SMPD-MERG: a hybrid downscaling model for high-resolution daily precipitation estimation via merging surface soil moisture and multi-source precipitation data

Currently, the poor spatial resolution (10-50 km) and accuracy of satellite-based precipitation products limit their applications at regional scales. To overcome these issues, a hybrid downscaling framework, named soil moisture-based precipitation downscaling and merging methods (SMPD-MERG), that merging soil moisture-based precipitation downscaling results with European Space Agency (ESA) Climate Change Initiative (CCI) soil moisture product and multi-source data from rain gauge measurements and European Center for Medium-Range Weather Forecasts ERA5-Land precipitation data with random forest model was proposed to derive high-resolution and high-accuracy precipitation data at daily scale. The method was successfully applied to the Global Precipitation Measurement (GPM) daily precipitation product and improved its spatial resolution from 10 km to 1 km in the central part of the Iberia Peninsula during 2016-2018. The validation with field measurements revealed that the proposed method has good performance with correlation coefficient (CC), relative bias (BIAS), root mean square error (RMSE), and the modified Kling-gupta efficiency (KGE') values of 0.94, 1.00%, 1.27 mm, and 0.88, respectively. Meanwhile, the intercomparison with other downscaling algorithms including geographically weighted regression and interpolation methods, highlights the significant advantages of the proposed method. It improves the CC from around 0.60 to over 0.90, reducing the RMSE to below 1.30 mm, and decreasing BIAS by nearly an order of magnitude. In general, different from previous empirical downscaling methods, the proposed method not only considers the physical dynamics of the precipitation process but also well integrates the advantage of multi-source data. According to the satisfactory downscaling accuracy, this method shows good potential for producing high-quality precipitation data with high spatiotemporal resolution.