Development and evaluation of a sugarcane yield forecasting system.
There is a need in the South African sugar industry to investigate improved techniques for forecasting seasonal sugarcane yields. An accurate and timely forecast of seasonal cane yield is of great value to the industry, and could potentially allow for substantial economic savings to be made. Advances by climatologists have resulted in increasingly accurate and timely seasonal climate forecasts. These advances, coupled with the ongoing advances made in the field of crop yield simulation modelling, present the sugar industry with the possibility of obtaining improved cane yield forecasts. In particular, the lead time of these forecasts would be improved relative to traditional techniques. Other factors, such as the flexibility offered by simulation modelling in the representation of a variety of seasonal scenarios, would also contribute to the possibility of obtaining improved cane yield forecasts. The potential of applying crop yield simulation models and seasonal rainfall forecasts in cane yield forecasting was assessed in this research project. The project was conducted in the form of a case study in the Eston Mill Supply Area. Two daily time step cane yield simulation models, namely the ACRU-Thompson and CANEGRO-DSSAT models, were initially evaluated to test their ability to accurately simulate historical yields given an observed rainfall record. The model found to be the more appropriate for yield forecasting at Eston, the ACRU-Thompson model, was then used to generate yield forecasts for a number of seasons, through the application of seasonal rainfall forecasts in the model. These rainfall forecasts had previously been translated into daily rainfall values for input into the model. The sugarcane yield forecasts were then evaluated against observed yields, as well as against forecasts generated by more traditional methods, these methods being represented by a simple rainfall model and Mill Group Board estimates. Although the seasonal rainfall forecasts used in yield forecasting were found not to be particularly accurate, the proposed method provided more reliable cane yield forecasts, on average, than those using the traditional forecasting methods. A simple cost-benefit analysis indicated that the proposed method could potentially give rise to the greatest net economic benefits compared to the other methods. Recommendations are made for the practical implementation of such a method. Future areas of research are also identified.