Modelling Nonlinear Daily Evapotranspiration using Variable Infiltration Capacity Model and Artificial Neural Network

Abstract

Evapotranspiration (ET) is a key variable for hydrologic, climatic and agricultural studies. Accurate quantification of this variable is of utmost importance for irrigation management and crop productivity. With the availability of only meteorological variables in climatic stations, reference gross evapotranspiration (ET_o) estimation is becoming a challenging task. Hence, there is a scope to estimate the ET_o using various physical and empirical methods. Among physical methods, FAO-56 PM method is best and Artificial Neural Network (ANN) models are accurate empirical methods. Further, ET_o can also be estimated using a water budget approach i.e. variable infiltration capacity (VIC) model, which accounts for the sub-grid variability of land use and land cover and soil moisture in a better way. In this study, the ET_o was estimated by two different methods, namely, VIC and ANN for Mohanpur climatic location in India. The results reveal that VIC- ET_o showed the correlation coefficient, r = 0.853, coefficient of determination, R² = 0.727 and index of agreement, d = 0.924; while ANN models showed better agreement with r = 0.999, R² = 0.998 and d = 0.999 with the FAO-56 PM method. Hence, it is concluded that the ANN showed better results as compared to VIC model for ET_o estimation in Mohanpur climatic location.