Plausibility of Training Artificial Neural Networks with Crop Models to Predict Wheat Phenology and Yield

Increasing the demand for agricultural products and increasing the pressure on water and soil resources, and the problems of access to field data on the other, will require the use of appropriate models to predict the yield of agricultural products. The most of the input parameters of crop models are not available in our country. On the other hand, with the emergence of strong statistical techniques and neural networks, predictive models of crop yields are rapidly developing. For this purpose, the present study was aimed to evaluate the capability of Artificial Neural Networks in learning from complex crop models and their performance in predicting wheat phenology and yield as well as some input parameters of the AquaCrop model. The evaluation of the models was done by statistical indexes, coefficient of determination (R²), normal root mean square error (SRMSE). The results showed that the neural network of model No. 9 (duration of growth period from flowering to harvest) was fitted with R2 and SRMSE, 0.98 and 4.79%, respectively, and the neural network of model No. 2 (Grain DM at maturity) was fitted with R2 and SRMSE, 0.97 and 9.69%, respectively, the best performance among predictive models of duration of growth periods and yield of wheat grain. According to this paper , The efficiency of artificial neural networks has been confirmed to predict yield and duration of growth periods of wheat using climate parameters.