A Comparative Assessment of Advanced Bias Correction Methods for GPM IMERG in Banjir Kanal Timur Watershed

Reliable precipitation inputs are essential for flood risk management in the Semarang lower Banjir Kanal Timur (BKT) catchment, where tidal effects and convective storms play a significant role. This study uses monthly data (2014–2023) from three gauges and GPM IMERG Final Run to evaluate five bias correction methods: Linear Scaling (LS), Linear Regression (LR), Correction Factor (CF), Spatiotemporal (ST), and Random Forest (RF). The analysis follows a reproducible gauge–satellite workflow. Skill was assessed with Nash–Sutcliffe efficiency (NSE), Pearson correlation coefficient (R), and the ratio of RMSE to standard deviation (RSR). RF yielded the highest performance (NSE = 0.61, R = 0.78, and RSR = 0.63), indicating improved variance reproduction and association. CF and ST attained satisfactory operational skills (both NSE = 0.53, R = 0.73, and RSR = 0.69), while LS and LR yielded modest gains (NSE = 0.37 and 0.28; R = 0.70 and 0.65; RSR = 0.79 and 0.84), reflecting persistent intensity and season-dependent biases and point pixel mismatch. This study provides a decade-long, practice-oriented benchmark for bias correcting IMERG in a tidally influenced tropical urban basin and establishes a consistent monthly performance hierarchy (RF > CF ≈ ST > LS > LR). The resulting bias-corrected series enhances GPM IMERG suitability for hydrological modelling, flood forecasting, and monitoring in data-scarce tropical cities, while highlighting needs for tail-focused evaluation and temporally blocked validation to strengthen operational deployment.