Lightning accompanied by inconsequential rainfall is the primary natural ignition source for wildfires globally. These fires pose a significant risk to the natural environment, livestock and human life and property. There is a paucity of knowledge about the form and strength of relationships between large-scale modes of climate variability and observed lightning-ignited wildfires within the Southern Hemisphere, including Australia. Such knowledge might be useful for developing operational forecasting schemes that can inform fire-fighting planning decisions (e.g. requisition, allocation, positioning and integration of local, interstate and overseas resources, provisions, vehicles and aircraft).

Here we will report the results of an investigation of the relationships between six large-scale modes of climate variability (El Niño, Indian Ocean Dipole, Madden-Julian Oscillation, Southern Annular mode, Interdecadal Pacific Oscillation and the Quasi-Biennial Oscillation) and a 41-year record of lightning-ignited wildfires in the Warren region of southern Western Australia. The purpose of our analysis is to explore and characterise these relationships using a count regression modelling approach. Four regression techniques are considered: Poisson, negative binomial, zero-inflated Poisson and zero-inflated negative binomial. The models were evaluated using model performance criteria, probabilistic scoring rules, and the repeat holdout method to ensure that data used for model fitting was not used for model testing. The predictive skill of these models is compared to that obtained from a baseline model that captures the temporal structure of the ignition count series without any additional covariates (predictive or explanatory variables).