Evaluation and Comparison of Empirical Methods and Machine Learning Models in Estimating Reference Evapotranspiration (Case study: Varamin)

Agriculture, accounting for over ۷۶% of water use, is the largest consumer of water resources in Iran and faces serious challenges due to climate change, drought, and low irrigation efficiency. Therefore, accurate determination of reference evapotranspiration (ETo) is essential for optimal water-resource management. This study compares reference evapotranspiration estimated by the weighing lysimeter (water-balance) method, the empirical FAO-Penman-Monteith method, and two machine-learning models (Decision Tree and k-Nearest Neighbors) in Varamin County. Climatic data were collected from the Varamin synoptic station, and the weighing-lysimeter method was used as the reference. Results showed that the average evapotranspiration by the weighing-lysimeter method was ۲۷.۵ mm day¹, and the FAO-Penman-Monteith method, at ۶۵.۴ mm day, was the closest to the reference values. When comparing the machine-learning models against the measured lysimeter values, the Decision Tree (DT) and k-Nearest Neighbors (KNN) algorithms ranked first and second with correlation coefficients of ۰.۹۲ and ۰.۸۶, respectively. Moreover, this study demonstrated that, due to their high accuracy and lower requirement for comprehensive meteorological inputs, machine-learning models can serve as reliable alternatives to empirical methods for estimating reference evapotranspiration in regions with limited meteorological data. Their adoption can enhance agricultural water management and improve irrigation efficiency.

کلیدواژه ها:

Reference Evapotranspiration ، Machine Learning ، Water Balance ، FAO-Penman-Monteith ، Prediction Algorithms