Identifying sensitivities and uncertainties in the attribution of global fire weather extremes using CMIP6 ensembles

In response to the occurrence of a number of large wildfire events across the world in recent years, the question of the extent to which climate change may be altering the meteorological conditions conducive to wildfires has become a hot topic of debate. Despite the development of attribution methodologies for extreme events in the last decade, attribution studies dedicated explicitly to wildfire, or otherwise extreme ‘fire weather’, are still relatively few. In turn, there is a lack of consensus on how to define fire risk in a meteorological context, posing a challenge for research in this subfield. Recent work has offered clarification on uncertainties associated with the choice of meteorological indicator to represent fire weather in the context of extreme event attribution but there are additional sensitivities that are still not fully understood.

Here, using established statistical methodologies applied to six large (>10-member) CMIP6 model ensembles, we conduct probabilistic attribution of fire weather extremes across the world’s fire-prone regions. We assess trends in extremes in the Canadian Fire Weather Index (FWI) using extreme value distributions, fitted with both annual maxima and peaks over a pre-defined threshold, and scaled to global mean surface temperature. An initial evaluation of model performance shows that, while all models are able to reasonably reproduce observed global patterns in extreme distribution parameters, there are some notable differences at the regional scale. Subsequently, we use probability ratio maps to quantify the influence of rising global temperatures on the changing frequency of FWI extremes. Our results highlight the sensitivity of probabilistic fire weather attribution methodologies to the choice of climate model ensemble. In conclusion, we therefore make a set of recommendations for future attribution of extreme fire weather episodes: (i) the use (and comparison) of multiple model ensembles; (ii) robust evaluation of model capacity to represent fire weather extremes.