Speaker
Description
Accurate and spatially comprehensive monitoring of changes in glacier elevation is crucial for understanding glaciological processes, such as snow depth and geodetic mass balance, as well as snow and ice dynamics. In-situ measurements, while precise, are spatially limited and logistically challenging. Alternatively, LiDAR sensors can provide highly accurate measurements but are constrained by limited spatial and temporal coverage due to sparse sampling and low revisit frequencies. In this context, interferometric synthetic aperture radar (InSAR) systems offer a valuable alternative for assessing those changes by providing high-resolution data over extended areas, independent of weather conditions or illumination. In this work, we address key associated challenges for deriving accurate elevation changes from InSAR Digital Surface Models (DSMs), specifically focusing on mitigating errors introduced by radar wave penetration into the snowpack and ensuring precise DSM calibration. Current approaches rely on simplified assumptions regarding the compensation for the penetration depth, often averaging DSMs acquired under varying conditions, which introduces substantial uncertainty, particularly in complex alpine environments. Similarly, accurate DSMs require a precise calibration due to the presence of residual offsets and tilts caused by a limited knowledge of the baseline, which is expensive and impractical for large-area monitoring. This contribution outlines a novel, end-to-end framework for robust glacier elevation change estimation from InSAR data. The core strategy is to compare precisely corrected DSMs to quantify seasonal and interseasonal elevation changes. A critical innovation lies in the correction through a three-stage methodology designed to tackle the limitations of existing techniques. First, we use an automated and reference-free approach for residual calibration of InSAR-derived DSMs by relying on natural calibration tie-points. This method eliminates the dependency on costly and potentially unavailable GPS or LiDAR ground truth data, while also compensating for residual offsets and tilts. Second, we implement a data-driven, deep learning-based approach to accurately estimate and correct for the radar penetration bias in snow-covered areas. Finally, after compensating the DSMs for both the calibration offset and tilts and the radar penetration bias, these corrected DSMs are used to perform DSM differencing to retrieve elevation changes. Preliminary results from the bistatic X-band TanDEM-X system indicate that the proposed framework enables accurate estimation of glacier elevation changes at decimeter-level accuracy. Additionally, we assess the potential for transferring the proposed approach to C-band using historical SRTM and available TanDEM-X data. This activity is currently under development within the ESA SUPSAR-Snow project and is potentially relevant for future InSAR bistatic missions such as the upcoming ESA Harmony mission. The proposed framework enables reliable, large-scale estimation of glacier elevation changes from InSAR DSM differencing, improving the current understanding of cryospheric processes and preparing for future InSAR spaceborne missions.