This research developed a neural network representation for the the USDA Soil & Water Assessment Tool (SWAT) using remotely sensed soil moisture to provide near-real-time soil-moisture predictions. A recurrent neural network (RNN) with long short-term memory (LSTM) architecture was used to create a mapping, targeting pattern recognition in time series data. This project uses several Python packages. SMAP data acquisition was performed using the pytesmo package, which has tools for importing and comparing soil moisture data from several remote sensing sources on-the-fly. Popular neural network frameworks such as Keras and PyTorch were used to build a neural network with RNN-LSTM architecture.

---