Enhancing Rainfall-Runoff Simulation in Data-Poor Watersheds: Integrating Remote Sensing and Hybrid Decomposition for Hydrologic Modelling

Accurate Rainfall-Runoff (R-R) modeling is essential for effective water resource management and flood mitigation, yet it remains challenging in data-scarce regions. This study investigates the feasibility of integrating remotely sensed precipitation and temperature data to enhance R-R modeling. The proposed framework is applied to the Alingar watershed, with limited observational data. The Climate Hazards Group InfraRed Precipitation with Station data (CHIRPS) and Moderate Resolution Imaging Spectroradiometer/land surface temperature (MODIS-LST) data were utilized. Remote Sensing (RS) data were complemented by in situ measurements. A modified bias correction (BC) method was applied to improve the accuracy of RS data. Three input scenarios were considered, including ground-truth data, raw RS data, and bias-corrected RS data. These inputs were employed for the machine learning (ML) models, specifically weighted voting ensemble-based method and the CatBoost algorithm. Time series decomposition techniques, including the standalone Wavelet Transform and a novel hybrid method that combines both Wavelet Transform and Gaussian Filter (GF-WT), were applied to filter out noise from the data. The results demonstrated that incorporating RS data significantly improved R-R modeling accuracy, with the proposed BC method further enhancing ML model performance. Among the ML models tested, the hybrid GF-WT decomposition method yielded the highest predictive accuracy. The weighted voting ensemble model outperformed the CatBoost algorithm. This study underscores the strong potential of hybrid decomposition techniques and RS data integration in improving R-R modeling, particularly in data-limited watersheds, offering valuable insights for hydrologic applications in data-scarce regions.