatial and temporal occurrence of wet snow. This, combined with the snow depth monitoring, provides key inputs for downstream flood susceptibility assessment and drought risk evaluation, as well as additional information for avalanche forecasting activities.

During summer, ground deformation in periglacial environments is monitored through SBAS InSAR, which enables both near real-time tracking and multi-year reconstruction of surface displacements associated with rock glacier dynamics and deep-seated landslides.

These SAR-based components, complemented by cost-effective Galileo-enabled GNSS receivers, AWS records, and the SNOWPACK model, enable a multi-scale synergetic integration across sensors with different spatial resolutions and acquisition frequencies.

Using the 2023/2024 and 2024/2025 seasons as case studies, characterized by exceptionally thick winter snowpacks, high liquid water content, significant avalanche cycles, and a subsequent rock glacier collapse, we demonstrate how these monitoring approaches can be combined into a unified seasonal hazard framework. This integration reveals cascading hazard dynamics that would remain undetected by any single monitoring approach, offering a replicable model for SAR-based risk management in remote alpine communities.

Glacial Lake Outburst Floods (GLOFs) are emerging as one of the most destructive high-mountain hazards in the Himalaya, where rapid glacier retreat and the proliferation of moraine-dammed lakes have intensified disaster risk. Sikkim, which hosts the highest concentration of glaciers in the Eastern Himalayas, has witnessed a rapid increase in glacial lake formation and recent catastrophic events, including the 2023 South Lhonak GLOF. Identifying lakes most susceptible to failure and detecting precursory instability remain challenging due to rugged terrain, persistent cloud cover, and limited field accessibility. This study develops an integrated multi-sensor remote-sensing framework to assess GLOF hazard in Sikkim using Synthetic Aperture Radar (SAR), Interferometric SAR (InSAR) time series, optical satellite imagery, digital elevation models (DEMs), and climatic indicators. PS- and SBAS-InSAR analyses from Sentinel-1 and TerraSAR-X datasets map millimeter-scale ground deformation, coherence loss, and slope creep around moraine dams, revealing instability zones and precursory deformation patterns. High-resolution optical time series (PlanetScope, Sentinel-2) capture seasonal to multi-year lake expansion, while DEM differencing between SRTM (2000) and Copernicus GLO-30 (2020) quantifies elevation change in dam crests and proglacial basins. Climatic parameters including extreme rainfall and melt-season anomalies are integrated to interpret hydrological loading and lake-level fluctuations. By synthesizing deformation rates, morphometric instability, lake-area evolution, and meteorological drivers, the study classifies lakes into risk tiers and identifies those most prone to failure. Results demonstrate that combining SAR-derived deformation precursors with optical and DEM-based lake evolution significantly enhances forecasting capability and forms a robust pathway for early warning. The proposed framework enables a shift from post-event assessment to proactive monitoring, strengthening disaster-risk reduction efforts across the vulnerable Himalayan basins.

Vertical motion from ocean tides causes a significant horizontal displacement in the range-direction of offset-tracked ice speed measurements on ice shelves from Synthetic Aperture Radar (SAR) data. This geophysical signal can be usefully used to differentiate between grounded and floating ice, but it is not helpful for studies investigating ice-velocity change and interpretation of ice-dynamical processes. This effect is particularly important in coastal Antarctica, where large tidal amplitudes of up to 9 meters in the Ronne Filchner Ice Shelf, and floating ice around 74 % if the ice sheet margin led to strong tidal signals in SAR ice speed measurements. There is a need for robust algorithms to be developed to tidally correct ice speed measurements so the signal can be removed for applications that do not require it. However, converting tidal height variations into radar range displacement remains a major challenge because the tidal response of ice shelves depends on ice thickness, grounding-line proximity, flexural behaviour and ice rheology, all of which vary spatially.

In this study we use an ocean tide model, inverse barometer corrections from ERA5 data and information on SAR imaging geometry to calculate a tide correction that can be applied to Sentinel-1 SAR offset-tracking measurements of ice speed. We then investigate two alternative approaches to model the tidal correction in the complex grounding zone region. The first method follows the ESA Antarctic Ice Sheet Climate Change Initiative (CCI) approach and applies an elastic beam model to estimate the relationship between tidal vertical displacement and radar range motion using ice thickness and distance from the grounding line. This approach provides a physics-based description of tidal flexure but relies on assumptions about ice thickness and buoyancy in the vicinity of the grounding zone which are poorly constrained. The second approach uses the output from the tidal motion offset correlation (TMOC) method of identifying tidal displacement and uses this result to derive a data-driven approach to correcting tide motion in the grounding zone.

Our results show that tide amplitude and the associated ice speed displacement can be accurately obtained from an ocean model and SAR imaging geometry data. We present results from the methods application across the Antarctic Peninsula and Amundsen Sea, and other floating ice in Antarctica. When we compare the spatial distribution and magnitude of tide corrections, we find that the tide correction ranges from 46 m/y ice speed displacement on the Larsen-C ice shelf for a relatively large 1.5 m tide amplitude, through to a 20 m/yr velocity correction on Pine Island Glacier for a more modest 0.9 m tide amplitude.

Our investigation into the two difference approaches for correcting speed measurements in the grounding zone shows that there are differences between the physic-based and data-driven tidal-response estimates. This shows that while there are modest differences between the two approaches to correcting grounding zone data, the performance of the data-driven method is likely stronger in regions where ice thickness and stiffness measurements are poorly known, such as in East Antarctica, where assumptions in the elastic beam approach are poorly constrained.

In conclusion, tide corrections are an important processing step for SAR derived ice speed measurements over ice shelves, especially when dense time series of speed measurements are required. The choice of tidal-response model in the critical grounding zone region is important. In the future, ice speed in situ calibration validation campaigns will provide important validation data to help further reduce the error on ice speed measurements from space. The results provide guidance for future generation satellite velocity products and contribute to improving the accuracy of earth observation datasets used to monitor the Antarctic Ice Sheet.

Subglacial lakes are bodies of water that exist beneath ice sheets, and their filling and drainage can induce uplift and subsidence of the ice-sheet surface. In fast-flowing glacier regions, differential interferometric synthetic aperture radar (DInSAR), which can detect ice surface elevation changes by differencing two interferograms under the assumption of steady glacier flow, has been used to identify subglacial lake activity and investigate its temporal behavior. However, because the timing, magnitude, and spatial pattern of lake-related elevation changes vary among lakes, the reference interferogram used for differencing has typically been selected separately for each lake. This has limited the application of DInSAR to catchment-scale analyses of interactions among subglacial lakes that may be driven by subglacial water flow. If a catchment-scale reference interferogram representing the glacier-flow component can be defined, DInSAR observations based on this reference could be used to detect ice elevation changes associated with subglacial lake activity more consistently and to extend the analysis to the catchment scale. Furthermore, when combined with the Small Baseline Subset (SBAS) approach, the statistical noise in interferograms can be reduced. By using a multi-temporal baseline network, the SBAS approach can also reduce the risk of gaps in SAR observations while enabling more continuous reconstruction of ice elevation changes associated with subglacial lake activity. In this study, we generated time-series interferograms with 12-day, 24-day, and 36-day temporal baselines from Sentinel-1 SAR images acquired over David Glacier, East Antarctica. We then estimated a reference interferogram representing the horizontal displacement associated with glacier flow as the median of the 12-day interferograms and subtracted it from the remaining interferograms to perform DInSAR. SBAS was then applied to the resulting DInSAR interferograms to investigate time-series ice-sheet surface displacement anomalies over David Glacier. A clear line-of-sight (LOS) displacement anomaly associated with subglacial lake activity was identified over a subglacial lake. In addition, several localized displacement anomalies were observed, suggesting the possible presence of previously undetected subglacial lakes. LOS displacement anomalies were also detected over a wide area along the main flow corridor of David Glacier. These anomalies may reflect acceleration of glacier flow or elevation changes associated with subglacial water flow along the main trunk of the glacier.

Grounding lines are flux gates through which ice discharges into the ocean and are essential for estimating of ice sheet mass balance. Their position reflects ice sheet stability, retreating landward or advancing seaward in response to changes in melting and accumulation, while also exhibiting ephemeral movement driven by tidal flexure of floating ice. While grounding lines derived from Differential Interferometric SAR (DInSAR) phase are regarded as the most accurate (Rignot et al., 2011), most existing products lack formal uncertainty estimates. Errors in their positions directly propagate into ice discharge calculations and can bias estimates of ice mass loss and sea level rise (Rignot et al., 2011). A further limitation is that DInSAR grounding lines are derived from interferograms combining three or four SAR acquisitions, such that each estimated position represents a superposition of multiple tidal states, making it difficult to establish causal links between their position and tidal state.

Our proposed framework addresses both of these issues by obtaining grounding lines from single-difference SAR range offsets and simultaneously providing position uncertainties. We used a time series of Sentinel-1 range offsets at a temporal sampling of 6 days for the Larsen C Ice Shelf. The range offsets are part of the operational processing pipeline used by ENVEO IT to produce monthly and annual Sentinel-1 ice-velocity maps, and were computed by tracking features between consecutive SAR backscatter images (Nagler et al., 2015, Wuite et al., 2026). Grounding line positions were estimated by fitting the range offsets to a one-dimensional Euler–Bernoulli elastic beam model (Holdsworth, 1969) and performing Bayesian inversion using the cross entropy based importance sampling for Bayesian updating (CEBU) algorithm (Engel et al., 2023). This novel algorithm allows for the incorporation of external datasets as priors on model parameters while also accounting for errors in SAR data accrued during offset tracking.

The resulting dataset provides a dense and high-frequency time series of grounding line points for Larsen C. Each point includes a Bayesian estimate of uncertainty, allowing a quantitative assessment of positional confidence. The grounding line points have a mean distance of 348.07 m from contemporaneous Sentinel-1 DInSAR grounding lines. Because the dataset is derived from SAR backscatter rather than interferometric phase, it is robust to coherence loss and can be used to fill gaps in DInSAR grounding line products over fast-flowing outlet glaciers and ice streams. The modular nature of this framework permits easy substitution of ice flexure models and data, thereby offering a promising pathway to derive ice‑thickness and ice‑elasticity estimates as well. Model comparison is also facilitated by the framework’s provision of model evidence, enabling rigorous model intercomparision.

References

1. Rignot, E., Mouginot, J. & Scheuchl, B. Antarctic Grounding Line Mapping from Differential Satellite Radar Interferometry: GROUNDING LINE OF ANTARCTICA. Geophysical Research Letters 38. ISSN: 00948276. (2011).
2. Nagler, T., Rott, H., Hetzenecker, M., Wuite, J. & Potin, P. The Sentinel-1 Mission: New Opportunities for Ice Sheet Observations. Remote Sensing 7, 9371–9389 (2015).
3. Wuite, J., Nagler, T., Hetzenecker, M. & Rott, H. Ten Years of Polar Ice Velocity Mapping Using Copernicus Sentinel-1. Remote Sensing of Environment 332, 115092. ISSN: 00344257. (2026).
4. Holdsworth, G. Flexure of a Floating Ice Tongue. Journal of Glaciology 8, 385–397. ISSN: 0022-1430, 1727-5652. (1969).
5. Engel, M., Kanjilal, O., Papaioannou, I. & Straub, D. Bayesian Updating and Marginal Likelihood Estimation by Cross Entropy Based Importance Sampling. Journal of Computational Physics 473, 111746. ISSN: 00219991. (2023).

Ice shelves exert a crucial buttressing force on marine terminating glaciers, influencing outlet glacier dynamics. When these shelves fracture or thin, this resistive force diminishes, sometimes triggering ice flow speedups upstream and increasing ice discharge to the ocean. Crevassing can also play a role in surface and dynamic mass balance, however the spatial distribution and temporal evolution of crevasses across large ice sheets is poorly quantified. Improved records on the distribution of crevasses over catchments of interest and on an ice sheet wide scale is essential to improve our understanding of these complex relationships. This work aims to begin addressing this knowledge gap, particularly as climate change driven changes to ablation zones may alter future fracture patterns.

Interferometric SAR (InSAR) is a powerful tool for examining fracture processes of individual events. Interferograms enable the detection of small surface displacements, this sensitivity to sub‑wavelength motion (~5.5 cm for Sentinel‑1) makes InSAR uniquely suited for detecting the early stages of fracture propagation and for characterising mechanical changes that precede calving or major rupture. Coherence maps further reveal dense fracture networks that are often invisible in intensity backscatter imagery, reflecting structural weakening.

While InSAR captures incredibly high-resolution crevasse propagation, monitoring crevasses over wider spatial and temporal scales requires automated, scalable methods. To this end, we apply deep learning approaches to Sentinel‑1 SAR backscatter imagery for crevasse detection over Greenland, adapting U‑Net convolutional neural networks originally developed for Antarctica. We apply two different neural net models for two categories of crevasse: ‘Type A’ crevasses which are large, multiple pixels in width and visible from many satellite look-angles; ‘Type B’ crevasses which are finer features that can be a single pixel in width and are most visible when look-angle is perpendicular to the crevasse. Training dedicated models for each type enables detection of major rifts, which on their own can have implications for buttressing, alongside wider patterns of distributed damage.

Our primary case study is Petermann Glacier in NW Greenland, host to one of the last major floating ice tongues on the island. Petermann Glacier has long been regarded as dynamically stable. The glacier’s ice shelf underwent three major calving events during 2010-2012, with the large 2010 calving event not producing notable changes to grounding line position or ice flow speed. Petermann Glacier has recently undergone another major calving event. Early indications suggest a subtle speedup coincident with this event, raising questions about the glacier’s long‑term stability. With a large catchment, significant ice volume above flotation, and susceptibility to ocean forcing, Petermann represents a critical system for examining the links between fracture development and dynamic response.

By combining event‑scale InSAR observations with deep‑learning‑based crevasse mapping across the full Sentinel‑1 mission period, we reconstruct the evolution of distributed damage throughout Petermann’s floating tongue and grounded ice. This framework enables assessment of whether recent changes reflect natural variability or a shift away from the glacier’s historically stable behaviour. Planned extensions include application of the methodology across wider areas of the Greenland Ice Sheet.

Systematic monitoring of rock glacier kinematics through Interferometric SAR (InSAR) is an essential component for investigating climate change impacts on mountain permafrost, as recognized by the inclusion of Rock Glacier Velocity (RGV) as an associated product of the Essential Climate Variable (ECV) Permafrost. Compared to field-based methods such as GNSS or total station surveys, InSAR offers broader spatial coverage but trades the accuracy and directness of point-based measurements for an area-wide, line-of-sight displacement signal that requires careful interpretation. Optical remote sensing approaches based on feature tracking can provide true 2D surface displacement fields and are less sensitive to temporal decorrelation, but are constrained by cloud cover, illumination conditions, and the availability of trackable surface features. Furthermore, obtaining optical imagery with sufficient spatial resolution (< 1 m) and regular temporal sampling remains operationally challenging. InSAR complements these methods by providing spatially dense displacement measurements with consistent revisit capability, though its utility depends strongly on coherence, which can be limited by for example snow cover or rapid surface change.

In this contribution, we present results from the development and application of best-practice guidelines for Sentinel-1 C-band InSAR-based RGV production across several rock glaciers in the Swiss Alps. Processing the complete Sentinel-1 archive (2014–2025) with the GAMMA software, our workflow encompasses SLC co-registration and multi-looking, interferogram generation, adaptive filtering, atmospheric correction, phase unwrapping, and geocoding. The core of the approach is a structured quality assessment framework incorporating coherence-based temporal filtering with frequency- and baseline-appropriate thresholds, pixel-based observation count criteria, and restriction to a consistent summer observation window (July–September) to ensure inter-annual comparability. Filtering parameters are configured independently per rock glacier and orbit/baseline combination. By systematically excluding pixels and interferograms affected by unwrapping errors or low coherence, the framework yields reliable, reproducible RGV time series. Applied across our study sites, we find an overall acceleration trend from 2017–2020, followed by stabilization or slight deceleration through 2023 and renewed acceleration in 2024–2025, consistent with independent field GNSS measurements.

For rock glaciers where velocities regularly exceed the C-band unwrapping thresholds, approximately 1.7 m/yr for 6-day and 0.8 m/yr for 12-day temporal baselines, the operational approach does not provide any data. In such cases, we investigated the potential of a manual point-selection approach using 2–4 strategically placed monitoring points per rock glacier with systematic unwrapping error correction. This extends the measurable velocity range and recovers periods otherwise lost, but requires substantial operator intervention and is not scalable to large inventories. In some cases this approach is also subjective. It is therefore best suited to individual rock glaciers of particular scientific or monitoring interest, where the effort is justified by the value of a continuous record.

To investigate the potential of next-generation L-band missions such as NISAR and ROSE-L, we additionally processed already available SAOCOM-1 data (23.5 cm wavelength) at 16-, 32-, and 48-day temporal baselines for some of the same sites. L-band interferograms maintain high coherence over surfaces where C-band fully decorrelates, particularly on fast-moving fronts, and the longer wavelength substantially raises the phase aliasing threshold, reducing data gaps on fast rock glaciers. At Muragl, a particularly dynamic, polymorphic rock glacier with annual creep rates locally exceeding 1 m/yr, both C-band and L-band velocity estimates agree well with GNSS-derived velocities when accounting for line-of-sight projection geometry, validating both approaches. The limited temporal density of the current SAOCOM-1 archive constrains the statistical robustness of L-band time series at the moment. Together, the two frequencies are complementary: C-band provides reliable rock glacier velocities and benefits from the depth of the Sentinel-1 archive, while L-band extends the observable velocity range and improves coherence on challenging surfaces.

The convergence of C-band and L-band SAR constellations, combined with standardized processing workflows and quality frameworks makes InSAR a useful tool for operational rock glacier monitoring across mountain regions worldwide.

Rock glaciers in the arid tropical Andes represent important cryospheric water reservoirs and sensitive indicators of high-elevation permafrost dynamics. In regions with low precipitation, they provide one of the few persistent sources of water, supporting local ecosystems and communities. Their debris-covered surface insulates the internal ice from atmospheric warming, enhancing their resilience to climate change. Consequently, rock glaciers are expected to play an increasingly significant role as natural water reservoirs in arid Andean environments, particularly in regions where other high-altitude ice sources are scarce. Despite their importance, the seasonal kinematic behavior of rock glaciers in arid mountain environments remains poorly constrained. We investigate rock glacier deformation in the Sajama region of northwestern Bolivia (18°S) using Sentinel-1 Persistent Scatterer Interferometry (2014–2022, extendable to 2026).

Sajama is a region of the Bolivian Andes bordering Chile, where rock glaciers are located at high elevations between 4,500 and 5,600 meters above sea level. The climate of the Sajama region is characterized by cold temperatures and pronounced seasonality in precipitation. Monthly mean temperatures remain close to freezing throughout the year. Precipitation is strongly seasonal, with most of the rainfall occurring during the austral summer months (100 mm/month) and a pronounced dry season in winter from May-August. This climatic setting is favorable for continuous InSAR time-series analysis, due to lack of snow cover making it an ideal environment for studying the seasonal kinematic behavior of rock glaciers.

Ascending and descending line-of-sight (LOS) time series were analyzed to derive multi-year velocity fields and temporally resolved displacement signals. Because a complete rock glacier inventory does not exist for the region, individual rock glaciers were manually identified through visual inspection of their distinct surface morphologies, including both previously known and newly discovered rock glaciers in the analysis. Actively deforming areas were then identified using velocity thresholding and density-based clustering. To investigate seasonal behavior, LOS time series were decomposed using harmonic modeling to extract annual amplitude and phase metrics.

The study assesses whether deformation magnitude and seasonal behavior vary systematically with elevation, slope orientation, and regional wet–dry climatic cycles. By combining multi-year InSAR analysis with detailed geomorphometric and climatic context, we aim to better constrain the climatic sensitivity and hydrological relevance of tropical rock glaciers in the Bolivian Andes, providing a foundation for future studies on high-elevation permafrost dynamics and cryospheric water resources.

Landslides frequently result in severe casualties and substantial economic losses worldwide, making reliable deformation monitoring a critical component of hazard mitigation and risk management. Interferometric synthetic aperture radar (InSAR) has emerged as a powerful remote sensing technique for landslide deformation monitoring. Nevertheless, phase unwrapping (PU) under high-gradient deformation remains a fundamental challenge. Conventional single-baseline (SB) PU methods predominantly rely on the phase continuity assumption, which is frequently violated in areas characterized by rapid deformation, such as active landslides induced by underground mining activities. In contrast, existing multi-baseline (MB) PU approaches are primarily developed for topographic phase recovery and typically operate in a pixel-wise manner, limiting their effectiveness in resolving severe deformation gradients.

To address these limitations, we propose improved multi-temporal baseline PU approaches specifically designed for high-gradient deformation scenarios and validate them using both simulated experiments and real landslide cases. Assuming that the deformation rate is approximately linear over short temporal intervals, we establish relationships among multi-temporal interferograms on Arcs. The initial ambiguity gradients on each Arc are determined via integer programming to ensure global consistency, after which the absolute phases of selected pixels from representative interferograms are optimally estimated using an L^p-norm optimization model, enhancing robustness against noise and local inconsistencies. To further improve computational efficiency, a quadtree segmentation strategy is introduced to adaptively partition the study area according to wrapped phases. Simulation experiments demonstrate that the proposed method improves phase recovery accuracy by approximately 23 times and computational efficiency by nearly one order of magnitude compared with conventional approaches. Furthermore, a typical mining-induced high-gradient landslide located in the mountainous region of southwestern China is selected as a real-case study using Sentinel-1 data. The proposed approach is systematically compared with existing one-dimensional to three-dimensional PU methods. Results indicate that our method is capable of recovering larger-magnitude and more reliable deformation signals under severe gradient conditions.

Second, we address the challenge of an unknown feasible solution space, which can significantly degrade the performance of multi-temporal baseline PU when deformation gradients exceed the assumed bounds. We analytically demonstrate that inaccurate FSS estimation results in solution instability, particularly under extreme deformation conditions. To mitigate this problem, a constrained FSS is constructed using a priori deformation information derived from SAR offset-tracking, thereby providing physically meaningful bounds for ambiguity estimation. By incorporating this constraint into the multi-temporal PU framework, the stability and robustness of phase recovery are substantially improved. A representative landslide case located in the mountainous region of southwestern China is further selected for validation using high-resolution ALOS/PALSAR-2 data. Performance comparisons in terms of single-interferogram phase quality, deformation rate estimation, and time-series reconstruction consistently confirm the superiority and reliability of the proposed constrained approach under high-gradient deformation conditions.

Unlike conventional PU approaches that depend on phase continuity assumptions, the proposed framework is tailored for deformation environments where such assumptions fail. Experiments on simulated and real datasets verify its robustness and scalability under varying gradient conditions. Although developed for landslide monitoring, the strategy is readily transferable to other severe deformation settings, including mining subsidence and earthquake-induced deformation, thereby broadening the applicability of multi-temporal InSAR for geohazard assessment and deformation monitoring.

Landslide disasters are a global geohazard phenomenon. Conducting early identification of landslides prior to their occurrence, along with timely detection and monitoring of surface deformation, constitutes a crucial technical approach for effective landslide prevention. However, with global warming and the expansion of human engineering activities, the frequency of landslide events has been increasing annually, and an increasing number of ancient landslides are being reactivated. This situation imposes new requirements on the timeliness of landslide deformation detection.

Time-series Interferometric Synthetic Aperture Radar (InSAR) offers the advantages of all-day, all-weather, and large-scale monitoring, and has therefore been widely applied in early landslide identification and monitoring applications. Nevertheless, due to the complex processing workflow of time-series InSAR, unavoidable phase unwrapping errors, and the uncertainties associated with error propagation, the accuracy, efficiency, and reliability of landslide detection results in wide-area complex mountainous regions often fall short of meeting engineering application requirements. For early landslide detection, obtaining only the boundary extent of the landslide feature field and the trend of deformation magnitude can suffice to locate potential hazard zones and assess danger levels, thereby enabling ground-based monitoring and prevention. Consequently, the Stacking method was proposed, which rapidly estimates landslide deformation rates by stacking time-series interferometric phases. Its straightforward and fast processing strategy has been widely adopted in early landslide detection. However, the Stacking method also necessitates phase unwrapping and struggles to effectively mitigate vertically stratified atmospheric disturbances in complex mountainous areas. Therefore, rapid landslide detection methods based on phase gradient stacking have gained further expansion and application in recent years.

Compared to traditional Stacking deformation rate estimation methods, landslide detection using phase gradient stacking still faces the following issues. Firstly, the results of phase gradient stacking lack a clearly corresponding physical quantity (such as deformation magnitude or rate), and the errors inherent in gradient stacking cannot be assessed. Consequently, the detection outcomes from gradient stacking fail to provide quantitative deformation accuracy. Secondly, the gradient field only reflects, to a certain extent, the boundary characteristics of the deforming area and does not correspond to a complete deformation field. This makes it difficult to identify areas of uniform or slow deformation within the landslide body. Moreover, as the gradient step increases, the boundaries of the stacked gradient field become increasingly blurred. Finally, since phase gradient stacking requires stacking the entire time-series InSAR interferometric phases, the landslide detection results lose temporal observation capability, thus failing to dynamically capture newly developed landslides.

Addressing the various difficulties encountered in rapid landslide detection using phase gradients, this paper proposes a landslide deformation detection method based on phase gradient analysis and stacking. This method extracts genuine deformation gradients from interferometric phases and recovers the spatial absolute deformation field through spatiotemporal double-difference phase analysis and time-series progressive gradient stacking. This approach circumvents phase unwrapping processing, enabling efficient extraction of the surface deformation field. Furthermore, the errors inherent in the phase gradient stacking process are also assessed for accuracy, facilitating fast, robust, and wide-area landslide deformation detection.

In mountainous landscapes, bedrock landslides represent a major natural hazard that threatens the safety of populations and infrastructure. In Himalaya, they are also widely recognized as the primary erosion mechanism driving hillslope erosion and landscape evolution. Diversity in rock weakening and geologic setting leads to a large diversity of landslides types, involving distinct depths and geometry of the failure surface. It also leads to a wide range of deformation rates, from rapid and catastrophic landslides to extremely slow deformations (a few m/kyr). While rapid landslides have been the subject of numerous studies, particularly because of the natural hazards they pose, slow-moving landslides have been investigated much less. However, slow-moving landslides can have a significant impact both on infrastructure and as precursors to rapid landslides: understanding their dynamics and the factors controlling their deformation therefore represents a significant research challenge.

Deep seated slow-moving landslides in the Himalaya reveals different types: Deep Seated Gravitational Slope Deformation (DSGSD), characterized by very slow movements, with a pervasive zone of deformation at depth; Deep seated slides, which affect the whole hillslope and displace panels over hundreds of meters, with intense fracturing and loss of cohesion; Secondary slides developing within the cohesionless mass of the previous type and associated with rapid retrogressive incision of a talweg into the cohesionless mass. When they undergo phases of activity, the last two types present average deformation rates ranging from cm/a to tens of m/a. However, they can display high spatial and temporal variabilities, controlled by external factors including earthquake and monsoon.

In Himalayan setting, the ability to identify, map on large scale and monitor slow deformations on hillslopes relies on remote sensing techniques. However, the many difficulties associated with widespread cloud cover during the monsoon season, steep slopes, and atmospheric problems for radar interferometry require the use of multiple sensors and methods to achieve the high-precision monitoring required.

Here, we demonstrate that a combination of optical and radar datasets over the Bhote Koshi Valley (Central Nepal), leveraging both image correlation and Sentinel-1 InSAR processing, is effectively capable of characterizing the ground displacement associated with slow-moving landslides in a non-favorable environment.

We established displacement records of several landslides, from ERS and ENVISAT radar archives (since 1992), high return time Sentinel-1 and 2 constellations since 2015 and 2016, as well as very high resolution optical Pléiades images (since 2014) and stripmap PAZ radar images (for the 2025 monsoon). Using image correlation and VHR Pléiades DSM differences, we show the effective detection of ground displacements from metric trends over several years for lower resolution sensors up to a fully characterizable displacement of tens of cm/yr for very high-resolution sensors. The high variability of the speckle pattern in Sentinel-1 images is mitigated using image stacking over 50 days and correlation refinement in frequency domain, effectively enhancing the signal-to-noise ratio of the correlation’s results. Different trends can be isolated and associated with annual events including monsoon, as well as punctual events like the 2015 Mw7.8 Gorkha earthquake, therefore allowing estimation of the response to landslide triggers.

Additionally, we processed Sentinel-1 InSAR data with additional steps to mitigate the low coherence of mountainous areas and the high deformation gradients associated with acceleration phases of landslides such as monsoon. These steps include multi-looking and filtering of interferograms integrating a proxy reflecting potential biais and coherence variation from the humidity, snow and vegetation. Besides, we apply a series of correction on the wrapped phase to reduce the variability of the phase and thus enable unwrapping. The corrections include: (1) empirical flattening of the interferogram in range and azimuth directions, (2) empirical phase-elevation estimations to remove stratified delays, (3) empirical removal of the deformation during monsoon. This last step relies on the correlation results deformation model to effectively reduce the wrapped phase gradient during landslide activity and therefore reduce aliasing of the phase.

By combining image correlation and de-ambiguated InSAR results over this unique dataset, 29 moving landslides were detected on the 760km2 study area with areas ranging from 0.1 to 10km², including 13 detectable from image correlation. Displacement characterization can be achieved for all landslide velocities even during acceleration periods, from cm/yr for InSAR detection and tens of cm/yr for high-resolution image correlation. One of the largest landslides detected, located in the Balephi Valley northeast of Kathmandu and covering an area of 2.2 km², has an average annual displacement of 1 to 2 m/y. This displacement is not continuous but mostly occurs during the monsoon season. Data from a GNSS station positioned on this landslide confirm that in 2025 the landslide responded to hydrology, with displacement limited to the very end of the monsoon and following the maximum elevation of the water table along the hillslope. These large displacements of 1 to 2 m are also systematically underestimated by conventional InSAR processing due to phase aliasing. This type of landslide with intermediate deformation rate clearly illustrates the complementary nature of the InSAR and image correlation approaches: the first method identifies active slopes, while the second accurately documents their spatial extent and average speed. Combining the two methods provides in addition precise information on temporal dynamics of the slow-moving landslides.

Landslides are a major environmental and socio-economic hazard, with their frequency and intensity expected to increase under changing climatic conditions. Their sudden occurrence and complex geological, hydrological, climatic, and human interactions make reliable prediction and monitoring challenging. In this context, systematic multi-source monitoring of surface changes and the identification of slope instability precursors are key components of landslide risk mitigation strategies.

This study evaluates the potential of integrating multi-source remote sensing data to monitor and identify landslide precursors. The research was conducted in two landslide-affected areas, combining radar and multispectral satellite data with very high-resolution RGB and multispectral imagery acquired using unmanned aerial vehicles (UAVs). Surface displacements were calculated using radar-based Persistent Scatterer Interferometric Synthetic Aperture Radar (PS-InSAR) and UAV-photogrammetry based on orthomosaics and digital terrain models (DEMs). The comparison of both methods revealed high agreement (RMSE = 0.017 m), confirming that PS-InSAR and UAV-photogrammetry can provide complementary monitoring of terrain deformation at regional and local scales.

In addition, multispectral data from Sentinel-2 and Landsat-9 satellites, as well as UAV platforms, were used to derive vegetation indices, including the Normalized Difference Vegetation Index (NDVI), Green Normalized Difference Vegetation Index (GNDVI), and Normalized Difference Water Index (NDWI). Mean vegetation indices derived from satellite observations were compared with high-resolution UAV results to assess their reliability for detecting early signs of slope instability at a regional scale. Quantitative analyses showed a strong correlation between satellite- and UAV-based results in the studied areas (r ≥ 0.88). Spatial analysis further confirmed that publicly available satellite observations are effective for regional-scale vegetation monitoring, while UAV imagery enables the detection of localized anomalies indicative of early slope destabilization.

A two-stage framework for landslide monitoring and precursor identification is introduced. The first stage focuses on regional-scale assessment using multi-temporal satellite radar and multispectral data to analyse surface displacements and vegetation index changes, supported by machine learning algorithms for automated preliminary identification of high-risk areas. The second stage involves detailed investigations of selected sites using high-resolution RGB and multispectral UAV data, enabling precise local-scale characterization of terrain deformation and surface changes. The framework also incorporates the analysis of historical data from stable periods, enabling the determination of accuracy thresholds for each technique. Changes exceeding these thresholds can then be identified as anomalies potentially indicating landslide processes.

By integrating multiple sensors and methods, the approach leverages the strengths and mitigates the limitations of individual techniques, enabling scalable and comprehensive monitoring strategies with potential applications in operational landslide early warning systems.

Landslide hazards in New Zealand are responsible for significant economic and social damage every year with an estimated cost between NZ$ 250 – 300 million and more than 1500 landslide related deaths since 1760. The most common trigger of landslides in New Zealand is heavy rainfall. Extreme weather events such as Cyclone Gabrielle in 2023 triggered an estimated 800,000 landslides, mainly across the north island.

Historically, monitoring landslides has been done with ground-based instrumentation. However, these methods are time-consuming and costly in nature and are unsuitable for a nationwide landslide study. The launch of the Sentinel-1 mission in 2014 has accelerated research surrounding the application of satellite data to monitor geohazards such as landslides. Interferometric Synthetic Aperture Radar (InSAR), from satellites such as Sentinel-1, is widely used in studies that identify and map landslide deformation due to the nature of freely available data, frequent revisit time, global coverage, and its ability to measure accurate deformation on the scale of millimetres.

However, solely relying on InSAR data makes estimating the 3D displacement of landslides challenging to resolve. This is because of InSAR’s relative insensitivity to north-south movement, estimates from InSAR observations are often limited to deriving the downslope component of motion. This typically assumes that movement is either along the same direction as the steepest slope, or that there is zero movement perpendicular to the maximum slope direction. However, for many landslides the assumption of downslope movement may not always be true. In the case of rotational landslides, there will be spatial variations in the slip vector with near vertical movement at the head and in some cases uplift of the toe. Here we exploit a suite of high-resolution, multi-mission InSAR datasets, digital elevation models and, where available, a-priori surface displacement datasets to help constrain the 3D-component of landslide movement across New Zealand. Through our analysis, we capture widespread, large-scale slope movements across New Zealand and find complex internal deformation within many of the landslide masses. Within individual landslides, there can be significant variations in their horizontal and vertical rates of movement highlighting zones of compression and extension. These data will be used to develop modelling tools for the rapid assessment of landslide behaviour in support of more complex numerical models.

Landslides represent some of the most destructive natural hazards encountered in unstable mountainous regions, such as the Western Ghats in India. The examination of landslides has garnered significant global attention due to their profound impacts on socio-economic activities. The utilization of remote sensing and geographic information systems has proven valuable for integrating the spatial factors that contribute to landslide occurrences. In this study, satellite imagery from Sentinel 1-C Band has been employed, and further interferometry techniques for detecting the landslide event in 2024 in the Wayanad district, Kerala. Leveraging Artificial Intelligence Techniques in RADAR remote sensing such as machine learning algorithms, particularly the Random Forest (RF) model, were utilized to classify the study area into affected and non-affected regions. The findings also indicate the affected land use and land cover in the given study area. In the end, it can be concluded that significant landslides on 30th July, 2024 in the Wayanad district were primarily precipitated by anthropogenic interventions, compounded by heavy precipitation and unstable topography. Activities such as stone quarrying and infrastructure development emerged as critical factors contributing to these landslides. This research provides valuable insights aimed at mitigating landslide hazards in the Wayanad district, thereby fostering sustainable development.

Landslides constitute a recurrent geomorphic hazard in the mountainous regions of Vietnam, where intense monsoonal rainfall, deeply weathered lithologies, and active tectonics promote widespread slope instability. In these humid tropical environments, dense vegetation cover, rapid biomass regeneration, and seasonal canopy variability introduce substantial challenges for satellite-based deformation monitoring. Temporal decorrelation caused by vegetation dynamics significantly reduces interferometric coherence, particularly in C-band systems, limiting the reliability of long-term displacement retrieval in forested terrain. These constraints are especially relevant for slow-moving landslides, where deformation rates are subtle and require high phase stability for robust detection.

In this study, we applied multi-year Sentinel-1 Interferometric Synthetic Aperture Radar (InSAR) time-series analysis to reconstruct the spatiotemporal evolution of slow-moving landslides in a selected region of northern Vietnam with low vegetation cover. Differential interferograms were generated using the cloud-based Hybrid Pluggable Processing Pipeline (HyP3) and processed through a Small Baseline Subset (SBAS) workflow implemented in the open-source MintPy software. Both ascending and descending Sentinel-1 datasets were analyzed to enhance the robustness of displacement estimates and to reduce line-of-sight (LOS) directional bias. Daily precipitation data were integrated to evaluate the relationship between deformation patterns and the Southeast Asian monsoon regime.

Given the dense tropical vegetation across many Vietnamese slopes, careful selection of interferometric pairs and strict baseline thresholds were implemented to minimize temporal and spatial decorrelation. Coherence filtering and masking strategies were applied to exclude low-quality pixels, resulting in deformation estimates primarily derived from partially vegetated slopes, exposed scarps, road cuts, and sparsely forested sectors. Although these measures improved signal reliability, coherence loss during peak monsoon months remained significant, leading to localized data gaps and increased uncertainty in some forested sectors. Consequently, deformation estimates in heavily vegetated zones should be interpreted conservatively, and the results highlight the inherent limitations of C-band InSAR in humid tropical settings.

Despite these challenges, the InSAR time series reveals sustained, low-magnitude slope deformation across all investigated sites, confirming ongoing activity. Distinct periods of acceleration were observed during the main monsoon season (typically September-October), with peak displacement rates occurring in months characterized by cumulative rainfall maxima. While deformation does not exhibit a strict one-to-one correspondence with individual rainfall events, multi-week-to-seasonal rainfall accumulation appears to act as a conditioning and amplifying factor, indicating a hydrologically modulated response typical of deep-seated, slow-moving landslides. Outside the monsoon season, displacement rates decrease but do not cease entirely, demonstrating persistent creep behavior.

This study demonstrates that, despite coherence limitations imposed by dense tropical vegetation, cloud-based InSAR processing combined with open-source SBAS time-series analysis can provide valuable quantitative insights into landslide kinematics in monsoon-dominated environments. The findings contribute to improved hazard characterization, seasonal risk assessment, and long-term infrastructure management strategies in Vietnam's landslide-prone regions.

The study has been supported by the National Science Center (project no 2023/49/B/ST10/02879).

The Kandersteg region in the Swiss Alps presents a highly active paraglacial landscape characterized by deepseated gravitational slope deformations, complex mass wasting processes, and evolving rock glaciers. Continuous spatial monitoring is essential for understanding these slope failure mechanisms and assessing regional hazard potential. However, quantifying surface kinematics in such extreme high-relief environments poses significant challenges for spaceborne Synthetic Aperture Radar (SAR) techniques. Prolonged seasonal snow cover, highly dynamic tropospheric conditions, and steep terrain gradients frequently cause severe temporal decorrelation, layover, and shadow effects in C-band data. To capture the complete kinematic diversity across this complex terrain, we integrated InSAR time-series analysis with SAR amplitude offset tracking. Our dataset encompasses a five-year time series (2020–2025) of Sentinel-1 acquisitions from both ascending and descending orbit. Utilizing this dual-geometry configuration is critical for mitigating inherent geometric distortions in alpine valleys. It also allows for the robust decomposition of Line-Of-Sight (LOS) measurements into two-dimensional East-West and vertical displacement vectors, yielding a much more realistic representation of the actual slope mechanics. The data processing relies on a multi-platform pipeline tailored to address specific deformation regimes. For stable slopes and regions exhibiting slow to moderate creep, we generated unwrapped interferograms using the HyP3 cloud-processing architecture. Conversely, in the most active kinematic zones where rapid displacement gradients cause complete phase aliasing and decorrelation, we applied SAR offset tracking using the ISCE2 processing framework. By performing sub-pixel cross-correlation on the amplitude imagery, we successfully extracted large-scale displacements that would otherwise remain undetected by standard phase analysis. To seamlessly merge these distinct datasets, both the HyP3 unwrapped phase products and the ISCE2 offset tracking fields were ingested into MintPy for time-series inversion. This combined integration allowed us to generate continuous, multi-temporal deformation maps bridging the gap between millimeter-scale creep and meter-scale sliding. To ensure the reliability of the remote sensing observations, the derived SAR velocity fields were validated against kinematic measurements documented in existing regional literature. Following this independent validation, we systematically cross-correlated the extracted displacement time-series with localized environmental variables to assess climatic drivers. Specifically, Ground Surface Temperature (GST) records provided by the PERMOS (Permafrost Monitoring Switzerland) network were utilized to evaluate the thermal control on slope kinematics. This correlation isolates long-term gravitational deformation trends from transient, temperature-dependent accelerations. Ultimately, this framework provides critical physical insights into how complex alpine slopes respond to external forcing, demonstrating a highly effective approach for operational hazard monitoring in steep terrain

In the framework of an ESA-PECS project (ID: 000114846/15/NL/Nde), the HUN-REN Institute of Earth Physics and Space Science in cooperation with the Budapest University of Technology and Economics and University of Leeds, a novel twin corner reflector design was developed to address the issue of limited coherent reflections in areas affected by landslide and other natural hazards, enabling accurate InSAR observations. This design was applied in an integrated approach combining InSAR and GNSS technologies to monitor landslides in Southern Hungary (Dunaszekcső settlement), where recurrent landslide has been active since 2007.

Before installing the reflectors, there were practically no coherent reflections, making it difficult to process InSAR data. The reflectors, installed on a geodynamical benchmark, serve as coherent reflecting objects, allowing for accurate co-located InSAR and GNSS measurements. Our results show that the combined use of InSAR and GNSS datasets can provide accurate estimates of surface movements, with an accuracy in the range of 1-5 mm in the line-of-sight direction. The integration of InSAR and campaign GNSS measurements using Kalman-filtering enables the reconstruction of detailed displacement time series. Campaign GNSS measurements, carried out approximately every 6 months between 2016 and 2017, can be used to identify the pixels dominated by the integrated benchmarks, detect unwrapping errors and missing cycles, and also provide boundary values for Kalman-filtering. The latter is essential for estimating the north component of movements which cannot be determined by InSAR technologies alone.

We have also installed low-cost GNSS receivers near the reflectors, which allowed us to collect multiple months of data, enabling a more comprehensive analysis of the landslide dynamics. During previous measurement campaigns it has become clear that landslide movement velocities often lead to phase unwrapping errors. Continuous GNSS measurements enable the detection of unwrapping errors, quality control and comparison with InSAR time series, providing a more complete understanding of landslide behavior. We will also compare the InSAR and GNSS derived time-series in combination and separately to assess the advantages and disadvantages of low-cost GNSS monitoring compared to the corner reflector solution. Based on the results a recommended strategy will also be presented for monitoring landslides and natural hazards using space-geodetic technologies.

InSAR-based landslide investigations remain constrained by manual parameter tuning. It limits their applicability at the sub-regional scale for susceptibility mapping and cataloguing of unstable terrain in mountainous regions. Open-access Sentinel-1 (S1) radar data offer extensive spatial and temporal coverage; however, their exploitation demands substantial time and computational resources across the full InSAR processing chain. Such requirements are often disproportionate to the comparatively smaller dimensions of the failure zones. Within the PEPR IRiMa program (a French government-funded Priority Research and Equipment Program) we developed an automated InSAR workflow tailored to gravitational movements. Our objective is to minimise manual interventions and deliver a rapid insight into the target area, while enabling the delineation of deformation extent (at a single burst level), which can be broader than in-situ observations.

This initiative is willing to complement the existing PSI-based EGMS yearly-update products, whose reliance on long-term signal stability reduces their coverage in low-coherence natural terrain, especially on low-rate deformed slopes. To mitigate these limitations, we employ a standard stacking approach¹ to detect spatially distributed hillslope motion and to derive (slow-to-moderate) mean-velocity maps, computed from a set of unwrapped differential interferometric phases. In this study, we also examine the phase gradient stacking approach² in a landslide context.

We implement the automated processing pipeline in the GAMMA software suite, using C-band S1 images acquired in the IW mode (Interferometric Wide swath), restricted to a reduced-swath segment (one to two bursts only). Additionally, we develop an in-house unwrapping procedure based on the phase gradient. We apply our methodology to the French-Alps region, leveraging existing landslide inventories³. In particular, we concentrate on two landslide case studies. Firstly, the Marie landslide, which is well known and instrumented on the ground. Secondly, the unknown -difficult to access La Pinée (Daluis) landslide whose kinematic behavior remains unknown today, yielding another unique opportunity to showcase the advantages of using spaceborne InSAR. We benchmark stacking-derived velocities against time-series methods to assess performance and identify biases. This evaluation provides a basis for refining our automated stacking strategy without adopting the heavy time-series processing. We further evaluate the results using ancillary datasets to support validation and interpretation.

Keywords: landslide, InSAR, automated workflow, slope, mean-velocity estimation.

References:

1) Ciuffi, P., Bayer, B., Berti, M., Franceschini, S., Simoni, A.: InSAR stacking to detect active landslides and investigate their relation to rainfalls in the Northern Apennines of Italy, Geomorphology, Volume 457, 2024, 109242, ISSN 0169-555X, https://doi.org/10.1016/j.geomorph.2024.109242.

2) Sandwell, D. T., & Price, E. J. (1998). Phase gradient approach to stacking interferograms. Journal of Geophysical Research, 103(B12), 30183–30204. https://doi.org/10.1029/1998JB900008

3) Aslan, G., Foumelis, M., Raucoules, D., De Michele, M., Bernardie, S., Cakir, Z.: Landslide Mapping and Monitoring Using Persistent Scatterer Interferometry (PSI) Technique in the French Alps. Remote Sens. 2020, 12, 1305. https://doi.org/10.3390/rs12081305.

Civita di Bagnoregio, located in Central Italy at the boundary between the Vulsini volcanic district and the Tiber river valley, represents one of the most emblematic examples of Cultural Heritage threatened by geomorphological processes. The historic settlement is built on a pyroclastic plateau resting on thick Plio-Pleistocene clay formations, a geological asset that prompt slope instability driven by superficial erosion, tension crack development, toe erosion, and progressive retreat of the whole tuff cliff. The interaction between lithological contrasts (plastic bedrock over rigid outcrops), intense fracturing of the volcanic deposits, climatic forcing, and anthropogenic cavities has produced a highly dynamic landscape characterised by rockfalls, rotational slides, earth flows, and shallow debris flow. The rapid erosion of the surrounding “Valle dei Calanchi” makes Civita di Bagnoregio an exceptional natural laboratory for testing advanced Earth Observation approaches to Cultural Heritage monitoring.

The present work shows the main results developed within the first Special National Plan for Monitoring and Conservation of Italian Cultural Heritage (NPMCCH), aimed at the monitoring, conservation, and proactive protection of cultural heritage, against the impacts of different hazards, both anthropogenic and natural, including climate-induced extreme events. Civita di Bagnoregio was selected as a priority case study to evaluate the potential of multi-mission InSAR approaches, using an A-DInSAR technique, and integrating remote sensing, GB radar monitoring and field surveys.

A layered ground-motion analysis workflow was implemented. Regional-scale products, including Copernicus European Ground Motion Service (EGMS) Sentinel-1 time series, were first used to characterise the long-term deformation context and identify sectors requiring higher-resolution investigation. Although EGMS provides a consistent baseline, the Civita case highlights the limitations of radar C-band observations in mostly natural environments characterised by localised, non-linear, or rapidly evolving geomorphological processes.

To overcome these constrains, high-resolution COSMO-SkyMed (X-band) time series covering the period 2022–2026 were processed using multi-temporal interferometric techniques. The analysis identified clusters of Persistent Scatterers showing millimetric deformation rates, particularly concentrated in two priority areas: the pedestrian bridge zone, connecting the village to the mainland and sectors of the historic centre, including structures facing the western side of the main square. Time-series analysis indicates deformation patterns consistent with structural stress redistribution and slope-controlled processes. Field inspections confirmed minor cracking and local structural distress corresponding to interferometric anomalies, supporting the reliability of the satellite-derived indicators.

The study adopts a multi-frequency InSAR framework integrating C-band regional monitoring, X-band structural observations, and ongoing L-band processing using SAOCOM data. Wavelength complementarity enables the investigation of deformation across spatial scales, bridging building-level responses and landscape-scale slope dynamics while improving temporal continuity in challenging environments.

L-band processing represents a central component of the study. Preliminary assessment confirms the expected improvement in temporal coherence over vegetated slopes and clay-rich terrains where shorter wavelengths experience decorrelation. L-band processing is currently ongoing and is expected to enhance temporal coherence and to enable the characterisation of slope deformations thanks to the multi-frequency framework.

Preliminary results highlight that ground deformation in Civita is spatially heterogeneous and strongly controlled by the geological asset of the tuff-clay system and by climatic forcing, particularly precipitation-driven erosion and thermo-hydrological cycles. The combined satellite analysis and field investigations allowed the identification of priority monitoring targets and supported the design of an integrated low-impact monitoring strategy calibrating A-DInSAR observations with GNSS measurements and integrating them with in situ sensors, including crack meters, meteorological stations, and corner reflectors. Such hybrid architectures facilitate the transition from periodic assessment toward near-real-time early warning.

The Civita di Bagnoregio case provides an experimental reference site to evaluate how multi band SAR observations can support conservation strategies in fragile cultural heritage environments. The proposed workflow positions EGMS as a screening layer, X-band as a structure deformation detector, and L-band as a sensor enabling the observation of spatially distributed geomorphological deformations in natural environments, contributing to the evolution of next generation ground motion services and future SAR mission exploitation.

Landslides result in numerous fatalities and significant infrastructure damage each year, leading to enormous individual and economic losses and severe damage to infrastructure worldwide. Due to climate change and its associated effects, the frequency and magnitude of landslides, rock falls, and mudflows increase globally. In March 2023, Tropical Cyclone Freddy brought extreme and prolonged rainfall to southern Malawi, triggering widespread landslides that affected settlements and critical infrastructure. Reliable information on the spatial extent of slope failures and their impact on infrastructure is essential for supporting humanitarian response and recovery planning.

Synthetic aperture radar (SAR) provides important advantages in such emergency contexts, as it is independent of cloud cover and daylight conditions. In this study, we investigate the potential of Sentinel-1 SAR data for infrastructure-focused landslide assessment after Cyclone Freddy. First, we generate pre- and post-event coherence maps to detect surface changes associated with landslide occurrence and structural damage. Coherence loss patterns are analysed in relation to road networks, built-up areas, and critical facilities to evaluate potential disruptions and identify places where coherence reduction may indicate mass movement activity or structural impact. Second, we perform Interferometric SAR (InSAR) processing using a Small Baseline Subset (SBAS) approach to investigate surface deformation signals. Interferogram generation and co-registration are carried out using the InSAR Scientific Computing Environment (ISCE) processing framework. The resulting interferograms are filtered and phase-unwrapped before being imported into the Miami InSAR Time-series software in Python (MintPy) for time series analysis. Within MintPy, we create a deformation time series to estimate line-of-sight (LOS) displacement and velocity fields, enabling the detection of post-event ground displacement associated with landslide movement.

The combined analysis of coherence change detection and deformation time series allows us to evaluate the strengths and limitations of radar-based approaches for landslide assessment under humanitarian constraints. We assess data availability, processing requirements and data quality in vegetated tropical terrain. By focusing on infrastructure impacts and practical use, this study links advanced radar methods to the needs of humanitarian organisations. The results demonstrate how radar-based techniques can contribute to targeted assessment of landslide impacts, thereby supporting decision-making during disaster response and recovery phases.

This paper presents an initial analysis of landslide deformation in the district, located in the southeast of Turkey near the borders of Iraq and Syria. The study area is situated at an altitude of approximately 1,400 meters and has a mostly dry environment. The study uses the StaMPS approach of the Persistent Scatterer Interferometry (PSI) method with long-term interferometric Synthetic Aperture Radar (InSAR) data. Copernicus Sentinel-1A data of Single Look Complex (SLC) Interferometric Wide swath (IW) data are acquired from the NASA Alaska Satellite Facility and used for the extraction of landslide movement in the region. The dataset covers image acquisition of both ascending and descending tracks between October 2017 and August 2022. In total, 168 and 149 images of descending and ascending mode images were processed, respectively. SAR data was processed using SNAP v9.0 software, and interferograms were produced. The data in vertical transmitted and vertical backscattered polarization were analyzed using the StaMPS 4.1 beta version to produce time series and obtain velocity vectors of the landslide region. TRAIN software was also used to remove the atmospheric phase in the interferograms. According to the results obtained by the five-year time series analysis, the displacement varies between approximately -9 mm/year and 6 mm/year. In both cases, the standard deviation is less than 0.7 mm/year. The total maximum movement was determined to be approximately 45 mm. The time series analysis shows the dynamics of the slowly developing landslide in the long term. Most of the movement obtained was observed in the city center and over the human-made structures. In the velocity map, another small local movement is observed in the northeast of Sirnak city center, apart from the main landslide pattern. The study area has a continental climate, and the general natural vegetation is steppe since the season's precipitation is low. Therefore, slight displacements have been determined in regions other than man-made areas. Observation of opposite movements in the results obtained from descending and ascending images shows preliminary information that there is horizontal movement. By using these two datasets, the results obtained along the line-of-sight (LOS) direction can be converted into horizontal and vertical movements, and detailed information about the direction and size of the movement can be obtained. This is the first study that has been conducted for this region using the InSAR time series. Local movements are also observed in Turkey, except for the big cities, which are constantly monitored, and it is revealed that these areas should also be examined. Especially slowly developing landslides should be considered risky areas as they will affect both buildings and infrastructure for a long period, and precautions should be taken by examining the geological structure of the region. As a further study, small baseline methods can also be tested in order to increase the number of PS points in the region. The prediction approaches can be beneficial in risk reduction activities and sustainable urban management strategies.

The Northern Apennines of Italy is a young, uplifting mountain chain where weak rocks are widely outcropping. Landsliding is a key geomorphological process actively shaping the landscape. Large complex landslides are widespread and characterized by long dormant phases interrupted by periodic catastrophic reactivation which pose a significant threat to human activities, due to their potential impact on houses, settlements and infrastructures. Catastrophic reactivations are invariably triggered by intense and prolonged rainfall events. However, they are often preceded by measurable deformations and are usually followed by a long deceleration phase leading to dormancy again.

The Cà di Sotto landslide in San Benedetto Val di Sambro (BO), Italy is a well-documented large complex landslide (> 45 hectares) that in 1994 destroyed some buildings and dammed the river at the toe, requiring drainage works to mitigate the risk associated to a potential flood. This landslide originates in the upper part of the slope as a rockslide and progresses downslope evolving into an earthflow with total length of about 2 km. The affected material, belonging to the Monte Venere Formation, consists of tectonized calcareous-marly turbidites interbedded with arenaceous-pelitic strata. After 30 years of dormancy, in October 2024, following a heavy rainfall event, the landslide underwent a new catastrophic reactivation initially triggered by a rockslide in the upper slope. The deformation quickly progressed downslope, with peak velocities of several meters per day, disrupting previously established mitigation measures and generating a large water impoundment, upstream of the landslide toe. After about a week, the deceleration phase occurred gradually and was interrupted by brief episodes of acceleration linked to major rainfall events in December 2024 and February 2025. Still today, the landslide has not yet reached its dormant phase and heavy rainfalls occurred during the last winter season have caused local retrogression of the crown area and/or partial reactivation episodes along the landslide body.

We use two-pass standard interferometry to investigate the pre- and post-failure behavior of the Cà di Sotto landslide. The two-pass technique retains all displacement information contained in interferograms and allows us to overcome some of the limitations associated with multi-temporal techniques, like sparse spatial coverage in highly deforming areas due to decorrelation and exceeding of measurable displacement rates. We take advantage of the short revisit time of Sentinel 1 satellites and process interferograms with temporal baselines of 6 to 24 days using gmtsar. Our analysis goes back to 2015 to investigate the dormant phase of the landslide and includes all phases of movement except the catastrophic reactivation phase (about 10 days) when displacement rates largely exceeded the upper detection limit of 6-days interferograms. The spatial extent of the catastrophic phase was instead resolved through amplitude change detection using pre- and post-event scenes. The results of our analysis are compared to available ground-based measures for comparison and validation. They include a GNSS-based monitoring system, comprising 31 evenly distributed periodic measuring points and three dual-frequency permanent GNSS stations. Few weeks later, also a robotic total station was installed as an early warning system for future possible reactivations. It provides hourly measurements with millimetric precision across 24 monitoring prisms regularly distributed along the landslide body.

Our interferograms show variable degrees of decorrelation, depending on the perpendicular baseline, atmospheric disturbances and vegetation cover. Some of them do not allow to distinguish the deformation signal from the noise and were, therefore, discarded based on visual inspection. Despite being fully aware that the displacement signals retained in the remaining selected interferograms cannot be treated as rigorous measures, we interpret such signals with the aim of extracting information about the spatio-temporal evolution of the gravitational slope movement. Interferograms were unwrapped using snaphu and combined in interferometric stacks before and after the catastrophic reactivation. During the acceleration period, unwrapping was problematic due to high spatial fringe frequency and decorrelation yet wrapped interferograms can still be used together with the change detection maps to determine where high deformation occurred.

Our results show that detectable displacements have been taking place in the upper part of the slope where the rockslide initially occurred, during the entire dormant phase, spanning from 2015 to 2024. During the same period, no clear displacement signals were detected along the landslide body nor were they detected on the ground and reported to authorities in charge of territorial management. During the catastrophic reactivation, peak displacement rates were several meters per day exceeding the detection capabilities of InSAR. However, low noise interferograms clearly delineate the spatial distribution of the active area with frequent phase jumps and decorrelation soon after the failure. During the post-failure stage, displacement rates progressively decreased and some interferograms have high enough coherence to map the deformation field. In early phases, the frequency distribution of displacement rates along the landslide approaches the theoretical upper limit of the technique (about 1 m/yr) and greatly exceed values typically measured by InSAR. When downslope-projected, displacement rates derived from our interferograms show a reasonably good agreement with ground-based data. Depending on the relative orientation between the line-of-sight and the direction of slope movement, the ascending and the descending geometry show variable capabilities to capture deformations associated to headscarp or lateral scarp retrogressions rather than the deformation of the earthflow body or toe.

The comparison between InSAR data and on-site ground measures helped us to understand and interpret the remotely sensed information and highlights the potential and the limits of dual-pass interferometry to identify and monitor active landslides. Although retrieved information cannot be treated as measures due to residual noise associated with low target coherence, InSAR provides an unbeatable areal perspective that fast accessible which adds spatial information for the interpretation of geomorphological processes and their evolution in the different stages of a landslide.

Although Poland is characterized by relatively stable geological conditions, it remains exposed to various natural hazards, including regularly occurring landslides. These phenomena constitute a significant environmental and economic problem, causing substantial material losses and posing a real threat to people and infrastructure. This situation highlights the need to implement effective monitoring tools to support efficient risk management.
Remote sensing techniques, particularly synthetic aperture radar interferometry (InSAR), offer an efficient solution for large-scale and long-term monitoring of ground deformation. An important advancement of this approach is Persistent Scatterer Interferometry (PSI), which enables the detection of millimetre-scale surface movements using stable radar targets. This study demonstrates the potential of the European Ground Motion Service (EGMS) data, based on Sentinel 1 SAR imagery and PSI processing, for analyzing landslide activity and identifying unstable slopes. The paper presents two methodological approaches for the interpretation and mathematical processing of PSI data to identify the most effective way to use InSAR for determining landslide activity: calculation of slope velocities and decomposition of LOS velocities into vertical and east–west components.
The results of one stack SAR images processing provide information on only one component of the true displacement vector, namely along the radar sensor’s line of sight (LOS), and the recorded deformation depends on the orbit geometry (ascending or descending). The relationship between slope geometry and the sensor’s viewing direction is therefore crucial when applying the PSI method in landslide analysis. Since most landslides move in the direction of the steepest slope, the LOS-measured deformation of PS points from the EGMS service was projected along the slope to estimate the actual ground displacement along the slopes.
Since landslides have complex kinematics, it is recommended to decompose LOS displacement to retrieve the vertical and horizontal motion components. LOS-derived deformation can be decomposed into vertical and east–west components in varying grid cell resolutions, taking into account radar acquisition geometry and terrain morphology. In this study, the effect of grid cell size on the decomposition of LOS deformation into east–west and vertical components, and its influence on the number of available PS points within landslide areas, was assessed.
The analyses were conducted in the Polish Carpathians, where the majority of landslides in the country are concentrated. This is due to the specific geological structure of the Carpathian massif and the dense network of tectonic discontinuities, such as faults and fractures, associated with the Alpine structure of the Carpathian orogen. The methodology was validated using landslides included in the SOPO (Landslide Counteracting System) database, based on the available data. The SOPO database is one of the largest landslide information inventories in Europe. Access to such an extensive and detailed dataset enabled high-accuracy analyses. Four landslides monitored within the SOPO project were selected for detailed investigation: Grybów, Zbyszyce, Szymbark Huciska, and Szymbark Zapadle, all located in the Carpathian region. The selection of these sites allowed for the assessment of the method’s applicability under diverse terrain conditions.

Landslides in mountainous regions represent a major hazard, and their early detection is essential for risk mitigation and infrastructure protection. Interferometric Synthetic Aperture Radar (InSAR) has become a key tool for monitoring slope dynamics due to its ability to measure ground deformation with millimetric precision, independent of weather or daylight conditions.

The shorter wavelength, such as C-band, is more sensitive to decorrelation from vegetation and snow, which limits performance in densely vegetated or seasonally variable mountain environments. In contrast longer wavelength, such as L‑band, can penetrate vegetation more effectively and maintain coherence over longer time intervals, thereby enhancing the detection of slow-moving, deep-seated landslides. Therefore, by jointly analyzing C‑ and L‑band datasets, it becomes possible to exploit the complementary strengths of each frequency: high temporal resolution from Sentinel-1 C‑band and high coherence stability from NISAR L‑band satellite.

This multi-wavelength approach expands monitoring capability across diverse land-cover conditions and increases the reliability of early-warning indicators in mountainous terrains. The results underscore the value of wavelength diversity for comprehensive landslide hazard assessment and long-term slope stability monitoring.

This study highlights the benefits of integrating multi-wavelength InSAR observation, specifically C‑band and L‑band, for improved landslide detection in complex alpine terrain. Moreover, different approaches for non-deformation phase terms mitigation in complex height variation scenarios, such as turbulent and non-turbulent atmospheric effects, are assessed in order to obtain the accurate deformation phase term.

The European Ground Motion Service (EGMS) provides ground displacement products across Europe through a standardized processing chain delivering Basic, Calibrated, and Ortho products derived from Sentinel-1 multi-temporal interferometry. This continental-scale framework ensures methodological consistency and comparability across regions; however, localized limitations may arise in rapidly evolving terrains, particularly within active and fast-moving landslide systems characterized by strong displacement gradients and temporal instability. In such environments, coherence degradation and high phase variability may significantly reduce point density, resulting in limited or absent coverage within the most actively displacing sectors. The Pissouri landslide in southern Cyprus represents a characteristic example of such conditions.

The Pissouri landslide constitutes the most active and fastest-moving landslide on the island. Complex and spatially heterogeneous kinematics generate pronounced displacement gradients and variable phase stability across the sliding body, particularly within its central sector. These characteristics make it an ideal test site for evaluating targeted InSAR reprocessing strategies aimed at extending continental-scale displacement products in highly dynamic landslide environments.

To overcome these limitations, a dedicated multi-temporal Sentinel-1 reprocessing for the 2022–2025 period was implemented, aligned with the EGMS Basic–Calibrated–Ortho product hierarchy while enhancing local validation and calibration capacity. The analysis was performed using an Interferometric Point Target Analysis (IPTA) approach. The workflow was initialized with a single-reference stack configuration to establish preliminary displacement rates and phase stability, followed by a multi-reference refinement to enhance spatial coverage, increase point density in high-gradient areas, and improve phase model consistency across the sliding body.

The processing chain continued with the generation of Basic Line-of-Sight (LOS) displacement products for both ascending and descending geometries. Atmospheric phase contributions were mitigated through spatial–temporal filtering within the multi-temporal framework to reduce long-wavelength artefacts affecting velocity estimation. An independent performance assessment of the Basic products was conducted using dedicated Corner Reflectors installed at the Souni test site, part of the CyCLOPS Strategic Research Infrastructure Unit. The reflectors enabled amplitude stability evaluation, phase referencing verification, and LOS velocity validation under controlled backscatter conditions.

Subsequently, Calibrated products were derived through integration with a localized GNSS velocity model developed within the CyCLOPS Strategic Research Infrastructure Unit. The CyCLOPS network comprises ten (10) continuously operating GNSS stations installed both inside and outside the active landslide area, providing stable geodetic reference conditions and independent velocity estimates. GNSS-based calibration was applied to align InSAR-derived LoS velocities with the local geodetic reference frame, correcting residual velocity offsets and improving georeferencing consistency across the sliding body. Following calibration, ascending and descending LoS datasets were combined to generate Ortho products through geometric decomposition, yielding East–West and vertical displacement components. These products were integrated into the Cyprus Ground Motion Service (CyGMS) as high-resolution displacement layers georeferenced to the ITRF2020 realization through GNSS-based frame alignment. The incorporation of locally calibrated Ortho components within CyGMS ensures geodetic consistency at national scale, complementing continental-scale EGMS outputs while enhancing spatial representation in dynamically active zones.

The updated 2022–2025 analysis reveals sustained high-gradient displacement within the central sliding mass, with mean annual velocities approaching 80 mm/year and cumulative displacements exceeding 40 cm over the observation period. Relative to continental-scale EGMS outputs, the reprocessed dataset improves spatial density and displacement representation within the most dynamically active areas, demonstrating the capacity of locally enhanced processing to resolve displacement gradients that remain underrepresented in continental-scale products.

The presented methodology demonstrates that localized, multi-sensor InSAR reprocessing supported by the CyCLOPS Strategic Research Infrastructure Unit can effectively extend continental-scale displacement services in rapidly evolving geohazard environments. Rather than replacing EGMS products, the approach complements them through high-resolution enhancement layers operationally integrated within CyGMS, illustrating a structured transition from research-grade processing to service-oriented national ground displacement monitoring. The same reprocessing framework is currently being applied to additional areas in Cyprus exhibiting similar displacement behaviour, demonstrating its reproducibility and scalability within the national ground motion service context.

ACKNOWLEDGMENTS

The authors would like to acknowledge the ‘CyCLOPS+’ (RIF/SMALL SCALE INFRASTRUCTURES/1222/0082) project, which is funded by the European Regional and Development Fund and the Republic of Cyprus through the Research and Innovation Foundation.

The authors would like to acknowledge the ‘CyCLOPS’(RIF/INFRASTRUCTURES/1216/0050) project (www.cyclops.cy), which is funded by the European Regional and Development Fund and the Republic of Cyprus through the Research and Innovation Foundation in the framework of the RESTART 2016-2020 programme.

The authors also acknowledge the ‘EXCELSIOR’: ERATOSTHENES: EΧcellence Research Centre for Earth Surveillance and Space-Based Monitoring of the Environment H2020 Widespread Teaming project (www.excelsior2020.eu). The ‘EXCELSIOR’ project has received funding from the European Union’s Horizon 2020 research and innovation programme under Grant Agreement No 857510, from the Government of the Republic of Cyprus through the Directorate General for the European Programmes, Coordination and Development and the Cyprus University of Technology.

The aim of the analyses is to assess the usefulness of satellite data from the European Ground Motion Service (EGMS), collected under the European Copernicus programme and based on Persistent Scatterer Interferometry (PSI), for analyzing manifestations of landslide activity in the Carpathian region. Observation and monitoring of landslides in this area constitute a significant organizational and financial challenge for local authorities.The study used EGMS Level L2B products as well as data from the SOPO (Counteracting Landslide System of Poland) database, including results from instrumental monitoring. As part of the work, Persistent Scatterer (PS) points were selected and subjected to geostatistical analyses. Subsequently, the ADAtools software was used to identify areas of active deformation.The results indicate that despite limitations related to forest cover and radar signal coherence loss, a considerable number of PS points from EGMS data are located within delineated landslides, enabling the analysis of deformation velocities and the preliminary inventory of landslide process activity.

Interferometric Synthetic Aperture Radar (InSAR) has become a well-established technique for monitoring slow-moving landslides, particularly in mountainous regions where in-situ measurements are challenging due to limited accessibility and harsh terrain. The Sentinel-1 mission, with its 12-day repeat acquisition cycle and open-access data policy, provides a reliable and cost-effective resource for continuous surface deformation monitoring at regional scales.

This study evaluates the applicability of Sentinel-1 data for landslide monitoring of the Mayoon landslide, located in Mayoon village within the Hunza-Nagar Valley, northern Pakistan. The landslide has evolved gradually over time but has shown signs of accelerated activity in recent years, raising significant concern. The area is highly vulnerable: approximately 120 families (∼1000 inhabitants) reside at the toe of the slope. Furthermore, the landslide is situated in close proximity to the Karakoram Highway, a strategic transportation corridor linked to the China–Pakistan Economic Corridor (CPEC).

The geomorphological setting further amplifies the associated hazard. The landslide mass is adjacent to a river corridor, and a potential large-scale slope failure could obstruct the river flow, forming a temporary landslide dam. Such an event could trigger catastrophic upstream flooding and downstream outburst floods, posing severe risks to local communities and critical infrastructure.

In this study, we investigated the Sentinel-1 InSAR time series using the Small Baseline Subset (SBAS) approach to retrieve surface displacement over the Mayoon landslide. The landslide exhibits an east-southeast facing slope, resulting in moderate geometric sensitivity with respect to the Sentinel-1 line-of-sight (LOS) configuration. Combined with generally good temporal mean coherence across large portions of the area of interest, these conditions make the site both challenging and suitable for evaluating the performance and limitations of Sentinel-1-based multi-temporal InSAR analysis in mountainous terrain.

The dataset spans the period 2023–2024 and comprises 59 interferograms generated using the Alaska Satellite Facility (ASF) Hybrid Pluggable Processing Pipeline (HyP3) on-demand service. To enhance phase stability, multilooking with a factor of 20 × 4 was applied prior to time-series inversion. The HyP3-generated interferograms were subsequently processed using the MintPy framework to perform SBAS time-series inversion, from which line-of-sight (LOS) displacement time series and mean velocity fields were derived.

The estimated mean LOS velocity of approximately −3.9 mm/year (negative values indicating motion away from the satellite) indicates ongoing active deformation within the area of interest. To further investigate the internal kinematics of the landslide body, transect-based time-series analyses were conducted by selecting three representative points across different sectors of the landslide. The results reveal spatially heterogeneous displacement behavior, with distinct sectors exhibiting varying displacement magnitudes and temporal evolution relative to the reference acquisition date, indicating non-uniform kinematic patterns across the slope.

A key methodological challenge encountered in this study relates to reference point selection for time-series inversion. The valley floor is characterized by vegetation cover that reduces interferometric coherence, particularly after multilooking, while persistently coherent zones are primarily located along upper slope and ridge areas that may themselves be deforming. Reference selection within such regions can introduce bias in displacement estimates and ambiguity in inferred deformation rates.

Landslides represent some of the most destructive natural hazards encountered in unstable mountainous regions, such as the Western Ghats in India. The examination of landslides has garnered significant global attention due to their profound impacts on socio-economic activities. The utilization of remote sensing and geographic information systems has proven valuable for integrating the spatial factors that contribute to landslide occurrences. In this study, satellite imagery from Sentinel 1-C Band has been employed, and further interferometry techniques for detecting the landslide event in 2024 in the Wayanad district, Kerala. Leveraging Artificial Intelligence Techniques in RADAR remote sensing such as machine learning algorithms, particularly the Random Forest (RF) model, were utilized to classify the study area into affected and non-affected regions. The findings also indicate the affected land use and land cover in the given study area. In the end, it can be concluded that significant landslides on 30th July, 2024 in the Wayanad district were primarily precipitated by anthropogenic interventions, compounded by heavy precipitation and unstable topography. Activities such as stone quarrying and infrastructure development emerged as critical factors contributing to these landslides. This research provides valuable insights aimed at mitigating landslide hazards in the Wayanad district, thereby fostering sustainable development.

Mountain regions are highly sensitive to climatic change, and the effects of changing environmental conditions can destabilise rock and soil slopes. This can lead to increased hazards to people and infrastructure as it can result in accelerations of slow-moving landslides and reactivations of dormant landslides. As these changes occur over large geographic areas, regional-scale monitoring of slope dynamics is becoming essential for both hazard assessment and for understanding ongoing landscape change. Freely available ESA Sentinel-1 data enables wide-area surface displacement monitoring with frequent revisit times. This data is typically processed using PS-InSAR techniques, which focus on stable backscatterers. However, direct analysis of wrapped interferograms allows additional information to be extracted with substantially less processing. It also offers analysis ready data from single interferograms instead of requiring long time series, and avoids the need to blindly mask areas affected by phase aliasing. However, despite the abundance of available data, interpreting interferograms remains labour-intensive while requiring expert knowledge. Deep learning based semantic segmentation algorithms have recently emerged as a promising tool to enable scalable exploitation of these datasets. However, segmentation models are often trained and or validated on labels created using optical imagery, rarely account for task-specific uncertainties and inter-operator variability inherent to interferometric interpretation, and currently still focus on monitoring known mass movements rather than also performing detection.

We address these gaps by comparing the performance of state-of-the-art semantic segmentation algorithms with the range of expert variability for mapping mass movement-related phase patterns. We processed Sentinel-1 SLC data over Valais, Switzerland (5600 km²), from tracks 088 (ascending) and 066 (descending) using 12 – 18 day temporal baselines during summer. We generated D-InSAR interferograms with Goldstein filtering and topography-based atmospheric correction. Expert delineations of ten selected case studies revealed low agreement (Intersection over Union (IoU): 0.21 – 0.41), highlighting the difficulty of consistently distinguishing coherent and incoherent phase patterns and separating deformation signals from noise. A custom training dataset (>1000 labels; velocity range ~15 – 210 cm/a) was created from four interferograms and used to evaluate different model configurations. The best-performing model, a U-Net++ with a ResNet-18 encoder, achieved IoU 0.531 ± 0.029 (Dice 0.693 ± 0.024) on positive-only samples and slightly lower performance (IoU 0.494 ± 0.045; Dice 0.661 ± 0.041) on mixed datasets. Comparison with expert mappings showed deep learning model performance within inter-expert variability (mean IoU 0.494 ± 0.045), with an IoU of 0.61 relative to training labels.

Once trained, the model segments interferograms in seconds, whereas experts required more than three hours for the twelve selected case examples. These results show that deep learning can reduce manual mapping effort by guiding attention to relevant signals within the vast amounts of available spaceborne radar interferometry data, supporting expert-driven mass movement cataloguing, and facilitating future large-scale detection and monitoring applications.

Paddy rice is essential for global food security, yet its accurate monitoring in tropical Asia remains challenging due to complex multi-season cultivation patterns and persistent cloud cover. Previous studies have achieved promising results on rice mapping using time series of Sentinel-1 data, but the multi-season cases have been rarely considered, and the evolution of different advanced SAR-derived features has not been fully investigated. In this study, we analyse the interferometric and polarimetric signals displayed by single- and double-season tropical rice fields, with the intention of finding useful information to indicate key phenological signals under tropical agronomy scenarios.

The study area is located in Hainan Province, China, whose agricultural landscape exhibits typical characteristics of tropical Asia: small in size and irregular in shape. Time series of backscattering coefficients (σ_VH, σ_VV) and Radar Vegetation Index (RVI) from February to November 2025 were obtained from Google Earth Engine. In addition, during the periods from 8^th February to 3^rd May, and 7^th August to 29^th November, 6-day repeated observations were also provided by the newly operational Sentinel-1C alongside Sentinel-1A. The 12-day coherence over the full period and the 6-day coherence from August to November were extracted to evaluate the sensitivity of interferometric coherence and decorrelation to the phenological evolution of rice. Moreover, two dual-polarimetric decompositions (the dual-polarimetric H-alpha decomposition and a model-based dual-polarimetric decomposition) were applied to inspect the evolution of scattering mechanisms for single- and double-season tropical rice.

Coherence is highly sensitive to ground variations and is usually low in paddy fields. Our preliminary experimental results indicate that the 12-day coherence of single-season rice in tropical Hainan displayed similar temporal trends to those of temperate climate zones. For both VH and VV polarizations, the 12-day coherence values were significantly higher in the field preparation and post-harvest stages than in the growth stage, and the coherence values of VH were lower compared to VV. The coherence of VV may exceed 0.5 during the start-of-season (SOS), then drops rapidly to a level around 0.2. This suggests that VV coherence may constitute a potential indicator for rice cultivation, especially under the circumstances where the flooding signals of early rice are not as significant as in temperate regions.

For the double-season rice, in contrast, the 12-day coherence of VH was always low from the beginning of the season. For the VV polarization, some parcels exhibited slightly higher coherence in the SOS, but the difference compared to the growing stage was not as distinctive as in single-season rice. These results were reasonable because of a shorter land-preparation period for the rush of early-rice sowing, and thus a faster coherence loss. However, it was found that for some parcels, the coherence of VV rose in July or August, which could be attributed to the compound effects of grain drying, harvest, and straw residues of early-season rice.

Furthermore, the 6-day coherence from August to November was inspected. Even though rapid biomass accumulation still leads to significant decorrelation, the 6-day coherence values were significantly higher than those of the 12-day baseline for both single- and double-season rice. The harvest signal of single-season rice was clearly reflected by the 6-day VV coherence, and for some parcels even in the VH polarization, where 6-day coherence exceeded 0.4 at the end-of-season (EOS).

Regarding polarimetric features, since the dual-polarimetric covariance matrix does not provide a full picture of scattering mechanisms, these features did not provide superior performance as scattering mechanism indicators to backscattering intensities. Our results revealed that dual-polarimetric alpha (α), entropy (H), and the volume scattering component (P_v) from model-based decomposition all followed trajectories similar to and RVI. For single-season rice, these curves reached troughs during the flooding stage, then started to increase as crop biomass accumulated through the vegetative stage. During the maturation stage, mainly from late August to September, the curves declined at first, and then returned to a second peak rapidly, which can be easily confused with the actual harvest signal occurred subsequently in October. For the double-season parcels, the harvest timepoint of the first season rice was more difficult to identify because it was closely followed by the sowing of the second season. Even if the land still went through preparation, the scattering intensities might not decline significantly and could return rapidly to levels similar to July. Consequently, the harvest and re-sowing signals of rice were not clearly identified, and the backscattering intensity just fluctuated around a ‘stable’ level, which could lead to an overestimation of the growth period, and thus a misclassification of the double-rice as other crops.

This research aims to supplement theoretical studies on multi-season rice monitoring in fragmented tropical agronomy landscapes. Our preliminary results indicate that polarimetric features reflected similar information to backscattering, whereas the timing and magnitude of coherence-loss might provide supplemental information for the identifications of the SOS or EOS.

We assume that with a denser time-series, the synergy of intensity and interferometric information (for instance, the timing, degree, and rate of decorrelation) might improve the discrimination of early- and late-season rice. This requires a stable coherence pattern of rice, which should be attributed to real biophysical changes and reduce the influence of the environmental factors such as tropical precipitation, which will be considered in our future works.

Interferometric Synthetic Aperture Radar (InSAR) time-series analysis is a recognized remote sensing method for measuring surface deformation associated with volcanic activity and landslides. While the Small Baseline Subset (SBAS) algorithm (Usai et al. 2001) has become a standard technique for estimating millimetric surface motion, its reliance on consistent viewing geometries restricts analysis to single-sensor datasets. Consequently, observation frequency is strictly constrained by the satellite's revisit cycle. High temporal sampling is particularly critical for disaster monitoring, where sparse sampling can lead to the aliasing or complete omission of rapidly evolving precursors such as volcanic inflation or landslide acceleration. While the recent rise of small satellite constellations aims to address this temporal gap by reducing revisit times, these systems predominantly operate in X-band. Although effective for urban infrastructure, the short wavelength of X-band radar interacts primarily with the canopy surface, limiting coherence preservation in vegetated environments due to wind-induced motion or dielectric changes. Sentinel-1 (C-band), the current global standard, generally maintains better coherence than X-band. However, it still exhibits instability in the disaster-prone regions of Japan and Southeast Asia. Here, dense vegetation and seasonal snow cover frequently cause severe temporal decorrelation, resulting in data deficit exactly where monitoring is most critical. On the other hand, L-band missions (e.g., ALOS-4/NISAR) utilize a longer wavelength that penetrates vegetation to interact with stable ground scatterers, demonstrating superior coherence preservation. Yet, L-band systems are currently limited by a smaller number of operational satellites and generally lower observation frequencies, making it difficult to construct dense InSAR time series independently. These complementary characteristics motivate a framework that can combine temporally frequent yet spatially fragmented C-band observations with a more coherence-robust L-band backbone in a principled time-series inversion.

The integration of such diverse datasets is formalized by the Multidimensional SBAS (MSBAS) framework (Samsonov & d’Oreye. 2012, Samsonov. 2024), which fuses observations onto a unified chronological timeline by estimating deformation velocities via singular value decomposition (SVD). However, conventional MSBAS processing typically relies on a common interferogram network designed for the full scene. This global selection creates a limitation in heterogeneous environments: locally valid measurements are often discarded if the overall scene coherence is low. Consequently, the inversion is forced to rely heavily on mathematical regularization (interpolation) rather than actual observational data to bridge gaps between accepted scenes. Independently, within the domain of single-sensor InSAR time-series analysis, Pixel-by-Pixel (PBP) approaches emerged to address similar coverage limitations. For instance, the Intermittent SBAS (ISBAS) method (Sowter et al., 2013) mitigates this by relaxing strict persistence requirements to exploit observations that are only intermittently coherent. By evaluating coherence on a layer-by-layer basis rather than enforcing a stack-wide threshold, ISBAS unwraps valid pixels individually and estimates velocities for points exceeding a minimum number of coherent layers. Similarly, Ishitsuka et al. 2016 introduced a pixel-based interferometric pair selection algorithm to tailor network connectivity based on local coherence statistics, further demonstrating the efficacy of spatially adaptive processing.

In this study, we present Multidimensional-Intermittent SBAS (MI-SBAS), a framework that integrates the multi-sensor data fusion capabilities of MSBAS with the pixel-wise network selection strategy. Unlike standard techniques that enforce a globally fixed network, MI-SBAS dynamically constructs a unique set of valid interferograms for every pixel based on local coherence thresholds. This design aims to maximize the density of actual observations integrated into the inversion. Specifically, it allows for: (i) the retention of locally valid acquisitions rejected by global criteria; and (ii) the minimization of temporal gaps that require regularization, even when multi-sensor observations are sparse and misaligned. The inversion employs an SVD-based solver to robustly handle rank deficiencies, ensuring that the retrieved time series is constrained by physical measurements rather than interpolation wherever possible.

We applied MI-SBAS to two Japanese volcanoes: Sakurajima (2015–2016) and Mt. Fuji (2021–2022). All interferograms were generated using the standard GMTSAR pipeline at a resolution of 50m, with a first-order polynomial fit applied to remove long-wavelength atmospheric and ionospheric phase delays. The Sakurajima dataset consisted of 34 Sentinel-1 scenes (405 interferograms) and 23 ALOS-2 scenes (252 interferograms). For Mt. Fuji, we utilized 60 Sentinel-1 scenes (989 interferograms) and 16 ALOS-2 scenes (105 interferograms), covering diverse surface conditions. GNSS observation data for validation were obtained from GNSS earth observation network system (GEONET) and Japan volcanological data network (JVDN). All computational processing benchmarks were conducted using a single core on an Apple M4 Pro Mac mini (14-core CPU, 64GB RAM).

The results for Sakurajima demonstrate that MI-SBAS enhanced the density of input interferograms by recovering pairs that were excluded by the global coherence criteria of the common-network baseline. This led to an increase in the number of acquisition dates integrated into the inversion by up to 20 days (1.6 days on average). Although computational runtime increased from 119 s to 158 s, the retrieval quality improved: the median time-series uncertainty (temporal standard deviation) was reduced from 1.82 mm/yr to 1.69 mm/yr, and spatial smoothness (spatial standard deviation) improved from 17.3 mm/yr to 10.9 mm/yr. Validation against four GNSS stations indicates that the pixel-wise selection maintains comparable accuracy overall; for example, RMSE at station 960720 improved (29.3 mm to 22.1 mm), whereas station J884 showed a degradation (12.6 mm to 12.9 mm).

Similarly, for Mt. Fuji, the number of valid dates in the input network expanded by up to 39 days (10 days on average) due to the inclusion of locally coherent interferograms. While the runtime increased from 377 s to 912 s, the method achieved a reduction in the median time-series uncertainty (from 3.74 mm/yr to 3.45 mm/yr) and better spatial smoothness (from 86.1 mm/yr to 58.6 mm/yr). GNSS validation across five stations showed substantial improvements in challenging areas, despite some degradation elsewhere; for example, the maximum RMSE reduction was recorded at station 021100 (126.2 mm to 55.2 mm), whereas station 93070 exhibited the largest degradation (20.3 mm to 32.4 mm). These metrics indicate that while the pixel-wise selection effectively maximizes observation density and recovers signals in difficult environments, it presents a trade-off with increased noise sensitivity in some areas.

In conclusion, our results highlight the advantage of MI-SBAS. Whereas conventional methods that rely on regularization for apparent time-series continuity, our framework maximizes the density of direct observations. By prioritizing local signal quality and avoiding interpolation, MI-SBAS ensures that the retrieved time series are derived from coherent phase information. However, this approach presents specific limitations: (1) the output time-series coverage may decrease in some pixels where gaps are not interpolated, and (2) the increased volume of input data used for velocity estimation can introduce noise, potentially degrading accuracy in certain scenarios. Despite these trade-offs, the framework offers significant advantages in yielding spatially smoother displacement fields and higher effective temporal sampling frequencies. Although the computational overhead increases, the total processing time remains well within practical limits for operational monitoring. Future work will focus on integrating robust noise-aware weighting schemes to maximize the spatiotemporal coverage of the retrieved time series while maintaining consistent geodetic accuracy.

Every year, millions of metric tons of plastic waste enter the oceans from land-based sources, and projections suggest that annual inputs will continue to rise substantially (Borelle et al., 2020). Under the influence of converging currents, this debris accumulates into plastic islands floating on the surface of oceans, seas, and rivers. These accumulations represent a serious threat to both marine ecosystems and human health. They endanger aquatic organisms through entanglement and ingestion, disrupt marine habitats, and contaminate the food chain, ultimately affecting human consumers. In addition to ecological impacts, large floating plastic islands can also interfere with maritime activities and pose risks to vessels and navigation. This study presents preliminary results from two experiments that aim to evaluate the capability of Synthetic Aperture Radar (SAR) satellite data to identify plastic marine litter. To address this objective, a small artificial floating plastic island was constructed using various plastic items, including water bottles, detergent bottles, boxes, and plastic bags. These materials were arranged along multiple ropes to reproduce as closely as possible the configuration typically observed in natural floating plastic litter, which often organizes into long, narrow stripes. The assembled structure was deployed at two separate sites to test its detectability under different environmental conditions. The first experimental campaign was carried out in October 2024 at Lake Massaciuccoli, near Viareggio in the Tuscany region. The second campaign took place in April 2025 in a marine environment near Portovenere, within the Gulf of La Spezia in the Liguria region. During both experiments, high-resolution X-band SAR imagery was acquired by the Cosmo-SkyMed and ICEYE satellite missions, which feature varying viewing geometries, polarizations, and spatial resolutions, to assess their influence on the visibility of the floating plastic target. The results indicate that the artificial plastic island is visible in SAR scenes as a thin linear feature spanning only a few pixels, characterized by backscattering values slightly higher than those of the surrounding water surface. This behavior is consistent with known scattering mechanisms: floating debris introduces localized surface roughness, enhancing diffuse backscatter toward the sensor, whereas calm water surfaces act predominantly as specular reflectors, redirecting most of the incident radar energy away from the sensor. Furthermore, the comparison among different SAR acquisition configurations enabled an initial assessment of how key parameters—such as type of polarization and incidence angle—affect the visibility of floating plastics. The statistical analyses conducted on these preliminary datasets provide valuable indications for identifying the most suitable SAR configurations, serving as a preparatory step toward the generation of change detection maps specifically designed to monitor the evolution of floating plastic accumulations. While these findings provide encouraging evidence of the feasibility of SAR-based monitoring of plastic marine litter, additional investigations are required to better understand the role of each acquisition parameter and to distinguish the radar signature of plastic materials from other floating elements such as wood, algae, foam, and similar debris.

This study was carried out within the Space It Up project funded by the Italian Space Agency, ASI, and the Ministry of University and Research, MUR, under contract n. 2024-5-E.0 - CUP n. I53D24000060005."

REFERENCES

Borelle, S.B.; Ringma, J.; Law, K.L.; Monnahan, C.C.; Lebreton, L. (2020). Predicted Growth in Plastic Waste Exceeds Efforts to Mitigate Plastic Pollution. Science 2020, 369, 1515–1518.

Vegetation-Specific Correction for Improved Soil Moisture Estimation Using NavIC-IR Multipath Phase Analysis Integrated with CORS GNSS Network Framework

Soil moisture is a fundamental variable in the Earth’s hydrological cycle and plays a decisive role in agricultural productivity, evapotranspiration, groundwater recharge, and land–atmosphere interactions. Accurate and continuous monitoring of soil moisture is essential for precision agriculture, drought assessment, irrigation planning, and climate modeling. Traditional in-situ methods provide high accuracy but limited spatial coverage, while satellite missions such as SMAP and SMOS offer broader coverage at relatively coarse spatial and temporal resolutions. Ground-based GNSS-Interferometric Reflectometry (GNSS-IR) provides a promising intermediate solution by enabling continuous, high-temporal-resolution monitoring using existing GNSS infrastructure.

This study presents a novel vegetation-specific correction methodology for soil moisture estimation using multipath phase observations derived from Navigation with Indian Constellation (NavIC) signals. The approach leverages the sensitivity of multipath interference patterns to near-surface dielectric variations, which are directly related to volumetric soil moisture content (VMC). Unlike conventional GNSS-IR studies primarily focused on GPS, this work utilizes NavIC L-band signals, whose longer wavelength enhances penetration capability through vegetation and improves robustness in cropped agricultural fields.

Theoretical Background

GNSS-IR relies on the interference between direct satellite signals and signals reflected from the ground surface. When a geodetic right-hand circularly polarized (RHCP) antenna is mounted above the ground, the reflected signal experiences a path delay relative to the direct signal. This path difference produces a sinusoidal modulation in the Carrier-to-Noise ratio (C/No).

The multipath phase is particularly sensitive to changes in surface dielectric properties, making it a strong indicator of soil moisture variations. However, vegetation growth modifies the reflective properties of the surface, introducing additional phase shifts and reducing estimation reliability if not properly compensated.

NavIC Signal Characteristics

NavIC satellites provide elevation angle coverage between approximately 15° and 30° in the usable GNSS-IR range. Within this low-elevation window, multipath effects are pronounced and suitable for soil moisture analysis. As elevation increases, direct signals dominate, diminishing multipath sensitivity. Therefore, this study restricts analysis to the optimal low-elevation regime.

Vegetation-Specific Phase Compensation

The novelty of this work lies in explicitly accounting for vegetation growth effects using NDVI (Normalized Difference Vegetation Index) as a vegetation proxy. The dataset was divided into two vegetation groups:

• Group 1: NDVI < 0.20 (early growth stage)
• Group 2: 0.20 ≤ NDVI < 0.35 (moderate vegetation stage)

Analysis revealed systematic phase shifts between these groups. The generalized multipath phase was expressed as:

where:

• = soil moisture contribution,
• = vegetation-induced phase component,
• = direct signal phase.

To compensate for vegetation effects, normalization was applied:

This normalization mitigates vegetation-induced bias while preserving soil moisture sensitivity.

Machine Learning-Based Soil Moisture Retrieval

A Random Forest Regressor was implemented to model nonlinear relationships between normalized multipath phase and measured soil moisture (VMC). The dataset was split 80:20 into training and testing subsets. Hyperparameter optimization using Grid Search Cross-Validation identified the optimal configuration:

• n_estimators = 150
• max_depth = None
• min_samples_split = 5
• min_samples_leaf = 1

The optimized model achieved a Mean Squared Error (MSE) of 1.63%, demonstrating high predictive capability and strong generalization performance across vegetation stages.

Scatter plot analysis between predicted and true VMC values confirmed close alignment along the ideal 1:1 regression line, validating the robustness of vegetation-compensated multipath phase modeling.

Integration with CORS GNSS Network

A significant advancement of this work is its potential integration within a Continuously Operating Reference Station (CORS) GNSS network. CORS infrastructure, typically deployed for geodetic positioning and crustal deformation monitoring, can serve as a cost-effective environmental sensing network when multipath observables are systematically analyzed.

In a CORS-based deployment:

1. Existing geodetic-grade antennas can be used without additional instrumentation.
2. Continuous multi-year datasets enable seasonal and inter-annual soil moisture monitoring.
3. Large-scale spatial coverage can be achieved by integrating multiple CORS stations across agricultural regions.
4. Real-time data streams can support near-real-time soil moisture monitoring.

The National Centre for Geodesy (NCG) CORS network framework provides an ideal testbed for scaling this methodology. By embedding multipath phase processing algorithms into CORS data pipelines, soil moisture retrieval can become an additional value-added service of national geodetic infrastructure.

Such integration transforms CORS networks from purely positioning infrastructure into multipurpose environmental observatories, supporting:

• Precision agriculture
• Drought early warning
• Irrigation optimization
• Land–atmosphere coupling studies
• Climate resilience modeling

Furthermore, NavIC-based implementation strengthens national capability by leveraging indigenous satellite infrastructure for agricultural monitoring applications.

Conclusion

This study introduces a vegetation-aware multipath phase correction methodology for improved soil moisture estimation using NavIC-IR signals. By categorizing data according to vegetation growth stages and applying group-specific normalization, the model significantly enhances estimation reliability under dynamic crop conditions.

The optimized Random Forest regression framework achieves high prediction accuracy (MSE = 1.63%), demonstrating that vegetation-compensated multipath phase is a robust proxy for soil moisture retrieval.

Importantly, the approach can be operationalized using existing CORS GNSS networks, enabling scalable, continuous, and cost-effective soil moisture monitoring without additional sensor deployment. This integration establishes a powerful synergy between geodetic infrastructure and environmental remote sensing, contributing to sustainable agriculture and climate-adaptive resource management.

Forest biodiversity comprises both vertical structural heterogeneity and canopy compositional variability, which arise from distinct physical processes and are differentially observable across remote sensing domains. Long-wavelength P-band SAR is sensitive to volume scattering and canopy vertical organization, whereas optical hyperspectral and multispectral data primarily capture biochemical and mixture-dependent reflectance variability. This study investigates whether these domains provide interchangeable or complementary information for forest biodiversity characterization in the Taita Hills, encompassing montane cloud forests and lowland savanna woodlands. 

Four hypotheses are tested: (H1) P-band PolSAR and PolInSAR observables are associated with independently measured vertical structural complexity; (H2) optical spectral diversity metrics are associated with in-situ compositional diversity; (H3) structural and spectral predictors explain partially independent variance in biodiversity metrics; and (H4) combined models outperform single-domain models. 

Polarimetric (PolSAR) and interferometric (PolInSAR) metrics are derived from multiple acquisitions of the BIOMASS mission, including entropy, anisotropy, cross-polarized backscatter, and coherence-based vertical structure proxies. Airborne LiDAR point clouds provide independent structural reference metrics, while forestry inventory plots containing species composition, DBH, and tree height measurements provide compositional and structural diversity indices. Spectral diversity metrics are derived from PRISMA hyperspectral and Sentinel-2 multispectral surface reflectance data. 

Random Forest regression with spatial block cross-validation is used to evaluate structural and compositional relationships and to quantify the complementarity of radar and optical predictors. The study clarifies the extent to which P-band interferometric observables represent vertical diversity and examines whether structural and spectral domains provide redundant or complementary information for biodiversity inference. 

Forests are essential for global climate regulation, carbon storage, and biodiversity preservation, making rapid disturbance monitoring a critical priority. However, traditional optical satellite sensors are often limited by cloud and smoke cover during active disturbance events. This project addresses this gap by leveraging Interferometric SAR (InSAR) coherence to provide a robust, weather-independent monitoring framework. The study specifically focuses on the 2022 forest fire in the Bohemian-Saxon Switzerland region, which represents the largest fire event in modern Czech history. This event devastated approximately 1,300 hectares of forest between July and August 2022, highlighting the urgent need for advanced remote sensing tools that can penetrate atmospheric obstacles to map biomass loss in real-time.

At its scientific core, InSAR coherence serves as a measure of the geometric and dielectric stability of the Earth's surface over time. In the context of forest monitoring, healthy and structurally complex forests typically maintain permanently low coherence values. This is primarily due to volume scattering, where microwave radiation penetrates the canopy and interacts with thousands of randomly oriented scatters such as leaves, needles, and small branches. These interactions cause a random summation of phases, leading to phase decorrelation. However, when a fire removes the forest canopy, the scattering process changes significantly, often resulting in increased coherence values as the remaining stable structures or the ground surface become the dominant reflectors.

The methodology utilized in this research involved the analysis of an extensive time series from 2019 to 2025, consisting of 206 interferometric pairs from the Sentinel-1A satellite with a 12-day temporal baseline. The data processing was automated through a pipeline using SNAP GPT and PowerShell. The final output products achieved a spatial resolution of 30 meters per pixel. To isolate the specific signal caused by the fire from environmental noise, the study employed a Generalized Least Squares (GLS) statistical framework. This model was used for adjusting for various decorrelation sources, including the perpendicular baseline of satellite orbits and meteorological variables such as wind speed, precipitation, temperature, and snow cover.

The results of the statistical modeling confirm that the 2022 fire had a significant impact on coherence levels within the affected areas. While the reference healthy forest showed no statistically significant change during the fire period, the burned areas exhibited a significant increase in coherence with an estimate of 0.045. This increase effectively marks the transition from a complex volume-scattering environment to a more stable, disturbed surface. Furthermore, the research identified that meteorological factors, particularly temperature and precipitation, were significant contributors to temporal decorrelation.

In conclusion, the application of Sentinel-1 InSAR coherence provides an indicator for identifying biomass loss and forest disturbances. Its ability to provide consistent data regardless of smoke or cloud cover offers a distinct advantage over optical sensors during active forest fires. However, the study also notes that challenges remain, particularly regarding the influence of terrain topography and other geometric uncertainties in complex landscapes.

Landfilling is widely recognised as one of the most environmentally harmful waste disposal methods. However, in many developing countries it is still commonly used because it is relatively low-cost. The key challenge is therefore how to identify acceptable landfill locations while taking socio-economic conditions into account.
In our case study area, waste management is under increasing pressure. Rapid economic growth and urbanisation have led to a continuous rise in both the volume and complexity of solid waste. At the same time, existing landfill infrastructure is approaching its environmental capacity. Under these conditions, the identification of new landfill sites becomes essential for sustainable urban development and public health protection.
This study applies a hybrid approach that combines remote sensing (including SAR data) with GIS-based Multi-Criteria Decision Analysis (GIS-MCDA) to assess landfill suitability. SAR information was used to characterise surface conditions and terrain constraints relevant to environmental risk assessment. Economic, environmental, topographic and social factors are evaluated together to support spatial decision-making. The results indicate that only 45% of the study area can be classified as highly or very highly suitable. Nevertheless, 19.78% of these areas still present notable environmental and human health risks. The proposed framework can also be adapted for other low- and middle-income countries facing similar municipal solid waste management challenges.

The increasing frequency and intensity of natural hazards, exacerbated by climate change and progressive urbanization, have amplified the need for robust, scalable, and financially interpretable risk assessment frameworks. In Europe, economic losses due to weather- and climate-related extremes have grown significantly over the past decades, while in Italy alone seismic recovery costs have reached hundreds of billions of euros in the last sixty years. These figures underline the urgency of transitioning from reactive post-disaster reconstruction strategies to proactive, data-driven risk mitigation and financial planning approaches. Within this context, the VALUESAFE framework (Vulnerability of Assets and Losses in mUltirisk Evaluations SAtellite data for Financial Estimation) proposes an integrated, multi-hazard and multi-scale platform designed to support both public authorities and private stakeholders in urban and territorial risk assessment, damage estimation, and asset depreciation forecasting.

Current state-of-the-art methodologies for disaster risk reduction present several structural limitations. Data fragmentation across heterogeneous repositories, inconsistencies in spatial resolution, and the frequent separation between qualitative resilience indicators and quantitative economic loss modeling reduce comparability and operational usability. Indicator-based resilience frameworks offer conceptual flexibility but often lack direct translation into monetary loss. Conversely, engineering-based vulnerability assessments, although physically grounded, are resource-intensive and difficult to scale across large building stocks. These gaps motivate the development of standardized, automated, and interoperable platforms capable of bridging territorial screening with asset-level diagnostics.

VALUESAFE addresses these limitations through a Software-as-a-Service (SaaS) Web-GIS architecture structured across three progressive levels of analysis - Explore, Insight, and Detail - each corresponding to increasing spatial resolution, data granularity, and computational depth. The framework is designed to ensure methodological coherence while maintaining scalability and adaptability to different stakeholder needs, ranging from preliminary territorial prioritization to facility-specific financial diagnostics.

At the Explore level, the system performs automated large-area screening based primarily on existing global and national hazard inventories, census data, open geospatial building databases, and cadastral or market-based real estate quotations. Outputs are generated on a grid-based territorial scale and provide integrated risk classifications derived from the additive combination of hazard, vulnerability, and exposure indicators. Although spatially generalized, this tier enables rapid identification of macro-scale criticalities, particularly useful for regional planning authorities, portfolio managers, and preliminary investment screening.

The Insight level refines the analysis at building or urban-cluster scale by integrating dynamic satellite-derived measurements and AI-driven image processing techniques. In particular, A-DInSAR time-series derived from Sentinel-1 data are incorporated to detect ground displacement patterns with millimetric precision, improving hazard characterization for landslide, subsidence, and ground instability phenomena. AI-based image segmentation algorithms extract building footprints, façade opening ratios, geometric attributes, and approximate construction age when not available from official datasets. This enrichment allows the calculation of hazard-specific vulnerability indices incorporating parameters such as building height, number of floors, structural typology, material class, and window-to-wall ratio. For seismic risk, fragility curves and EMS-98 classifications are used to estimate expected damage levels and condemnation probabilities. Flood vulnerability assessments integrate building height and hydraulic hazard layers to approximate inundation exposure. The resulting outputs include quantitative risk scores, expected damage scenarios, and estimated economic depreciation expressed in absolute and percentage terms, enabling operational decision-making for insurance providers, municipalities, and asset managers.

The Detail level focuses on facility-specific diagnostics, leveraging high-resolution satellite data such as COSMO-SkyMed imagery to detect differential deformation at sub-building scale. When available, on-site surveys and user-provided structural information are integrated to refine vulnerability modeling, including foundation typology and structural system characteristics. Scenario-based analyses are performed under alternative hazard intensities (e.g., different peak ground accelerations for seismic events), producing detailed damage probability distributions and associated financial loss estimates. For flooding scenarios, local topographic profiles and cross-sectional analyses are combined with hydraulic hazard maps to simulate water depth variations and asset-specific exposure. Subsidence assessments incorporate time-series deformation trends and satellite-based stability indicators to evaluate potential structural impacts due to differential settlement. This tier delivers hazard-focused synthesis sheets that directly connect physical performance to financial consequences, offering actionable insight for high-value or critical infrastructure assets.

The computational backbone of VALUESAFE is based on an additive risk formulation in which Hazard (IH), Vulnerability (IV), and Exposure (IE) indices are computed separately and combined to derive a global Risk index. Each index is structured upon a multi-tier severity scale and tailored to the hazard type and service level. Hazard indices integrate both static inventories and dynamic satellite-based measurements. Vulnerability indices evolve from multifactorial empirical formulations at Explore level to parameter-rich engineering-informed models at Insight and Detail levels. Exposure indices consider built volume, resident population, territorial extent, and asset economic value, with customizable weighting factors to prioritize human safety or financial assets depending on stakeholder objectives.

A distinguishing feature of the framework is the explicit translation of risk indicators into financial loss and depreciation estimates. Economic exposure is derived from official real estate quotations and cadastral data, while reconstruction cost references and regulatory compensation frameworks are used to calibrate damage-to-loss conversion. This ensures that projected losses align with realistic post-disaster recovery mechanisms. The resulting depreciation indices quantify expected reduction in asset value under specified hazard scenarios, bridging engineering risk modeling and financial analytics.

The methodology has been tested on case studies in central Italy, including large-scale applications in Rome municipality (Explore level) and multi-tier analyses in the Mugello area (Insight and Detail levels). In the Rome test case, grid-based territorial screening successfully captured macro-scale flood and landslide patterns near major river basins, demonstrating the effectiveness of automated large-area classification. In Mugello, a region affected by recent seismic events, building-scale risk assessments integrated fragility-based damage modeling with satellite-derived deformation velocities. Outputs included detailed numerical indicators of hazard, vulnerability, exposure, expected damage classes, condemnation probabilities, and associated depreciation values. High-resolution analyses at Detail level further identified localized deformation trends and differential structural behavior within single facilities. Preliminary validation against previous ad hoc engineering studies conducted in the same area showed substantial agreement in risk classification and damage estimation, while achieving significant reductions in computational time and resource expenditure.

Overall, VALUESAFE demonstrates that the integration of Satellite Earth Observation (SatEO), AI-based building characterization, and engineering-based vulnerability modeling within a unified Web-GIS environment can overcome critical methodological gaps in multi-hazard assessment. The three-tier architecture ensures continuity between territorial screening and asset-level diagnostics, preserving consistency of indicators while adapting to increasing data resolution. By embedding financial estimation directly into the computational workflow, the framework supports not only disaster risk reduction strategies but also investment planning, insurance underwriting, and portfolio risk management.

From a scientific perspective, the contribution of VALUESAFE lies in three main advancements: (i) the operational integration of dynamic InSAR-derived ground motion data into multi-hazard risk indices across different territorial scales; (ii) the extraction of vulnerability-related building parameters at urban scale and their processing to mitigate data fragmentation; and (iii) the systematic coupling of risk modeling with regulatory-aligned economic depreciation estimation. These elements collectively enable a scalable, standardized, and decision-oriented approach to urban resilience analytics.

Future developments will focus on extending validation across diverse geographical contexts, incorporating probabilistic multi-scenario modeling, and enhancing dynamic monitoring capabilities through near-real-time satellite data streams. Additional research will also address uncertainty quantification within each index component and explore machine learning approaches for adaptive calibration of fragility and loss functions. By continuously refining its methodological robustness and operational usability, VALUESAFE aims to contribute to international resilience objectives and to provide a transferable model for integrated multi-risk assessment in climate-vulnerable territories.

The research and operational developments presented in the contribution have been carried out within the framework of the European Space Agency (ESA) InCubed programme, fostering the transition from research-driven methodologies for innovative Earth Observation–based solutions to market-ready multi-risk assessment services.

Rapid and reliable damage mapping is essential for effective emergency response after natural disasters. With the situational awareness that damage maps provide, responders can make plans which use their resources efficiently, make decisions on transport logistics, and identify locations that require more specific targeted attention. In the medium to long term, rapidly-produced damage estimates can be used to identify priority areas for follow-up damage surveys and for making informed decisions about disaster recovery operations.
Synthetic aperture radar (SAR) satellite remote sensing is a valuable tool for the task of damage mapping thanks to three key advantages: It has all-weather, day-and-night data collection capabilities, wide spatial coverage, and it is sensitive to changes in the structure or topography of the ground surface, enabling the detection of damage-related disturbance and deformation. Current semi-operational damage proxy mapping (DPM) algorithms primarily rely on detecting atypical decreases in SAR’s interferometric coherence. These algorithms perform well in high-coherence areas such as urban environments. However, their sensitivity drops sharply in regions with naturally low or variable pre-event coherence such as forests, agricultural land, or some mixed-use land cover.
In this research we present a new damage proxy mapping algorithm that fuses interferometric coherence and SAR intensity similarity using a spatially adaptive weighting scheme. While intensity similarity is less sensitive than coherence to damage in built-up areas, it proves to be more robust across low-coherence land cover types. We evaluate the algorithm’s performance using labelled data from independent sources including ground surveys, aerial surveys, and very high-resolution optical satellite imagery. To make sure that the algorithm has good transferability, our suite of test cases contains six diverse disaster events, including landslides, liquefaction, earthquake, tsunami, and fires.
Our results show that the new method consistently outperforms or matches the benchmark coherence-based algorithm, particularly in vegetated and non-urban regions. The new algorithm performs well on both C-band radar data from the Sentinel 1 satellites, and L-band data from ALOS 2. It also performs well relative to the benchmark under limited data conditions using only two pre-event scenes. This versatile approach improves damage mapping across heterogeneous landscapes, offering significant potential for operational disaster response.

This study investigates the water-surface dynamics of three lakes in the Wkra Forest (NW Poland): Piaski (32 ha), Piaszynko (2 ha), and Karpino (35 ha). The research was initiated following the complete disappearance of Lake Piaski in 2020, caused by a severe regional precipitation deficit. A multitemporal analysis of water surface area changes throughout 2020 was performed using Sentinel-1 C-band SAR data. Images acquired at six-day intervals were retrieved from the Copernicus Data Space Ecosystem and processed with SNAP. Thresholding classification was applied within the lake basins to delineate water extent. Derived surface areas were correlated with meteorological data (precipitation, temperature) from the Szczecin station. Results indicate that the drying process of Lake Piaski commenced in April/May 2020, resulting in total water loss by late July, which persisted until November. A partial recovery of water resources was observed in December 2020. Concurrently, Lake Karpino and Lake Piaszynko experienced reductions in surface area of 17% (6 ha) and 20% (0.3 ha), respectively. The Climatic Water Balance (CWB) during the summer of 2020 dropped below -150 mm, reaching a minimum of < -200 mm in July. The study demonstrates the high potential of high-temporal-resolution SAR data for monitoring hydrological drought in small lake ecosystems.

The use of Synthetic Aperture Radar (SAR) has proven valuable in the context of agricultural monitoring due to its generally consistent acquisition schedule, on account of not being affected by illumination conditions and the presence of clouds. Additionally, SAR measurements are sensitive to structural and dielectric characteristics of the surface (soil roughness, presence and structure of the vegetation, vegetation water content and soil moisture), offering complementary information to the one provided by optical data.

In agricultural applications, SAR-based monitoring commonly relies on time series of backscatter intensity and, in some cases, on polarimetric indicators to characterise crop development and soil conditions. In contrast, repeat-pass interferometric coherence has been comparatively less exploited, as its processing is computationally expensive and its interpretation in this context less intuitive. One frequent limitation in remote sensing studies is the scarcity of detailed ground truth data. The availability of comprehensive ground data in this study provides a valuable opportunity to investigate the complementarity of different radar observables for crop monitoring and to better understand the mechanisms driving temporal decorrelation over cultivated surfaces.

This study focuses on the use of multi-year time series of Sentinel-1 backscattering coefficient and repeat-pass coherence amplitude over agricultural fields. The test site for the study is a set of agricultural parcels in the Cherkasy Oblast, Ukraine, comprising more than 15000 hectares of crops. The available dataset includes field boundaries, crop type, cultivated area, planting and harvesting dates, yield, and ancillary information on agricultural activities and some anomalies. Meteorological observations (daily temperatures, precipitation and snowfall) from nearby weather stations are also considered.

This work takes advantage of the behaviour of coherence amplitude over vegetated surfaces. As vegetation grows, temporal decorrelation increases and coherence rapidly decreases, giving SAR sensitivity to the emergence of plants. Moreover, coherence is sensitive to other sources of temporal decorrelation, such as precipitation, snowfall and farmer activities in the fields, such as ploughing, sowing or harvesting, which are manifested as anomalies in the time series. Finally, changes over time in soil moisture, vegetation water content and its distribution affect both backscatter and coherence, giving information about water stress and plant health.

The results highlight the potential of repeat-pass interferometric coherence derived from Sentinel-1 for crop monitoring applications. Paired with backscatter, coherence provides complementary information on vegetation structure and temporal dynamics, and it can be a useful tool for both field-level management and food security decisions.

Time-Series Interferometric Synthetic Aperture Radar (TS-InSAR) is a powerful remote sensing technique for monitoring Ground Deformation (GD), which poses threats to geological stability and civil infrastructure worldwide. However, many existing TS-InSAR processing software tools face limitations, such as restricted geographic applicability, commercial licensing, and a lack of full end-to-end processing workflow support, forcing users to combine multiple processing platform, resulting in format incompatibility and workflow inefficiencies.
While GMTSAR avoids some of these challenges, it still relies heavily on manual user intervention, requires the execution of C-shell commands, lacks a user-friendly interface, and misses key processing steps such as interferogram network pruning and anchoring of unwrapped interferograms, all of which hinder both accessibility and accuracy. This paper introduces DefoEye (v1), an open-source Python-based software that wraps GMTSAR to facilitate the full TS-InSAR processing chain for Sentinel-1 data through a user-friendly interface.
DefoEye provides an end-to-end workflow, and the main contributions can be summarized as:
• A streamlined, end-to-end TS-InSAR workflow that wraps GMTSAR, from SAR data download to final GD map generation and export, implemented through a structured sequence of processing steps and a user-friendly interface.
• Parallelized job processing of key computational steps, including image co-registration, interferogram generation and correction, sub-swath merging, and phase unwrapping, significantly accelerating TS-InSAR analysis.
• Newly added features, including automatic directory setup, interferogram network pruning using triplet loop closure, flexible anchoring methods (including point/region referencing and average-phase referencing), self-checks at different steps to ensure correct execution, and an enhanced export module with automatic validation using GNSS.
Its applicability and reliability were evaluated in four regions with differing geological settings, causes of GD, and atmospheric and climatic conditions over the 2020-2024 period. For three regions, Bologna (Italy), Gotland (Sweden), and Houston (USA), DefoEye results were compared against 10 available GNSS stations, demonstrating strong consistency. For Karaj (Iran), where no GNSS stations were available, the DefoEye results were compared with those obtained from a widely used Python-based software, MintPy, for TS-InSAR analysis. MintPy requires a stack of unwrapped interferograms, which in this study were generated using GAMMA software. The same zero-reference points, processing parameters, and interferogram networks were used in the GAMMA + MintPy analysis as in the DefoEye processing to ensure a fair comparison between the two platforms. Comparisons show a very strong similarity between DefoEye and GAMMA + MintPy outputs, with DefoEye conducting the process through a unified workflow without the need for additional external software.
Overall, the findings show that DefoEye delivers reliable TS-InSAR results suitable for a variety of geological, hydrological, and environmental applications. DefoEye software and a detailed step-by-step tutorial are available at: https://github.com/ATDehkordi/DefoEye.

The operational utility of Interferometric Synthetic Aperture Radar (InSAR) for regional geohazard and environmental monitoring is strongly influenced by the latency between data acquisition and delivery of analysis-ready time-series products. As Earth observation workflows evolve toward Digital Twin Earth concepts, there is increasing need for processing systems that combine scalability, reproducibility, and predictable response times. While individual algorithmic components of MT-InSAR processing have been accelerated, system-level characterization of end-to-end time-series production using containerized deployments, Slurm-based orchestration, and GPU-enabled HPC and cloud infrastructures remains limited.

This study presents the development and benchmarking of a containerized, modular MT-InSAR processing pipeline designed for high-throughput execution across HPC clusters and cloud environments. The architecture employs CUDA acceleration for computationally intensive kernels and supports elastic scaling across GPU-enabled nodes. Performance evaluation is conducted on the AAU Strato system using a regional Sentinel-1 dataset over Emilia-Romagna, where the single-node wall time is reduced from 14.0 hours to 7.8 hours under GPU offloading. Strong-scaling experiments across 32 nodes achieve 68% parallel efficiency, reducing processing duration for large stacks from days to hours.

To evaluate performance under application-driven load, a hydrology-oriented inversion surrogate is appended as a downstream workload. A 3D U-Net designed for groundwater storage retrieval is trained and tested using InSAR deformation data from Punjab, India, representing a region with intense groundwater depletion and complex deformation dynamics. This surrogate serves as a controlled proxy for emerging environmental InSAR applications and highlights the additional computational burden beyond deformation mapping alone.

Benchmarking results show that end-to-end response time is frequently constrained more by data movement and storage behavior than by raw compute performance. Transitioning from shared parallel file systems to node-local NVMe staging reduces I/O share by approximately 10%, improving latency stability.

The study provides a technical roadmap for deploying scalable, reproducible MT-InSAR processing systems across HPC and cloud GPU platforms and demonstrates how such infrastructures enable integration of computationally demanding application legs, including hydrological analysis.

Key words: MT-InSAR; cloud computing; HPC; GPU acceleration; containerization; Deep Learning; operational latency; Sentinel-1

Wetlands provide key ecosystem services to humans. However, most remain unmonitored, and the costs to monitor them are enormous. Monitoring is essential as unmonitored still waters are already facing accelerated Earth system change, driven by human activities and climate change, with unknown consequences. Interferometric Synthetic Aperture Radar (InSAR) is a promising technology for observing these resources from space. It employs the differences in the path length of two satellite acquisitions taken from the same orbit to generate maps of spatial and temporal changes of the water or land surfaces. Here we test two hypotheses concerning its application: First, besides determining water levels, InSAR can be used to track water surface extent in wetlands. Second, InSAR can be used to locate hydrological barriers, even those found below vegetation or not obvious from satellite imagery. To test both, we perform InSAR on ALOS PALSAR 1-2 imagery in several large wetlands worldwide and capply a convolutional neural network model to identify hydrological barriers in the wetlands. The model can successfully locate flow barriers by seeing abrupt patterns of differences in phase, enabling mapping of the hydrological barriers to flow in wetlands such as roads, ditches, or embankments. We further assess hydrological connectivity in other wetlands of the Ramsar Convention, identifying permanent and seasonal barriers and developing an indicator of connectivity.

Regarding water surface extent, we combine InSAR with polarimetric analysis to detect flooded vegetation in a tropical wetland in Colombia. We find that InSAR detects water surfaces even in areas where conventional backscatter-based methods are limited by dense canopy and complex scattering. The method reveals extensive inundation beneath vegetation that is larger than that obtained from common SAR mapping approaches and allows to construct probabilistic flooded vegetation maps. In this time of rapid Earth system change and the availability of SAR sensors increasing worldwide, we show the unknown potentials of InSAR for the monitoring and hydrological assessment of the functioning of wetlands.

Linear infrastructure networks (highways, railways, pipelines and related structures) are increasingly exposed to interacting ground-related hazards—slow and rapid landslides, regional subsidence, and earthquake-induced liquefaction—whose impacts challenge both safety and long-term serviceability, and whose management is often hindered not by lack of information but by fragmented workflows that do not translate heterogeneous geospatial evidence into consistent, segment-level maintenance priorities.

In this contribution, we present SGAM (Smart Geotechnical Asset Management) as a semi-automated, modular decision-support pipeline that couples Earth Observation (EO) InSAR ground deformation time series with thematic geodata (e.g., topographic derivatives, lithology, land cover, seismic parameters), inventories and engineering knowledge to deliver operational products for predictive maintenance and monitoring planning at network scale; beyond operations, SGAM can also support preliminary design by screening alternative corridors and highlighting geohazard-constrained sectors to help choose the least hazard prone option.

SGAM is designed around hazard-specific modules—rather than a single monolithic model—so each process is represented with predictors, logic and outputs coherent with its physical drivers and data constraints, and results are then intersected with infrastructure geometries to produce standardized hazard/attention classes for assets or asset segments that can be directly ingested in asset-management workflows and GIS-based reporting. The approach has been tested on a ~110 km highway pilot corridor in Northern Italy, where multi-source datasets are fused to support both hazard characterization and interpretation of ongoing activity along the right-of-way.

For landslides, SGAM combines machine learning-based susceptibility mapping (learned from inventories and terrain/environmental predictors such as slope, morphometry, lithology and land use) with systematic exploitation of Persistent Scatterers (PS) InSAR to infer activity state and refine attention attribution, using explicit class-matrix logic that links susceptibility levels to observed deformation ranges (e.g., velocity thresholds) so that slow-moving processes—where InSAR provides actionable kinematic evidence—are distinguished from rapid phenomena for which EO sampling may be intrinsically limited and thus treated conservatively. The outcome is a segment-ready Landslide Attention Index that supports targeted inspections, monitoring densification and prioritization of stabilization needs; for subsidence, SGAM leverages PS InSAR velocity patterns—interpreted within geological and geomorphological context—to identify and classify affected areas and to assign hazard levels along the corridor based on maximum vertical deformation rates and setting-specific constraints, providing an intuitive screening layer for diagnosing underlying drivers and planning follow-up investigations.

For liquefaction, where persistent deformation signals are not the primary indicator, SGAM adopts a susceptibility framework grounded in seismic and geotechnical predictors and implements a segmentation-aligned hazard index with quantitative thresholds, allowing network-level comparison and harmonization with other hazard outputs while remaining updatable as new geophysical and seismic information becomes available.

Beyond the hazard-specific results, SGAM introduces a summary geospatial synthesis layer that consolidates landslide, subsidence and liquefaction outcomes into a single, decision-ready product highlighting priority segments where hazards converge or where one mechanism reaches a severity threshold that warrants immediate action, implemented through transparent rule-based logic to preserve interpretability and traceability from portfolio-level priorities back to underlying drivers.

SGAM also includes a guideline-oriented landslide extension that systematizes parameter calculation and attention-class assignment for linear infrastructure elements (including bridges) in alignment with established national practices, incorporating InSAR-derived kinematic evidence to infer activity state where inventories are incomplete and supporting infrastructure-focused analyses (e.g., buffer-based upslope assessment) for ranking and reporting.

Overall, SGAM demonstrates a practical pathway from Copernicus-era EO deformation streams and multi-source geodata to actionable, segment-level prioritization for predictive maintenance, scalable to large networks and designed to evolve as continuous EO time series and complementary datasets (e.g., LiDAR, drones) enable periodic re-evaluation, and more proactive resilience planning for critical transport corridors.

Terrestrial ecosystems play a critical role in biodiversity conservation and climate regulation through carbon storage and biogeochemical processes. Accordingly, land cover dynamics, including forests, vegetation, and surface water, serve as key indicators for sustainable ecosystem management and climate change studies, reflecting environmental responses to spatiotemporal variability. Remote sensing enables consistent large-scale monitoring of land cover structure and dynamics. At the same time, Synthetic Aperture Radar (SAR) is particularly indispensable due to its sensitivity to structural properties and its all-weather, day–night imaging capability.

The synergy between SAR and optical remote sensing has become a standard methodology for vegetation monitoring, leveraging the complementary characteristics of active and passive sensors. While optical indices such as NDVI and EVI are reliable proxies for photosynthetic activity, their utility is often restricted by cloud persistence and atmospheric interference. Conversely, SAR observations provide moisture-sensitive and structural information independent of solar illumination or weather conditions, enabling continuous monitoring across heterogeneous and cloud-prone regions. In cloud-dominated areas, SAR-driven gap-filling of optical time series has significantly enhanced the operational performance of yield-prediction and vegetation-monitoring models. Nevertheless, the correlation between SAR-derived indices and optical metrics is known to vary with land-cover type and phenological status.

The estimation of NDVI and EVI from SAR data has gained substantial traction with the evolution of machine learning and deep learning. Convolutional neural networks trained on paired SAR–optical datasets have demonstrated the feasibility of predicting NDVI directly from radar imagery, often outperforming conventional regression models. Similarly, temporal and cross-sensor modeling approaches exploiting SAR–optical time series have improved estimation performance by capturing non-linear relationships between scattering mechanisms and spectral vegetation signals.

The study was conducted in the Gönen district of Balıkesir, northwestern Türkiye. The district features a Mediterranean climate with mild winters and long dry summers, and is centered on the Gönen Basin, where the Gönen Stream sustains irrigation across surrounding agricultural lands. Rice is the dominant crop in the region, covering approximately 8,450 hectares as of 2019 data from the Turkish Statistical Institute (TÜİK), and holds significant economic and ecological importance. This study investigates the cross-modal predictive capacity of S-band and C-band SAR observations for reconstructing optical vegetation indices derived from Sentinel-2 imagery. The target variables include NDVI, NDVI-red, and EVI. The primary objective is to assess whether multi-frequency SAR backscatter and radar-derived indices can reliably approximate optical vegetation metrics across contrasting land-cover conditions. To ensure statistical robustness and class balance, a stratified sampling strategy was implemented across three representative land-cover types: Rice, Water, and Forest, using 150 randomly distributed samples per class.

Sentinel-1 was selected as the C-band source owing to its systematic global coverage and open-access data policy, while NovaSAR-1 was chosen as the S-band counterpart due to its ~10 cm wavelength sensitivity to vegetation structure and surface moisture. SAR acquisitions were temporally co-registered with Sentinel-2 optical imagery to minimize phenological discrepancy, with both datasets acquired on 11 July 2024. The predictor space consisted of six radar-derived variables. From Sentinel-1 (C-band), VV and VH backscattering coefficients were extracted, while NovaSAR-1 (S-band) provided HH and VV backscattering coefficients. In addition, the Radar Vegetation Index (RVI) was computed for each sensor independently. Given the intentionally low-dimensional structure of the feature set, no explicit feature selection or dimensionality reduction was performed. This methodological choice enabled a direct examination of intrinsic radar–optical relationships without introducing bias driven by optimization.

Model development and validation were conducted using a 5-fold cross-validation framework to enhance generalizability and reduce the risk of overfitting. Six regression algorithms representing diverse statistical and ensemble learning paradigms were implemented: Random Forest, ExtraTrees, Gradient Boosting, Gaussian Process Regression, NGBoost, and CatBoost. Model performance was evaluated using mean coefficient of determination (R²), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Pearson’s correlation coefficient (r). Statistical significance of the radar–optical associations was assessed using p-values to determine whether observed relationships were distinguishable from random variability.

The results reveal a pronounced dependence on land cover in SAR-to-optical translation performance. In the Rice and Water classes, regression models demonstrated strong predictive performance. In the Rice class, CatBoost achieved the strongest performance (R²=0.74), whereas ExtraTrees yielded the highest accuracy for Water surfaces (R²=0.76). Pearson correlation coefficients ranged between 0.85 and 0.88, with highly significant p-values (p<0.001), confirming that the radar–optical linkage in these relatively homogeneous or moisture-dominated environments is statistically robust and non-random. These findings indicate that multi-frequency backscatter intensities effectively capture surface roughness and dielectric variations characteristic of agricultural and aquatic systems.

Analysis of the top-performing feature combinations further highlighted the critical contribution of S-band metrics. For the Water class, a parsimonious three-feature configuration consisting exclusively of NovaSAR-1 HH, VV, and S-band-derived RVI achieved an R² of 0.766, virtually identical to more complex mixed-sensor configurations. In the Rice class, the most effective models for NDVI-red estimation integrated NovaSAR-1 S-band backscatter with Sentinel-1 VH and the S-band RVI (R² = 0.74). Conversely, feature combinations lacking S-band RVI or NovaSAR-1 HH/VV components exhibited substantially lower predictive stability, underscoring the added value of the approximately 10 cm wavelength for characterizing vegetation structure and moisture dynamics in wetland and agricultural environments.

In contrast, predictive performance over Forest areas was weak and inconsistent. R² values were near zero or negative, and correlation metrics were generally low and statistically insignificant. Both linear and non-linear regression approaches failed to establish a stable functional relationship between SAR metrics and optical vegetation indices in forested environments, suggesting that the employed radar descriptors were insufficient to represent the underlying biophysical complexity.

The performance discrepancy can be primarily attributed to structural heterogeneity and volumetric scattering within forest canopies. Forest ecosystems exhibit pronounced vertical stratification, multiple scattering, and substantial volume contributions, resulting in highly variable backscatter responses. Standard intensity-based metrics (HH, VV, VH) lack the structural sensitivity needed to disentangle canopy-volume scattering, understory effects, and ground contributions. Furthermore, the limited predictor set likely imposed a “small feature bias,” constraining the models’ ability to capture higher-order scattering mechanisms and subtle structural variations.

Feature importance analyses across high-performing models consistently highlighted the contribution of S-band backscatter components and S-band-derived RVI. The approximately 10 cm wavelength of NovaSAR-1 enhances canopy penetration and sensitivity to biomass and moisture-related structural properties, particularly in wetland and agricultural settings. This demonstrates the added value of S-band observations when synergistically combined with C-band data, especially in environments characterized by moderate structural complexity.

Overall, the findings demonstrate an explicit land-cover dependency in SAR-based reconstruction of optical vegetation indices. While multi-frequency synergy enables robust cross-modal estimation in homogeneous and moisture-driven landscapes, it remains insufficient for structurally complex forest systems under low-dimensional feature constraints. These results underscore the need for incorporating additional polarimetric observables, multi-temporal acquisitions, or advanced structural descriptors to improve forest canopy modeling within SAR–optical integration frameworks. The study provides empirical evidence that wavelength diversity is a critical factor in multimodal Earth observation modeling, particularly for vegetation monitoring in moisture-driven ecosystems.

Operational multi-temporal InSAR over geomorphologically complex scenes is often limited by (i) long runtimes, (ii) repeated manual parameter tuning, and (iii) hard-to-audit quality decisions. We present a modular, end-to-end SBAS DInSAR pipeline designed for horizontal scalability and transparent automation, targeting near-operational reprocessing and large parameter exploration.

The workflow implements the full SBAS chain with multiple quality-control checks. These include coherence-driven acquisition screening, phase-consistency checks, statistical validation of intermediate products, and reliability metrics derived from inversion residuals. A multi-stage atmospheric correction cascade combines external reanalysis fields with scene-driven residual corrections to improve time-series interpretability in complex topography.

To reduce turnaround time, the most expensive spatial stages are executed with a cloud-native, vendor-neutral distributed model. An orchestrator dispatches independent tasks through a centralized queue/status service and elastically provisions short-lived containerized workers, enabling high concurrency while preserving fault tolerance via retries, per-task timeouts and automatic resource cleanup.

On three representative test sites covering distinct geomorphological settings, we process SAOCOM L-band stacks of 40-55 acquisitions generating up to 338 interferograms per viewing geometry. Cloud-distributed unwrapping of these networks completes in approximately one hour using up to 200 concurrent containerized workers, compared to multi-day sequential execution on a single node.

Beyond cloud orchestration, the pipeline leverages Artificial Intelligence as an active decision-support system at every critical processing stage. Rather than relying on manual parameter tuning, traditionally one of the most time-consuming and expertise-dependent aspects of advanced InSAR workflows, the system employs Human-AI (HAI) interaction to dynamically manage algorithmic complexity. This AI integration operates across multiple dimensions of the processing chain. For adaptive parameter selection, embedded intelligence analyzes scene and terrain characteristics to suggest optimal processing thresholds, such as baseline constraints derived from geomorphological classification, or stable-point ensemble criteria for systematic bias estimation, ensuring that each dataset is processed with parameters calibrated to its specific conditions. For automated quality assessment, AI-driven routines continuously evaluate data reliability at each processing gate: screening acquisitions based on statistical coherence distributions, identifying unreliable pixels through phase-consistency analysis and verifying network integrity after each filtering step, all without requiring continuous manual intervention. For signal processing, intelligent estimation techniques guide the execution of computationally complex algorithms, from frequency-domain analysis for orbital artifact removal to multi-scale spatial interpolation for residual atmospheric signal correction, where AI feedback dynamically selects methods and parameters based on intermediate data characteristics.

Critically, this level of automation is designed to maintain high-quality outputs. The system keeps the domain expert in the loop through transparent AI-suggested recommendations: the specialist can inspect, override, or validate any automated decision at each stage. Intelligent quality-control routines flag anomalies and low-confidence areas rather than silently discarding them, preserving traceability and enabling informed expert judgment. The result is a pipeline that scales operationally while maintaining the rigor expected of advanced InSAR analyses.

The rapid realization of this complex, multi-disciplinary system was itself enabled by a novel AI-assisted development methodology. We adopted a "deep context" documentation approach: encoding scientific domain knowledge, algorithmic requirements and expert feedback into structured, persistent context files that make AI coding assistants fully domain-aware. This technique allowed rapid implement-test-refine cycles, enabling specialists and non-specialists alike to translate theoretical InSAR concepts into production-ready, parallelized code with continuous AI-driven feedback. The development impact was substantial: a processing environment of this complexity, encompassing cloud-distributed computing, multi-sensor support, multi-stage atmospheric correction and dozens of specialized tools, was built and iterated to operational readiness by a small multidisciplinary team within a timeframe that would be unrealistic under conventional development workflows. We propose this HAI development approach as a replicable model for multidisciplinary teams building complex Earth Observation processing environments, demonstrating that the combination of structured domain context and AI-assisted coding significantly accelerates the path from scientific literature to operational software.

Interferometric SAR (InSAR) has been widely used for monitoring ground deformation [1], which is crucial in disaster management through preparedness and mitigation. The European Ground Motion Service (EGMS), a pre-processed InSAR product provided by European Environmental Agency (EEA), has democratized the access to continental-scale Persistent Scatterers (PSs) for deformation monitoring [2]. However, interpreting this data often requires remote sensing expertise, limiting its usability for disaster managers. To address this gap, we introduce Visual Question Answering (VQA) [3], a system that generally accepts images and natural-language questions to produce answers. We aim to adapt VQA models to EGMS data and improve its usability in disaster management.

EGMS data contains PS points with time series indicating land deformation. Its data format differs from traditional dense, natural images used in VQA, posing challenges in handling point modality and temporal context. We propose a novel VQA architecture specialized for this modality. First, we employ a temporal encoder Presto [4], a lightweight transformer architecture pre-trained on remote sensing data, to capture long-term dependencies in the time series at each point. Next, a spatial encoder Point-BERT [5] models spatial relationships among sparsely distributed points. The encoded spatio-temporal features are fed into an open-source Large Language Model (LLM) backbone (LLaVA-Vicuna [6]). Through instruction tuning with questions, the LLM learns to map specific latent deformation patterns to human-understandable concepts, and generate answers with high-level reasoning.

A major bottleneck in developing VQA systems for EGMS is the lack of labeled instruction data. A semi-automated Knowledge Distillation pipeline is proposed to address this, which bypasses the need for expensive expert annotation. In the pipeline, GPT-4 [7] acts as a teacher. We feed structured physical attributes extracted from EGMS samples (e.g., velocity gradients, anomalies, spatial variability) into GPT-4 to generate high-quality, physically grounded question-answer pairs. This knowledge distillation process transfers the expert interpretation patterns simulated by GPT-4 into the resulting dataset, termed EGMS-Instruct. This dataset acts as a bridge, enabling our student model (Vicuna) to align EGMS-specific visual tokens with their corresponding linguistic semantics.

The framework is currently being prototyped on strategic test sites across Europe. EGMS dataset has been collected, and its structural attributes have been defined. By the time of the conference, we will present preliminary experimental results demonstrating the framework’s effectiveness. Specifically, we will release the EGMS-Instruct dataset, a benchmark for EGMS-language tasks. We will also bring case studies on critical areas and highlight how the system enables disaster managers to rapidly interpret EGMS dataset and assess potential risks. In conclusion, this study proposes a paradigm shift in how we interact with EO data. By synergizingInSAR information with the reasoning power of LLMs, we aim to provide a user-friendly interface to improve the usability of EGMS dataset in disaster management.

Keywords: InSAR, European Ground Motion Service (EGMS), Visual Question Answering (VQA), Large Language Models (LLMs), Disaster Management.

References:

[1] R. Bürgmann, P. A. Rosen, and E. J. Fielding, “Synthetic Aperture Radar Interferometry to Measure Earth’s Surface Topography and Its Deformation,” Annual Review of Earth and Planetary Sciences, vol. 28, no. Volume 28, 2000, pp. 169–209, May 2000, ISSN: 0084-6597, 1545-4495. DOI: 10.1146/annurev.earth.28.1.169. Accessed: Nov. 9, 2025.

[2] M. Costantini et al., “European Ground Motion Service (EGMS),” in 2021 IEEE International Geoscience and Remote Sensing Symposium IGARSS, Jul. 2021, pp. 3293–3296. DOI: 10.1109/IGARSS47720.2021.9553562. Accessed: Nov. 22, 2025.

[3] S. Antol et al., “VQA: Visual Question Answering,” in 2015 IEEE International Conference on Computer Vision (ICCV), Santiago, Chile: IEEE, Dec. 2015, pp. 2425–2433, ISBN: 978-1-4673-8391-2. DOI: 10.1109/ICCV.2015.279. Accessed: Oct. 28, 2025.

[4] G. Tseng, R. Cartuyvels, I. Zvonkov, M. Purohit, D. Rolnick, and H. Kerner, Lightweight, Pre-trained Transformers for Remote Sensing Timeseries, Feb. 2024. DOI: 10 . 48550 / arXiv . 2304 . 14065. arXiv: 2304.14065 [cs]. Accessed: Dec. 16, 2025.

[5] X. Yu, L. Tang, Y. Rao, T. Huang, J. Zhou, and J. Lu, Point-BERT: Pre-training 3D Point Cloud Transformers with Masked Point Modeling, Jun. 2022. DOI: 10.48550/arXiv.2111.14819. arXiv: 2111.14819 [cs]. Accessed: Dec. 15, 2025.

[6] H. Liu, C. Li, Q. Wu, and Y. J. Lee, Visual Instruction Tuning, Dec. 2023. DOI: 10.48550/arXiv.2304.08485. arXiv: 2304.08485 [cs]. Accessed: Oct. 29, 2025.

[7] OpenAI et al., GPT-4 Technical Report, Mar. 2024. DOI: 10.48550/arXiv.2303.08774. arXiv: 2303.08774 [cs]. Accessed: Dec. 16, 2025.

Soil moisture (SM) is a fundamental variable in the evolution and functioning of agricultural and hydrological systems. Its spatial and temporal dynamics influence plant growth, irrigation scheduling, crop yield variability, and the onset and intensification of drought events. Accurate SM estimation is equally important for hydrological modelling, and sustainable water resource management.

Conventional in-situ monitoring networks deliver high accuracy point measurements but are limited by their spatial sparsity and the cost of installation and maintenance. As a complement to ground based observations, Synthetic Aperture Radar (SAR) remote sensing offers a spatially continuous, and weather-independent means of mapping surface soil moisture over large areas. Microwave backscatter is sensitive to soil dielectric constant, which is strongly dependant on volumetric water content. This provides the physical basis for SM retrieval from SAR data.

Over the years, retrieval strategies have evolved from physically-based scattering model inversion [1] to semi-empirical change detection approaches such as the TU Wien change detection algorithm [2], and more recently to data-driven machine and deep learning frameworks that exploit SAR time series [3] [4] [5]. Building on these developments, we investigate the capacity of Sentinel-1 dual-polarized SAR time series (VV and VH) to retrieve surface soil moisture during dry season conditions using a supervised neural network regression. In-situ SM measurements from the International Soil Moisture Network (ISMN) serve as the target variable.

We conduct our analysis over the REMEDHUS soil moisture monitoring network in Spain, composed of 18 stations for our study year providing volumetric surface SM measurements, distributed across the provinces of Zamora and Salamanca [6]. The region has a Mediterranean climate with pronounced summer drought and agricultural land use. Importantly, the terrain exhibits topographic variability including slopes, piedmont zones, and drainage convergence areas, which induces lateral water redistribution.

For this study, we selected six Sentinel-1 Interferometric Wide Swath (IW) Ground Range Detected (GRD) acquired during a dry season (July–September) time series; a period specifically chosen to limit the effect of vegetation on SAR backscatter. We carried out pre-processing using ESA SNAP software, including precise orbit correction, radiometric calibration to sigma-naught (σ⁰), speckle filtering, Range-Doppler terrain correction, and co-registration of all scenes to ensure temporal consistency across acquisitions. In addition to the SAR data, Local Incidence Angle and elevation were added, both derived from SNAP's terrain correction module. The pixels used for the database were extracted per station using a small spatial window centered on each station's coordinates.

We define the regression feature vector as: X = [σ⁰_VH, σ⁰_VV, Local Incidence Angle, Elevation]. We deliberately restrict the feature space to SAR and geometric/topographic descriptors in order to isolate the dielectric sensitivity of C-band backscatter and evaluate the retrieval potential of Sentinel-1 dual polarization.

We designed a feedforward neural network to model the relationship between the SAR observables and in-situ SM. Our architecture comprises an input layer, three hidden layers of 32 neurons each with Rectified Linear Unit (ReLU) activation, and a single output neuron representing predicted volumetric SM. We applied feature normalization prior to training, and performed model optimization by minimizing the Mean Squared Error (MSE) loss function via Adaptive Moment Estimation (Adam).

Our model achieved R = 0.87, RMSE = 0.026 m³/m³, and R² = 0.76 (Fig. 1). The correlation reflects a strong relationship between predicted and observed SM; consistent with the expected sensitivity of C-band backscatter to soil dielectric contrasts under sparse vegetation cover. The RMSE is comparable to recently published SAR-based SM retrievals [4][5], and remains competitive with studies incorporating more features, including optical indices or surface roughness parameters [7][8]. These results are particularly encouraging given the limited number of acquisitions and features.

We then applied our model to a full Sentinel-1 scene to generate a spatially continuous SM map shown in Fig.2. Given the limited geographic coverage of the in-situ network, pixel-wise quantitative validation across the entire scene was not feasible; the assessment then remains qualitative. Nevertheless, the resulting moisture distribution is well correlated with the expected hydrological behaviour of the terrain. Specifically, areas at relatively lower elevations tend to display higher SM values which coincide with slope outlets, where surface runoff accumulation and subsurface lateral flow convergence are expected to sustain higher moisture levels. This spatial correspondence between predicted SM and hydrological phenomena shows the reliability of the soil moisture retrieval.

References

[1] Dubois, P. C., van Zyl, J., & Engman, T. (1995). Measuring soil moisture with imaging radars. IEEE Transactions on Geoscience and Remote Sensing, 33(4), 915–926.

[2] Wagner, W., Lemoine, G., & Rott, H. (1999). A method for estimating soil moisture from ERS scatterometer and soil data. Remote Sensing of Environment, 70(2), 191–207.

[3] Zhu, L., Dai, J., Liu, Y., Yuan, S., Qin, T., & Walker, J. P. (2023). A cross-resolution transfer learning approach for soil moisture retrieval from Sentinel-1 using limited training samples. Remote Sensing of Environment, 301, 113944.

[4] Lakra, D., Pipil, S., Srivastava, P. K., Singh, S. K., Gupta, M., & Prasad, R. (2025) Soil moisture retrieval over agricultural region through machine learning and sentinel 1 observations. Front. Remote Sens. 5:1513620.

[5] Raut, D. B., Misal, V., Pangarkar, R., Raut, S., & Sayyad, S. (2025). Nonlinear soil moisture retrieval from Sentinel-1 SAR using ensemble machine learning. International Journal of Scientific Research in Science and Technology, 12(6), 616–626.

[6] Sánchez, N., Martínez-Fernández, J., Scaini, A., & Pérez-Gutiérrez, C. (2012). Validation of the SMOS L2 soil moisture data in the REMEDHUS network (Spain). IEEE Transactions on Geoscience and Remote Sensing, 50(5), 1602–1611.

[7] Celik, M. F., Isik, M. S., Yuzugullu, O., Fajraoui, N., & Erten, E. (2022). Soil Moisture Prediction from Remote Sensing Images Coupled with Climate, Soil Texture and Topography via Deep Learning. Remote Sensing, 14(21), 5584.

[8] Liu, J., Xu, Y., Li, H., & Guo, J. (2021). Soil Moisture Retrieval in Farmland Areas with Sentinel Multi-Source Data Based on Regression Convolutional Neural Networks. Sensors, 21(3), 877.

The growing demand for Critical Raw Materials (CRMs) driven by the energy and digital transition has led European countries to reassess their national mineral potential and resource recovery strategies. In Italy, the National Exploration Programme (PNE), developed in response to the Critical Raw Materials Act, aims to update national knowledge of the strategic and critical mineral potentiality through non-invasive exploration methodologies and the integration of previous datasets. In parallel, the URBan mining and Extractive waste information System (URBES) project the systematic mapping and characterisation of extractive waste deposits as potential secondary sources of SCRMs, in accordance with the regulatory framework governing mining residues. These parallel initiatives reflect a dual exploration strategy addressing both primary mineral systems and secondary resource reservoirs.

Italy hosts a highly diverse geo-metallogenic framework shaped by the inheritance of Variscan basement domains, the development of Mesozoic passive margins, subsequent Alpine and Apennine tectonic evolution, widespread magmatism, and long-lived hydrothermal systems. These processes generated a broad spectrum of mineral occurrences including several materials currently classified as Critical Raw Materials, distributed across multiple metallogenic provinces. Extensive historical mining footprints have left thousands of abandoned sites and large volumes of extractive waste, particularly concentrated in districts such as Sardinia, Tuscany, the Alpine arc, parts of the Apennines and Sicily. These areas represent both environmental challenges and strategic opportunities for the identification of primary mineral systems and secondary resource reservoirs within contemporary exploration frameworks.

The spatial diffusion of minerogenetic provinces, the presence of extensive leftover mining infrastructures, and the need for low impact exploration methodologies highlight the importance of large-scale Earth Observation (EO) approaches.

Within this context, Synthetic Aperture Radar Interferometry (InSAR) can provide complementary information during early-stage mineral exploration and extractive waste assessment through non-invasive monitoring. Multi-temporal interferometric analyses enable the detection of millimetric surface deformation associated with processes relevant to mineral systems, such as subsidence linked to underground voids, slope instability in former mining districts, and ground responses controlled by structural discontinuities or fluid circulation. Rather than directly identifying mineral deposits, InSAR may help refine investigation areas by highlighting sectors characterised by active deformation or structurally controlled dynamics, thereby supporting geological and geophysical interpretation.

In geothermal districts potentially hosting lithium-bearing fluids, for example, deformation signals may reflect ongoing tectonic or hydrothermal processes that are relevant for understanding the structural framework controlling fluid pathways. Similarly, in historical mining regions such as those of Sardinia or the antimony districts of Tuscany, deformation time series can assist in evaluating the stability of extractive waste deposits and in identifying geomorphological processes influencing material redistribution.

Beyond phase-based measurements, SAR amplitude analysis provides additional insights into surface conditions. Variations in backscatter intensity may be related to surface disturbance, material texture, and moisture conditions, offering a means to characterise extractive waste deposits and their temporal evolution. Although still underexplored in the context of mineral exploration, amplitude-based approaches may contribute to the assessment of secondary resource potential and environmental risk. The integration of InSAR products within national geological databases and extractive waste inventories supports a multi-scale approach, from regional screening to site-specific analysis. In this perspective, satellite-derived deformation and surface characterisation are not presented as stand-alone exploration tools, but as elements that can complement traditional geological mapping, geophysical surveys, and field investigations within a responsible and low-impact exploration strategy.

Satellite-based Interferometric Synthetic Aperture Radar (InSAR), particularly using the Sentinel-1 constellation, has revolutionised global volcanic monitoring, providing an unprecedented volume of routinely acquired, open-access data. Automated systems, e.g., Looking into Continents from Space with Synthetic Aperture Radar (LiCSAR) and its timeseries processing system LiCSBAS, now continuously process interferograms and regularly update deformation timeseries, providing a valuable dataset for monitoring volcanoes globally. However, seasonal snow leads to coherence loss and subsequent unwrapping errors in interferograms. This results in network disconnections in the automated timeseries analysis, reducing deformation accuracy. We identified that 484 of 1183 (~41%) subaerial Holocene volcanoes globally exhibit seasonal snow cover, predominantly situated in high-latitude and high-altitude regions. Therefore, optimising InSAR processing of volcanoes with seasonal snow cover will substantially improve the monitoring of volcanic activity globally.

In this study, we analyse whether an external optical remote sensing dataset (MODIS 8-Day Snow Products) can be used to predict which pixels will be coherent in the interferograms. Based on this, we develop an optimised InSAR timeseries processing workflow that combines InSAR data with the MODIS 8-Day Snow Products. Using the relationship between MODIS snow coverage map and Sentinel-1 InSAR coherence map, we adapted the standard LiCSBAS time-series processing strategy and optimised the network. We use a network selection algorithm that imposes dual constraints: 1) restricting the minimum and maximum connections of each SAR acquisition; and 2) ensuring every interferometric pair is embedded within a minimum of two closed loops. This algorithm guarantees the connectivity of the inversion network and ensures unwrapping errors can be detected.

We test the workflow at Laguna del Maule, Chile, a deforming caldera volcano with seasonal snow cover (182 days in 2017). We process line-of-sight deformation using LiCSAR Frame 083D_12636_131313 and compare the LiCSBAS timeseries outputs between 10/2014 and 06/2023 with data from five continuous GNSS stations. The mask automatically generated by LiCSBAS default settings masked out signals near the deformation centre, primarily due to low average coherence, poor spatiotemporal consistency and a high prevalence of unwrapping errors causing unclosed loops. This results in the loss of critical deformation information, including that at three of the five operating GNSS stations. We calculate the RMS misfit between the unmasked InSAR timeseries and the GNSS timeseries. At the MAU2 station, which is located closest to the deformation centre, the misfit was ~167 mm. We then generated a manually-improved LiCSBAS timeseries, using over 2200 high-quality interferograms. The result shows that the crucial signals around the deformation centre are no longer masked, capturing ~1.04 m of cumulative deformation and the RMS misfit is reduced to ~15 mm at MAU2.

Furthermore, we evaluate the ability of MODIS Snow Products to predict whether the interferometric coherence will exceed a customised threshold. For each interferometric pair, we identify snow-free pixels in both corresponding MODIS snow maps. Using a confusion matrix, we calculate a prediction accuracy of 87%. The network optimisation strategy reduced the required data by ~90% and the LiCSBAS processing time by ~80%. In the resulting timeseries, the LiCSBAS mask retains the critical pixels around the deformation centre and the values match GNSS observations well, with a maximum RMS misfit of ~16 mm at MAU2.

Finally, we consider whether this approach would be applicable to other volcanoes with seasonal snow cover. Vegetation also causes loss of coherence and cloud cover can limit optical satellite observations. To evaluate the global applicability of this methodology, we calculate the Normalised Difference Vegetation Index (NDVI) and Cloud Cover Duration (CCD) for all subaerial Holocene volcanoes globally using MODIS products. We find that more than 50 of 484 global seasonally snow-covered volcanoes exhibit lower NDVI and CCD than LdM (NDVI=0.16; CCD=144 days), suggesting that seasonal snow is their dominant source of coherence loss and MODIS products are applicable. We then tested the applicability of our workflow to a diverse set of seasonally snow-covered volcanoes with different NDVI and CCD, such as the Ashikule volcanic field (China), Askja (Iceland), and Okmok (Alaska, USA). Where applicable, we process long-term deformation using Sentinel-1 data by LiCSBAS, and the results are compared with GNSS observations and the LiCSBAS default processing results. Looking ahead, the quantitative relationships established between MODIS snow products and interferometric coherence can be applied to optimise automated InSAR processing pipelines such as LiCSAR and LiCSBAS.

Volcano monitoring has advanced substantially in recent years, driven by the growing use of remote sensing techniques that overcome limitations inherent to ground-based instruments, such as restricted access, data-transmission constraints, and hazardous field conditions. Among these techniques, Interferometric Synthetic Aperture Radar (InSAR) has become particularly valuable for detecting volcanic deformation and improving our understanding of magmatic and subsurface processes. The development of the LiCSAR system [1], which automatically generates global deformation products at ~100 m resolution—now provides consistent resampled SLC (RSLC) image stacks for all volcanic regions worldwide from 2014 to 2026.

In this study, we introduce a methodology to design a coherence-optimized interferogram network at ~30 m resolution for each volcano. The workflow compiles all RSLC images and generates coherence maps at 6–12 day intervals. Using Smithsonian volcano coordinates and a DEM, we determine summit elevation, delineate the volcanic edifice down to its base (defined as pixels below the 10th percentile of summit height), and compute summit-to-base distances to characterize edifice extent and set a stable analysis window. Mean coherence over this area yields a list of interferograms and an adaptive threshold to retain the most reliable pairs, producing full and filtered coherence matrices. We then select recurring acquisition dates and enforce temporal-density constraints to avoid gaps between acquisitions, generating long interferograms (6, 9, 12 months) as well as short-term interferograms between each date and its four consecutive acquisitions. Interferograms are processed with LiCSBAS [2,3] to produce deformation time series, with optional atmospheric corrections from GACOS and ERA5.

We are currently applying this workflow globally, with particular emphasis on volcanoes active during 2025, while progressively extending the approach to the entire global volcanic catalog. We also compare the resulting InSAR time series with GPS observations from publicly available volcanic monitoring networks. The methodology enhances the efficiency of interferogram-network construction, reduces overall processing time, and introduces an incremental-update strategy that incorporates each new SAR acquisition without reprocessing the full archive. Together, these advances significantly improve the feasibility of near-real-time global volcano monitoring with InSAR.

References:

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

[2] Morishita, Y., M. Lazecky, T. Wright, J. Weiss, J. Elliott, and A. Hooper (2020). “LiCSBAS: An open-source InSAR time series analysis package integrated with the LiCSAR automated Sentinel-1 InSAR processor”. Remote Sensing 12(3), page 424. doi: 10.3390/rs12030424.330

[3] Lazecký, M., Q. Ou, L. Shen, J. McGrath, J. Payne, P. Espín, A. Hooper, and T. Wright (2024). “Strategies for improving and correcting unwrapped interferograms implemented in LiCSBAS”. Procedia Computer Science 239, pages 2408–2412. doi: 10.1016/j.procs.2024.03.258

Mt. Baekdu is an active stratovolcano located along the border between the Korean Peninsula and northeastern China. Depending on meteorological conditions, an eruption could pose a significant hazard to the Korean Peninsula; therefore, continuous monitoring is essential to assess volcanic activity levels and prepare for potential eruptions. Satellite-based Synthetic Aperture Radar (SAR) enables day and night observation regardless of weather conditions and is particularly effective in detecting surface changes in geopolitically restricted environments such as Mt. Baekdu. This study presents a SAR-based volcanic monitoring system developed and operated by the Korea Meteorological Administration (KMA). The system consists of three core monitoring technologies: (1) quarterly average surface displacement analysis, (2) AI-based caldera lake monitoring, and (3) time-series InSAR-based magma chamber volumetric change modeling. First, quarterly average surface displacement monitoring is conducted using Sentinel-1 C-band SAR data through InSAR Stacking analysis. Line of sight (LOS) displacement measurements are utilized to estimate mean displacement values and evaluate anomalous deformation patterns on a quarterly basis. Monitoring results from 2017 to the present indicate that observed displacements remain within a few centimeters, falling within the historical range of variability with no clear evidence of sustained inflation or deflation trends. For Cheonji Lake monitoring, SAR datasets (Sentinel-1, ALOS-2, and TerraSAR-X) were utilized to design and train a U-Net-based segmentation model. A water body label dataset was constructed using SAR intensity images from these satellites to train the model. The trained model was then applied to multi temporal SAR images to segment water bodies and calculate the area of Cheonji Lake using the generated masks. Based on these results, long-term changes in the lake area were quantitatively analyzed, showing no statistically significant changes over the analyzed period. The final method utilized time-series surface displacement data obtained from Sentinel-1 InSAR to model the magma chamber using the Mogi model to interpret the magmatic activity process beneath Baekdu Mountain. To determine initial parameters for volumetric change estimation, InSAR stacking results were first used to estimate the depth and location of the subsurface pressure source by minimizing the residual difference between observations and simulated models through Monte Carlo simulation. With the estimated source location fixed, time-series volumetric changes were subsequently calculated. The results indicate an estimated magma chamber volume change of approximately 0.005 km³, suggesting negligible long-term volumetric variation. By integrating short-term quarterly deformation analysis, time-series magma chamber modeling, and caldera lake area monitoring, this framework enables a multi-indicator assessment of volcanic activity at Mt. Baekdu. This study demonstrates the effectiveness of satellite SAR-based multi-parameter remote sensing for volcano monitoring in environments with limited accessibility and its potential for the quantitative assessment of volcanic activity.

On August 2, 2025 (16:38 UTC), Krasheninnikov volcano (Kamchatka, Russia) resumed eruptive activity after several centuries of quiescence. The first visual evidence was an ash plume rising 3–4 km above the summit (~5–6 km a.s.l.; KVERT), followed by effusive lava emplacement. The eruption occurred less than a week after the largest earthquake in Kamchatka in more than 70 years, a Mw 8.8 event on July 29, 2025, at 23:24 UTC.

The August 2025 eruption occurred during an optimal observation period, allowing the use of Sentinel-1/2. Because seasonal snow cover typically masks the upper flanks from autumn through late spring, the summer 2025 eruption provided a rare opportunity to investigate syn-eruptive deformation using satellite radar observations. From October 2024 to late June 2025, heavy snow cover prevented detection of small-scale thermal or deformation signals from orbit.

We analyze volcano deformation during the first two weeks of August 2025 using Sentinel-1A and 1C C-band SAR data from four ascending and descending tracks, incorporating pre- and post-eruptive acquisitions between June and October 2025 to constrain the temporal evolution of the deformation signal. More than 30 interferograms were generated using GMTSAR, SNAP, and ISCE, with and without tropospheric delay correction.

We constrain the deformation to a short-lived episode that began after the Sentinel-1 acquisition on July 31, 2025 (~19:30 UTC). Independent interferometric pairs indicate that most line-of-sight (LOS) displacement occurred within ~24 hours and had ceased on or before August 5, 2025. The spatial pattern consistently reveals a localized LOS displacement centered on the summit, with no evidence of prolonged pre-eruptive inflation.

Using published estimates of lava-flow extent between 2–12 August (~2.7 km²) and assuming plausible mean flow thicknesses of 2–5 m, we estimate an erupted bulk volume of ~5–14 Mm³. This volume is consistent with a short-lived, localized intrusion rather than a sustained magma supply episode.

To explain the observed displacement field, we tested elastic source models using the four available satellite tracks, including spherical point sources (Mogi), sub-vertical dyke intrusions, and combined geometries to assess model robustness. The preferred solution is a shallow, sub-vertical dyke intrusion with a modeled volume change of ~20–40 Mm³ and an opening of 1–2 m, initially without surface breach. This source geometry is consistent with the observed eruptive behavior and its close temporal proximity to the July 2025 Kamchatka earthquake sequence, suggesting a short-lived intrusion occurring in close temporal association with the July 2025 earthquake sequence.

High-resolution time-series InSAR provides the opportunity to resolve volcanic flank deformation beyond edifice-scale interpretations. We analyze 193 ascending and 208 descending TerraSAR-X Spotlight scenes (2012–2021) over Lastarria Volcano in the Central Andes to investigate the spatial organization and persistence of flank motion at meter-scale resolution.

Line-of-sight velocities range from approximately –10 to +10 mm yr⁻¹ and reveal a distinctly non-radial deformation pattern. A polarity reversal across the summit ridge indicates opposing horizontal components on eastern and western flanks, consistent with downslope-directed motion rather than coherent radial spreading.

On the eastern flank, a prehistoric debris-avalanche sector exhibits internally coherent but temporally variable creep. On the northwestern–western flank, deformation is strongly segmented by lithological and geomorphic contrasts. Four wedge-shaped mid-slope domains display sustained velocities of ~5–8 mm yr⁻¹, whereas coherent ignimbrite bodies and lava outcrops correspond to zones of minimal displacement. Deformation preferentially localizes at lithological transitions rather than on the steepest slopes and internally follows downslope-diverging channel networks, forming structured fan-like patterns.

These spatial patterns persist throughout the available TerraSAR-X record and are detectable in earlier satellite observations, indicating multi-decadal gravitational creep organized into mechanically distinct compartments. Time-series processing was performed using an SBAS approach with quadratic ramp removal and geometry-dependent coherence thresholds; residual RMS statistics confirm that displacement signals exceed background noise levels.

Across spatial scales, deformation is hierarchically organized: non-radial at the edifice scale, sector-confined at the flank scale, and internally structured within sub-kilometer domains. The results demonstrate that volcanic flanks may deform as mosaics of partially decoupled compartments rather than as coherent sliding bodies, highlighting the capability of high-resolution InSAR to resolve persistent, compartmentalized gravitational instability in mechanically heterogeneous volcanic systems.

Modern-day volcanic deformation in the Central Andes is anomalous, characterised by spatially extensive displacements, deformation sources that remain stationary over decades, no geomorphological evidence of net uplift, and little correlation between deformation and Holocene volcanic activity [1]. It has been proposed that uplift is caused by the ascent and then temporary accumulation of magmatic volatiles from the mid-crustal Altiplano-Puna and Southern Puna Magmatic Bodies; when volatiles are released, the ground subsides back to its initial position [1]. Here, we test this idea against 15 years of Sentinel-1 satellite radar observations of surface deformation in the Atacama. A ~50 km region within the Pastos Grandes Caldera, Bolivia, has been uplifting since early 2023 with a maximum rate of 70 mm/yr. This has started to decelerate to a rate of 40 mm/yr in late 2025, although this is still the fastest volcanic deformation that has been observed to date in the Central Andes. Preliminary analytical elastic half-space modelling suggests the pressurisation at Pastos Grandes is caused by a 1e8 m3 volume increase at about 13km below the surface. This is deeper than the 4-8 km depth of pre-eruptive plutons derived by thermo-barometry [2,3]. It is unclear if a 13 km depth is within the Altiplano-Puna magmatic body under Pastos Grandes.

Pastos Grandes is a Pliocene caldera that last erupted 2.89 Ma, producing the 1500 km3 Pastos Grandes Ignimbrite [4]. There is still an active hydrothermal system with thermal springs surfacing in Laguna Pastos Grandes [5]. We are currently investigating the response of these thermal systems using median thermal anomalies derived from MODIS [6]. A thermal anomaly, with an onset in early 2023, is visible across Laguna Pastos Grandes and Laguna Q’ara. This suggests that degassing and hydrothermal activity increased at roughly at the same time as the inflation started.

We also discuss new observations of small-scale (<6 km) uplift at the Bolivian volcanoes of Cerros de Tocorpuri, Parinacota and Jatun Mundo Quri Warani, observations of a second uplift period at Cerro Overo, and provide updated timeseries for Cerro El Cóndor, Socompa and Lazufre. Uplift in the Central Andes has historically been episodic, with some events lasting a few years (e.g., Socompa and Sillajhuay) and some continuing for decades (e.g., Uturuncu and Lazufre) [1].

References:

[1] Pritchard, ME. et al. (2018) Geosphere, https://doi.org/10.1130/GES01578.1

[2] Muller, E. et al. (2020) Geochimica et Cosmochimica Acta, https://doi.org/10.1016/j.gca.2020.03.020

[3] Schmitt, A. et al. (2001) Contributions to Mineralogy and Petrology, https://doi.org/10.1007/s004100000214

[4] Salisbury, MJ. et al. (2011) GSA Bulletin, https://doi.org/10.1130/B30280.1

[5] Bougeault, C. et al. (2019) Minerals, https://doi.org/10.3390/min9060380

[6] Girona, T. et al. (2021) Nature Geosciences, https://doi.org/10.1038/s41561-021-00705-4

InSAR time series analysis of ground deformation in the Chiles-Cerro Negro region reveals a complex sequence of temporally and spatially linked magmatic and tectonic events. InSAR has detected several areas of uplift; preliminary interpretation supports the presence of shallow sill-type magmatic sources beneath and south of the volcanic complex (depth < 2km) and Potrerillos (depth < 5km). Mw5.8 and Mw5.6 earthquakes occurred in October 2014 and on 22 July 2022, respectively. Constant inflation of the volcanic complex and a 3 cm uplift event at the Potrerillos caldera a month prior the July 2022 earthquake appear to have promoted the earthquake. Furthermore, the noticeable increase in uplift following the July 2022 event (~8 cm in the caldera and south of Chiles) suggests a positive feedback not only between magma intrusion and earthquakes but also between the earthquake and the magmatic system. Stress changes due to magma intrusion promote the faulting while stress changes due to faulting promote further magma intrusion. To better understand these interactions, we conducted magmatic source inversions and Coulomb stress modeling to evaluate the spatial relationship between inferred pressure sources, fault geometries, and earthquake timing. The processes before and after the 2022 earthquake are somewhat similar to the October 2014 ones. Ebmeier et al. (2016) noted that the earthquake was a consequence of regional surface displacement around the complex.

The magmatic unrest at Mount Agung preceding the November 2017 eruption was identified by previous InSAR studies in September 2017, but the precursory signal had begun much earlier, masked by a concurrent tectonic signal. Applying Variational Bayesian Independent Component Analysis to the full Interferometric Synthetic Aperture Radar (InSAR) time series, we separate the magmatic deformation from atmospheric noise and regional tectonic subsidence driven by interseismic loading on the Flores back-arc thrust, and reveal that magmatic inflation began in May 2017, six months before the eruption. The separated signals are independently verified by Global Navigation Satellite Systems displacement and seismicity patterns. Signal separation of tectonic and magmatic deformation, demonstrated retrospectively at Mount Agung, provides a framework for early precursor detection at tectonically complex arc volcanoes.

The Subdirectorate General for for Monitoring, Warning and Geophysical Surveys, belonging to the National Geographic Institute of Spain (IGN) is responsible for planning and management of systems for observation, monitoring and communication to institutions of volcanic activity and determination of associated hazards within Spanish territory and around Spanish Antarctic bases, as well as conducting research in volcanology and the early detection of eruptive processes.

Within this framework, the IGN operates a comprehensive suite of interdisciplinary observation systems encompassing geodesy, seismology, gravimetry, geochemistry, and geomagnetism. In order to monitor ground deformations, Spaceborne SAR interferometry (InSAR) is integrated with other techniques such as GNSS, tiltmeters or robotic total stations.

In this context, InSAR is applied through several complementary methodologies. An automated processing workflow, in continuous operation for the past eight years, generates interferograms with each new satellite acquisition, supporting near‑real‑time detection of deformation signals. In addition, long‑term deformation time series are produced to characterize the temporal evolution of volcanic processes and to identify ground‑motion trends. Both processing strategies are implemented in the Canary Islands using data from the Sentinel‑1 and PAZ satellites.

Due to the particular atmospheric and topographical characteristics of the Canary Islands, displacement and interferometric phase maps often exhibit a significant contribution from atmospheric artifacts. These effects are especially common on volcanic islands such as the Canaries, where strong vertical gradients in water‑vapor distribution and dominant moisture‑bearing winds from the ocean generate substantial atmospheric phase delays.

In parallel, several processing workflows are in operation for the GNSS station network. In addition to providing precise ground‑deformation measurements, GNSS data supply zenith tropospheric delay (ZTD) estimates that can be used to correct atmospheric artifacts in the interferograms.

In this work, we present the results obtained from applying several methodologies—such as GACOS atmospheric products, phase–topography correlation analysis, and GNSS‑derived ZTD corrections—to mitigate the impact of atmospheric variability on interferograms. To evaluate their performance, these approaches were tested on islands with contrasting atmospheric and topographic conditions, as well as different expected deformation patterns, in order to identify the most suitable correction strategy for each scenario. Based on these results, our aim is to incorporate atmospheric correction into the automatic processing workflow enabling the selection of the most appropriate correction method in each case.

Geodetic models of volcanic deformation are commonly interpreted using simplified analytical sources (e.g., Mogi or sill geometries), implicitly assuming melt-dominated and spatially discrete magma storage. At Eyjafjallajökull, such approaches have supported interpretations of multiple shallow intrusions or stacked sills.

Here, we revisit the 2010 eruption using integrated InSAR and GNSS time series, focusing on co-eruptive and post-eruptive deflation. We show that previously proposed sill-like geometries systematically underpredict the observed deformation amplitudes. In contrast, a single ellipsoidal source at ~6–7 km depth simultaneously reproduces the spatial and temporal evolution of subsidence.

However, this geometry is inconsistent with a purely molten reservoir, as seismicity persists within the inferred source region. We reconcile these observations by interpreting the source as a mechanically heterogeneous, crystal-rich mush, in which deformation reflects bulk volume change while seismicity is accommodated within the rigid framework. Petrological and geochemical evidence further supports this interpretation, indicating magma mixing between newly intruded basalt and remnant silicic material stored in the system. Additionally, the discrepancy between erupted volume and geodetically inferred contraction suggests that magma compressibility and fracture compliance play a significant role in modulating the deformation signal.

Together, these results challenge interpretations based on discrete melt bodies and instead support a vertically extensive, mush-dominated magmatic system beneath Eyjafjallajökull.

Volcanic eruptions occur when magma generated by partial melting of the mantle rises and is released on the Earth’s surface in the form of lava, pyroclastic material, and volcanic gases. Understanding eruption mechanisms and subsurface magma structures, including magma chambers and pressure sources, is essential for mitigating volcanic hazards. Mauna Loa, located in Hawaii, is the largest active volcano on Earth and has experienced frequent eruptions since 1843. One of the most significant eruptions began on November 27, 2022, providing an opportunity to investigate eruption-related surface deformation using remote sensing techniques.

Synthetic Aperture Radar (SAR) is a remote-sensing technique that enables data acquisition regardless of weather conditions, making it particularly useful for monitoring inaccessible areas, such as volcanic regions. Differential interferometric SAR enables precise deformation measurements with centimeter- to millimeter-scale accuracy. These techniques are widely used to monitor surface displacements associated with volcanic activity, earthquakes, landslides, and subsidence. In this study, we monitored surface deformation associated with the 2022 Mauna Loa eruption using time-series Interferometric SAR (InSAR) and assessed how well a simple point-source model explains the observed spatiotemporal deformation complexity. Small Baseline Subset (SBAS) analysis was conducted using Sentinel-1 datasets acquired between June 2022 and December 2023.

Since differential interferograms measure surface deformation along the satellite line-of-sight (LOS), observations from different viewing geometries are required to interpret the deformation components. In this study, both ascending and descending Sentinel-1 SAR tracks were used for SBAS time-series analysis and displacement decomposition. A total of 47 ascending and 47 descending SAR images were processed to generate cumulative LOS deformation maps. The time-series deformation results were validated using Global Navigation Satellite System observations from four stations near the summit region, which showed consistent spatial and temporal displacement patterns. Assuming negligible north-south displacement, east-west and up-down displacement components were estimated using three-dimensional deformation analysis before, during, and after the eruption. Although deformation in the north-south direction may exist in the Mauna Loa summit region, this assumption is commonly adopted because Sentinel-1 orbits nearly poleward and InSAR has limited sensitivity to the north-south component; thus, our three-dimensional results primarily constrain the east–west and vertical motions.

The time-series results indicate deformation concentrated near the Mauna Loa summit caldera during the eruption, with LOS displacements ranging from approximately +0.32 m to –0.29 m over the study period. To distinguish the short-term eruptive signal from longer-term cumulative deformation, we also examined an event-focused differential interferogram spanning the eruption, which reached approximately 0.31 m and 0.34 m in the ascending and descending LOS geometries, respectively. The three-dimensional deformation analysis reveals complex displacement patterns with both horizontal and vertical components, consistent with previously reported deformation mechanisms of Mauna Loa associated with magma migration and the 2022 eruption.

To further interpret the deformation source, magma-source modeling was performed using the Geodetic Bayesian Inversion Software based on the Mogi point-source model. The inversion was conducted separately for deformation fields derived from the ascending and descending SBAS time series displacement fields. The modeling results indicate a primary deformation point source beneath the summit of Mauna Loa, which may represent a pressure source associated with the subsurface magma system. The estimated source parameters show relatively stable behavior across inversions, suggesting that the best-fitting source is consistently centered beneath the summit area.

However, a comparison of the modeled displacement field with the observed SBAS displacement revealed localized residual deformation near the caldera. While the point-source model can explain the first-order deformation pattern associated with the initial eruption, it does not fully capture the spatial complexity observed in the SBAS results. These residual signals are interpreted as deformation components that are not directly related to the initial magma-pressure source but may be associated with shallower subsurface deformation processes or surface displacement caused by lava-flow emplacement following the initial eruption. This result suggests that the 2022 Mauna Loa eruption involved multiple deformation mechanisms that cannot be fully represented by a single point source in a homogeneous elastic half-space, highlighting the limitations of single-source modeling and the importance of integrating time-series InSAR analysis with geodetic inversion.

Keyword: Mauna Loa eruption, SBAS, Sentinel-1, Mogi point-source model

Volcanic activity can rapidly and completely reshape the Earth’s surface, from the emplacement of lava flows and ash to catastrophic explosive events that destroy volcanic edifices. Monitoring lava flow progression during eruptions is crucial for understanding volcanic behaviour and mitigating associated hazards, including the destruction of infrastructure, livelihoods, and loss of life. Satellite imaging has emerged as a powerful tool for monitoring volcanic eruptions, providing new insights where other remote sensing or ground-based observations may be limited due to inaccessibility, the hazardous environment or cloud coverage (e.g., meteorological or volcanic). However, tracking the progression and morphology of lava flows, especially during ongoing eruptions, remains a challenge.

This study demonstrates the potential of SAR backscatter and coherence imagery for monitoring lava flow emplacement and morphology. Using high-resolution COSMO-SkyMed (CSK) data, we examine the 2017 eruption of Erta ‘Ale, Ethiopia and the 2016 lava flow at Kilauea, Hawai’i. SAR backscatter and coherence offer complementary strengths for detecting lava flow changes. Backscatter is sensitive to variations in the surface roughness, which highlights newly emplaced lava and provides insights into flow surface morphology. Coherence responds to temporal changes in the surface stability, with freshly emplaced lava producing high decorrelation. Individually, these datasets can identify volcanic activity, but combined they provide a more robust basis for detecting and characterising lava flows in various environments.

We develop and evaluate an automated lava flow extraction framework based on pixel SAR timeseries. To improve detection accuracy, we assess the impact of spatial filtering through three scenarios: (1) no additional filtering, (2) bilateral filtering, which locally reduces speckle while preserving sharp flow edges, and (3) a combination of multi-looking and super-resolution processing, which improves the overall signal-to-noise ratio across the image while maintaining spatial detail.

We apply a sequential change detection algorithm using the Cumulative Sum (CUSUM) method to automatically identify newly emplaced lava flows. CUSUM detects statistically significant deviations in pixel backscatter and coherence over time. It identifies abrupt shifts in the mean signal relative to a pre-determined background level that we correlate with lava flow emplacement. We evaluate detected changes against other datasets (e.g., field observations) at the pixel level or with an additional spatial aggregation of neighbouring pixels to better capture continuous flows and reduce isolated noise.

This pixel-based temporal analysis on the SAR backscatter achieved 79% and 82% of pixels correctly classified compared to manually derived flow maps for Erta ‘Ale and Hawai’i respectively. Here, we observe rougher flow surfaces emplaced farthest from vent. Integrating backscatter and coherence provides a more comprehensive characterisation of lava flow extent and morphology than either dataset alone. In more complex environments, their combined strengths are expected to improve flow extraction and reduce misclassification, increasing confidence in mapping extents and morphological interpretation.

To assess transferability, we test the approach on Sentinel-1 datasets at multiple volcanoes (e.g., Nyiragongo, Fuego, and El Reventador), which vary in environmental conditions, vegetation and seasonal dynamics. The comparison between CSK and Sentinel-1, allows us to explore the sensitivity of the method to radar polarisation (co- and dual-polarisation) and wavelength differences (C- and X-band), which influence backscatter sensitivity to surface roughness and vegetation, affecting accuracy of lava flow detection. We demonstrate that SAR timeseries analysis can provide a robust and transferable method for tracking lava flow progression and morphology, providing the frequent observations needed for volcano monitoring.

During volcanic unrest, inverse modeling of InSAR surface displacements is often used to retrieve the parameters of the source characteristics of volcanoes. Such approach could be costly in terms of computing time, and therefore not always adapted to routine modeling. Here, we are interested to explore the potential and limitations of deep learning for the characterization of volcanic sources. We rely on our previous proof-of-concept work (Lopez Uroz et al., 2024), which has demonstrated the potential of neural networks in case of a point source model (Mogi 1958).

This work will be used as a basis for expanding research into the benefits of deep learning in volcanic inverse modeling. In this study, we will address issues of image resolution, patch dimension, and SAR acquisition geometry in order to properly prepare the learning dataset and to further improve the prediction accuracy of neural networks.

To do this, we continue with the Mogi model (Mogi, 1958) described by the depth of the volcanic source and its volume change. We propose multi-channel input, including multi-resolution and multi-geometry (i.e. descending and ascending) InSAR displacement fields in the case of previously deployed ResNet to estimate the depth and the volume change of the volcanic source. For results evaluation, besides the mean squared error of estimated parameters compared to the truth, we also consider structural similarity (SSIM) between the input displacement field and the reconstructed one as metric. Moreover, we perfome a sensitivity analysis in order to highlight the impact of resolution on each estimated parameter.

The multi-channel ResNet model is trained and first evaluated by synthetic datasets where the ground truth allows for quantitative assessment of the learning performance. The trained model is also applied to Sentinel-1 InSAR datasets covering the Suswa volcano during the period of 2016-2020 (Albino & Biggs, 2021). The results are compared to those of the conventional inversion approach for validation.

According to the obtained results, the input image resolution to the neural network has significant impact on its effectiveness. As shown by the sensitivity analysis, the depth of the volcanic source is more sensitive to near-field displacement information, while its volume change is more sensitive to far-field information. Given the fixed input image size, a multi-resolution input therefore seems optimal to ensure both near-field and far-field information. On the other hand, taking both ascending and descending geometries as input to the ResNet model is highly interesting for improving the inversion performance.

1. Introduction

In April 2025 ESA launched the BIOMASS satellite with the first orbital P-band SAR as the main instrument. The main objective of ESA’s BIOMASS mission is the mapping of forest biomass as a contribution to the understanding of the carbon cycle and climate system. In addition, the Biomass mission offers the opportunity to develop other applications of orbital P-band SAR data.

Early results of the mission confirm the interferometric capability of the BIOMASS P-band SAR. But they also show that many of the interferograms are significantly affected by ionospheric effects. The main ionospheric effects are: (1) positional offsets in the azimuth direction caused by ionospheric path delay gradients along the synthetic aperture, (2) ionospheric path delay phase, and (3) Faraday rotation effects. The third effect affects the cross-polarization backscatter and polarimetric parameters. The Faraday rotation depends on the total free electron concentration along the imaging path. The first two effects depend on the spatial variation of the free electron concentration.

In our contribution, we investigate the ionospheric effects on positional offsets and the interferometric phase. We first discuss methods to identify the presence of ionospheric effects in an interferometric pair. Then we present methods to mitigate ionosphere-related positional offsets and path delays and discuss the applicability of these methods.

2. Identification of ionospheric effects

For an interferometric SLC pair, the presence of ionospheric effects can be identified based on the co-registration offset field or based on the sub-band double difference interferogram.

Our co-registration offset-field based approach starts with the co-registration of the second SLC to the reference SLC geometry based on the orbital parameters and a digital elevation model. The related geometric model is refined with a single range and azimuth offset (not an offset field). Then, in a second step, matching techniques are used to determine an offset field between the reference SLC and the transformed second SLC. Spatially varying non-zero azimuth offsets indicate either ionospheric effects or ground-displacements. In the case of ionospheric effects, the corresponding range offset field does not show significant non-zero values. In the case of ground deformation, on the other hand, the range offset field also shows offsets for the instable areas. In addition, the shape of deformation patterns typically differs from the shape of patterns caused by ionospheric effects, and the displacement effects related to the topography (landslides, glaciers), natural processes (earthquakes, volcanoes, ice motion) or man-made activities (mining, oil, or water extraction).

The split-beam double difference interferogram procedure includes the azimuth bandpass filtering of the two SLCs, the calculation of differential interferograms for the two azimuth sub-band SLC pairs, and the calculation of the double difference interferogram, which can be done in a combination of the two complex valued differential interferograms. It is relevant that the co-registration is not done with offset-field refinement, as this will remove the split-beam double difference interferogram. The resulting non-zero phase relates either to ionospheric effects – the signal of the two sub-bands propagates through different parts of the ionosphere to the same pixel on the surface – or to an along-track ground motion. Typically, motion effects and ionospheric effects can be discriminated since the prior ones relate to the surface characteristics or processes such as earthquakes or ice motion.

3. Mitigation of positional offsets

Resampling the transformed second SLC with the determined offset-field permits achieving an accurate co-registration in the presence of ionospheric effects or ground-displacements.

In the case of a very strong spatial variation of the ionospheric path delay, with azimuth offsets > 1 SLC pixel, the azimuth offset differs between different parts of the azimuth spectrum. In such cases it is possible to apply the procedure separately to smaller fractions of the azimuth spectrum This improves the co-registration in such cases, as confirmed by the higher coherence obtained, nevertheless we could not get perfect results in cases with very “wild” ionospheric effects as observed mainly in arctic regions.

4. Mitigation of ionospheric path delay phase effects

For the estimation of the ionospheric path delay phase, we apply the split-spectrum method as described in [1,2]. The full bandwidth differential interferogram is generated and spatially unwrapped. The split-spectrum double difference interferogram is generated by combining the two complex valued sub-band differential interferograms. The resulting phase variations are typically smaller than half a phase cycle, permitting to directly extract the “unwrapped” phase values without spatial phase unwrapping step. In the BIOMASS case and using the lowest and highest third of the 6 MHz chirp bandwidth for the split-spectrum double difference interferogram, the factors used to calculate the ionospheric path delay phase based on the unwrapped full bandwidth differential interferogram phase and the split-spectrum double difference interferogram phase are 0.5 and -54.38. This ionospheric path delay phase can then be subtracted from the full bandwidth differential interferogram to get the “ionosphere corrected” differential interferogram. At P-band, the coherence of pairs with short intervals is typically very high, which facilitates the unwrapping step. In cases with low or intermediate level path delay variations the ionospheric path delay phase variations can typically be estimated and used to subtract this ionospheric path delay phase term from the BIOMASS interferograms. But there are cases with “too wild” ionospheric effects, observed in arctic regions, where the mitigation procedure used was not successful.

5. Discussion and conclusions

The two approaches used to identify ionospheric effects are robust, reliable and can be automated. We typically use the co-registration refinement offset-field based approach as it also provides the refined co-registered SLCs. The co-registration quality is good enough to get coherence values interpretable with respect to the scatterer characteristics and phases that can be unwrapped – except for pairs affected by spatially very strongly varying ionospheric effects as observed sometimes in arctic areas. The mitigation of the ionospheric path delay using the described split-spectrum based approach usually works fine, again except for the cases with spatially very strongly varying ionospheric effects. Overall, the presence of ionospheric effects can be identified and, in most cases, corrected for. The cases with “wild distortions” can be identified. In these cases, fully correcting the data for ionospheric effects was not achieved.

At the ESA BIOMASS workshop held in Ljubljana in early 2026 several presenters addressed the estimation of ionospheric phase screens, but nobody else seemed to use the split-spectrum method, which is typically used for this purpose with L-band and higher frequency data. Their reason for not trying it was the narrow 6 MHz chirp bandwidth of the BIOMASS data. Nevertheless, because of the lower frequency, the applicability is quite good, with scaling factors similar to L-band data with about 20 MHz bandwidth. Our results clearly confirm the applicability of the technique.

6. Acknowledgements

ESA is acknowledged for providing us with access to the BIOMASS data used in our work.

7. References

[1] Wegmüller, U., Werner, C., Frey, O., Magnard, C. and Strozzi, T.: Reformulating the Split-Spectrum Method to Facilitate the Estimation and Compensation of the Ionospheric Phase in SAR Interferograms. Procedia Computer Science, pp. 318–325, 2018. doi:10.1016/j.procs.2018.10.045

[2] Wegmüller, U., Werner, C., Frey, O., and Magnard, C., “Estimation and Compensation of the Ionospheric Path Delay Phase in PALSAR-3 and NISAR-L Interferograms,” Atmosphere, vol. 15, no. 6, p. 632, May 2024, doi: 10.3390/atmos15060632

Airbus is the market leader in the production and dissemination of global-scale, commercial off-the-shelf digital elevation datasets, derived predominantly from interferometric processing of imagery collected by the Airbus Radar Constellation (TerraSAR-X and TanDEM-X). Airbus’ flagship elevation product – WorldDEM Neo – provides customers with both a Digital Surface Model (DSM) and a Digital Terrain Model (DTM). The former is sensitive to both vegetation (including, most notably, tree cover) and man-made objects, whereas the latter is derived from the DSM via the removal of these objects. As the highest resolution, truly global elevation product on the market, WorldDEM Neo is provided with 5 m horizontal pixel spacing and has an absolute vertical accuracy of 1.4m at 90% confidence (LE90).

In terms of digital terrain information, precise measurements are a critical requirement for Airbus’ customers given the foundational importance it plays across a wide range of downstream applications. Key use cases for terrain information (DTMs) include:

• Precision land use management (including agriculture and forestry);

• Carbon crediting and environmental / climate change monitoring;

• Water resource management (e.g. for the design of water supply, flood control and hydrological structures);

• Civil Engineering (including critical infrastructure development and monitoring);

• Insurance, Environmental Impact Assessments and Disaster Mitigation Planning;

• Geological and Archaeological exploration;

• Defence and Security applications (e.g. for environmental domain awareness and strategic planning).

Despite the important use case applications listed above, little to no high accuracy terrain measurements exist across the world’s rainforest regions which, together with the planet’s other densely tree-covered regions, cover over 30% of Earth’s land area (FAO, 2025: https://www.fao.org/state-of-forests/en/). There, terrain information is typically interpolated from sparse field-based measurements or is ‘guesstimated’ from satellite-derived DSMs with the support of point-wise laser altimetry data, the latter of which can themselves be subject to significant measurement error. This error is due primarily to limitations in our ability to image the complex structure of the rainforests from space (let alone the terrain below), and the absolute vertical error associated with several market-leading DSM products can exceed 10 metres over selected areas of rainforest.

Built by Airbus Defence and Space, ESA’s new BIOMASS P-band forest mission presents an exciting opportunity to peer through the rainforest canopy for the first time from space, and in doing so opens new doors for rainforest-scale DTM refinement. Here, we present ongoing progress towards the production of precision DTMs over the rainforests, with a view towards their integration into our Airbus WorldDEM Neo products and services. This progress exploits imagery acquired during the first (tomographic) phase of the BIOMASS mission, whose unique orbital configuration permits the mapping of pseudo-3D forest structure and underlying terrain in unprecedented detail.

In light of the above, we aim to:

1. Develop a sub-rainforest processor capable of creating pseudo-three dimensions maps of forest structure and the terrain below. In particular, we are developing a Tomographic SAR processor based on the first phase of the BIOMASS mission.

2. To integrate the BIOMASS DTM into existing Airbus WorldDEM Neo workflows and elevation products (e.g. CO3D). A particular focus will be given to oversampling and interpolating the BIOMASS data in order to achieve the same spatial resolution of the WorldDEM Neo.

3. To compare and benchmark the BIOMASS DSM against Copernicus DEM Global data.

At conference time, the authors aim to show relevant examples of BIOMASS elevation data derived from the exploitation of BIOMASS Level-1C stacks over the rainforest areas (i.e. Amazonia).

Measuring the Earth’s above-ground biomass is a mission-critical requirement for understanding the global carbon cycle. While high-frequency sensors often saturate in dense forests, the ESA BIOMASS mission’s use of P-band SAR offers a breakthrough by providing the canopy penetration necessary to interact with large woody structures. However, this perspective is clouded by speckle noise a physical corruption that threatens the integrity of the scattering matrix. This work represents a specialized evolution of a polarization-agnostic methodology, transitioning a general purpose framework into a mission specific tool for P-band PolSAR. By adapting a shared convolutional backbone with a dynamic masking data loader, we enable a unified approach to despeckling that maintains the physical and statistical authenticity of the radar return. Training on real multi-temporal data and utilizing a spatio-temporal MuLog-filtered supervision strategy, we demonstrate that this evolved architecture effectively preserves the sub-canopy textures essential for carbon density estimation.

Introduction and Motivation

The estimation of biomass is a critical variable in climate modeling, where the polarimetric signature serves as the primary signal for woody structures. Traditional filters often smooth forest textures, masking small-scale variations. Deep learning offers a solution, but adoption has been hindered by rigid architectures. Our approach builds upon an established polarization-agnostic framework, evolving its core logic to meet the rigorous physical demands of the BIOMASS mission. While BIOMASS data is primarily characterized by its Full-Pol (quad-polarization) nature, we retain the methodology’s underlying flexibility to ensure the network learns universal spatial manifolds that are not tethered to a single modality, thereby increasing the robustness of the feature extraction layers.

Proposed Methodology

The core innovation, inherited and refined for this application, is the ability to handle heterogeneous polarimetric inputs within a single unified architecture. Rather than using fixed-input dimensions, we implement a specialized masking-based data loader.

In this approach, the network is designed to accept a maximum-capacity input tensor (e.g., a 9-channel stack representing the 3 x 3 covariance matrix C). During the data loading phase, a binary mask is applied to the input channels. For the BIOMASS mission's Full-Pol requirements, the mask acts as a gate that ensures all polarimetric relationships are preserved while the shared convolutional filters learn features that are invariant to channel presence. This masking strategy allows the model to leverage spatial textures found across the scattering matrix while remaining structurally flexible a direct evolution of previous work designed to process diverse mission modes without the need for costly retraining.

Training on Real Data: The Multi-Temporal MuLog Strategy

One of the primary critiques of DL-based despeckling is the reliance on synthetic Gaussian noise for training. To address this, we leverage multi-temporal stacks from real data:

Spatio-Temporal Averaging: We take N co-registered images of the same forest track. By averaging them in the temporal domain, we significantly reduce speckle while maintaining the spatial resolution of the original sensor.

MuLog Integration: The resulting images are processed using the Multi-channel Logarithmic framework. This transforms the multiplicative Wishart noise into an additive Gaussian-like space, allowing the CNN to optimize its weights more effectively.

The resulting "pseudo-clean" labels represent the true statistical behavior of the scattering, providing a much more rigorous training signal than synthetic noise models.

Physical Consistency and Loss Functions

A significant portion of our technical development focused on engineering a loss function L that transcends simple pixel-wise comparisons. In the context of P-band PolSAR, a standard Mean Squared Error often leads to over-smoothed images that strip away the texture essential for distinguishing forest types. To avoid this, we utilize a multi-objective optimization strategy centered on the Frobenius loss and the Gradient loss. In PolSAR, the physical properties of the target are encapsulated in the matrix structure; therefore, the network must minimize the distance between the estimated matrix and the pseudo-clean reference across all elements simultaneously. While the Frobenius loss handles the "physics" of the signal, the gradient loss handles the "geography" of the scene. P-band images are characterized by significant texture variations caused by the forest canopy’s uneven structure. To prevent the convolutional layers from blurring these vital edges, the gradient loss penalizes the difference between the spatial gradients of the denoised image and the reference.

Conclusion

Ultimately, this framework provides the rigorous, scalable data processing needed for the BIOMASS mission ground segment. By evolving a polarization-agnostic model, we address the operational reality that while BIOMASS is a Full-Pol mission, the robust feature extraction granted by this methodology drastically reduces computational overhead and increases reliability. This approach demonstrates that we can clean the radar signal without destroying the delicate sub-canopy textures that signal carbon density. By bridging the gap between high-performance deep learning and the strict laws of radar physics, this evolved framework ensures that the BIOMASS mission provides the high-fidelity measurements needed to accurately track the pulse of our planet's forests.

Polarimetric interferometric Synthetic Aperture Radar (PolInSAR) and Tomographic SAR (TomoSAR) are the advanced SAR techniques used to assess canopy height and forest vertical structure and canopy height. PolInSAR exploits the polarimetric dependance of interferometric coherence and phase to constrain parametric scattering models. In contracts, TomoSAR, reconstructs the vertical reflectivity profile using multi-baseline acquisitions, thereby separating the different scattering contribution at different elevations within the canopy.

Comparative studies highlight the potential of P-band SAR for estimating vertical forest structure due to its strong canopy penetration and its ability to interact with canopy, ground, and canopy–ground scattering components. However, existing P-band studies have relied predominantly on airborne sensors, which are limited in spatial coverage and temporal consistency. The recently launched ESA BIOMASS mission, operating at P-band (~435 MHz) and launched in 2025, opens a new era for Earth observation–based retrieval of forest vertical structure over broader spatial and temporal scales.

The main objective of this study is twofold: (1) to retrieve forest vertical structure parameters using using PolInSAR and TomoSAR frameworks; and (2) to quantitatively assess and compare the reliability and accuracy of the forest vertical structure information derived from each framework, using statistical performance indicators, including the coefficient of determination (R²), the Root Mean Square (RMSE), and completeness. Retrieved forest height and related vertical structure parameters are validated against reference measurements obtained from in-situ LiDAR data to evaluate estimation bias, dispersion, and overall retrieval performance.

The study area is located in the Taita Hills, which is surrounded by semi-arid savanna and scrubland at 600-900 m a.s.l., creating strong climatic and structural contrasts over short distances. The focused study area was selected based on the overlapping spatial and temporal coverage of the available ESA BIOMASS, Level 1 Single-look Complex Slant Range (SCS), acquired in 2026 and in-situ LiDAR data with closely matched acquisition dates.

In the framework of PolInSAR, forest height was estimated to use the Random Volume over Ground (RVoG) scattering model within three-stage inversion framework. First, the complex interferometric coherences were derived for multiple polarization channels, and the coherence region was constructed to characterize the volume-ground scattering mixture. Second, the ground phase was estimated by identifying the coherence point corresponding to the surface scattering contribution within the coherence region. Lastly, forest height and extinction parameters were retrieved through inversion of the RVoG model by fitting the observed polarimetric coherences to the theoretical volume-over-ground coherence formulation.

The TomoSAR processing includes the implementation of single- and fully polarized CAPON, Multiple Signal Classification (MUSIC) and Expectation-Maximization (EM) followed by geometric-based volume delineation techniques to derive height cubes, and reflectivity maps. The forest vertical structure information derived from each technique then is validated against in-situ LiDAR data, and the best performance workflow is identified.

The result of this study will improve the understanding of the applicability of forest vertical structure information retrieved from TomoSAR and PolInSAR techniques for forest characterization, with focus on Taita Hills. Furthermore, the proposed frameworks will provide an EO-based solution for continuous forest height and vertical structure characterization, supporting policy and monitoring initiatives, such as the European Union Deforestation Regulation (EUDR).

Synthetic Aperture Radar (SAR) imagery is inherently corrupted by speckle, a multiplicative noise arising from coherent signal interaction. Speckle degrades radiometric accuracy, visual interpretability, and downstream tasks such as classification and parameter retrieval. It must be suppressed across all SAR modalities, including single-channel SAR, Polarimetric SAR (PolSAR), Interferometric SAR (InSAR), and Tomographic SAR (TomoSAR), while preserving both amplitude and phase information in complex-valued data.

Local mean filters reduce speckle through spatial averaging but sacrifice spatial resolution and edge detail. Deep learning approaches achieve strong despeckling performance but require ground truth training data, which particularly lacks in case of multi-temporal data, thus limiting their generalizability to unseen sensor configurations and acquisition conditions. Non-Local Means (NLM) addresses these limitations by estimating each pixel as a weighted average of similar pixels within a broader search neighbourhood, where similarity is assessed at the patch level. NLM requires no training data and preserves structural detail, making it particularly attractive for SAR despeckling.

Existing NLM implementations for different SAR modalities rely on different similarity metrics, such as Log-Euclidean, Wishart, and Kullback-Leibler distances. Each metric assumes a specific statistical model and data structure. Single-channel SAR filters differ from PolSAR or multi-channel (InSAR/TomoSAR) filters. Although some unified frameworks have been proposed, they mostly depend on explicit statistical modelling of each data type.

This work introduces a new Non-Local Means framework for adaptive despeckling across different SAR modalities, from single-channel to multi-channel data. The approach is inspired by neuroscience, where biological neurons encode stimulus intensity or input features, such as polarimetric or interferometric phase, through spike timing. Two stimuli are considered similar when their spikes occur close in time. This coincidence-based similarity assessment is efficient and robust to noise. The proposed pipeline operates within a unified framework applicable to multiple SAR data modalities.

The method has been evaluated on P-band coherency matrices from ESA’s BIOMASS mission over Mumbai, India, and initial results are competitive with established state of the art techniques such as MULOG. A more detailed quantitative analysis is ongoing, and the framework will next be assessed on interferometric, tomographic, and multi-temporal SAR datasets. As SAR data volumes continue to grow, computationally intensive model-based approaches face scalability challenges. In contrast, the proposed method is designed to remain computationally efficient and lightweight.

The main contribution of this work is the introduction of a simple, model-free, and computationally efficient similarity metric for identifying similar pixels in SAR imagery. The metric is applicable across different SAR modalities while maintaining low computational complexity.

Landslides represent a major geohazard in high mountainous regions of Asia, where extreme precipitation events such as cloudbursts frequently trigger large‑scale slope failures. Climate change is expected to further intensify the frequency and magnitude of these events, increasing risks to infrastructure, agriculture, and sustainable development. The Eastern Himalayas, particularly Arunachal Pradesh (India), are highly susceptible to rainfall‑induced landslides, which often remain undocumented yet cause significant downstream impacts, including artificial dam formation and secondary flooding in lowland areas.

Mudslides, landslides, and severe convective storms occur frequently in Arunachal Pradesh. Although more than 70% of the region is forest‑covered and sparsely populated—limiting direct loss of life—the impacts on infrastructure and agriculture are substantial, resulting in considerable economic and environmental losses, especially during the monsoon season. The region’s geological setting is dominated by the Eastern Himalayan syntaxis, characterized by a complex assemblage of Precambrian to Tertiary lithologies organized into four major tectonic belts: the Sub‑Himalayan, Lesser Himalayan, Higher Himalayan, and Tethys Himalayan domains. While fragile lithology and active geodynamics provide preparatory conditions for slope instability, landslide occurrence is strongly concentrated during the monsoon months (June–September - JJAS), when intense rainfall acts as the primary triggering factor. Annual precipitation ranges from approximately 2,000 to 5,000 mm, with up to 60% falling during the JJAS period, frequently inducing shallow slope failures in steep, geologically young terrain.

In densely vegetated mountainous environments, conventional pre‑failure deformation monitoring using Synthetic Aperture Radar (SAR) interferometry is hindered by low coherence, restricting analyses largely to post‑event conditions following vegetation removal and soil exposure. In this study, we assess the feasibility of integrating multitemporal Sentinel‑1 InSAR techniques, including Persistent Scatterer Interferometry (PSI) and Small Baseline Subset (SBAS) methods, with P‑band BIOMASS tomography data to characterize post‑failure landslide dynamics. While InSAR provides millimeter‑scale deformation measurements on exposed colluvial deposits, BIOMASS tomography offers complementary insight into the three‑dimensional structure of land cover and subsurface scattering mechanisms.

Arunachal Pradesh was selected as a test site due to its high landslide incidence, extensive documentation in the literature, availability of Sentinel‑1 data, and the unique access to BIOMASS tomographic datasets as of early 2026. The proposed integrated approach demonstrates strong potential for improving post‑event landslide characterization in heavily vegetated mountainous regions.

The Copernicus Precise Orbit Determination (CPOD) Service has been a cornerstone of the Copernicus Earth Observation programme, routinely delivering high-accuracy orbit and auxiliary products for Sentinel-1, Sentinel-2, Sentinel-3 and Sentinel-6. Through sustained operational experience in GNSS-based precise orbit determination (POD) for Low Earth Orbit (LEO) satellites, CPOD has developed robust processing standards relying on multi-frequency GNSS measurements. Building on this knowledge and added value for the Copernicus missions, a proposal was submitted in response to the Biomass In-Orbit Commissioning (IOC) Call for Proposals to assess the feasibility of generating an independent precise orbit solution for ESA’s Biomass mission.

ESA provided access to GNSS data acquired by the onboard LEORIX receiver. Biomass is equipped with a single-frequency receiver tracking GPS and Galileo satellites on L1, delivering C/A code and carrier phase observations. Unlike the dual-frequency configurations available on other Copernicus Sentinel missions, this single-frequency setup prevents the direct formation of ionosphere-free linear combinations. Consequently, first-order ionospheric effects cannot be eliminated using standard approaches, posing specific challenges for high-accuracy POD in a LEO environment.

To overcome this limitation, a POD strategy based on the GRAPHIC (Group and Phase Ionospheric Correction) combination has been implemented. By combining code and phase measurements, the GRAPHIC approach mitigates first-order ionospheric effects while preserving the precision of carrier phase observations. The processing chain includes detailed force modelling (Earth gravity field, third-body perturbations, solar radiation pressure and atmospheric drag), observation modelling (antenna phase centre corrections, relativistic effects and Earth orientation parameters, and a dynamic orbit determination scheme based on batch least-squares estimation. Particular emphasis is placed on the challenges inherent to single-frequency GNSS processing, including reduced observability and sensitivity to ionospheric variability.

Preliminary results demonstrate that a dynamically consistent precise orbit solution can be achieved. Orbit convergence behaviour and internal consistency metrics confirm the viability of the approach. However, several open issues remain, including unexpected features in code and phase residuals and sensitivities in the dynamic parameter estimation. These effects are analysed and discussed to identify modelling limitations and areas requiring further refinement.

Finally, a Test Data Set (TDS) of Biomass POD products has been generated and will be made available to the community for independent assessment. Although the current solutions do not yet reach the maturity of established multi-frequency CPOD products, they represent a first step toward independent precise orbit products for Biomass and provide a foundation for continued improvement and collaboration.

Spaceborne SAR tomography has entered an operational phase with the BIOMASS missions providing multi-baseline and repeat-pass interferometric observations over forested regions. A central assumption underlying the tomographic inversion schemes is that the vertical reflectivity distribution remains invariant during the acquisition interval. While this hypothesis is often justified for airborne campaigns acquired within a single flight, it becomes increasingly fragile for repeat-pass satellite systems, even when temporal baselines are short. Subtle variations in canopy structure induced by wind, moisture dynamics, or precipitation can modify the effective scattering distribution, leading to systematic phase inconsistencies across interferometric pairs. These inconsistencies propagate into tomographic reconstructions, degrading vertical focusing and biasing phase-center estimation.

In this contribution, we revisit the concept of interferometric closure phase as a diagnostic observable for assessing temporal reliability in SAR tomography. Closure phase, defined over closed loops of interferometric measurements, is traditionally interpreted in InSAR as an indicator of decorrelation or multi-scattering effects. However, in volumetric media such as forests, non-zero closure phase arises naturally from the coherent superposition of multiple vertical scattering contributions. Consequently, the mere presence of closure phase does not imply temporal instability. We propose instead to exploit the structural symmetries of closure phase under a stable reflectivity profile as a consistency constraint for tomographic stacks. We analyze the dependence of closure phase on forest structural parameters and TomoSAR geometry. The analysis reveals that, for a temporally stable vertical reflectivity profile, closure phases exhibit invariant behavior. Based on this principle, we introduce a circular dispersion metric. Unlike conventional interferometric coherence, which conflates structural decorrelation and temporal effects, the proposed metric is designed to be insensitive to purely volumetric decorrelation under stationary conditions. Instead, it selectively responds to deviations from reflectivity-profile invariance, thereby offering a targeted indicator of temporal reliability for tomographic inversion.

Beyond diagnostic analysis, we demonstrate how closure-phase residuals can be mapped onto pairwise phase correction terms for the interferometric data. The resulting correction framework enforces loop consistency. This approach aims to mitigate systematic phase biases that accumulate across interferometric pairs and to restore internal consistency prior to tomographic inversion.

The methodology is assessed using airborne P-band data and also BIOMASS acquisitions over forest regions. The closure-based dispersion index highlights areas impacted by temporal variability, in agreement with coherence patterns but with enhanced selectivity. In pixels exhibiting low dispersion, tomographic reflectivity profiles before and after data correction remain nearly indistinguishable. In contrast, higher-dispersion areas show moderate redistribution of vertical power after correction, including adjustments in sidelobe structure and empowering the secondary scatterer (e.g. ground or canopy) and, in some cases, shifts of the dominant phase center. Importantly, the correction does not introduce artificial structural distortions, but rather reconciles inconsistencies already present in the interferometric stack.

The results indicate that closure-phase offers a physically interpretable and geometry-aware tool for evaluating the temporal consistency of SAR tomographic stacks. As multi-baseline acquisitions from missions such as BIOMASS and Sentinel-1 continue to expand, closure-driven consistency analysis may play a key role in quality control, data screening, and pre-inversion process for operational SAR tomography