Comparison of multivariate linear regression and artificial neural networks models for estimating of rainfed wheat yield in some central Zagros areas

Given the importance of wheat in human nutrition and its cultivation in large-area under rainfed in Iran, this study was aimed to evaluate the efficiency of artificial neural networks and linear multiple regression models to predict biomass and grain yields of wheat (cv. Sardari), in two-year study. In two stations (Koohrang and Ardal), 202 sampling points were selectedin the various hillslopes includes summit, shoulder, back slope, foot slope and toe slope. Atthe harvesting stage, the soil and plant samples were collected. Primary and secondary terrain attributes were extracted from digital elevation models, and meteorological data were used in two regions. Topography, 54 different soil characteristics, rainfall and management as the inputs as well as biomass and grain yields were considered as the outputs of both models. Artificial neural networks and multiple linear regression models, respectively, accounted for 84% and 15% of variations (R2) in grain yield prediction, and 76% and 6% in prediction of biomass yield. The root mean square error (RMSE) of the models also were equal to 0.033 and 0.092 to predict grain yield, and 0.037 and 0.102 to predict the biomass based on artificial neural network and multiple linear regression models, respectively. The results showed a better ability of artificial neural networks in comparison with multiple linear regression to estimate grain and biomass yields of wheat in the target areas.