Application of Heuristic Methods in Prediction of Wheat Yield

Increasing demand for agricultural products and lack of appropriate soil and water resources with problems of field research reveals the application of efficient models to predict crop yield. This research aimed to examines the efficiency of artificial neural networks, comparative fuzzy neural network, adaptive nero fuzzy inference system and particle swarm optimization algorithm models for estimating the wheat yield through soil and land properties. For this purpose, 80 soil profiles were drilled in wheat fields’area in East Azerbaijan province with temperature and moisture regimes of mesic and aridic border to xeric, respectively. Soil samples were collected from each genetic horizon. The results of sensitivity analysis showed that total nitrogen, absorbable phosphorus, slope, gravel, soil reaction and organic matter are effective soil properties in wheat yields. The hybrid model of PSO-ANFIS was the best model from the viewpoint of statistical indices including R2 (0.89) and RMSE (213.5). Also, neuro-fuzzy method has a R2 (0.84) and RMSD (243.2) and artificial neural networks have a R2 (0.81) and RMSD (274.5), respectively. The GMER index also indicated overestimation of artificial neural network (0.24) and nero fuzzy (0.53) and underestimation of PSO-ANFIS model (1.13). The results indicated that the hybrid neuro-fuzzy-swarm particles model performed better than other models that can be used a powerful tool for estimating wheat yield.