Comparative Evaluation of Machine-Learning Models for Predicting Daily Evapotranspiration in a Naturally Ventilated Greenhouse

Accurate determination of reference evapotranspiration (ET0) is crucial for optimizing irrigation scheduling in greenhouse environments, ensuring optimal plant growth and resource management. This study aims to identify the most accurate method for predicting ET0 in naturally ventilated greenhouse conditions. Four machine learning (ML) models were analyzed: stand-alone adaptive neuro-fuzzy inference system (ANFIS), decision tree (DTREE), support vector machine (SVM), and ANFIS with improved reptile search algorithm (IRSA). These models were evaluated for predictive accuracy using performance metrics, including R-squared (R2), root mean squared error (RMSE), mean absolute error (MAE), BIAS, and scatter index (SI) on training and validation data sets. The results showed variations in the precision and efficacy of the different estimation equations, but the ANFIS IRSA model demonstrated superior performance across the evaluated metrics. This model’s success highlights its potential for advancing precision agriculture by accurately predicting ET0 in greenhouse conditions, paving the way for further research in this field.