A Comparative Evaluation of Bias Correction Techniques for Improving GPM-IMERG Precipitation Data in the Welang Watershed, Indonesia

Abstract

Precise precipitation information underpins hydrological modeling, water resource planning, and hazard mitigation, yet, gauge coverage in many Indonesian catchments is sparse. The GPM-IMERG product provides 0.1°/30-min rainfall estimates, however, systematic biases limit its operational value. Five benchmark correction techniques were evaluated: Linear Scaling (LS), Linear Regression (LR), Genetic- Algorithm-based Correction Factor (GA-CF), Local Intensity Scaling (LOCI), and Power Transformation (PT) against daily observations from seven gauges in the Welang Watershed (2001–2020). LS delivered the most consistent improvement (NSE = 0.87, R = 0.92, RSR = 0.36), reducing the residual error by 30% relative to the next-best method. LR, GA-CF, and LOCI enhanced seasonal patterns (NSE ≈ 0.85), while PT provided complementary gains for moderate events but remained sub-optimal for extremes. The refined IMERG series met the accuracy thresholds proposed for reservoir operations, providing a readily deployable rainfall input for data-scarce, topographically complex tropical watersheds.