Evaluation of Spatial Rainfall Products in Sparsely Gauged Region Using Copula Uncertainty Modeling with Triple Collocation

Abstract

This study presents a topography-dependent error adjustment technique for observed spatial rainfall estimate (OSRE) with the application of a multivariate t-copula model in a sparsely gauged region in Bangladesh. Multiple satellite precipitation products, including TRMM-3B42v7, TRMM-3B42RT, IMERG-Daily (Final run), PERSIANN-CCS, and PERSIANN-CDR, and reanalysis products, including MERRA-2, ERA-Interim, and JRA55, were tested to remove the topography-dependent errors in OSREs. The Triple Collocation (TC) method was employed as an alternative evaluation method of OSRE in the absence of adequate rain gauges. To address topography-dependent errors, the error adjustment model was separately run over the mountainous and plain zones. After bias correction, the correlation coefficient (CC) values were improved by 17.39%–38.6% for the mountainous region and 23.69%–47.83% for the plain basin. Based on categorical and volumetric performance evaluation matrices, the satellite product IMERG-Daily performed more reasonably than any other OSREs in both the plain and mountainous zones. CC values obtained for IMERG-Daily were high, whereas PERSIANN products exhibited low CC values within the observed datasets. Among the reanalysis products, JRA55 performed poorly, whereas ERA-Interim showed satisfactory results. All reanalysis and precipitation products except for TRMM-3B42v7 provided significant random errors in the coastal area and mountainous basin relative to the plain zone. The TC method provides a similar performance ranking to traditional methods, that is, comparing OSRE with rain gauges. This study works to fill the existing knowledge gap by developing a zone-wise copula uncertainty model for multisource satellite and reanalysis precipitation products. An evaluation technique in a sparsely gauged region using the TC method was adopted to measure OSREs’ performance in the absence of gauge rainfall data.