15 May 2025 in IEEE Transactions on Geoscience and Remote Sensing
There is a growing trend towards making Synthetic Aperture Radar (SAR), SAR Interferometry (InSAR), and their applications more accessible to end users. Directly delivering geocoded, co-registered, and flattened SLC data (GSLC) eliminates the need for the complex geocoding and co-registration procedures which require professional domain knowledge and software. GSLC products dramatically simplify InSAR processing flows, making InSAR products easily available to a wide range of users. However, challenges still exist for SAR/InSAR analysis using GSLC datasets. In this paper, we analyze the feasibility of using GSLC for deformation measurements based on offset tracking, both in theory and practice. We find that correct GSLC offset tracking requires the input GSLCs to be 1) unflattened, 2) deramped, and 3) adequately sampled. We also show that the direct result from GSLC offset tracking is a projection of displacement in the slant range and azimuth directions. We can transform the offset measurement from GSLCs to the deformation field, but the current transformation relation is not precise enough. This research may help deepen the understanding of GSLCs and their applications.
30 March 2026 in Journal of Geophysical Research: Earth Surface
Rock glacier kinematics are indicators of alpine climate change, yet the relative influence of thermal and hydrological processes on their motion remains poorly constrained. This uncertainty limits our ability to predict how these permafrost landforms will respond to ongoing climate change. Here, we present a mechanistic case study of seven active rock glaciers in Wrangell–St. Elias National Park, Alaska (USA) was selected to span contrasting aspects, snow insulation, and kinematic behavior. Using multi‐temporal Sentinel‐1 interferometric synthetic aperture radar (InSAR) data from 2018 to 2022, we derived high‐resolution (∼30 m) velocity maps and deformation time series by combining ascending and descending observations and assuming predominantly downslope motion. Independent Component Analysis (ICA) was applied to separate long‐term trends from seasonal signals, enabling the characterization of deformation components associated with environmental forcing. We then quantified the water infiltration effects by estimating hydraulic conductivity from the observed lag times between snowmelt and seasonal deformation peaks. Subsurface thermal regimes were simulated using the CryoGrid land surface and permafrost model. Results reveal that rock glaciers with greater solar radiation and warmer subsurface conditions exhibit enhanced meltwater infiltration and stronger seasonal deformation, while colder, shaded sites remain relatively stable. This combined InSAR, ICA, hydraulic, and thermal analysis highlights how hydrological and thermal controls govern spatial variability in rock glacier kinematics, offering insights into the sensitivity of permafrost landforms to ongoing climate change.
28 May 2026 in IEEE Transactions on Geoscience and Remote Sensing
Deep learning is revolutionizing the way InSAR data are processed. Supervised deep learning models trained on synthetic datasets have demonstrated strong performance according to conventional evaluation metrics. However, in practice, many supervised InSAR phase denoising filters tend to oversmooth fringe patterns and generate ghost fringes in decorrelated regions, raising concerns about the reliability of models trained solely on synthetic data in real-world applications. In this work, we introduce Noise2Fringe, an unsupervised learning framework that enables the direct training of denoising filters on noisy interferograms, eliminating the need for clean references, while achieving performance comparable to that of supervised models. This framework allows a fair comparison between models trained exclusively on real versus synthetic data, independent of the training scheme. On synthetic benchmarks, models trained on real data appear to underperform compared to those trained on synthetic data. However, in real-world interferograms, the opposite holds: synthetic-trained models oversmooth and generate ghost fringes, whereas real-trained models avoid these artifacts. These findings reveal (i) a much larger gap between synthetic and real data than previously assumed, (ii) an overestimation of the effectiveness of synthetic-trained models, and (iii) deficiencies in current evaluation methods that rely heavily on simulated benchmarks. We further find that incorporating auxiliary information, such as the Amplitude Dispersion Index, into the input can enhance denoising performance. Overall, this study demonstrates that models trained solely on synthetic data may be far less reliable in real-world applications than suggested by their benchmark results, and emphasizes that generating more realistic training datasets should take precedence over designing increasingly complex model architectures.
8 December 2025 in TechRxiv
Deep learning is revolutionizing the way InSAR data are processed. Supervised deep learning models trained on synthetic datasets have demonstrated strong performance according to conventional evaluation metrics. However, in practice, many supervised InSAR phase denoising filters tend to oversmooth fringe patterns and generate ghost fringes in decorrelated regions, raising concerns about the reliability of models trained solely on synthetic data in real-world applications. In this work, we introduce Noise2Fringe, an unsupervised learning framework that enables the direct training of denoising filters on noisy interferograms, eliminating the need for clean references, while achieving performance comparable to that of supervised models. This framework allows a fair comparison between models trained exclusively on real versus synthetic data, independent of the training scheme. On synthetic benchmarks, models trained on real data appear to underperform compared to those trained on synthetic data. However, in real-world interferograms, the opposite holds: synthetic-trained models oversmooth and generate ghost fringes, whereas real-trained models avoid these artifacts. These findings reveal (i) a much larger gap between synthetic and real data than previously assumed, (ii) an overestimation of the effectiveness of synthetic-trained models, and (iii) deficiencies in current evaluation methods that rely heavily on simulated benchmarks. We further find that incorporating auxiliary information, such as the Amplitude Dispersion Index, into the input can enhance denoising performance. Overall, this study demonstrates that models trained solely on synthetic data may be far less reliable in real-world applications than suggested by their benchmark results, and emphasizes that generating more realistic training datasets should take precedence over designing increasingly complex model architectures.