A Multivariable Approach for Estimating Soil Moisture from Microwave Radiation Imager (MWRI)

Accurate measurements of soil moisture are beneficial to our understanding of hydrological processes in the earth system. A multivariable approach using the random forest (RF) machine learning technique is proposed to estimate the soil moisture from Microwave Radiation Imager (MWRI) onboard Fengyun-3C satellite. In this study, Soil Moisture Operational Products System (SMOPS) products disseminated from NOAA are used as a truth to train the algorithm with the input of MWRI brightness temperatures (TBs) at 10.65, 18.7, 23.8, 36.5, and 89.0 GHz, TB polarization ratios (PRs) at 10.65, 18.7, and 23.8 GHz, height in digital elevation model (DEM), and soil porosity. The retrieved soil moisture is also validated against the independent SMOPS data, and the correlation coefficient is about 0.8 and mean bias is 0.002 m³ m⁻³ over the period from 1 August 2017 to 31 May 2019. Our retrieval of soil moisture also has a higher correlation with ECMWF ERA5 soil moisture data than the MWRI operational products. In the western part of China, the spatial distribution of MWRI soil moisture is much improved, compared to the MWRI operational products.