An Australian Climatology of Severe Convective Wind Storms Using Bayesian Hierarchical Modelling

Climatological analysis is necessary when trying to develop a stochastic model that simulates an extended period of convective windstorm activity. Unfortunately, severe weather event records (including wind gusts) in the Australian Bureau of Meteorology’s Severe Storm Archive are unreliable. Moreover the wind data from automatic weather stations (AWS) are spatial incomplete and contain biases. These biases occur because of differences in station density over space and time.

Severe weather parameters (i.e. Convective Available Potential Energy and 0-6 km wind shear) are increasingly being used by climate researchers to either compliment or replace poor observational records of severe weather events and facilitate the development of convective windstorm climatology information. The parameters chosen to do this, though, are not typically tailored for use in specific climate regions or for estimating the occurrence of different storm types. A Bayesian hierarchical framework may facilitate such tailoring and this work explores the use of such a model to estimate the probability of severe wind events occurring in different parts of Australia.

The hierarchical Bayesian model used in this study couples AWS density estimates and severe weather parameters to correct for biases in the AWS convective storm day counts. Lightning data is utilised in conjunction with daily AWS wind gusts to identify convective wind days between 2005-2015. Combinations of different severe weather parameters are examined to determine which provide the greatest explanatory power. The spatio-temporal variability of the underlying factors inherent to severe weather formation are often unaccounted for in severe weather parameters. A conditional autoregressive term is incorporated into the model to consider these factors and better determine the expected number of severe wind events across different seasons and climate regions of Australia. Different model estimates of ‘true’ storm occurrence counts are compared and analysed to assemble a nation-wide severe windstorm climatology.