Present and Future Interannual Variability in Wildfire Occurrence: A Large Ensemble Application to the United States

Theodore R Keeping ^1,2* Boya Zhou ³ Wenjia Cai ³ Theodore G Shepherd ⁴ Iain Colin Prentice ^2,3 Karin Van Der Wiel ⁵ Sandy Harrison ^1,2

• ¹ Department of Geography and Environmental Science, School of Archaeology, Geography and Environmental Science, University of Reading, Reading, England, United Kingdom
• ² Leverhulme Centre for Wildfires, Environment and Society, London, England, United Kingdom
• ³ Georgina Mace Centre for the Living Planet, Faculty of Natural Sciences, Imperial College London, London, England, United Kingdom
• ⁴ Department of Meteorology, School of Mathematical, Physical and Computational Sciences, University of Reading, Reading, England, United Kingdom
• ⁵ Royal Netherlands Meteorological Institute, De Bilt, Netherlands

Realistic projections of future wildfires need to account for both the stochastic nature of climate and the randomness of individual fire events. Here we adopt a probabilistic approach to predict current and future fire probabilities using a large ensemble of 1600 modelled years representing different stochastic realizations of the climate during a modern reference period (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009) and a future characterised by an additional 2°C global warming. This allows us to characterise the distribution of fire years for the contiguous US, including extreme years when the number of fires or the length of the fire season exceeded those seen in the short observational record. We show that spread in the distribution of fire years in the reference period is higher in areas with a high mean number of fires, but that there is variation in this relationship with regions of proportionally higher variability in the Great Plains and southwestern US. The principal drivers of variability in simulated fire years are related either to interannual variability in fuel production or atmospheric moisture controls on fuel drying, but there are distinct geographic patterns in which each of these is the dominant control. The ensemble also shows considerable spread in fire season length, with regions such as the southwestern US being vulnerable to very long fire seasons in extreme fire years. The mean number of fires increases with an additional 2°C warming, but the spread of the distribution increases even more across three quarters of the contiguous US. Warming has a strong effect on the likelihood of less fireprone regions of the northern US to experience extreme fire years. It also has a strong amplifying effect on annual fire occurrence and fire season length in already fire-prone regions of the western US. The area in which fuel availability is the dominant control on fire occurrence increases substantially with warming. These analyses demonstrate the importance of taking account of the stochasticity of both climate and fire in characterising wildfire regimes, and the utility of large climate ensembles for making projections of the likelihood of extreme years or extreme fire seasons under future climate change.