Estimating sub-daily resolution soil moisture using Fengyun satellite data and machine learning

Soil moisture (SM) is a critical parameter influencing hydrological cycles, evaporation, and plant transpiration, connecting land surface and atmospheric interactions. However, traditional SM inversion methods mainly offer daily resolution data, potentially overlooking diurnal fluctuations due to factors such as precipitation and human activities. This study addresses this limitation by shifting to sub-daily (four times per day) SM data, utilizing artificial neural networks (ANN) with microwave brightness temperature data obtained from Fengyun-3C and Fengyun-3D (FY-3C and FY-3D) satellites, alongside the microwave vegetation index (MVI) to correct for vegetation effects. The ANN was trained from July 2018 to December 2019 (FY-3C) and January 2019 to December 2022 (FY-3D) using the International Soil Moisture Network as the training target. The ANN method demonstrates favorable global performance, as indicated by r = 0.751–0.805, NSE = 0.56–0.64, RMSE = 0.069–0.077 m3/m3, ubRMSE = 0.066–0.071 m3/m3, and mean Bias = 0.002–0.007 m3/m3 under the cross-validation mode. It can capture significant diurnal variations in SM, especially in regions like central Asia, western Australia, and South America. This research presents the feasibility of producing sub-daily high-temporal-resolution SM products with potential applications in large-scale agricultural drought and flood disaster monitoring, thereby enhancing national disaster management and mitigation strategies.