Using seasonal climate forecasts in forecasting the Australian wheat crop

Recent research has pointed to the potential to forecast the Australian wheat crop using seasonal climate forecasts. Over the last decade seasonal climate outlooks have been issued for the wheat-growing season and regional crop yield modelling has progressed to an operational stage. To quantify the potential benefits to crop production forecasting of combining seasonal climate forecasts and regional crop yield models, long-term wheat production across the Australian wheatbelt was simulated using an agroclimatic stress index model (STIN) and grouped according to the phase of the SOI in April-May (around planting). Throughout the wheatbelt there was a clear tendency for lower production in SOI phase 1 (consistently negative SOI) or phase 3 (rapidly falling SOI) years and for higher production in phase 2 or 4 years (consistently positive or rapidly rising SOI). There was greater spread in the simulated yields associated with some phases and greater certainty of obtaining production anomalies with others. This differentiation of productivity early in the crop season suggested that seasonal climate forecasts based on SOI phases would likely provide useful skill when used in real-time yield forecasting. To test this capability in a forecasting mode, an operational yield forecasting model (Weighted Rainfall Index - WRI) was tested with hindcasts using 1986-95 rainfall and compared with forecasts issued by the Australian Bureau of Agricultural and Resource Economics (ABARE). These hindcasts showed that use of the WRI model generally improved on ABARE forecasts, with mean prediction errors reduced 25-30% when average rainfall was assumed throughout the growing season. The addition of rainfall forecasts based on SOI phases, however, gave little improvement in yield forecasts. This may have been due to the unrepresentativeness of the short hindcast test period and differences between the WRI and STIN models. A single analogue forecast system was also tested, but this highlighted the need for a number of analogues to stabilise mid-season forecasts and mean prediction errors. Further studies to examine these aspects are needed to better test the utility of current seasonal forecasting capability, before definitive conclusions are possible.