Improved downscaling of microwave-based surface soil moisture over a typical subtropical monsoon region

Abstract

Microwave remote sensing has great potential for capturing surface soil moisture (SSM) but suffers from its coarse spatial resolution. In this study, we combined the disaggregation based on physical and theoretical change (DISPATCH) and random forest (RF) approaches to downscale microwave-based SSM in the Yangtze River Delta region, China. Results showed that downscaled SSM by DISPATCH approach was superior to that by RF approach with Pearson correlation coefficient (R), unbiased root mean square error (ubRMSE), and the mean bias (Bias) of 0.778, 0.022 m3 m-3 and 0.061 m3 m-3, respectively. However, the DISPATCH approach was inapplicable in pixels with weak link between SSM and evapotranspiration, while the RF approach did not have such restriction. Therefore, the DISPATCH approach was supplemented with the RF approach to downscale the SSM for the entire study area. In addition, this downscaling process was conducted for the four different seasons and for the entire study period to test whether considering seasonality was necessary in the downscaling. The downscaled SSM was the best in autumn with R = 0.843, ubRMSE = 0.022 m3 m-3, and Bias = 0.048 m3 m-3, followed by summer and spring, and poorest in winter. However, the seasonally downscaled SSM performed better than that downscaled for the entire period. Moreover, over the entire study period, land surface temperature, land cover heterogeneity, soil texture, and water area percentage exerted significant influences on errors. However, contribution order of these environmental factors on errors varied with seasons. This study would be helpful to widen DISPATCH applicability in the semi-humid/humid region and maximize the potential of various data in generating high accurate and spatiotemporally continuous SSM estimates.