Production Potential Prediction for Wheat, Barley and Maize Based on Soil Characteristics Using Artificial Neural Networks in Varamin Region, Iran

Abstract

Abstract. Due to increasing demands for agricultural products and the problems to generate new data, using proper models to predict the performance of agricultural products seems necessary. The objective of this study was to assess the ability of artificial neural networks (ANN) for yield prediction of wheat, barley and maize to determine the most important soil properties for land production potential in Varamin region, Iran. In this study, potential production was calculated using the AEZ model. ANN model inputs were soil characteristics, including the percentage of calcium carbonate, coarse fragments, gypsum, clay, silt and sand, electrical conductivity, sodium absorption ratio and pH. ANN model output was production potential. Mean absolute error, root means square error and coefficient of determination criteria were used to evaluate the performance of the ANN. The obtained results showed that the ANN models with two hidden layers provided the most accurate prediction of production potential. Mean land production potentials for wheat, barley and maize, using ANN models were predicted respectively, as 7417.73, 6810.41 and 8922.48 kg/ha. Sensitivity analysis of models showed that wheat production potential has the most dependence on sodium absorption ratio and clay percent. Sand and coarse fragments rates were the most important parameters for predicting the barley production potential. The sodium absorption ratio and electrical conductivity were determined as the relevant and effective soil properties on maize production potential. Also, the optimum model for maize production potential prediction has less accuracy than wheat and barley. The reason of this may be differences in the environmental compatibility of various products to adverse conditions.