全球被动微波遥感反演雪水当量数据集（2013-2020）

Accurate quantitative snow water equivalent (SWE) information on a large global scale is very important for hydrological model improvement, snowmelt runoff prediction, and global climate change studies. There are numerous and heterogeneous parameters affecting the microwave radiation characteristics of snow cover, which is a great challenge for snow depth and SWE retrieval from passive microwave remote sensing observations. At present, the inversion algorithms used by most snow depth or SWE products usually fix the snow parameters, especially the snow microstructure parameter and snow density inside the snowpacks. However, the variation of snow parameters brings great uncertainty with respect to the snow metamorphosis and the spatial distribution of snow mass. In this paper, a global snow water equivalent algorithm is developed by combining the Microwave Emission Model of Layered Snowpacks (MEMLS) and machine learning method by considering the snow parameters of snow microstructure parameter and snow density. The snow microstructure parameter of exponential correlation length (Pec) was optimized utilizing the integrated snow radiative transfer model consisting of the snow emission model of MEMLS, atmospheric transmittance model, and Forests transmittance by minimizing the difference between the simulations and AMSR2 observations. Six selected parameters including vertical polarized temperature brightness differences between 18.7GHz and 36.5GHz，Pec，snow density from ERA5 (ρ), Forests cover fraction (f)，elevation，and longitude together with target output of snow depth were applied to train the machine learning model. Compared with the in-situ observations, the retrieval accuracy of SWE is 26.59 mm when SWE is lower than 500mm, which is better than those of the international SWE remote sensing products, assimilation, and reanalysis data in the same period.