Support-Vector-Machine-Based Models for Modeling Daily Reference Evapotranspiration With Limited Climatic Data in Extreme Arid Regions

Evapotranspiration is a major factor that controls hydrological process and its accurate estimation provides valuable information for water resources planning and management, particularly in extremely arid regions. The objective of this research was to evaluate the use of a support vector machine (SVM) to model daily reference evapotranspiration (ET0) using limited climatic data. For the SVM, four combinations of maximum air temperature (T-max ), minimum air temperature (T-min ), wind speed (U-2 ) and daily solar radiation (R-s ) in the extremely arid region of Ejina basin, China, were used as inputs with T(max)and T-min as the base data set. The results of SVM models were evaluated by comparing the output with the ET0 calculated using Penman-Monteith FAO 56 equation (PMF-56). We found that the ET0 estimated using SVM with limited climatic data was in good agreement with those obtained using the conventional PMF-56 equation employing the full complement of meteorological data. In particular, three climatic parameters, T-max , T-min , and R-s were enough to predict the daily ET0 satisfactorily. Moreover, the performance of SVM method was also compared with that of artificial neural network (ANN) and three empirical models including Priestley-Taylor, Hargreaves, and Ritchie. The results showed that the performance of SVM method was the best among these models. This offers significant potential for more accurate estimation of the ET0 with scarce data in extreme arid regions.Evapotranspiration is a major factor that controls hydrological process and its accurate estimation provides valuable information for water resources planning and management, particularly in extremely arid regions. The objective of this research was to evaluate the use of a support vector machine (SVM) to model daily reference evapotranspiration (ET0) using limited climatic data. For the SVM, four combinations of maximum air temperature (T-max ), minimum air temperature (T-min ), wind speed (U-2 ) and daily solar radiation (R-s ) in the extremely arid region of Ejina basin, China, were used as inputs with T(max)and T-min as the base data set. The results of SVM models were evaluated by comparing the output with the ET0 calculated using Penman-Monteith FAO 56 equation (PMF-56). We found that the ET0 estimated using SVM with limited climatic data was in good agreement with those obtained using the conventional PMF-56 equation employing the full complement of meteorological data. In particular, three climatic parameters, T-max , T-min , and R-s were enough to predict the daily ET0 satisfactorily. Moreover, the performance of SVM method was also compared with that of artificial neural network (ANN) and three empirical models including Priestley-Taylor, Hargreaves, and Ritchie. The results showed that the performance of SVM method was the best among these models. This offers significant potential for more accurate estimation of the ET0 with scarce data in extreme arid regions