Reconstructing Holocene fire history in a southern Appalachian forest using soil charcoal


  • Corresponding Editor: A. H. Lloyd.


Lacking long-term dendrochronological and lake sediment data, little is known regarding the history of fire in southern Appalachian forests through the Holocene. Here we used 82 radiocarbon ages for soil charcoal collected from local depositional sites along a topographic gradient from mixed hardwood (Liriodendron tulipifera and Quercus spp.) to oak–pine (Quercus prinus and Pinus rigida) forest to provide a coarse-grained picture of changes in fire frequency within a 10-ha area during the Holocene. Charcoal ages ranged from 0 to 10 570 yr BP, with a single date older than 4000 yr BP. The data indicate that fires occurred regularly across the study area over the past 4000 yr. Further, such fires were not confined to dry oak–pine dominated ridges, but extended downslope into areas that are today dominated by mesic hardwood forest. Summed probability analysis taking into account radiometric errors suggests that fires became more frequent ∼1000 yr ago, coinciding with the appearance of Woodland Tradition Native Americans in this region. We provide a direct demonstration of relatively frequent fire at the forest stand scale in Appalachian forests over a significant portion of the Holocene. Our results are consistent with the widely held view that fires have become less frequent in this region over the past 250 years. We discuss the difficulties in calculating the inbuilt error associated with estimating actual fire dates from charcoal fragments. But we conclude that such analysis of soil charcoal is a promising approach for reconstructing general trends in fire behavior within forest stands in this region.


It has long been understood that the species and communities of the pine-dominated ridge tops of the southern Appalachians are adapted to recurrent fires (Whittaker 1956, Barden and Woods 1976, Harrod and White 1999). The thick bark, basal sprouting, and serotinous cones of the pines (Pinus rigida and P. pungens) are well-known fire adaptations (Burns and Honkala 1990), and their fine, needled leaves promote or perpetuate the spread of fires (Nowacki and Abrams 2008). Further, germination and early growth of these conifers are best suited to the high-light, duff-free conditions typical of burned areas (Barden and Woods 1976).

However, the nature and frequency of fire are less apparent in hardwood-dominated communities that occur at lower topographic positions (slopes, coves) where moisture availability is higher (Runkle 1985, van Lear and Waldrop 1989, Delcourt and Delcourt 1997). Early explorers and settlers often described these communities as woodlands with open grassy understory conditions indicative of high-frequency, low-intensity fire regimes. The prevalence of fire-adapted oaks (Quercus), hickory (Carya), and chestnut (Castanea) supports this assessment. Occasional stand replacement fires may have been important during times of severe drought, particularly if occurring after canopy disturbance (e.g., blowdown) when dead and down fuel levels were high (Abrams 1992, Brown 2002).

Whatever fire regimes prevailed along these gradients through the Holocene, there is general agreement that they changed considerably with European settlement and subsequent changes in land use, forest management, and ignition patterns (MacCleery 1996, Yarnell 1998, Nowacki and Abrams 2008). During the 17th, 18th, and 19th centuries, increased human access, land clearing, and altered fuel conditions likely increased the frequency of fires (Jurgelski 2008). Over most of the last century, however, active fire suppression and landscape fragmentation have reduced fire frequency in most areas (Jurgelski 2008). Indeed, Nowacki and Abrams (2008) argue that successional change in the absence of fire has altered forest composition and structure in ways that have greatly diminished flammability, and therefore fire likelihood, across this entire region.

Knowledge of the nature of presettlement fire regimes and the changes that have occurred in those regimes since European settlement is quite relevant to modern forest management. It is fundamental to our interpretation of the likely composition and structure of presettlement forests, and thus, to the establishment of targets for forest restoration (Nowacki and Abrams 2008). It is important to our understanding of the likely responses of species to the increased use of prescribed fire for understory fuel management and forest restoration. Finally, the information generated by this study serves as a core data set for land managers who are trying to emulate historic disturbance regimes through silvicultural practices (Kimball et al. 1995, Engstrom et al. 1999, Vose et al. 1999).

Evidence for frequent, anthropogenic fires in the southern Appalachians in presettlement times has relied heavily on anecdotal and first-person descriptions of the landscape following European contact with the region beginning around 1690. These sources include William Bartram's journey through the Little Tennessee River valley of Georgia and North Carolina in the 1770s, Andrew Ellicott's similar route in 1812, and Robert Love's survey of Macon and Swain counties, North Carolina in 1820. All of these accounts include descriptions of open, sunny landscapes with well-spaced trees that suggest fire, and most attribute the ignition of fires to Native Americans. This historical evidence is weakened by the limited number and spatial extent of such observations, and by potential observer bias. Furthermore, Native American populations and cultures were experiencing significant change during this era, including changes in their use of fire (Abrams and Nowacki 2008, Jurgelski 2008).

Dendrochronological studies in the Appalachians have provided useful information regarding fire frequency in oak–pine dominated forests during the past several centuries (Guyette and Cutter 1991, Abrams et al. 1995, Nowacki and Abrams 1997, Shumway et al. 2001, Ruffner and Abrams 2002, Schuler and McClain 2003). However, fire-scarred trees older than 400 yr are exceedingly rare in this region.

Charcoal in wetland or lake sediments has been used to evaluate changes in fire occurrence at watershed or regional scales (Delcourt et al. 1986, Delcourt and Delcourt 1997), but appropriate depositional environments are very rare in this area. White (2007) combined analyses of isotopic anomalies in stalagmite carbon with radiocarbon dates of sediment charcoal from a West Virginia cave to infer changes in fire regimes and human land use in the area surrounding the cave. These studies provide insights into long-term changes in the frequency of fire on landscapes surrounding the sample sites. They do not reveal much information about fire regimes at the scale of individual stands, or variation in fire regimes among forest communities within a watershed.

In summary, assertions regarding presettlement southern Appalachian fire regimes within individual forest types are based largely on a very limited number of historical accounts and on the fact that regeneration of many common tree species (e.g., pines, oaks, and hickories) and a variety of herbs appears to require open forest structures often associated with fire. Indeed, the same could be said for assertions about fire regimes in much of the eastern deciduous forest (Nowacki and Abrams 2008).

In order to reconstruct the presettlement, stand-level fire history of a southern Appalachian forest, we collected and radiocarbon dated charcoal from soil in small (1–5 m2) depositional locations across a topographic gradient that included pine forest, xeric oak-dominated hardwood forest, and mesic cove forest. Wood charcoal is an inert and recalcitrant form of carbon that can persist in soils for millennia following fires. Although charred material may be transported significant distances by wind and water, the vast majority of macroscopic charcoal that accumulates in depositional locations (such as small depressions) within a forest likely originated within a few tens of meters of those locations (Blackford 2000, Ohlson and Tryterud 2000, Lynch et al. 2004, Higuera et al. 2007). Thus, the range of charcoal ages at sites likely reflects the history of fire at or very near that site.

Carcaillet et al. (2002) review the use of soil charcoal to reconstruct the history of Holocene fires and biomass burning. Radiocarbon dating of soil charcoal fragments has been used to establish the approximate age of fire events in tropical forests (Turner 1984, Piperno and Becker 1996, Hammond et al. 2006), high-elevation forests in the Sierra Nevada and French Alps (Anderson and Smith 1997, Carcaillet 1998, Hajdas et al. 2007), and old-growth Douglas fir–hemlock in the Pacific Northwest (Lertzman et al. 2002, Gavin et al. 2003, 2006, Higuera et al. 2005). In the Eastern Deciduous Biome, stand-level fire history studies using soil charcoal have focused on old-growth maple/birch forests in southern Quebec (Talon et al. 2005) and mixed hardwood forests in the Cumberland Plateau of middle Tennessee (Hart et al. 2008). Here we present the first such stand-level study in forests of the Appalachian Mountains.


Study area

This study was carried out in the Wine Spring Creek Ecosystem Management Area (WSCEMA) of the Nantahala National Forest in Macon County, North Carolina (35° 12′ N latitude, 83° 37′ W longitude) (Fig. 1). The WSCEMA is located on the western slope of the Nantahala Mountains. Sampling was done over an area of ∼10 ha at 1280–1430 m elevation, extending from a perennial stream across a south–southwest–west-facing slope to a ridge top. Soils for the site are Soco–Stecoah Complex and Rock Outcrop–Cataska Complex, coarse, mixed, mesic Typic Dystrochrepts. The sample area was selected because it included the gradient of vegetation from hemlock–hardwood cove near streams through mixed-oak hardwood on hill slopes to chestnut oak–pine forests on the ridge top. A few live pitch pine trees were widely scattered on the ridge top; dead boles and logs of this species were more abundant, however, suggesting that this species' relative importance has decreased over the past several decades. The well-developed soil A horizon indicated that this site had never been subjected to agriculture. A 1997 prescribed fire at this site provided abundant charcoal of recent age for comparison. Other research from WSCEMA has focused on the effects of this prescribed fire and is described in Elliott et al. (1999) and Vose et al. (1999). There is no record nor is there any evidence at this site of any other fires in the past century.

Figure 1.

Sample locations of soil cores at Wine Spring Creek Ecosystem Management Area in the Nantahala National Forest, Macon County, North Carolina, USA. The star on the inset map shows the location of the Ecosystem Management Area.

As classified by the LANDFIRE (2007a) remotely sensed data products, the existing vegetation type at the site is composed of Central and Southern Appalachian Montane Oak Forest (47%), Southern Appalachian Northern Hardwood Forest (36%), and Southern and Central Appalachian Cove Forest (17%). Based on biophysical characteristic modeling and the presumed natural fire regime for oak and oak–pine dominated forests in this region (Keane et al. 2006), the historical Fire Regime Group for >90% of this site is Group I, ≤35-year fire return interval, low and mixed severity (Keane et al. 2006, LANDFIRE 2007b).

Soil and charcoal sampling

Soil charcoal studies in other regions have generally relied on samples taken from sediments associated with local depressions (e.g., Lertzman et al. 2002, Gavin et al. 2003, but see Carcaillet and Thinon 1996). Such depressions are not common across the steep terrain of the southern Appalachians. Our reconnaissance revealed that soils containing significant amounts of charcoal often accumulated immediately upslope and downslope from rock outcrops (height = 0.5–3 m). Large rocks and rock outcrops act as barriers to overland colluvial soil movement, especially on a sloping landscape.

Using a slide hammer soil sampler we obtained 5 cm diameter intact soil cores from 33 rock outcrop locations. After removing the O horizon (duff layer), we collected soil cores varying from 10 to 30 cm, depending on the depth to bedrock at each sample location (total soil volume collected thus ranged from 785 to 2355 cm3). Sample distribution across the study site was dependent on the location of such depositional sites and was, therefore, nonrandom. Furthermore, we could not ascertain the organization of these soils (e.g., the correlation of soil age and depth) in advance of this study. For purposes of comparison, sample sites were divided into two groups: xeric, ridge top and upper slope oak–pine sites and mesic, downslope hardwood sites.

In the laboratory, soil cores were divided into 2-cm sections, taking care to preserve the stratigraphy of the sample. Each sample section was soaked for two hours in a 10% KOH (mass/volume) solution to disaggregate the organic matter. This slurry was then passed and rinsed through a 5-μm sieve. Samples were dried (100°C for 48 hours) and, with the aid of a dissecting microscope, the largest charcoal fragments were removed from the residual material. These charcoal pieces ranged in mass from 1 to 50 mg. Soil core sections contained 0–5 charcoal fragments >1 mg, the minimum size for 14C age determination.

In total, >250 charcoal fragments were isolated from 33 sample locations distributed across the sample area. Of these, 83 individual charcoal fragments from 18 of the 33 locations were selected for carbon dating: 65 fragments from 13 xeric oak–pine sites and 18 fragments from 5 mesic hardwood sites (Fig. 1). The total number was determined by funding available for carbon dating. Fragments were selected so as to represent entire core horizons from among different vegetation types on the site.

Radiocarbon dating (14C) using accelerator mass spectrometry (AMS) was done by Beta Analytic, Incorporated (Miami, Florida, USA) after additional pretreatment cleaning with KOH and HCl. Charcoal fragments as small as 1 mg can be dated using AMS. The calibrated age and its radiometric error (2 sigma = 2 SD) of each charcoal fragment were derived from the INTCAL 04 calibration curve (Reimer et al. 2004). The radiometric (2 sigma) error range associated with AMS dates varied among samples from as little as 100 years to as much as 470 years, and contained both continuous and irregular/broken error ranges.

These data were subjected to cumulative probabilities analysis (Meyer et al. 1992) using the “sum probabilities” option in CALIB 5.0.1 (Stuiver et al. 2005, Lafortune et al. 2006). This analysis allowed us to represent for a given year the relative probability that any fire represented in our data set occurred at the site. Additionally, we used the “test sample significance” feature in CALIB 5.0.1 to assess whether samples collected from the same sites were significantly different (Ward and Wilson 1978, Stuiver et al. 2005). We applied this test solely to our consideration of the charcoal age vs. sample depth relationship.

The calibrated radiometric age and error of a charcoal fragment corresponds not to the age of a fire event, but to the time when the wood that comprises a charcoal fragment was actually produced. The so-called “inbuilt error” of a date is estimated by the probability distribution of the difference between the date of carbon assimilation into wood and the date of the fire event that consumed and converted the wood to charcoal (Carcaillet 1998, Gavin 2001). The inbuilt error is additive to the radiometric error, and it is typically assumed to depend on stand age structure and the rate of wood decay in the ecosystem (Gavin 2001, Gavin et al. 2003). However, the age distribution of charcoal remaining after a fire is also influenced by the prevailing fire regime itself, because fire frequency and intensity affect both the amount of available fuel and how much is actually charred (Higuera et al. 2005).


Calibrated carbon dates for all but two of these 83 charcoal fragments ranged in age from 0 (probably corresponding to the 1997 prescribed fire in this area) to 4000 yr BP. Dates, as measured radiocarbon age, conventional date, and median probability of calibrated date, and their radiometric errors are presented in the Appendix. A single charcoal fragment from site 60 (oak–pine, 14–16 cm) dated at 10 570 (+120/−130) yr BP. This was so far outside the range for all other samples that it was not considered in subsequent analyses. However, this very old fragment is evidence that fires occurred at this site in the early Holocene. An additional hardwood sample was eliminated from analysis because it contained material younger than 0 yr BP (1950 CE) and could not be dated.

Across the study area the age of charcoal fragments demonstrated a range of relationships with their depth in soil cores (Fig. 2). Seven sites show no pattern (cf. Fig. 2, Site 60, 70, 110, 210, 304, 306, 307); one site shows a decreasing age/depth relationship (cf. Fig. 2, Site 302); and 10 sites show an increasing age/depth relationship. Of these 10 sites with increasing age/depth relationships, only 2 sites had a trend with more than two dates significantly different across depths (cf. Fig. 2, Site 80, 204, 215). We tested the similarity of charcoal dates from the same sample site and depth (11 pairwise combinations) using chi-square testing at the 0.05 level (Table 1). Among the 11 comparisons, 8 differed significantly in age. In 4 cases, age differences between charcoal fragments exceeded 1000 years.

Figure 2.

Charcoal age (±2-sigma calibration error) compared to depth in the soil at which the charcoal fragment was collected at each sample location. Sites that are currently dominated by xeric oak–pine forest are designated OP, and those dominated by mesic hardwood forest are designated HW.

Table 1. Pairwise chi-square comparison ages of charcoal fragments taken from the same sample site and depth at Wine Spring Creek Ecosystem Management Area (WSCEMA).Thumbnail image of

Charcoal sample ages (±2 sigma) are arrayed according to age and site, and the summed probability distributions for the two site groups are plotted in Fig. 3. This distribution of relative probabilities suggests that fires have regularly occurred at xeric oak–pine sites over the past 4000 yr. The record of fire events at mesic hardwood sites begins ca. 2020 yr BP. There is a relatively abrupt increase in the relative probability of a year being represented in our charcoal sample at ∼1000 yr BP in both xeric and mesic sites.

Figure 3.

(A) Ages of 81 charcoal fragments arrayed by calibrated median probability of the calibrated age with 2-sigma error bars. Note that 2-sigma errors are discontinuous for some ages. (B) The frequency of charcoal ages in 250-yr time bins. (C) The summed probability that any fire represented in our data set occurred at the site.

Using Gavin's (Gavin 2001, Gavin et al. 2003) methodology and age distributions for existing forests, and wood decay rates for this region suggested by Harmon et al. (1986), the inbuilt error for the oak–pine forests is likely between +50 and +100 years, although without knowledge of the actual fire regime, this estimate is quite uncertain. Thus, we did not attempt to include this inbuilt error in our summed probability analysis. Therefore, the most probable dates of fire events are, on average, likely 50–100 yr more recent than the most probable radiometric dates of charcoal fragments. Furthermore, the radiometric errors used in the summed probability analysis underestimate the total error surrounding estimated dates of fire events. This does not, however, alter the overall trends represented in Fig. 3, nor the general conclusions we draw from them.


Despite our hope that the soil deposited around stable rock outcrops would provide a uniform chronology of depositional ages at individual sites and thus facilitate cross-referencing fire ages among sites, we report a limited relationship between charcoal age and sample depth. This finding is consistent with Carcaillet (2001a, b), and may be attributable to bioturbation (Fisher and Binkley 2000), soil disturbance associated with treefall pits, and differences in patterns of particle movement and deposition. Furthermore, within sampling depths with multiple individually dated charcoal fragments, the majority of pairs proved significantly different. This suggests that aggregating fragments collected from similar depths is inappropriate in our region.

It is clear from Fig. 3 that fires have occurred regularly on this forested slope over the past 4000 yr. This is the first direct demonstration of relatively frequent fire within a single Appalachian forest stand spanning a significant portion of the Holocene. Furthermore, fires were not confined to ridge top sites that currently support oak–pine forests, but they extended into areas that are today dominated by mesic hardwood communities. Although no charcoal fragments >2020 yr BP were found in mesic hardwood sites, we speculate that this may reflect our unequal sampling effort rather than inherent differences in fire regime between site types.

The single 10 570-yr-old charcoal fragment is evidence that fire occurred at this site in the early Holocene. The plant communities at this elevation were likely dominated by spruce, fir, and pine at this time (Delcourt and Delcourt 1981, 1983, Anderson 1995). The fire event represented by this fragment occurred at approximately the same time that humans first moved into this region (Gragson et al. 2008).

Interpretation of temporal trends in the likelihood of fire is complicated by the fact that the number of samples represented in particular age bins diminishes with increasing charcoal age. Among 82 samples, 35 were 0–499 yr BP, 24 were 500–999 yr BP, 10 were 1000–1999 yr BP, 8 were 2000–2999 yr BP, 4 were 3000–3999 yr BP, and 1 was >10 000 yr BP. This pattern of diminishing data with time is inherent in virtually all historical data sets (Egan and Howell 2001). Nevertheless, the abrupt appearance at 4000 yr BP of regular fires and the marked increase in the summed relative probability of charcoal ages at 1000 years are consistent with historical trends in human activity and fire frequency hypothesized for the southern Appalachians by Delcourt and Delcourt (1986, 1998).

Abrams and Nowacki (2008) argue that Native American use of fire to manipulate vegetation in the eastern United States was ubiquitous, driving the composition of plant communities at the regional and biome scales. The palynological record from Horse Cove Bog, North Carolina (located ∼20 km from our study area), shows that chestnut, oak, and pine were prevalent in the area surrounding the bog from 4000 yr BP to the present (Delcourt and Delcourt 1997). Charcoal is also present through this time, but is especially abundant in peat dating after 1000 yr BP. Delcourt and Delcourt (1986, 1998) note that Native Americans of the Woodland Tradition appeared in this region ∼4000 yr ago, and they attribute fires during that period to these hunter-gatherers. Mississippian people appeared in this area ca. 1000 yr BP, coinciding with the appearance of Zea mays and a number of weedy herbs in the pollen record. The widespread use of fire by Mississippian Native Americans is well documented (Hatley 1993, Delcourt et al. 1998).

In eastern Kentucky, Delcourt et al. (1998) linked Native American use of fire to the dominance of oak–hickory forests starting 3000 yr ago (also see Ison 2000). White's (2007) recent analysis of pollen and charcoal deposits in a West Virginia cave suggests an increase in fire in that location beginning 4000 yr BP and lasting until the arrival of Europeans. Here too, Native Americans were implicated. In the only additional soil charcoal study in the southern portion of the Eastern Deciduous Biome, Hart et al. (2008) describe a comparable range of fire occurrence (five fire events spanning 6785 to 174 yr BP) in a hardwood deciduous forest on the Cumberland Plateau of middle Tennessee (located 185 km from our study area).

Because the charcoal ages presented here are from fire events within a single forest stand, it is tempting to try to infer changes in fire return times from these data. However, this is not possible because the uncertainties associated with assigning fire dates from radiometric ages of charcoal fragments are too large. Certainly, the summed probability plot (Fig. 3) is consistent with frequent, low-severity fires during the past millennium. If our charcoal samples are representative of the fire events during the period 4000–1000 yr BP, then we might speculate that fires were less frequent, and therefore, more severe during that time interval.

After accounting for the prescribed fire, a simple histogram of charcoal ages suggests that fires have become less frequent during the past 250 yr (Fig. 3B). This would coincide with the decline in Native American populations in this region after the beginning of European exploration and settlement. However, radiometric errors are particularly high for samples more modern than 400 yr BP, so this suggestion cannot be verified statistically. Consistent with known fire records for this area, there is no evidence of fire at this site in the century prior to the 1997 prescribed fire. This absence of fire is the likely cause for the lack of recruitment of pitch pine on the ridge top.

In summary, we conclude that fires have regularly burned across this slope over the past 4000 years. Fires appear to have become more frequent and widespread ca. 1000 yr ago, coinciding with the appearance of Woodland Tradition Native Americans in this area. The actual composition of the forests that burned during these four millennia remains unknown, although this amount of fire suggests that pines were likely far more prevalent in presettlement times than today. The nature of presettlement fires, e.g., their seasonality and severity, also remains unknown. The virtual absence of fire across most of the southern Appalachians over the past 75 years suggests that this once important ecological process is now lacking (Brose et al. 2001). Species composition of forests has undoubtedly changed as a consequence. As Nowacki and Abrams (2008) suggest, current structure and composition of fuels may prevent prescribed fires from replicating the effects of historic fires.


We thank James Vose and Katherine Elliot for advice and assistance in locating our field site. We are also grateful to Jim Clark and Ben Poulter for helpful advice on analyses and data interpretation. Paul Heine provided assistance with laboratory analyses. We are especially appreciative of the thoughtful comments and criticisms we received from two anonymous reviewers. Funding for this research was provided by the USDA Forest Service Southeastern Experimental Station.


Charcoal age data by sample location and depth (Ecological Archives E091-049-A1).