Evapotranspiration Transfer-Function-Noise Modeling

Abstract

Water-resource projects require a correct estimation of evapotranspiration for their successful planning, design, and operation. The accuracy of evapotranspiration is particularly important in, for example, water-quality models. Evapotranspiration estimation is very sensitive to the degree of forecasting error and even small errors can produce highly misleading results. Unfortunately, the time-dependent behavior of evapotranspiration makes it very hard to be represented by simpler expressions and forecasting based on such relations will naturally be inaccurate. In this study, the inherently stochastic process of reference evapotranspiration is modeled in a dynamic regression environment, also known as transfer-function-noise (TFN) modeling. TFN models have the capability to relate different time series and the capacity to allow delays between system inputs and responses, which make them quite attractive. In this study, various single input-single output TFN models are developed for reference evapotranspiration. It is shown that evapotranspiration can be adequately represented and forecasted by TFN models.