Evaluating Linear Scaling and Random Forest Bias Correction of GPM-IMERG V7 Rainfall in Moluccas Islands Watersheds

Reliable, high-resolution daily rainfall data are essential for hydrological assessments and flood-risk management in tropical island watersheds; however, rain-gauge networks in the Moluccas Islands are sparse and unevenly distributed. This study assesses the daily performance of the GPM-IMERG Version 7 (Final Run) precipitation product. It develops bias-correction strategies for six coastal watersheds (Dodaga, Sangaji, Kobe, Inggol, Wai Ruhu, and Wai Batumerah) using daily gauge observations from 2022 to 2025 as reference. A conventional mean-based Linear Scaling (LS) approach is implemented as a benchmark and compared with a non-parametric Random Forest (RF) regression model developed using matched IMERG-gauge pairs and evaluated against gauge observations over 2022–2025. Model skill is measured at the daily level using the Nash-Sutcliffe efficiency (NSE), Pearson correlation coefficient (R), and the RMSE-to-standard-deviation ratio (RSR). LS shows poor daily performance (NSE −0.53 to 0.18; R 0.12-0.49; RSR 0.91-1.40), indicating that mean adjustment alone is insufficient to address event-scale errors and intensity-related biases. In contrast, RF significantly enhances agreement with gauge observations across all watersheds (NSE 0.85-0.90; R 0.96-0.98; RSR 0.32-0.39). These results suggest that RF-corrected IMERG V7 offers a reliable daily rainfall input suitable for hydrological applications in data-limited tropical islands. Additionally, it provides a correction workflow that can be adapted to similar coastal watersheds.