A decision framework for selecting IMERG bias-correction methods in complex tropical watersheds

Abstract

Accurate satellite precipitation estimates are critical for hydrological modelling and climate-risk management in the tropics, where sparse gauges and complex storm regimes increase uncertainty. This PRISMA-guided systematic review synthesizes advances in bias correction for GPM-IMERG across monsoonal, equatorial, orographic, and coastal basins. Evidence is organized into bias controls, classical and distribution-based corrections, machine-learning/hybrid methods, and hydrological impacts. Reported IMERG errors vary widely, exceeding 70% in high-relief watersheds and approaching 200% in strongly convective regions. Classical scaling reduces mean bias, whereas distribution-based methods better reproduce variability and extremes. ML models (ANN, LSTM/Bi-LSTM) improve correlation, RMSE, and event detection, particularly at hourly scales. Hydrological evaluations show improvements in streamflow, intensity-duration-frequency curves, and drought indices, with the Nash-Sutcliffe efficiency increasing to approximately 0.89 and peak-flow estimates improving. Despite improvements in IMERG V07, local correction remains essential; the choice of method should align with the regime, basin attributes, product version, and application objectives.