Role of air-sea interaction for seasonal prediction of Australian monsoon rainfall

Seasonal prediction of Australian rainfall predominantly stems from the strong influence that El Niño \/Southern Oscillation exerts on Australian climate (e.g., McBride and Nicholls 1983). Eastern and northern Australia tend to experience drought during El Niño (and pluvial during La Nina), especially during winter and spring. Although El Niño tends to peak in austral summer, the impact of El Niño on Australian monsoon rainfall (AMR), especially in the monsoonal region in the north, weakens dramatically from spring into summer (e.g., Nicholls et al. 1982, McBride and Nicholls 1983). Insight into the lack of relationship between AMR in summer and ENSO is gained by examining the seasonal relationship of AMR with SST. During the pre-monsoon (spring season) the pattern of SST correlation associated with AMR is indicative of La Niña (or El Niño): enhanced (decreased) AMR is associated with strong negative (positive) correlation in the eastern equatorial Pacific and strong positive (negative) correlation in the seas to the north and east of Australia. However, these correlations weaken dramatically in summer, and the local positive correlation to the north of Australia even changes sign. That is, in the pre-monsoon AMR is associated with locally warm SST to the north of Australia but during the monsoon AMR has little or even negative relationship with local SST.

Seasonally varying air-sea interaction to the north of Australia has been posited to explain the dramatic drop in correlation between AMR and local and remote SST in going from the pre-monsoon into the monsoon season (e.g., Nicholls 1981; Hendon 2003; Wu and Kirtman 2007). The strong positive correlation of AMR with local SST in the pre-monsoon season occurs before the summer monsoon circulation is established (i.e., when northern Australia is still in a trade wind regime). During the winter (JJA) and spring (SON), the Australian monsoon region experiences trade easterlies. Anomalous easterlies (for instance, as driven remotely by El Niño in the Pacific) at this time of year, then would act to increase the total windspeed (easterly anomaly acting on an easterly basic state) thereby producing surface cooling through increase latent and sensible heat flux. Thus, a positive feedback is produced with negative local SST anomalies acting to raise surface pressure and producing stronger easterly anomalies. Once the Australian summer monsoon onsets and the mean winds to the north of Australia become westerly (e.g, Troup 1961), anomalous easterlies, now acting on a westerly basic state, will decrease the total windspeed, thereby acting to warm the ocean surface. Hence, the easterly anomalies during the summer monsoon will produce a negative air-sea feedback. The positive feedback during the pre-monsoon and negative feedback during monsoon is offered as an explanation for the strong correlation between El Niño and onset date (and El Niño and pre-monsoon rainfall) and for a weakening of the negative correlation between El Niño and northern Australia rainfall once the monsoon onsets.

Coupled climate models are now routinely used to make seasonal climate predictions (e.g., Wang et al. 2007). Here we focus on forecast skill in the Australian monsoon using the Bureau of Meteorology coupled model forecast system, POAMA (Predictive Ocean Atmosphere Model for Australia; Alves et al. 2003). Forecast skill is assessed using the ensemble mean of retrospective 9 month forecasts for the period 1980-2006. The ensemble mean was formed from ten forecasts that were initialized on the first of each month. POAMA can skilfully predict El Niño 2-3 seasons in advance (i.e., anomaly correlation of Nino34 SST index remains above 0.6 to beyond 9 month lead time). Mean state drift, especially related to the over-development of the equatorial Pacific cold tongue, does limit utility of these El Niño forecasts because the atmospheric teleconnection of ENSO degrades with increasing forecast lead time. The forecast model also severely underestimates the mean land based monsoonal rainfall, although the seasonality is well depicted. On the other hand, the intensity of the monsoonal circulation, as depicted by the seasonal development of the monsoonal westerlies to the north of Australia (e.g., Troup 1961), is well simulated), which presumably reflects a good depiction of the seasonal evolution of convection across the broader maritime continent region.

POAMA has no skill at predicting monsoon rainfall at any lead time (0-8 months) during the monsoon, while pre-monsoon rainfall is skilfully predicted at short lead times (1 and 3 month). The skill in the pre-monsoon stems from POAMA's ability to predict ENSO and to then simulate ENSO's teleconnection to Australian rainfall. Interestingly, POAMA overestimates the impact of ENSO on Australian rainfall during the monsoon especially at longer lead time (i.e., POAMA simulates a stronger negative correlation of monsoon rainfall with Nino34). On the other hand, POAMA does simulate the strong seasonality of the relationship of local SST and rainfall in the monsoon region (i.e., strong positive correlation in the pre-monsoon that then weakens or changes sign during the monsoon). Hence, it would appear that the seasonality of the critical air-sea interaction in the region to the north of Australia is captured in the forecasts, which suggests that future improvement of model bias may alleviate the erroneously strong ENSO teleconnection during the monsoon.

To conclude, seasonal mean Australian summer monsoon rainfall is not predictable with the POAMA model. However, pre-monsoon rainfall is predictable and predictions of the pre-monsoon (including onset of the wet-season) are of practical use. Seasonally varying air-sea interaction acts to promote local SST anomalies to the north of Australia in the pre-monsoon that compliment remote forcing of rainfall by ENSO, thus adding to the predictability of rainfall stemming from predictability of ENSO. During the summer monsoon air-sea interaction acts to suppress local SST anomalies to the north of Australia, thereby detracting from the predictability stemming from remote forcing of ENSO. There would appear to be some prospect for improved dynamical model prediction during the monsoon by reduction of mean-model bias and a better representation of land-based rainfall. However, the nature of air-sea feedbacks during the monsoon suggests an upper limit of predictability that is much reduced compared to that during the pre-monsoon.