Linking tree-ring and sediment-charcoal
records to reconstruct fire occurrence
and area burned in subalpine forests of
Yellowstone National Park, USA
Philip E. Higuera, Cathy Whitlock and Josh A. Gage1
Montana State University, USA
Abstract
Reconstructing specific fire-history metrics with charcoal records has been difficult, in part because calibration data sets are rare. We calibrated charcoal
accumulation in sediments from three medium (14–19 ha) and one large (4250 ha) lake with a 300 yr tree-ring-based fire-history reconstruction from
central Yellowstone National Park (YNP) to reconstruct local fire occurrence and area burned within a 128 840 ha study area. Charcoal peaks most
accurately reflected fires within 1.2–3.0 km of coring sites, whereas total charcoal accumulation correlated best with area burned within 6.0–51 km
(r2=0.22–0.62, p<0.05). To reconstruct area burned for the entire study area, we developed a statistical model based on a composite charcoal record. The
model explained 64–79% of the variability in area burned from ad 1675 to 1960 and was robust to cross-validation. Reconstructed area burned from ad
1240–1975 was significantly higher during periods including extreme annual drought (p=0.05), and area burned varied significantly at ≈ 60 yr timescales
(p<0.05), similar to the variability in an independent precipitation reconstruction covering the same period. Widespread burning (>10 000 ha) occurred
at 150–300 yr intervals, and at the site level, fire probability increased with stand age (composite Weibull c parameter = 1.61 [95% CI 1.36–2.54]), both
suggesting that post-fire stand development played an important intermediary role between climate and fire by increasing fuel abundance and probability
of fire spread. Our study illustrates the possibility of reconstructing area burned with multiple charcoal records, and results imply that future fire regimes
in YNP will be governed by direct impacts of altered moisture regimes and by vegetation dynamics affecting the abundance and continuity of fuels.
Keywords
calibration, charcoal analysis, climate change, fire history, Pinus contorta, wildfires
The Holocene
21(2) 327 –341
© The Author(s) 2010
Reprints and permission:
sagepub.co.uk/journalsPermissions.nav
DOI: 10.1177/0959683610374882
http://hol.sagepub.com
Introduction
Fire is a dominant natural disturbance in forested ecosystems
linking climate change and biosphere response. Understanding
these linkages has become an important priority, particularly in
western North America, where numerous large stand-replacing
fires have occurred in recent years (e.g. National Interagency Fire
Center, 2008: http://www.nifc.gov/fire_info/fire_stats.htm). Fire-
history reconstructions from tree-ring and lake-sediment records
are the primary source of information for evaluating the prece-
dence of current and future fire activity (Conedera et al., 2009).
Tree-ring records provide fire-history information with high tem-
poral resolution for the last 300–500 yr, whereas high-resolution
charcoal records are more widely spaced geographically, offer
decadal- to multidecadal-scale resolution, and span millennia.
Testing the assumptions of charcoal-based fire history studies
is a critical goal for paleofire research. In the western USA, this
work began in Yellowstone National Park (YNP), where Whitlock
and Millspaugh (1996) collected sediment samples in small lakes
in burned and unburned watersheds to understand post-fire char-
coal accumulation. Millspaugh and Whitlock (1995) used these
insights to develop a 750-yr-long fire history based on sediment
charcoal from five lakes in central YNP, comparing the age of
charcoal peaks with a regional fire chronology based on tree-ring
data (Romme and Despain, 1989). These initial studies did not
quantify the spatial domain represented by charcoal peaks, nor
did they interpret overall charcoal abundance.
Since the early work in YNP, our understanding of sediment-
charcoal records and the techniques used to infer fire history have
greatly improved. For example, statistical analyses have allowed
reconstructions of fire history at small spatial scales by decom-
posing charcoal time series into ‘background’ and ‘peak’ compo-
nents, and applying a threshold based on (semi-) objective criteria
to identify samples likely associated with local fire events (e.g.
Clark et al., 1996; Gavin et al., 2006; Higuera et al., 2009; Long
et al., 1998). Support for these techniques comes from empirical
studies linking known fires with charcoal accumulation (Gardner
and Whitlock, 2001; Gavin et al., 2003; Higuera et al., 2005;
Lynch et al., 2004; Tinner et al., 1998) and process-based models
that consider charcoal production, dispersal, and deposition
Received 2 October 2009; revised manuscript accepted 26 April 2010
1Present address: American Wildlands, P.O. Box 6669, Bozeman MT
59715, USA
Corresponding author:
Philip E. Higuera, Department of Forest Ecology and Biogeosciences,
Box 441133, University of Idaho, Moscow ID 83844, USA
Email: phiguera@uidaho.edu
Research paper

328 The Holocene 21(2)
(Clark, 1988b; Higuera et al., 2007; Peters and Higuera, 2007).
CharSim (Higuera et al., 2007) is a recently developed process-
based model that incorporates fire history, primary and secondary
charcoal transport, sediment mixing, and sediment sampling to
create sediment-charcoal records based on a user-defined fire
regime. In addition to supporting the assumption that charcoal
peaks reflect fire occurrence at small spatial scales (c. 1 km
radius), a major prediction from CharSim is that overall trends in
macroscopic charcoal reflect area burned at larger spatial scales
(c. 10+ km radius). Although this prediction is valuable for the
interpretation of charcoal records (e.g. Brubaker et al., 2009;
Marlon et al., 2008, 2009), it has received little direct or indirect
empirical testing (but see Duffin et al., 2008).
The original YNP data set analyzed by Millspaugh and Whit-
lock (1995) remains one of the most complete calibration data sets
for testing this and other key questions about the nature of
sediment-charcoal records. In this study, we revisit these data and
apply new techniques for their comparison and interpretation. Spe-
cifically, spatially explicit comparisons between charcoal and tree-
ring records help to (1) statistically calibrate charcoal records to
detect both ‘local’ fire occurrence and ‘extra local’ area burned and
(2) test the theoretical relationships between area burned and char-
coal accumulation inferred from CharSim. We then develop a
composite record of fire occurrence and a statistical model recon-
structing area burned from ad 1225 to 1975. This exercise serves
as a case study to demonstrate the potential for reconstructing fire
occurrence and area burned at well-defined spatial scales, where
both charcoal and tree-ring data sets are available. The value of
this approach is illustrated by a comparison of the area-burned
reconstruction to a precipitation reconstruction for the YNP region
(Gray et al., 2007). Our objectives were to answer the following
questions: (1) What metrics of fire regimes (e.g. fire occurrence,
fire size, area burned) are recorded by charcoal data and at what
spatial scales? (2) How can combining charcoal records yield
information on regional fire history, and in particular area burned?
And (3) how do changes in regional fire history compare with
variations in precipitation over the last 750 yr in central YNP?
Methods and rationale
Study area and stand-age reconstruction
Four sites were used in this study (Figure 1). Duck (44°25'N,
110°35'W, elev. 2374 m, 14.2 ha, water depth 18.5 m), Mallard
(44°28'N, 110°47'W, elev. 2454 m, 13.7 ha, water depth 9.1 m),
and Dryad (44°33'N, 110°31'W, elev. 2530 m, 18.5 ha, water depth
8.5 m) lakes are fed by small inlet streams and have no outlet
streams. West Thumb (44°26'N, 110°32'W, elev. 2357 m, 4250 ha,
water depth 81 m) is a large sub-basin of Yellowstone Lake. The
Central Plateau of YNP is dominated by infertile substrates from
rhyolite volcanic rocks, which support lodgepole pine (Pinus con-
torta Dougl. var. latifolia) in about 80% of the forested area
(Despain, 1990). The combination of homogenous vegetation and
subdued topography should minimize the impact that these vari-
ables have on charcoal dispersal and deposition. In combination
with a historic regime of large, infrequent, stand-replacing fires
(Romme and Despain, 1989), these attributes make the Central
Plateau an ideal location for linking tree-ring and sediment-char-
coal records and comparing results to CharSim.
A tree-ring-based stand-age map developed by Romme and
Despain (1989) and published by Tinker et al. (2003) was used to
estimate area burned from ad 1675 to 1975 (Figure 1). The coverage
was obtained from DB Tinker as an Arc/Info shapefile and con-
verted to raster format with a cell size of 100 m using ArcGIS
(ESRI Inc., Redlands CA). Stand ages within the 128 840 ha
study area were estimated from 5–10 increment cores from the
dominant lodgepole pine in forest patches > 4 ha. We summarized
this reconstruction considering two important sources of error.
First, stand ages provide a minimum time-since-fire, because
post-fire lodgepole pine regeneration occurs over several years,
and tree cores taken above the root crown underestimate tree age
(Agee, 1993). We thus binned stand ages in 15-yr intervals. Sec-
ond, more recent fires erase evidence of older fires, and area
burned by older fires is likely underestimated. We accounted for
this fading record by using ad 1675 as the oldest stand age in our
analysis, even though the data set includes trees dating to ad 1425
(Romme and Despain, 1989; Tinker et al., 2003; Figure 1).
Sediment records
Sediment-charcoal data collected by Millspaugh and Whitlock
(1995) were analyzed with newly developed methods to estimate fire
history over the past 500–750 yr. Briefly, charcoal particles 125–250
μm in diameter were quantified from 5 cm3 samples taken from con-
tinuous 1 cm intervals. Chronologies for the last 150–200 yr were
based on 210Pb methods using the constant rate of supply model
(Appleby and Oldfield, 1978); the age of older sediments was esti-
mated using the cumulative dry weight of the interval and the mean
sediment accumulation rate for the dated portion of the cores
(Whitlock and Millspaugh, 1996). The original age extrapolation
was justified given similarity between estimated sedimentation rates
over the past several centuries and those in better-dated sites in YNP
(Huerta et al., 2009; Millspaugh et al., 2000). Sediment accumula-
tion rates were used to calculate charcoal accumulation rates (CHAR,
pieces/cm2 per yr) and resulted in median sample resolutions of 8, 10,
7, and 13 yr/sample at Duck, Mallard, Dryad, and West Thumb,
respectively. Prior to analysis, all records were interpolated to 15 yr
intervals to account for varying sampling resolution within and
between sites and for comparisons with the tree-ring record.
Calibrating charcoal records to detect fire occurrence
and defining ‘local’ spatial scales
To identify charcoal peaks potentially related to ‘local’ fire occur-
rence, we decomposed each charcoal series by subtracting 450 yr
trends (aka ‘background’ charcoal; estimated with a locally
weighted regression robust to outliers) to obtain a ‘peak’ CHAR
series. We used a Gaussian mixture model to define noise-related
variations in the entire peak series (Gavin et al., 2006) using the
program CharAnalysis (Higuera et al., 2009; available online at
http://code.google.com/p/charanalysis/). For each record, we con-
sidered three possible global threshold values (‘peak thresholds’)
for separating noise-related from fire-related variations in peak
CHAR, defined by the 95th, 99th, and 99.9th percentiles of the
noise-related Gaussian distribution.
We evaluated the accuracy of peak-inferred fires by compar-
ing identified peaks with tree-ring-inferred fires (‘true fires’;
Appendix 1, Figure A1). Accuracy was defined as the true-posi-
tive rate (proportion of peaks correctly identifying fires) minus
the false-positive rate (proportion of peaks incorrectly identifying
fires). For example, consider a record that has a total of four char-
coal peaks; three of these peaks match three true fires, while one