Large-scale ocean-atmospheric processes and seasonal rainfall variability in South Australia: accounting for non-linearity and establishing the hierarchy of influence
Tozer, CR and Kiem, AS, Large-scale ocean-atmospheric processes and seasonal rainfall variability in South Australia: accounting for non-linearity and establishing the hierarchy of influence, International Journal of Climatology, 37, (suppl 1) pp. 1180-1198. ISSN 0899-8418 (2017) [Refereed Article]
An important step in developing climate forecasts is the identification of an appropriate set of predictors. Seasonal rainfall forecasts for South Australia (SA) currently have low predictive skill. It is hypothesised that this is because the large-scale climate processes influencing SA's rainfall variability have yet to be completely identified and therefore are not adequately represented in forecast models. Here, relationships between large-scale climate influences and rainfall in SA are assessed with a focus on methods that account for non-linearities inherent in the climate system. A threshold method and also a novel method for climate predictor selection, Partial Mutual Information, are used to establish the hierarchy of importance of the key influences. From a large suite of potential predictors, variability in the Subtropical Ridge intensity, blocking and the gradient of sea surface temperatures (SSTs) in the tropical central and eastern Indian Ocean are found to be key indicators of seasonal rainfall variability in SA. Interactions between processes are shown to increase the amount of variability accounted for and therefore need to be better considered when producing seasonal forecasts. Notably, the view of which climate process(es) is(are) important changes depending on the input variables used, suggesting that it is not wise to make a priori assumptions about what is or is not important. Furthermore, even a perfect forecast of El Niño Southern Oscillation (ENSO) and Indian Ocean Dipole (IOD), the focus of forecasting efforts in Australia, will not necessarily lead to an improvement to forecast skill in SA given that these processes are not key for the region. The results of this analysis can inform the development of existing and new statistical and dynamical seasonal forecasting systems for SA, and for other regions where the impacts of climate variability are significant but not necessarily dominated by ENSO and IOD.