Forecasting of garlic (Allium sativum L.) evapotranspiration by using multiple modeling

Estimation and evaluation of crop water requirement is important key in scheduling, management and using from agricultural water resources. Many methods for estimation of phenomena are available that in this regard multiple modeling methods are prevalent. In this study Adaptive Neural Network based Genetic Algorithm and Co-Active Neuro-Fuzzy Inference System in side Artificial Neural Network for estimation of garlic evapotranspiration were used. For modeling from meteorological data including minimum and maximum temperature, minimum and maximum relative humidity, wind speed and sunshine hours and lysimeter data in climatology Station of Bu-Ali Sina University Agriculture Faculty during 2008-2009 for modeling were used. The garlic lysimeter Evapotranspiration maximum and average were 11.22 and 4.79 mmd-1 respectively. Results showed that Adaptive Neural Network based Genetic Algorithm were more accurate than Co-Active Neuro-Fuzzy Inference System. In other words, optimization of Artificial Neural Network more successful operation of fuzzy logic can be. Also, Artificial Neural Network was more appropriate to Co-Active Neuro-Fuzzy Inference System. So that Adaptive Neural Network based Genetic Algorithm than Co-Active Neuro-Fuzzy Inference System reduced 24 percent the testing error. But Co-Active Neuro-Fuzzy Inference System not only improves the results of artificial neural network was not, but reduced 12 percent the prediction accuracy. For the best architecture of Adaptive Neural Network based Genetic Algorithm, statistical parameters were RMSE= 0.12 mmd-1 and r=0.88. Coefficient of determination criteria, the width of the origin and the slope of line fitted to data estimated the Adaptive Neural Network based Genetic Algorithm model period (two years) compared with the Co-Active Neuro-Fuzzy Inference System model also showed, this model is efficiency.