Document Type : Special Issue: New Approaches to Water and Soil Management and Modeling
Authors
1
Graduated M.Sc., Department of Water Engineering, Faculty of Agriculture and Natural Resources, University of Mohaghegh Ardabili, Ardabil, Iran
2
Associate professor, Department of Water Engineering, Faculty of Agriculture and Natural Resources, University of Mohaghegh Ardabili, Ardabil, Iran
3
Professor, Department of Water Engineering, Faculty of Agriculture and Natural Resources, University of Mohaghegh Ardabili, Ardabil, Iran
Abstract
Evapotranspiration is one of the main sources of water loss in agricultural lands, and its accurate estimation plays a key role in reducing water wastage in the agricultural sector. In this study, several empirical equations, including FAO Penman–Monteith, Blaney–Criddle, Hargreaves–Samani, Irmak, Dalton, Romanenko, and Jensen–Haise, were used to estimate monthly evapotranspiration at two meteorological stations in Iran: Ardabil (semi-arid climate) and Zabol (arid climate). To improve the accuracy of these equations, both linear regression and nonlinear optimization methods were applied for calibration. In addition to the empirical equations, data-driven models, including a multilayer perceptron artificial neural network (MLP) and a hybrid MLP model combined with the ant colony optimization algorithm (MLP–ACO), were also developed. Model performance was evaluated using the root mean square error (RMSE), mean absolute error (MAE), coefficient of determination (R²), and general performance index (GPI). The results indicated that calibration using both linear and nonlinear methods significantly reduced RMSE (96.46% to 98.08%) and MAE (96.78% to 98.27%) values and increased R² (up to 5.49%) for all equations at both stations. The nonlinear optimization method showed greater performance improvement compared to linear regression. Among the empirical equations, the calibrated Blaney–Criddle equation exhibited the best performance. Furthermore, the MLP–ACO model outperformed the standalone MLP model and the non-calibrated empirical equations. Overall, the results demonstrated that equation calibration was highly effective, as the calibrated empirical equations outperformed both standalone and hybrid intelligent models in most cases under both climatic conditions.
Keywords
Main Subjects
)