Artificial neural networks, regression and empirical methods for reference evapotranspiration modeling in Inhambane City, Mozambique

Abstract

Precise estimation of reference evapotranspiration (ETo) is important for designing and managing irrigation systems. Methods of ETo estimation (11 empirical methods, 10 multiple regression models: RLM and 10 artificial neural networks: RNAs) were evaluated against Penman Monteith FAO 56 method using the following indexes: MBE (Mean Bias Error), RMSE (Root Mean Square Error) and R2, and RMSE was used as the main criterion of method selection. The significance of the methods was evaluated on the basis of the t test using data from 1985 to 2009. The meteorological data used (maximum temperature: Tmax, minimum temperature: Tmin and average temperature: T, relative air humidity, wind speed and solar brightness), from 1985 to 2009, are from the conventional meteorological station of the city of Inhambane, Mozambique. The results showed that the RLM4 model presented better performance (MBE = 0.01 mm.d-1; RMSE = 0.15 mm.d-1; R2 = 0.99). In the absence of global solar radiation, RLM6 (MBE =-0.01 mm.d-1; RMSE = 0.23 mm.d-1; R2 = 0.97) and RLM10 (MBE = 0.01 mm. d-1; RMSE = 0.23 mm.d-1; R2 = 0.97) can be used, which require measurement of T, and Tmax and Tmin, respectively. These models were not statistically different from the standard method.