Subseanonal Prediction of Summer Monsoon Rainfall in South America and the MJO as a Source of Subseasonal Predictability

The subseasonal prediction of monsoon anomalies in South America (SA) and the contribution of the Madden Julian Oscillation (MJO) to subseasonal predictability are assessed for two models (ECMWF and CFSv2).

In parts of the core monsoon region the MJO-related daily rainfall anomalies reach 30% of the intense climatological precipitation and the frequency of extreme events doubles, thanks to tropics-tropics and tropics-extratropics teleconnections.

Two indices are defined to represent the monsoon variability and to evaluate the models skill in predicting it: the monsoon precipitation index (MPI) and the monsoon wind index (MWI). The MPI is the standardized anomaly of rainfall in the core monsoon region and defines active and break monsoon periods when it is greater (less) than +1(-1). It captures well the subseasonal variability of the monsoon rainfall, since the anomaly patterns for active and break episodes reproduce the pattern of the first mode of synoptic and intraseasonal variability of the observed summer daily rainfall over SA. The two indices are dynamically linked, and the models simulate well the relationships between them.

Active and break monsoon rainfall episodes are associated with circulation anomalies that are very similar to those associated with MJO-related positive and negative precipitation anomalies in the core monsoon region. Therefore, the MJO is a source of monsoon subseasonal predictability. The observed connection between the MJO phases and the active and break episodes is well represented by the models up to week 3, but shifted by one phase, showing highest number of active episodes in phase 8, while observations show it in phase 1. Coherently, the crucial teleconnection between the central subtropical Pacific and SA is stronger in observations (models) in phase 1 (8), apparently taking shorter time in the models.

The prediction correlation skill for the MWI is higher than for MPI, but after three weeks it drops much below 0.5 for both. For the first 3 weeks, the skill for the models is higher than for a MJO-based 'perfect' forecast, since the models capture predictable weather phenomena and other intraseasonal variability besides the MJO. Beyond 3 weeks, however, the 'perfect' MJO component provides higher predictive skill than the models, accounting for much of the predictable rainfall.