Seven hundred years of human-driven and climate-influenced fire activity in a British Columbia coastal temperate rainforest

While wildland fire is globally most common at the savannah-grassland ecotone, there is little evidence of fire in coastal temperate rainforests. We reconstructed fire activity with a ca 700-year fire history derived from fire scars and stand establishment from 30 sites in a very wet (up to 4000 mm annual precipitation) temperate rainforest in coastal British Columbia, Canada. Drought and warmer temperatures in the year prior were positively associated with fire events though there was little coherence of climate indices on the years of fires. At the decadal scale, fires were more likely to occur after positive El Niño-Southern Oscillation and Pacific Decadal Oscillation phases and exhibited 30-year periods of synchrony with the negative phase of the Arctic Oscillation. Fire frequency was significantly inversely correlated with the distance from former Indigenous habitation sites and fires ceased following cultural disorganization caused by disease and other European impacts in the late nineteenth century. Indigenous people were likely to have been the primary ignition source in this and many coastal temperate rainforest settings. These data are directly relevant to contemporary forest management and discredit the myth of coastal temperate rainforests as pristine landscapes.


Introduction
Palaeofire studies have found evidence that both humans and climate influence fire regimes, but it is often difficult to discern their relative importance, and the drivers of fire and its ecosystem effects remain poorly understood [1,2]. Humans affect fire regimes by controlling ignition and suppression, and through altering vegetation and the availability of fuel types by practices such as land clearing and habitat modification [2][3][4][5]. Climate drives fire activity through interactions with local-scale, bottom-up controls such as available fuels and topography to affect fuel drying and fire spread and by influencing rates of natural ignitions [1,6,7]. In the Pacific Northwest (PNW) of North America, it is often difficult to attribute historic drivers of fire activity and causes of fire ignitions due to both the abundance of natural ignitions from lightning and the impact of logging, grazing and fire suppression that decreased fire occurrence soon after the loss of Indigenous fire management [1][2][3].
Indigenous peoples used fire for diverse purposes such as clearing land, creating fodder for game, and increasing the productivity of specific plants for food and medicine [8,9]. The cultural importance of fire has been mostly examined in the savannah-grassland ecotone [2,3,8] and in fire-adapted forests dominated by ponderosa pine (Pinus ponderosa) and Garry oak (Quercus garryana) [10][11][12]. There has been very little documented evidence of historic fire activity in coastal temperate rainforests located in the PNW and previous disturbance studies have asserted that both human-and lightning-caused fires are infrequent and become progressively rarer in high-latitude regions (more than 50°N) [13][14][15][16]. Low fire activity is assumed due to the presence of old growth forests [14], the lack of fire adaptations in dominant conifer species [13], the decreased flammability of large-diameter, moisture-laden surface fuels [1,3] and the absence of twentieth-century fire activity [17]. There have been no reconstructions of historic fire activity with fire-scarred trees in these wet and dense forests, and fires are hypothesized to be large, stand-replacing events occurring during unusually dry years when low fuel moisture promotes rapid fire spread [2,13]. A handful of studies in coastal temperate rainforests of British Columbia have described temporally coarse reconstructions of Holocene fire activity with charcoal stratigraphy in lake sediments [18] and radiocarbon dating of soil charcoal [19][20][21]. These studies depict patchy fires during the last few centuries embedded within landscapes experiencing long (sometimes more than 1000 year) fire intervals as well as widespread increases in fire activity in the Mid-and Late Holocene [19][20][21] that may be related to increases in the use of fire by Indigenous peoples [21].
Wildfire occurrence is also influenced by variability in modes of sea surface temperatures (SSTs) such as the El Niño-Southern Oscillation (ENSO) and the Pacific Decadal Oscillation (PDO) that influence winter temperature and precipitation patterns and indirectly affect summer moisture availability [22][23][24]. The ENSO is characterized by oscillating phases of warmer (El Niño) and cooler (La Niña) than average winter SSTs in the equatorial Pacific at interannual (2-7 year) frequencies [25]. The ENSO affects climate and fire in distinct ways depending on ocean-atmosphere conditions and their associated circulation patterns [24]. The warm (El Niño) phase produces warmer and drier winter and spring conditions in the PNW, which enhances wildfire activity in summer months [23,24]. The PDO is characterized by variations in SSTs in the North Pacific and positive (negative) PDO values produce relatively similar climate and circulation patterns to El Niño (La Niña) conditions though the periodicity is decadal (approx. 20 years) [26].
The Arctic Oscillation (AO) is the dominant mode of sea-level pressure (SLP) variation north of 20°N and is similar to ENSO in its interannual periodicity [27]. Variability in SLP has been shown to indirectly affect fire activity in high-latitude regions by affecting the position of the jet stream [28]. It is characterized in its positive (negative) phase by negative (positive) pressure anomalies that form over the Arctic and a circumpolar belt of positive (negative) temperature anomalies at mid-latitudes, with the exception of low-temperature anomalies that form over the PNW [27][28][29]. In Western Canada and Southeast Alaska, large fire incidents have been linked to the negative phase of the AO when high-pressure anomalies centred over the Arctic extend south to 30-50°N and result in warmer and drier conditions that are more conducive to fire initiation and spread [29]. Only a few studies have assessed fire-climate relationships between the ENSO, PDO and AO in British Columbia. These studies have omitted coastal temperate rainforests from their analyses, suggesting that these forests are too wet to show summer moisture deficits and are therefore unable to support wildfires [17,30].
In this study, we overcome the limitations of past studies by reconstructing the fire history of a coastal island with extremely low rates of lightning and a short summer dry season. Indigenous peoples abandoned the island more than 100 years ago, and it never had Euro-American settlement, logging, grazing or fire suppression [31,32]. We hypothesize that if Indigenous people frequently used fire, the fire history record would be temporally and spatially associated with former habitation sites at decadal intervals and absent in the twentieth century. If Indigenous people rarely used fire, fire would be consistently rare, occurring at centennial and millennial intervals and mostly explained by topographic and climatic controls. We evaluate the relationship between climate variability and humans on fire occurrence over ca 700 years in a high-latitude coastal temperate rainforest. We use a network of fire scars and stand establishment to address the following questions: (i) What is the relationship between the temporal occurrence of fire events with drought (The Palmer Drought Severity Index (PDSI)) and with single and interacting phases of climate drivers such as the ENSO, PDO and AO? (ii) Is fire occurrence best explained by human presence, regional climate variability, topography or more likely, interactions among these drivers?

Study area
The study area is located on Hecate Island (latitude 51°39 N, longitude 128°04 W) within the Hakai Lúxvbálís Conservancy on the central coast of British Columbia, Canada (figure 1). This region is characterized by a temperate climate with year-round cool temperatures (average annual approx. 7°C, average summer approx. 12°C) and annual rainfall sometimes exceeding 4000 mm [33]. Hecate Island (67 km 2 ) is located in the very wet hypermaritime subzone of the Coastal Western Hemlock biogeoclimatic classification [34]. Excess soil water regulates this environment and subtle variations in slope or drainage result in significant differences in forest productivity [33]. Four general vegetation types defined by dominant species and closely associated landforms are found in the study area [33]. Productive (zonal) forests found in near shore and riparian areas are characterized by large-diameter western red-cedar (Thuja plicata Donn ex D. Don.) and western hemlock (Tsuga heterophylla (Raf.) Sarg.), with lesser amounts of yellow-cedar (Cupressus nootkatensis (D. Don) Farjon & Harder) and Sitka spruce (Picea sitchensis (Bong.) Carr.) [34]. Bog forests exhibit stunted growth forms and are located on hill slopes dominated by western red-cedar, yellow-cedar, western hemlock and shore pine (Pinus contorta var. contorta Douglas ex Louden) [35]. Bog woodlands are the most common vegetation type and comprise patchy mosaics of forested and unforested sites in subdued or rolling terrain [33]. These forests contain roughly equal densities of western red-cedar, yellow-cedar and shore pine with lesser amounts of mountain hemlock (Tsuga mertensiana (Bong.)) [35]. Blanket bogs are nutrient-poor, sparsely forested wetland areas that contain small amounts of shore pine and yellow-cedar [33]. Zonal and bog forest vegetation types have closed canopies with larger diameter and more vertical fuel structures that are blanketed by thick moss compared with bog woodland and blanket bog vegetation types, which are more open, drier and contain finer, more flammable fuel assemblages [14,21]. Elevations in the study area range from sea level to 150 m, and the geological substrate is homogeneous quartz diorite and granodiorite [36].
Although there are no oral histories of fire use by Indigenous people on Hecate Island, Indigenous peoples continuously inhabited the study area for at least 13 000 years until European contact in the late eighteenth and nineteenth centuries [6,7]. A team of archaeologists used radiocarbon dating of shell, faunal and charcoal deposits to determine the locations of former habitation sites [31,32]. This evidence together with the presence of culturally modified trees and near shore features, including shell middens, clam gardens and fish traps confirm that Hecate Island supported three year-round habitation sites that were consistently used throughout the Holocene [31,32] and established millennia prior to the rise of contemporary vegetation in the region [37]. Although Indigenous groups continue to use the study area for seasonal resource gathering, habitation sites have not been occupied since the late nineteenth century coinciding with smallpox epidemics and the relocation of people to government-imposed reservations [37][38][39].

Fire history sampling design
To reconstruct a ca 700-year fire history, we sampled fire scar and stand establishment dates from 30 plots (11.28 m radius (0.04 ha)) selected using a stratified random sampling design representing the range of elevations, aspects and vegetation types within a 300 ha study area spanning three watersheds on Hecate Island (figure 1). To reconstruct low and mixed severity fires, we used a chainsaw to remove partial wedge sections of fire scars from the bases of the oldest sound living trees (determined after coring and measuring) [40]. Fire-scarred trees were sampled in the 1 ha area surrounding every plot. We validated the year of each fire event recorded on fire scar wedges by sampling two 5 mm increment cores roughly 1.3 m from the ground from approximately 300 trees which were collected outside the fire area to ensure cross-dating accuracy in the fire scar chronology [41]. In every plot, we recorded diameter at breast height (DBH) and removed two 5 mm increment cores from the root collar of every living tree more than 7.5 cm DBH to define the ages of post-fire cohorts within each plot and to estimate the year of the fire in which the cohort established [41,42]. Because fire scars provide more accurate records of fire occurrence, we assigned post-fire cohorts to the same year as nearby fire events if they occurred within the 10-year period of a nearby fire scar [42].
In the laboratory, cores and wedges were processed using standard dendrochronological techniques [43]. Samples were first visually cross-dated and then statistically verified using the computer program COFECHA [44]. We cross-dated fire scars to species-specific chronologies to obtain exact fire years (the calendar year in which the scar formed) [7]. We were unable to determine the intra-ring position (seasonality) of the majority of samples due to rot and narrowing of tree rings at fire scar margins [45]. We determined the fire history in each plot by compiling fire scars into a composite chronology, and identified fire events as those in which at least two trees had fire scars [41]. Composite fire histories were graphed using Fire History Analysis and Exploration Software [46].

Spatial controls of the fire regime
We used a generalized linear model (GLM) with a Poisson distribution and model selection via the Akaike information criterion (AIC c ) to predict two response variables, (i) the abundance of fire scars and (ii) the frequency of fire events, using the explanatory variables (vegetation type, slope, aspect, elevation, distance to shoreline and distance to former habitation site) in the 'MuMIn' package in R statistical software [47,48] following the methods of Harrell [49] and Burnham & Anderson [50]. Because there were no recorded fire events in the study area after 1893, we only calculated the mean fire interval (MFI) from the first detected fire in 1376 to the last detected fire in 1893 (517 years), and we assessed the composite MFI by combining all fire scar data from the study area over this period of analysis [41]. We were able to estimate fire effects on vegetation (severity) using the presence of fire-scarred trees, the MFI and whether cohorts had established after fire events (indicative of mixed severity fires) [45].

Climate datasets
The PDSI is a composite monthly index of regional climate calculated from current and archived instrumental data of precipitation and temperature [51]. Negative PDSI values indicate dry years and positive values indicate wet years (range −6 to +6). To characterize historic patterns of drought in the study area, we used an extended reconstruction of the PDSI from the gridded North American Drought Atlas (grid point 17 (127.5°W × 52.5°N)), which combines 28 tree-ring reconstructions of summer (June-August) temperature and precipitation [52]. In our analyses, we used previous reconstructions of climate variability (PDSI, ENSO, PDO and AO), which correlated to instrumental records and tree-ring chronologies sampled from the study area (electronic supplementary material, table S1).

Annual fire-climate analyses
At the annual scale, frequency analysis was evaluated with a G-test of goodness of fit in the 'DescTools' package in R statistical software [48,53] to determine whether fire events were more likely to occur during extreme climate years of single climatic indexes (negative and positive phases of the PDSI, ENSO, PDO and AO). We used this test again to consider the influence of two-and three-way interactions between the ENSO, PDO and AO with fire occurrence [54]. We compared the number of fires during phase combinations assuming an equal likelihood of burning in all years. Sixty per cent of fire events occurred between 1700 and 1893, therefore, we chose to analyse three-way interacting climate indexes during this period and compared all two-way interactions using their respective temporal windows (e.g. ENSO 1300-2004 with PDSI 1240-2006 (common period of analysis 1300-1893); electronic supplementary material, figure S1).

Interannual and multidecadal fire-climate analyses
To characterize interannual and multidecadal relationships between fire occurrences and the PDSI, ENSO, PDO and AO, we used bivariate event analysis (BEA) with K1D software [55]. BEA is a temporal variant of spatial point pattern analysis, which employs Ripley's K function to examine one-dimensional time-series data [56]. This method is advantageous when testing time-series data that are subject to serial autocorrelation and BEA assesses the response of fire years (synchrony, independence or asynchrony) to climate events within a range of temporal windows. Monte Carlo simulations with 1000 replicates were used to randomize climate time series with 95% and 99% confidence envelopes [57]. BEA, as implemented here, assumes a one-directional process where fire events only respond to previous and current climate events [58].
We first tested single climatic indexes and fire years by defining extreme climate events as the 50 most positive or negative annual values of the PDSI, ENSO, PDO and AO [54]. We then tested associations between two-and three-way combinations among the phases of the ENSO, PDO and AO, and we defined extreme climate interactions by combining the 100 highest or lowest ranked annual values [54].

Fire history
We detected 13 fire events and three additional fires (recorded with one fire-scarred tree and verified with post-fire cohorts) from 99 fire scars on 45 partial sections of living, fire-scarred trees (figure 2). The majority of partial sections were from western red-cedar (73%) and we found an average of two fire scars  per sample (maximum 7) and 27% of trees had only one scar. We cored over 3000 trees (age range 52-953 years) and sampled 9-92 trees (mean = 52 trees) from each of 30 plots (electronic supplementary material, table S2). Our fire evidence was derived from fire scars and confirmed with stand establishment data, which reliably extended to 1325 (689 years, sample depth more than five trees). The year of maximum fire synchrony and the last detected fire event in the study area was in 1893, when 90% of plots burned in a 287 ha mixed severity fire, in which 66% of trees survived (figure 2). The point fire interval (PFI (the MFI at the tree scale)) was 95 years and the composite MFI was 39 years in the period 1376-1893 (electronic supplementary material, table S2).

Controls on the spatial variation of fires
The abundance of fire-scarred trees was found to consistently decrease with distance from former habitation sites (GLM Poisson regression: b = −1.810 × 10 −3 , z = −2.677, p < 0.01; figure 3a). The most parsimonious model related the abundance of fire scars to distance from former habitation sites only (on the basis of AIC c ), and model averaging indicated that distance to former habitation site was significantly more influential than any other predictor or combination of predictor variables. Model validation confirmed that distance to former habitation site explained 65% of the variation (pseudo R 2 ) in the abundance of fire-scarred trees (electronic supplementary material,  type; on the basis of AIC c ; table 1b). Model validation confirmed that distance to former habitation site was the most important predictor and together with aspect and bog forest vegetation type explained 52% of the variation in the frequency of fire events (table 1b and electronic supplementary material,  table S4).

Fire occurrence and interacting climate indexes
Fire frequency analyses of observed versus expected two-and three-way interactions between phase combinations of the ENSO, PDO and AO in the year of fire occurrence were non-significant (electronic supplementary material, table S5 and figure S2). This demonstrated that fire years were not more likely to occur during specific phase combinations and their associated climate signatures. Although fire and climate interactions did not follow an observable pattern in the year of fire events, interannual and multidecadal interactions between two-and three-way combinations of the ENSO, PDO and AO were significant (figure 4). Fires were synchronous with the combined positive phases of the ENSO and PDO 3-14 and 20-30 years prior to fire events (BEA, p = 0.005; figure 4e). Similarly, the negative phase of ENSO (La Niña) and the positive phase of the PDO occurred more often than expected 1-12 years prior to fire events (BEA, p = 0.001; figure 4f ). Fires were consistently synchronous with the positive phase of the PDO and the negative phase of the AO during the 6-33-year period prior to fire events (BEA, p = 0.001; figure 4g). BEA analysis of interacting three-way phase combinations of the ENSO, PDO and AO were synchronous with fire events only when the ENSO and PDO were positive and the AO was negative 8-36 years prior to fire events (BEA, p = 0.001; figure 4h). All other phase combinations were non-significant (electronic supplementary material, figure S3).

Discussion
Drivers of historic fire activity remain largely unknown in coastal temperate rainforests in the PNW of North America [1,4,59]. We present the first examination of the influences of and interrelationships between humans and climate on fire activity in a very wet coastal temperate rainforest using a ca 700year record of fire scars and stand establishment. Our data describe a low and mixed severity fire regime in which fires occurred on average every 39 years over the 517-year period from the first detected fire in 1376 to the last fire in 1893 ( figure 2). Although the frequency of fire events on Hecate Island may seem low relative to fire-prone landscapes, our data indicate that fire occurrence was 25 times more likely  than previous estimates of ca 1000-year intervals [14][15][16]. Such a fire frequency would have had large impacts on forest composition and structure, specifically in the AD 1600-1893 period when fires were more frequent (figure 2). The combination of low and mixed severity fires recurring at multidecadal intervals, the absence of fire after Indigenous abandonment (figure 2), and the spatial pattern of fire relative to habitation sites (figure 3) suggests that fire may have been intentionally used as a tool for resource management [9,60]. In addition, significant lagged effects of climate indexes prior to fire events indicate an added role of climate variability on fire occurrence ( figure 4).
There is no evidence of fire activity after the last recorded fire in our study area in 1893 ( figure 2). This 123-year fire-free interval (1893-present day) is distinct from the reconstructed fire history and is the longest fire-free period documented in the study area in seven centuries ( figure 2). This period also coincides with the only time in roughly 13 000 years that the study area was not continuously inhabited by Indigenous peoples [31,32]. Unlike many regions of British Columbia that experienced widespread fire suppression in the twentieth century, there are no known records of ignitions or suppressions in this remote coastal temperate rainforest during the twentieth century, and very few in the entire coastal temperate rainforest as a whole [15,61]. However, several cycles of in-phase climate oscillations resulting in warmer and drier conditions conducive to fire spread occurred in the study area after the last recorded fire in 1893 (electronic supplementary material, table S1).
Fire-climate studies in the PNW relate synchronous fire years to the year's climate and cite warm and dry spring conditions [22], combined with low winter snowpack [7], as factors affecting fire years by drying fuels and resulting in longer fire seasons. Contrary to this, we found that antecedent drought in the year prior to fire events was significantly related to fire occurrence (figure 4a) and a closer examination of individual fire years revealed that fire events corresponded to 2-7 years of dry conditions (negative PDSI; electronic supplementary material, figure S1). The largest fire in the study area (287 ha) occurred in 1893, during one of the wettest (10th percentile) reconstructed years on record and one of the strongest negative ENSO (5th percentile) and PDO (25th percentile) climate years. Although the combined effects of these climate indexes suggest unsuitable conditions for fuel ignition and spread, a closer examination of the 10-year period prior to the fire year reveals a period of sustained drought and three of the strongest reconstructed positive ENSO (5th percentile) and PDO (10th percentile) years. This pattern of antecedent drought in combination with positive in-phase ENSO and PDO indexes prior to large fire events in the study area repeats through the dataset (figure 4) and reveals that antecedent drought may be a necessary precondition for fire initiation and spread at this study site. The effects of interannual and decadal drought on forest growth and fuel dynamics are not well documented in this region. Such antecedent climate events may affect the available fuel profile (as occurs in grassy fuel types such as in the southwest USA [24,62]) or other unknown aspects of the fire regime.
Our results are comparable to other studies in the PNW and indicate that interannual drought and interactions with regional climate variability may have been an important prerequisite for fuel combustion and fire spread [23], but our results are unique in concluding that fires would not likely have occurred without human ignitions (figure 3). For example, in our analyses, we found that fire events were not spatially biased to expected locations such as south-facing aspects [42], steep slopes [20] or rocky outcrops at middle and high elevations [19], but were strongly associated with the locations of former habitation sites (figure 3). We also found no difference in the frequency of fire events across the four vegetation types despite dissimilarities in fuel load, fuel availability and potential flammability (electronic supplementary material, figure 4) [21]. Although the effects of topography and vegetation on fire frequency were not readily apparent, the variables aspect and bog forest vegetation type were included in our final model selection (based on weighted probabilities) along with distance to former habitation site (table 1b). These variables may suggest other potential relationships between the spatial patterns of fire, vegetation and humans such as the influence of prevailing summer winds, natural fuel breaks and territorial boundaries. Despite being able to reconstruct historical fire events and gain insights into fire behaviour with the presence of fire-scarred trees [45], we could not confirm whether anthropogenic burning was accidental or intentional, or how often burning was attempted but unsuccessful because fire weather was not conducive to fuel ignition and fire spread [11].
Historically, low severity fires may have been intentionally set near habitation sites to clear land in otherwise dense forests, harvest timber and promote important and accessible food species such as berry-producing shrubs and bracken fern (Pteridium aquilinium), a fire follower species and an important starch in the diet of coastal Indigenous peoples [9,60]. In these very wet forests, centuries of repeat burning surrounding habitation sites may have affected fuel loads, seral stages and the abundance and distribution of flammable vegetation [1,2,24]. Because biomass is not limited in coastal temperate rainforests, humans can exert control over natural conditions that promote or reduce limits on fire activity. Therefore, human drivers and their ecosystem effects are probably more pronounced relative to long-term climate in these very wet ecosystems [1,2,4]. As climate change continues to promote prolonged warm and dry conditions and human footprints expand [8], fires in these ignition-limited systems may occur again, but have the potential to become large and uncontrollable [1,2,4,8]. We propose that a greater understanding of the cultural traditions of burning and their ecological effects would provide more insight into human controls on historic fire activity and help assess vulnerabilities to future human-fire-climate interactions [2].

Conclusion
Fires and their drivers have never been reconstructed with tree rings at centennial scales in coastal temperate rainforests. Although our study comprises only 300 ha in British Columbia, our results provide insight into the role of fire in very wet coastal temperate rainforests and the need to re-examine the fire regime concept to include human ignitions, fire as a tool for resource management and longer analysis periods [2]. Our results are consistent with other studies that suggest fire is rare in coastal temperate rainforest settings in the twentieth century [15][16][17], but we also found that historical fires occurred on average every 39 years during the AD 1376-1893 period and were synchronous with large-scale climate oscillations. Both Indigenous land use and climate variation appear to have interacted to create longterm trends in fire occurrence over the past 700 years, and we find the fire history of our study site is best explained by relating cycles of anthropogenic burning, which varied on decadal timescales, to decadal variability in climate. The 'ecological legacies' of historic anthropogenic burning, in the form of persistent effects on contemporary forest structure, probably abound in this region, and thus discredit the myth of coastal temperate rainforests as pristine landscapes. Our clear evidence of precontact anthropogenic burning calls into question our current understanding of fire history in coastal temperate rainforests and how we define landscapes that were historically managed with fire.