After a policy of aggressive fire suppression in most of North America during the 20^th century, increasing aridity has driven widespread, synchronous fire occurrence in recent decades. A lack of historical (pre-1880) fire records limits our ability to understand long-term continental fire-climate dynamics. The goal of this study is to use tree-ring reconstructions to determine the relationships between spatio-temporal patterns in historical climate and widespread fire occurrence in North America, and whether they are stable through time. 

We applied regionalization methods to tree-ring reconstructions of historical summer soil moisture and annual fire occurrence to independently identify broad- and fine-scale climate and fire regions based on common inter-annual variability. We then tested whether the regions were stable through time and for spatial correspondence between the climate and fire regions. Last, we used correlation analysis to quantify the strength of the fire-climate associations through time.

We found that broad-scale historical patterns in climate and fire have strong spatial coherence. Although climate and fire regions vary over time, large core areas of the regions were stable. The association between climate and fire varied through time and was strongest in western North America, likely due to a combination of factors, such as the magnitude of drought frequency and severity, as well as varying use of fire by human communities.

The historical perspective gained through tree-ring reconstructions of climate and fire patterns and their association suggests that the recent climate-driven synchrony of fire across large areas in recent decades is not unprecedented, will likely continue into the future, and may exhibit similar spatial patterns.

Our study area includes North America between 20°N and 60°N (Figure 1). We used this latitudinal range because outside of this area there are few fire history sites with sufficient data prior to 1880 CE (Margolis et al. 2022). The fire data reconstructed from tree rings are located in forested regions; consequently, our study may not be representative of fire regimes in non-forest vegetation.

We analyzed records of fire occurrence for the period 1750-1880 from NAFSN (Margolis et al. 2022), using tree-level records of the year of fire occurrence from 1,159 sites. The start year of the analysis, 1750, was chosen to optimize the longest possible period with the broadest geographic coverage of sites that were continuously recording fire. Continuously recorded fire data are necessary to prevent biases in the cluster analysis related to decreasing sample depth associated with tree-ring sample decay (Swetnam et al. 1999). The analysis period ended in the year 1880 due to the strong influence of fire exclusion from changing land use, disrupted Indigenous burning, and fire suppression (Swetnam et al. 2016), although the timing, cause, and level of fire exclusion varies across the study area (Margolis et al. 2022).

We used the burnr package in R (v0.6.1, Malevich et al. 2018) and applied two filters to the site-level fire-scar data to include fire records that were: 1) replicated within each site (> 2 trees scarred per site) and 2) recorded by 10% or more of trees within each site. The minimum-tree filter has a larger influence on small sites with few trees, while the percent-scarred filter reduces the influence of large sites (e.g., > 20 trees; Falk et al. 2007). This combination of filters is commonly used in fire history analyses (e.g., Swetnam and Baisan 1996) to minimize the effects of varying sample depth among sites and reduce the contribution of small fires that may obscure the signal of widespread synchronous fires. To normalize for the higher density of NAFSN sites in the West, we aggregated fire data from the sites into hexels with a diameter of 100 km (hexel area 8660 km²; Figure 1) and produced a time series of percent of sites scarred in each hexel per year.

To represent historical climate (1750-1880), we used an existing tree-ring reconstruction of soil moisture as a proxy for fuel aridity (Williams et al. 2022a). Summer soil moisture (June-August; JJA) was reconstructed by relating climate-sensitive tree-ring widths to modern gridded soil moisture products using point-by-point regression, producing annual grids (°1 latitude-longitude). Anomalies were calculated as standardized (z-scored) values relative to the 1901-2000 mean at each point.