Soil Moisture Prediction Based on Spatiotemporal Graph Convolution Deep Learning

Soil moisture(SM) content is a key factor in agriculture, ecosystem and environmental management. Aiming at the difficulty of obtaining high-quality spatiotemporal continuous soil moisture information, Oklahoma region was selected as our test area. Temporal Graph Convolutional Network (TGCN) deep learning method based on spatiotemporal fusion was used to predict spatiotemporal continuous deep soil moisture information in Oklahoma region, and multiple indicators were used to evaluate this method. The results demonstrate that the soil moisture predicted by TGCN method at four depths in Oklahoma is extremely close to the measured data, and the fitting degree R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> of TGCN model is 0.99, which is 0.02 and 0.05 higher than that of GRU model and GCN model, respectively. According to the results, the method of soil moisture prediction based on deep learning is feasible, which provides a potentially prediction method for obtaining multi-level and high-precision soil moisture in Oklahoma, and also provides another new thinking for soil moisture prediction in other areas.