Development of a common definition approach for multi-event attribution of fire weather extremes

In response to the occurrence of large wildfire events across both hemispheres in recent years, the effort to understand the extent to which climate change may be altering the frequency of fire-conducive meteorological conditions has become an emerging subfield of attribution science. However, to date, the relative paucity of wildfire attribution studies, coupled with limited observational records, makes it difficult to draw solid and collective conclusions to better inform forest management strategies. The inter-study differences that emerge due to the choice of methodology and event definition are common to many attribution studies; for wildfire attribution in particular, the lack of consensus on how fire danger should be defined in a meteorological context presents an additional challenge.

Here, we present a framework for the simultaneous attribution of multiple extreme fire weather episodes of using an empirical-statistical methodology. Key to this framework is the development of a common spatiotemporal definition for extreme fire weather events. With reference to the fourth version of Global Fire Emissions Dataset (GFED4), we focus on all parts of the world that have experienced fires during the period 1995-2016. At each target grid point, we fit a Generalized Extreme Value (GEV) distribution, scaled by global mean surface temperature (smoothed over 4 years), to the annual maxima of a series of reanalysis-derived fire danger indicators (including the fire weather index) for the period 1980-2018. Using global maps of risk ratios and percentage of changes, we quantify the influence of recent global warming on the frequency and magnitude of fire weather extremes according to a common ‘event type’ definition, irrespective of their spatiotemporal occurrence. We subsequently conduct a collective attribution analysis of a series of recent exceptional events. We conclude with suggestions for further application to climate model ensembles and a discussion of the potential of our findings to inform decision-makers and practitioners.