Enhancing Global Streamflow Predictions: Integrating Remote Sensing Data into a Hybrid PCR-GLOBWB and Random Forest Modeling Approach

Abstract

Streamflow predictions are essential for effective water management, enabling assessment of water availability, maintenance of agricultural practices, flood mitigation, and the overall impact on society. Traditional hydrological models have a difficult time comprehending some of the complex behavior of the hydrological cycle, leading them to have less accurate streamflow predictions. A hybrid modeling approach that couples the PCR-GLOBWB global hydrological model with the Random Forest machine learning algorithm was developed to enhance streamflow predictions. This study presents an innovative method for improving hydrological simulations by integrating satellite-derived precipitation and evaporation data into this hybrid modeling setup. The research explores the impact of this integration across different global contexts and uncovers the varying efficacy of satellite data in enhancing model accuracy, especially in regions with limited data. Through a comparative analysis of model performances using both global and local training datasets, the study emphasizes the critical importance of using satellite-based data and the strategic use of localized data for optimal predictions. The findings of this study show that model performance did not improve with the integration of satellite-based evaporation and precipitation globally. However, it suggests that satellite data integration offers significant benefits in certain contexts, even though its overall impact depends on the specific hydrological and geographical characteristics of the target region. This research provides valuable insights into the potential use of satellite data to enhance the accuracy and reliability of hydrological predictions, creating opportunities for more informed water resource management strategies amid global environmental changes.