SCIENCE OF SEASONAL CLIMATE PREDICTION

Imperfect forecasts and forecast value
by Andrew L Vizard, School of Veterinary Science, The University of Melbourne

Associate Professor Andrew Vizard has a background in research and consultancy. He is an Associate Professor of Veterinary Epidemiology at The University of Melbourne (part-time) and a senior consultant and former Director of the Mackinnon Project at the same university. The Mackinnon Project is recognised as a leader in delivering practical advice to farmers and agribusiness on a wide range of agricultural and economic issues. He is the author of over 50 scientific papers. He is also a director of several ASX listed companies and government instrumentalities.

The utility of probabilistic seasonal rainfall forecasts to assist with rational-decision making is highly dependent upon the reliability and the resolution of the forecasting system. Value score curves can be used to quantify the expected impact of unreliable or poor resolution forecasting systems for any given decision that an end-user may face. Using this approach, it has been shown that unreliable forecasts can impart negative value to users of the forecasts. Reliable forecasts should therefore be the primary aim of any seasonal rainfall forecasting system. Similarly, it has been shown that the value of forecasting systems with poor resolution is highly eroded and limited to decisions that are triggered by a small shift in the forecast from climatology. Analyses have demonstrated that the resolution of both the Australian Bureau of Meteorology seasonal rainfall forecasting system and the seasonal rainfall forecasting system based on the five phases of the Southern Oscillation Index is relatively poor, consequently constraining their utility to end users. Efforts should therefore be made to improve the resolution of these systems. However, any future minor improvement is unlikely to generate significant and widespread benefits to users. To deliver uniform and widespread value to users of forecasts, new lead indicators with markedly better predictive characteristics may need to be developed.

References