Estimating daily reference evapotranspiration using available and estimated climatic data by adaptive neuro-fuzzy inference system (ANFIS) and artificial neural network (ANN)

• Published in: Hydrology Research 2013-01, Vol.44 (1), p.131-146
• Main Authors: BABA, Ana Pour-Ali, SHIRI, Jalal, KISI, Ozgur, FAKHERI FARD, Ahmad, KIM, Sungwon, AMINI, Rouhallah
• Format: Article
• Language: English
• Subjects: Adaptive systems; Air temperature; Artificial neural networks; Climatic data; Coefficients; Daily; Data processing; Earth sciences; Earth, ocean, space; Engineering and environment geology. Geothermics; Estimating; Estimation; Evapotranspiration; Exact sciences and technology; Fuzzy logic; Fuzzy systems; Goodness of fit; Hydrology; Hydrology. Hydrogeology; Independent variables; Inference; Learning theory; Mathematical models; Neural networks; Pollution, environment geology; Radiation data; Relative humidity; Root-mean-square errors; Solar radiation; Solar radiation data; Sunlight; Weather stations; Wind speed
• ISSN: 1998-9563; 0029-1277; 2224-7955
• DOI: 10.2166/nh.2012.074

Daily reference evapotranspiration (ET0), as a dependent variable, was estimated for two weather stations in South Korea, using 8 years (1985–1992) of measurements of independent variables of air temperature, sunshine hours, wind speed and relative humidity. The model uses the adaptive neuro-fuzzy inference system (ANFIS) and artificial neural networks (ANNs) for estimating daily ET0. In the first part of the study, the applied models were trained, tested and validated using various combinations of the recorded independent variables, which corresponded to the Hargreaves–Samani, Priestly–Taylor and FAO56-PM equations. The goodness of fit for the models was evaluated in terms of the coefficient of determination (R2), root mean square error (RMSE), mean absolute error (MAE) and Nash–Sutcliffe coefficient (NS). In the second part of the study, the estimated solar radiation data were applied as input parameters (for the same input combinations, as the first part), instead of recorded sunshine values. The results indicated that the both applied ANFIS and ANN models performed quite well in ET processes from the available climatic data. The results also showed that the application of estimated solar radiation data instead of the recorded sunshine values decreases the models’ accuracy.