Overview This repository contains the code for a novel multiscale deep learning (LSTM-based) scheme that learns simultaneously from satellite and in-situ soil moisture data to produce 9 km daily soil moisture estimates at 5 cm depth. The model was evaluated over the conterminous United States using spatial cross-validation, achieving a median correlation of 0.901 and RMSE of 0.034 m³/m³ — outperforming SMAP's 9 km product, single-source deep learning models, and land surface models. Notably, our 9 km product showed better accuracy than previous 1 km satellite downscaling products, highlighting limited impacts of improving resolution. The multiscale scheme is generically applicable to geoscientific domains with data on multiple scales, breaking the confines of individual datasets. Citation If you use this code, please cite the following paper: Liu, J., Rahmani, F., Lawson, K., and Shen, C.: A multiscale deep learning model for soil moisture integrating satellite and in situ data, Geophysical Research Letters, 49(7), e2021GL096847, https://doi.org/10.1029/2021GL096847, 2022. BibTeX: bibtex @article{liu2022multiscale, title={A multiscale deep learning model for soil moisture integrating satellite and in situ data}, author={Liu, Jiangtao and Rahmani, Farshid and Lawson, Kathryn and Shen, Chaopeng}, journal={Geophysical Research Letters}, volume={49}, number={7}, pages={e2021GL096847}, year={2022}, doi={10.1029/2021GL096847}, publisher={Wiley Online Library} } Related Work This multiscale approach was later extended to a global model (GSM3) covering 2015–2020: Paper: Liu, J., et al. (2023). Evaluating a global soil moisture dataset from a multitask model (GSM3 v1.0). Geoscientific Model Development, 16, 1553–1567. https://doi.org/10.5194/gmd-16-1553-2023 Dataset: https://doi.org/10.5281/zenodo.7026036 Interactive Viewer: https://jayhydro.github.io/gsm3-viewer/