Application of Artificial Intelligence in Agrometeorology
Abstract Replacing irrigated with rainfed crops and sustainable production of major rainfed plants (such as wheat) can be an efficient strategy to restore water resources that are drying up. Identifying plant response to climate is essential to advancing this strategy and planning for precision agriculture. Wheat is the main plant of Saqez in the Lake Urmia basin of Iran, whose yield is associated with severe fluctuations. This study was conducted to investigate the climate effect on wheat yield fluctuation. For this purpose, the method of growing degree days (GDDs) and the Zadoks scale were used to divide the wheat growth period into seven stages. Forty-seven climatic variables of the first six stages were used to do factor analysis and to develop the model for forecasting pre-harvest yield. Gene expression programming (GEP), artificial neural networks (ANNs), and multivariate linear regression (MLR) methods were applied to develop the model. The results showed that 90.7% of the total variance of 47 variables can be explained by 10 factors. Eighty-two percent of yield variations were modeled by these 10 factors (r = 0.91). The mean absolute percentage error (MAPE) for the models developed by the GEP and ANN methods was 26%, and its amount for the MLR model was 35%. In this study, for the first time, the GEP method was used to model rainfed wheat yield. Comparison with MLR and ANN methods shows that GEP is suitable for modeling in this field.