Hybrid Data-driven and Process-based Modeling for Streamflow Forecasting Under Climate Change in the Upper Genale River Basin of Ethiopia

<title>Abstract</title> <p>This study evaluates streamflow forecasting in Ethiopia's Upper Genale River Basin, focusing on the challenges posed by climate change, regional environmental shifts, and the need for highly accurate data. The research developed a hybrid hydrological framework that integrates the process-based QSWAT+ model with Random Forest, XGBoost, LightGBM, and Long Short-Term Memory (LSTM) algorithms to improve streamflow prediction in the Upper Genale River Basin. Historical climate and hydrological data (1985–2014) were combined with CMIP6 climate projections under SSP2-4.5 and SSP5-8.5 scenarios to assess future hydrological responses. The findings reveal that QSWAT+ achieved satisfactory calibration and validation performance (NSE = 0.74 and 0.71; R² = 0.77 and 0.85, respectively), indicating reliable representation of watershed processes. The hybrid framework substantially improved predictive accuracy, with the stacking ensemble model achieving NSE and R² values of 0.98. Projections under the SSP2-4.5 and SSP5-8.5 scenarios indicate a clear upward trend in streamflow. Evapotranspiration increased under all scenarios, suggesting growing pressure on water resources. These findings demonstrate that hybrid process-based and machine-learning approaches enhance streamflow forecasting and provide valuable information for climate adaptation, irrigation planning, and sustainable water-resources management in Ethiopian river basins</p>