Improving the seasonal prediction of Northern Australian rainfall onset to help with grazing management decisions

Abstract

The development of the Australian Community Climate and Earth-System Simulator-Seasonal prediction system version 1 (ACCESS-S1) signifies a major step towards addressing predictive limitations in multi-week to seasonal forecasting throughout Australia. It is anticipated that moving to ACCESS-S1 will provide improved skill in rainfall prediction during the dry to wet season transition period across tropical northern Australia. This is an important time for northern Australian livestock producers in terms of the decisions they make around pasture and livestock management. This study quantifies the hindcast skill of ACCESS-S1 for the northern rainfall onset (NRO), defined as the date when 50 mm of precipitation has accumulated at a given location from the 1st of September, heralding the shift towards greener pastures. We evaluate the raw model hindcasts, and compare them to hindcasts corrected for mean biases and those calibrated against observations. It is found that the raw ACCESS-S1 hindcasts broadly replicate the observed median NRO over the period 1990–2012, despite a tendency for earlier than observed onsets. In terms of forecasting the interannual variability of the NRO, the calibrated hindcasts show the greatest skill, with the largest improvements over a climatological forecast in their probabilistic forecasts of an earlier or later than usual onset, with a large portion of northern Australian showing more than 10% improvement. With real-time NRO forecasts now generated by ACCESS-S1, it is expected that the calibrated predictions will help northern Australian graziers make better informed decisions around livestock management prior to the wet season.