A Bayesian modelling approach for long lead sugarcane yield forecasts for the Australian sugar industry

Abstract

For marketers, advance knowledge on sugarcane crop size permits more confidence in implementing forward selling, pricing, and logistics activities. In Australia, marketing plans tend to be initialised in December, approximately 7 months prior to commencement of the next harvest. Improved knowledge about crop size at such an early lead time allows marketers to develop and implement a more advanced marketing plan earlier in the season. Producing accurate crop size forecasts at such an early lead time is an on-going challenge for industry. Rather than trying to predict the exact size of the crop, a Bayesian discriminant analysis procedure was applied to determine the likelihood of a small, medium, or large crop across 4 major sugarcane-growing regions in Australia: Ingham, Ayr, Mackay, and Bundaberg. The Bayesian model considers simulated potential yields, climate forecasting indices, and the size of the crop from the previous year. Compared with the current industry approach, the discriminant procedure provided a substantial improvement for Ayr and a moderate improvement over current forecasting methods for the remaining regions, with the added advantage of providing probabilistic forecasts of crop categories.