Multi-week prediction of the Madden–Julian oscillation with ACCESS-S1

Abstract

We assess the ability of the Bureau of Meteorology’s new ACCESS-S1 dynamical forecast system to predict the MJO using retrospective forecasts for the period 1990–2012. Compared to the benchmark POAMA-2 system, ACCESS-S1 demonstrates improved skill in predicting the ensemble mean bivariate RMM index by about 4 days lead time in austral summer and 5 days in boreal summer. Probabilistic forecast scores further demonstrate improved skill in predicting MJO amplitude by at least 7 days, and MJO phase by about 9 days. However, the ensemble from ACCESS-S1 for the MJO is underdispersed, indicating further gains in forecast skill can still be achieved. Improvements in the regional depiction of MJO rainfall in ACCESS-S1 over POAMA-2 include a more realistic southward extension of austral summer rainfall over Northern Australia, and a better overall spatial distribution and eastward extension of boreal summer rainfall over the tropical Indo-Pacific region. Both models depict well the northward propagation of boreal summer rainfall over the Indian Ocean warm pool. Overall, ACCESS-S1 simulates the MJO signature in global rainfall at least as well as, if not better than, POAMA-2.