Spatial Downscaling of the FY3B Soil Moisture Using Random Forest Regression

Soil moisture (SM) plays a vital role in regulating the feedback between the terrestrial water, carbon, and energy cycles. However, the passive microwave SM product can hardly satisfy many applications, owing to their coarse spatial resolution. In this study, a random forest (RF) -based downscaling approach was applied to downscale the FY3B L2 soil moisture data from 25 -km to 1 -km, synergistically using the optical and thermal infrared (TIR) observations from the Moderate-Resolution Imaging Spectro-radiometer (MODIS). The RF algorithm used various surface variables to construct the SM relationship model, such as surface temperature, leaf area index, albedo, water index, vegetation index, and elevation, comparing with the widely used polynomial-based relationship model. The correlation coefficient (R) and the root-mean-square deviation (RMSD) of RF-based method reached 0.93 and 0.051 m3/m3, respectively. Four blends of data were used to retrieve the downscaled SM through the RF-based downscaling method. The downscaling results were validated by the in-situ soil moisture from REMEDHUS network. The temporal changing pattern of the downscaled SM was assessed with the precipitation time series. This study suggests that the RF-based downscaling method can characterize the variation of SM and is helpful to improve accuracy of the passive microwave SM product.