Evaluation of Artificial Neural Network and Adaptive Neuro Fuzzy Inference System in Decreasing of Reference Evapotranspiration Parameters

AbstractThe present study is attempted to present the minimum required meteorological parameters for reference evapotranspiration estimation at Hamedan region of Iran from 1997 to 1998. Employing Pierson test, six meteorological parameters which are used by Penman-Montieth FAO-56 method including maximum and minimum air temperature, maximum and minimum relative humidity, wind speed and daily sunshine were composed and considered as 4 difference scenarios (called 1, 2, 3 and 4). These scenarios were applied to artificial neural network (ANN) and adaptive neuro-fuzzy inference system (ANFIS) for reference evapotranspiration estimation of the area using the Matlab software. The results of the scenarios were evaluated using the actual reference evapotranspiration (lysimeter data). The results showed that increasing of number of input layers data could not be based as obtaining the more exact results. Using the scenario 2, which was based on minimum and maximum temperature as well as daily sunshine, showed more reliable results using the ANN and ANFIS methods. The root mean square error (RMSE), mean absolute error (MAE) and R2 of examination step of this scenario were 0.09, 0.07 mm/day and 0.9, respectively. Overall, the statistic performances revealed that ANN and ANFIS had the same results and similar input layer sensitivity. The iteration times of the ANN and ANFIS methods to reach the best results were 26 and 40, respectively. Comparison between ANN (RMSE= 0.09 mm/day) and standard Penman-Montieth method (RMSE= 0.34 mm/day) confirmed that the intelligence approaches such as ANN are more accurate for reference evapotranspiration estimation.