Improving daily spatial precipitation estimates by merging gauge observation with multiple satellite-based precipitation products based on the geographically weighted ridge regression method

Merging gauge observation with a single original satellite-based precipitation product (SPP) is a common approach to generate spatial precipitation estimates. For the generation of high-quality precipitation maps, however, this common method has two drawbacks: (1) the spatial resolutions of original SPPs are still too coarse; and (2) a single SPP can't capture the spatial pattern of precipitation well. To overcome these drawbacks, a two-step scheme consisting of downscaling and fusion was proposed to merge gauge observation with multiple SPPs. In both downscaling and fusion steps, the geographically weighted ridge regression (GWRR) method, which is a combination of the geographically weighted regression (GWR) method and the ridge regression method, is proposed and implemented to generate improved spatial precipitation estimates by overcoming the collinearity problem of the pure GWR method. The proposed two-step merging scheme was applied to Xijiang Basin of China for deriving daily precipitation estimates from the data of both gauge observation and four near real-time SPPs (i.e., TMPA-3B42RT, CMORPH, PERSIANN and GSMaP_NRT) during the period of 2010-2017. The results showed that: (1) the collinearity problem caused by GWR was not serious in downscaling but serious enough to prevent GWR from being directly used in the fusion; and (2) the proposed two-step merging scheme significantly improved the spatial resolution and accuracy of precipitation estimates over the original SPPs. Comparisons also showed that, in the second step (fusion) of the merging scheme, the use of multiple SPPs provided more reliable spatial precipitation estimates than using a single SPP.