Application of Bayesian Model Averaging (BMA) Approach to Estimating Wheat Yield in Golestan Province

Wheat is a strategic crop for food security, so accurate forecasting of its performance is important for planning and adopting appropriate policies for import or export. Crop models use climatic, soil, cultivar type and crop management data to predict crop growth, development and yield. The purpose of this study was to improve the accuracy of estimating wheat yield using a combination of recommended crop models in Golestan province using Bayesian model averaging (BMA). In this study, first, the ability of DSSAT, CropSyst and SSM-Wheat models was evaluated to estimate wheat yield in Golestan province. According to the results, the DSSAT model with root mean square error (RMSE) equal to 290 kg ha-1, coefficient of determination (R2) equal to 96%, mean square root of normalized error (NRMSE) equal to 6.34 and the efficiency of the model (EF) equal to 0.9 has the most accurate estimation as compared to two other models. In the next step, using the BMA approach, binary and ternary combinations were taken from three crop models, which led to the production of four new models. Comparing the performance of BMA models with individual models, showed that the combination of models improves the accuracy of estimation. Thus, the amount of RMSE in the C23 model (which is the result of combining DSSAT and SSM-Wheat models (was reduced by 30% compared to the amount of yield calculated by the best single model (DSSAT) and reached 202 kg ha-1. Also, the value of R2 in the C23 model reached 97%. Therefore, the combination of models improves the accuracy, estimating wheat yield.