Speaker
Description
Snowpack characterization is of significant interest for hydrological and climate applications, particularly in mountainous environments, where conventional remote sensing techniques may suffer from limited spatial resolution. Synthetic Aperture Radar (SAR) tomography provides the capability to resolve scattering contributions along the vertical direction and therefore represents a promising technique for retrieving both snowpack height and internal stratigraphy.
This work presents the results of an experimental X-band UAV-SAR campaign conducted in February 2026 in Cheneil, Valle d'Aosta, Italy. Thirty approximately parallel trajectories were flown at different altitudes using an FMCW radar operating at 10 GHz with a bandwidth of 400 MHz, providing the spatial baseline diversity required for tomographic imaging. The reconstructed snowpack structure is validated against manual snowpit observations and independent ground-based radar measurements. The in-situ measurements indicate a dry and vertically stratified snowpack with a total thickness of approximately 0.95 m.
Conventional tomographic processing based on beamforming and spectral estimation techniques, such as Capon and MUSIC, allows the ground and snow-air interfaces to be identified, providing snow-depth estimates in good agreement with the in-situ measurements. However, the available vertical aperture results in a resolution comparable to the total snowpack thickness, preventing a reliable characterization of its internal stratigraphy.
By modeling the interfaces between different snow layers as planar distributed scatterers located at different heights, a covariance-based compressive sensing approach is introduced to exploit the expected sparsity of the vertical scattering power profile and achieve super-resolution in the vertical direction. Covariance estimates obtained from the multi-baseline SAR stack are improved by jointly exploiting observations acquired at multiple squint angles, thereby increasing the effective number of looks while preserving the tomographic information. The vertical scattering power profile is then retrieved through sparse reconstruction directly from the estimated covariance matrix.
Application to the Cheneil dataset shows a substantial improvement in the separation of vertical scattering contributions compared with conventional tomographic processing. In addition to identifying the ground-snow and snow-air interfaces, the proposed method consistently reveals intermediate scattering layers within the snowpack. These features show good agreement with the stratigraphy observed in the snowpit measurements, demonstrating the potential of covariance-based sparse reconstruction for high-resolution characterization of snowpacks from UAV-SAR observations.