Hyperparameter optimization of machine learning models for predicting actual evapotranspiration

Actual evapotranspiration (AET) is a critical component of the water cycle, essential for effective water resource planning and management. While meteorological variables can predict AET, determining the reduction of potential evapotranspiration due to soil dryness remains challenging. AET data are often unavailable, necessitating reliance on more commonly measured variables. This study uses two groups of predictor variables for AET prediction. The first group, selected through Pearson correlation, tolerance, and VIF scores to address multicollinearity, includes net CO 2 , sensible heat flux, air temperature, relative humidity, and wind speed. The second group, chosen for practical applicability, consists of soil surface temperature, air temperature, relative humidity, and wind speed. Three predictive approaches are proposed: (i) deep learning models (LSTM, GRU, CNN) and (ii) classical machine learning models (SVR, RF). Hyperparameters were optimized using Bayesian optimization and compared with grid search. Bayesian optimization demonstrated higher performance and reduced computation time. Model performance was evaluated using statistical indicators (RMSE, MSE, MAE, R 2 ). Deep learning methods outperformed classical methods, with LSTM achieving the best results (Bayesian optimization: RMSE=0.0230, MSE=0.0005, MAE=0.0139, R 2 =0.8861). Performance decreased with fewer predictors. LSTM maintained superiority, achieving R 2 =0.8861 with five predictors and R 2 =0.8467 with four. LSTM also slightly outperformed SVR with fewer predictors. Overall, deep learning methods, especially with Bayesian optimization, proved more effective than classical methods for AET prediction. Future research could explore additional predictors and machine learning models to further improve AET prediction accuracy.