ExplainableETStack: An Explainable Hybrid Machine Learning Framework for Reference Evapotranspiration Estimation

ABSTRACT Estimating reference evapotranspiration (ET 0 ) becomes more difficult in situations with numerous features, requiring careful modelling. Due to swift advancements in machine learning (ML) models, several robust techniques have been introduced in the last few decades. This study leverages an explainable ensemble approach for ET 0 prediction to enhance the interpretability of the output predictions. The present approach includes ensemble stacking of ML models coupled with a gravitational search algorithm, where the meta‐model learns to combine the predictions from the best versions of the multiple models. This effectively reduces individual model weaknesses and builds a stronger, more accurate model. The study also leverages a combination of recursive feature elimination (RFE) and artificial bee colony (ABC) algorithms for optimal feature selection. Among the state‐of‐the‐art models, the XGBoost ( R 2 = 0.995, RMSE = 0.046 mm/day, MAE = 0.031 mm/day) and RF ( R 2 = 0.994, RMSE = 0.099 mm/day, MAE = 0.074 mm/day) models exhibited the highest performance for predicting ET 0 with default parameters. The proposed explainable meta‐learning–based stacked regression model (EESTK) exhibited an improved R 2 of 0.999, RMSE of 0.021 mm/day, MAE of 0.01 mm/day and NSE of 0.999 with complete input features. The proposed approach not only enhances performance but also introduces a reliable model for ET predictive modelling.