Downscaling of Snow Depth at Moderate Spatial Resolution of 1-km for FY-3D/MWRI Using a Linear Unmixing Method

Passive microwave (PMW) snow depth (SD) products currently face significant uncertainty due to their coarse spatial resolution, and existing downscaling algorithms that utilize optical data often encounter issues such as signal saturation and a lack of a foundational physical mechanism. We proposed a novel downscaling algorithm: Linear Unmixing-based Snow Depth Downscaling (LUSDD), which leverages fractional landcover, obtained from daily seamless optical remote sensing, to derive brightness temperature (BT) endmembers of different landcover classes. This facilitates the reconstruction of BT at 1-km spatial resolution, subsequently enabling the retrieval of downscaled SD through the 1-km BT. Validations conducted at 53 ground SD stations demonstrate that the LUSDD algorithm significantly enhances both the spatial resolution and the accuracy of the original FY-3D SD product, with its root-mean-square error (RMSE) reduced from 4.60 cm of the original product to 2.56 cm. As an efficient and reliable method capable of downscaling SD data to a moderate spatial resolution of 1-km, the LUSDD algorithm is valuable to meet the needs of applications requiring higher spatial resolution.