A k-means clustering approach to assess wheat yield prediction uncertainty with a HYDRUS-1D coupled crop model

Soil moisture, especially under drought conditions, is a factor that is known to impact crop yield predictions. Crop growth models used to make these predictions rely on soil texture estimates, which influence simulated soil moisture and ultimately crop growth. The purpose of this research was to implement a k-means clustering approach to address the uncertainty of the soil texture estimates. By grouping similar soil textures based on their simulated responses, clustering reveals how soil texture uncertainty may impact yield estimates. Wheat growth simulations were conducted using a HYDRUS 1D and coupled crop model for soils defined on the USDA soil texture triangle. A k-means clustering algorithm was applied to the simulated biophysical data for each soil texture. Resulting clusters were different from traditional soil type classifications. The k-means clustering approach proved useful for investigating the relationship to soil texture that crop yield may have. This research shows that the impact of soil texture variation should be considered when conducting crop growth simulation for the purposes of yield forecasting.