Click on “Download PDF” for the PDF version or on the title for the HTML version.
If you are not an ASABE member or if your employer has not arranged for access to the full-text, Click here for options.
OPTIMIZATION OF FUZZY EVAPOTRANSPIRATION MODEL THROUGH NEURAL TRAINING WITH INPUT–OUTPUT EXAMPLES
Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan www.asabe.orgCitation: Transactions of the ASAE. 44(6): 1625–1633. (doi: 10.13031/2013.7049) @2001
Authors: L. O. Odhiambo, R. E. Yoder, D. C. Yoder, J. W. Hines
Keywords: Evapotranspiration estimation, Fuzzy logic, Fuzzy–neural–model, Neural network
In a previous study, we demonstrated that fuzzy evapotranspiration (ET) models can achieve accurate estimation of daily ET comparable to the FAO PenmanMonteith equation, and showed the advantages of the fuzzy approach over other methods. The estimation accuracy of the fuzzy models, however, depended on the shape of the membership functions and the control rules built by trialanderror methods. This paper shows how the trial and error drawback is eliminated with the application of a fuzzyneural system, which combines the advantages of fuzzy logic (FL) and artificial neural networks (ANN). The strategy consisted of fusing the FL and ANN on a conceptual and structural basis. The neural component provided supervised learning capabilities for optimizing the membership functions and extracting fuzzy rules from a set of inputoutput examples selected to cover the data hyperspace of the sites evaluated. The model input parameters were solar irradiance, relative humidity, wind speed, and air temperature difference. The optimized model was applied to estimate reference ET using independent climatic data from the sites, and the estimates were compared with direct ET measurements from grasscovered lysimeters and estimations with the FAO PenmanMonteith equation. The modelestimated ET vs. lysimetermeasured ET gave a coefficient of determination (r 2 ) value of 0.88 and a standard error of the estimate (Syx) of 0.48 mm d 1 . For the same set of independent data, the FAO PenmanMonteithestimated ET vs. lysimetermeasured ET gave an r 2 value of 0.85 and an Syx value of 0.56 mm d 1 . These results show that the optimized fuzzyneuralmodel is reasonably accurate, and is comparable to the FAO PenmanMonteith equation. This approach can provide an easy and efficient means of tuning fuzzy ET models.(Download PDF) (Export to EndNotes)