Modelling the non-stationary, asymmetric impact of ENSO on seasonal rainfall totals and daily rainfall extremes

Abstract

The El Niño Southern Oscillation (ENSO) is known to impact on Australian rainfall, in particular, summer rainfall in Southeast Queensland (SEQ). A non-stationary statistical Bayesian framework is used to quantify this influence on rainfall totals and daily rainfall extremes. With this framework, it is possible to analyse the influence of ENSO on both the mean and the variability of rainfall. The Southern Oscillation Index (SOI), a measure of ENSO, was considered as a covariate. In order to account for different influence during La Niña and El Niño episodes, an asymmetric model was used to analyse the at-site observation data. During a La Nĩa episode, SOI had a statistically significant impact on summer rainfall totals and maximum daily rainfall for the majority of the 16 study sites. Conversely, during an El Niño episode, the SOI did not have a statistically significant impact on rainfall for the majority of sites. The value of using this asymmetric model was demonstrated because it identified a greater number of sites with a statistically significant SOI influence than the symmetric model. Furthermore, the asymmetric model demonstrated that the posterior mean of the increase in 1 in 100 year rainfall due to the strength of a La Nĩa was 4-10 mm/unit SOI, whereas the symmetric model showed a considerably lower rainfall increase of 1-5 mm/unit SOI. During a strong La Niña, the asymmetric model estimated the posterior median of the 1 in 100 year rainfall can be up to approximately 33% higher than the symmetric model and 50% higher for a stationary model. Further research on developing regional approaches will aim to reduce the large uncertainties in the estimation of impact of ENSO on extreme rainfalls.