Comparative Evaluation of Machine Learning Models for Evapotranspiration Prediction

Abstract

Accurate estimation of reference evapotranspiration (ET₀) is important for irrigation scheduling, crop water requirement estimation, and agricultural water management. Reliable prediction of ET₀ helps improve irrigation efficiency and supports effective utilization of available water resources. Conventional evapotranspiration estimation methods often require complete meteorological datasets, which may not always be available in many agricultural regions. Therefore, machine learning approaches have emerged as alternative tools for predicting evapotranspiration using available climatic information. The present study evaluated the performance of four machine learning algorithms, namely Random Forest (RF), Gradient Boosting (GB), Extreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM), for predicting daily evapotranspiration values. Climatic data representing the College of Agricultural Engineering, Madakasira, Andhra Pradesh, India, were collected from the NASA POWER database for a period of 34 years. The collected parameters included maximum temperature, minimum temperature, rainfall, relative humidity, wind speed, and sunshine hours. Reference evapotranspiration values were estimated using the FAO Penman–Monteith equation through the CROPWAT model and used as the target variable for model development. Model performance was evaluated using Root Mean Square Error (RMSE) and coefficient of determination (R²). The results indicated that all evaluated models were capable of predicting evapotranspiration with satisfactory accuracy; however, differences in performance were observed among the algorithms. The LightGBM model produced the lowest RMSE value (0.2791) and highest R² score (0.9527), followed by XGBoost with an RMSE value of 0.2831 and an R² score of 0.9514, while Random Forest and Gradient Boosting models showed comparatively lower prediction performance. The close agreement between observed and predicted ET₀ values demonstrated the ability of machine learning algorithms to model the relationship between climatic variables and evapotranspiration. The findings of the study indicate that the LightGBM model is suitable for daily evapotranspiration prediction and can support irrigation scheduling and agricultural water management applications.