Daily Evapotranspiration Modeling from Limited Weather Data by Using Neuro-Fuzzy Computing Technique

Abstract

Evapotranspiration is an integral part of the hydrologic cycle and an important component in water resource development and management, especially in the arid and semiarid conditions such as those found in Iran in which water resources are limited. The standard Food and Agricultural Organization of the United Nations (FAO)-56 Penman-Monteith (PM) equation requires several meteorological inputs for estimating reference evapotranspiration (ETo) that are not usually available in most of the stations. This paper investigates the potential of the adaptive neuro-fuzzy computing technique (ANFIS) for daily reference evapotranspiration modeling under arid conditions from limited weather data. The gamma test technique is applied to find the best input combination and number of sufficient data points for the model calibration. The training and testing data sets are chosen on the basis of the K-fold method of cross-validation to obtain the optimal classifier. The estimates of ANFIS models are compared with calibrated FAO-56 reduced-set PM ETo approaches and some calibrated empirical ETo equations such as Hargreaves, Priestley-Tailor, Makkink, and Blaney-Criddle equations. The FAO-56 full-set PM is adopted as the reference ETo equation, and it is applied to calibrate other ETo equations and ANFIS models. The comparison results indicate that when similar meteorological inputs are used, the ANFIS models performed better than all the methods pursued. This fact strongly suggests using ANFIS technique as an accurate ETo estimator method even in the absence of complete weather data. The minimum required data to construct a good ANFIS model under arid conditions are the minimum and maximum air temperatures and wind speed data.