Soil Moisture Forecasting and Hindcasting Using Deep Learning and SMAP Satellite Data

The Soil Moisture Active Passive (SMAP) mission has delivered valuable sensing of surface soil moisture since 2015. This surface soil moisture data has significant value for agricultural and weather prediction applications. However, it has a short time span and irregular revisit schedule. On the other hand, land surface models’ predictions are biased, with incorrect temporal dynamics. Utilizing a state-of-the-art time-series deep learning neural network, Long Short-Term Memory (LSTM), we created a system that predicts SMAP level-3 soil moisture data with atmospheric forcing, model-simulated moisture, and static physiographic attributes as inputs. The system removes most of the bias with model simulations and also improves predicted moisture climatology, achieving a test root-mean-squared-difference of less than 0.037 in 75 percent areas of Continental United States, including the forested Southeast. As the first application of LSTM in hydrology, we show that it is more robust than simpler methods in either temporal or spatial extrapolation tests. LSTM generalizes better across regions with distinct climates and physiography. We demonstrate the value of this system for agricultural soil moisture forecasting.