Improving Local Rainfall Projections in Bali through Bias Correction of CMIP6 Data: A Contribution to Hydro-Meteorological Disaster Mitigation

Accurate rainfall projections are crucial for flood and landslide preparedness especially for regions with high exposure of hydro-meteorological hazards, such as Bali. We evaluated four statistical bias correction methods: Linear Scaling (LS), Empirical Quantile Mapping (EQM), Quantile Delta Mapping (QDM), and Quantile Mapping with Boostrapping (QMB), using daily CMIP6 TaiESM1 outputs (2000-2022) and observations from four BMKG stations in Bali. Data were split into calibration (2000-2015) and validation (2016-2022) sets and considered only a subset of wet days. Evaluation metrics included MAE, RMSE, mean bias, and high-quantile errors (P₉₅ and P₉₉). LS achieved the smallest MAE (17.7 to 19.9 mm), lowest RMSE (26.6 to 29.8 mm), and near-zero bias (-1.9 to +1.8 mm), confirming its effectiveness for mean correction. Notably, QMB, which incorporates block bootstrap into the quantile mapping process, showed the most consistent behaviour in representing upper-tail rainfall. Although QDM yielded the lowest average P₉₅ and P₉₉ errors, QMB more effectively minimized worst-case errors across stations, highlighting its robustness in extreme quantile estimation. Our findings point to a practical workflow: applying LS effectively adjusts the mean, while QMB for more reliable extreme quantiles. That combined approach improves the potential use of GCM rainfall data for hydrometeorological disaster risk planning on Bali.