Previous work has shown that the Madden-Julian Oscillation (MJO), as depicted by the Real-time Multivariate MJO (RMM) indices, is well forecast by the POAMA model to a lead of ~3 weeks. However, the global scale RMM index does not necessarily reflect local conditions in the monsoons, so here we consider the prediction of a more local index of convection.
Hindcasts with POAMA have been performed for the period 1980-2010, initialised on the 1st, 11th and 21st of each month. Each hindcast consists of an ensemble of 33 members using perturbed initial conditions generated by the Coupled Ensemble Initialisation Scheme. The measure of convection we use is outgoing longwave radiation (OLR) averaged over a box covering the Australian monsoon region (120ºE-150ºE, 5ºS-17.5ºS). This averaging serves to focus on the intraseasonal and longer time scales, and is an area of interest to monsoon forecasters.
The POAMA hindcasts of daily OLR show a strong systematic adjustment away from their initial values during the first week, and then converge to a mean seasonal cycle of similar amplitude and phase to observations. Hence, forecast OLR anomalies are formed by removing the model's own seasonal cycle of OLR, which is a function of start time and lead time.
Skill of the hindcast anomalies is assessed against both climatology and persistence. Over all hindcasts, POAMA exhibits skill, as measured by the RMSE, over climatology to at least 35 days, and against persistence, as measured by RMSE or correlation, at all lead times beyond 2 days. Similar results are found for just those hindcasts beginning in the "wet season" months from December through to March, when the greatest variability in OLR occurs.
A real-time forecast display is generated whereby the ensemble mean anomaly and each ensemble member is plotted relative to the observed climatology providing a total measure of convection that includes the seasonal cycle.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner