Predicting Yield and Water Use Efficiency in Saffron Using Models of Artificial Neural Network Based on Climate Factors and Water

Abstract

The predicted models for crops yield are developing rapidly by the creation of new statistical techniques and neural networks. For this purpose, a research was carried out in the Torbat-e-Heydarieh region for predicting yield and water use efficiency of saffron by using an artificial neural network model. The model was calibrated and validated by using crop yield and climate parameters data during 2009-2010. The models were evaluated by using indices of correlation coefficient (R2), root mean squares error normalized (RMSEn), and mean squares error (MSE). The results showed that the suggested neural network (model No. 9) with having 2 hidden layers, 8 neurons, and R2= 0.97 (for saffron yield); and 1 hidden layer, 7 neurons, and R2= 0.90 (for water use efficiency) had a high accommodation with these two factors. Also, according to the indices RMSEn and MSE, model No. 9 simulated the yield and WUE of saffron with a high accuracy, such that RMSEn and MSE for yield in this model obtained were 2.78% and 0.0041, respectively; and for WUE they were calculated to be 5.41% and 0.0073, respectively. Also, the results of sensitivity analysis indicated that irrigation is the most important parameter for predicting yield and WUE, and after that is precipitation and solar radiation. Generally, use of the suggested neural network in this research can improve saffron cultivation in the Torbat-e-Heydarieh region.

Keywords

• Model
• simulation
• stigma
• Torbat-e-Heydarieh