Conference Agenda

Tropospheric delays are a major source of error in radar remote sensing applications such as InSAR. Access to state of the art models and computing corrections requires in-depth expertise and resources. We present an global tropospheric delay dataset derived from ECMWF-HRES, developed under NASA's Observational Product For End Users from Remote Sensing Analysis (OPERA) project with the archive starting in 2015 and kept up to date. The dataset provides zenith delays every six hours and is validated against GNSS-derived ZTD measurements over the period 2016-2025. It is SAR-agnostic and supports multiple satellite and airborne radar missions. TROPO products are computed using the RAIDER software package, where products are produced by OPERA typically within hours from new model data being available, and where products are distributed free and open as cloud optimized netcdf datacubes to the broader community through the NASA ASF DAAC archives. We will provide an overview of the dataset, its validation results, and demonstrate its utility by applying it to OPERA DISP products, showing reduced long-wavelength noise and improved time-series stability.

Atmospheric delays remain one of the primary error sources affecting the accuracy of deformation measurements derived from Interferometric Synthetic Aperture Radar (InSAR). Variations in the troposphere introduce phase artifacts that can significantly degrade the interpretation of surface displacement signals. To improve the reliability of InSAR observations, independent atmospheric measurements are required. Global Navigation Satellite Systems (GNSS) provide continuous estimates of Zenith Tropospheric Delay (ZTD), which can support robust atmospheric correction of SAR interferograms. However, the spatial density of geodetic-grade GNSS networks is often limited by their high installation and operational costs. To address the need for scalable atmospheric monitoring, this study investigates the capability of low-cost GNSS systems to reliably estimate ZTD, ultimately evaluating their potential to densify atmospheric monitoring infrastructures. Specifically, this research examines the accuracy of precise point positioning with ambiguity resolution (PPP-AR) using the low-cost dual-frequency GNSS receiver, u-blox ZED-F9P. While previous studies have demonstrated the utility of this receiver for positioning applications, its performance for tropospheric delay estimation, particularly when using PPP-AR solutions, requires rigorous validation. Static experiments were conducted over three consecutive days using three-hour observation sessions with a 30-second sampling interval. The low-cost ZED-F9P receiver was deployed alongside a geodetic-grade reference receiver to establish an accurate baseline for validation, ensuring identical atmospheric conditions for both instruments. Observations were processed using PPP with GPS and Galileo data, employing final precise orbit and clock products. Processing was performed using both the Net_Diff software and the Natural Resources Canada CSRS-PPP online service. The results demonstrate that the low-cost receiver successfully achieved fixed ambiguity solutions for a substantial portion of the observation period, with ZTD estimates agreeing within 3–5 mm root mean square error relative to the geodetic-grade reference. This level of agreement is particularly noteworthy given the substantial cost difference between the two systems and suggests that modern low-cost receivers can approach the performance of geodetic equipment for atmospheric sensing applications. These findings indicate that low-cost GNSS receivers are capable of estimating tropospheric delays with accuracy sufficient for operational atmospheric monitoring and InSAR correction. Integrating such observations into InSAR processing workflows could enable scalable and cost-effective monitoring networks, particularly in regions where geodetic infrastructure is sparse, thereby improving the reliability of geohazard and geodynamic analyses.

Dispersive ionospheric phase can severely bias deformation signals in long-wavelength InSAR, for example in L-band missions. Conventional range split-spectrum methods (SSM) often require strong spatial smoothing to obtain reliable dispersive ionospheric phase estimates, because the center-frequency separation between range sub-bands is limited by acquisition bandwidth. In narrow-band modes this separation becomes small, reducing sensitivity to the dispersive term and amplifying noise in the SSM estimate. For example, in the 14‑MHz ALOS‑2 ScanSAR mode the usable sub-band separation is only a few MHz. To mitigate these limitations, we present an operator-learning approach to estimate the dispersive ionospheric phase from range sub-band interferograms using a Fourier Neural Operator (FNO). Unlike convolutional neural networks (CNNs) that learn mappings tied to a specific pixel grid, the FNO is designed to learn an operator between function spaces. It parameterizes this operator using Fourier-domain (spectral) convolutions, which efficiently capture multiscale and nonlocal structure. This operator-learning formulation makes the model naturally discretization- and resolution-independent, enabling the same trained model to be applied across different image sizes and pixel spacings without retraining.

To avoid reliance on real-data labels, we train the model exclusively on simulated L-band interferograms, where the ground-truth dispersive ionospheric phase is known. The simulated interferograms include deformation, stratified and turbulent tropospheric delays, long-wavelength components, and dispersive ionospheric phase, and are further corrupted with realistic artifacts such as random phase noise, localized pseudo-unwrapping errors, and spatially contiguous missing-data gaps. In particular, the ionospheric component is generated by simulating a TEC field via a weighted superposition of plane waves and converting it to ionospheric phase using a ΔTEC-to-phase conversion at the radar center frequency. We randomize the amplitude and spatial correlation length of each component to generate diverse training conditions.

The network is trained to map (i) a pair of low- and high-frequency sub-band interferograms and (ii) a coherence-based mask (coherence < 0.1) to the dispersive ionospheric phase component only. By explicitly providing a low-coherence mask, the model can handle spatially contiguous data gaps that often degrade split-spectrum estimates. Using simulation provides physically consistent supervision and reduces the risk of learning spurious phase contributions, while also eliminating the substantial effort required to build and quality-control labeled real-data training sets.

We then applied the trained model to ALOS-2 interferograms acquired in the 84-MHz mode, enabling direct comparison with high-quality ionospheric phase estimates from conventional SSM. We evaluate three real-data cases from northeastern Siberia, the Noto Peninsula, Japan, and Fairbanks, Alaska, spanning markedly different coherence patterns and geographic settings. Performance is quantified using mean squared error (MSE) and structural similarity (SSIM) with the SSM-derived ionospheric phase as a reference. For Siberia (mean coherence of the sub-band interferogram pair: 0.76), the model achieves an MSE of 0.08 rad and an SSIM of 0.97; for Fairbanks (0.19), it achieves an MSE of 0.37 rad and an SSIM of 0.92. Metrics are computed over the full interferogram, including low-coherence regions. Across these cases, the FNO-based estimates are consistent with SSM estimates while requiring only standard multilooking and no additional spatial smoothing, because denoising is handled implicitly by the learned estimator rather than by case-by-case tuning of filters.

We are now extending the architecture to take acquisition center frequency as an explicit input, aiming to support narrow-band modes and to improve cross-mission applicability to L-band systems including ALOS‑4, NISAR, and the future ROSE‑L. As a preliminary test of cross-mission applicability, we applied the frequency-conditioned model to a NISAR L-band dataset over Ethiopia (20 MHz mode) and obtained qualitative ionospheric-phase estimates that capture the dominant spatial patterns. An important next step is to quantify the estimator’s operating limits and uncertainty under severe noise, decorrelation, and data gaps. Updated results will be presented.

Differential Synthetic Aperture Radar Interferometry (DInSAR) is a well-established technique for investigating surface deformation, enabling the detection and measurement of ground motions at centimeter-to-millimeter scale over large areas [1]. In recent years, space agencies have strongly supported the development of new satellite SAR systems operating at L-band (about 23 cm wavelength), which can play a key role in interferometric monitoring of the Earth’s surface. Indeed, the use of low-frequency SAR sensors offers significant advantages in terms of interferometric coherence improvements in several scenarios as, for instance, in rural and vegetated areas. However, the effective exploitation of the L-band interferometric measurements peculiarities requires careful consideration of possible ionospheric effects, which can significantly affect both the phase and the amplitude of the radar signals. In particular, they can cause decorrelations due to azimuthal shits (azimuth streaks) of the SAR images and spurious Line of Sight (LOS) phase contributions within the interferograms, thus reducing the accuracy of the displacement estimates in DInSAR applications.

In order to mitigate ionosphere-induced LOS phase contributions in L-band DInSAR measurements, the range split-spectrum (RSS) technique, jointly with specific noise filtering procedures, is widely used. The rationale of the RSS technique is based on splitting the range bandwidth of SAR images into multiple sub-bands and on properly combining the interferometric phase retrieved from each sub-band; this permits to estimate (following an appropriate filtering operation) and to subsequently correct the range dispersive signal component that is directly related to the ionospheric phase contribution [2],[3].

In recent years, different RSS implementation approaches coexist in the literature [2]-[5]. Their formulations mainly differ in terms of the adopted noise filtering procedures and phase unwrapping strategies. However, although these different formulations are often presented as formally equivalent, in realistic DInSAR scenarios affected by the investigated ground deformation, their performance can be significantly different. Consequently, the straightforward exploitation of these solutions, without a careful assessment of the underlying assumptions and operational choices, can lead to corrupted estimates of both the ionospheric phase contributions and the retrieved deformation signals.

This work will provide an analysis on the exploited RSS-based approaches for ionospheric phase estimation in DInSAR applications. Specifically, through a brief overview of the currently adopted RSS formulations, it will highlight their underlying assumptions, operational differences, and practical implications in realistic DInSAR scenarios. Particular attention will be devoted to investigating the impact of deformation signals and possible noise amplification on the reliability of the estimated ionospheric phase. Overall, the aim of the presented work is to try to clarify the conditions under which different RSS formulations can be considered practically equivalent, as well as the scenarios where their behaviour may significantly diverge. The resulting discussion is intended to finally identify a consensus approach for the RSS-based implementations in operational L-band DInSAR processing chains.

Furthermore, a final discussion will be devoted to presenting an innovative RSS-based processing strategy based on properly exploiting the degree of freedom deriving from the above-mentioned splitting of the range bandwidth SAR images into multiple sub-bands and on its possible implications to address the critical role of the noise filtering procedures for the ionospheric phase component retrieval.

The overall analysis will be carried out by exploiting the L-band SAR images acquired by the Argentine SAOCOM-1 constellation.

REFERENCES

[1] A. K. Gabriel, R. M. Goldstein, and H. A. Zebker, “Mapping small elevation changes over large areas: Differential interferometry,” J. Geophys. Res., vol. 94, no. B7, pp. 9183–9191, 1989.

[2] P.Rosen, S. Hensley and C. Chen, "Measurement and mitigation of the ionosphere in L-band interferometric SAR data," in IEEE Radar Conf., 2010.

[3] G. Gomba, A. Parizzi, F. D. Zan, M. Eineder and R. Bamler, "Toward operational compensation of ionospheric effects in SAR interferograms: The split-spectrum method," IEEE Trans. Geosci. Remote Sens, vol. 54, no. 3, p. 1446–1461, 2016.

[4] G. Gomba, "Estimation of ionosphere-compensated azimuth ground motion with sentinel-1." EUSAR 2018; 12th European Conference on Synthetic Aperture Radar. VDE, 2018.

[5] Wegmüller, Urs, et al. "Estimation and compensation of the ionospheric path delay phase in PALSAR-3 and NISAR-L interferograms." Atmosphere 15.6 (2024): 632.

Over the past few decades, Differential Synthetic Aperture Radar Interferometry (DInSAR) has established itself as a pivotal Earth Observation technique for ground motion detection and monitoring. DInSAR is a microwave remote sensing technique capable of measuring surface displacements with centimetric/millimetric accuracy over wide areas (hundreds of km), regardless of weather conditions or solar illumination and at affordable costs. Thanks to these characteristics, nowadays it is routinely employed in operational monitoring frameworks across a broad spectrum of geohazard scenarios — including volcanic unrest, seismic deformation, landslide dynamics, and subsidence — as well as in the structural monitoring of the built environment [1], [2].

The current SAR landscape is characterized by an ever-growing availability of data from multiple satellite missions operating at different wavelengths and acquisition modes. Alongside the free and open-access European Sentinel-1 C-band constellation, specifically designed for interferometric applications, a number of other systems significantly contribute to the available data pool — including the Italian COSMO-SkyMed first and second generation constellations (CSK and CSG, X-band) and the Argentinian SAOCOM mission (L-band). This rich multi-sensor scenario has fostered the development of automated, cloud-based processing chains capable of handling and effectively exploiting these large data volumes [3], [4]. Furthermore, both software tools and ancillary external datasets aimed at applying suites of corrections to SAR data or interferometric products are rapidly evolving, enabling increasingly accurate DInSAR measurements [5].

Currently one of the main challenges in retrieving accurate DInSAR measurements is the presence of the Atmospheric Phase Screen (APS), arising from the temporal and spatial variability of atmospheric properties. Variations in radar signal propagation velocity through the inhomogeneous atmosphere introduce delay signals that can easily be mistaken for surface deformation, making it particularly challenging to separate and filter out the APS contribution from displacements within DInSAR measurements [6].

A well-established approach to APS mitigation relies on the use of meteorological data provided by Numerical Weather Prediction (NWP) models. Among such data, the most widely used are the ERA-5 reanalysis products from ECMWF, which provide estimates of key atmospheric parameters — pressure, temperature, and humidity profiles at multiple vertical levels — on a global scale with hourly temporal sampling and a horizontal spatial resolution of approximately 31 km [7], [8], and the GACOS (Generic Atmospheric Correction Online Service) products, which combine NWP data with GNSS-derived tropospheric delays to provide interpolated correction maps [9]. More recently, the ETAD (Enhanced Temporal and Atmospheric Delay) product, developed within the ESA Sentinel-1 mission framework, has also become available [10], [11]. However, in practice, thanks to their global coverage and immediate accessibility, ERA-5 data remain the most widely used for NWP-based atmospheric correction in DInSAR applications. Nevertheless, the main limitation of ERA-5 data lies in their coarse spatial resolution relative to that of DInSAR products, which significantly constrains the effectiveness of the atmospheric correction. Indeed, ERA-5 atmospheric corrections have proved effective in retrieving the stratified APS component — the large-scale contribution correlated with the scene topography — while failing to capture smaller-scale atmospheric features, such as the turbulent component [11]. Because the latter can significantly contaminate deformation signals, it represents a critical source of error in DInSAR measurements.

In this work we address the exploitation of the recently released CERRA (Copernicus European Regional ReAnalysis) dataset for APS estimation in DInSAR products. CERRA is a regional atmospheric reanalysis product developed by the Copernicus Climate Change Service (C3S), covering Europe with a horizontal spatial resolution of approximately 5.5 km and a 3-hourly temporal sampling, representing a significant resolution improvement over global reanalysis products such as ERA-5. In the final presentation of the work we will show an extensive experimental analysis to evaluate the performance of CERRA-based APS correction on both DInSAR interferograms and displacement time series. To this end, we consider two large Sentinel-1 datasets spanning areas with diverse morphological and atmospheric characteristics, processed by means of the P-SBAS technique [3]. The first test site is the Mt. Etna volcanic area (~500 acquisitions, ascending orbit, April 2015 – Jan 2026), characterized by a complex and steep topography that generates a pronounced stratified APS component strongly correlated with the local relief. The second test site is the Ligurian coast (410 acquisitions, ascending orbit, July 2016 – June 2025), a coastal area subject to highly variable atmospheric conditions and frequent precipitation events, where the turbulent APS component is expected to be dominant. This choice of complementary test sites allows a comprehensive and robust evaluation of the CERRA-based atmospheric corrections across diverse APS regimes, providing insights into the added value of high-resolution reanalysis data under different geomorphological and climatic conditions. The APS filtering performance will be assessed through appropriate statistical metrics, such as variogram and standard deviation analyses [11], applied both to interferograms and displacement time series, aimed at quantifying the impact of the improved spatial resolution of CERRA data on the retrieval of the different atmospheric delay contributions. The obtained results will be systematically compared against those derived from ERA-5-based atmospheric corrections, providing a quantitative benchmark against the current standard for NWP-based atmospheric correction in DInSAR applications.

[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. 1, pp. 169-209, 2000.

[2] D. Massonnet and K. L. Feigl, "Radar interferometry and its application to changes in the Earth's surface," Reviews of geophysics, vol. 36, no. 4, pp. 441-500, 1998.

[3] M. Manunta et al., "The parallel SBAS approach for Sentinel-1 interferometric wide swath deformation time-series generation: Algorithm description and products quality assessment," IEEE Transactions on Geoscience and Remote Sensing, vol. 57, no. 9, pp. 6259-6281, 2019.

[4] I. Zinno et al., "National scale surface deformation time series generation through advanced DInSAR processing of sentinel-1 data within a cloud computing environment," IEEE Transactions on Big Data, vol. 6, no. 3, pp. 558-571, 2018.

[5] Mulder, G., van Leijen, F. J., Lopez-Dekker, P., and Hanssen, R. F., “RIPPL, a Python-based InSAR stack and tropospheric delay software package”, Computers and Geosciences, vol. 207, Art. no. 106069, Elsevier, 2026. doi:10.1016/j.cageo.2025.106069.

[6] A. Parizzi, R. Brcic, and F. De Zan, "InSAR performance for large-scale deformation measurement," IEEE Transactions on Geoscience and Remote Sensing, vol. 59, no. 10, pp. 8510-8520, 2020.

[7] R. Jolivet, R. Grandin, C. Lasserre, M. P. Doin, and G. Peltzer, "Systematic InSAR tropospheric phase delay corrections from global meteorological reanalysis data," Geophysical Research Letters, vol. 38, no. 17, 2011.

[8] Z. Hu and J. J. Mallorquí, "An accurate method to correct atmospheric phase delay for InSAR with the ERA5 global atmospheric model," Remote Sensing, vol. 11, no. 17, p. 1969, 2019.

[9] Yu, C., Li, Z., Penna, N. T., & Crippa, P. (2018). Generic atmospheric correction model for Interferometric Synthetic Aperture Radar observations. Journal of Geophysical Research: Solid Earth, 123(10), 9202–9222.

[10] C. Gisinger et al., "The Extended Timing Annotation Dataset for Sentinel-1—Product Description and First Evaluation Results," in IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1-22, 2022, Art no. 5232622, doi: 10.1109/TGRS.2022.3194216.

[11] I. Zinno, F. Casamento and R. Lanari, "On the Exploitation of the ETAD Product for Filtering Out the Atmospheric Phase Screen From Medium Resolution DInSAR Measurements: An Extensive Performance Analysis," in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 18, pp. 712-727, 2025, doi: 10.1109/JSTARS.2024.3488494.