Long-term drivers of forest composition in a boreonemoral region: the relative importance of climate...
Transcript of Long-term drivers of forest composition in a boreonemoral region: the relative importance of climate...
ORIGINALARTICLE
Long-term drivers of forest compositionin a boreonemoral region: the relativeimportance of climate and humanimpactTriin Reitalu1*, Heikki Sepp€a2, Shinya Sugita3, Mihkel Kangur3,
Tiiu Koff3, Eve Avel3, Kersti Kihno4,5, J€uri Vassiljev1, Hans Renssen6,
Dan Hammarlund7, Maija Heikkil€a8,9,10, Leili Saarse1,
Anneli Poska1 and Siim Veski1
1Institute of Geology, Tallinn University of
Technology, EE-19086, Tallinn, Estonia,2Department of Geosciences and Geography,
University of Helsinki, FI-00014, Helsinki,
Finland, 3Institute of Ecology, Tallinn
University, EE-10120, Tallinn, Estonia,4Institute of Ecology and Earth Sciences,
University of Tartu, EE-50411, Tartu,
Estonia, 5Institute of History, Tallinn
University, EE-10130, Tallinn, Estonia,6Department of Earth Sciences, VU
Amsterdam, NL-1081HV, Amsterdam, The
Netherlands, 7Department of Geology, Lund
University, SE-22362, Lund, Sweden, 8Centre
for Earth Observation Science, Department of
Environment and Geography, University of
Manitoba, Winnipeg, MB, R3T 2N2, Canada,9Freshwater Institute, Fisheries and Oceans
Canada, Winnipeg, MB, R3T 2N6, Canada,10Department of Environmental Sciences,
ECRU, FI-00014 University of Helsinki,
Helsinki, Finland
*Correspondence: Triin Reitalu, Institute of
Geology, Tallinn University of Technology,
Ehitajate tee 5, EE-19086 Tallinn, Estonia.
E-mail: [email protected]
ABSTRACT
Aim To assess statistically the relative importance of climate and human
impact on forest composition in the late Holocene.
Location Estonia, boreonemoral Europe.
Methods Data on forest composition (10 most abundant tree and shrub taxa)
for the late Holocene (5100–50 calibrated years before 1950) were derived from
18 pollen records and then transformed into land-cover estimates using the
REVEALS vegetation reconstruction model. Human impact was quantified with
palaeoecological estimates of openness, frequencies of hemerophilous pollen
types (taxa growing in habitats influenced by human activities) and
microscopic charcoal particles. Climate data generated with the ECBilt-CLIO-
VECODE climate model provided summer and winter temperature data. The
modelled data were supported by sedimentary stable oxygen isotope (d18O)records. Redundancy analysis (RDA), variation partitioning and linear mixed
effects (LME) models were applied for statistical analyses.
Results Both climate and human impact were statistically significant predic-
tors of forest compositional change during the late Holocene. While climate
exerted a dominant influence on forest composition in the beginning of the
study period, human impact was the strongest driver of forest composition
change in the middle of the study period, c. 4000–2000 years ago, when per-
manent agriculture became established and expanded. The late Holocene cool-
ing negatively affected populations of nemoral deciduous taxa (Tilia, Corylus,
Ulmus, Quercus, Alnus and Fraxinus), allowing boreal taxa (Betula, Salix, Picea
and Pinus) to succeed. Whereas human impact has favoured populations of
early-successional taxa that colonize abandoned agricultural fields (Betula,
Salix, Alnus) or that can grow on less fertile soils (Pinus), it has limited taxa
such as Picea that tend to grow on more mesic and fertile soils.
Main conclusions Combining palaeoecological and palaeoclimatological data
from multiple sources facilitates quantitative characterization of factors driving
forest composition dynamics on millennial time-scales. Our results suggest that
in addition to the climatic influence on forest composition, the relative abun-
dance of individual forest taxa has been significantly influenced by human
impact over the last four millennia.
Keywords
Anthropogenic impact, community history, Estonia, forest composition, forest
ecology, meta-analysis, millennial time-scale, palaeoecology, pollen data, varia-
tion partitioning.
ª 2013 Blackwell Publishing Ltd http://wileyonlinelibrary.com/journal/jbi 1doi:10.1111/jbi.12092
Journal of Biogeography (J. Biogeogr.) (2013)
INTRODUCTION
Forest composition changes in Europe during the Holocene
are attributable to changing climate (Birks, 1986; Huntley,
1990), climate-associated migration processes with possible
time-lags (Huntley & Birks, 1983; Svenning & Skov, 2005),
species competition and replacement (Bennett & Lamb,
1988; Sepp€a et al., 2009), and long-term soil development
(Birks, 1986). In addition, human activities have influenced
European forest composition, especially during the late
Holocene (Iversen, 1973; Behre, 1988; Bradshaw & Hannon,
1992; Keller et al., 2002; Magny et al., 2002). The impact of
growing human population and agriculture on boreonemoral
forests is clearly demonstrated by an increase of crop-fields
and other open habitats (Poska et al., 2004; Mitchell, 2005;
Overland & Hjelle, 2009; Fyfe et al., 2010). However, the
possible long-term influence of human impact on species
composition in the remaining tree-covered areas is less clear,
and the relative importance of human impact and climatic
factors in determining the late Holocene forest composition
is largely unknown. One way to assess the relative impor-
tance of climatic and anthropogenic factors on long-term
community dynamics is to use dynamic vegetation models in
combination with pollen records (Keller et al., 2002; Lischke
et al., 2002; Bradshaw, 2008; Miller et al., 2008). This model-
ling approach requires considerable knowledge about the
bioclimatic optima and tolerances, dispersal capacities and
disturbance responses of the species. Pollen records are then
used to assess simulation outcomes. This approach has been
effective in evaluating model performance in relation to past
and future climate changes; however, the incorporation of
human impacts into dynamic vegetation modelling has
hitherto been limited (but see Keller et al., 2002).
Here, we provide a novel approach for evaluating the
relative importance of climate change and human impact on
forest dynamics based on fossil pollen. We utilized recent
advances in pollen-based vegetation reconstruction and in
palaeoclimatic modelling to test statistically the significance
of climatic and anthropogenic factors on long-term vegeta-
tion dynamics in Estonia. Estonia is situated in the boreo-
nemoral zone of northern Europe and is well suited for this
study because of an exceptionally dense network of pollen
sites and a long history of human impact. Furthermore,
independent palaeoclimatological proxy records are available
from regions adjacent to Estonia. We focused on the late
Holocene (5150–50 cal. yr bp, calibrated years before 1950)
because it was the period of expanding anthropogenic land
use in Estonia (Poska & Saarse, 2002; Poska et al., 2004) and
the pollen records show that all the major tree taxa currently
growing in Estonia were present in the region over the entire
study period (Saarse et al., 1999; Saarse & Veski, 2001).
Sedimentary pollen records provide insight into past
vegetation dynamics, but interpreting these data in terms of
vegetation composition is not straightforward. Notably, the
representation of surrounding vegetation in the pollen sam-
ples depends on numerous factors such as intertaxonomic
differences in pollen productivity and dispersal, the patchi-
ness of the surrounding vegetation and the size of the sedi-
mentary basin. However, recent theoretical and modelling
advances in pollen analysis (Sugita, 2007a,b) have improved
reconstructions of regional and local vegetation composition
(Hellman et al., 2008; Nielsen & Odgaard, 2010; Sugita et al.,
2010). We used the REVEALS (regional estimates of vegeta-
tion abundance from large sites) model (Sugita, 2007a) to
reconstruct regional vegetation composition at 18 sites in
Estonia during the 5100-year study period. The REVEALS-
reconstructed tree and shrub composition was compared
with simulated and reconstructed records of climate and
human impact. The simulated temperatures were based on
the palaeoclimatic model ECBilt-CLIO-VECODE (Renssen
et al., 2009). As complementary palaeoclimatic data indepen-
dent of vegetation dynamics, we used stable oxygen isotope
records from lacustrine sediments in neighbouring regions
(Hammarlund et al., 2003; Heikkil€a et al., 2010).
The quantity of past archaeological finds has been used to
indicate the intensity of past human impact on the natural
environment (Magny et al., 2002; P€artel et al., 2007; Talla-
vaara & Sepp€a, 2011). However, such data are seldom avail-
able for larger regions and at sufficiently high temporal
resolution. In Estonia, numerous case studies show a corre-
spondence between agricultural establishment phases of
human populations as revealed by archaeological finds and
increases in landscape openness as inferred from pollen data,
particularly from the occurrence of hemerophilous (growing
in habitats influenced by human activities) and cultivated
plant taxa (e.g. Poska & Saarse, 1999; Poska et al., 2004). The
abundance of charcoal particles in lake sediments is also a
good indicator of human activities in northern Europe
(Behre, 1981; Bradshaw & Hannon, 1992; Koff et al., 1998).
Thus, we used landscape openness, percentage cover of
hemerophilous and cultivated taxa, and charcoal concentra-
tion as indicators of human impact.
This study addresses the following questions through sta-
tistical analyses of the combined long-term records of forest
composition, climate and human impact.
1. Are climate and human impact statistically significant
predictors of late Holocene forest composition change?
2. To what extent was the forest composition change during
the late Holocene influenced by climate, human impact, and
their combined effect?
3. What are the responses of individual tree and shrub taxa
to climate and human impact?
MATERIALS AND METHODS
Study area
Estonia (58–60° N, 22–28° E) is situated in the boreonemor-
al forest zone (Fig. 1). The vegetation is characterized by a
mixture of boreal coniferous and deciduous tree species
(Picea abies, Pinus sylvestris, Betula pendula and Betula pubes-
cens), with nemoral deciduous species (e.g. Corylus avellana,
Journal of Biogeographyª 2013 Blackwell Publishing Ltd
2
T. Reitalu et al.
Tilia cordata, Ulmus glabra and Quercus robur) being also
widespread but much less frequent. Climatically, Estonia lies
in the northern part of the temperate climate zone and
at the transition between moderately continental (to the
east) and slightly more maritime climate (to the west). At
present (mean values of 1971–2000), mean temperature is
16.7 °C in July and �4.7 °C in February, and annual precip-
itation is 650 mm (EMHI, 2011). Based on sedimentary
pollen records, the warmest period in the Holocene was
c. 8000–4400 years ago, when mean annual temperature was
c. 2–3 °C warmer and effective moisture was lower than
today (Sepp€a & Poska, 2004).
The earliest records of Stone Age human activities date to
11,000–10,000 cal. yr bp. The early anthropogenic impact on
vegetation was probably limited and only small areas of for-
est were cleared in the immediate surroundings of the settle-
ments (Veski et al., 2005). The earliest evidence of
cultivation in different parts of Estonia dates back to the
Neolithic (6900–3800 cal. yr bp) (Poska et al., 2004). Agri-
culture became a major source of subsistence during the
Bronze Age (3800–2500 cal. yr bp) (Lang, 2007). The study
period (5150–50 cal. yr bp) covers most of the period of
agricultural development in Estonia except for the last
100 years when rapid political and socio-economic changes
caused major land-cover shifts.
Site selection
Pollen records from small lakes and bogs reflect the local
vegetation within c. 10–100 km2 and those from large sites
reflect the regional vegetation within c. 10,000–100,000 km2
(Sugita, 1993). We focused on regional-scale changes in for-
est composition and selected pollen records from sites 250–
2800 m in radius (Fig. 1, and Appendix S1 in Supporting
Information). In two regions where there were no data
available from large sites – Viitna (site no. 18) and southern
Estonia (site no. 17) (Fig. 1) – data from small sites were
pooled by summing up the pollen counts from respective
time periods (Sugita, 2007a). In the case of site 5 (Kiilaspere/
Velise, Fig. 1), two closely situated pollen records that cover
different time periods were combined and treated as one site.
Chronologies
Based on the conventional and accelerator mass spectrometry
(AMS) radiocarbon dates, the chronologies of the selected
sequences were calculated using a standardized methodology
based on the weighted averages of the ranges of the cali-
brated radiocarbon dates using the IntCal09 calibration data
set (Reimer et al., 2009) and the OxCal 4.1 program (Bronk
Ramsey, 2009). To combine radiocarbon ages and lithologi-
cal data, sedimentary deposition models (Bronk Ramsey,
2008) were used. The 5100-year study period was divided
into seventeen 300-year time windows, which were used as
study units. All data were averaged within each of the 300-
year time windows.
REVEALS modelling
We ‘translated’ pollen count data to abundances (% cover) of
the corresponding plant taxa in the surroundings of the study
sites using the REVEALS model (Sugita, 2007a). The details of
the model are provided in Appendix S2. Similar to earlier
studies that have used the REVEALS model (e.g. Hellman
et al., 2008; Nielsen & Odgaard, 2010; Sugita et al., 2010), the
assumed extent of reconstructed vegetation was approximately
100 km 9 100 km for each site. In total, the REVEALS
reconstruction included 22 taxa (trees, shrubs and herbs) that
were abundant in our pollen records (Appendix S2).
Forest composition
Forest composition was characterized by the REVEALS abun-
dance estimates of the 10 most common shrub and tree taxa:
Figure 1 Map of the study area (Estoniawith county borders). Locations of the sites
included in the study: (1) Ermistu; (2)Kahala lake; (3) Kahala bog; (4) Kalsa; (5)
Kiilaspere/Velise; (6) Konsu; (7) Liivj€arve;(8) Maardu; (9) Mustj€arve; (10) Parika;(11) Raigastvere; (12) Ruila; (13) Saha; (14)
Saviku; (15) Surusoo; (16) Vedruka; (17)southern Estonia, combining data from
Kirikum€ae, Plaani and Verij€arv; and (18)Viitna, combining data from Viitna Linaj€arvand Viitna Pikkj€arv. Appendix S1 providesdetailed information about the pollen sites.
Journal of Biogeographyª 2013 Blackwell Publishing Ltd
3
Forest composition, climate and human impact
one shrub (Corylus), eight trees (Alnus, Betula, Fraxinus,
Pinus, Picea, Quercus, Tilia and Ulmus), and one taxon
including both shrubs and trees (Salix). This taxonomic sim-
plification was necessary because pollen identification is
mostly at the genus level (Appendix S2). The REVEALS-esti-
mated percentages of the 10 taxa were recalculated to sum to
100% to quantify the taxon composition of the forested area
instead of the entire landscape.
Indicators of human impact
To estimate the openness (OPEN), we used the sum
of REVEALS estimates of the pollen taxa that indicate non-
forested conditions (Artemisia, Filipendula, Plantago lanceola-
ta, Plantago major/media, Poaceae, Cyperaceae, Rumex
acetosa/acetosella, Juniperus and Cerealia). Because OPEN was
strongly dependent on Poaceae and Cyperaceae that are also
present in natural habitats, we used an additional variable
(HEMER) to reflect the occurrence of taxa that are more
closely related to human impact. HEMER was calculated as
the sum of REVEALS estimates of hemerophilous (Artemisia,
Plantago major/media, Plantago lanceolata, Rumex acetosa/
acetosella) and cultivated taxa (Cerealia).
Charcoal particle counts from pollen samples spiked with
Lycopodium marker grains were available from seven lakes in
Estonia (Appendix S1). In each sample, charcoal concentra-
tion was calculated as: (added Lycopodium 9 counted char-
coal)/(counted Lycopodium 9 sample volume), giving the
number of charcoal particles per cm3 of sediment. The aver-
age charcoal concentration for each of the 300-year time
windows gave an estimate of CHARC.
Climate
For the temperature changes over the study period, we used
the results from a transient simulation experiment with the
ECBilt-CLIO-VECODE climate model, forced by annually
changing values of orbital parameters, atmospheric CO2 and
CH4 concentrations, ice sheet surface albedo, topography
and melt flux. An overview of this simulation (named OG-
MELTICE) has been published by Renssen et al. (2009). The
model reconstructions cover the whole world at 5.6° 9 5.6°spatial resolution and monthly temporal resolution. The
average summer (May–August) and winter (December–Feb-
ruary) temperatures (SUMM_T and WINT_T, respectively)
for Estonia (two grid-cells) were used for the 300-year time
windows. Temperatures were expressed as the differences
from the pre-industrial mean simulated in the climate model
(250–550 cal. yr bp).
In addition to the simulated temperatures, we used stable
oxygen isotope (d18O) records from two sites near Estonia: a
lacustrine carbonate record from Lake Igelsj€on (d18O_IGEL)
in southern Sweden (Hammarlund et al., 2003; Jessen et al.,
2005) and an aquatic cellulose record from Lake Saarikko
(d18O_SAAR) in southern Finland (Heikkil€a et al., 2010). At
these two sites, d18O reflects mainly centennial-scale changes
in effective moisture. In general, high d18O values reflect
drier and warmer climate, and low d18O values reflect colder
and moister climate (Hammarlund et al., 2003; Heikkil€a
et al., 2010).
The present-day climate in Estonia is relatively maritime
in the coastal areas and more continental further inland. We
assume that this gradient has been consistent over the late
Holocene. To account for that gradient, we calculated a coef-
ficient of continentality for each pollen site based on the
present-day winter and summer temperatures. The mean
temperature of the coldest quarter of the year and the mean
temperature of the warmest quarter of the year for the pollen
sites were extracted from the WorldClim global climate data-
base (BIOCLIM database at 2.5-arc-minute resolution, http://
www.worldclim.com/; Hijmans et al., 2005). The difference
between the temperature of the warmest and coldest quarters
gave an estimate of continentality (CONT) (Tuhkanen,
1980).
Statistical analyses
For all the statistical analyses we used the 300-year time win-
dows as study units. Some pollen records did not cover the
entire 5100-year study period (Appendix S1) and altogether
the data set used for statistical analyses consisted of 277
study units (n = 277).
To evaluate general trends in the associations among forest
composition, human impact and climate, we used redun-
dancy analysis (RDA). We assigned forest composition data
as the response matrix and climate and human impact data
as constraining variables. The forest composition data were
square-root transformed prior to the RDA to ensure a more
even distribution of data in the ordination space. Signifi-
cances of the marginal effects of single constraining variables
were evaluated using Monte Carlo permutation tests (999
randomizations) and non-significant (P < 0.05) constraining
variables were removed by a backward selection procedure.
To illustrate the community change through time at individ-
ual sites, we superimposed the time-trajectories of the sites
on the RDA plot.
Variation partitioning (Borcard et al., 1992) was used to
provide insight into the relative impacts of the three groups
of variables – ‘human impact’ (OPEN, HEMER, CHARC),
‘climate’ (WINT_T, SUMM_T, d18O_IGEL, d18O_SAAR,CONT) and SITE. To understand how the relative roles of
climate and human impact change over time, we repeated
the variation partitioning analysis in subsets of the data
defined by a moving period of 1200 years (four 300-year
study units). This resulted in 14 analyses of overlapping data
sets. The significances of the individual effects of the parti-
tions were estimated with Monte Carlo permutation tests
(999 randomizations) in each of the analyses.
To explore the individual responses of different forest taxa
to the factors reflecting human impact and climate, we used
linear mixed effects (LME) models with each of the 10 taxa
(Fig. 2a) as response variables. LME models allow one to
Journal of Biogeographyª 2013 Blackwell Publishing Ltd
4
T. Reitalu et al.
estimate the effects of fixed explanatory variables (OPEN,
HEMER, CHARC, SUMM_T, WINT_T, d18O_IGEL,
d18O_SAAR, CONT) and to account for variation due to
random effects, such as among-site variability and temporal
autocorrelation (Zuur et al., 2009). The abundances of forest
taxa were square-root transformed prior to the analysis and
the explanatory variables were standardized to zero mean
and unit variance. In the random part of the models, we
included SITE to account for the fact that samples from one
site tended to be similar and temporal autocorrelation struc-
ture with autoregressive moving average (corARMA) model
where the number of autoregressive parameters and the
number of moving average parameters were both set to one.
We used a stepwise backward selection procedure with the
likelihood ratio test (Zuur et al., 2009) where, at each step,
the variable (or quadratic term) giving the highest P-value
was left out until only significant (P < 0.05) terms remained.
To evaluate the goodness of fit of the LME model, we calcu-
lated squared correlation coefficient (r2) between the
response variable and the model-estimated values.
We used the R environment (version 2.10.1) (R Develop-
ment Core Team, 2009) with the packages vegan (RDA and
variation partitioning), and nlme (LME models) for statisti-
cal analyses.
RESULTS
All environmental variables except CHARC and WINT_T were
significantly (P < 0.05) associated with forest composition in
the RDA. The triplot (Fig. 3a) indicated that the nemoral
deciduous taxa were mainly related to the climate variables;
among boreal taxa, Picea was positively associated with conti-
nentality, and Pinus, Betula and Salix were mainly related to
the indicators of human impact. Time-series of individual sites
(Fig. 3b–d, Appendix S3) showed that in the beginning of the
5100-year period, all sites were dominated by nemoral decidu-
ous taxa (Ulmus, Tilia, Corylus and Fraxinus). Thereafter,
three distinctive trajectories can be separated: (1) development
of Picea-dominated forest, followed by the dominance of
Pinus, Betula and Salix (exemplified by Kiilaspere/Velise,
Fig. 3b); (2) development and persistence of Picea-dominated
forest (exemplified by Parika, Fig. 3c); (3) development of
forest dominated by Pinus, Betula and Salix without the Picea-
dominated phase (exemplified by Surusoo, Fig. 3b). The tra-
jectories of all 18 sites are given in Appendix S3.
The results of the variation partitioning analysis showed
that the variable groups ‘climate’ (SUMM_T, WINT_T,
d18O_SAAR, d18O_IGEL, CONT) and ‘human impact’
(CHARC, OPEN, HEMER), together with SITE, explained
69.0% of the variation in the forest composition across the
whole 5100-year study period (Fig. 4a). The decomposition
of the variation showed that the largest fraction of the vari-
ability was accounted for by SITE (27.7%), followed by the
shared effect of ‘climate’ and ‘human impact’ (21.3%). The
individual effect of ‘climate’ on forest composition was
10.2% and the individual effect of ‘human impact’ was 3.8%.
Variation partitioning within the 1200-year time periods
indicated a rapid increase in the relative importance of the
%%
%%
%%
%
δ18O_SAAR
δ18O_IGEL
Alnus
Betula
Corylus
Fraxinus
Picea
Pinus
Quercus
Salix
Tilia
Ulmus
6
(a) (b)
54
3
35
25
15
12
6
321
50
40
30
21
15
9
21
21
9
15
3
9
15
3
OPEN
HEMER
CHARC
WINT_T
SUMM_T
%%
%
%
30
20
10
6
3
%
5000
-7
-9
-14
-18
0.2
0.1
0.5
0.2
°C‰
‰
5000 4000 3000 2000 1000 0Time, cal. years BP
Time, cal. years BP
35000
Particles/cm
3
CONT20
21
22
°C°C
5000 4000 3000 2000 1000 0
Figure 2 Forest composition and
explanatory variables in Estonia through
time: (a) REVEALS estimates of theabundances of forest taxa, and (b)
indicators of human impact (OPEN,HEMER, CHARC) and climate
(d18O_SAAR, d18O_IGEL, SUMM_T,WINT_T, CONT). Mean (solid lines)
and � 2 standard errors of the mean(dashed lines) characterizing the amount of
variation.
Journal of Biogeographyª 2013 Blackwell Publishing Ltd
5
Forest composition, climate and human impact
human impact from around 0% in the first time period (5150
–3950 cal. yr bp) to 14% in the fifth time period (3950–
2750 cal. yr bp) (Fig. 4b), after which the importance of
human impact slowly decreased. Climate explained 17–20% of
the variation in forest composition during the first two time
periods and < 10% during the rest of the study (Fig. 4b).
Separate LME models for each taxon produced roughly
the same results as the RDA. Nemoral deciduous taxa (Cory-
lus, Tilia, Fraxinus, Quercus and Ulmus) were positively asso-
ciated with summer temperature and d18O (Table 1). Alnus,
Quercus and Ulmus were also positively associated with land-
scape openness. Betula, Pinus and Salix were strongly posi-
RDA1
RDA2
-0.6 -0.2 0.2 0.6
-0.8
-0.4
0.0
0.4
OPEN
HEMERBetula
PinusSalix
CONT Picea
Corylus
SUMM_TUlmus
Tilia
FraxinusAlnus
RDA1
RDA2
-0.6 -0.2 0.2 0.6
-0.8
-0.4
0.0
0.4
RDA1
RDA2
-0.6 -0.2 0.2 0.6
-0.8
-0.4
0.0
0.4
RDA1
RDA2
-0.6 -0.2 0.2 0.6
-0.8
-0.4
0.0
0.4
Kiilaspere/Velise
Parika Surusoo
(a) (b)
(c) (d)
δ18O_IGELδ18O_SAAR
Quercus
Figure 3 Results of redundancy analysis
(RDA) of Estonian forest composition: (a)RDA triplot indicating the positions of
forest taxa, climate and human impactvariables and sites along the first two
ordination axes; (b–d) ecological trajectoriesof forest composition from 5150 cal. yr bp
(open circle) to 50 cal. yr bp (circle with across) at three selected sites.
3950-2750
4250-3050
4550-3350
Time period (years BP)
Explainedvariation(%)
Climate10.2 %
Site27.7 %
3.9 %1.2 %
Undeterminedvariation31.0 %
Humanimpact3.8 %
0.9 %
21.3 %
ClimateHuman impact
(a) (b)30
25
20
15
10
5
0
5150-3950
4850-3650
3650-2450
3350-2150
3050-1850
2750-1550
2450-1250
2150-950
1850-650
1550-350
1250-50
Figure 4 Partitioning of Estonian forest composition in terms of the percentage of variation explained by ‘human impact’ (HEMER,OPEN, CHARC), ‘climate’ (SUMM_T, WINT_T, d18O_SAAR, d18O_IGEL, CONT) and SITE. The results are presented (a) for the
whole 5100-year study period where all three groups of predictors are highly significant (P < 0.001), and (b) for 14 subsets of data in
1200-year periods where black symbols indicate significant (P < 0.05) and grey symbols indicate non-significant individual effects ofclimate and human impact.
Journal of Biogeographyª 2013 Blackwell Publishing Ltd
6
T. Reitalu et al.
tively associated with human impact, as shown by the posi-
tive association with the proportion of hemerophiles, with
Betula and Pinus also negatively associated with summer
temperature. The only taxon that was negatively associated
with human impact was Picea.
DISCUSSION
We demonstrated that over the 5100-year study period in
the late Holocene, climate explained about four times more
variation in forest composition in boreonemoral Estonia
than human impact. Many earlier palaeoecological studies
have qualitatively shown that the importance of human
impact on overall vegetation composition in Europe has
increased over the last few millennia with the expansion of
agriculture and forest management (e.g. Behre, 1988; Lind-
bladh et al., 2000; Poska et al., 2004; Feurdean et al., 2010).
In accordance with dynamic modelling studies (Easterling
et al., 2001; Keller et al., 2002), our analysis showed that
human activities have not only caused increased landscape
openness, but have also affected the species composition in
the remaining forests. The results of the variation partition-
ing analysis within shorter time periods revealed that human
impact became a dominating driver of forest compositional
change at 3950–1850 cal. yr bp, which is the period of estab-
lishment and expansion of permanent agriculture in Estonia
(Poska et al., 2004). Our results thus suggest that the first
opening of the landscape had effects on forest composition
exceeding the effects of increasing non-forested agricultural
land area over the last 2000 years of the study period.
In addition to the statistically significant individual effects
of climate and human impact, there is a large shared effect of
these two groups of variables on forest composition. The
shared effect can be attributed to climate influencing agricul-
ture and other human activities. In northern Europe, warmer
periods are believed to promote the expansion of agriculture
(Berglund, 2003). In our data set, however, all three indica-
tors of human impact increase when temperature decreases
(Fig. 2b), indicating that agricultural expansion occurred
irrespective of climatic trends (cf. Sillasoo et al., 2009). The
shared effect in our analysis suggests that, coincidentally, cli-
mate change and human impact have simultaneously affected
forest composition: the same tree taxa that benefited from
human disturbances (Betula, Pinus and Salix) were able to
better withstand the late Holocene cooling than the nemoral
deciduous taxa (Sepp€a et al., 2009). Similar parallel ecological
trajectories between human influence and climate have been
observed in the Mediterranean region (Jalut et al., 2009).
The RDA results (Fig. 3, Appendix S3) showed that the
forest composition was relatively homogeneous at the begin-
ning of the study period, being mainly characterized by ne-
moral deciduous taxa, such as Tilia, Ulmus, Corylus, Fraxinus
and Quercus. Around 4000 cal. yr bp, however, geographical
differences in the vegetation started to evolve. Coastal sites
in northern and western Estonia became dominated by Pi-
nus, Betula and Salix, whereas sites in southern and eastern
Estonia became first characterized by Picea and thereafter by
Pinus, Betula and Salix at about half of the sites (Fig. 3b,
Appendix S3). One possible explanation for this geographical
pattern may be related to the differences in soil properties.
The coastal areas of Estonia have calcareous bedrock and the
soils are shallow and less suitable for the growth of Picea
than the deeper soils on top of sandstone and clay in south-
ern Estonia (Laasimer, 1965). Picea is also the latest of the
Table 1 Summary of the LME models explaining the occurrences of forest taxa in Estonia over the last 5100 years. The random part of
the models includes SITE (random intercept) and temporal autocorrelation with autoregressive moving average model. In the fixed partof the models, linear terms are marked with ‘+’ for positive and ‘�’ for negative associations; quadratic terms are marked with ‘∩’ forhump-back associations and with ‘∪’ for U-shaped associations; r2 is the squared correlation between response variable and modelestimation. WINT_T and CHARC were always excluded during the backward selection procedure and are the only fixed variables not
included in any of the final models. The significances of the fixed factors are given as: ***P < 0.001; **P < 0.01; *P < 0.05.
TaxonHuman impact Climate
HEMER OPEN d18O_IGEL d18O_SAAR SUMM_T CONT r2
Corylus + *** + *** � *** 0.54
Fraxinus + ***∪ *
+ *** 0.31
Tilia + *** + *** + *** 0.48
Ulmus + *** + *** + ***∪ **
+ *** 0.55
Quercus + ** + *** + *** 0.33
Alnus + **∪ **
+ **∪ *
0.18
Picea � *** � ***∩ ***
0.22
Betula + ** + *** � ** 0.52
Pinus + **∪ *
� *** � *** 0.64
Salix + *** + ** 0.34
Journal of Biogeographyª 2013 Blackwell Publishing Ltd
7
Forest composition, climate and human impact
major tree species to immigrate to Estonia and began to
spread from the most continental part in the south-east
around 8000 cal. yr bp (Saarse et al., 1999). It is possible
that Picea never really became dominant in the coastal areas
because the relatively maritime climate favoured the nemoral
deciduous species and postponed the dominance of Picea
until the fertile areas suitable for the growth of Picea had
already become arable fields. In southern and eastern
Estonia, on the other hand, Picea had already become domi-
nant before the expansion of arable fields occurred. The sites
where Picea dominated throughout the end of the study
period (Saviku, Konsu, Ermistu and Parika; Appendix S3)
are the sites with the lowest human impact in the surround-
ings until the present day. It is likely that without human
impact, Picea would be more dominant in Estonia and else-
where in the boreonemoral zone.
The taxon-wise analyses of the 10 major tree and shrub taxa
showed that the nemoral deciduous taxa (Tilia, Corylus, Fraxi-
nus, Quercus and Ulmus) were positively associated with war-
mer (and drier) climate, as indicated by the significant
positive associations with SUMM_T and d18O_SAAR. The
decline of nemoral deciduous taxa during the late Holocene
has often been associated with increasing human impact in
central Europe (e.g. Behre, 1988; Bradshaw & Hannon, 1992;
Lindbladh et al., 2000; Cowling et al., 2001), but in Estonia
these taxa occur close to their climatic distribution limits
(Hult�en & Fries, 1986), and the influence of cooling is likely to
overshadow the influence of human impact. Sepp€a et al.
(2009) suggested that the interspecific competition associated
with the mid-Holocene expansion of Picea played a major role
in forest ecosystem change in Fennoscandia. Picea is a shade-
tolerant and competitively superior taxon, and its rapid popu-
lation rises may have suppressed nemoral deciduous taxa,
especially Tilia and Corylus. We did not explicitly test the asso-
ciations between Picea and deciduous taxa in our data but it is
notable that the initial decline of Tilia and Corylus coincided
with an increase of Picea around 4500 cal. yr bp (Fig. 2).
In addition to the strong associations with climate vari-
ables, three nemoral deciduous taxa (Quercus, Ulmus and
Alnus) were positively associated with landscape openness
(Table 1). The positive association of Quercus with open-
ness may be the result of its demand of light (Diekmann,
1996). Quercus was also preferred by human management
in the wooded meadows and wood pastures, which were
half-open seminatural habitat types common in Estonia
(covering up to 19% of the land area) and elsewhere in
northern Europe for thousands of years (Iversen, 1973;
Kukk & Kull, 1997; Fyfe et al., 2010). The declines of
Ulmus and Alnus have previously been associated with
increasing human impact in Estonia (Poska et al., 2004;
Saarse et al., 2010). Our present analysis, using multiple
linear models, however, revealed that when the climate
variables were accounted for, both Ulmus and Alnus were
more common in areas with high levels of openness. The
species of both taxa (especially A. glutinosa, A. incana and
U. laevis) are known to be associated with forest edges,
river banks and lake shores in the present-day landscape
(Kukk, 2004), which may explain their relative commonness
in more open areas. A. glutinosa and A. incana are both
early-successional species that are common invaders of
abandoned arable fields and pastures.
Contrary to the findings for the nemoral deciduous taxa
(Ulmus, Tilia, Corylus, Quercus and Fraxinus), the boreal taxa
(Picea, Pinus and Betula) were significantly negatively associ-
ated with SUMM_T and d18O_SAAR, showing that the
cooler and wetter climate has favoured these taxa. Human
impact was also significantly associated with all boreal taxa,
positively with Pinus, Salix and Betula, and negatively with
Picea. Agriculture has focused primarily on fertile and mesic
soils (Poska et al., 2004), causing the decline of Picea, and
nutrient-poor and wet soils have remained forested, explain-
ing the increasing relative abundances of Pinus, Betula and
Salix. Betula and Salix are also usual invaders of abandoned
arable fields, further increasing the relative importance of
these taxa in the agricultural landscape.
CONCLUSIONS
Although the effects of climate and human impact cannot be
fully decoupled, our results show that both climate and
human impact are statistically significant predictors of long-
term forest compositional change during the late Holocene.
While climatic influence on forest composition is dominant
in the beginning of the study period, the importance of
human impact becomes a dominating driver behind forest
compositional change in the middle of the study period
(c. 4000–2000 years ago), which is the period of establish-
ment and expansion of permanent agriculture. Late Holocene
cooling negatively affected the populations of nemoral decid-
uous taxa (Tilia, Corylus, Ulmus, Quercus, Alnus and Fraxi-
nus) and favoured the expansion of boreal taxa (Betula,
Salix, Picea and Pinus). Human impact in Estonia favoured
early-successional fast-growing taxa that can colonize aban-
doned fields (Betula, Salix and Alnus) or can grow on less
fertile soils (Pinus). Increased landscape openness and possi-
bly selective logging have favoured light-demanding taxa
such as Quercus.
Our study demonstrates that meta-analysis of palaeoeco-
logical and palaeoclimatic data has the potential to provide
an alternative and comparative approach to dynamic vegeta-
tion modelling in characterizing long-term ecological pro-
cesses. Compiling a large data set of pollen records helps to
overcome inter-site variation in pollen data and enables the
evaluation of general regional trends in vegetation types and
species composition. Combining palaeoecological and palaeo-
climatic data with statistical analyses allows us to reveal asso-
ciations that are unobservable in purely descriptive pollen
studies. The increasing availability of high-quality pollen
records (Fyfe et al., 2009) makes it possible to expand our
synthetic approach to larger data sets and other forest
ecosystems to assess the generality of the trends demon-
strated in Estonia.
Journal of Biogeographyª 2013 Blackwell Publishing Ltd
8
T. Reitalu et al.
ACKNOWLEDGEMENTS
We thank Kai Kimmel for sharing pollen data, Meelis P€artel
and the LANDCLIM group for discussions, Oliver Purschke
for statistical advice, Aveliina Helm and Jesse Morris for
comments, and Jens-Christian Svenning, Thomas Giesecke
and Petr Kune�s for constructive criticism. This project was
supported by Estonian Science Foundation Mobilitas Pro-
gramme MJD4 to T.R. and MTT3 to S.S. and by ETF9031.
REFERENCES
Behre, K.-E. (1981) The interpretation of anthropogenic indi-
cators in pollen diagrams. Pollen et Spores, 23, 225–245.
Behre, K.-E. (1988) The role of man in European vegetation
history. Vegetation history (ed. by B. Huntley and T. Webb
III), pp. 634–672. Kluwer, Dordrecht.
Bennett, K.D. & Lamb, H.F. (1988) Holocene pollen sequences
as a record of competitive interactions among tree popula-
tions. Trends in Ecology and Evolution, 3, 141–144.
Berglund, B.E. (2003) Human impact and climate changes –
synchronous events and a causal link? Quaternary Interna-
tional, 105, 7–12.
Birks, H.J.B. (1986) Late Quaternary biotic changes in terres-
trial and lacustrine environments, with particular reference
to north-west Europe. Handbook of Holocene palaeoecology
and palaeohydrology (ed. by B.E. Berglund), pp. 3–65. John
Wiley & Sons, Chichester, UK.
Borcard, D., Legendre, P. & Drapeau, P. (1992) Partialling
out the spatial component of ecological variation. Ecology,
73, 1045–1055.
Bradshaw, R.H.W. (2008) Detecting human impact in the
pollen record using data-model comparison. Vegetation
History and Archaeobotany, 17, 597–603.
Bradshaw, R.H.W. & Hannon, G. (1992) Climatic change,
human influence and disturbance regime in the control of
vegetation dynamics within Fiby Forest, Sweden. Journal of
Ecology, 80, 625–632.
Bronk Ramsey, C. (2008) Deposition models for chronologi-
cal records. Quaternary Science Reviews, 27, 42–60.
Bronk Ramsey, C. (2009) Bayesian analysis of radiocarbon
dates. Radiocarbon, 51, 337–360.
Cowling, S.A., Sykes, M.T. & Bradshaw, R.H.W. (2001) Pal-
aeovegetation-model comparisons, climate change and tree
succession in Scandinavia over the past 1500 years. Journal
of Ecology, 89, 227–236.
Diekmann, M. (1996) Ecological behaviour of deciduous
hardwood trees in Boreo-nemoral Sweden in relation to
light and soil conditions. Forest Ecology and Management,
86, 1–14.
Easterling, W.E., Brandle, J.R., Hays, C.J., Guo, Q. & Guer-
tin, D.S. (2001) Simulating the impact of human land use
change on forest composition in the Great Plains agroeco-
systems with the Seedscape model. Ecological Modelling,
140, 163–176.
EMHI (2011) Kliimanormid. Available at: http://www.emhi.
ee/?ide=6,299 (accessed 16 November 2011).
Feurdean, A., Willis, K.J., Parr, C.L., Tant��au, I. & F�arcas�, S.(2010) Post-glacial patterns in vegetation dynamics in
Romania: homogenization or differentiation? Journal of
Biogeography, 37, 2197–2208.
Fyfe, R.M., de Beaulieu, J.-L., Binney, H., Bradshaw, R.H.W.,
Brewer, S., Le Flao, A., Finsinger, W., Gaillard, M.-J., Gies-
ecke, T., Gil-Romera, G., Grimm, E.C., Huntley, B., Kune�s,
P., K€uhl, N., Leydet, M., Lotter, A.F., Tarasov, P.E. & Ton-
kov, S. (2009) The European Pollen Database: past efforts
and current activities. Vegetation History and Archaeobota-
ny, 18, 417–424.
Fyfe, R.M., Roberts, N. & Woodbridge, J. (2010) A pollen-
based pseudobiomisation approach to anthropogenic land-
cover change. The Holocene, 20, 1165–1171.
Hammarlund, D., Bj€orck, S., Buchardt, B., Israelson, C. &
Thomsen, C.T. (2003) Rapid hydrological changes during
the Holocene revealed by stable isotope records of lacus-
trine carbonates from Lake Igelsj€on, southern Sweden.
Quaternary Science Reviews, 22, 353–370.
Heikkil€a, M., Edwards, T.W.D., Sepp€a, H. & Sonninen, E.
(2010) Sediment isotope tracers from Lake Saarikko, Fin-
land, and implications for Holocene hydroclimatology.
Quaternary Science Reviews, 29, 2146–2160.
Hellman, S., Gaillard, M.-J., Brostr€om, A. & Sugita, S.
(2008) The REVEALS model, a new tool to estimate past
regional plant abundance from pollen data in large lakes:
validation in southern Sweden. Journal of Quaternary Sci-
ence, 23, 21–41.
Hijmans, R.J., Cameron, S.E., Parra, J.L., Jones, P.G. & Jarvis,
A. (2005) Very high resolution interpolated climate
surfaces for global land areas. International Journal of
Climatology, 25, 1965–1978.
Hult�en, E. & Fries, M. (1986) Atlas of North European vascu-
lar plants: north of the Tropic of Cancer. Koeltz Scientific
Books, K€onigstein, Germany.
Huntley, B. (1990) European post-glacial forests: composi-
tional changes in response to climatic change. Journal of
Vegetation Science, 1, 507–518.
Huntley, B. & Birks, H.J.B. (1983) An atlas of past and pres-
ent pollen maps for Europe: 0–13000 years ago. Cambridge
University Press, Cambridge.
Iversen, J. (1973) The development of Denmark’s nature since
the last glacial. Danmarks Geologiske Undersøgelse, V Series.
C.A. Reitzels Forlag, Copenhagen.
Jalut, G., Dedoubat, J.J., Fontugne, M. & Otto, T. (2009)
Holocene circum-Mediterranean vegetation changes: cli-
mate forcing and human impact. Quaternary International,
200, 4–18.
Jessen, C.A., Rundgren, M., Bj€orck, S. & Hammarlund, D.
(2005) Abrupt climatic changes and an unstable transition
into a late Holocene Thermal Decline: a multiproxy lacus-
trine record from southern Sweden. Journal of Quaternary
Science, 20, 349–362.
Journal of Biogeographyª 2013 Blackwell Publishing Ltd
9
Forest composition, climate and human impact
Keller, F., Lischke, H., Mathis, T., M€ohl, A., Wick, L., Am-
mann, B. & Kienast, F. (2002) Effects of climate, fire, and
humans on forest dynamics: forest simulations compared
to the palaeological record. Ecological Modelling, 152, 109–
127.
Koff, T., Punning, J.-M. & Yli-Halla, M. (1998) Human
impact on a paludified landscape in northern Estonia.
Landscape and Urban Planning, 41, 263–272.
Kukk, T. (2004) Eesti taimede kukeaabits. Varrak, Tallinn.
Kukk, T. & Kull, K. (1997) Puisniidud [Wooded meadows].
Estonia Maritima, 2, 1–249.
Laasimer, L. (1965) Eesti NSV taimkate. Valgus, Tallinn.
Lang, V. (2007) The Bronze and Early Iron Ages in Estonia.
Estonian Archaeology 3. Tartu €Ulikooli Kirjastus, Tartu.
Lindbladh, M., Bradshaw, R. & Holmqvist, B.H. (2000) Pat-
tern and process in south Swedish forests during the last
3000 years, sensed at stand and regional scales. Journal of
Ecology, 88, 113–128.
Lischke, H., Lotter, A.F. & Fischlin, A. (2002) Untangling
a Holocene pollen record with forest model simulations
and independent climate data. Ecological Modelling, 150,
1–21.
Magny, M., Miramont, C. & Sivan, O. (2002) Assessment of
the impact of climate and anthropogenic factors on Holo-
cene Mediterranean vegetation in Europe on the basis of
palaeohydrological records. Palaeogeography, Palaeoclima-
tology, Palaeoecology, 186, 47–59.
Miller, P.A., Giesecke, T., Hickler, T., Bradshaw, R.H.W.,
Smith, B., Sepp€a, H., Valdes, P.J. & Sykes, M.T. (2008)
Exploring climatic and biotic controls on Holocene
vegetation change in Fennoscandia. Journal of Ecology, 96,
247–259.
Mitchell, F.J.G. (2005) How open were European primeval
forests? Hypothesis testing using palaeoecological data.
Journal of Ecology, 93, 168–177.
Nielsen, A.B. & Odgaard, B.V. (2010) Quantitative land-
scape dynamics in Denmark through the last three
millennia based on the Landscape Reconstruction Algo-
rithm approach. Vegetation History and Archaeobotany,
19, 375–387.
Overland, A. & Hjelle, K.L. (2009) From forest to open pas-
tures and fields: cultural landscape development in western
Norway inferred from two pollen records representing dif-
ferent spatial scales of vegetation. Vegetation History and
Archaeobotany, 18, 459–476.
P€artel, M., Helm, A., Reitalu, T., Liira, J. & Zobel, M. (2007)
Grassland diversity related to the Late Iron Age human
population density. Journal of Ecology, 95, 574–582.
Poska, A. & Saarse, L. (1999) Holocene vegetation and
land-use history in the environs of Lake Kahala, north-
ern Estonia. Vegetation History and Archaeobotany, 8,
185–197.
Poska, A. & Saarse, L. (2002) Vegetation development and
introduction of agriculture to Saaremaa Island, Estonia:
the human response to shore displacement. The Holocene,
12, 555–568.
Poska, A., Saarse, L. & Veski, S. (2004) Reflections of pre-
and early-agrarian human impact in the pollen diagrams
of Estonia. Palaeogeography, Palaeoclimatology, Palaeo-
ecology, 209, 37–50.
R Development Core Team (2009) R: a language and envi-
ronment for statistical computing. R Foundation of Statisti-
cal Computing, Vienna.
Reimer, P.J., Baillie, M.G.L., Bard, E. et al. (2009)
IntCal09 and Marine09 radiocarbon age calibration curves,
0–50,000 years cal BP. Radiocarbon, 51, 1111–1150.
Renssen, H., Sepp€a, H., Heiri, O., Roche, D.M., Goosse, H.
& Fichefet, T. (2009) The spatial and temporal complexity
of the Holocene thermal maximum. Nature Geoscience, 2,
411–414.
Saarse, L. & Veski, S. (2001) Spread of broad-leaved trees in
Estonia. Proceedings of the Estonian Academy of Sciences,
Geology, 50, 51–65.
Saarse, L., Poska, A. & Veski, S. (1999) Spread of Alnus and
Picea in Estonia. Proceedings of the Estonian Academy of
Sciences. Geology, 48, 170–186.
Saarse, L., Niinemets, E., Poska, A. & Veski, S. (2010) Is
there a relationship between crop farming and the Alnus
decline in the eastern Baltic region? Vegetation History and
Archaeobotany, 19, 17–28.
Sepp€a, H. & Poska, A. (2004) Holocene annual mean tem-
perature changes in Estonia and their relationship to solar
insolation and atmospheric circulation patterns. Quater-
nary Research, 61, 22–31.
Sepp€a, H., Alenius, T., Bradshaw, R.H.W., Giesecke, T.,
Heikkil€a, M. & Muukkonen, P. (2009) Invasion of Norway
spruce (Picea abies) and the rise of the boreal ecosystem in
Fennoscandia. Journal of Ecology, 97, 629–640.
Sillasoo, €U., Poska, A., Sepp€a, H., Blaauw, M. & Chambers,
F.M. (2009) Linking past cultural developments to palaeo-
environmental changes in Estonia. Vegetation History and
Archaeobotany, 18, 315–327.
Sugita, S. (1993) A model of pollen source area for an entire
lake surface. Quaternary Research, 39, 239–244.
Sugita, S. (2007a) Theory of quantitative reconstruction of
vegetation I: pollen from large sites REVEALS regional
vegetation composition. The Holocene, 17, 229–241.
Sugita, S. (2007b) Theory of quantitative reconstruction of
vegetation II: all you need is LOVE. The Holocene, 17,
243–257.
Sugita, S., Parshall, T., Calcote, R. & Walker, K. (2010) Test-
ing the Landscape Reconstruction Algorithm for spatially
explicit reconstruction of vegetation in northern Michigan
and Wisconsin. Quaternary Research, 74, 289–300.
Svenning, J.-C. & Skov, F. (2005) The relative roles of envi-
ronment and history as controls of tree species composi-
tion and richness in Europe. Journal of Biogeography, 32,
1019–1033.
Tallavaara, M. & Sepp€a, H. (2011) Did the mid-Holocene
environmental changes cause the boom and bust of hun-
ter-gatherer population size in eastern Fennoscandia? The
Holocene, 22, 215–225.
Journal of Biogeographyª 2013 Blackwell Publishing Ltd
10
T. Reitalu et al.
Tuhkanen, S. (1980) Climatic parameters and indices in
plant geography. Acta Phytogeographica Suecica, 67, 1–110.
Veski, S., Heinsalu, A., Klassen, V., Kriiska, A., L~ougas, L.,
Poska, A. & Salu€a€ar, U. (2005) Early Holocene coastal set-
tlements and palaeoenvironment on the shore of the Baltic
Sea at P€arnu, southwestern Estonia. Quaternary Interna-
tional, 130, 75–85.
Zuur, A.F., Ieno, E.N., Walker, N.J., Saveliev, A.A. &
Smith, G.M. (2009) Mixed effects models and extensions
in ecology with R. Springer, New York.
SUPPORTING INFORMATION
Additional Supporting Information may be found in the
online version of this article:
Appendix S1 Details of pollen and charcoal data sets.
Appendix S2 Details of the REVEALS model.
Appendix S3 Site trajectories on RDA triplot.
BIOSKETCH
Triin Reitalu is a plant ecologist interested in historical pro-
cesses that influence the present-day vegetation patterns, with a
special interest in the role of human impact in the formation of
diversity patterns. Currently, she is a post-doctoral researcher
at the Institute of Geology, Tallinn University of Technology in
the group of Siim Veski, exploring various ecological hypothe-
ses using palaeoecological and palaeoclimatological records.
Author contributions: T.R. and S.V. conceived the ideas; M.
K., T.K., E.A., K.K., L.S., A.P. and S.V. provided the pollen
data; J.V. calibrated the chronologies of the pollen stratig-
raphies; H.R. provided the modelled climate data; H.S., M.H.
and D.H. provided the d18O data; T.R. conducted the statis-
tical analyses and wrote the paper, with substantial input
from S.S., H.S. and S.V.
Editor: Jens-Christian Svenning
Journal of Biogeographyª 2013 Blackwell Publishing Ltd
11
Forest composition, climate and human impact