Modelling heatwaves: Connecting an empirical Markov process model with an autoregressive model

Abstract

The frequency, intensity and duration of heatwaves at sites in mid-latitude Australia are modelled by a Markov process model and an autoregressive model with both producing typical relative errors of ten to fifteen per cent. The Markov model requires a location-specific empirical coefficient and this coefficient is shown to have a dependency on the summer-time autocorrelation and a seasonality index. Applying the autoregressive model to simple idealised climates, a very similar dependency is found which serves to demonstrate a connection between the two models. The use of detrended temperature, rather than actual temperature, has no material effect on the results.