Human augmentation of historical red pine fire regimes in the Boundary Waters Canoe Area Wilderness
Corresponding Editor: Carrie R. Levine.
Abstract
The Border Lakes Region of Minnesota and Ontario has long been viewed as a fire-dependent ecosystem. High-severity fire in the region's near-boreal forests has been a focus of ecological research and public fascination. However, the surface fire history within this transnational wilderness landscape has received more limited attention. We used an interdisciplinary, dendroecological approach to characterize the surface fire history of the region, assess potential drivers of historical surface fires, and document the ecological legacies of frequent fires within the red pine forests of the Boundary Waters Canoe Area Wilderness (BWCAW) in northern Minnesota. We used tree-ring and fire atlas data to reconstruct multi-century surface fire records for 101 sites and document age structure and composition at 32 sites across the BWCAW. Stratification of these sites relative to their proximity to a primary travel and trade corridor used first by Indigenous groups and later by Euro-American fur traders through the late 1800s provided strong evidence of human augmentation of fires. The patterns of fire activity, fire–climate relationships, and forest development indicate that traditional landuse by Anishinaabeg (Ojibwe) increased rates of local surface fire and played an important role in shaping the landscape. The decline of traditional subsistence practices by the Border Lakes Anishinaabeg coincided with a sharp decline in surface fires and a period of abundant tree establishment. In the absence of repeat surface fires, many red pine sites have shifted in composition, increased in stem density, and grown vulnerable to forest-type conversion through future high-severity fire. These results highlight the need for active fire reintroduction to red pine forests of the Great Lakes Region and underscore the importance of collaboration and guidance from Indigenous Knowledge Keepers in this process. A blended knowledge approach to fire restoration that directly engages with Indigenous perspectives and cultural practices can perpetuate the distinctive character of the largest remaining tracts of long-lived pine forest in the Great Lakes Region. Carefully developed fire restoration practices would enhance the visitor use experience within one of the most frequently visited wilderness areas in the United States while re-engaging directly with Indigenous knowledge and traditional cultural practices.
Introduction
Fire as both a biophysical and cultural phenomenon presents difficulties in distinguishing the variable effects of lightning and human ignitions on historical fire regimes across North America (Abrams and Nowacki 2020, Oswald et al. 2020a, b, Roos 2020). Understanding the balance among human and non-human factors in past fire regimes is key to recognizing how fire and people interact as integral components of forest communities. Human-augmented fire regimes are distinct from lightning-dominated fire regimes in their frequency, spatial patterns, seasonal timing, and associated effects (Granström and Niklasson 2007). Prior to widespread fire suppression over the 20th century, Indigenous fire stewardship helped maintain and enhance pyrodiverse and heterogeneous forested landscapes, particularly in ignition-limited systems (Lewis and Ferguson 1988). By seeking nuanced understanding of all the factors that shaped historical fire regimes, we can better identify vegetation communities and landscape settings where planned fire is needed to restore the structure and function of fire-dependent forest types.
The fire ecology and fire history of red pine (Pinus resinosa Ait.) woodlands and forests have been studied for at least the past 80 yr (Kittredge 1934, Eyre and Zehngraff 1948, Spurr 1954, Buckman 1964, Van Wagner 1970, Frissell 1973, Heinselman 1973a, Hansen et al. 1974, Loope and Anderton 1998, Drobyshev et al. 2008a, Kipfmueller et al. 2017, Larson and Green 2017, Meunier et al. 2019a). While the role of fire as a key process in the life cycle of red pine has long been evident, the importance of historical fires in shaping red pine forest types remains drastically underappreciated in the Great Lakes Region. Heavy logging followed by high-severity fires and subsequent fire exclusion in red pine forest types for over a century has resulted in landscape-scale patterns of forest composition and structure that are uncharacteristic of historical red pine woodlands and forests present at the time of Euro-American colonization (Schulte et al. 2007). Today, red pine forest types in the region predominantly exist in dense, artificially regenerated single-aged stands and less commonly in passively managed fire-origin stands with uncharacteristically high stem densities of fire-intolerant species and heavy understory fuel loads (Palik and D'Amato 2019). Remnant pine woodlands and forests that established prior to widespread Euro-American colonization demonstrate a range of historical stand structures and species compositions (Fraver and Palik 2012) shaped by relatively frequent low-to-mixed severity fire (Schantz-Hansen 1931). Many of these mixed-age pine forest types lack natural regeneration of red pine today (Dietrich et al. 2017). A primary factor implicated in the diminished presence of naturally regenerated red pine forests across the region is a dramatic reduction of fire over the 20th century.
Critical to the restoration of fire to Great Lakes pine forests is the need to understand the historical drivers of fire occurrence in red pine forest types. Climate, fuel availability, fire exclusion, and fire suppression are often considered the primary controls on 20th-century fire occurrence in the Great Lakes Region (Heinselman 1996, Meunier et al. 2019b). A diminished role of traditional Indigenous landuse as a source of frequent fire can also explain changes in fire occurrence over the last century in many places (Anderton 1999, Muzika et al. 2015, Guyette et al. 2016, Kipfmueller et al. 2017) and has important implications for understanding the eco-cultural factors that led to the diminished presence of red pine on the landscape since Euro-American colonization (Buckman et al. 2006). More place-based fire history records are needed to characterize and call attention to the complexities of historical red pine fire regimes and provide resource managers actionable metrics to guide the reintroduction of fire to the region's fire-adapted landscapes.
- document evidence of a frequent fire regime in the red pine forests of the BWCAW;
- identify the relative roles of people and climate in shaping this fire regime; and
- determine the links between historical fire activity and forest development.
We discuss the implications of our results with respect to understanding and perpetuating red pine forest types in the BWCAW, specifically, and more broadly as they relate to managing fire and forests across the Great Lakes Region. Our interpretation of the patterns evident in the results was shaped through close discussions, site visits, and shared interpretation with traditional Knowledge Keepers from multiple Indigenous Communities connected to this landscape. Stories and insights shared through these discussions were key for contextualizing our data, while the physical tree-ring samples provided an important foundation on which to build ongoing relationships and conversations. We close by considering the implications of this work for active fire management in the BWCAW, a landscape defined as wilderness by federal law but that carries long-term ecological legacies created through cultural engagement by Indigenous communities with their ancestral lands.
Materials and Methods
Study area
Our work in the BWCAW was conducted on the ceded territory and the ancestral lands of many peoples including the Border Lakes Anishinaabeg (the cultural groups also referred to as Ojibwe or Chippewa, but who refer to themselves as Anishinaabe (s) or Anishinaabeg (pl)), Očeti Šakówiŋ, Métis, Cree, and Assiniboine. The BWCAW is a 328,000-ha federally protected wilderness area that contains some of the largest tracts of old-growth forest in the eastern United States and is part of a larger transnational forest ecosystem referred to as the Border Lakes Region of Minnesota and Ontario (Fig. 1). The Border Lakes Region is a near-boreal ecosystem where northern conifer forests and northern deciduous forests merge and intermingle in a landscape laced with a vast labyrinth of interconnected lakes, rivers, and streams. The climate of the BWCAW is continental, with short growing seasons and long, cold winters. Snow begins to accumulate in the study area in November and can linger well into May. Convective storms are common in the summer months that can deliver locally heavy precipitation. There is a slight moisture gradient west to east (dry to wet) across the landscape, but vegetation patterns are largely expressions of landform, soil depth to bedrock, and disturbance history (Aaseng et al. 2011).

The forests of the Border Lakes Region gradually shifted in species composition following the recession of the glaciers (Swain 1973, Gajewski 1988). Spruce (Picea spp.) forests developed between 10,700 and 10,300 cal yr BP. Pine species began to persist in the region around 9000 cal yr BP, with jack pine (Pinus banksiana) arriving somewhat earlier than red pine and white pine (Pinus strobus). Composition has been relatively stable since, with birch (Betula papyrifera), aspen (Populus spp.), ash (Fraxinus spp.), alder (Alnus spp.), and oak (Quercus spp.) present in smaller numbers in a conifer-dominated forest commonly referred to as near-boreal. Hardwoods tend to occur in uplands, while conifers often surround the shorelines of lakes (Heinselman 1996). Successional patterns in the BWCAW suggest that following stand-replacing disturbances, aspen, pine, and/or birch dominate early followed by a gradual transition to fir (Abies balsamea), spruce (Picea glauca and Picea mariana), and cedar (Thuja occidentalis), though site conditions, disturbance type (e.g., fire vs. wind), and competitive interactions influence these general patterns (Frelich and Reich 1995). Fire is the primary ecological process that shaped the forest mosaic of the Border Lakes Region over recent millennia, as evidenced by charcoal preserved in lake sediment (Ahlgren 1960, Swain 1973, Heinselman 1996) and the evolutionary adaptations to fire exhibited by dominant tree species, such as serotinous cones, thick bark, and/or sprouting tendencies (Heinselman 1973a).
Stands of red pine are found across the BWCAW but the present distribution and abundance of red pine have been shaped by a combination of edaphic factors, past disturbance regimes, and logging history (Heinselman 1996). Red pine germination and seedling establishment are highest on exposed mineral soils (McRae et al. 1994, Nyamai et al. 2014), and while these conditions are facilitated by fires, the same seedlings are relatively intolerant of large amounts of charcoal and ash (Ahlgren 1976). Therefore, conditions most conducive to red pine regeneration are found on the dry south-facing slopes of coarse glacial deposits, on bedrock exposures with thin soils, or in areas affected by frequent surface fires that reduce competition with other fire-intolerant tree and shrub species, limit fuel accumulation, and burn frequently enough to result in relatively light post-fire ash accumulations (Buckman 1964, Heinselman 1973a, Scherer et al. 2016, 2018). The pace of succession at all but the driest sites therefore can quickly lead to conditions that limit red pine regeneration and result in the decline of red pine over time (Frelich and Reich 1995). Red pine is therefore considered to be a fire-dependent species (Van Wagner 1970).
Early fire history research in the BWCAW
Groundbreaking fire history work conducted in the BWCAW by Miron ‘Bud’ Heinselman (1973a) and others (1973) fundamentally advanced the field of fire ecology by establishing the concept of a landscape-scale shifting mosaic created by tree establishment following infrequent, stand-replacing fires. Heinselman's efforts served as an important catalyst for scientists and managers working from a Western perspective to consider the important ecological role of fire in forested landscapes and began laying the groundwork for fire as an important restoration tool (Heinselman 1973b). His emphasis on landscape-scale patterns in fire activity explicitly highlighted the role of relatively infrequent, stand-initiating fire events and illustrated the fire-dependent nature of northern forests, while also foreshadowing the implications of fire exclusion for the forests of the BWCAW. Though Heinselman observed and noted the occurrence of past surface fires in the BWCAW (1973a:337), this aspect of the system was neither a focus of his work nor within the resolution of his methods to document.
Heinselman's limited quantitative treatment of surface fires in favor of an emphasis on stand-replacing fires helped establish a narrative around the role of fire in the forests of the Border Lakes Region that centered on infrequent, large, stand-reinitiating fires. This narrative dominated thinking and management actions concerning the role played by fire in the BWCAW specifically, and northern forests more generally, while shaping later efforts to model the impacts of fire on vegetation patterns and landscape configuration (Baker 1989, 1992, 1995) and shifts in fire patterns due to climate change (Scheller et al. 2005, Xu et al. 2009, Shinneman et al. 2010). Regional media coverage of recent severe fires further reinforced this perspective, while the existence of fine-scale patterns in forest composition potentially linked to past surface fires was often overlooked.
An additional consideration lacking from the narrative that emerged around Heinselman's work was the potential influence of people on historical fire activity in the BWCAW. Heinselman acknowledged the potential for humans to augment fire occurrence but considered the answer to this question unobtainable from the available records (Heinselman and Wright 1973:325). Recent research in the Border Lakes Region (Johnson and Kipfmueller 2016, Kipfmueller et al. 2017, Larson et al. 2019) and broader Great Lakes Region (Muzika et al. 2015, Guyette et al. 2016), however, documented widespread evidence of surface fires, particularly in upland–lakeshore environments, suggestive of people as an important driver of past fire activity.
Cultural history of the Border Lakes Region
Contrary to its current reputation for wilderness solitude, the Border Lakes Region was once a focal point of Indigenous and European travel and trade and carries a deep eco-cultural history that spans millennia. A vast network of traditional water routes in the Border Lakes allowed Indigenous people, namely the Border Lakes Anishinaabeg and later Euro-Americans, to access and interact with the landscape by water. Until the early 1900s, the main mode of travel in the ice-free season was by birch-bark canoe, which concentrated human landuse along these water corridors (Birk 1991). Many Indigenous water routes were adopted and used extensively by European and American explorers from at least the mid-1700s to the 1840s (Nute 1941) and are still paddled by wilderness visitors today.
Anishinaabe presence in the western Border Lakes began at least by 1736 (Richner 2002). Prior to Anishinaabe control of the region, Cree and Assiniboine and other proto-Ojibwe groups (Richner 2002) along with the Očeti Šakówiŋ (the Indigenous group also referred to as Dakota) maintained a presence in the Border Lakes (available: https://native-land.ca/). The Border Lakes Anishinaabeg ceded most of their traditional lands in treaties signed with the U.S. government in 1854 and 1866 and the Canadian government in 1873. These treaties established Indian reservations in the United States and Indian reserves in Canada, but Anishinaabe families did not strictly reside within their reserve boundaries. Instead, they moved about extensively while maintaining their traditional seasonal rounds into the late 1800s and early 1900s (Child 2011). These movements utilized seasonally available resources and included overland and water travel, hunting, trapping, fishing, harvesting, and gathering on both sides of the established Minnesota–Ontario border. Historical documentation of Indigenous fire use in the Great Lakes Region is limited (Bigsby 1850:315, Anderton 1999), but research with some contemporary Indigenous communities in northwest Ontario demonstrates the importance of planned fire as part of the traditional seasonal round (Berkes and Davidson-Hunt 2006). Eventually, forest reserve management activities in today's Quetico Provincial Park and the Superior National Forest constrained these traditional subsistence and landuse practices (Andrews 1900, Nelson 2009). Today, local Anishinaabe communities on both sides of the Minnesota–Ontario border continue to view the Border Lakes Region as traditional lands where they retain treaty rights to hunt, fish, and gather.
Field reconnaissance and mapping
Our fieldwork focused on locations within the BWCAW that supported primary-growth forest with evidence of past surface fires. We first used stand-origin and fire-scar data developed by Heinselman (1973a) to identify stands that contained evidence of two or more past fire events. We then used information from Heinselman's hand-annotated stand-origin maps and field journals and mapped historical logging extents, Superior National Forest fire atlas data, and historical and modern aerial photographs to create a map of the remaining primary-growth forest in the BWCAW over 120 yr in age (Fig. 1). Areas of primary-growth forest identified through this process were prioritized first for reconnaissance and sampling; however, areas that had been logged or burned since 1900 were visited in many cases to search for remaining evidence of past fires. Through this process, Heinselman's hand-annotated stand-origin maps were scanned and made publicly accessible through the University of Minnesota Digital Conservancy (available: http://conservancy.umn.edu/handle/11299/168077).
Of the 2401 stands delimited by Heinselman that intersected the remaining primary forest in the BWCAW, 174 were reported as having evidence of two or more fire years. We inspected 104 of these stands for fire scarred trees and stumps indicative of past surface fires. Many of these stands contained living fire-scarred trees, but only 32 contained sampleable fire-scarred material from dead and downed or remnant red pine. Additionally, we actively searched red pine-dominated shorelines of visited lakes, walked additional old-growth pine stands away from lakeshores, and visited young forest sites with evidence of red pine remnants that potentially contained multiple fire scars. Areas identified for field reconnaissance were typically accessed by canoe but sometimes by motorboat as allowed by wilderness regulations. Site conditions that encouraged on-the-ground reconnaissance included stand structure with two or more visually apparent cohorts of likely fire-origin, living and dead trees exhibiting fire-damage, fire-prone site physiography, and the strategic geography of habitable sites as informed by the historical record and field observations.
Modern fire atlas records
We used fire atlas records maintained by the Superior National Forest to characterize the modern BWCAW fire regime with respect to the roles of people and lightning as sources of ignition. The modern fire data were used to determine the relationship between fire and instrumental climate and to augment the tree-ring fire history record over the most recent century. This information was then used to aid interpretation of our tree ring-based fire history records. The fire atlas records were shared by USDA Forest Service staff as geospatial data sets and included the date, location, cause, and area burned of fires that occurred in the BWCAW from 1923 to 2017 (USDA Forest Service 2017). Strict policies of fire suppression, known as the 10 AM Policy, were implemented by the USDA Forest Service in 1935 and were coupled with aerial detection (Trygg 1946), granting relatively high confidence in the record from that time to the present. We therefore only considered fire atlas records for the period 1935–2017 to ensure a robust representation of fire activity. The estimated ignition locations of all fires included in the fire atlas records were represented in a point feature layer. The estimated burn perimeter for most fires with an area burned of >0.4 ha was included in a polygon feature layer.
We compared annual ignition counts, by cause (lightning or human), to instrumental summer (June–August) drought as represented by seasonalized Palmer's Drought Severity Index (PDSI) data for NCDC Minnesota Climate Division 3 (NCDC 2017). The purpose of the comparison to summer PDSI over the instrumental record was to examine relationships between our tree ring-based fire history and available climate reconstructions for the region, which focused on the same summer season (Cook et al. 2004). The spatial distributions of ignition and burn perimeters by cause were mapped for qualitative interpretation and quantitatively assessed for proximity to lake shores by calculating the distance to the nearest waterbody >1 ha in surface area for each ignition point using the near tool in ESRI ArcGIS Pro v2.5 (ESRI, Redlands, California, USA) relative to lake or pond features in the Minnesota Department of Natural Resources hydrography data set (Minnesota Department of Natural Resources 2020). The distances to lakeshore were compared between causes using a non-parametric two-sample Kolmogorov-Smirnov (K-S) test in the base package of R (R Development Core Team 2019).
Forest age structure and composition
Tree ring-based fire history
Federal regulations required careful selection of fire-scarred materials to sample for this reconstruction to limit impacts on wilderness character and required that fire-scarred specimens only be collected from dead and down trees (i.e., stumps and logs). We therefore used a targeted sampling approach as much by necessity as for reasons of practical efficiency. Full or partial cross sections were cut from the base of dead and downed fire-scarred red pine and remnant stumps with a two-person crosscut or arborist saw. Samples were photo-documented, annotated to aid reassembly and interpretation in the laboratory, and protected with stretch wrap for transportation out of the wilderness. In stands with abundant fire-scarred material, remnant pine that exhibited the highest number of well-preserved basal scars were sampled. Experimental work by others indicated that this targeted approach to sampling fire scars yields a robust minimum estimate of fire activity in a stand (Van Horne and Fulé 2006, Farris et al. 2013). In some cases, culturally modified trees (CMTs) exhibiting healing scars related to bark peeling were co-located with fire-scarred cross sections (Johnson et al. 2018, Larson et al. 2019, 2021) These were differentiated from fire-scar injuries by characteristic tool marks (e.g., axe cuts, patterned scrapings), location on the tree bole, and overall shape. We collected samples from these CMTs where wilderness regulations allowed and, with permission from Anishinaabe Knowledge Keepers, appointed tribal community representatives, and U.S. Forest Service Cultural Resource specialists.
Tree-ring sample processing
The tree-ring samples collected for this work were prepared following procedures described by Stokes and Smiley (1968). Samples were air-dried. We glued fragile or fragmented cross sections to plywood and all increment cores to wooden core mounts. Each sample was sanded using progressively finer grits of abrasive paper until individual tracheid cells, resin ducts, and annual-ring boundaries were easily distinguishable under 4×–40× magnification. Well-prepared, highly polished surfaces were essential in the identification of false, micro, and locally absent growth rings and to determine the placement of fire scars within the associated annual ring. Most samples were visually cross-dated using a list of marker years established from cores taken from living trees for this project and previous research in the study area (Kipfmueller et al. 2010, Johnson and Kipfmueller 2016). A limited number of increment core samples were collected from species that did not have a regional chronology available, such as paper birch (B. papyrifera), and were ring-counted. Red pine samples that could not be cross-dated with confidence using visual dating techniques were measured to the nearest 0.01 mm using a Velmex measuring system (Velmex, Bloomfield, New York, USA). Measured ring-width series were statistically compared to ring widths with an existing regional master chronology of red pine growth (Kipfmueller et al. 2010) using the program COFECHA (Holmes 1983). Potential dates identified by the statistical approach were visually verified in all cases. Widely occurring false rings were particularly useful in this process. Samples for age structure were included for analysis if the cores intersected pith or exhibited sufficient inner-ring curvature to geometrically estimate rings to pith (Applequist 1958). All other age structure samples were excluded from our analyses.
All fire scars contained in our samples were examined under 40× magnification to determine annual placement and, where possible, intra-ring position to infer fire seasonality. Intra-annual fire-scar positions included dormant, earlywood, latewood, or indeterminate placements. Dormant season scars were assigned to the calendar year following the scar. This inference was made due to a greater frequency of spring fires in northern Minnesota over the modern era and informed by separate analysis of red pine radial growth phenology indicating that red pine ceased growth in early October and that onset of radial growth near the base of most trees was delayed until early June (K. F. Kipfmueller, unpublished data). All tree-ring samples and fire scars included in this project were independently inspected and dated by two or more dendrochronologists. Seventy-one samples collected and dated in central Lac La Croix as part of a separate study (Johnson and Kipfmueller 2016) were combined with this data set to augment our sample size. Those samples were collected and analyzed using a similar approach to that employed here with the exception that fire seasonality was not recorded.
Fire-scar sample aggregation and site identification
The fragmented terrestrial landscape of the BWCAW makes necessary the identification of relatively discrete sites for the purposes of data analysis. Site fire composites were constructed from the aggregated samples to aid in capturing a more complete inventory of fires at the site scale (Dieterich and Swetnam 1984). This compositing procedure was employed because individual trees can be imperfect recorders of fire history at any given site for a variety of reasons (Dieterich 1980, Dieterich and Swetnam 1984, Baker and Ehle 2001). By compositing records from multiple trees in a particular area that likely experienced a common disturbance history, a more accurate estimation of the population of fires can be obtained.
Composite fire chronologies were created by aggregating nearby samples using a set of quantitative rules coupled with consideration of the landscape context in which individual samples were collected. First, we considered sites on islands less than approximately 2 km2 to be a single site to reflect the decreased likelihood of fire spreading over water from other areas. Although spotting of fires is possible, we felt this was a reasonably objective means of identifying site independence. Second, where forests were more contiguous or on larger islands, we used Euclidean distances of UTM coordinates and complete linkage clustering to group samples based on the distance between them (Legendre and Legendre 1998). A 1-km threshold was utilized in the clustering algorithm to differentiate between relatively discrete groups of sites on larger islands or mainland landscapes. The results of the clustering were inspected and subsequently subdivided where physiographic geographic barriers to fire spread existed within an identified cluster such as open water, lowlands, or wetlands. For example, samples separated by narrow water channels were considered separate sites despite being within a 1-km distance.
We included dates from the Superior National Forest fire atlas records into our site-scale composite fire chronologies to ensure a robust representation of fire activity over the 20th century, a period of lower sample depth in our tree-ring data set because we did not sample any living fire-scarred trees. Fire dates to be included were identified by first calculating the geographic centroid of a minimum bounding polygon that contained all individual fire-scar samples for each site in ArcGIS Pro v2.5. We then conducted a spatial join operation to identify all ignitions and burn perimeters recorded in the Superior National Forest fire atlas records that occurred within 100 m of each centroid. All fire dates within this distance of a site centroid were included in the composite fire history of that site.
Quantifying the human dimension of BWCAW fire regimes
Three conditions should be considered to identify the influence of humans as unique from other important driving factors of historical fire regimes: (1) temporal or spatial differences in fire activity that (2) are unaligned with changes in climate or fuels that could create such patterns but that (3) coincide with documented events and changes in human history (Bowman et al. 2011). Fire history researchers using a variety of methods (e.g., dendrochronology, palynology, historical records) have attempted to meet these conditions through a variety of qualitative and quantitative means that (ideally) reflected the specific ecological, cultural, and historical conditions of the site in which they worked (Savage and Swetnam 1990; Delcourt and Delcourt 1997; Kitzberger and Veblen 1997; Liebmann et al. 2016; McWethy et al. 2016; Taylor et al. 2016; Roos et al. 2018). Our approach categorized each fire history site with respect to the estimated potential of human impacts based on the physical landscape and its relationship to archeological sites, historical accounts, and oral records. We then used our categorization to examine the temporal and spatial patterns of fire occurrence and fire–climate relationships to identify the human influence on past fire activity.
Assigning cultural context to fire history data
We collaborated with Lee Johnson, Heritage Program Manager and Archaeologist for the Superior National Forest, to conduct a post hoc categorization of each fire history site informed by Johnson's expert knowledge and insights derived from a network of over 4000 archeological cultural sites across the BWCAW and Superior National Forest. In our study area, historical human activity centered on the Border Route, a primary water travel corridor that served as the focus of canoe travel and commerce during the fur trade era, beginning with the establishment of Grand Portage in the 1780s and extending through the 1860s (Morse 1968). The Border Route follows the northern boundary of our study area and extends along the present-day Minnesota–Ontario international border from Lake Superior and the Pigeon River west to Lac La Croix, Rainy Lake, and Lake of the Woods (Fig. 1). The Border Lakes Anishinaabeg used this corridor extensively from at least the 1730s through the early 1900s as part of their annual patterns of subsistence and for travel among dispersed tribal communities.
The classification of our sites with respect to this route resulted in individual sites separated into one of two categories: on-route for those sites on water bodies directly along the Border Route and off-route for all others (Fig. 1). The two site classifications were made entirely based on geographic location and without regard to the fire data of the sites. We recognize that this coarse binary classification oversimplifies the complexity of relationships between people and landscape. Other areas of the BWCAW surely experienced use either as seasonal encampments or as travel corridors from more interior sites to the Border Route for trading opportunities and myriad other socioeconomic activities. However, our categorization broadly represents an important pattern of differential intensity in historical human presence and impacts for our study area and therefore provides a reasonable initial approximation of human influence on the fire history of the BWCAW.
Testing for temporal and spatial differences in fire activity
In our study area, quantitative identification of temporal and spatial differences in fire activity that could be related to human influences was complicated by the fundamental characteristics of our data set. Reduced sample depth in the earliest part of our tree-ring record, a precipitous decline in fire occurrence over the 1900s (similar to what occurred in almost all fire-dependent systems across the coterminous United States and southern Canada), and an uneven spatial distribution of fire history sites made most of the commonly used metrics of fire frequency, such as mean fire return interval and Weibull median interval, inappropriate for examining change over time in our fire history data. There were simply too few fire intervals over some periods to justify comparison of fire frequency measures, particularly during the most recent century. Instead, we calculated cumulative distributions of the number of sites recording a fire among on- and off-route sites to illustrate temporal and spatial differences in fire activity. We graphed the cumulative distributions as the absolute number of sites recording a fire in each year to compare overall differences in fire activity. In addition, we graphed proportions to more closely compare the shapes of each curve while accounting for differences in sample depth between on- and off-route sites. We examined each curve for changes in slope that would indicate changes in the rate of burning over time (Johnson and Gutsell 1994). We tested for differences between the on- and off-route cumulative distributions using a non-parametric two-sample K-S test. Site-scale fire interval distributions were created using 5-yr bins, and the current fire-free interval was calculated as the time since the last fire recorded at each site in either the tree ring or fire atlas data.
The analyses described above were only arrived at after extensive consideration of the data set and the challenges posed by an unequal distribution of fire scar samples and sites between the on- and off-route categories, as detailed below in the Results and Discussion. We were concerned that differences in the number of samples and sites between the on- and off-route categories could produce differences in fire activity simply as a function of sample depth, rather than because real differences in historical fire activity existed across our study area. This idea is similar to the concept of the collector's curve in tree ring-based fire history reconstructions, where including more sites in one category would capture more fire years, independent of any effect of proximity to the Border Route and associated influence of human-caused ignitions (Falk et al. 2007). We therefore developed a Monte Carlo simulation that used a randomization process to determine whether the differences we observed in fire activity between on- and off-route categories would be expected by chance alone, given different sample sizes.
The model combined the entire pool of fire history sites and randomly sorted them into two groups equal in number to the observed on-route and off-route categories without considering initial group membership. The number of unique fire years across all of the sites included in each randomly created group was calculated, and the difference in the number of fire events between groups was determined. This process was repeated 1000 times and compared to the values and differences observed in the actual fire scar data for on- and off-route sites. Interpretation of the simulation results was guided by the following logic. If the differences in the number of fires observed between the on- and off-route sites were primarily the result of disparate sample sizes, then large observed differences in the number of fires should be consistently reproduced by the model. If, however, the differences in fire activity observed in our data set were expressions of actual differences in fire activity, then the randomization of sites among categories should diminish these differences and result in smaller differences in the number of fires than those observed between on- and off-route sites. The latter case would be expressed by the model as lower numbers of unique fire events for on-route sites, higher numbers of unique fire events for off-route sites, and lower differences between the groups.
Assessing the influence of climate on fire activity
We employed differences in scale and our site classification to determine whether variations in fire–climate relationships existed among our sites that could indicate the influence of people on the historical fire regime. At inter-annual scales, if human agency played a significant role in driving fire events, it would be expected that sites with greater hypothesized human influence might have a weaker fire–climate relationship (Muzika et al. 2015). Among our sites in the BWCAW, this would translate to stronger fire–climate relationships at off-route sites where human activities were hypothesized to be less intense, and weaker fire–climate relationships at on-route sites where human actions would carry greater influence, perhaps masking or limiting climate as a driver of fire activity.
Based on previous research in the region, drought is a key driver of fire activity in red pine forests (Drobyshev et al. 2008a, Johnson and Kipfmueller 2016, Kipfmueller et al. 2017). We therefore focused on the relationship between fire and drought, with drought represented by a reconstruction of summer (June–August) Palmer's Drought Severity Index for Grid Point 197 of the North American Drought Atlas (Cook et al. 2004). We broadly examined the relationship between fire and climate by graphing the total number of sites burned per year relative to a time series of PDSI and by creating a histogram of the number of fire events per 0.1 PDSI interval over the extent of our data. We established a more nuanced perspective on the inter-annual relationship between fire events and drought using superposed epoch analysis (SEA). Superposed epoch analysis is a compositing technique widely applied in fire–climate analyses that computes the mean departure of a climate parameter of interest for each year in a window around fire events (Baisan and Swetnam 1990, Brown and Schoettle 2008). In our analyses, we considered a 4-yr window of PDSI around fire events that included the fire year, two years prior to fire occurrence, and one year following fire events to consider potential lagged fire–climate relationships (Swetnam and Betancourt 1990). First, the mean PDSI for each year of the window was calculated based on all events included in the analysis. A Monte Carlo simulation was employed to generate confidence limits to which the observed mean departures in climate were compared. The simulation conducted 1000 random draws of 4-yr windows from the PDSI time series data set, calculated the mean PDSI for each year of the window based on the random draws, and subtracted the observed mean values from the simulated mean. The mean departure values from the window of interest were then compared to the 95% confidence limits to assess whether departures in the actual minus simulated values in the window of interest were significantly different from zero.
Event years used in the SEA represented different potential degrees of hypothesized human or climate influence: (1) fire years identified only at on-route sites to represent a possible human-augmented fire regime that includes both climate-driven events and a number of human-set fires, (2) fire years identified only at off-route sites to represent a primarily climate-driven fire regime where human impacts were potentially more limited, (3) fire years recorded at both on- and off-route sites to represent the overall fire regime of the BWCAW, and (4) fire years where 4+ sites, including both on- and off-route sites, recorded fire events to represent potential years of widespread climate-influenced fire activity. The SEA results were interpreted on the assumption that, in the absence of human influences, fire–climate relationships in red pine forests should be similar across the spatial extent of the BWCAW and should therefore return similar SEA results. Differences in SEA results would suggest different drivers of fire activity across what is otherwise a climatically homogeneous fire environment.
We conducted the SEA using a MATLAB script based on an early version of the analysis programmed by Holmes and Swetnam (1994) and verified our results using a double bootstrapping approach developed by Rao et al. (2019). The double bootstrapping approach used random draws with the number of key events set at half the number of event years in each of the original event year sets. This reduction in key events forced a more stringent consideration of statistical tables used to determine significance and helped to reduce bias potentially introduced by outlier values where sample sizes were small.
Linking human influences on fire activity to forest development
If humans influenced fire frequency, and fire has been an important factor in shaping red pine forest types, changes in human use of fire would lead to changes in forest composition and structure. To test this, we used both qualitative and quantitative techniques to characterize forest structure and demographics and to compare identified patterns to past fire and human activity. We first compiled fire chronologies specific to each age structure plot by combining (1) the fire dates of the nearest fire history site within 250 m of the plot origin, (2) all stand origin dates from Heinselman (1973a) within 250 m of the plot origin, and (3) all ignitions and fire perimeters in the Superior National Forest fire atlas records within 100 m of the plot origin (USDA Forest Service 2017); the more limited search distance for fire atlas events reflected the greater spatial precision of these records relative to Heinselman's reconstructed perimeters. The data were combined through spatial joins in ArcGIS Pro v2.5. Some fire years in the Heinselman data set that were within the search radius were recorded as ranges of years (e.g., 1755–1759). In these cases, we included the fire event as the first year of the listed window to account for potential lagged demographic effects following severe fires and because we considered any temporal bias introduced through this approach to be insignificant relative to the purposes of this aspect of our analyses. We then assigned each age structure plot as either on- or off-route following the same approach used for our fire history sites.
The relationship between tree establishment and fire occurrence was examined at the landscape and stand scales. Landscape-scale demographic patterns and their relationship to fire activity were visualized by calculating the total number of tree establishment dates and total number of sites recording a fire event per year among all age structure plots. We refined this approach to specifically examine patterns in red pine tree establishment over time by using vectors to represent the lifespan of all dated trees within the age structure plots, stratified to show samples from red pine in contrast to all other species, and by calculating the proportion of tree establishment dates that were red pine by decade. Stand-scale demographics were compared to fire activity by overlaying age–diameter plots with site-specific fire chronologies for each n-tree plot. Initial examination of the age–diameter plots suggested potentially weak relationships between tree age and diameter among the red pine inventoried, drawing into question the relationship between the visual hierarchy in canopy structure of many stands in which we worked and the demographic processes at play. To explore this pattern more fully, we fit linear regressions to tree age-dbh plots for (1) all red pine inventoried across our sites, (2) only red pine trees older than 150 yr old, and (3) only red pine younger than 150 yr old. Differences in the resulting regression coefficients were compared using the Chow test (Chow 1960), as implemented in the gap package (version 1.2.2) for R (Zhao 2007).
We tested for possible legacies in forest structure related to human activities by considering both the current age structure and composition as well as the successional trajectory evident at each site. Composite tree establishment records including the establishment dates for trees sampled in the n-tree plots, all pith dates included in the associated fire history site, and all fire-scar dates were created for each site and graphed to visually compare patterns between on- and off-route sites. Based on the general understanding that red pine is a long-lived, early successional species (Ahlgren 1976), we examined successional advancement by calculating (1) the time since fire for each plot based on the composite fire scar record, Heinselman data, and fire atlas records and (2) the change in IV of red pine from the canopy to subcanopy layers of each stand (Kipfmueller and Kupfer 2005). These data were graphed and compared between on- and off-route sites. To ensure that any observed differences were unrelated to differences in stand age or the number of plots considered, we screened the plots included in this comparison by fitting Weibull distributions to the tree ages of each plot using the fitdistrplus package in R (Delignette-Muller and Dutang 2015). We then compared a subset of sites for which the scale (α) was within a common interval between both on- and off-route sites. This process helped identify a similar number of sites to compare between on- and off-route with similar overall demographics. Changes in red pine IV were calculated as ΔIV = IVS − IVC, where IVS is the IV of red pine in the subcanopy, IVC is the IV of red pine in the canopy, and negative ΔIV values suggest advancing succession, or a shift in species composition from red pine dominated to more mesic or fire-intolerant species. The ΔIV was plotted against time since fire and examined visually for any apparent patterns.
Results
Fire history
Over two field seasons, we visited and explored the shorelines of 211 lakes across the BWCAW (Fig. 1). When material was available, fire-scarred samples were collected in most stands encountered with visual evidence of two or more fire events and in several instances in stands with fewer than two fires. Often, sites with evidence of only 1–2 fires were encountered but not sampled because the fire scars were only found on living trees that could not be sampled because of wilderness research restrictions. As a rule, samples were collected at all sites where there were opportunities for replication from multiple dead-downed trees with evidence of two or more fires recorded. We did not conduct a census of fire scar data across the sites visited, and several stands of red pine with evidence of a single fire event were scouted but not sampled. However, no stand we encountered that contained remnant material with evidence of multiple scars was excluded from our sampling effort.
In all, we collected one or more fire-scarred samples from 362 trees across the study area. Of these, we successfully cross-dated 356 samples from 343 trees, which, when combined with the 71 cross-dated samples reported by Johnson and Kipfmueller (2016), produced a fire history based on 414 trees that spanned 1489–2016 CE. The record included 1219 fire scars that represented 185 unique fire years, the earliest of which was recorded in 1546 and the last in 1967 (Fig. 2). Fire occurrence varied through time, with the highest occurrence of fire scars recorded between the years 1700 and 1899 when a fire was recorded in 76% of all years (Fig. 2). Legacies of large, stand-initiating fires were evident in our fire scar data set as step-like increases in sample depth that represented pulses of tree establishment following widespread fire events identified in both our fire-scar record and the stand reconstructions of Heinselman, such as the fires of 1681/82 and 1736/39 (Heinselman 1973a).

Seasonality was determined for 696 fire scars (57%). Extremely narrow growth rings, weathering, and decay created distinct challenges for determining seasonality in the case of indeterminate scars. Of the scars where seasonality could be confidently identified and replicated by multiple dendrochronologists, 33% occurred as dormant scars along ring boundaries, 26% occurred in earlywood, and 38% occurred in latewood. The relatively small proportion of fire scars with positively identified seasonality, coupled with the uncertainty of growth phenology in Border Lakes red pine, led us to exclude this aspect of our fire scar data from further analysis. We do acknowledge that fire seasonality may be a useful indicator of human influence on fire activity in some landscapes and represents a potentially valuable line of future research.
Fire atlas records for the BWCAW included 1090 ignitions and 92 fire perimeters over the time period 1935–2017. Comparison with patterns of drought during that time indicated that some large fires occurred during drought years, such as the lightning-caused 12,775-ha Cavity Lake fire of 2006 and the human-caused 30,686-ha Ham Lake fire of 2007, while many drought years did not result in widespread burning and the largest fire on record, the late summer lightning-caused 37,484-ha Pagami Creek fire of 2011, burned during a year of moderate PDSI values (Fig. 3A, B). Lightning accounted for 532 (49%) of the total ignitions (Fig. 3C) and 558 (51%) were human-caused (Fig. 3D). Compared to instrumental PDSI records, more recorded fires from both lightning- and human-caused occurred during dry summers, though this relationship was noticeably stronger for lightning-caused fires (Fig. 3E). The distribution of large fires in the BWCAW was generally too few to characterize patterns, though human-caused fires largely occurred near the wilderness boundary and along what is known as the Gunflint Trail road corridor that cuts through the northeast portion of the Wilderness (Fig. 4A). Lightning-caused ignitions were generally distributed across the study area with the exception of visual clusters in the northwest of the Wilderness and along an area of higher elevation in the southeast (Fig. 4B). In contrast, human-caused ignitions were concentrated around lakes (Fig. 4C). Quantitative comparisons of distance-to-lake measurements indicated significantly different distributions between ignition causes (D = 0.52, P < 0.0001), with human-caused ignitions occurring closer to lake shores than lightning ignitions (Fig. 4D).


Aggregation of cross-dated samples resulted in 101 discrete sites that included fire history information based on 1–24 trees (see Discussion; Appendix S1: Table S1). The period of tree-ring data recorded at individual fire history sites ranged from 99 to 505 yr and the number of fire events recorded at individual sites ranged from one fire year, which occurred at 11 sites, up to 19 fire years at one site (Fig. 5A). Comparison with Superior National Forest fire atlas records from 1935 to 2017 identified 16 ignitions that occurred within 100 m of 14 different fire history sites, none of which were represented in the fire-scar record (Fig. 5A). Additionally, during fieldwork it was evident that one additional site burned during the 2007 Ham Lake fire that was not represented in the fire atlas records. These events resulted in an additional 17 fire dates included in our analyses, 12 of which were attributed to humans and five to lightning. Based on the composite fire chronologies, most fire years occurred at only one (n = 76 fire years) or two sites (n = 48 yr). Thirty-eight fire years were recorded at four or more sites with the maximum number of sites recording fire during a single year occurring in 1739, when 23 of 47 sites recorded a fire event (Fig. 5B).

Quantification of the human influence on fire activity
We classified 69 fire history sites as on-route and 32 as off-route based on discussions with Cultural Resource Management staff with the Superior National Forest (Fig. 1). Stratification of the fire history data by on- and off-route sites illustrated striking differences in fire activity (Fig. 5A). Of the 185 unique fire years recorded among the 101 sites, 172 were recorded at on-route sites. Of these 172 fire years, 129 (75%) were recorded only at on-route sites. Only 56 fire years were recorded at off-route sites. Of these 56 fire years, 13 (23%) were unique to off-route sites. Forty-three fire years were shared by both on- and off-route sites. Cumulative fire event curves were significantly different (k = 0.517, P < 0.001, n = 475) and illustrated these differences, in terms of both the absolute number of fire events recorded at on- and off-route sites (inset of Fig. 5B), and the shape of the curves, with a nearly continuous accumulation of events among on-route sites contrasting with the step-like accumulation of fire events among off-route sites (Fig. 5B). Fire intervals recorded for both on- and off-route sites were generally short, with approximately 45% of the fire intervals at on-route sites less than 20 yr (Fig. 6A) and 50% of fire intervals at off-route sites less than 20 yr (Fig. 6B). The current fire interval at each site, relative to the most recent fire scar or fire atlas ignition date, was 10–344 yr, with a mean time since fire of 134 yr (Fig. 6C). The last fire recorded at each site was prior to 1900 at 69 of the 101 sites (70%).

The Monte Carlo simulation of fire activity strongly supported the notion that the observed differences in fire activity between on- and off-route sites were expressions of actual spatial differences in the historical occurrence of fire. Compared to the simulation results, the observed number of unique fire years recorded is much higher than expected at on-route sites and much lower than expected at off-route sites. The median number of unique fire events for the simulated on-route category was 158, less than the 172 observed in our record, while the median number of unique fire events for the simulated off-route category was 101, nearly twice as many as the 56 observed in our reconstruction (Fig. 7A). Likewise, only 5 of 1000 iterations produced more than the observed number of fires among on-route sites. The maximum number of fires simulated for the off-route category was 137, and no single iteration produced less than 69 unique fire events (Fig. 7B). The difference between the number of fire years simulated for on-route and off-route sites never, in 1000 simulations, exceeded the observed difference in fire activity between on-route and off-route sites.

The relationship between drought and fire in the BWCAW varied according to the scale of the fire event considered and by proximity to the Border Route. The median PDSI across all years during which a fire event was recorded in our study area was 0.02, with fire years spanning a broad range of PDSI values (−4.5 to +4.9) and years when more sites recorded fire generally associated with drier conditions (Fig. 8A). Considering fire activity by route, more sites recorded fire events during dry years than wet years among off-route sites (n = 34 vs. 22), while fire events were split relatively evenly between dry and wet years among on-route sites (n = 84 vs. 88; Fig. 8B).

Superposed epoch analyses identified significant but variable relationships between drought and fire among the different categories of key events considered (Fig. 9). Significant dry departures were identified during the event year for fires recorded at only off-route sites, for fires recorded at both off- and on-route sites, and for fires recorded at 4+ sites among both off- and on-route sites. Significant dry departures were identified at a −1 lag for fires recorded at both off- and on-route sites and at a +1 lag for fires recorded only at off-route sites, indicating the important influence of multi-year droughts on fire activity in the region. The SEA for fires recorded at on-route sites only contrasted with all of these results and indicated significantly wet conditions during event years.

Forest inventory and changing forest conditions
We inventoried 32 n-tree plots that contained 1093 trees and 502 seedlings and saplings (Fig. 1, Table 1). Plot area ranged from 70 to 380 m2 with a median of 158 m2. Corresponding stand densities were 886–4429 trees/ha with basal areas of 28–151 m2/ha (Table 1; Appendix S1: Fig. S1). The majority of trees inventoried in the plots were living (n = 995), with only occasional declining trees (n = 31) or standing dead snags (n = 67). A total of 79 out-of-plot trees were cored in substitution of within plot trees that yielded rotten cores. Red pine was the dominant canopy species at all but one site, with IV of 0.55–1.0, while subcanopy dominants included red pine (n = 26), white pine (n = 5), and paper birch (n = 1; Table 1). Regeneration was sparse at most sites, with 0–51 seedlings and saplings inventoried per plot. We therefore summed the tallies of seedlings and saplings to represent total regeneration by species, translating to total regeneration densities of 0–2579 stems/ha (Table 1). Though dominant in the canopy of all inventoried plots, red pine regeneration was only observed in 17 plots, with a total of only 29 red pine seedling and saplings across 1001 m2 surveyed. The most abundant tree species present in our regeneration tallies both in terms of the frequency of plots in which it was present and the abundance of stems within each plot was white pine, with 164 stems counted across 24 plots. Balsam fir (A. balsamea), paper birch, and red maple (Acer rubrum) regeneration were also relatively abundant.
Plot ID | Lake | Plot area (m2) | Density (trees/ha) | Basal area (m/ha) | Canopy† | Subcanopy‡ | All regen (n/ha) | PIRE regen (n/ha) | ||
---|---|---|---|---|---|---|---|---|---|---|
Spp. | IV | Spp. | IV | |||||||
AGN-01 | Agnes | 135 | 2222 | 99 | PIRE | 0.70 | PIRE | 0.46 | 74 | 0 |
BIG-01 | Big Moose | 210 | 1571 | 105 | PIRE | 0.94 | PIRE | 0.83 | 381 | 48 |
BIG-02 | Big Moose | 190 | 1579 | 57 | PIRE | 1.00 | PIRE | 0.75 | 2579 | 0 |
CRO-01 | Crooked | 250 | 1240 | 83 | BEPA | 0.83 | PIRE | 0.62 | 1440 | 0 |
CRO-02 | Crooked | 140 | 2214 | 96 | PIRE | 0.60 | PIST | 0.82 | 3643 | 0 |
CRO-03 | Crooked | 115 | 2870 | 83 | PIRE | 1.00 | PIRE | 0.82 | 1478 | 87 |
CUM-01 | Cummings | 255 | 1176 | 41 | PIRE | 0.82 | PIRE | 0.75 | 1451 | 39 |
CUM-02 | Cummings | 70 | 4429 | 103 | PIRE | 1.00 | PIRE | 1.00 | 286 | 286 |
FIN-01 | Finger | 200 | 1600 | 75 | PIRE | 1.00 | PIRE | 0.90 | 500 | 0 |
GUN-01 | Gun | 225 | 1289 | 66 | PIRE | 1.00 | PIRE | 0.90 | 578 | 44 |
IRN-01 | Iron | 380 | 921 | 47 | PIRE | 1.00 | PIRE | 0.72 | 211 | 0 |
LAC-01 | Lac La Croix | 200 | 1700 | 131 | PIRE | 1.00 | PIRE | 0.88 | 850 | 50 |
LAC-02 | Lac La Croix | 95 | 3368 | 106 | PIRE | 1.00 | PIRE | 0.89 | 632 | 0 |
LAC-03 | Lac La Croix | 165 | 1879 | 77 | PIRE | 1.00 | PIRE | 1.00 | 303 | 0 |
LAC-04 | Lac La Croix | 150 | 2333 | 113 | PIRE | 0.82 | PIST | 0.55 | 1533 | 267 |
LAC-05 | Lac La Croix | 100 | 3000 | 73 | PIRE | 1.00 | PIRE | 0.93 | 600 | 500 |
LAC-06 | Lac La Croix | 150 | 2133 | 91 | PIRE | 0.88 | PIST | 0.53 | 867 | 67 |
LAC-07 | Lac La Croix | 80 | 3875 | 56 | PIRE | 1.00 | PIRE | 0.77 | 625 | 0 |
LAC-08 | Lac La Croix | 260 | 1154 | 52 | PIRE | 0.96 | BEPA | 0.38 | 846 | 0 |
LAC-09 | Lac La Croix | 130 | 2615 | 99 | PIRE | 1.00 | PIRE | 0.52 | 1231 | 462 |
LAC-10 | Lac La Croix | 130 | 3077 | 110 | PIRE | 1.00 | PIRE | 1.00 | 462 | 0 |
NEE-01 | Neewin | 120 | 2750 | 67 | PIRE | 0.84 | PIRE | 0.77 | 750 | 0 |
SAG-01 | Saganaga | 200 | 1450 | 55 | PIRE | 0.93 | PIST | 0.76 | 500 | 50 |
SAG-02 | Saganaga | 175 | 1600 | 28 | PIRE | 0.83 | PIRE | 0.48 | 914 | 0 |
SAG-03 | Saganaga | 350 | 886 | 51 | PIRE | 0.71 | PIST | 0.94 | 514 | 0 |
SAG-04 | Saganaga | 150 | 2133 | 80 | PIRE | 0.85 | PIRE | 0.79 | 0 | 0 |
SAG-05 | Saganaga | 75 | 4000 | 122 | PIRE | 0.93 | PIRE | 1.00 | 1200 | 0 |
SAG-06 | Saganaga | 170 | 1765 | 37 | PIRE | 0.55 | PIRE | 0.57 | 647 | 59 |
SAG-07 | Saganaga | 145 | 2138 | 115 | PIRE | 1.00 | PIRE | 0.45 | 1724 | 69 |
SHE-01 | Shell | 140 | 2214 | 96 | PIRE | 1.00 | PIRE | 0.65 | 1000 | 143 |
STU-01 | Stuart | 165 | 1879 | 151 | PIRE | 1.00 | PIRE | 0.87 | 1939 | 61 |
ZEP-01 | Zephyr | 220 | 1409 | 74 | PIRE | 1.00 | PIRE | 0.86 | 318 | 0 |
- † Canopy dominant species as determined by importance values (IV).
- ‡ Subcanopy dominant species as determined by IV.
Inner date estimates for 899 trees in the n-tree plots were determined and included in our age structure analyses, with a mean pith correction of 7 yr (0–29 yr). At the scale of the study area, tree survivorship from the 1700s and early 1800s was relatively sparse and visually dominated by a pulse of tree establishment that peaked in the late 1800s and diminished over the 1900s (Fig. 10A). Fire occurrence in and around the age structure plots was broadly similar to the more extensive fire history record, with frequent fires among the plots from c. 1650 to the end of the 1800s (Fig. 10A). The peak in tree establishment closely followed the cessation of fire activity among the plots. Species representation varied over time, with most older trees being red pine and an increasing proportion of other species among more recently established trees (Fig. 10B).

The age structure of individual stands reflected different process-based effects of fire on red pine woodland and forest communities. Specific examples illustrate the occasional even-aged cohorts that established in gaps created after a fire (Fig. 11A), multiple cohorts indicative of repeated non-lethal burns that left many surviving trees while creating openings for red pine regeneration (Fig. 11B), and occasional tree establishment where frequent fires served a maintenance role in stand development (Fig. 11C). In a few cases, clear cohorts of fire-dependent species such as red pine and jack pine had no apparent associated fire event in our tree-ring record, suggesting that additional fires may have burned at these sites that went unrecorded in the samples we collected and that our fire history record is a minimum estimate of fire activity at these sites (see LAC-08 in Appendix S1: Fig. S1 for an example of this). Although other disturbances could result in the regeneration of red pine, such as wind (Fraver and Palik 2012), fire-initiated cohorts are likely more common based on the life-history characteristics of red pine and the documented abundance of fire prior to the 1900s.

The relationship between tree age and dbh differed between trees of different ages (Table 1; Appendix S1: Fig. S1). This pattern was particularly evident among red pine trees, where the relationship between tree age and dbh for trees younger than 150 yr (r2 = 0.05, n = 548) was significantly different than the relationship among trees older than 150 yr (r2 = 0.25, n = 133; Chow test F = 447.9, P < 0.0001; Fig. 12A). The importance of this result is directly related to interpreting canopy dominance, as it indicates the canopy position of red pine at our sites does not reflect stand age structure. Therefore, subcanopy red pine that established since the late 1800s may visually mask ongoing demographic and compositional changes.

When considered by proximity to the Border Route, stand histories showed similar patterns of abundant tree establishment and a general shift to non-red pine composition beginning in the late 1800s, but starkly different patterns in fire history and tree establishment over the early part of the record (Fig. 12B). In general, maximum tree ages were greater in plots on the Border Route than off the route. Fires were more abundant at on-route sites than off-route sites from the early 1700s through the 1800s, and this period also saw more frequent tree establishment at on-route sites. A distinct gap in tree establishment is evident at off-route sites during the mid-1800s that is present to a lesser degree among on-route sites. Red pine made up a greater proportion of tree establishments at off-route sites than on-route sites within the late 1800s and early 1900s cohort. This difference is reflected in the changes of IV from the canopy to subcanopy layers of sites, with three stands showing increased red pine IV in the subcanopy, three stands indicating no difference, and 26 stands highlighting declines in red pine IV between canopy levels. The median change in red pine IV from the canopy to the subcanopy was −0.14 and ranged from +0.52 to −0.78. Comparing these changes between on- and off-route sites suggested distinct differences in how stands in these groups are changing. Considering only those sites with overlapping Weibull scale measures of 73–138, indicative of similar demographic patterns, declines in red pine IV from the canopy to subcanopy were greater at on-route sites when compared to off-route sites (Fig. 13A, B). These differences in IV were unrelated to time-since-fire measures within either group and suggested that successional changes were occurring more rapidly at on-route sites than at off-route sites.

Discussion
An important story materializes from our research that adds to the understanding of fire and fire-dependent forests in the Border Lakes Region and illustrates the importance of Indigenous fire stewardship in shaping red pine forests over time. We documented abundant evidence of frequent surface fires in the red pine stands of the BWCAW. The patterns of frequent fire strongly suggest people played an important role in increasing fire frequency in the past, just as they increase ignition probabilities today. Our fire history also identifies important links between human-augmented fire regimes and the development of the open stands of red pine that are emblematic of the wilderness character of the BWCAW. In the absence of human-set fires, our results collectively indicate a diminished presence of red pine in the future forests of the Border Lakes Region. The environmental narrative that emerges from this work conveys the need and provides justification for proactive prescribed fire programs that, when considered in the cultural context of the Border Lakes Region, offers an opportunity to promote conversations on social and environmental justice and a critical reexamination of Western perspectives of wilderness.
Abundant evidence of frequent surface fires
We documented abundant evidence of frequent surface fires in many of the red pine stands of the BWCAW with ongoing fire-free periods that are unique over the multi-century history of most existing stands. Our data illustrate a fire regime where occasional widespread fires set the broad context in a landscape largely dominated by aspen (Populus spp.), paper birch, and jack pine forests, but within which a complex array of smaller red pine stands, maintained by frequent fire, persisted and thereby enhanced the fine-scale heterogeneity and diversity of the fire regime and forest mosaic (Miller and Urban 1999, 2000). Short-interval surface fires, although possibly impacting only a relatively small areal portion of the study area, are important to consider in the overall functioning and composition of the ecosystem. Our results build on the foundational work of Miron Heinselman (1973a) and show that in addition to the landscape-scale fire events he documented, a network of sites supporting open, red pine forest types was maintained by fires burning as often as every 5–10 yr through much of the 1700s and 1800s. Evidence to this point is provided by the expansive swaths of similarly aged forest documented by Heinselman that often align with step-like increases in the inner dates of our fire scar and age structure samples. The widespread fire scarring that occurred during these same events implicitly indicates low-to-mixed-severity fire effects at the site level during even the most extreme fire events over the past 400 yr. The more nuanced perspective we advocate for with respect to the historical fire regime of the BWCAW contributes to a growing literature that suggests a more widespread presence of surface fires in the pre-Euro-American forests of the western Great Lakes than previously thought (Drobyshev et al. 2008b, Muzika et al. 2015, Guyette et al. 2016, Kipfmueller et al. 2017, Meunier et al. 2019b).
The historical presence of frequent surface fires clearly documented in our fire scar data stand in stark contrast to the low frequency of fire recorded at our sites over the modern fire atlas record. The sharp decline in fire occurrence that began in the 1890s is generally similar to patterns documented elsewhere in the region (Frissell 1973, Clark 1990, Drobyshev et al. 2008a, Guyette et al. 2016). This decline in frequent fire has been attributed to many factors that vary from region to region, including changing climate (Fauria and Johnson 2006), decreased fuel availability due to grazing (Savage and Swetnam 1990, Guiterman et al. 2019), landuse change, and industrialization (Bowman et al. 2011), and active fire suppression (Stephens and Ruth 2005). Heinselman (1973a) and others conducting fire history in northern Minnesota suggested that a decline in late 19th-century fire activity was the result of fire suppression and exclusion (Clark 1990). The critical assumptions underlying these studies are that lightning was the primary source of historical ignitions and that effective fire suppression and exclusion have led to the observed reduction in fire activity over the 20th century. Although fire suppression over the 20th century has clearly played an important role in the overall reduction of fire activity, this conceptualization reflects the more limited focus among Western scientists and land managers on the role Indigenous people played in historical fire regimes at the time wilderness legislation was drafted in the 1960s. Similarly, fire is primarily conceived of as a natural process and described as such throughout management guidelines (e.g., see descriptions of wildfire in the Superior National Forest Plan, USDA Forest Service 2004). However, if people were important sources of ignition in the past, disregarding the role of people in historical fire regimes will result in lightning-based ignition frequencies inconsistent with the level of fire activity that created the vegetation patterns encountered by early Euro-American explorers and settlers. Understanding the relative roles of people and lightning in the historical fire regime is therefore critical for guiding management decisions around fire restoration and the conservation of biodiversity associated with fire-dependent ecological communities.
Human ignitions are an important driver of surface fires in the BWCAW
Human presence has increased rates of ignition in the BWCAW, regardless of cultural affiliation or time period, past or present. Human-caused ignitions increased fire frequency and created spatial and temporal heterogeneity within the BWCAW fire regime, which in turn increased the heterogeneity of less frequent landscape-scale fires and their effects. In short, humans greatly increased past pyrodiversity in the Border Lakes Region and helped create greater forest diversity.
Our data show (1) an increase in fire activity in the early 1700s, (2) a sharp decline in fire occurrence that began in the 1890s and ended with the near-complete cessation of fires by the early 1900s, (3) substantial differences in fire occurrence between sites on and off of the historical Border Route travel corridor, (4) different fire–climate relationships among categories of fire events that indicated drought was an important driver of widespread fire events and at sites with less human influence while fires recorded only at sites with the highest level of human activity occurred during wet years, and (5) differences in the fire–climate relationships and location of fire events in the fire-scar record were consistent with patterns evident in fire atlas records that clearly distinguished the impact of human-set fires. Climate cannot explain these patterns, with no clear change in regional hydrologic or temperature conditions evident for these specific time periods (Gajewski 1988, St George 2002, Shuman et al. 2009). Cultural changes do help contextualize each of these changes in fire activity. We do not suggest that all of the fires recorded in the tree-ring record are of Indigenous origin, but we do argue that many likely were, and that the frequent-fire component of the BWCAW fire regime documented here is the result of both lightning and Anishinaabe landuse.
Great Lakes Anishinaabeg carry a traditional story that describes their migration from the Gulf of Saint Lawrence to the western Great Lakes Region where they found Manoomin, the food that grows on the water (Benton-Banai 1979). Ethnographic and historical records describe the lands now referred to as the BWCAW as contested between Anishinaabe and Očeti Šakówiŋ groups for most of the 1600s, with Anishinaabe control gradually established over much of northern Minnesota beginning in the mid-1700s (Zedeño et al. 2001, Warren 2009). Federal U.S. and state policies that aimed to systematically assimilate Indigenous people and their cultures began in earnest in the mid-1800s, with the most immediately relevant legislative acts for our discussion being the 1887 Dawes Act (U.S. Congress 1887) and 1889 Nelson Act (U.S. Congress 1889) that disregarded past treaties and privatized reserved Indigenous Lands through a process of allotment with surplus land sold to white colonizers and commercial interests. These two acts of legislation and their immense cultural impacts align with the onset of the sharp decline in fire activity in the BWCAW, decades before fire suppression efforts began in the area. The establishment of significant fines for setting fires in the early 1900s clearly illustrates efforts to curtail Indigenous fire stewardship and implicitly indicates that human-ignited fires were viewed as inherently problematic (Fig. 14). The establishment of the Superior Forest Reserve in 1909 and a concerted investment in total fire suppression in the 1930s were the last key events driving the decline of fire activity in our study area (Superior National Forest 1931). While fire suppression policies likely reduced the frequency of human-caused ignitions and certainly reduced the spread of many fires, the precipitous decline of fire activity in the BWCAW took place before federal fire suppression was effective in the region. Rather, our data strongly support the notion that the loss of frequent fire is the result of reduced ignitions caused by the active disruption of traditional Indigenous fire stewardship, not solely or even principally fire suppression (Lake and Christianson 2019).

Photograph by Howard Greene, included in Greene-Phillips (2017), and provided courtesy of Martha Greene Young.
We acknowledge that caveats exist with respect to our use of temporal changes in fire activity as evidence of human augmentation. We argue that our interpretations remain robust despite these considerations. First, it is important to recognize that the records of fire in tree-ring reconstructions typically diminish over time as decomposition or subsequent fires eliminate sampleable materials. This led Heinselman to label the earliest part of his record from 1542 to 1727 as the period of fading record (1973a:343). Our sample depth prior to the rise of fire activity in the early 1700s remains relatively large; however, with 10 sites (12 trees) extending to 1563, 28 sites (64 trees) present to record fires in 1650, and 47% (n = 194) of our samples reaching back to the 1600s. We believe that the tree-ring record we developed is adequate to represent actual rates of surface fire at our sites through most of the 1600s and that lower rates of fire prior to the 1730s are not an artifact of sample depth. Second, suggestions that the changes in fire occurrence were driven by fur trade activities and are the result of fires set by Voyageurs and other Euro-Americans bear some consideration and agree with our overall argument that simply having people present on a landscape increases pyrodiversity. However, to consider the fur trade to be the primary cause of increased fire activity diminishes the intentional use of fire by many Anishinaabe groups for blueberry production (Anderton 1999) and other approaches to traditional forest stewardship (Davidson-Hunt and Berkes 2003). This conclusion is strongly supported by the evidence of frequent fires continuing to burn in the BWCAW following a regional decline in the fur trade and continuing markedly until a series of disruptions to traditional Anishinaabe subsistence practices that began in the late 1800s, decades before effective fire suppression and exclusion. Furthermore, the close spatial association of frequent fire sites in the BWCAW to CMTs (Larson et al. 2019), represents the clear association of both fire scars and CMTs with areas of traditional Indigenous use that were off the Border Route (Larson et al. 2021). Conversations we have had with Indigenous Knowledge Keepers and references to Indigenous fire stewardship in the journals of early European explorers (Bigsby 1850) and settlers (Anderton 1999 and references therein) help reinforce the notion that many Anishinaabe communities had a long-standing and intimate relationship with and understanding of fire. To ignore this, truth is disrespectful of the region's Indigenous communities and dismissive of vital cultures.
The fire atlas records provide important support for our interpretations of the fire-scar data as reflecting human influence on past fire activity. First, it should be noted that humans remained an important cause of fire in the BWCAW between 1935 and 2017 despite this being the time of maximum fire suppression efforts in the region and peak influence of anti-fire educational campaigns (Doerr and Santín 2016, USDA Forest Service 2017). Second, the proximity of human-caused ignitions to lakeshores contrasts with the more dispersed lightning-caused ignitions that occurred across the landscape. This pattern of elevated lake shore ignitions reflects innate human preference for certain environments as well as the practicalities of traveling through a rugged forested landscape via navigable waterways (Birk 1991). These considerations have applied for as long as people have interacted with this landscape and have demonstrably resulted in certain shoreline sites becoming long-term foci of human impacts. It is reasonable to assume that the overall pattern of human ignitions observed in the fire atlas data is at least somewhat similar to ignition patterns included in our multi-century tree-ring record since federal wilderness regulations today require people to interact with the landscape in a similar, non-motorized manner. This assumption is strongly supported by the rich ethnographic literature documenting myriad uses of fire as a way of interacting with boreal landscapes by Anishinaabeg and other Indigenous groups (Lewis and Ferguson 1988, Miller and Davidson-Hunt 2010). Collectively, these results support the notion that while the large, stand-replacing fires documented by Heinselman were likely the primary process shaping interior forest dynamics, the concentration of human-set fires in lakeshore and island settings was an important mechanism for diversifying the BWCAW fire regime at local scales.
Finally, our fire–climate analyses and blending of fire-scar and fire atlas records enable the conceptualization of the human component of the BWCAW fire regime. First, the significant dry departures associated with fire events recorded only at off-route sites, at both on- and off-route sites during the same year, and at 4+ sites among both categories identified through SEA clearly indicate that, like with fire regimes globally, landscape fires are usually associated with droughts (Baisan and Swetnam 1990, Westerling et al. 2003, Drobyshev and Niklasson 2004, Heyerdahl et al. 2008, Aakala et al. 2017). In contrast, the significant wet departure during years when fires were only recorded at on-route sites suggests a different mechanism.
Comparing the fire–climate relationships apparent in the fire atlas records to the fire scar data help to illustrate this result and clarify the human dimension of the BWCAW fire regime. During the instrumental record, lightning ignitions occurred across a range of PDSI values but were more common during drier years, whereas human-caused ignitions occurred nearly evenly across all PDSI conditions (Fig. 15A). Comparing the area included within the distribution of each ignition source indicates that during the instrumental record, people were responsible for most of the fires recorded during years of higher PDSI values. In the fire-scar record, the distribution of fire events recorded at off-route sites included more fires during years of lower PDSI, while the distribution of fire events among on-route sites included more fires during years of higher PDSI values (Fig. 15B). Comparing the distributions across data types shows the shape of the off-route fire scar-based events distribution to be nearly identical to that of fire atlas-based lightning ignitions, while the distribution of on-route fire scar-based events is better fit when the distributions of lightning and human ignitions are taken together (Fig. 15C). This suggests that if patterns of lightning ignition were consistent through time, lightning was likely the primary source of fire events at off-route sites. The distribution of on-route fire events relative to PDSI essentially reflects the combined occurrence of lightning- and human-caused fires in the fire atlas data. Taken further, the combined density of ignitions from lightning- and human-caused fires helps explain the greater fire occurrence identified among on-route sites that was borne out by our simulation. Importantly, the on-route fire regime identified in our fire-scar data is the product of both human and lightning ignitions. Isolating the years when fires were only recorded among on-route sites focuses the analysis on the wetter years when lightning-caused fires were simply less likely to occur. The human-caused ignitions during years of higher PDSI explain the wet departure identified in the SEA and illuminate how the influence of people elevates rates of fire occurrence that would not exist in their absence.

Considered collectively, our fire-scar data and their eco-cultural context add to mounting evidence that the historical red pine fires of the Great Lakes Region are the product of both biophysical phenomena and recurring patterns of human behavior and landuse. The explicit consideration of cultural ecology is necessary when assessing and interpreting tree ring-based fire records in the Great Lakes landscape. Specifically, using evidence such as modern lighting flash densities (e.g., the U.S. National Lightning Detection Network) to suggest that historical patterns of surface fire activity were the product of lightning-caused fires alone is not always an appropriate explanation for tree ring-based records of historical surface fire (Meunier et al. 2019a:6). Patterns of lightning flash density in the Great Lakes Region do not directly translate to patterns of lightning ignition rates. For those who interpret historical fire frequencies to be absent of a cultural signal in the Great Lakes Region, the burden of proof is on them to demonstrate as such. We argue that the existence of red pine surface fire histories in the Great Lakes Region is de facto evidence of human-augmented fire regimes given both the relatively limited rates of lightning-fire occurrence across the Great Lakes landscape over the modern era (Balch et al. 2017) and the historically ubiquitous use of many fire-dependent landscapes by fire-adapted Indigenous cultures prior to Euro-American settlement (Tanner and Pinther 1987).
Human-augmented surface fire created open pine woodlands that have infilled with fire absence
Open stands of red pine developed in the presence of frequent fires that were, at least in part, the result of human activities. The absence of fires caused by the suppression of traditional landuse practices among Border Lakes Anishinaabeg led to an increase in stand densities but at a pace often undetected by casual observation. The structural changes resulting from the reduced frequency of human-set fires adversely impact the long-term conservation and perpetuation of fire-adapted pine forest types in the BWCAW. In this case, the length of ecological memory long surpasses that of human memory operating in a relatively recent settler-colonial system.
Our comparison of the successional changes occurring at on- and off-route sites also highlights important spatial differences in the conditions relevant to forest development. The dominance of red pine at off-route sites is more stable than among on-route sites and may be related to edaphic conditions at these sites that help perpetuate red pine. In contrast, the more advanced successional changes documented at on-route sites are indicative of relatively mesic site conditions driving gradual successional change in the absence of frequent fires that would reduce fire-sensitive species and favor red pine. These results support the notion that many red pine stands we sampled along the Border Route established on comparatively mesic sites, when compared to off-route sites, and that this was likely enabled by human-caused surface fires. Frequent fires appear to have amplified local edaphic factors to expand the niche of red pine in the BWCAW through a process described as xerification (Larson et al. 2021).
The physical impacts of frequent fires on the landscape and to forest composition and structure included increased pyrodiversity that likely led to altered fire behavior and effects at local scales. Evidence from both proxy-based reconstructions and modern fire management in western North America indicates that structural changes achieved through either frequent fires or a combination of mechanical thinning and prescribed burns can effectively reduce fire intensity in all but the most extreme fire conditions (Weaver 1943, Stephens et al. 2009). Historically, open-grown, fire-maintained stands of red pine appear to have had similar effects by reducing local fuel loads enough to increase post-fire survivorship following extreme fire events. This is also indicated by widespread fire scars in our record that date to severe landscape fire episodes reconstructed by Heinselman (1973a) and are evidence of abundant, dispersed post-fire refugia. The diversity represented and supported by these patches of open pine, though relatively limited in area, is directly relevant to the status of numerous species of conservation concern in the Border Lakes Region that require open pine woodland habitat (Aaseng et al. 2011). The integral role of people in recurrent surface fires therefore connects the presence of people and pine in the BWCAW with the broader community of life.
In addition to the differences between red pine stands at on- and off-route sites described above, demographic changes evident across our study area suggest the broad homogenization of forest conditions over the past century and clearly illustrate the actualization of concerns raised by Heinselman about how the lack of fire would affect BWCAW ecosystems (Heinselman 1973b). To reiterate a key result from our work here, only 29 red pine seedlings and saplings were documented across 1001 m2 of regeneration plots despite the fact that we were consistently working in red pine-dominated stands growing on landforms that by most measures should be the most amenable to red pine establishment. This discordance between canopy and seedling layers within red pine forests has been documented throughout the species range and creates a high potential for stand transitions to dominance by other species (Vickers et al. 2019). This result alone suggests that current conditions will lead to a long-term decline in the presence of red pine in the BWCAW assuming a business-as-usual scenario.
The demographic patterns documented among our age structure plots provide further evidence of synchronous and unprecedented changes in red pine stands across the BWCAW that will likely reduce the number of red pine-dominated stands in the future. The establishment of a widespread cohort of trees across the BWCAW in the last few decades of the 1800s and into the early 1900s is distinct in our data. In the case of red pine, a 50-yr gap in fire occurrence is considered a potential threshold to enable successful tree establishment and recruitment (Van Wagner 1970). This pattern is evident in our comparison of fire-free intervals and tree establishment in the n-tree plots. Clearly, the fire-free intervals that are now exceeding 100 yr at almost all of the sites we surveyed are long enough to ensure abundant tree establishment and are entirely unique in geographic scope within our data. In many fire-dependent systems, a gap in fire activity and the associated episodic tree regeneration at this scale and prior to the establishment of fire suppression practices can sometimes be linked to climate conditions that reduced the occurrence of fire long enough to enable trees to establish and grow resistant to subsequent surface fires (Brown and Wu 2005). In the BWCAW, widespread tree establishment was triggered by a gap in fire occurrence that began with the suppression of Anishinaabe landuse, rather than climate, and this fire gap continues to today at most sites. The late 1800s cohort of trees that established in the BWCAW did so shortly after the last fire to occur at most sites and can be viewed as the result of a small fire deficit (Parks et al. 2015) and an ecological legacy of a time when surface fires were more frequent (Foster et al. 1998). The infill of previously open stands and the transition in tree establishment to non-pine species over the 1900s is strikingly similar to changes observed in other fire-dependent systems, such as the ponderosa pine (Pinus ponderosa) forests of the southwestern United States (Cooper 1960), and potentially indicates novel trajectories for sites that had been dominated by fire-maintained red pine for centuries.
Indeed, it is possible that the fire regime described by Heinselman (1973a) is more broadly reflective of a fire regime with little human influence. Long intervals between successive fires would lead to increased homogenization of fuels and forest structure (Frelich and Reich 1995), conditions conducive to widespread stand-replacing fires when climate, ignitions, and short-term weather coincide (Bessie and Johnson 1995). The confluence of these patterns would likely be more rare, but their impacts could be broad and made more uniform by the changes in forest structure documented in our study. A fire regime typified by large and infrequent fires may make it challenging to perpetuate red pine across much of the landscape.
Caveats to the case of a human-augmented fire regime
Discussions about human–fire relationships often occur through broad generalities that lack detailed consideration of the complexity of human–environment interactions, often due to a relative dearth of place-based and direct observations (Baker 2002, Roos 2020). The development of indices to broadly describe the potential influence of people on fire regimes can be useful (Muzika et al. 2015, Guyette et al. 2016), but our data clearly illustrate the risk of blurring site-specific evidence through coarse-scale interpretations that may lead to questionable conclusions about the role of people in historical fire regimes (Oswald et al. 2020a). Our overall results emphasize that place matters when considering the effects of people on historical fire regimes. We will critique our own findings help to illustrate this point. The binary classification between on- and off-route sites we used as a coarse proxy for human impacts may be instructive, but it disregarded the dynamic relationship and shifting perspectives likely held by people living on a landscape that itself changes over time (Davidson-Hunt 2003). It must be noted that the nature of human modification of fire regimes is neither uniform nor static in space or time, regardless of proximity to a main travel route. More detailed consideration of the patterns in fire activity identified among our sites begins to illustrate a more nuanced set of relationships among people, fire, and pines that reflects patterns influenced by the land itself (Lewis and Ferguson 1988, Larson et al. 2021).
It is evident from a more careful analysis of our fire history data that not all on-route sites exhibit clear human influences in terms of relatively short fire intervals and reduced importance of climate. That is, just because a site is situated on the Border Route does not mean that the dominant control of the fire regime was human. Likewise, some off-route sites were likely used by Anishinaabeg family groups for a variety of purposes over the seasonal rounds that governed many landuse practices (Richner 2002). Two examples of off-route areas with likely human impacts include sites at Cummings Lake and Upper Pauness Lake, both of which recorded more frequent fire events than might be expected for being so far removed from the Border Route. However, both areas are connected by the Little Indian Sioux river that served as a known travel route between Anishinaabe communities (Boshey 1996) and have indications of Anishinaabe landuse in the form of CMTs with peel scars dating to 1788 and 1791 at Cummings Lake and throughout the 1800s at Upper Pauness Lake (Larson et al. 2021). Our coarse approach that assigned these sites as off-route in the sense of having fewer human impacts was an oversimplification. The richness of the fire-scar data set and recognition of the complex patterns of historical landuse by the Anishinaabe encourages a more refined understanding of relationships between historical Indigenous communities and the landscapes in which they lived, particularly how ecological variability over space and time would relate to shifting resource availability and the broader perspective of interactions among the beings that make up the community of life in an area (Davidson-Hunt and Berkes 2003).
Red pine forest types cannot be perpetuated without fire
The close association of red pine with evidence of surface fire in the BWCAW and elsewhere is indicative of how short fire intervals promote red pine dominance and resilience (Engstrom and Mann 1991, Meunier et al. 2019a). Similar to changes in forest structure documented across the eastern United States and Great Lakes Region, the loss of frequent surface fire has led to reduced regeneration opportunities for red pine, increased competition from fire-intolerant species, and increased stand density over the past century (Hanberry et al. 2012, 2020, Young et al. 2017). Indeed, the identification of substantial increases in red pine establishment beginning in the 1890s is reflective of the impacts of reduced fire activity for more than a century.
Red pine woodland infill and the continued absence of surface fires in these sites have implications for the long-term persistence and resilience of the BWCAW forests we currently know. Higher pine stand densities, increased fuel loads, and ladder fuels will increase the severity of the next fire to burn in these stands and potentially convert long-lived red pine woodlands to forest conditions more reflective of severe, stand-replacing fires such as aspen or jack pine. Furthermore, red pine stand density is positively linked to drought susceptibility both at the scale of individual trees (Curzon et al. 2016) and stands (Gleason et al. 2017), with denser stands less resistant and resilient to drought (Magruder et al. 2013, Bottero et al. 2017). The current status of nearly every red pine-dominated stand we surveyed included negligible red pine regeneration, higher stand density than at any point in recent centuries, a transition to non-red pine species, and fuel loads more amenable to catastrophic stand-replacing fire than maintenance surface fire, reflecting findings in other red pine-dominated landscapes (Meunier et al. 2019a). These data illustrate the mechanisms that will likely drive significant future losses of red pine from the BWCAW ecosystem (Frelich and Reich 2009) and potentially in forests across the Great Lakes Region.
Restoration of frequent surface fire is needed to maintain and perpetuate red pine forest types in the Border Lakes Region. Calls to proactively employ fire in the management of red pine have been long-standing (Van Wagner 1970, Heinselman 1973b, Methven and Murray 1974) and deserve to be considered in the evolution of wilderness management in the BWCAW. In this vein, the work of Miron Heinselman is considered a landmark advancement in the field of fire ecology (Heinselman 1973a, 1996), but there has been little action with respect to his strong advocacy of fire reintroduction to the fire-adapted forests of the BWCAW (Heinselman 1973b). A half-century after Heinselman's impassioned pleas for us to move beyond hands-off wilderness management and develop a prescribed fire program for the BWCAW (Heinselman 1965), too little has been done to restore the full range of variation in historical fires across the Wilderness. If past is prologue, our results collectively imply that humans and the fires they set played a crucial role in the development of what is now the reduced presence of open-grown red pine stands in the BWCAW. Therefore, human-set fires will be necessary to promote the future persistence of red pine at a level near that of the pre-Euro-American landscape, reflective of the forest conditions used to justify its Wilderness status.
Active management for the restoration of fire to red pine and other fire-adapted forests is needed to sustain fundamental ecological functions and enhance the adaptive capacity of the landscape (Handler et al. 2017). If we continue to delay action, we will eventually enter a situation where no action can be effectively taken to restore the unique eco-cultural landscapes of the Border Lakes Region. At present, a concerted effort is needed to strategically reduce forest density and fuel loads that contribute to the potential for future catastrophic wildfires such as the 2007 Ham Lake Fire and 2011 Pagami Creek Fire. We recognize that similarly widespread fires burned in the past and are an important aspect of the BWCAW's fire regime. However, the homogenization of forest structure and fuel loads produced by over a century of fire absence has undoubtedly increased fire severity in previously open stands of red pine during these recent events and likely reduced the patchiness of fire effects characteristic of many stand-replacing fire regimes (Turner and Romme 1994). Our own observations during fieldwork along with anecdotal evidence shared by wilderness managers support this notion.
Fire restoration in red pine forests as an eco-cultural act
The recognition that fire is an essential ecological process in almost all temperate forest ecosystems and that fire suppression policies implemented over the 1900s led to dangerous fuel build-ups and compositional shifts emerged in forest management communities in North America in the early 1900s (Weaver 1943, 1947). Translating this recognition to on-the-ground management activities has taken time, however, despite significant advances in understanding the role of fire in forest ecosystems such as those included in the 1973 special issue of Quaternary Research that contained the seminal work of Bud Heinselman (Heinselman 1973a). Since then, evidence continues to emerge across a range of forest types demonstrating not only the vital ecological role of fire in forest systems, but also the utility of prescribed fire as a key tool for addressing the impacts of a century of fire suppression, reducing fuels in wildland–urban interfaces to reduce fire-related risks, and improving forest biodiversity and resiliency to climate change.
Building on this understanding of fire as an important ecological process, our results contribute to the growing recognition that restoring fire is also a fundamentally cultural process fundamentally linked to reciprocal relationships between humans and fire-adapted biota (Kimmerer and Lake 2001, Lake et al. 2018). Inclusion of Indigenous voices and perspectives in forest stewardship can help practitioners embrace fire as a key management tool, particularly in wilderness settings, while also promoting the revitalization of language, ceremony, and the tending of traditional non-timber forest products (Lake et al. 2017). In the BWCAW specifically, an active, collaborative fire program co-created with Indigenous perspectives would foster opportunities to publicly acknowledge the deep connections between Anishinaabe people and their ancestral lands while simultaneously achieving other wilderness management goals. Specifically, BWCAW red pine woodlands developed over centuries, and potentially millennia, with direct human impacts mediated through the process of fire. Red pine-dominated landscapes are emblematic of the BWCAW experience, with canoeist campsite preferences often coinciding with red pine stands (Frissell and Duncan 1965). Modern visitor preferences for red pine woodlands continue a centuries-long pattern of landuse by Border Lakes Anishinaabeg and reflect cultural, emotional, and spiritual connections to the environment afforded by red pine forests. If active fire management within the BWCAW continues to be restricted by the language of the 1964 Wilderness Act that embodies Western concepts of a human-nature divide shaped by scientists and policymakers at the time of its inception, there are vast opportunities outside of the BWCAW for land management agencies to embrace fire as a tool for eco-cultural restoration and forest adaptation.
The BWCAW fire histories reported here re-emphasize that the presence of red pine at many sites was made possible by frequent, low-severity fires in the past that created site conditions appropriate for red pine persistence (Van Wagner 1970). Given that the patterns of human history in the Border Lakes Region are consistent with the broader Great Lakes landscape (Tanner and Pinther 1987), we suggest that human augmentation of red pine surface fire occurrence was historically common, and a key ecological factor that shaped the composition and structure of fire-dependent systems across the Great Lakes Region (Riley 2013). There is great potential for red pine fire history reconstructions to be linked to site-specific historical patterns of Indigenous landuse beyond the BWCAW (Loope and Anderton 1998, Anderton 1999) and to serve as biocultural records that can be used as catalysts for cross-cultural conversation on fire stewardship priorities across ownership types (Lake et al. 2017). In landscapes where Indigenous histories have been overlooked, or underappreciated, tree ring-based fire records offer tangible evidence of traditional Indigenous landuse, corroborate tribal histories, and give resource managers actionable data that can shape future forest stewardship to be consistent with the needs and desires of local Indigenous communities (Bussey et al. 2016).
Conclusion
The decline in red pine presence and regeneration across the Border Lakes Region poses significant questions regarding the legacy of fire disturbance in these stands and highlights an opportunity for the reintroduction of frequent surface fire to sustain this economically, ecologically, and culturally significant species. Fire is a key process in the life cycle of red pine, but the importance of fire in shaping these forests remains underappreciated in the BWCAW and across the Great Lakes Region.
We offer multiple lines of evidence to demonstrate the historical presence and influences of surface fire in the red pine stands in the BWCAW. The development of open red pine forest types characteristic of the BWCAW is the result of human-augmented surface fire occurrence coupled with more episodic, climatically driven fires. While lightning has been assumed to be the most important source of fire ignitions, we conclude that human presence increases ignition probability by default and increases fire frequency in the absence of effective fire suppression. Our results indicate that fire history at sites located along the Border Route is the product of both human and lightning ignitions with human-caused ignitions occurring during wetter years when lightning fires were less likely to occur. In comparison, lightning was the primary source of fire events at sites located off the Border Route. Further, while the decline of surface fires over the early 20th century is related to fire suppression efforts, we contend that the decline in fire activity in the BWCAW over this same period is at least partly, and perhaps largely, the result of reduced ignitions caused by the active disruption of traditional Anishinaabeg landuse practices. The legacies of human-influenced surface fires are still evident in the BWCAW, but the disruption of Indigenous fire stewardship over the early 20th century and the near-complete cessation of fires in these stands suggests a diminished presence of red pine in the future.
In the absence of frequent surface fire, the remaining red pine woodlands and forests in the Border Lakes will continue to be lost to ongoing successional change, catastrophic fire, severe windstorms, and potentially climate change-induced drought stress. The human linkage to the occurrence of historically frequent surface fire clearly illustrates the need and justification for proactive prescribed fire programs that embrace the cultural context of the Border Lakes Region. More important, our results highlight an opportunity for collaboration among natural resource specialists and Anishinaabeg communities to advance conversations around the restoration of fire to the red pine forests of the Border Lakes Region as both an ecological and cultural process.
Acknowledgments
We acknowledge that this research was carried out on the ceded territory of the Anishinaabeg and that these are ancestral lands to many peoples including the Anishinaabeg, Očeti Šakówiŋ, Métis, Cree, and Assiniboine. Our efforts to conduct this research benefited in many ways from settler colonialism, and we hope that our work can contribute to the truthful recognition of this history. We also acknowledge the formative influence of Bud Heinselman's legacy on our perspectives and appreciation of fire as an integral part of northern forests. The ideas presented here were strongly shaped by conversations with our Coalition for Archaeological Synthesis collaborators Jessica Atatise (Lac La Croix First Nation), Brian Jackson (Quetico Provincial Park), Lee Johnson (Superior National Forest), Robin Kimmerer (Citizen Nation Potawatomi, SUNY College of Environmental Science and Forestry), Melonee Montano (Red Cliff Band of Ojibwe, Great Lakes Indian Fish and Wildlife Commission), Damon Panek (White Earth Band of Ojibwe, Apostle Islands National Lakeshore), and Jeff Savage (Fond du Lac Band of Lake Superior Chippewa). We deeply appreciate Bill Latady's role in facilitating meetings with members of the Bois Forte Band of Chippewa and Trevor Gibb of Quetico Provincial Park for helping to connect with members of the Lac La Croix First Nation. We gratefully acknowledge the generosity in time, guidance, and wisdom of many members of these communities, as well as their patience with us as we work to understand the connections of their knowledge and our work. Thanks go to the many hands who helped collect and transport tree-ring samples out of the wilderness, including Ben Matthys, Tom Wilding, Kalina Hildebrandt, Liam Martin, Elizabeth Tanner, Nick Harnish, John Eads, Danica Larson, Bryn Larson, Mara Larson, Shelley Larson, and Eric Larson. The planning and implementation of the project benefited from the expertise of Ann Schwaller, Superior National Forest Wilderness Specialist, and current and past wilderness research coordinators for the Superior National Forest Katie Frerker and Pooja Kanwar. Lee Frelich and John Almendinger provided access to Bud Heinselman's stand-origin maps and field books. Douglas Smith (U.S. NPS) provided helpful comments on an earlier version of this manuscript to improve its utility for managers interested in fire ecology in red pine forest types. Funding for this study was provided by National Science Foundation grants 1359868, 1359863, and 1560919, and the Coalition for Archaeological Synthesis.