Temporal Downscaling of TRMM Rain-Rate Images Using Principal Component Analysis during Heavy Tropical Thunderstorm Seasons

Abstract

his paper examines the utility of principal component analysis (PCA) in obtaining accurate daily rainfall estimates from 3-hourly Tropical Rainfall Measuring Mission (TRMM) satellite data during heavy precipitation in a humid tropical environment. A large bias during heavy thunderstorms in humid tropical catchments is indicated by the TRMM satellite and is of profound concern because it is a conspicuous constraint for practical hydrology applications and requires proper treatment, particularly in areas with sparse rain gauges. The common procedure of calculating daily rainfall estimates by direct accumulation (DA) of a series of 3-hourly rainfall estimates caused a large bias because of temporal uncertainties, upscaling effects, and different mechanisms. In this study, PCA was used to transform correlated 3-hourly rain-rate images into a minimum effective principal component and to compute the corresponding rain-rate proportion based on correlation strength. This study was conducted on 91 rainy days of various intensity, acquired from three different years, during the wettest season on the eastern coast of peninsular Malaysia. Results showed that PCA reduced the bias and daily root-mean-square error by an average of 62% and 22%, respectively, compared with the DA approach. The PCA transformation was able to produce more precise daily rainfall estimates compared to the DA approach without the use of any rain gauge references. However, the performance was varied by the threshold selection and rainfall intensity. The results of this study indicate that PCA can be a useful tool in effective temporal downscaling of TRMM satellite data during heavy thunderstorm seasons in areas where rain gauges are sparse and satellite data are pivotal as a secondary source of rainfall data.