Modeling reference evapotranspiration using machine learning and remote sensing techniques for semi-arid subtropical climate of Indian Punjab

Abstract A study was carried out to develop and evaluate the performance of different machine learning (ML) models for predicting reference evapotranspiration (ET0). The included multiple linear regression (MLR), least square-support vector (LS-SVM), artificial neural networks (ANNs) adaptive neuro-fuzzy inference system (ANFIS). daily meteorological data 50 years (1970–2019) were used estimate ET0 using FAO-ET calculator. calculator compared with ML investigate best-fit model ET. Thereafter, ET predicted by satellite (Moderate Resolution Imaging Spectroradiometer – MODIS) ET, which finally mapped a larger landscape (over entire Punjab Haryana). Modeling best performed through LS-SVM followed ANN2, ANN1, ANFIS10, ANFIS2, MLR ANFIS9 models. Among developed models, coefficient determination (R2) value varied from 0.800 0.998, being highest (0.998) under model. MODIS overestimated when having R2 root mean square error (RMSE) values 0.73 3.95 mm, respectively. After applying bias correction factor, RMSE 0.74 1.19 satellite-based estimation would be useful timely water budgeting manage scarcity problems local regional levels.