Evaluating performances of SVR, HBA–SVR, and COOT–SVR techniques: monthly evaporation modeling in Algeria

Accurate estimation of pan evaporation is essential for effective water resource management, especially in arid and semiarid regions. This study investigates the performance of a standalone support vector regression (SVR) model and its hybrid versions optimized with the honey badger algorithm (HBA–SVR) and the coot optimization algorithm (COOT–SVR) for monthly pan evaporation prediction in Algeria from 1978 to 2023. Nine modeling scenarios (M1–M9) were designed using different combinations of meteorological inputs such as precipitation, temperature, humidity, wind speed, and sunshine duration. The HBA and COOT optimizers were selected for their strong global search capabilities and proven effectiveness in optimizing nonlinear models. Model scenarios M1–M9 represent a structured progression, with each incorporating additional meteorological variables to assess model performance under varying data richness. The results demonstrated that SVR outperformed in scenarios with fewer input variables (M1–M5), while COOT–SVR achieved superior accuracy in more complex multivariable scenarios (M6–M9), with a peak Nash–Sutcliffe efficiency of 0.713. These findings highlight the potential of hybrid machine learning models for long-term evaporation forecasting and offer valuable support for irrigation planning, drought response, and water allocation strategies in arid regions.