Comparative Analysis of Shallow Regression Models for Evopotranspiration Estimation with Climate Variables
2025 Innovations in Intelligent Systems and Applications Conference, ASYU 2025, Bursa, Türkiye, 10 - 12 Eylül 2025, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/asyu67174.2025.11208404
- Basıldığı Şehir: Bursa
- Basıldığı Ülke: Türkiye
- Anahtar Kelimeler: Evapotranspiration, Meteorological data, PenmanMonteith, Regression models, Shallow Learning
- Bursa Uludağ Üniversitesi Adresli: Evet
Özet
Accurate estimation of reference evapotranspiration (ETo) is essential for sustainable agricultural water management, particularly in regions where field-based measurements are limited and remote sensing data are either unavailable or impractical to use. This study investigates the predictive capability of six shallow learning regression algorithms, extreme gradient boosting (XGBoost), random forest, multi-layer perceptron regressor (MLP), linear regression (LR), support vector regression (SVR), and decision tree (DT), for daily ETo estimation using only five commonly recorded meteorological variables: air temperature, soil temperature, relative humidity, wind speed, and shortwave radiation. The dataset, derived from the Open-Meteo platform for Batman, Turkiye. The models were trained, evaluated and assessed based on R 2, MAE, MSE, and RMSE metrics. Among the tested models, XGBoost achieved the highest accuracy (R 2=0.9946, RMSE =0.0208), followed closely by LR and MLP methods. In contrast, SVR and DT models exhibited lower precision and generalization capability. Correlation analysis and scatter plots confirmed that shortwave radiation, air temperature, and soil temperature were the most influential predictors. The results demonstrate that accurate and interpretable ETo prediction is feasible using a limited set of accessible meteorological inputs and computationally efficient shallow learning models.