Jump to Main Content
A simple and parsimonious generalised additive model for predicting wheat yield in a decision support tool
- Chen, Kefei, O'Leary, Rebecca A., Evans, Fiona H.
- Agricultural systems 2019 v.173 pp. 140-150
- data collection, decision support systems, disease control, farmers, grain yield, growing season, land classification, meteorological data, pastures, prediction, rain, simulation models, soil types, weeds, wheat, yield forecasting, Western Australia
- Yield prediction is a major determinant of many management decisions for crop production. Farmers and their advisors want user-friendly decision support tools for predicting yield. Simulation models can be used to accurately predict yield, but they are complex and difficult to parameterise. The goal of this study is to build a simple and parsimonious model for predicting wheat yields that can be implemented in a decision tool to be used by farmers at a paddock level.A large yield data set accumulated from trials on commonly grown varieties in Western Australia is used to build and validate a generalised additive model (GAM) for predicting wheat yield. Explanatory variables tested included weather data and derivatives, geolocation, soil type, land capability, and wheat varieties. Model selection followed a forward stepwise approach in combination with cross-validation to select the smallest set of explanatory variables. The predictive performance is also evaluated using independent data.The final model uses seasonal water availability, location and year to predict wheat yield. Because the GAM model has minimal inputs, it can be easily employed in a decision tool to predict yield throughout the growing season using rainfall data up to the prediction date and either climatological averages or seasonal forecasts of rainfall for the remainder of the growing season. It also has the potential to be used as an input to agronomic models that predict the effect on yield of various management choices for fertiliser, pest, weed and disease management.