Comparison of bias correction techniques for rainfall over the Lower Mekong River Basin in Cambodia

Abstract

Rainfall is a key factor for water resources management and crucial for informed disaster management planning. The analysis of flood and drought conditions heavily relies on accurate and comprehensive rainfall datasets. According to the limitation of rainfall data scarcity in certain regions, using secondary datasets of rainfall data from satellites and reanalysis products could offer useful alternative in providing a wider coverage with acceptable resolution. Several bias correction techniques have been applied to further improve the accuracy of secondary rainfall data. This study aims to compare two techniques of bias correction which are quantile mapping (QM), and Bayesian theorem applied between observed rainfall and reanalysis rainfall. QM is a statistical transformation that attempts to match between model output and observation data using cumulative distribution function (CDF), while Bayesian theorem is a probabilistic framework that allows the integration of prior information with new data to generate an improved posterior distribution. The comparison of the two bias correction techniques is demonstrated using a case study in the Lower Mekong River Basin in Cambodia, which covers the area of 122,000 km² with available rainfall data from 1985-2022. The results illustrate Bayesian method performs better than QM method for all statistical performance indicators including R (0.90-1.00), NSE (0.81-0.99), and RMSE (6.00-37.50mm) over the regions of the study area. The findings from this study demonstrate the potential of using bias-corrected data in enhancing disaster planning and mitigation efforts, particularly for drought and flood in data scarce regions.