Enhanced prediction of monthly reference evapotranspiration by optimizing SVR with the bagging algorithm in Indian Punjab

Abstract

In this study, support vector regression (SVR) was enhanced with the Bagging (B-SVR) algorithm for monthly ET0 prediction at Ludhiana and Faridkot stations. The three different scenarios (S1, S2, and S3) were developed to predict monthly ET0 at study locations. The outcomes of the B-SVR model were compared against the SVR and Valiantzas models. The results showed that the highest ET0 prediction accuracy was achieved with B-SVR-S1 model (MAE = 0.062 mm/month, RMSE = 0.078 mm/month, IOS = 0.021, NSE = 0.998, R = 0.999, and IOA = 0.999) for Ludhiana and Faridkot (MAE = 0.089 mm/month, RMSE = 0.113 mm/month, IOS = 0.031, NSE = 0.996, R = 0.998, and IOA = 0.999) than other deployed models. Moreover, the results of this research suggested that the SVR with the bagging algorithm is a promising alternative for monthly ET0 prediction in sub-humid and semi-arid climates of Indian Punjab.