LAUSR

Abstract Because seasonal rainfall is the largest driver of grain production in Australia, forecasts of can increase the profitability of grain farms when used to inform on-farm decision making. Seasonal forecasting in the grainbelt of Western Australia (WA) is difficult because climate drivers that affect the Australian continent have little influence on the south-west corner. The Statistical Seasonal Forecast system was developed to provide skilful forecasts to assist on-farm decision making in the south-west of WA. It has now been operational from 2013 to 2018. This article describes the models and methods used to create forecasts, the results of forecast validation, and verification of the accuracy of six years of operational seasonal forecasts. Forecast validation was performed using leave-one-out cross-validation. It showed positive skill, lower for late spring and summer forecasts that rises in autumn and is highest for forecasting winter rainfall. Skill for forecasting growing season (May to October) is good for lead times of up to three months. Verified skill of the operational forecasts is lower than estimated using cross-validation. The operational system performed reasonably well at forecasting 3-month periods, and for most years it showed better skill than climatology. Verified skill for growing season forecasts is higher than that of forecasting any 3-month period, with higher skill in the south. With only a small sample of six years of forecasts, longer-term accuracy may be better or worse. To test whether declining accuracy is a risk, use of a less generous measure of forecast validation is recommended.