Using data-driven forecasts of sugarcane yield to optimise variable N-fertiliser application

The Australian sugar industry is under international scrutiny due to nutrient-rich runoff that flows into the Great Barrier Reef. In response, there have been industry-wide efforts to introduce precision agriculture (PA) into sugarcane (Saccharum spp.) production. The overall aim of PA is to variably apply agricultural inputs according to crop needs in space and time, which are reflected by the potential yield and sugar content of the crop. Current methods to digitally forecast sugarcane yield incorporate a range of data; from crop simulation models (such as APSIM-sugar) to satellite imagery. However, simulation models require many strict inputs, and satellite imagery approaches are only able to accurately predict in-season yield well beyond the nutrient application and management intervention deadline for growers. Commercial cane sugar (CCS) is rarely modelled despite being a key factor in the price return for sugarcane. This project aimed to include a wider range of publicly available spatio-temporal data (e.g. satellite imagery, climate, terrain attributes and soil maps) in tandem with data collected by growers and mills (e.g. past yield, relative CCS, ratoon number, planting and harvest dates) to more closely represent the factors that drive yield and CCS.