Machine Learning-Based Estimation of Reference Evapotranspiration and Crop Coefficients for Wheat Under Diverse Climatic Conditions

Accurate estimation of reference evapotranspiration (ET0) and crop coefficients (Kc) is critical for irrigation planning, particularly in data-limited regions where agriculture dominates freshwater consumption. Although machine learning (ML) methods have been widely applied to ET0 and Kc estimation, most studies address these parameters separately or focus on a single seasonal Kc value rather than stage-specific coefficients. This study presents an integrated machine learning framework that simultaneously estimates monthly ET0 and wheat stage-specific crop coefficients (Kc_ini, Kc_mid, Kc_end) using long-term district-level meteorological data from Türkiye. A total of 19 algorithms, including linear models, tree-based ensembles, boosting methods, support vector machines (SVR) and deep learning models, were evaluated using 10-fold cross-validation and standard performance metrics. The results showed that linear models provided the most consistent performance for monthly and annual ET0 estimation (R2 ≈0.85–0.87), while nonlinear ensemble models achieved superior accuracy for Kc prediction (R2 > 0.95). Feature importance analysis identified solar radiation and temperature as dominant ET0 drivers, whereas elevation and relative humidity were key controls for stage-specific Kc. The proposed framework provides a scalable approach for regional agricultural water management and supports model selection under data-limited conditions.