MACHINE LEARNING MODELS (MLP, RANDOM FOREST, LIGHTGBM) FOR DAILY ET₀ ESTIMATION WITH LIMITED DATA IN HUMID MEDITERRANEAN REGION (JIJEL) ALGERIA

Accurate estimation of the reference evapotranspiration (ET₀) is essential for effective irrigation planning and sustainable water resource management, particularly in climatically heterogeneous regions. This study evaluated the performance of three machine learning models— multilayer perceptron (MLP), Random Forest (RF), and Light Gradient Boosting Machine (LightGBM)—for predicting daily FAO-56 Penman–Monteith ET₀ in the humid Mediterranean region of Jijel, northeastern Algeria. Daily meteorological data from ten stations for the period 2000–2024 were used, incorporating six input variables: air temperature, relative humidity, wind speed, sunshine duration, solar radiation, and vapor pressure deficit. The model performance was assessed using multiple statistical metrics across the training, validation, and independent testing datasets. All models achieved high predictive accuracy, with R² values exceeding 0.97. RF exhibited the highest training performance (R² = 0.997, RMSE ≈ 0.09 mm day⁻¹) but showed signs of mild overfitting on test data. In contrast, MLP demonstrated the best generalization capability (test R² = 0.983, RMSE = 0.21 mm day⁻¹, NSE = 0.983), closely followed by LightGBM (test R² ≈ 0.980). Trend analysis revealed no significant long-term change in annual ET₀ (p = 0.907). The results confirm the robustness of machine learning approaches, particularly MLP and LightGBM, for reliable ET₀ estimation in humid Mediterranean environments.