Snow depth estimation model calibration and validation for high-altitude glacier valleys in the Indian Himalaya.

Seasonal snowpack is one of the primary sources of freshwater for rivers in the Indian Himalaya. It plays a vital role in regional hydrology, climate variability, and water resource management. To understand these processes and their impact on the community, spatial and temporal monitoring of snow is essential. Snow depth is a key parameter for monitoring snow. However, in the Himalayas, due to accessibility challenges and logistical constraints,  limited snow depth observations are available. To address this gap and estimate snow depth at high spatial and temporal resolution, we develop a model using polarimetric parameters derived from Sentinel-1 SAR data, topographic and auxiliary data, integrated with field-based observations in the European Alps and Grand Mesa, USA. Field observations are filtered to match the Sentinel-1 pass, ensuring consistency between field-based observations and satellite acquisition. Our model employs topographic data (e.g., elevation, slope, and aspect) from the Copernicus 30 m digital elevation model, auxiliary parameters (such as day of the season (DoS)), Forest cover fraction from MODIS, and Sentinel-1 SAR-based polarimetric parameters (cross-ratio, entropy, Stokes parameters, alpha), ensuring a topographically dependent snow depth distribution. Sensitivity analysis is performed using SHAP (SHapley Additive Explanations) to identify the most critical parameters for estimating snow depth. The model shows a Mean Absolute Error (MAE) of 0.04m, a root mean square error (RMSE) of 0.15m, with a test R-squared (R2) of 0.95 and a cross-validation correlation coefficient (R) of 0.98 in the European Alps. We transfer the model to the mountains in the Chandra Bagha basin (33°01′N°, 76°40′E) of the Indian Himalayas. Our transferred model highlights the potential of estimating snow depth in data-scarce regions while resolving the spatial and temporal details.