High-resolution snow water equivalent estimation derived from downscaled snow depth and non-constant snow density in Chinese Altai Mountains

Abstract

Snow water equivalent (SWE) plays a critical role in managing freshwater resources in mountainous areas. Due to the spatial heterogeneity of snow, high-resolution SWE mapping is essential for accurate mountainous snow monitoring. However, the coarse spatial resolution of commonly used SWE datasets introduces uncertainties regarding their accuracy and applicability in these areas. This study combined multi-source remote sensing data fusion downscaling algorithm (MDFDA), Random Forest (RF) model and non-constant snow density (SDE) to reconstruct SWE at 1-km resolution in Chinese Altai mountains (CAM), demonstrating nearly 50 % and 17 % reductions in relative error compared to 25-km reference SWE data and SWE derived from the constant SDE. First, we applied MDFDA to downscale the 25-km passive microwave-derived snow depth (SD) data to 1-km. Next, we input the downscaled SD, along with spatiotemporal and climatological covariates, into RF model to obtain accuracy-optimized downscaled SD (RFSD). Finally, we converted the RFSD to SWE by utilizing two non-constant SDE conversion methods (Power law formula and SDE gridded data). The inclusion of covariates in RF model significantly improved the SD estimation accuracy, with the Pearson correlation coefficient (R) increasing from 0.61 to 0.96. The SWE derived from the Power law formula showed R of 0.85, whereas SDE gridded data yielded improved R of 0.9. Based on reconstructed SWE data, we found statistically significant differences in SWE (p < 0.01) between CMA and non-mountainous areas (NMA) in February and April. Our results are helpful for enhancing high-resolution SD/SWE estimations and hydrological research in mountainous areas.