Statistical models

Statistical models have the benefit of being accessible and quick to run, and their simplicity means that a variety of future climate scenarios from different GCMs can be generated and evaluated.

Statistical models combine historical wildfire and climate information into relationships which can then be applied to future climate information to estimate how future climate will impact wildfire activity – beyond simply stating that future climate will increase the number of days conducive to wildfire. Some studies include McKenzie et al. (2004), Balshi et al. (2009), Littell et al. (2009), Moritz et al. (2012), Yue et al. (2013), and Barbero et al. (2015). Results from those studies range from 24% to 150% increases in annual burned area, and from two- to four-fold increases in very large fires, by mid-century. Sousounis et al. (2021) analyzed fire data from the Fire Occurrence Database for the period 1992-2015 in conjunction with historical VPD data to develop region-specific burned area-VPD relationships for the Western U.S. The relationships were then applied to future GCM output of VPD to determine changes in burned area by mid-century under Representative Concentration Pathway (RCP) 8.5.¹ The study found significant increases in mean annual burned area in excess of 600% for parts of the Intermountain West and lesser, although still significant, increases for other parts of the Western U.S.

Some statistical model studies have also been conducted to estimate future loss from wildfire. Westerling and Bryant (2008) developed a statistical model for estimating loss from future wildfire activity for the state of California. Their regression model was based on 11 different climate and geographic variables including prior year snowfall, concurrent-year precipitation, elevation, etc. They used output from two different GCMs and from two climate scenarios – one representing a greenhouse gas concentration between the RCPs 6.0 and 8.5, and the other representing an RCP of 8.5. For the most extreme climate-model combination they found a 96% increase in loss for California by end of century. Sousounis et al. (2021) used estimates of future changes in burned area to estimate changes in annual average loss and other loss metrics based on existing distribution of property.