Multifactor Spatial Downscaling of Satellite Precipitation Based on Vegetation Index and Elevation

Many environmental applications, such as hydrological and agricultural studies, require high spatial resolution and accurate precipitation data. In many places, ground observations of precipitation are unavailable. Therefore, data on precipitation are often obtained from satellite remote sensing. Unfortunately, precipitation estimates from remote sensing products are often too coarse in space and require further downscaling to match the required spatial resolution. We developed the Landcover similarity and Distance Weighted Regression with Residual correction model (LDWRR) that incorporates a vegetation index based on an adaptive lag phase and topography, and also includes a residual correction procedure that is combined with ground observations. The performance of the model is assessed by downscaling precipitation data obtained from the Integrated Multi-satellitE Retrievals for Global Precipitation Measurement (IMERG) from 10 km to 1 km spatial resolution over the Pearl River Basin in Southern China from 2010 to 2019 at daily resolution. Validation of the downscaled precipitation is done using ground observations. The LDWRR model is able to reproduce the precipitation pattern at high resolution, and it provides improved precipitation accuracy compared with the original coarse precipitation data. The correlation coefficients of the precipitation downscaled using LDWRR are larger than 0.7 at daily resolution. Moreover, the accuracy improved by more than 4% on average compared with IMERG. The LDWRR model is general and can be used in other regions and on any remote sensing device to spatially downscale precipitation data.