Assessing the efficacy of various predictive models in simulating monthly reference evapotranspiration patterns and its impact on water resource management for agriculture in the Kebir-West watershed, North-East of Algeria

The estimation of monthly reference evapotranspiration (ETo) is important for the efficient management of water resources, especially in regions with limited meteorological station coverage such as Algeria. For this purpose, different prediction models including support vector machines, multiple regression, bagged trees, and neural networks were applied to estimate Penman-Monteith FAO-56-based monthly ETo in the Oued El Kebir watershed in northeastern Algeria. Eight combinations of climate inputs, including wind speed, relative humidity, and maximum and minimum temperatures, were examined. Four metrics were used to assess the models' performance: coefficient of determination (R²), mean relative error (MRE), mean absolute error (MAE), and root mean square error (RMSE). Sobol sensitivity analysis was conducted to determine the most influential parameter in ETo estimation. According to the results, the variable with the highest impact was maximum temperature. The findings indicate that the proposed models achieved high estimation accuracy. Among them, neural networks outperformed the other models, with a correlation coefficient of 0.99, RMSE = 0.37, MAE = 0.28, and MRE = 0.005, for a period spanning (1984-2022). This superior performance is attributed to their ability to simulate complex and nonlinear relationships between climatic variables and ETo. These results contribute to improved irrigation planning and more efficient water resource management for farmers, climate scientists, and water managers. © 2025 Rayane Saci et al., published by Sciendo.