Hossien Riahi Modavar; Abbas Khashei-Siuki; Akram Seifi
Abstract
Because of saffron yield sensitivity and the effects of climate on its performance, and also due to the nonlinear nature of crop yield functions, the Artificial Neural Network (ANN) model is employed in this study for prediction and uncertainty analysis of saffron yield in the South Khorasan province ...
Read More
Because of saffron yield sensitivity and the effects of climate on its performance, and also due to the nonlinear nature of crop yield functions, the Artificial Neural Network (ANN) model is employed in this study for prediction and uncertainty analysis of saffron yield in the South Khorasan province based on 20 years of data. The input vector of the ANN model was optimized from 37 parameters through correlation and variance inflation. The optimum architecture of the model was derived as 1-2-4-11 with a sigmoidal activation function based on the results at three stages of training, testing and verification. The root mean square error (RMSE) and mean absolute error (MAE) were equal to 0.3 and 0.5 in the training step and 0.7 and 1 in the test step, respectively. These results indicate that the ANN is a suitable model for predicting saffron yield. Uncertainty analysis based on R2, d-factor and 95%PPU showed that despite use of inadequate data, model prediction showed acceptable prediction bounds and predicted a satisfactorily saffron yield trend. The R2 values were equal to 0.92 and 0.58 in the training and test steps, respectively, which are statistically significant at the P
