Operational forecasting of South African sugarcane production: Part 2 - System evaluation

The performance of a model-based crop forecasting system is assessed in this paper. The operational error associated with a forecast originates from two independent sources. First, the system error reflects the system's ability to match yields simulated from historic data to actual yields. The system error is due to factors such as model and data inaccuracies, incorrect aggregation assumptions and the system's inability to reflect all the compelling factors, like pest and diseases, climatic disasters and suboptimal crop management. Second, the climate error reflects inaccuracies of the operational yield forecasts associated with the assumed future climate. The purpose of the study was to assess the performance of a system to forecast sugarcane yields by quantifying the accuracy of (1) estimates based on complete sets of actual weather data and (2) operational system forecasts with incomplete sets of actual weather data. Estimates and forecasts were compared to actual yields recorded from 1980 to 2004. Industry production data from 1980 to 2002, corrected for various time trends, were used to calculate the system error for mills and the industry. The skill of estimation was calculated by comparing the size of the system error with the observed seasonal variation. On an industry scale, estimates captured 57% of inter-annual variability. Production at most mills was also simulated well, with some exceptions in irrigated areas. Operational forecasts issued between January and April for the industry between 1998 and 2004 had an average forecast error of 4.0%, which is 2.2% lower than the equivalent mill committee forecasts. The study provides ample evidence that industry stakeholders could use information from this system to enhance their management of sugarcane production.