A Creative Computing Approach to Forecasting Yield Shock of Winter Wheat

The United States stands as one of the world's largest wheat exporters, holding more than 10% of the global export market. Recognising the significant global economic repercussions that would arise from a negative yield shock of wheat, this study aims to use machine learning techniques to determine the key climate drivers of the winter wheat yield shock to facilitate forecasting three to six-month yield shock. Winter wheat is a strain of wheat that constitutes approximately 55% of total wheat plantations in the US. To accomplish this goal, this research uses a machine learning approach and employs support vector machines as a foundation. This research's findings reveal the designed model can achieve over 90% accuracy and ROC AUC score classification results. Notably, the temperature in the winter and spring months are identified, as well as the total precipitation of the autumn months would influence the forecast of yield shock the most. The result of this research validates the proposed approach that machine learning can be used to provide valuable information and thus can be applied when analysing other regions, e.g., the UK.