Artificial intelligence based and linear conventional techniques for reference evapotranspiration modeling

Abstract

This study was aimed at investigating the potentials of Artificial Neural Network (ANN), Support Vector Regression (SVR) and Multiple Linear Regression (MLR) techniques for the estimation of reference evapotranspiration (ET0 ) in a semiarid station of Nigeria. To do so, 34 years daily monthly average data including maximum, minimum and mean temperatures (Tmax, Tmin, and Tmean ), relative humidity (RH) and wind speed (U2) were used as input parameters. Three models were developed from each technique using three different input combinations. FAO Penman Monteith method was used as the basis upon which the performances of the models were assessed. The results revealed that models developed using Tmin, Tmax and U2 produced better performance. The results also depicted that with the unique capability of each technique, different results would be obtained, both ANN and SVR models could lead to efficient and reliable results, but MLR model could not produce reliable performance due to its inability to deal with nonlinear aspect of ET0.