Spatial Dispersion and Non-Negative Matrix Factorization of SAR Backscattering as Tools for Monitoring Snow Depth Evolution in Mountain Areas: A Case Study at Central Pyrenees (Spain)

Accurate knowledge of snow cover extent, depth (SD), and water equivalent is essential for studying the global water cycle, climate, and energy–mass exchange in the Earth–atmosphere system, as well as for water resources management. Ratio between SAR cross-and co-polarization backscattering (σVH /σVV ) was used to monitor SD during snowy months in mountain areas; however, published results refer to short periods and show lack of correlation during non-snowy months. We analyze Sentinel-1A images from a study area in Central Pyrenees to generate and investigate (i) time series of σVH /σVV spatial dispersion, (ii) spatial distribution of pixelwise σVH /σVV temporal standard deviation, and (iii) fundamental modes of σVH /σVV evolution by non-negative matrix factorization. The spatial dispersion evolution and the first mode are highly correlated (correlation coefficients larger than 0.9) to SD evolution during the whole seven-year-long period, including snowy and non-snowy months. The local incidence angle strongly affects how accurately σVH /σVV locally follows the first mode; thus, areas where it predominates are orbit-dependent. When combining ascending-and descending-orbit images in a single data matrix, the first mode becomes predominant almost everywhere snow pack persists during winter. Capability of our approach to reproduce SD evolution makes it a very effective tool.