A Deep Neural Network Architecture to Model Reference Evapotranspiration Using a Single Input Meteorological Parameter

Hydro-agrological research considers the reference evapotranspiration (ETo), driven by meteorological variables, crucial for achieving precise irrigation in precision agriculture. ETo modelling based on a single parameter would be beneficial places where collection of climatic parameters is challenging. The aim this to develop deep neural network (DNN) architecture that predicts daily with input selected feature importance (FI) score generated machine learning techniques, random forest (RF), and extreme gradient boosting (XGBoost). This study also investigated potential SHapley Additive exPlanations interpret validate outcomes selection methods assessing contributions each prediction. These recommended solar radiation as significant datasets three California Irrigation Management System (CIMIS) weather stations located distinct zones. Three models (DNN-Ret, XGB-Ret, RF-Ret) were built using sole input, CIMIS output. performance evaluation developed proved DNN-Ret outperformed XGB-Ret RF-Ret regardless dataset, coefficients determination (R2) ranging from 0.914 0.954 local scenario, an average decrease 8–9.5% mean absolute error root squared error, improvement 2.6–2.9% Nash–Sutcliffe efficiency 1.7–2% increase R2. overall result analysis highlighted diverse