Can Bias correction Techniques Improve Remote Sensing-based Rainfall Estimates in a Semi-Arid Context: Case of the Oum Er-Rbia River Basin in Morocco

Abstract

In semi-arid contexts, the strong spatiotemporal fluctuation of rainfall and the sparsity of the rain gauge (RG) measurement networks are the main limitations for water resources management. Freely available satellite-based rainfall estimates can be a potential source of information to cop data limitations over poorly gauged regions. Thus, the main aim of this work was to investigate how eight Spatial Rainfall Products (SRP, ARC-2, CHIRPSp25, CHIRPSp5, CMORPH-CRT, GPM-IMERG, PERSIANN-CDR, RFE-2, and TRMM-3B42) can be able to reproduce the observed monthly rainfall over a semi-arid context. The SRP estimates were directly evaluated against the RG observations. Then, bias correction techniques were used to account for the bias in the SRPs. The results indicated that the SRPs poorly correlate with the daily rainfall patterns (with Pearson Correlation Coefficients (PCCs) mostly below 0.5) but agreed with the monthly observations. The agreement was stronger over the lowlands than over the mountainous region. Overall, out of all the considered SRPs, IMERG (with a short-term record) and PERSIANN (with a long-term record) performed the best. Still, the monthly SRP estimates were significantly biased as the large rainfall totals were frequently underestimated. However, when the bias correction was applied remarkable improvement in the SRP’s performance was observed. The different adopted correction techniques yielded close results, with a slight prevalence of the Cumulative Distribution Function (CDF) over the Linear Scaling (LS), and Simple Linear Regression (SLR) techniques. Still, to reliably adjust the bias in the SRP estimates, LS and SLR should be preferred over the CDF technique, as they demonstrated more spatially consistent performance after validation.