Estimating spatially downscaled rainfall by regression kriging using TRMM precipitation and elevation in Zhejiang Province, southeast China

Abstract

Estimating regional daily rainfall accurately is of prime importance for many environmental applications, such as hydrology, meteorology, and ecology. The rainfall product from the Tropical Rainfall Monitoring Mission (TRMM) satellite is better able to estimate rainfall than rain gauge interpolation in some regions with coarse rain gauge spatial resolution. In the present article, analyses were made at 1379 rain gauge stations in Zhejiang Province, China, during January 2011 to July 2012 (536 days). A good relationship was found between the rain gauge data and the data analysis from the TRMM, especially for the precipitation that was between 2 and 10 mm day–1. However, gaps exist between TRMM products and rain gauge records, which could be considered as uncertainty. To predict rainfall more precisely, four categories of daily rainfall and three regression kriging (RK) models were selected for analysis. TRMM and elevation data were used as auxiliary variables to construct RK1. The auxiliary variable in RK2 and RK3 was TRMM and elevation data, respectively. Residuals (four rainfall categories × three RK models) of RK models showed spatial auto-correlation. Compared with RK2, which has a 0.25° resolution, RK1 and RK3 are predicted at a finer 1 km spatial resolution. However, RK1 has the best performance in rainfall prediction according to validation statistics. The root mean square error was decreased from 0.667 to 0.437 and the mean of error was improved from –0.250 to –0.007 in the prediction of mean daily rainfall. RK1 may facilitate easy downscaling of precipitation and capture the trends in daily rainfall variability.