Bias Correction of CMIP6 Models for Assessment of Wet and Dry Conditions Over Sumatra

Abstract

The performance of CMIP6 models in capturing local and regional precipitation patterns often requires refinement due to inherent biases. This study evaluates eleven CMIP6 models for their applicability over Sumatra Island and applies two bias correction methods namely Linear Scaling (LS) and Quantile Delta Mapping (QDM).  We used ERA5 precipitation datasets as a reference bias correction during 1981-2014. The performance was assessed using MAE, correlation, and PBIAS. Results reveals that raw model of CMIP6 generally underestimate precipitation, particularly during the DJF and SON seasons, with the largest errors over the mountainous western Sumatra. LS tends to overcorrect and shift precipitation estimates toward a wetter bias, while QDM significantly improves the accuracy and seasonal consistency of the simulations.  The multi-model ensemble mean (CMIP6-avg) outperforms individual models, and its performance is further enhanced with QDM, yielding higher correlation and lower error metrics. Spatial and seasonal analyses demonstrate that QDM more effectively reduces both dry and wet biases, especially during peak rainfall seasons. These findings underscore the importance of robust bias correction techniques to improve climate projections for hydrological and climate impact studies in Sumatra and other tropical regions with complex terrain.