Spatiotemporal trend detection and modeling of reference evapotranspiration using machine learning techniques for the Marathwada region, Maharashtra, India

ABSTRACT Predicting water loss accurately through evapotranspiration, using machine learning (ML) models, is crucial for irrigation scheduling. In this study, predictions and trends in the reference evapotranspiration (ET0) series were investigated for all eight districts of the Marathwada region of Maharashtra, India. ML models, including the random forest (RF), support vector regression (SVR) and linear regression (LR), were applied to predict ET0 under minimum climatic input. The modified Mann–Kendall (mMK) test and innovative trend analysis (ITA) detected a significant declining trend in annual and seasonal ET0 over the districts of the Marathwada region. The sensitivity analysis showed that maximum temperature (Tmax) and relative humidity (RHmean) are the most sensitive climatic factors that affect seasonal and annual ET0. Among the applied ML models, the RF model outperformed the SVR and LR models. For weather input combination of Tmax, RHmean, minimum temperature (Tmin) and wind speed, the RF algorithm predicted the ET0 with the highest R2 (0.97) and lowest RMSE (9.6 mm/month) during the testing period. The analysis provides a comprehensive approach that could help agricultural planners and water managers to understand ET0 dynamics and accurately predict crop water requirement, using fewer climatic parameters to make informed decisions on irrigation scheduling.