Snow wetness retrieval for snowmelt monitoring using dualpolarization SAR data

Snow wetness is crucial for the quantitative monitoring of snow melt, but methods for retrieving snow wetness from Sentinel-1 data are still lacking. This study proposes a contextguided snow wetness retrieval method using VV and VH dualpolarization data. First, we simulated three different scattering components and the total scattering of snow in both the VV and VH polarizations. Then, an error equation was constructed by subtracting the simulated total snow scattering from the corresponding SAR measured values for each polarization. As there are more than one unknown variables in each equation, we established an error equation group for retrieving snow wetness by combining the error equations for the proximity pixels. To address the challenge of unknown variables reduction, we proposed a proximity similarity principle based on Tobler’s First Law of Geography and the high sensitivity of backscattering to changes in snow wetness. Based on this principle, we obtained the snow wetness of each proximity pixel by solving the error equation group. Sentinel-1 data and Radarsat-2 data were used to retrieve the snow wetness in the central Tianshan Mountains, China, for verification. The mean absolute errors (MAE) for the Sentinel-1 and Radarsat-2 data were 0.77% and 0.58%, respectively, showing that the proposed method can perform large-scale snow wetness retrieval in mountainous areas, characterized by reliable applicability in different data modes. Subsequently, snowmelt area mapping and snowmelt phase division were achieved using the retrieved timeseries snow wetness with Sentinel-1 data. Compared with snowmelt monitoring methods relying on changes in backscattering, we demonstrated that the retrieved snow wetness not only directly reflects the snowmelt state, but also avoids the bias caused by only attributing the change in backscattering to the change in wetness.