Ecology, 90(7), 2009, pp. 1788–1801
 2009 by the Ecological Society of America
Linking sediment-charcoal records and ecological modeling
to understand causes of fire-regime change in boreal forests
LINDA B. BRUBAKER,1,7 PHILIP E. HIGUERA,1,2,3 T. SCOTT RUPP,4 MARK A. OLSON,5 PATRICIA M. ANDERSON,6
AND FENG SHENG HU3
1College of Forest Resources, Box 352100, University of Washington, Seattle, Washington 98195-2100 USA
2Department of Earth Sciences, 200 Traphagen Hall, Montana State University, Bozeman, Montana 58717 USA
3Departments of Plant Biology and Geology, University of Illinois, 265 Moril Hall, 505 Goodwin Avenue, Urbana, Illinois 61801 USA
4Department of Forest Sciences, 368 O’Neill Building, Box 757200, University of Alaska, Fairbanks, Alaska 99775 USA
5P.O. Box 83021, Fairbanks, Alaska 99708 USA
6Department of Earth and Space Sciences and the Quaternary Research Center, Box 351310, University of Washington,
Seattle, Washington 98195-1310 USA
Abstract. Interactions between vegetation and fire have the potential to overshadow
direct effects of climate change on fire regimes in boreal forests of North America. We develop
methods to compare sediment-charcoal records with fire regimes simulated by an ecological
model, ALFRESCO (Alaskan Frame-based Ecosystem Code) and apply these methods to
evaluate potential causes of a mid-Holocene fire-regime shift in boreal forests of the south-
central Brooks Range, Alaska, USA. Fire-return intervals (FRIs, number of years between
fires) are estimated over the past 7000 calibrated 14C years (7–0 kyr BP [before present]) from
short-term variations in charcoal accumulation rates (CHARs) at three lakes, and an index of
area burned is inferred from long-term CHARs at these sites. ALFRESCO simulations of
FRIs and annual area burned are based on prescribed vegetation and climate for 7–5 kyr BP
and 5–0 kyr BP, inferred from pollen and stomata records and qualitative paleoclimate
proxies. Two sets of experiments examine potential causes of increased burning between 7–5
and 5–0 kyr BP. (1) Static-vegetation scenarios: white spruce dominates with static mean
temperature and total precipitation of the growing season for 7–0 kyr BP or with decreased
temperature and/or increased precipitation for 5–0 kyr BP. (2) Changed-vegetation scenarios:
black spruce dominates 5–0 kyr BP, with static temperature and precipitation or decreased
temperature and/or increased precipitation. Median FRIs decreased between 7–5 and 5–0 kyr
BP in empirical data and changed-vegetation scenarios but remained relatively constant in
static-vegetation scenarios. Median empirical and simulated FRIs are not statistically different
for 7–5 kyr BP and for two changed-vegetation scenarios (temperature decrease, precipitation
increase) for 5–0 kyr BP. In these scenarios, cooler temperatures or increased precipitation
dampened the effect of increased landscape flammability resulting from the increase in black
spruce. CHAR records and all changed-vegetation scenarios indicate long-term increases in
area burned between 7–5 and 5–0 kyr BP. The similarity of CHAR and ALFRESCO results
demonstrates the compatibility of these independent data sets for investigating ecological
mechanisms causing past fire-regime changes. The finding that vegetation flammability was a
major driver of Holocene fire regimes is consistent with other investigations that suggest that
landscape fuel characteristics will mediate the direct effects of future climate change on boreal
fire regimes.
Key words: Alaska, USA; Alaska Frame-based Ecosystem Code; ALFRESCO; black spruce; boreal
forest; Brooks Range; charcoal records; data-model comparison; fire regime; Picea; white spruce.
INTRODUCTION
Several recent investigations suggest that the effects of
climate warming on Alaskan boreal fire regimes will be
partially indirect due to strong interactions between
climate, vegetation, and fire (Rupp et al. 2002, Duffy et
al. 2007, Higuera et al. 2009). For example, analyses of
fire events since CE (Common Era) 1950 show that
forest type interacts with climate to modify the direct
effects of climate on fire occurrence and spread (e.g.,
Kasischke et al. 2002, Duffy et al. 2007). This occurs
because differences in the foliage and branching patterns
of dominant tree species cause marked differences in
vegetation flammability (Rupp et al. 2006), leading to
unexpected relationships between local climate and fire.
In particular, within Alaskan boreal forests, highly
flammable black spruce stands occupy wet soils on
north-facing slopes (Viereck et al. 1986), causing the
shortest fire rotations to occur on cold, wet sites (Drury
Manuscript received 27 April 2008; revised 8 October 2008;
accepted 14 October 2008. Corresponding Editor: A. H. Lloyd.
7 Present address: Professor L. B. Brubaker, 474 Eagle
Crest Road, Camano Island, Washington 98282 USA.
E-mail: lbru@u.washington.edu
1788

and Grissom 2008). Conversely, white spruce stands
occupy warmer sites on well-drained soil (Viereck et al.
1986), but the low flammability of this species results in
long fire-return intervals (Drury and Grissom 2008).
Given such interactions, vegetation responses to future
climate warming may cause complex and unexpected
changes in fire regimes of Alaska and other boreal
regions (Balshi et al. 2007, Soja et al. 2007, Ruckstuhl et
al. 2008).
Ecological modeling and paleorecords provide insights
about the potential effects of climate change on Alaskan
fire regimes (e.g., Lynch et al. 2002, Rupp et al. 2002, Hu
et al. 2006, Tinner et al. 2008, Higuera et al. 2009). Model
simulations using ALFRESCO (Alaskan Frame-based
Ecosystem Code), a spatially explicit model which
depicts fire and vegetation-recruitment dynamics across
Alaskan landscapes (Rupp et al. 2000 a, b), highlight the
importance of vegetation flammability to boreal fire
regimes (Rupp et al. 2002). These simulations emphasize
that climate-induced changes in species composition can
alter fire frequency and fire size, changing landscape-level
vegetation patterns and feedbacks to fire regimes (Rupp
et al. 2002, Chapin et al. 2003). Sediment-charcoal
records document that major shifts in fire regimes during
the Holocene corresponded more closely to vegetation
characteristics than to inferred climate (Lynch et al.
2002, Higuera et al. 2009). Thus both model simulations
and empirical data suggest that vegetation can modify
the direct impacts of climate change on fire regimes by
altering landscape flammability.
Although both ecological modeling and paleorecords
can reveal fire-regime shifts associated with changes in
climate and vegetation, neither approach provides a
rigorous assessment of the relative effects of climate and
vegetation on landscape burning. Paleorecords preserve
long-term fire histories across a variety of climate and
vegetation types, but do not provide direct evidence of
the inferred roles of climate or vegetation. Ecological
modeling predicts the consequences of vegetation–fire
interactions under different climate conditions, but
typically cannot check the realism of simulated climate
and/or vegetation effects on fire occurrence and spread.
Comparisons of paleorecords and model simulations
offer a powerful means to evaluate alternative mecha-
nisms of past ecosystem change (Anderson et al. 2006),
and comparisons between pollen data and forest
simulations have been used to examine climate, distur-
bance, and human impacts on forest ecosystems
worldwide (e.g., Cowling et al. 2001, Hall and McGlone
2001, Keller et al. 2002). However, no study has used a
data–model approach to evaluate the extent to which
vegetation has modified the direct effect of climate on
past fire regimes.
Here we link high-resolution lake-sediment-charcoal
records (Higuera et al. 2009) to fire-history simulations
from the landscape model ALFRESCO. Our main goal
is to develop analytical methods for comparing AL-
FRESCO simulations with charcoal records as a proof
of concept that such comparisons can reveal ecological
processes underlying shifts in past fire regimes. The
empirical and simulated data sets are then used to
examine several potential causes of a fire-regime shift
associated with the mid-Holocene (;5 kyr BP; calibrat-
ed 14C ages are used throughout the paper) expansion of
black spruce (Picea mariana) into areas dominated by
white spruce (P. glauca) and shrub birch (Betula
glandulosa) in the south-central Brooks Range of
Alaska, USA. Previous paleoclimate studies suggest
climate cooling and/or moistening during the mid-
Holocene (see Anderson et al. 2003), and sediment
records show increased charcoal content and inferred
fire frequencies (Lynch et al. 2002, Higuera et al. 2009).
Since the fire-regime shift was in an opposite direction
than would be expected by climate cooling and/or
moistening, investigators have proposed that the ex-
treme flammability of black spruce drove the change in
mid-Holocene fire regimes (Lynch et al. 2002, Hu et al.
2006, Higuera et al. 2009).
METHODS AND RATIONALE
Study sites and paleorecords
Vegetation and fire histories are reconstructed from
pollen, stomata, and charcoal records at three sites
(Ruppert, Wild Tussock, and Code Lakes) along the
southern flanks of the central Brooks Range, Alaska,
USA (Fig. 1). See Higuera et al. (2009) for a detailed
description of the sediment records and analytical
methods. The present vegetation near each site is a
mosaic of black spruce (Picea mariana), white spruce
(Picea glauca), paper birch (Betula papyrifera), and
aspen (Populus tremuloides), with shrub birch (Betula
glandulosa) tundra in non-forested areas. Fire is the
primary disturbance, with an estimated fire rotation
period for the study area of 175 yr, based on data for CE
(Common Era) 1950–2001 (Kasischke et al. 2002).
Pollen percentages and stomata presence/absence were
used to characterize past vegetation and to detect the
development of black-spruce dominated forest. Analog
analyses (Gavin et al. 2003) compare fossil and modern
pollen assemblages and indicate the probability of a given
past vegetation assemblage matching present boreal
forests types of North America. The first presence of
spruce stomata (Carlson 2003) confirmed analog inter-
pretations of modern boreal forest development at the
transition between the white and black spruce periods.
Past fire regimes are reconstructed from variations in
charcoal accumulation rates (CHARs) in sediment
cores. The temporal resolution of the analyzed sedi-
ments (;15 yr/sample; Higuera et al. 2009) is optimal
for detecting CHAR peaks resulting from fires with
return intervals .75 yr (Higuera et al. 2007). Peaks in
CHARs were used to estimate fire-return intervals
(FRIs) according to methods described by Higuera et
al. (2009).
Knowledge of Holocene climate change in Alaska
remains limited. Recent geochemical studies have
July 2009 1789CHARCOAL RECORDS AND ECOLOGICAL MODELING

revealed marked variations in temperature and moisture
during the mid- and late-Holocene (Anderson et al.
2001, Hu et al. 2001, 2003). Recent summaries based on
fossil, lake-level, and glacial records (Anderson et al.
2003, Kaufman et al. 2004) have suggested that central
Alaska was generally warmer and/or drier than the
present during the early Holocene (;10–5 kyr BP) and
reached modern conditions by the late Holocene (;4–0
kyr BP). These two periods correspond roughly to the
white spruce and black spruce zones of pollen records
from the study area (Higuera et al. 2009). We examine
the 7–0 kyr BP period because it spans the white and
black spruce pollen zones and is documented by high-
resolution charcoal records at all sites.
ALFRESCO model
ALFRESCO was originally developed to simulate the
response of subarctic vegetation to a changing climate
and disturbance regime (Rupp et al. 2000a, b). The
boreal forest version of ALFRESCO was developed to
explore the interactions and feedbacks among fire,
climate, and vegetation in interior Alaska (Rupp et al.
2002). A detailed description of ALFRESCO can be
obtained from the literature (Rupp et al. 2000a, b, 2002).
For the purpose of our research, we focus our
description on details relevant to this specific model
application.
Rather than modeling fire behavior, ALFRESCO
models the empirical relationship between growing-
season climate (e.g., mean temperature and total
precipitation for May–September) and total annual area
burned (i.e., the footprint of fire on the landscape; Duffy
et al. 2005, Rupp et al. 2007). It models changes in
vegetation flammability during succession through a
flammability coefficient that changes with vegetation
type and stand age (i.e., fuel accumulation; Chapin et al.
2003). The fire regime is simulated stochastically as a
function of climate (as defined here), vegetation type
(i.e., fuel structure and loading), and time-since-last-fire.
‘‘Ignition’’ of a pixel is determined using a random
number generator and a flammability value for each
pixel, and an ignited pixel may spread to any of the eight
surrounding pixels. Fire ‘‘spread’’ depends on the
flammability of the receptor pixel and effects of natural
firebreaks, including non-vegetated mountain slopes and
large water bodies. ALFRESCO’s coarse pixel resolu-
tion (1 km2) precludes inclusion of fine-resolution
factors such as wind and topography that are typically
important components of fire behavior models.
Simulation landscape.—Because simulating the land-
scape of all sites was computationally too expensive, we
chose a 100 3 100 km area centered on Ruppert Lake
(Fig. 1). This area is large enough to capture parameters
of Alaskan fire regimes, appropriate for comparison
with sediment-charcoal records (Higuera et al. 2007),
and representative of the other study sites. ALFRESCO
models four vegetation types: upland tundra, black
spruce forest, white spruce forest, and early successional
deciduous vegetation (Rupp et al. 2000a, b, 2002). The
current (;CE 1990) composition of the simulation
landscape is black spruce (28%), white spruce (9%),
and early successional deciduous vegetation (1%) in
lowlands near Ruppert Lake, with tundra (62%)
predominantly on mountain slopes to the north (Fig. 1).
Climate input.—ALFRESCO maintains mean tem-
perature and total precipitation of the growing season
(May–September) for each pixel because these climate
variables are used to determine internal variables of
vegetation types and to model fire (Rupp et al. 2000a, b).
Geographically defined climate data for Alaska (CE
1960–1990; Fleming et al. 2000), which account for
variability due to topography (e.g., aspect, elevation)
and regional influences (e.g., synoptic weather patterns),
were used to set up the spatial variability in climate
across the simulated landscape. A temporal offset from
the current climate was selected at each time step from a
normal deviate for the entire simulation landscape (i.e.,
climate variability is assumed constant over the study
region; Rupp et al. 2000b).
ALFRESCO requires quantitative climate estimates
as inputs for both the white and black spruce periods,
but only qualitative Holocene climate records are
available from the study area. In our experiments, we
prescribed a climate shift to cooler and/or moister
conditions at the beginning of the black spruce period,
which is qualitatively consistent with paleo evidence.
However, two aspects of our choice may be unrealistic.
First, we chose temperature and precipitation shifts that
were large enough to affect ALFRESCO-simulated fire
regimes but may have been larger than actual Holocene
temperature and/or precipitation change (since the
simulated 5–0 kyr BP climate is somewhat cooler and
wetter than the closest climate station to the study area,
Bettles, Alaska, USA). As a result, while the ALFRES-
CO experiments represent the qualitative effects of
Holocene temperature decrease and/or precipitation
increase, they may overestimate the magnitude of those
effects. Second, centennial- to millennial-scale fluctua-
tions in temperature and moisture occurred elsewhere in
Alaska within the past 7 kyr BP (Anderson et al. 2001,
Hu et al. 2001, 2003), and some of these fluctuations
altered fire regimes (Tinner et al. 2008). Although these
climate variations may have also affected the fire
regimes of our study area, their effects are not
addressed by our experiments. Given the exploratory
stage of our research and the time and expense of
running multiple ALFRESCO runs, we felt that our
rather simple experiments were justified. We address the
implication of these limitations to our results in the
Discussion.
We generated climate input in three steps. First,
climate for the white spruce period (7–5 kyr BP) was
defined based on climate station data (growing season
temperature, precipitation, and variability) in the
southeastern Brooks Range (Arctic Village, Alaska,
USA; data available online), where modern boreal forests
LINDA B. BRUBAKER ET AL.1790 Ecology, Vol. 90, No. 7

are dominated by white spruce.8 These climate data were
also used in simulations of the black spruce period under
conditions of static temperature and/or precipitation.
Second, we generated temperature and precipitation
values for each time step of the simulation, accounting
for observed interannual variability, by developing
temporal offset vectors for each climate variable for
the length of the simulation period. This was done by
sampling with replacement from the station data and
then summarizing by decade. Third, climate for the
black spruce period was generated by applying a linear-
ramping algorithm to the simulated climate data for 5–4
kyr BP, representing a 28C decrease in temperature and
an 86 mm (50%) increase in precipitation.
Model simulations
ALFRESCO scenarios reveal the separate and inter-
active effects of climate (shifts in growing-season mean
temperature and/or total precipitation) and vegetation
(presence/absence of black spruce) on the fire regimes of
non-tundra vegetation types in the study area. Prior to
simulations, we conducted model calibrations and
followed a standard ‘‘spin-up’’ methodology (Rupp et
al. 2007) to generate realistic modern vegetation
patterns (i.e., vegetation type, age, and patch dynamics),
which were modified in the simulation experiments.
Model calibration (Rupp et al. 2006) focused on
calibrating the fire routine to generate fire numbers
and area burned similar to modern observations (;6
ignitions and 19 200 ha burned per decade, CE 1950–
2000). Single-replicate simulations were conducted at a 1
3 1 km pixel resolution on the Ruppert Lake landscape,
run at decadal time steps for 7–0 kyr BP. Fire
occurrence (or absence) for each pixel and each time
step was archived to calculate FRIs.
Fire history from each ALFRESCO simulation was
summarized using FRIs and mean annual area burned.
Each simulation produced a series of FRIs for each non-
tundra pixel, yielding many thousands of FRIs. FRIs
were calculated for non-tundra pixels at the spatial scale
of each ALFRESCO pixel. When a non-tundra pixel
FIG. 1. Ruppert Lake, Alaska, USA, study area. The dark circle near the center of both panels shows Ruppert Lake. The left
panel shows site location and real topography. The right panel shows simulation landscape with simulated vegetation types in
1-km2 pixels. White indicates no vegetation cover. This simulated landscape was modified in ALFRESCO (Alaskan Frame-based
Ecosystem Code) experiments.
8 hhttp://www.wrcc.dri.edu/cgi-bin/cliMAIN.pl?akarcti
July 2009 1791CHARCOAL RECORDS AND ECOLOGICAL MODELING

burned in a simulation, the time (in years) since the last
fire occurred was recorded. Annual area burned
represents the total number of pixels (1 km2) burned in
each simulated decade, divided by 10 yr.
Vegetation and climate periods.—The white spruce
and black spruce periods are defined as 7–5 and 5–0 kyr
BP, respectively. Climate and vegetation changes are
depicted as ramped transitions between 5–4 kyr BP,
generally similar to the time of black spruce pollen
increase in sediment records (Fig. 2). Our results did not
differ when we examined the impact of using other
transition periods. Because our goal was to examine the
responses of fire regimes to alternative vegetation and
climate scenarios, rather than to simulate transient
responses to a specific paleoenvironmental proxy, the
time of the transitions in paleodata and model
simulations did not need to match.
To create input for the white spruce period (7–5 kyr
BP), we generated a spin-up landscape by converting all
modern black spruce pixels to tundra, resulting in
roughly 90% tundra and 10% white spruce. This
simulated landscape is consistent with the finding (see
Results) that the closest modern analogs to fossil pollen
assemblages during the white spruce period are in
forest–tundra ecotones (Higuera et al. 2009), where
black spruce is absent and white spruce is a minor
component of the vegetation. For the 5–4 kyr BP
transition period, when modern pollen analogs are
located in black spruce-dominated forests, we forced
the converted tundra cells back to black spruce using a
linear-ramping algorithm that re-established black
spruce to be consistent with the modern landscape
(Fig. 1). For the remainder of this period, the vegetation
cover of all cells was allowed to change in response to
burning. While these vegetation changes are not the
focus of the present study, they are mentioned as context
for discussing past vegetation–fire interactions.
Experiments.—The 7–5 kyr BP baseline vegetation
and climate are modeled as just described. Two sets of
experiments (Table 1) were designed to address hypoth-
eses about the independent and interactive effects of
vegetation and climate on the shift in fire regimes
between the white and black spruce periods (e.g., Fig. 3).
Comparison of charcoal-based and ALFRESCO
fire history records
We compared FRIs from charcoal and ALFRESCO
records based on two well-founded assumptions: (1) the
spatial scale represented by peaks identified in the
sediment-charcoal records is approximated by the
1-km2 cell size of ALFRESCO, based on empirical
(Gavin et al. 2003, Lynch et al. 2004) and theoretical
considerations (Higuera et al. 2007) of how charcoal
data record fire occurrence and (2) at least part of the
landscape was covered by trees for the past 7 kyr BP
(Fig. 2; Higuera et al. 2009).
FIG. 2. Pollen percentages of spruce, birch, and alder; probability of analog values for comparisons between fossil samples and
modern boreal forest, forest–tundra, and arctic tundra vegetation zones; and charcoal accumulation rate (CHAR) for Ruppert
Lake over time (calibrated years before present). Solid and open circles on the top panel represent presence or absence of spruce
stomata. Triangles along the x-axis indicate locations of radiocarbon dates.
LINDA B. BRUBAKER ET AL.1792 Ecology, Vol. 90, No. 7

Fire-return intervals.—For each vegetation period, we
pooled FRIs inferred from CHAR records at individual
sites to form a distribution representing FRIs across the
study transect as a whole. Given the small number of
inferred fires at each site, this step enhanced the rigor of
comparisons between CHAR and ALFRESCO records
and is justified because FRI distributions of individual
sites did not differ statistically (P . 0.05) within
vegetation periods, based on statistical comparisons
used by Higuera et al. (2009). To compensate for
inherent biases in the ALFRESCO data set, FRIs from
ALFRESCO records were truncated to remove FRIs .
500 yr. This step is justified because ALFRESCO does
not model age-dependent mortality within the late
successional spruce forest trajectories, thus allowing a
small number of pixels to record unrealistically long
FRIs in ALFRESCO simulations that encompass multi-
millennial time periods. We chose to truncate the
ALFRESCO FRIs at 500 yr because this was the
longest FRI in the CHAR data set. Additionally, we
removed charcoal-based FRIs , 90 yr from the 7–5 kyr
BP period because ALFRESCO rules that govern
succession on white spruce sites do not allow the
early-successional deciduous state to switch to white
spruce until 90 yr after fire (Rupp et al. 2000b). While
reasonable on the modern landscape, this may be a
limitation of ALFRESCO when simulating paleofire
regimes during the white spruce zone, as paleorecords
show evidence of shorter FRIs during this period
(Higuera et al. 2009). Nevertheless, this truncation step
should have little impact on our interpretation of fire
regimes within the white spruce period because FRI
distributions including vs. excluding FRIs , 90 yr are
not statistically different based on Kolmogorov-
Smirnov or likelihood-ratio tests (Higuera et al. 2009;
medianallFRI FRIs¼ 135, range¼ 90–210 years between
fires; medianFRI90 ¼ 188, range ¼ 142–263 years
between fires; PK-S ¼ 0.17; Plk-rat ¼ 0.26).
We developed a three-step statistical test for compar-
ing charcoal-based FRIs and ALFRESCO FRIs for
each period: (1) calculate pixel-specific FRIs (i.e., fire-
return intervals for each pixel) for all non-tundra pixels
on the ALFRESCO landscape, (2) combine these into a
single FRI distribution, and (3) compare the cumulative
distribution function (CDF) of the ALFRESCO FRIs to
the CDF of pooled FRIs from the charcoal data set. We
compared CDFs using a two-sample Anderson-Darling
(AD) test (Anderson and Darling 1954, Pettitt 1976) and
evaluated the null hypothesis that both populations
(charcoal and ALFRESCO FRIs) come from the same
continuous distribution. To compare the CHAR and
ALFRESCO data sets under conditions of similar
spatial sample densities, we randomly sampled FRIs
from non-tundra pixels in each ALFRESCO scenario
with the number of sampled pixels equal to the number
of sites recording charcoal accumulation (i.e., three). As
with the charcoal data set, FRIs from the ALFRESCO
pixels were combined to form a pooled FRI record. We
used a Monte Carlo approach to compare FRIs from
the pooled CHAR and ALFRESCO data sets to derive a
‘‘model score’’ that quantifies the similarity of these data
sets. For each vegetation zone, the subsample of FRIs
from the ALFRESCO landscape was compared to the
charcoal-based FRI data set with the Anderson-Darling
test. If the comparisons yielded a P value , 0.10, the
null hypothesis of no difference between charcoal-based
and ALFRESCO FRI distributions was rejected and 0
was recorded; if P . 0.10, the null hypothesis was not
rejected and 1 was recorded (note that using a P value of
0.10 in this case is more conservative that using a P
value of 0.05). The model score equals the proportion of
1s recorded when this procedure was repeated 10 000
times. Thus the model score ranges from 0 to 1, with 0
representing perfect dissimilarity (100% rejection of null
hypothesis of no difference between FRI distributions
estimated by CHAR and ALFRESCO data sets) and 1
TABLE 1. Alaskan Frame-based Ecosystem Code (ALFRESCO) scenarios and rationales.
ALFRESCO
scenario Factors Rationale
Static vegetation
Static climate sV, sT, sP Did fire regimes change in the absence of vegetation and climate change? A control
for identifying vegetation and/or climate effects. Vegetation and climate
are modeled as for 7–5 kyr BP.
Changed climate sV, sT, DP
sV, DT, sP
sV, DT, DP
Did climatic cooling and/or moistening alone cause past fire-regime shifts? The initial
vegetation is set at 7–5 kyr BP and allowed to change in response to temperature
decrease (28C) and/or precipitation increase (86 mm) ramped over 5–4 kyr BP.
Changed vegetation
Static climate DV, sT, sP Did increased black spruce alone cause past fire-regime shifts? Black spruce cover is
increased from 0% to 28% to match the modern landscape near Ruppert Lake,
Alaska, USA. Climate is modeled as for 7–5 kyr BP.
Changed climate DV, DT, sP
DV, sT, DP
DV, DT, DP
Did both increased black spruce and climate change influence past fire regimes?
Black spruce, precipitation, and/or temperature are ramped over 5–4 kyr BP.
Note: Each scenario represents an alternative hypothesis about the cause of mid-Holocene fire-regime shifts documented in
charcoal accumulation rate (CHAR) records from the south-central Brooks Range, Alaska, USA.
 Abbreviations are: V, vegetation; T, temperature; P, precipitation; s, static; D, change.
July 2009 1793CHARCOAL RECORDS AND ECOLOGICAL MODELING

representing perfect similarity (100% acceptance of null
hypothesis) in CDFs.
Area burned.—We compared long-term variations in
CHARs (generally termed ‘‘background CHAR’’), as a
proxy for area burned, to long-term variations in area
burned calculated from ALFRESCO simulations. Al-
though background CHAR has often been considered to
be the ‘‘noise’’ component of charcoal records, recent
theoretical work (Higuera et al. 2007) indicates that for
a given vegetation type over long time periods, charcoal
deposition to a lake should be related to area burned.
Thus we explored whether these theoretical inferences
are supported by comparisons between background
charcoal in empirical data and ALFRESCO simulations
of area burned. In our comparison, the CHAR and
ALFRESCO series were treated identically to isolate
long-term trends and standardized to facilitate visual
comparisons. Each time series was smoothed with a
locally weighted regression using a 500-yr window
(Cleveland 1979), and the resulting time series were
standardized to a mean of 0 and a standard deviation of
1. For the CHAR time series, each record was
interpolated to uniform 15-yr intervals (Higuera et al.
2009) and a composite time series was developed by
averaging the three standardized time series.
RESULTS
Vegetation and fire history
Vegetation and fire histories are summarized here and
described in full by Higuera et al. (2009). Between 7 and
5 kyr BP, forest–tundra dominated the landscape near
Ruppert Lake. The strongest modern analogs to fossil
pollen assemblages are in North American forest–tundra
vegetation; further, the absence of stomata indicates a
sparse tree cover during this period (Fig. 2). Spruce
pollen during 7–5 kyr BP (Brubaker et al. 1983, Higuera
et al. 2009) is almost exclusively white spruce. Charcoal
accumulation rates (CHARs) for the white spruce period
are low (Fig. 4), with means at Ruppert, Wild Tussock,
and Code Lakes of 0.031–0.044 piecescm2yr1. The
median fire-return interval (medFRI) of the charcoal
data set is 135 years (95% CI¼ 90–210); the medFRI of
the truncated data set is 188 years (95% CI ¼ 142–263;
Table 2). The establishment of black spruce forests ;5
kyr BP is evidenced by the high probability of analog
(.75%) to pollen assemblages in modern black spruce
forests of North American and by the consistent
presence of spruce stomata (Fig. 2), which implies
continuous forest cover (Carlson 2003). Charcoal is
most abundant 5–0 kyr BP (Fig. 4): mean CHAR
increases to 0.09–0.19 piecescm2yr1, and medFRI
decreases to 120 years (95% CI ¼ 120–150; Table 2).
ALFRESCO simulations
7–5 kyr BP period.—As just described, vegetation
consisted of ;10% white spruce, with tundra dominat-
ing the remaining landscape, particularly north of
Ruppert Lake (Fig. 3). Mean annual area burned was
28 km2 (Figs. 5 and 6, Table 2); medFRI was 180 years
(Table 2). Descriptions of simulations will use abbrevi-
ations: vegetation as V, temperature as T, precipitation
as P, static as s, and change as D.
5–0 kyr BP period: Static vegetation.—
1. Static climates (V, sT, sP).—Vegetation cover did
not change, and fire regimes differed slightly from 7–5
kyr BP (mean annual area burned¼ 29 km2, medFRI¼
150 years; Table 2; Figs. 5 and 6) due to stochastic
processes in ALFRESCO.
2. Changed climates (V, DT, sP; sV, sT, DP; sV, DT,
DP).—Mean annual area burned remained constant or
decreased slightly (16–29 km2; Table 2, Fig. 6), and
medFRIs increased somewhat (190–200 years; Table 2,
Fig. 5). In each case, the changes were greatest with
shifts in both temperature and precipitation (sV, DT,
DP). Deciduous cover declined and white spruce cover
increased slightly with a change in both climate
variables, but neither responded to an individual shift
in temperature or precipitation.
5–0 kyr BP period: Changed vegetation.—
1. Static climate (DV, sT, sP).—Median FRIs de-
creased markedly to 110 years and mean annual area
burned increased to 181 km2 (Table 2, Figs. 5 and 6).
Deciduous cover increased and white spruce cover
declined markedly.
2. Changed climate (DV, sT, DP; DV, DT, sP [see Fig.
3]; DV, DT, DP).—Median FRIs and mean annual area
burned changed less dramatically in these scenarios
(medFRIs, 120–150 yr; mean area burned, 79–129 km2),
with the DV, DT, DP scenario showing the smallest
changes (Table 2, Figs. 5 and 6). Deciduous cover
increased in all scenarios. White spruce cover declined
somewhat in DV, sT, DP and DV, DT, sP, but remained
relatively stable in DV, DT, DP.
Comparison of ALFRESCO simulations
and charcoal records
FRIs.—For 7–5 kyr BP, high model scores (0.967)
indicate strong similarity in CDFs of empirical and
simulated FRIs, and low Anderson-Darling (AD)
statistics (0.65, P . 0.10; Table 2) indicate no statistical
differences in these records. For 5–0 kyr BP, empirical
FRIs are lower than simulated FRIs in all static-
vegetation scenarios (medFRI ¼ 120, medFRIs ¼ 150–
200, respectively). Low model scores (0.014–0.065) and
high AD statistics (17.21–34.16) confirm significant
differences in these data sets (Table 2). Empirical and
simulated FRIs are more similar for changed-vegetation
scenarios (model scores ¼ 0.328–0.460; Table 2), with
no statistical difference for the DV, sT, DP and DV, DT,
sP scenarios (AD ¼ 1.32, 1.37, respectively; P . 0.10;
Table 2).
Area burned.—Both the individual and composite
CHAR records and all ALFRESCO changed-vegetation
scenarios show long-term increases between 7–5 and 5–0
kyr BP (Fig. 6). However, the CHAR series increase
more gradually and lack a peak that is evident in
LINDA B. BRUBAKER ET AL.1794 Ecology, Vol. 90, No. 7

ALFRESCO data sets between 4.5–4.0 kyr BP. In
contrast, all static-vegetation scenarios differ from
CHAR series by showing no long-term trends in area
burned between 7–5 and 5–0 kyr BP.
DISCUSSION
Using data-model comparisons to investigate
past fire regimes
The comparison of empirical and simulated records is
a critical step in investigations linking paleodata and
ecosystem models because the degree of similarity
between data sets can help distinguish between alterna-
tive hypotheses to explain past change. However, given
the disparate metrics and spatial/temporal scales of
many empirical and simulated records, this step is not
straightforward. Even the most recent comparisons
between pollen records and forest models (Cowling et
al. 2001, Keller et al. 2002, Heiri et al. 2006) mention the
challenges of meshing pollen and simulated data sets. As
discussed next, we paid particular attention to the
metrics of charcoal and ALFRESCO data sets when
selecting methods for data-model comparisons.
Qualitative (visual) comparisons were used to evaluate
the similarity of the unlike metrics of long-term charcoal
accumulation rates (CHARs) and area burned by
ALFRESCO. At the most general level, the similarity
of multi-centennial variations in CHARs and mean
annual area burned supports the hypothesis that
changes in sediment-charcoal content reflect changes in
area burned. This inference is supported by three recent
investigations that indicate variations in background
charcoal reflect aspects of landscape burning. The
Higuera et al. (2007) model of charcoal dispersal and
incorporation into lake sediments predicts that at multi-
centennial to millennial time scales the amount of
charcoal deposited in lake sediments is a function of
FIG. 3. Example ALFRESCO output for the DV, DT, sP scenario. (Abbreviations are: V, vegetation; T, temperature; P,
precipitation; s, static; D, change.) Time series A shows temperature and precipitation over the simulation period (simulated years
BP). Time series B shows cover (km2) of simulated vegetation types and annual area burned (km2) on the Ruppert Lake landscape,
with the panels of C illustrating the pattern of vegetation types on simulated landscape. The dashed line in panels A and B separates
the white spruce period (7000–5000 yr BP) from the black spruce period (5000–0 yr BP) within the simulation.
July 2009 1795CHARCOAL RECORDS AND ECOLOGICAL MODELING

the area burned within ;10–20 km of the lake.
Empirical studies at regional and global scales suggest
a correspondence between long-term CHARs and the
amount of woody fuels on the landscape (Marlon et al.
2006) or the influence of climate and human drivers
(Marlon et al. 2008). Thus our qualitative comparisons
agree with recent findings that CHAR records have
potential to expand the understanding of past fire
regimes beyond the conventional inference of fire
frequencies.
FIG. 4. CHAR records for Ruppert, Code, and Wild Tussock lakes over time (calibrated years before present). Plus signs (þ)
indicate inferred fire events at each site.
TABLE 2. Mean annual area burned for ALFRESCO scenarios and results of Anderson-Darling (AD) test for no difference
between fire-return interval (FRI) distributions for truncated CHAR and ALFRESCO data sets.
Period
(kyr BP) Scenario
Mean annual area burned
Median FRI (yr) Model score AD§ P}7–5 kyr BP 5–0 kyr BP
Paleo
7–5 na na na 188 (142, 263) na na na
5–0 na na na 120 (120, 150) na na na
ALFRESCO
7–5 baseline 28 na 180 0.967 0.65 .0.10
5–0 sV, sT, sP na 29 150 0.065 17.21 ,0.01
sV, sT, DP na 29 200 0.014 34.16 ,0.01
sV, DT, sP na 23 190 0.023 27.85 ,0.01
sV, DT, DP na 16 200 0.036 32.99 ,0.01
5–0 DV, sT, sP na 181 110 0.328 4.14 ,0.01
DV, sT, DP na 127 120 0.457 1.32 .0.10
DV, DT, sP na 129 120 0.460 1.37 .0.10
DV, DT, DP na 79 150 0.421 4.39 ,0.01
Note: Scenario abbreviations are: V, vegetation; T, temperature; P, precipitation; s, static; D, change; na, not applicable.
 Fire-return intervals (FRIs), with 95% confidence intervals for the paleodata in parentheses. Confidence intervals are not given
for the ALFRESCO median FRI because the populations of FRIs are completely sampled, making the median FRI a known rather
than estimated statistic.
 Proportion of comparisons between paleo and ALFRESCO FRIs resulting in nonsignificant differences at alpha¼ 0.10, from
10 000 bootstrapped samples of the ALFRESCO data set. Model score ranges from 0 (imperfect) to 1 (perfect).
§ Anderson-Darling statistic for comparisons between ALFRESCO scenario and the appropriate paleo time period.
} P values .0.10 are significant.
LINDA B. BRUBAKER ET AL.1796 Ecology, Vol. 90, No. 7

We developed a quantitative measure (model scores) to
compare FRIs, which could be calculated from both
CHAR records andALFRESCO simulations. Themodel
score is a versatile similarity index that accounts for
differences in sample sizes and spatial variability in the
simulated landscape. Model scores are particularly useful
when comparing multiple simulations (i.e., different
hypotheses) to one set of empirical data. For example,
model scores show that charcoal-based FRIs are more
similar to the FRIs of the DV, DT, DP (0.421) than the sV,
sT, sP (0.065) scenario (Table 2), even thoughmedFRIs in
both scenarios are identical (150 yr). This example also
illustrates that comparisons of FRI distributions can lead
to different conclusions than comparisons of simple
metrics such as the median FRI, providing a potentially
more rigorous test of the similarity between empirical and
simulated data sets.
Our second quantitative comparison applied the
Anderson-Darling statistic to test the similarity of
empirical and simulated FRI series. To our knowledge,
this is the first time a statistical test has been used to
evaluate the adequacy of alternative causal hypotheses
to account for variations in paleodata (in our study, the
adequacy of ALFRESCO scenarios to describe shifts in
FIG. 5. Time series of area burned for ALFRESCO simulations. The gray line separates the two simulation periods.
July 2009 1797CHARCOAL RECORDS AND ECOLOGICAL MODELING

charcoal-based FRIs between 7–5 and 5–0 kyr BP). The
AD statistic was critical to interpreting the similarity of
FRIs in charcoal records and changed-vegetation
scenarios. AD tests identified only the DV, sT, DP and
DV, DT, sP scenarios as being indistinct from empirical
FRIs, leading to the inference that temperature or
precipitation interacted with vegetation to cause the
mid-Holocene shift in fire regimes (see Discussion:
Implications to fire ecology and vegetation responses to
future climate change for ecological implications).
Because statistical tests set a rigorous standard for
recognizing the similarity of data sets derived from
disparate approaches, the absence of statistical differ-
ences between empirical and simulated FRIs in two
scenarios supports a central premise of our approach:
that the similar spatial and temporal scales of CHAR
records and ALFRESCO simulations allow a rigorous
exploration of hypotheses about the causes of past fire-
regime shifts. The compatibility of CHAR and AL-
FRESCO data sets is due to several factors. First,
FIG. 6. Time series of standardized CHAR for Ruppert, Code, and Wild Tussock Lakes; composite standardized CHAR
(originally measured as pieces of charcoal per cm2 per year) for all lakes; and standardized annual area burned (originally measured
at km2/yr) simulated by ALFRESCO. All series have been smoothed with a locally weighted regression using a 500-yr window.
LINDA B. BRUBAKER ET AL.1798 Ecology, Vol. 90, No. 7

CHAR series are derived from contiguous, high-
resolution samples of sediment cores. Since no sediment
levels are skipped within a sediment record and sample
resolution is at decadal time scales, these series
continuously register charcoal deposition from the
theoretical number of fires within the potential charcoal
source area (Higuera et al. 2007, Peters and Higuera
2007). As a result, CHAR series can be used to calculate
the same metrics of fire regimes as estimated by
ALFRESCO (e.g., FRIs, potentially annual area
burned), permitting direct comparisons between empir-
ical and simulated data sets. Second, the temporal scales
of CHAR records are suitable for investigating the
dynamics of fire regimes with long fire-return intervals
(such as boreal forests). The limiting constraints are
slow sediment accumulation and sediment mixing,
which prevent charcoal peaks from being isolated by
laboratory sampling and/or dampen high-frequency
variations in charcoal records (Higuera et al. 2007).
Third, the spatial scales of sediment-charcoal records
and ALFRESCO are similar. Theoretical considerations
(Higuera et al. 2007) indicate that short-term variations
in CHARs in boreal forests register fire occurrence
within ;500 m of lakes (corresponding to the spatial
scale of 1 km2 pixels in ALFRESCO) and long-term
trends in total CHARs (‘‘background CHAR’’) register
area burned within ;10–20 km.
Inferring causes of Holocene fire-regime shifts
By depicting ecological interactions in vegetation/
climate scenarios, ALFRESCO simulations provide
insights about the vegetation–fire feedbacks driving
changes in past fire regimes. For example, in the static-
vegetation scenarios, decreased temperature and in-
creased precipitation (sV, DT, DP) resulted in fewer
ignitions and lower rates of fire spread, leading to fewer
fires and less area burned. The cover of deciduous vegeta-
tion decreased in response to the decline in area burned,
which favored the development of late-successional white
spruce forests. In contrast, when black spruce increased
in the changed-vegetation scenarios, landscape burning
increased dramatically in response to fuel buildup that
occurred when flammable black spruce forests became
extensive. Although the extent and continuity of
vegetation types influence the fire regime (Rupp et al.
2000b), species-specific flammability (i.e., differences
between white and black spruce forests) is the dominant
factor driving increased burning in the simulated records
(Rupp et al. 2002a). In these scenarios, decreases in
temperature and/or increases in moisture dampened the
effects of increasing black spruce, resulting in fewer fires
and less area burned than when climate did not change.
Overall, our results support evidence of several other
paleoecological investigations (Clark et al. 1996, Lynch
et al. 2002, Gavin et al. 2007), suggesting that in some
instances the quality and quantity of fuels have
overridden the direct effect of climate on Holocene fire
regimes.
Comparisons of empirical and simulated data allowed
us to identify which of the ALFRESCO scenarios best
describe the climate–vegetation–fire interactions respon-
sible for the fire-regime shift in paleorecords. For ex-
ample, model scores indicate that all changed-vegetation
scenarios are more similar than static-vegetation sce-
narios to the paleodata (Table 2), implying that vegeta-
tion change had a greater influence than climate on the
mid-Holocene fire-regime shift. However, AD tests
indicate that vegetation change was not the sole ex-
planation of altered fire regimes, as empirical and
simulated FRIs were statistically indistinct only when
temperature or precipitation shifts accompanied vegeta-
tion change. The inference that cooler temperature or
greater precipitation (but not both) dampened the effect
of increased landscape flammability is a novel interpre-
tation of the cause of mid-Holocene increased burning.
This interpretation illustrates that data-model compar-
isons have potential to identify the relative importance
and interactive effects of climate and vegetation on
landscape burning, a difficult task using pollen and
charcoal records alone.
We emphasize two cautionary notes to these results,
however. First, although the response of fire regimes to
prescribed temperature and precipitation changes are
plausible, given the direct effects of decreased temper-
ature and increased precipitation on landscape burning,
the temperature and precipitation influences may be too
large due to our choice of climate offsets in these
simulations. We suspect that with smaller temperature
and precipitation changes, the DV, DT, DP scenario
(which represents modern) would be most similar to
paleorecords for 5–0 kyr BP. Second, although the
magnitude of black spruce increase in our simulations is
reasonable (pollen assemblages became similar to
modern black spruce forests 5–0 kyr BP; Higuera et al.
2009) and simulated vegetation matches current vegeta-
tion near Ruppert Lake, our results do not address the
degree to which climate forced the mid-Holocene
increase in landscape flammability because the increase
in black spruce was not a function of climate in
ALFRESCO simulations. To improve the realism of
our simulations, we are developing independent quan-
titative climate proxies of mid-Holocene climate change
and species-specific climate responses for black spruce in
ALFRESCO.
Implications to fire ecology and vegetation responses
to future climate change
Our comparisons between simulated and empirical
records suggest that the mid-Holocene expansion of
black spruce fundamentally altered landscape flamma-
bility and caused fire regimes to change in a direction
opposite to that predicted by the direct effects of climate
(see also Higuera et al. 2009). Our study, therefore, adds
to growing evidence that changes in species- and
community-level traits can alter ecosystem processes
(e.g., Suding et al. 2008), which in turn may have
July 2009 1799CHARCOAL RECORDS AND ECOLOGICAL MODELING

consequences to the climate system (Chapin et al. 2008).
Although the expansion of highly flammable species is
known to increase fire occurrence in diverse ecosystems
(e.g., Brooks et al. 2004), vegetation-mediated changes
in fire regimes are not a universal finding of modern or
paleologic studies (Brooks et al. 2004, Gavin et al.
2007). The extent to which species-level traits affect
larger scales of ecological organization (including fire
regimes) is poorly understood and difficult to predict in
many regions (Suding et al. 2008). In black spruce
ecosystems, the influence of vegetation on fire regimes
results from a suite of species-level traits, including
flammable foliage and tree architecture, semi-serotinous
cones, and rapid recruitment following fire (Rupp et al.
2002).
The results of our study provide insights to potential
ecosystem responses to climate change at high latitudes,
where major changes in fuel types are predicted for the
future (ACIA 2004). For example, flammable shrubs are
currently increasing in Alaskan tundra, with potential to
facilitate fire occurrence and spread (Tape et al. 2006,
Higuera et al. 2008). In addition, some future scenarios
for boreal forests (Calef et al. 2005, Flannigan et al.
2005, Johnstone and Chapin 2006) predict an increase in
deciduous trees. With the high water content of
deciduous compared to coniferous fuels, this vegetation
change could lead to reduced fire occurrence. Our
findings imply that such fuel changes are potentially
important for future fire regimes and that ALFRESCO
modeling can explore the range of conditions in which
vegetation can substantially alter the direct effect of
climate on fire regimes.
ACKNOWLEDGMENTS
This research was funded by NSF collaborative grants to
L. B. Brubaker and P. M. Anderson (OPP-0112586), F. S. Hu
(OPP-0108702, ARC-0612366), and T. S. Rupp (OPP-
0108237), and an NSF Graduate Research Fellowship to
P. E. Higuera.
LITERATURE CITED
ACIA. 2004. Impacts of a warming arctic: arctic climate impact
assessment. Cambridge University Press, Cambridge, UK.
Anderson, L., M. B. Abbott, and P. B. Finney. 2001. Holocene
climate inferred from oxygen isotope ratios in lake sediments,
Central Brooks Range, Alaska. Quaternary Research 55:313–
321.
Anderson, N. J., H. Bugmann, J. A. Dearing, and M. J.
Gaillard. 2006. Linking palaeoenvironmental data and
models to understand the past and to predict the future.
Trends in Ecology and Evolution 21:696–704.
Anderson, P. M., M. E. Edwards, and L. B. Brubaker. 2003.
Results and paleoclimate implications of 35 years of
paleoecological research in Alaska. Pages 427–440 in A. R.
Gillespie, S. C. Porter, and B. F. Atwater, editors.
Developments in quaternary science, Vol. 1. Elsevier, New
York, New York, USA.
Anderson, T. W., and D. A. Darling. 1954. A test of goodness
of fit. Journal of the American Statistical Association 49:765–
769.
Balshi, M. S., et al. 2007. The role of historical fire disturbance
in the carbon dynamics of the pan-boreal region: a process-
based analysis. Journal of Geophysical Research–Biogeo-
sciences 112.
Brooks, M. L., C. M. D’Antonio, D. M. Richardson, J. B.
Grace, J. E. Keeley, J. M. DiTomaso, R. J. Hobbs, M.
Pellant, and D. Pyke. 2004. Effects of invasive alien plants on
fire regimes. BioScience 54:677–688.
Brubaker, L. B., H. L. Garfinkel, and M. E. Edwards. 1983. A
late Wisconsin and Holocene vegetation history from the
central Brooks Range: implications for Alaskan paleoecolo-
gy. Quaternary Research 20:194–214.
Calef, M. P., A. D. McGuire, H. E. Epstein, T. S. Rupp, and
H. H. Shugart. 2005. Analysis of vegetation distribution in
interior Alaska and sensitivity to climate change using a
logistic regression approach. Journal of Biogeography 32:
863–878.
Carlson, L. J. 2003. Describing the postglacial pattern and rate
of Picea expansion in Alaska using paleoecological records.
Dissertation. University of Washington, Seattle, Washington,
USA.
Chapin, F. S., J. T. Randerson, A. D. McGuire, J. A. Foley,
and C. B. Field. 2008. Changing feedbacks in the climate–
biosphere system. Frontiers in Ecology and the Environment
6:313–320.
Chapin, F. S., T. S. Rupp, A. M. Starfield, L. O. DeWilde, E. S.
Zavaleta, N. Fresco, J. Henkelman, and A. D. McGuire.
2003. Planning for resilience: modeling change in human–fire
interactions in the Alaskan boreal forest. Frontiers in
Ecology and the Environment 1:255–261.
Clark, J. S., P. D. Royall, and C. Chumbley. 1996. The role of
fire during climate change in an eastern deciduous forest at
Devil’s Bathtub, New York. Ecology 77:2148–2166.
Cleveland, W. S. 1979. Robust locally weighted regression and
smoothing scatterplots. Journal of the American Statistical
Association 74:829–836.
Cowling, S. A., M. T. Sykes, and R. H. W. Bradshaw. 2001.
Palaeovegetation-model comparisons, climate change and
tree succession in Scandinavia over the past 1500 years.
Journal of Ecology 89:227–236.
Drury, S. A., and P. J. Grissom. 2008. Fire history and fire
management implications in the Yukon Flats National
Wildlife Refuge, interior Alaska. Forest Ecology and
Management 256:304–312.
Duffy, P. A., J. Epting, J. M. Graham, T. S. Rupp, and A. D.
McGuire. 2007. Analysis of Alaskan burn severity patterns
using remotely sensed data. International Journal of Wild-
land Fire 16:277–284.
Flannigan, M. D., K. A. Logan, B. D. Amiro, W. R. Skinner,
and B. J. Stocks. 2005. Future area burned in Canada.
Climatic Change 72:1–16.
Fleming, M. D., F. S. Chapin, W. Cramer, G. L. Hufford, and
M. C. Serreze. 2000. Geographic patterns and dynamics of
Alaskan climate interpolated from a sparse station record.
Global Change Biology 6:49–58.
Gavin, D. G., D. J. Hallett, F. S. Hu, K. P. Lertzman, S. J.
Prichard, K. J. Brown, J. A. Lynch, P. Bartlein, and D. L.
Peterson. 2007. Forest fire and climate change in western
North America: insights from sediment charcoal records.
Frontiers in Ecology and the Environment 5:499–506.
Gavin, D. G., W. W. Oswald, E. R. Wahl, and J. W. Williams.
2003. A statistical approach to evaluating distance metrics
and analog assignments for pollen records. Quaternary
Research 60:356–367.
Hall, G. M. J., and M. S. McGlone. 2001. Forest reconstruction
and past climatic estimates for a deforested region of south-
eastern New Zealand. Landscape Ecology 16:501–521.
Heiri, C., H. Bugmann, W. Tinner, O. Heiri, and H. Lischke.
2006. A model-based reconstruction of Holocene treeline
dynamics in the Central Swiss Alps. Journal of Ecology 94:
206–216.
Higuera, P. E., L. B. Brubaker, P. M. Anderson, T. A. Brown,
A. T. Kennedy, and F. S. Hu. 2008. Frequent fires in ancient
LINDA B. BRUBAKER ET AL.1800 Ecology, Vol. 90, No. 7

shrub tundra: implications of paleorecords for arctic
environmental change. PLoS ONE 3:e0001744.
Higuera, P. E., L. B. Brubaker, P. M. Anderson, F. S. Hu, and
T. A. Brown. 2009. Vegetation mediated the impacts of
postglacial climate change on fire regimes in the south-central
Brooks Range, Alaska. Ecological Monographs 79:201–219.
Higuera, P. E., M. E. Peters, L. B. Brubaker, and D. G. Gavin.
2007. Understanding the origin and analysis of sediment-
charcoal records with a simulation model. Quaternary
Science Reviews 26:1790–1809.
Hu, F. S., L. B. Brubaker, D. G. Gavin, P. E. Higuera, J. A.
Lynch, T. S. Rupp, and W. Tinner. 2006. How climate and
vegetation influence the fire regime of the Alaskan Boreal
Biome: the Holocene perspective. Mitigation and Adaptation
Strategies for Global Change 11:829–846.
Hu, F. S., E. Ito, T. A. Brown, B. B. Curry, and D. R.
Engstrom. 2001. Pronounced climatic variations in Alaska
during the last two millennia. Proceedings of the National
Academy of Sciences (USA) 98:10552–10556.
Hu, F. S., D. S. Kaufman, S. Yoneji, D. E. Nelson, A.
Shemesh, Y. Hauang, J. Tian, G. Bond, B. Clegg, and T.
Brown. 2003. Cyclic variation and solar forcing of Holocene
climate in the Alaskan subarctic. Science 301:1890–1893.
Johnstone, J. F., and F. S. Chapin. 2006. Fire interval effects on
successional trajectory in boreal forests of northwest Canada.
Ecosystems 9:268–277.
Kasischke, E. S., D. Williams, and D. Barry. 2002. Analysis of
the patterns of large fires in the boreal forest region of
Alaska. International Journal of Wildland Fire 11:131–144.
Kaufman, D. S., et al. 2004. Holocene thermal maximum in the
western Arctic (0–180 degrees W). Quaternary Science
Reviews 23:529–560.
Keller, F., H. Lischke, T. Mathis, A. Mohl, L. Wick, B.
Ammann, and F. Kienast. 2002. Effects of climate, fire, and
humans on forest dynamics: forest simulations compared to
the palaeological record. Ecological Modelling 152:109–127.
Lynch, J. A., J. S. Clark, N. H. Bigelow, M. E. Edwards, and
B. P. Finney. 2002. Geographic and temporal variations in
fire history in boreal ecosystems of Alaska. Journal of
Geophysical Research 108:FFR8-1–FFR8-17.
Lynch, J. A., J. S. Clark, and B. J. Stocks. 2004. Charcoal
production, dispersal and deposition from the Fort Provi-
dence experimental fire: interpreting fire regimes from
charcoal records in boreal forests. Canadian Journal of
Forest Research 34:1642–1656.
Marlon, J. R., P. J. Bartlein, C. Carcaillet, D. G. Gavin, S. P.
Harrison, P. E. Higuera, F. Joos, M. J. Power, and I. C.
Prentice. 2008. Climate and human influences on global
biomass burning over the past two millennia. Nature
Geoscience 1:697–702.
Marlon, J., P. J. Bartlein, and C. Whitlock. 2006. Fire–fuel–
climate linkages in the northwestern USA during the
Holocene. Holocene 16:1059–1071.
Peters, M. E., and P. E. Higuera. 2007. Quantifying the source
area of macroscopic charcoal with a particle dispersal model.
Quaternary Research 67:304–310.
Pettitt, A. N. 1976. A two-sample Anderson-Darling rank
statistic. Biometrika 63:161–168.
Ruckstuhl, K. E., E. A. Johnson, and K. Miyanishi. 2008.
Introduction. The boreal forest and global change. Philo-
sophical Transactions of the Royal Society B 363:2245–2249.
Rupp, T. S., F. S. Chapin, and A. M. Starfield. 2000a.
Response of subarctic vegetation to transient climatic change
on the Seward Peninsula in northwest Alaska. Global
Change Biology 6:541–555.
Rupp, T. S., X. Chen, M. Olson, and A. D. McGuire. 2007.
Sensitivity of simulated boreal fire dynamics to uncertainties
in climate drivers. Earth Interactions 11.
Rupp, T. S., M. Olson, L. G. Adams, B. W. Dale, K. Joly, J.
Henkelman, W. B. Collins, and A. M. Starfield. 2006.
Simulating the influences of various fire regimes on caribou
winter habitat. Ecological Applications 16:1730–1743.
Rupp, T. S., A. M. Starfield, and F. S. Chapin. 2000b. A frame-
based spatially explicit model of subarctic vegetation
response to climatic change: comparison with a point model.
Landscape Ecology 15:383–400.
Rupp, T. S., A. M. Starfield, F. S. Chapin, and P. Duffy. 2002.
Modeling the impact of black spruce on the fire regime of
Alaskan boreal forest. Climatic Change 55:213–233.
Soja, A. J., N. M. Tchebakova, N. H. F. French, M. D.
Flannigan, H. H. Shugart, B. J. Stocks, A. I. Sukhinin, E. I.
Varfenova, F. S. Chapin, and P.W. Stackhouse. 2007. Climate-
induced boreal forest change: predictions versus current
observations. Global and Planetary Change 56:274–296.
Suding, K. N., S. Lavorel, F. S. Chapin, J. H. C. Cornelissen, S.
Diaz, E. Garnier, D. Goldberg, D. U. Hooper, S. T. Jackson,
and M. L. Navas. 2008. Scaling environmental change
through the community level: a trait-based response-and-
effect framework for plants. Global Change Biology 14:
1125–1140.
Tape, K., M. Sturm, and C. Racine. 2006. The evidence for
shrub expansion in Northern Alaska and the Pan-Arctic.
Global Change Biology 12:686–702.
Tinner, W., C. Bigler, S. Gedye, I. Gregory-Eaves, R. T. Jones,
P. Kaltenrieder, U. Krahenbuhl, and F. S. Hu. 2008. A 700-
year paleoecological record of boreal ecosystem responses to
climatic variation from Alaska. Ecology 89:729–743.
Viereck, L. A., K. Van Cleve, and C. T. Dyrness. 1986. Forest
ecosystem distribution in the taiga environment. Pages 22–43
in K. Van Cleve, F. S. Chapin, P. W. Flanagan, L. A.
Viereck, and C. T. Dyrness, editors. Forest ecosystems in the
Alaskan taiga. Springer-Verlag, New York, New York, USA.
July 2009 1801CHARCOAL RECORDS AND ECOLOGICAL MODELING