Global Change Biology (2010) 16, 2186–2197, doi: 10.1111/j.1365-2486.2009.02103.x
Rising concentrations of atmospheric CO2 have increased
growth in natural stands of quaking aspen
(Populus tremuloides)
C H R I S T O P H E R T . C O L E *, J O N E . A N D E R S O N *, R I C H A R D L . L I N D R O T H w and
D O N A L D M . WA L L E R z
*Division of Science and Mathematics, University of Minnesota Morris; Morris, MN 56267, USA, wDepartment of Entomology,
University of Wisconsin, Madison, WI 53706, USA, zBotany Department, University of Wisconsin, Madison, WI 53706, USA
Abstract
As atmospheric CO2 levels rise, temperate and boreal forests in the Northern Hemisphere are
gaining importance as carbon sinks. Quantification of that role, however, has been difficult
due to the confounding effects of climate change. Recent large-scale experiments with
quaking aspen (Populus tremuloides), a dominant species in many northern forest ecosystems,
indicate that elevated CO2 levels can enhance net primary production. Field studies also
reveal that droughts contribute to extensive aspen mortality. To complement this work, we
analyzed how the growth of wild aspen clones in Wisconsin has responded to historical shifts
in CO2 and climate, accounting for age, genotype (microsatellite heterozygosity), and other
factors. Aspen growth has increased an average of 53% over the past five decades, primarily in
response to the 19.2% rise in ambient CO2 levels. CO2-induced growth is particularly
enhanced during periods of high moisture availability. The analysis accounts for the highly
nonlinear changes in growth rate with age, and is unaffected by sex or location sampled.
Growth also increases with individual heterozygosity, but this heterozygote advantage has
not changed with rising levels of CO2 or moisture. Thus, increases in future growth predicted
from previous large-scale, common-garden work are already evident in this abundant and
ecologically important tree species. Owing to aspen’s role as a foundation species in many
North American forest ecosystems, CO2-stimulated growth is likely to have repercussions for
numerous associated species and ecosystem processes.
Keywords: climate change, CO2 fertilization, heterozygosity-fitness correlation, northern hardwood
forests, tree-rings
Received 13 July 2009 and accepted 10 September 2009
Introduction
Forests cover approximately 30% of earth’s terrestrial
surface and play critically important roles regulating
local environmental and climatic conditions and sequestering greenhouse gases linked to global climate
change (Bonan, 2008). Forests of the Northern Hemisphere are particular global sinks for CO2 and their role
has increased in recent decades (Myeni et al., 2001;
Boisvenue & Running, 2006; but see Stephens et al.,
2007). Despite this importance, we still lack a complete
picture of how historic increases in atmospheric CO2
and changes in climate are together affecting tree
growth, productivity, and thus their capacity to sequester further CO2.
Correspondence: Christopher T. Cole, e-mail:
[email protected]
2186
Quaking aspen (Populus tremuloides Michx.; here referred to simply as aspen) is a dominant forest type in
north-temperate, montane, and boreal regions of North
America. The most widely distributed tree species on
the continent, P. tremuloides is the dominant species on
about 17 million ha of Canadian forest (Peterson &
Peterson, 1992); within Wisconsin, USA, the location
of the present study, it covers ca. 1 million ha (http://
www.dnr.state.wi.us/forestry/usesof/aspen.htm). Aspen
and related poplars are thus quintessential foundation
species (Ellison et al., 2005), shaping the structure and
function of the communities and ecosystems in which
they occur (Whitham et al., 2006; Schweitzer et al., 2008;
Madritch et al., 2009).
In addition to its ecological importance in native
forests, Populus shows strong responses to enriched
CO2 in greenhouse (e.g. Lindroth et al., 2001) and
open-top chamber (e.g. Curtis et al., 2000) studies. Work
r 2009 Blackwell Publishing Ltd
C O 2 A N D C L I M AT E C H A N G E I N C R E A S E A S P E N G R O W T H
at the Aspen FACE site in northern Wisconsin, USA, has
sought to assess the response of aspen genotypes to CO2
and ozone enrichment (Karnosky et al., 2003), while
parallel work at POPFACE in Italy has evaluated responses of three species of Populus to CO2 (Wittig et al.,
2005). Both studies document enhanced primary production in trees exposed to enriched CO2; in the Aspen
FACE experiment, for example, elevating CO2 by 55%
increased aspen growth rates by 40% (Karnosky et al.,
2003). These results resemble those of other temperate
forest FACE studies, showing that CO2 enhances net
primary production across a broad range of productivity levels (Norby et al., 2005).
Our ability to infer how carbon sequestration in
forests will respond to higher CO2 levels is limited,
however, by the nature and design of fumigation experiments. The FACE studies have been relatively
short-term (o10 years) and focus on forests of young
to intermediate age. They also can incorporate only a
limited number of relevant and potentially interacting
environmental factors (e.g. soil nutrient and water
availability, temperature, ozone, etc.). Finally, CO2 augmentation studies do not identify how past changes in
CO2 and climate and their interactions with other environmental variables may together influence tree
growth under competitive field conditions. Research is
needed to investigate how tree species respond to
elevated CO2 levels at larger scales and over longer
time periods.
To complement recent experimental and prospective
FACE studies, we employed a retrospective approach,
measuring tree rings and employing multivariate statistical analyses to assess how growth patterns in quaking aspen have already responded to historical shifts in
CO2 and climate. Because genetically distinct aspen
clones differ in their growth responses to experimentally elevated CO2 (Wang et al., 2000; McDonald et al.,
2002) and because growth rates in aspen are known to
increase with individual heterozygosity (Jelinski, 1993;
Mitton & Grant, 1996), we also investigated how individual heterozygosity affects aspen’s responses to
CO2 and climate. This task was complicated not only
because growth rates vary among genotypes, and
change as trees age, but also because both climates
and CO2 levels are changing, particularly in the northtemperate and boreal regions of North America (Cayan
et al., 2001). Ever since the study by LaMarche et al.
(1984) first purported to observe increased tree growth
rates in response to rising CO2 levels, a major difficulty
has been to determine whether the increases reflect
increasing CO2 or changes in climate. Such work is
further complicated by variation in other factors, including age, site (aspect, slope, soil, elevation), and
morphology (strip-bark vs. continuous cambium).
2187
Annual rings in aspen are more difficult to measure
than those of conifers, which have been the main focus
of research into the effects of past increases in CO2 on
tree growth. Nevertheless, aspens are well-suited for
this work both because the species is known to respond
strongly to augmented CO2 and moisture (Fralish &
Loucks, 1975; Girardin & Tardif, 2005), and because its
clonal growth form allows one to measure multiple
ramets (trees) within each genet (clone). In addition,
P. trichocarpa is the first tree species to have its genome
sequenced (Tuskan et al., 2006), providing an abundance
of genomic information with which to assess individual
variation.
Given that this major forest species shows strong,
genotype-specific responses to elevated CO2, in this
work we ask whether past increases in ambient CO2
have affected the growth of quaking aspen, and how
rising CO2 may have interacted with moisture availability. We further ask whether age, sex, and genetic
variation play a role in this species’ response to rising
CO2. Although the study reported here encompasses
only a small portion of the total range of this
broad-ranged species, the use of wild-grown trees from
various soil types, weather patterns, and associated
vegetation, means that inferences from this work
should apply broadly.
Materials and methods
Field collections and sampling
We collected branch samples from 919 trees after leaves
dropped in the fall. These samples represent 189 genets
(clones; 5 trees clone1) at 11 sites distributed among
three regions in Wisconsin, USA (Fig. 1, Table S1 in
Supporting Information online). We sampled trees 5–76
years old (mean: 22.5) from forests dominated by aspen
and birch, noting GPS coordinates for each clone. Our
sites represent second-growth, unmanaged forests
south of the areas defoliated by forest tent caterpillars
in 1980–1982, 1989–1990, and 2001–2002. The sampled
clones were separated by tens to hundreds of meters of
other vegetation. We only sampled neighboring clones
if they were of different sexes, and we confirmed the
unique identity of each genet by comparing all multilocus genotypes within each population. We used a
10 m pole pruner to collect one branch from one tree
(ramet) of each clone and then collected buds for DNA
extraction. We recorded sex as male, female, or nonreproductive (rare for full-grown trees). We also collected increment cores and measured diameters at
breast height from the five largest trees (ramets) in each
clone (genet), avoiding any obviously damaged or
suppressed trees, in order to minimize potential
r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 2186–2197
2188 C . T . C O L E et al.
Fig. 1 Sampled locations in Wisconsin, USA. Circles mark sampled populations, open square indicates the Aspen FACE site, and
shaded area represents the range of Populus tremuloides; adapted from United States Geological Survey range map at: esp.cr.usgs.gov/
data/atlas/little/poputrem.pdf
shading effects. Aboveground growth as measured by
diameter at breast height, which directly reflects annual
rings, is very highly correlated with total biomass in
aspen (Bond-Lamberty et al., 2002).
Tree ring analysis
Cores were dried at 100 1C for ca. 48 h, then glued into
grooves in plywood so the vessels were vertical, sanded
with increasingly fine sandpaper (to 400 grit), and
stained with Fehling’s solution. Ring widths were measured under 7–30 magnification on a binocular dissecting microscope with a crosshair reticule, using a
data logger that recorded increments to 0.001 mm. For
off-center cores, the distance to the center and number
of rings were estimated by overlaying a plastic sheet
printed with circles of different radii so that the arc
segments of the rings present matched a circle on the
plastic, and dividing the radius of this circle by the
mean of the last four counted rings, to estimate the total
age of the stem. Graphs of annual ring widths for all
five ramets of a clone were plotted together and examined for anomalies that would indicate any error in
identifying a ring boundary; when any errors were
suspected, the cores were re-examined to correct or
confirm the measurements.
Microsatellite analysis
DNA purification, polymerase chain reaction (PCR)
reactions, and capillary electrophoresis are described
elsewhere (Cole, 2005). Microsatellite amplifications
were conducted on 16 loci, using previously developed
primers (van der Schoot et al., 2001; Tuskan et al., 2004,
http://www.ornl.bov/sci/ipgc/ssr-resource.htm). All
loci were polymorphic, and the level of heterozygosity
observed (Hobs) was calculated as the proportion of loci
for which each individual was heterozygous.
r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 2186–2197
C O 2 A N D C L I M AT E C H A N G E I N C R E A S E A S P E N G R O W T H
Atmospheric CO2 data were obtained from the Mauna
Loa observatory records (http://cdiac.esd.ornl.gov/
trends/CO2/sio-info.htm), which have been recorded
monthly since 1958; for each year, we used the mean of
the April to September values, which span the growing
season. We obtained moisture information from the
North American summer values of the Palmer Drought
Severity Index (PDSI) (ftp.ngdc.noaa.gov/paleo/
drought/pdsi2004/data-by-gridpoint) for locations
No. 198 (92.51W, 45.01N), 207 (90.01W, 45.01N), and
208 (90.01W, 428142.5 0 N). Mean annual precipitation at
these sites ranges from 81 to 86 cm (http://
www.crh.noaa.gov/mkx/climate/wipcpn.gif). Because
the three regions are roughly equidistant between these
three locations, we used the mean PDSI value calculated
from the two adjacent PDSI sites. Thus, the central
Lakes Coulee population (91.161W, 44.161N) used the
mean PDSI values from sites 198 and 207. PDSI values
for these three regions are highly similar (data not
shown). Temperature during the growing season was
not explicitly included in the model (except indirectly,
as it contributes to the Palmer Drought Severity Index),
since summer temperatures during the 1950–2000 period have not changed significantly (Feng & Hu, 2004).
We were able to analyze the separate and joint effects
of CO2 and moisture (PDSI) on tree growth because they
have varied in different ways in recent decades. That is,
while CO2 has increased monotonically, moisture (PDSI)
has varied widely among years. Thus, consecutive years
could experience an increase in CO2 but either a decrease or an increase in PDSI. In addition, over time,
years occurred with the same PDSI but very different
CO2 levels (the frequency of observations at different
levels of and PDSI is shown in Fig. S1). This asynchrony
between the PDSI and CO2 trends produced a statistical
independence that gave the model statistical strength.
etc.). Thus, ring width is composed of an overall value
common to all measurements, with an adjustment for
each genet and for each ramet within that genet. This
produced an age-specific growth curve (Fig. 2). The
model then analyzes how this age-specific growth
would be affected by each of the other factors analyzed,
alone and in combination. The final form of the statistical model was developed primarily through likelihood ratio testing of model components. Competing
model formulations were also compared using the
Akaike Information Criterion (AIC), the Bayesian Information Criterion (BIC), and examination of model
residuals. We further tested the model by randomly
subdividing the dataset into ‘training’ and ‘validation’
subsets, and estimating the residual error between the
subsets as well as the AIC values for different levels of
model complexity; this random subdivision was replicated 10 times (details and results in online Supporting
Information). The predictive strength of the model, as
measured by these criteria, greatly increased when the
model included higher-order functions of CO2 and
PDSI, as well as their interactions. The model was also
substantially strengthened by incorporating an autocorrelation function [AR(3)], which accounts for the effects
of changing environmental conditions persisting
through subsequent years.
The general form of the model used is
yij ¼ fðHobs ; CO2 ; PDSI; yÞ þ gðAge; a; b; g; d0 Þ
þ hðsex; region; yÞ þ eij ;
2.0
1.8
Ring width (mm)1/2
CO2 and climate
Statistical model and analysis
Our modeling process sought to determine the independent and interactive effects of each of the factors
listed above. As a response variable, we used (ring
width)1/2 to stabilize the small amount of heteroskedasticity in the ring width measures. We modeled this
response for each ramet (tree) in a nonlinear, mixed
effects model, i.e. one that allows for factors to have
both fixed effects (having the same magnitude for all
individuals) as well as random effects, distributed as
random variates among individuals. In this way, we
accounted for the strong and highly nonlinear effect of
age on growth, as well as variation arising from both
genetic and environmental sources (e.g. soil type, slope,
2189
1.6
1.4
1.2
0
10
20
30
Age (years)
40
Fig. 2 Age-specific growth of quaking aspen trees. Highly nonlinear growth patterns are shown in this graph of ring width
(square root of increment in mm) vs. tree (ramet) age (years).
Data points are mean values for all trees of each age, uncorrected
for any other factors; the dashed line shows the model function
at mean values for other factors.
r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 2186–2197
2190 C . T . C O L E et al.
where yij refers to growth of an individual ramet (tree) i
in genet j during each year, and eij is random error
associated with that individual at the specified Age. The
f and g refer to functions involving Age and (CO2, PDSI,
and Hobs), all of which proved to be significant; the h
refers to functions involving sex and region (of Wisconsin), which proved to be nonsignificant. The model is a
fourth-degree polynomial with linear, quadratic, cubic,
and quartic terms, and interaction effects between CO2
and PDSI, described more fully in the Appendix A (we
did not find evidence of other significant interactions).
Other factors known to affect aspen growth (ozone,
nitrogen deposition, herbivory) could be excluded from
the analysis for a priori reasons discussed in the online
Supporting Information, along with further details of
the analysis. Preliminary linear regression analysis
identified the significant first-order effects reported
here, but the mixed-effects model approach provides
more accurate expression of nonlinear response patterns, components of variation, and predictor effects.
Overall growth increase during 1958–2003 (the period
for which CO2 data are available) was estimated using
ring widths, their square roots, and the model described
above. In this latter case, age was held constant, and
default values for categorical variables ‘sex’ and ‘region’ were used, along with mean values for Hobs and
PDSI; the model was then evaluated for mean summer
CO2 levels of 1958 and 2003 (316 and 376 ppm, respectively).
affected growth: age, individual heterozygosity (Hobs),
CO2, and moisture (PDSI).
Growth rate varies strongly and nonlinearly with age,
peaking at about age 9 and declining thereafter (Fig. 2).
Age-specific ring width increased over time, and the
greatest increase occurred for relatively young trees, so
that young trees grow faster in recent years than did
young trees several decades ago. For example, during
the past five decades, growth of trees 11–20 years old
increased by 60% (Table S2). The overall increase shown
in Fig. 3 cannot be explained by age of the trees
sampled, since the average age of the trees increases
during most of the time period sampled (data not
shown). Estimates of changes in growth rates for older
trees are strongly influenced by the small number of
trees present at the beginning of the period studied [e.g.
only one tree (ramet) was 30 years old by 1960; Table
S2].
Clonal genotype (specifically individual-level heterozygosity, Hobs) has the simplest effect on aspen growth
rate (Fig. 4). As in previous studies with isozymes in
aspen (Mitton & Grant, 1980; Jelinski, 1993; Liu &
Furnier, 1993), microsatellite loci reveal high levels of
genetic diversity at the population level, as well as
individual heterozygosity (Hobs; mean 5 0.410; range:
0.125–0.688). Individual heterozygosity had a simple,
linear effect enhancing tree growth in these natural
stands (P 5 0.038). While the data show high levels of
2.5
Results
Ring width (mm)1/2
Aspen growth rates have accelerated markedly since
1935 (Fig. 3). Four factors in the model significantly
Mean ring width (mm)
5.0
4.0
3.0
2.0
1.5
2.0
1.0
0.0
1935
1945
1955
1965 1975
Year
1985
1995
0.1
2005
Fig. 3 Mean ring width over time. The overall increase in ring
width is evident even in the raw data, as are the effects of dry
(e.g. 1988) and wet (e.g. 1993) years. The number of trees
included in the data set increases with time, as does the mean
age of trees in the sample.
0.2
0.3
0.4
H OBS
0.5
0.6
0.7
Fig. 4 Individual heterozygosity (Hobs) increases growth rates.
Data points are mean ring widths (square root transform) for
each of 189 aspen genets, averaged across all ages and years for
all ramets. The solid line indicates the significant (P 5 0.0376)
linear effect of HObs on aspen growth.
r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 2186–2197
C O 2 A N D C L I M AT E C H A N G E I N C R E A S E A S P E N G R O W T H
variability arise from other factors (including age, CO2
level, and moisture, as well as other factors not measured), the effect of Hobs was significant in all versions
of the model tested. We found no evidence of interactions between heterozygosity and the other factors
affecting growth.
The effects of rising CO2 are especially strong as
reflected in Figs 5 and 6, and the model coefficients
(Table 1). Note that although many y coefficients are
small, they are multiplied by one or both factors in the
model raised to a second, third, or fourth power, so
their combination makes a major and significant con-
(a)
Ring width (mm)1/2
1.0
0.8
0.6
0.4
0.2
37
0.0
35
m)
2
M
ois
0
tu
re
33
DS
I)
–2
(P
31
5
5
CO 2
5
5
(pp
(b)
Ring width (mm)1/2
1.0
0.8
0.6
0.4
0.2
37
0.0
35
m)
2
M
ois
0
tu
re
33
DS
I)
–2
(P
31
5
5
CO 2
5
5
(pp
Fig. 5 Combined effects of CO2 and moisture on aspen growth.
CO2 is expressed in parts per million (ppm), moisture is given as
the Palmer Drought Severity Index (PDSI), and ring width is
shown in mm1/2. Although rising CO2 levels allow greater
water-use efficiency, growth promotion is particularly evident
at high moisture levels. Figure 5a shows the response surface for
the complete, fourth-order model; Fig. 5b shows the basic effects
of the linear components of the model
2191
tribution to the model (Fig. S6). Growth responses to
CO2 interact strongly with moisture (Po0.001, Table 1),
reflecting greater growth increases at higher moisture
levels. As shown in Fig. 5, rising CO2 causes ring width
to increase at all moisture levels, apparently resulting
from improved water use efficiency, but the overall
increase shown in Fig. 3 results from historical increases
in both CO2 and water availability. For a tree of average
age (22 years in this set), the effect of rising CO2 has
been to increase ring width by about 53%. This change
corresponds to a 19.2% increase in ambient CO2 levels
during the growing season, from 315.8 mL L1 in 1958
(when CO2 records began) to 376.4 mL L1 in 2003, a
period that spans the ages of all but 12 of the 919 trees
studied.
While the effects of individual factors are quantified
in Table 1, their joint effects may be better conveyed
graphically. For example, Fig. 6a illustrates the same
historical record of average ring width over time shown
in Fig. 3, along with a curve illustrating the ring width
that would be predicted by a model that included only
the age of the trees sampled at each time. Figure 6b
illustrates the ring width that would be predicted when
the suite of minor factors is added (i.e. sex, region, and
Hobs) – factors that do not change over time. Since these
constant factors do not increase the predictive power of
the model, their addition increases the AIC value
slightly. Figure 6c shows the dramatic increase in predictive power when moisture (PDSI) is added to the
model– an effect particularly noticeable in extremely
dry and wet years (see Supporting Information Fig. S3).
The model illustrated in Fig. 6c also includes the nonlinear functions of PDSI, both singly and interacting
with CO2. Finally, Fig. 6d shows the improvement
arising from the complete model, which includes CO2
along with the factors previously mentioned. Marked
declines in the AIC values, inset in each panel, reflect
the substantial improvements in model strength when
PDSI and CO2 are added; relative importance of
different components of the model are also indicated
in Fig. S6.
As might be expected in such longitudinal data, we
observed significant correlation over time between the
residuals of ramets (trees). We found evidence that a
degree 3 autoregressive process (described further in
the Appendix A) was a satisfactory model for the
autocorrelation between observations within ramets.
Figure S2 shows the autocorrelation patterns for three
models of the error correlation process: AR(3), AR(1),
and a model assuming no correlation between the
errors within a ramet. The figure shows the substantial
reduction in correlation between the residuals for using
the AR(3) model formulation. These differences were
tested with likelihood ratio tests, and the AR(3) formu-
r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 2186–2197
2192 C . T . C O L E et al.
(b)
5.0
Observed mean
Age structure only
AIC=9789.4
4.5
4.0
Mean ring width (mm)
Mean ring width (mm)
(a)
3.5
3.0
2.5
2.0
1.5
Observed mean
Minor factors added
AIC=9794.5
4.5
4.0
3.5
3.0
2.5
2.0
1.5
1960
1970
1980 1990
Year
2000
(c)
1960
1970
1980 1990
Year
2000
(d)
5.0
5.0
Observed mean
PDSI added
AIC=8754.3
4.5
4.0
Mean ring width (mm)
Mean ring width (mm)
5.0
3.5
3.0
2.5
2.0
1.5
1960
1970
1980
Year
1990
2000
Observed mean
Full model
AIC=8062.8
4.5
4.0
3.5
3.0
2.5
2.0
1.5
1960
1970
1980 1990
Year
2000
Fig. 6 Improving predictive power of the model. Solid line in each panel indicates mean ring width for aspen trees sampled; dashed
line indicates ring width that would be predicted by model components; strength of each version of the model is indicated by Akaike
Information Criterion (AIC) value inset in each panel, as factors are added to the version in the previous panel. (a) Model with age
structure only [i.e. Hobs, Palmer Drought Severity Index (PDSI), CO2, and other factors omitted]. (b) Addition of minor, constant factors
(Hobs, sex, region) makes little improvement in the model, and the AIC value increases from inclusion of additional factors. (c) Addition
of moisture (PDSI) makes a substantial improvement in strength of the model, particularly for the effects of wet and dry years. (d)
Addition of CO2 to the previous model markedly improves strength of the model.
lation was significantly better than lower order autoregressive structures.
Neither sex nor region had a significant effect on
growth. We did find that the amount of variation among
genets was about three times greater than among ramets. This result reflects both the genetic differences
among clones (genets) as well as the greater environmental uniformity found within clones relative to site
and environmental differences that exist among dispersed clones. Table 1 shows several nonsignificant
parameter estimates (y values) from the complex polynomial function for PDSI and CO2. We included nonsignificant terms in the polynomial function when these
terms belong to a parameter set that is statistically
significant by likelihood ratio testing. For example, the
parameter estimates for degree 2 degree 4 interactions
for PDSI and CO2 are not statistically significant individually (P values 0.1182 and 0.4874, respectively; see
Table 1), but a likelihood ratio test for including this set
of parameters gave a P-value of 0.06, sufficiently small
to suggest including these terms in the final model.
Discussion
Increases in both CO2 and precipitation have contributed to the higher growth rates we observe in Wisconsin
aspen (Figs 3, 5 and 6, Fig. S6). Although strong drought
conditions in some years (e.g. 1977, 1988) reduced aspen
growth, precipitation in Wisconsin has generally increased in recent decades (Feng & Hu, 2004; Fig. S3).
This increase contrasts with the prolonged drought in
the prairie provinces and northern plains states that has
contributed to widespread aspen mortality in that region (Hogg et al., 2005) and to the recent, acute drought
that has contributed to sudden aspen decline (SAD) in
the central Intermountain West (Worrall et al., 2008). The
improved growth under low moisture conditions
shown here may reflect higher water use efficiencies
in CO2-enriched atmospheres (Ceulemans & Mousseau,
1994).
The substantial, CO2-mediated growth increase identified in this work stands in contrast to previous studies
that have sought to document how rising atmospheric
r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 2186–2197
C O 2 A N D C L I M AT E C H A N G E I N C R E A S E A S P E N G R O W T H
Table 1 Fixed and random effects on growth of 919 ramets
representing 189 genets of Populus tremuloides
Standard
Parameter
Value
Error
P-value
a
b
g
d
yR1
yR2
ySM
ySF
yH1
yC1
yP1
yC2
yP2
yC3
yP3
yC4
yP4
yCP
yC1P2
yC2P1
yC1P3
yC3P1
yC4P1
yC1P4
yC2P4
yC4P2
0.17999
0.19576
0.88758
1.81726
0.04998
0.01986
0.09245
0.08840
0.26229
0.00561
0.00294
0.00012
0.02423
2.5 10–6
0.00087
5.8 10–8
0.00043
0.00620
0.00064
0.00049
0.00031
1.4 10–5
7.1 10–7
1.3 10–5
3.7 10–6
4.5 10–8
0.07654
0.00960
0.00492
0.10105
0.03128
0.05901
0.05981
0.06060
0.12613
0.00124
0.00449
8.0 10–5
0.00316
1.7 10–6
0.00121
1.0 10–8
0.00057
0.00049
0.00034
0.00004
0.00007
8.2 10–7
5.0 10–8
7.4 10–5
5.3 10–6
3.0 10–8
0.0187
o0.0001
o0.0001
o0.0001
0.1101
0.7364
0.1222
0.1447
0.0376
o0.0001
0.5128
0.1507
o0.0001
0.1286
0.4742
0.5805
0.4555
o0.0001
0.0610
o0.0001
o0.0001
o0.0001
o0.0001
0.8565
0.4874
0.1182
y coefficients from the nonlinear mixed-effects model account
for CO2, moisture, heterozygosity, and their interactions; and
for non-significant factors (region, sex). Coefficients a, b, g, and
d are used in function g describing the effect of age on growth.
Coefficients y are subscripted to indicate which factors they
refer to: R1 and R2 refer to region 1 (north) and region 2
(central), compared with region 3 (south); SM and SF refer to
male and female, compared with nonreproductive, and H1
refers to the effect of individual heterozygosity. C1–C4 are
first- through fourth-order effects of CO2, P1–P4 are firstthrough fourth-order effects of precipitation. Subscripts with
C-P- refer to the interaction between CO2 and precipitation,
with numerals indicating the level of the interactions. Nonsignificant parameters are retained when their interaction
parameters are significant. Note that while many y coefficients
are very small, in the model they are multiplied by their factor
(e.g. CO2 level) raised to a second, third, or fourth power.
CO2 levels may affect tree growth (summarized by
Huang et al., 2007). Only a few studies have attempted
to incorporate effects of both climate and rising CO2
levels on tree growth in natural systems. Among these,
work by Voelker et al. (2006), analyzing the effects of
climate (PDSI), tree age, and CO2 on oaks and pines of
the central United States, found strong evidence that the
2193
growth of these trees has also increased with rising CO2,
and these effects decline with age. These results are
similar to those we found in P. tremuloides, although the
linear regression model they used did not have the
same power to quantify the interactions among these
factors or how those interactions changed over time.
Several tree species in Europe (Körner et al., 2005) and
in the southern United States exposed to elevated CO2
show little increase in aboveground growth, perhaps
due to greater allocation to fine roots (Norby et al., 2004)
or higher rates of respiration (Feeley et al., 2007). However, aspen trees grown in the FACE experiment in
northern Wisconsin, which show a marked growth
response to CO2 enrichment, have little or no change
in their size-specific allocation to roots as a result of the
added CO2 (King et al., 2005). Also, while the length of
the growing season affects aspen growth (Yu et al.,
2001), climate records indicate that the growing season
in Wisconsin has increased only 2.5–5 days since 1951
(Feng & Hu, 2004), a change too small to account for the
large growth increases we have observed.
Most efforts to determine how rising CO2 affects tree
growth have focused on conifers in the western United
States (e.g. LaMarche et al., 1984). Such results tend to be
equivocal, possibly because annual increments are also
influenced by elevation and cambium morphology
(strip- or continuous-bark), or because they are hampered by incomplete CO2 and weather records (LaMarche et al., 1984; Graumlich, 1991; Graybill & Idso,
1993; Jacoby & D’Arrigo, 1997; Soule & Knapp, 2006).
Other studies implicating a role for CO2 have used
models to predict the response of tree growth to
changes in climate, and then noted elevated rates of
growth in recent decades that are difficult to explain in
terms of the climate models (West et al., 1993; Knapp
et al., 2001; Wang et al., 2006).
By including a 3-year autocorrelation function, our
model incorporates the effects of growth in each of the 3
previous years on growth in the current year (Fig. S2).
Although we explored using mean PDSI values from
previous years, we found the model to be stronger if it
explicitly represented the effect of previous year’s
growth using the AR(3) autocorrelation function instead. Leonelli et al. (2008), who also found that aspen
growth increases with moisture, concluded that the
strongest effect of moisture on growth came from
moisture during the previous summer rather than the
current year. Their analysis was based on monthly
weather records for the current and previous year
rather than on yearly means. We used mean values
for moisture through the growing season in order
to incorporate weather, CO2, age, genotype, and higher-order interactions; analyses including monthly
moisture were too complex. The positive response to
r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 2186–2197
2194 C . T . C O L E et al.
moisture reported here for aspen is quite different than
that of Pinus ponderosa, which showed increased growth
in response to rising CO2 on dry but not moist sites
(Soule & Knapp, 2006). Other studies of aspen, balsam
poplar, and birch suggest that broadleafed species may
be generally more sensitive to moisture and temperature than coniferous trees (Girardin & Tardif, 2005).
The gradual reduction of CO2-mediated growth enhancement (the plateau seen in Fig. 5a) could reflect a
saturating growth response to rising CO2 levels,
although trees grown under elevated CO2 at the aspen
FACE site have not shown saturation (King et al., 2005;
Kubiske et al., 2007). Alternatively, growth could become limited by some parallel, external factor such as
available N (Luo et al., 2004) or increased exposure to
O3. However, aspen trees growing at the Wisconsin
FACE site, as well as other species there and at other
forest FACE sites, have responded to elevated CO2 by
increasing their uptake of N, rather than showing
progressive N-limitation (Finzi et al., 2007). Addition
of ozone to the CO2 enrichment experiments at AspenFACE has shown that O3 reduces CO2-induced growth
enhancement (King et al., 2005; Kubiske et al., 2006),
especially under dry, high-light conditions. Ozone
levels generally follow a gradient that is low in northwest Wisconsin and high in the southeast (http://www.
epa.gov/oar/oaqps/uscanpl.pdf, dnr.wi.gov/org/aw/
air/monitor/ozonetrends.html). In summer, ozone can
reach levels that affect aspen growth (Berang et al., 1991;
King et al., 2005). Although not statistically significant, the
marginally lower growth rates of aspen trees in southern
Wisconsin relative to northern Wisconsin may reflect
slightly higher ozone exposures there.
Individual heterozygosity (Hobs) had a modest but
significant effect on growth rate, even though addition
of Hobs did not improve the predictive value of the
model, since it (like the nonsignificant factors sex and
region) does not change over time. Thus, we found no
significant interactions between Hobs and those factors
that did change over time, i.e. age, CO2, and moisture.
Previous isozyme studies in wild aspen that account for
environmental factors such as age and elevation also
found higher growth in more heterozygous individuals
(Mitton & Grant, 1980; Jelinski, 1993). Although it is
possible that heterozygosity per se at (or near) isozyme
loci might increase growth, our finding that microsatellite Hobs enhances growth (Fig. 4) suggests instead
that both sets of markers reflect overall levels of inbreeding across substantial portions of the genome
(‘associative overdominance’ – Keller & Waller, 2002).
Although aspen is dioecious and cannot self-fertilize,
localized pollen and seed dispersal could cause enough
biparental inbreeding in some individuals for inbreeding effects to be expressed.
The importance of the interactive effects (CO2 PDSI)
presented here indicate that the ‘fertilization’ effect of
rising CO2 is strongly contingent on precipitation. The
higher-order terms must be viewed with some caution,
since they can be influenced by vagaries of the particular CO2 PDSI values provided historically (Fig. S1);
furthermore, they can also reflect the interaction between the growth-promoting effects of moisture and the
growth-limiting effects of cloudy days and cool temperatures, which can markedly influence aspen’s response to elevated CO2 (Kubiske et al., 2006). Even at
the linear level, though (Fig. 5b), their interaction indicates that predictions of the effects of future CO2
increases must be conditioned by moisture availability
– a conclusion underscored by the extensive mortality
of aspens in western North America, which have experienced the same CO2 increase as the trees studied
here. These results also suggest the importance of extending this kind of analysis to cover greater changes in
moisture, over greater geographical and temporal
scales. Because both age and moisture influence responses to CO2 enrichment, large-scale common-garden
work (such as the Aspen FACE project) should be
extended to cover greater ranges of both climatic (especially moisture) flux as well as tree age, to provide more
robust predictions of how aspen-dominated forests will
respond to future CO2 increases. Finally, the magnitude
of the growth increase uncovered by this analysis raises
the question of how much other major forest species
have responded to the joint effects of long-term changes
in CO2 and precipitation.
Quaking aspen constitutes a major foundation species
in forest ecosystems throughout much of North America, influencing plant, herbivore, and decomposer communities. The species contributes substantially to the
productivity of northern temperate forests, which has
increased over the past half-century (Myeni et al., 2001;
Boisvenue & Running, 2006). Several indirect estimates
of forest growth rates in the eastern United States
suggest that this increase arises primarily from changes
in land use (e.g. afforestation), while the effects of rising
CO2 have been modest (Schimel et al., 2000) or virtually
nil (Casperson et al., 2000). The more direct analysis
presented here reveals that rising concentrations of CO2
during the past five decades have already strongly
influenced growth rates of this major component of
North American forests, especially under high-moisture
conditions.
Acknowledgements
We thank E. Rodne-Cole and B. Cole for help with field work;
D. Schlatter for help with ring measurements; P. Wyckoff and
T. Givnish for use of their laboratory equipment; and S. DiFazio,
r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 2186–2197
C O 2 A N D C L I M AT E C H A N G E I N C R E A S E A S P E N G R O W T H
C. Lorimer, T. Sharkey, and G. Slavov for comments on the
manuscript. Financial assistance was provided by a National
Science Foundation Research Opportunity Award to C. T. C. and
R. L. L (supplement to DEB-0074427), sabbatical support from
the University of Minnesota, Morris, and the Morris Academic
Partners program.
References
Berang P, Karnosky DF, Bennett JP (1991) Natural selection for ozone
tolerance in Populus tremuloides: an evaluation of nationwide trends.
Canadian Journal of Forest Research, 21, 1091–1097.
Boisvenue C, Running SW (2006) Impacts of climate change on natural
forest productivity– evidence since the middle of the 20th century.
Global Change Biology, 12, 862–882.
Bonan GB (2008) Forests and climate change: forcings, feedbacks, and the
climate benefit of forests. Science, 320, 1444–1449.
Bond-Lamberty B, Wang C, Gower ST (2002) Aboveground and belowground biomass and sapwood area allometric equations for six boreal
tree species of northern Manitoba. Canadian Journal of Forest Research,
32, 1441–1450.
Casperson JP, Pacala SW, Jenkins JC, Hurtt GC, Moorcroft PR, Birdsey RA
(2000) Contributions of land-use history to carbon accumulation in US
forests. Science, 290, 1148–1151.
Cayan DR, Kammerdiener SA, Dettinger MD, Caprio JM, Peterson DH
(2001) Changes in the onset of spring in the western United States.
Bulletin of the American Meterological Society, 82, 399–415.
Ceulemans R, Mousseau M (1994) Effects of elevated atmospheric CO2 on
woody plants. New Phytologist, 127, 425–446.
Cole CT (2005) Allelic and population variation of microsatellite loci in
aspen (Populus tremuloides). New Phytologist, 167, 155–164.
Curtis PS, Vogel CS, Wang XZ et al. (2000) Gas exchange, leaf nitrogen,
and growth efficiency of Populus tremuloides in a CO2-enriched atmosphere. Ecological Applications, 10, 3–17.
Ellison AM, Bank MS, Clinton BD et al. (2005) Loss of foundation species:
consequences for the structure and dynamics of forested ecosystems.
Frontiers in Ecology and the Environment, 3, 479–486.
Feeley KJ, Wright SJ, Nur Supardi MN, Kassim AR, Davies SJ (2007)
Decelerating growth in tropical forest trees. Ecology Letters, 10, 461–469.
Feng S, Hu Q (2004) Changes in agro-meterological indicators in the
contiguous United States: 1951–2000. Theoretical and Applied Climatology,
78, 247–264.
Finzi AC, Norby RJ, Calfapietra C et al. (2007) Increases in nitrogen uptake
rather than nitrogen-use efficiency support higher rates of temperate
forest productivity under elevated CO2. Proceedings of the National
Academy of Science, USA, 104, 14014–14019.
Fralish JS, Loucks OL (1975) Site quality evaluation models for aspen
(Populus tremuloides Michx.) in Wisconsin. Canadian Journal of Forest
Research, 5, 523–528.
Girardin M-P, Tardif J (2005) Sensitivity of tree growth to the atmospheric
vertical profile in the Boreal Plains of Manitoba, Canada. Canadian
Journal of Forest Research, 35, 48–64.
Graumlich LJ (1991) Subalpine tree growth, climate, and increasing CO2:
an assessment of recent growth trends. Ecology, 72, 1–11.
Graybill DA, Idso SB (1993) Detecting the aerial fertilization effect of
atmospheric CO2 enrichment in tree-ring chronologies. Global Biogeochemical Cycles, 7, 81–95.
Hogg EH, Brandt JP, Kochtubajda B (2005) Factors affecting interannual
variation in growth of western Canadian aspen forests during 1951–
2000. Canadian Journal of Forest Research, 35, 610–622.
Huang J-G, Bergeron Y, Denneler B, Berninger F, Tardif J (2007) Response
of forest trees to increased atmospheric CO2. Critical Reviews in Plant
Science, 26, 265–283.
2195
Jacoby GC, D’Arrigo RD (1997) Tree rings, carbon dioxide, and climate
change. Proceedings of the National Academy of Science, USA, 94,
8350–8353.
Jelinski DE (1993) Associations between environmental heterogeneity,
heterozygosity, and growth rates of Populus tremuloides in a cordilleran
landscape. Arctic and Alpine Research, 25, 183–188.
Karnosky DF, Zak DR, Pregitzer KS et al. (2003) Tropospheric O3 moderates responses of temperate hardwood forests to elevated CO2: a
synthesis of molecular to ecosystem results from the Aspen FACE
project. Functional Ecology, 17, 289–304.
Keller LF, Waller DM (2002) Inbreeding effects in wild populations. Trends
in Ecology and Evolution, 17, 230–241.
King JS, Kubiske ME, Pregitzer KS et al. (2005) Tropospheric O3 compromises net primary production in young stands of trembling aspen,
paper birch and sugar maple in response to elevated CO2. New
Phytologist, 168, 623–636.
Knapp PA, Soule PT, Grissino-Mayer HD (2001) Detecting potential
regional effects of increased atmospheric CO2 on growth rates of
western juniper. Global Change Biology, 7, 903–917.
Körner C, Asshoff R, Bignucolo O et al. (2005) Carbon flux and growth in
mature deciduous forest trees exposed to elevated CO2. Science, 309,
1360–1362.
Kubiske ME, Quinn VS, Heilman WE et al. (2006) Interannual climatic
variation mediates elevated CO2 and O3 effects on forest growth. Global
Change Biology, 12, 1054–1068.
Kubiske ME, Quinn VS, Marquardt PE, Karnosky DF (2007) Effects of
elevated atmospheric CO2 and/or O3 on intra- and interspecific competitive ability of aspen. Plant Biology, 9, 342–355.
LaMarche VC Jr, Graybill DA, Fritts HC, Rose MR (1984) Increasing
atmospheric carbon dioxide: tree ring evidence for growth enhancement in natural vegetation. Science, 225, 1019–1021.
Leonelli G, Denneler B, Bergeron Y (2008) Climate sensitivity of trembling
aspen radial growth along a productivity gradient in northeastern British
Columbia, Canada. Canadian Journal of Forest Research, 38, 1211–1222.
Lindroth RL, Roth S, Nordheim EV (2001) Genotypic variation in response of quaking aspen (Populus tremuloides) to atmospheric CO2
enrichment. Oecologia, 126, 371–379.
Liu Z, Furnier GR (1993) Comparison of allozyme, RFLP, and RAPD
markers for revealing genetic variation within and between trembling
aspen and bigtooth aspen. Theoretical and Applied Genetics, 87, 97–105.
Luo Y, Su B, Currie WS et al. (2004) Progressive nitrogen limitation of
ecosystem responses to rising atmospheric carbon dioxide. BioScience,
54, 731–739.
Madritch MD, Greene SG, Lindroth RL (2009) Genetic mosaics of ecosystem
functioning across aspen-dominated landscapes. Oecologia, 160, 119–127.
McDonald EP, Kruger EL, Riemenschneider DE, Isebrands JG (2002)
Competitive status influences tree-growth responses of Populus tremuloides to elevated atmospheric CO2 and O3 in aggrading aspen stands.
Functional Ecology, 16, 792–801.
Mitton JB, Grant MC (1980) Observations on the ecology and evolution of
quaking aspen, Populus tremuloides, in the Colorado Front Range.
American Journal of Botany, 67, 202–209.
Mitton JB, Grant MC (1996) Genetic variation and the natural history of
quaking aspen. BioScience, 46, 25–31.
Myeni RB, Dong J, Tucker CJ et al. (2001) A large carbon sink in the woody
biomass of Northern forests. Proceedings of the National Academy of
Science, USA, 98, 14784–14789.
Norby RJ, DeLuci EH, Gielen B et al. (2005) Forest response to elevated
CO2 is conserved across a broad range of productivity. Proceedings of the
National Academy of Science, USA, 102, 18052–18056.
Norby RJ, Ledford J, Reilly CD, Miller NE, O’Neill EG (2004) Fine-root
production dominates response of a deciduous forest to atmospheric
CO2 enrichment. Proceedings of the National Academy of Science, USA,
101, 9689–9693.
r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 2186–2197
2196 C . T . C O L E et al.
Peterson EB, Peterson NM (1992) Ecology, management, and use of aspen and
balsam poplar in the prairie provinces, Canada. Special Report 1. Forestry
Canada, Northwest Region, Northern Forestry Centre, Edmonton, Alberta.
Schimel D, Melillo J, Tian H et al. (2000) Contribution of increasing CO2
and climate to carbon storage by ecosystems in the United States.
Science, 287, 2004–2006.
Schweitzer JA, Madritch MD, Bailey JK et al. (2008) The genetic basis of
condensed tannins and their role in nutrient regulation in a Populus
model system. Ecosystems, 11, 1005–1020.
Soule PT, Knapp PA (2006) Radial growth increases in naturally occurring
ponderosa pine trees: a late-20th century CO2 fertilization effect? New
Phytologist, 171, 379–390.
Stephens BB, Gurney KR, Tans PP et al. (2007) Weak northern and strong
tropical land carbon uptake from vertical profiles of atmospheric CO2.
Science, 316, 1732–1735.
Tuskan GA, DiFazio S, Jansson S et al. (2006) The genome of black
cottonwood, Populus trichocarpa. Science, 313, 1596–1604.
Tuskan GA, Gunter LE, Yang ZK, Yin TM, Sewell MM, DiFazio SP (2004)
Characterization of microsatellites revealed by genomic sequencing of
Populus trichocarpa. Canadian Journal of Forest Research, 34, 85–93.
van der Schoot J, Popsikova M, Vosman B, Smulders MJM (2001) Development and characterization of microsatellite markers in black poplar
(Populus nigra L.). Theoretical and Applied Genetics, 101, 317–322.
Voelker SL, Muzika R-M, Guyette RP, Stambaugh MC (2006) Historical
CO2 growth enhancement declines with age in Quercus and Pinus.
Ecological Monographs, 76, 549–564.
Wang GG, Chhin SS, Bauerle WL (2006) Effect of natural atmospheric CO2
fertilization suggested by open-grown white spruce in a dry environment. Global Change Biology, 12, 601–610.
Wang XZ, Curtis PS, Pretziger KS, Zak DR (2000) Genotypic variation in
physiological and growth responses of Populus tremuloides to elevated
atmospheric CO2 concentration. Tree Physiology, 20, 1019–1028.
West DC, Doyle TW, Tharp ML, Beauchamp JJ, Platt WJ, Downing DJ
(1993) Recent growth increases in old-growth longleaf pine. Canadian
Journal of Forest Research, 23, 846–853.
Whitham TG, Bailey JK, Schweitzer JA (2006) A framework for community and ecosystem genetics: from genes to ecosystems. Nature Reviews
Genetics, 7, 510–523.
Wittig VE, Bernacchi CJ, Zhu X-G et al. (2005) Gross primary production is
stimulated for three Populus species grown under free-air CO2 enrichment from planting through canopy closure. Global Change Biology, 11,
644–656.
Worrall JJ, Egeland L, Eager T, Mask RA, Johnson EW, Kemp PA,
Shepperd WD (2008) Rapid mortality of Populus tremuloides in southwestern Colorado, USA. Forest Ecology and Management, 255, 686–696.
Yu Q, Tigerstedt PMA, Haapanen M (2001) Growth and phenology of
hybrid aspen clones (Populus tremula L. Populus tremuloides Michx.).
Silva Fennica, 35, 15–25.
Appendix A: statistical model development
We analyzed the data using a nonlinear mixed effects model that
accounted for the five samples from each genet, nonlinear
relationships between age and growth rate, and the gradual
increase in average tree age that occurs toward the end of the
sampled time period. This model accommodates factors that
are extrinsic or intrinsic, and fixed or varying, as well as
autocorrelation among years. Analyzing growth patterns via
individual annual rings allows us to remove the confounding
effects of age-specific changes in growth when exploring
interannual changes in the effects of genotype, environment,
and their interactions.
The model accommodates the inherently nonlinear relationship
between age and ring width characteristic of trees as well as
the multiple samples (five) from each genet. In this model, the
response variable yij is (ring width)1/2 (the transformation
stabilizes heteroskedasticity in the ring width measures). The
statistical model used is
yij ¼ fðHobs ; CO2 ; PDSI; yÞ þ gðAge; a; b; g; d0 Þ
þ hðsex; region; yÞ þ eij ;
where yij refers to (ring width)1/2 for an individual ramet (tree)
i in genet (clone) j during each year, and eij is random error
associated with that individual at the specified Age, assumed
to be a random variable with mean zero and standard deviation s. Hobs is the observed heterozygosity as described above.
The f and g refer to functions involving Age and (CO2, PDSI,
and Hobs). Specifically, the function f is defined as
fðHobs ; CO2 ; PDSI; yÞ ¼ yH1 Hobs þ yC1 CO2 þ yP1 PDSI
þ yCP CO2 PDSI þ yC2 ðCO2 Þ2
þ yP2 ðPDSIÞ2 þ yC1P2 CO2 ðPDSIÞ2
þ yC2P1 ðCO2 Þ2 PDSI þ yC3 ðCO2 Þ3
þ yP3 ðPDSIÞ3 þ yC1P3 CO2 ðPDSIÞ3
þ yC3P1 ðCO2 Þ3 PDSI þ yC4 ðCO2 Þ4
þ yP4 ðPDSIÞ4 þ yC1P4 CO2 ðPDSIÞ4
þ yC4P1 ðCO2 Þ4 PDSI þ yC2P4 ðCO2 Þ2
ðPDSIÞ4 þ yC4P2 ðCO2 Þ4 ðPDSIÞ2 :
This function is a fourth-degree polynomial with linear, quadratic, cubic, and quartic terms, and interaction effects between
CO2 and PDSI. We did not find evidence of other significant
interactions. Higher-degree terms were included if statistically
significant by likelihood ratio testing, and lower order terms
were kept in the model unless a set of terms was insignificant
by likelihood ratio testing. For example, under this system,
second degree by fourth degree interaction terms for PDSI and
CO2 were included because the joint likelihood ratio test for
these terms gave a P-value near statistical significance, approximately 0.06, even though the individual terms were not
statistically significant individually (P-values 0.1182 and
0.4874, respectively; Table 1). The second degree by third
degree interaction terms were not near statistical significance
in a likelihood ratio test, and were not included in our final
model. Individual parameter estimates are difficult to interpret in this complex model, but each term contributes to the
joint impact of PDSI and CO2 on growth as illustrated in Figs 5
and 6.
The function g, which represents the highly nonlinear change
of growth rate with age, is defined as
gðAge; a0 ; b; g; d0 Þ ¼ d0 þ ða þ b AgeÞgAge ;
where dN 5 d 1 di 1 dj, di represents a random variable that is a
random adjustment for the ith ramet, and dj represents a
random adjustment for the jth genet. In this model, d is a
fixed effect that is common for all observations, and the d
values are random effects adjusting the d parameter for each
ramet and genet. This parameter shifts the ring width up or
r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 2186–2197
C O 2 A N D C L I M AT E C H A N G E I N C R E A S E A S P E N G R O W T H
down, varying among stems. Thus ring width is composed of
an overall value common to all measurements (d), with
adjustment for each genet (dj) and each ramet (di). Similarly,
a 0 5 a 1 aj, where a represents the fixed effect for all observations and aj represents variation among genet values for the a
parameter. These a 0 parameters control the amount of curvature in the age-specific growth curve. In our investigation, we
did not find significant variation in the a parameter among
ramets within genets. The random effects defined here for
each ramet and genet can be assumed to have a general
correlation pattern that includes independence as a possible
structure.
The other parameters in this model, b and g, are fixed effects that
hold for all observations in the study. In other words, the
model says that the growth pattern across age follows the
same basic pattern with respect to the parameters controlled
by b and g, but variation among ramets and genets determines
how the a and d parameters affect individual growth curves.
Function h accounts for the effects of sex of each genet (staminate, pistillate, or nonreproductive) and the region where each
population was located (north, central, or south Wisconsin,
listed in Table S1), and is defined as
hðsex; region; y ¼ ySM SXM þ ySF SXF
þ yR1 REG1 þ yR2 REG2
where variables SXM and SXF are indicator variables taking
values of 0 or 1 depending on the sex of the genet: SXM 5 1 for
males, else 0, and SXF 5 1 for females, else 0. Thus the few
nonreproductive genets have values of 0 for both indicators,
and the values of ySM and ySF reflect gender-specific growth
rates relative to those of nonreproductive trees. Similar indicator variables represent regions: REG1 5 1 for Region 1
(southern Wisconsin), else 0, and REG2 5 1 for Region 2
(central Wisconsin), else 0. These y coefficients measure the
impact of region on ring width. For example, yR1 measures the
impact of growing in Region 1 compared with Region 3
(northern Wisconsin), after adjusting for the other terms in
the model. A similar interpretation applies for yR2, the impact
arising from growing in region 2 compared with region 3.
Because ring width observations are made over time, they form a
time series record. In many applications similar to this, and in
these data, we observe correlations between observations
taken near each other in time. For this reason, we have
r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 2186–2197
2197
modified the error correlation structure to allow correlation
between ring width measurements. Several correlation structures were tested, and a third-order autoregressive process
[AR(3)] seems to give a reasonable representation of the
correlation pattern. An AR(3) correlation structure implies
that for a particular ramet, the ring width error et, at time t,
can be written as
et ¼ j1 et1 þ j1 et2 þ j1 et3 þ at
where at is a pure noise term with mean zero, and independent
of previous observations. The data suggest such a structure is
reasonable; in turn this means errors in previous periods carry
some information about the error term in the current period. In
terms of aspen biology, this suggests several possibilities, one
being that growing conditions during one year have effects on
growth for several subsequent years, from the storage of
photosynthates, production of vegetative buds, etc.
Supporting Information
Additional Supporting Information may be found in the
online version of this article:
Figure S1. Sunflower plot of frequency of observations.
Figure S2. Residuals from autoregression functions.
Figure S3. Annual moisture changes over time in the region
studied.
Figure S4. Age-specific growth rates over time.
Figure S5. Model residuals by year for different age groups.
Figure S6. Relative strength of different versions of the
model.
Table S1. Sample locations and sizes.
Table S2. Mean ring widths for trees of different ages over
time.
Table S3. Model testing by subdivision into ‘training’ and
‘validation’ subsets.
Appendix S1. Supplementary analytical methods and
modeling information.
Please note: Wiley-Blackwell are not responsible for the
content or functionality of any supporting materials supplied
by the authors. Any queries (other than missing material)
should be directed to the corresponding author for the article.