Daily reference evapotranspiration estimation based on least squares support vector machines

Abstract

As a key hydrological parameter, daily reference evapotranspiration (ETo) determines the accuracy of the hydrological number of the crop, and, consequently, the regional optimization disposition of water resources. At present, the main methods for ETo estimation are the Penman-Monteith (PM) equation and its modified formula, both of which are based on climatic factors such as temperature, radiation, humidity, and wind velocity, among others. Unfortunately, these required data are not always available in Xinjiang Uighur Autonomous Region, China, which is a semiarid area. Hence, this paper puts forward, for the first time, a least squares support vector machine (LSSVM) model for estimating ETo. The LSSVM model used in this study considers climatic factors as input variables and the ETo calculated by the Penman-Monteith equation as an output variable. Compared with the artificial neural network (ANN) model, which was developed with the same data, LSSVM prediction shows higher accuracy, efficiency, and generalization performance. Therefore, it can be used as a complementary ETo estimation method. © 2012 IFIP International Federation for Information Processing.