Global Daily Surface Soil Moisture Data from 2015-2020 generated by Multi-Collocation Fusion and Spatiotemporal Deep Learning

We used multiple collocation analysis and least-squares linear weighted averaging to combine soil moisture estimates from SMAP, AMSR2, ASCAT, and SMOS microwave missions and GLDAS, MERRA2, and ERA5-Land model simulations. We then used a ConvLSTM model to fill in the spatiotemporal gaps in the fused soil moisture data, ultimately obtaining a seamless daily global surface soil moisture dataset spanning from March 31, 2015 to December 31, 2020. Due to limited computational resources, we trained the data by continent, dividing each continent into 1~4 tiles. Therefore, the final results are presented as tile-based compressed packages for each continent. The files named "MCA" and "ConvLSTM" contain our codes.