Snow avalanche mapping using sentinel-1 SAR change detection

Abstract

Snow avalanches pose a persistent threat to mountain communities, yet systematic inventories remain scarce where cloud cover and rugged topography limit optical remote sensing. Leveraging the all-weather capability of C-band radar, we design an automated workflow that transforms Sentinel-1 imagery into a season-scale avalanche record for the Zailiysky Alatau range (Northern Tien Shan, Kazakhstan). A dual-polarization Interferometric-Wide pair acquired in March–April 2024 was co-registered in Google Earth Engine, speckle-suppressed with an adaptive Enhanced Lee filter, and converted to a VV–VH polarization-difference layer. Temporal differencing highlighted fresh debris as negative anomalies. Layover, radar shadow, and permanent water were masked using the 30 m SRTM DEM and the JRC Global Surface Water product. Further, decision tree classifiers were used for delineation of avalanche from non-avalanche pixels. Validation against PlanetScope (3 m) and Sentinel-2 (10 m) imagery acquired within ± 2 days returned a detection completeness. Results confirm that Sentinel-1 change detection can retrieve most medium-to-large avalanches even under persistent cloud cover, offering a cost-free complement to sparse field observations in Central Asia. The workflow fully implemented in a cloud platform requires no scene-specific tuning and is transferable to other snow-covered mountain regions for near-real-time hazard assessment.