<![CDATA[Rainfall prediction in the tropics is highly challenging due to strong convective variability and limited observations. This study evaluates the Weather Research and Forecasting (WRF) model over southern Sumatra, using ERA5 and GFS-FNL as boundary datasets. Ten combinations of microphysics and cumulus parameterizations were tested, followed by bias correction, ensemble construction, and verification against AWS observations. Two bias correction methods were compared: Linear Scaling (LS) and Quantile Mapping (QM). LS consistently improved correlation and reduced RMSE, while QM often degraded performance. For rainfall intensity, ERA5-driven simulations outperformed GFS, with ERA5 Member 9 (WSM6 + BMJ) showing the highest skill (r = 0.91; RMSE ≈ 20 mm). Rainfall occurrence verification showed that ERA5 ensembles, particularly Ensemble 3 with POD–FAR weighting, achieved the most balanced skill (CSI ≈ 0.60), while GFS Member 5 occasionally surpassed the ensemble mean. ROC analysis revealed complementary strengths: ERA5 was more effective for moderate to heavy rainfall, whereas GFS showed greater sensitivity to light events. These results underscore ERA5’s advantage for tropical downscaling and highlight the potential of multisource ensembles to improve rainfall prediction in data-scarce regions.]]>
Copyrights © 2026
