A novel approach for predicting daily pan evaporation in the coastal regions of Iran using support vector regression coupled with krill herd algorithm model

Evaporation is one of the vital components of hydrological cycle. Precise estimation of pan evaporation (E_pan) is essential for the sustainable water resources management. The current study proposed a novel approach to estimate daily E_pan across the humid region of Iran using support vector regression (SVR) technique coupled with Krill Herd Algorithm (SVR-KHA). Meteorological data were collected from three stations (Bandar Abbas, Rudsar, and Osku) over a period from 2008 to 2018 and used for application. Of the data, 70% were used for training and remaining 30% were used for testing. The study considered seven different combinations of input variables for predicting daily E_pan at each station. The influence of KHA hybridization is examined by comparing results of SVR-KHA algorithm with simple SVR through a multitude of statistical performance evaluation criteria such as coefficient of determination (R²), Wilmot’s index (WI), root-mean-square error (RMSE), Mean Absolute Error (MAE), Relative Root Mean Square Error (RRMSE), Mean Absolute Relative Error (MARE), and several graphical tools. Single input SVR1 model hybrid with KHA (SVR-KHA1) showed improved performance (R² of 0.717 and RMSE of 1.032 mm/day) as compared with multi-input SVR models, e.g., SVR5 (with RMSE and MAE of 1.037 mm/day and 0.773 mm/day), while SVR7 model hybridized with KHA (SVR-KHA7), which considers seven meteorological variables as input, performed best as compared with other models considered in this study. E_pan estimates at Bandar Abbas and Rudsar by SVR and SVR-KHA are similar (with R² statistics values of 0.82 and 0.84 at Bandar Abbas station, and 0.88 and 0.9 at Rudsar station, respectively). However, better improvements in E_pan estimates are observed at Osku station (with R² of 0.91 and 0.86, respectively), which is situated at interior geographical location with a higher altitude than the other two coastal stations. Overall, the results showed consistent performance of SVR-KHA model with stable residuals of lower magnitude as compared with standalone SVR models.