Exploring climatic and biotic controls on Holocene vegetation change in Fennoscandia


*Correspondence author. E-mail: paul.miller@nateko.lu.se


  • 1We investigated the potential drivers of Holocene vegetation changes recorded at four Scandinavian pollen sites, two in Sweden and two in Finland, at a time when they were largely free of anthropogenic influence.
  • 2We used the generalized dynamic vegetation model LPJ-GUESS forced with climate anomaly output from an atmospheric general circulation model to simulate tree species dynamics from 10 000 years ago to the present. The model results were compared to high-resolution pollen accumulation rates gathered at the sites.
  • 3Our results indicate that both the observed northern distributional limits of temperate trees, and the limits of Pinus sylvestris and Alnus incana at the tree line, are a result of millennial variations in summer and winter temperatures. The simulation of several distinct trends in species occurrence observed in the pollen record indicates that a time lag due to the slow spreading of species need not be invoked for most species.
  • 4Sensitivity studies indicate that competition, natural disturbance and the magnitude of interannual variability play key roles in determining the biomass, establishment and even the presence of species near their bioclimatic limits. However, neither disturbance due to fire nor limits on establishment due to drought were likely to have been major determinants of the observed trends on the timescales considered.
  • 5We were unable to limit the modelled occurrence of Picea abies at the study sites to the periods at which it was observed in the pollen records, indicating that we have still not completely understood the driving or limiting factors for Holocene changes in Picea abies abundance.
  • 6Synthesis. This study shows that by combining quantitative vegetation reconstructions with a modern, process-based dynamic vegetation model, we may gain new insights into the potential biotic and abiotic drivers of Holocene vegetation dynamics, and their relative importance. This knowledge will be crucial in enabling us to assess more confidently the response of northern European vegetation to future climate change.


Although there have been many studies documenting the changes in plant cover and forest composition that have occurred in Europe through the Holocene, for example see Huntley (1990) and Huntley & Prentice (1993), the reconstructions provide little insight into their cause. In order to understand past vegetation change, palaeoecologists often evaluate competing hypotheses to arrive at a most likely cause (e.g. Bennett et al. 1992; Giesecke 2006; Tinner & Lotter 2006), but the great variety of potential drivers and dynamic possibilities rarely permits firm conclusions to be drawn (Foster 1992; Giesecke et al. 2007). Dynamic vegetation models make it possible to explore the effects of changing biotic and abiotic factors (Bradshaw 2008; Prentice et al. 2007), and the comparison of model results with palaeoecological reconstructions facilitates a more rigorous testing of competing hypotheses.

A number of studies comparing vegetation reconstructions with results from vegetation models have shown how model studies can corroborate interpretations based on observations and yield further insight into the drivers of vegetation change. Cowling et al. (2001) used the FORSKA2 (Prentice et al. 1993) gap model to study palaeoecological records from Sweden and Denmark over the past 1500 years. The dieback of species in response to sub-millennial climatic variations such as the Little Ice Age was successfully simulated by driving the model with a series of 30-year-long annual mean temperature anomalies and winter and summer precipitation reconstructed from palaeoecological proxy records. Through the comparison of modelled and reconstructed abundance of Fagus sylvatica and Tilia cordata, Cowling et al. (2001) were also able to demonstrate a strong human influence on the composition of Scandinavian temperate forests from the beginning of the 17th century. Heiri et al. (2006) used the FORCLIM gap model to simulate tree-line dynamics in the Central European Alps over a 12 000-year period. Assuming that a smoothed 62-point record of July air temperature anomalies reconstructed from nearby chironomid records was valid for all months, they successfully modelled the early to mid-Holocene fluctuations in a local tree line inferred from pollen and macrofossil data, and could therefore interpret late-Holocene discrepancies as being due to human influence. Also using FORCLIM, Lischke et al. (2002) conducted a series of sensitivity studies to investigate a 13 000-year long pollen record from Soppensee on the central Swiss Plateau. Using summer temperature reconstructions from nearby aquatic vegetation, insect and oxygen isotope records, and again assuming the anomalies to be uniformly distributed over the year, they confirmed the important influence of millennial climate change on observed vegetation composition.

The objective of this study was to investigate the effects of independently changing Holocene climate parameters on the vegetation history in Fennoscandia to explore the hypothesis that climatic factors alone can account for the observed vegetation dynamics. In particular we explored the role of millennial and interannual climate variability as well as disturbance and competition. To gain the most insight into the relative importance of biotic and abiotic drivers, we chose four sites belonging to different biogeographical zones for our study, each of which has been subject to little or no human interference through most of the Holocene. The published pollen stratigraphies for the chosen sites also exhibit the expansion and population contraction of at least one major tree species during the study period.

To meet this objective, we modelled the observed, regional vegetation changes at tree species level using LPJ-GUESS (Smith et al. 2001), a dynamic, process-based vegetation model. LPJ-GUESS simulates the dynamics of forest stands, accounting for competition between tree individuals and populations in the manner of a forest gap model (Shugart 1984), while using mechanistic representations of biophysical and physiological processes. In a novel approach, however, we forced the model for this study with independent Holocene climate data derived from the output of a modern atmospheric general circulation model (AGCM). This has the advantage of allowing us to examine the effects of asynchronous monthly climate changes, while simultaneously avoiding the circularity inherent in driving the model with climate variables reconstructed from the palaeoecological records we wished to simulate.

Throughout this paper, vascular plant nomenclature follows Tutin et al. (1964–80), and both model and pollen dates are expressed in calendar years, where we assume 0 kyr ago to represent pre-industrial times, i.e. year 1750 ad, approximately.


site descriptions

Four published pollen stratigraphies from Fennoscandia were selected to represent different forest regions (Fig. 1, Table 1). All sites show little human impact on the vegetation and each shows the expansion and population contraction of at least one tree species during the Holocene, which could be a result of climate change. The northernmost site, Tsuolbmajavri, lies at the ecotone between boreal forest and the arctic tundra and is thus near the limits of Pinus sylvestris, Betula pubescens and Alnus incana (Seppä & Birks 2001). The ecotone between the temperate and the boreal forest is represented by Laihalampi (Heikkilä & Seppä 2003) and Holtjärnen (Giesecke 2005a), situated just north of or near the present-day limits of Quercus robur, Tilia cordata and Corylus avellana. Abborrtjärnen was chosen because it showed the Holocene population expansion and contraction of Ulmus glabra and is situated north of the distribution limits of the above temperate species.

Figure 1.

Location of the four study sites considered in this study, overlain with the UM AGCM land points (grey shading) from 6 kyr ago.

Table 1.  Summary description of the study sites. MASL: metres above sea level; MASL CRU: metres above sea level as given by the Climate Research Unit's (CRU) 0.5° spatial resolution TS 2.1 forcing data set (Mitchell & Jones 2005); GDD5: average annual growing degree-days between 1901 and 1950 (above a 5 °C base); Pann: average annual precipitation. Both GDD5 and Pann were calculated using CRU TS 2.1 1901–1950 monthly climate data
Site nameLat. (°N)Lon. (°E)MASL (m)MASL CRU (m)GDD5 (°C days)Pann (mm)Dominant speciesSecondary species
Tsuolbmajavri68.6922.08526455 423439BetulaJuniperus, Pinus
Abborrtjärnen63.8814.45387455 654542Picea, PinusBetula, Populus, Sorbus, Alnus
Holtjärnen60.6514.92232245 918593Pinus, PiceaBetula, Alnus, Populus, Ulmus, Corylus
Laihalampi61.4926.081371071152589Pinus, PiceaBetula, Alnus, Populus, Sorbus, Tilia

The four pollen diagrams were constructed from the sediments of small to medium sized lakes, and thus record integrated forest composition on the order of 103 km2 around the lakes (Sugita 1993). All four profiles show near linear age–depth relationships that are secured with at least six radiocarbon age determinations and the assumption that the sediment top is contemporary. Deglaciation occurred at different times during the early Holocene around the four sites, but all were ice-free by 10 000 calendar years ago.

the lpj-guess model

LPJ-GUESS (Smith et al. 2001) is a dynamic vegetation model that simulates the dynamics of tree populations in the manner of a forest gap model (Shugart 1984; Prentice et al. 1993; Bugmann 2001). Trees belonging to different plant functional types (PFTs) or species (Hickler et al. 2004; Koca et al. 2006) may be simulated. Biophysical and physiological processes are represented mechanistically, based on the same formulations as the well-evaluated (Lucht et al. 2002; Sitch et al. 2003; Morales et al. 2005; Zaehle et al. 2005) Lund-Potsdam-Jena dynamic global vegetation model (LPJ-DGVM; Sitch et al. 2003; see also Gerten et al. (2004) for updates to the model's hydrological processes). In contrast to the simplified, area-averaged representation of vegetation structure in LPJ-DGVM, however, LPJ-GUESS simulates cohorts of trees of different species, ages and structures that compete for light and soil resources on a number of replicate patches (50 in the present study). A minimum set of bioclimatic limits (Prentice et al. 1992; Sykes et al. 1996) define the climate space in which each species may occur. In this study we choose to model the Holocene dynamics of the main tree species observed in each pollen record to facilitate a direct comparison with vegetation reconstructions using pollen accumulation rate (PAR) data gathered at individual sites.

In Table 2 we list the 10 tree species modelled, and their main parameters and bioclimatic limits. Further details of the model and our updates to it can be found in Appendix S1 in the Supplementary Material. The C3 grass is a generic PFT, intended to represent the numerous understorey species that are not considered in this paper, but that compete with trees for water and nutrients. Tree species are described as either boreal (Bo) or temperate (Te), either broadleaf summergreen (BS) or needleleaf evergreen (NE), and either shade tolerant (St), intermediately shade tolerant (ISt) or shade intolerant (Si) (Smith et al. 2001). Summergreen species require varying periods of chilling to induce budburst (Appendix S1; Murray et al. 1989).

Table 2.  Selected species parameters and bioclimatic limits used in the simulations and factors used in the adjustment of pollen accumulation rates. GDD5: minimum growing degree-day sum (5 °C base), Tc_min (Tc_max): minimum (maximum) temperature of the coldest month. We use ‘–’ to indicate that no limit is applied. Rfire: fraction of a species’ patch population and litter that survives a fire. Wmin: minimum soil water content (averaged over the growing season and expressed as a fraction of available water holding capacity) needed for establishment. Rpoll: multiplicative factor representing the differences in pollen productivity and dispersal between species. Arepr: age at which a tree species begins to allocate 10% of its net primary production to reproduction
SpeciesDescriptionDD5 (°C days)Tc_min (°C)Tc_max (°C)RfireWmin (mm day−1)RpollArepr (year)
Alnus incanaBo,BS,Ist500–30 –
Betula pendulaTe,BS,Si700–30 70.20.42210
Betula pubescensBo,BS,Si300 60.10.50210
Corylus avellanaTe,BS,Ist700–13100.20.300.520
Picea abiesBo,NE,St650–30 –
Pinus sylvestrisBo,NE,Ist450–30 –
Populus tremulaTe,BS,Si500–30 60.20.400.12520
Quercus roburTe,BS,Ist1100–9
Tilia cordataTe,BS,Ist1100–11 50.10.330.12520
Ulmus glabraTe,BS,Ist850–9.5 60.10.400.2520
C3 grass ,–,–,–01.00.01 0

The maximum range limits of the tree species under a given climate situation are determined in the model by four species-specific bioclimatic constraints, GDD5 (minimum growing degree-day sum (5 °C base)), Tc_min (minimum temperature of the coldest month), Tc_max (maximum temperature of the coldest month) and Wmin (minimum growing season soil water content (as a fraction of available water holding capacity) needed for establishment) (Table 2), each of which represents a known or likely physiological limiting mechanism (Appendix S1). The values in Table 2 were taken from the forestry literature (Prentice & Helmisaari 1991) and by comparison of modern species distributions with bioclimatic variables such as GDD5 and Wmin. This comparison goes beyond mere correlation between climatic variables and species ranges, which might not remain robust through time, but seeks to incorporate documented examples of climatic controls on species limits (Woodward 1987). The northern distributional limits of Tilia cordata, for example, are modelled by a GDD5 of 1100 °C days, which is underpinned by the need for sufficient late summer warmth to enable pollen tubes to reach and fertilize the ovules (Pigott & Huntley 1981) and the for the fertilized seeds to mature (Pigott 1991). The ability of Tilia cordata to survive more extreme winter temperatures than Quercus robur is documented from field observations of Quercus trees splitting longitudinally on very cold nights with a noise like a gunshot, with no such effect on Tilia (Jones 1959). Thus LPJ-GUESS summarizes current physiological and bioclimatic understanding of the controls of tree distributions (Table 2, Appendix S1), even though there are still significant gaps in our knowledge.

For a species within its bioclimatic limits, cohort establishment and mortality are modelled as stochastic processes within each patch. In addition, two types of stochastic disturbance are simulated in LPJ-GUESS. First, patch-destroying disturbances, representing processes such as herbivory and storm damage, result in all vegetation in a patch being transferred to the patch's litter pool with a certain annual probability, the inverse of the average disturbance interval. In the present study, a disturbance interval of 100 years, an approximate average value for European ecosystems, was assumed. Second, fire disturbance is modelled prognostically as in Thonicke et al. (2001), where the yearly probability of a fire depends on a minimum litter fuel load, litter flammability and the water content of the upper soil layer. Each modelled species has a specific fire resistance (Rfire, Table 2), with higher values resulting in both a higher probability of survival in the event of a fire, and in less of its litter being burned. In general, warmer and drier conditions result in more fires, so long as the vegetation productivity is sufficient to provide an adequate fuel load.

Finally, trees in the model allocate a fraction, 10%, of their net primary production (NPP) to reproduction (Harper 1977). However, to account for the fact that trees do not produce pollen until mature (Savill 1991), this is delayed until the species cohort has reached a minimum age (Arepr, Table 2) prescribed for each species. Trees younger than Arepr were not included in the model output.

Further changes to the model parameters described by Smith et al. (2001) and Hickler et al. (2004), as well as further species parameters, are listed in Appendix S1.

model forcing and simulation protocol

For simulations of modern potential natural vegetation, LPJ-GUESS is forced with monthly temperature, precipitation and cloudiness data from the Climate Research Unit's (CRU) 0.5° × 0.5° spatial resolution (TS 2.1) forcing data set (Mitchell & Jones 2005), available from http://www.cru.uea.ac.uk/. Averaged GDD5 and annual precipitation for the period 1901–50 are given in Table 1, and data for the grid cell encapsulating each study site were taken as representative for the site climate.

In the case of Tsuolbmajavri, the nearest grid cell had a lower average elevation and, consequently, warmer temperatures than the site itself. The adjacent grid cell to the west (henceforth referred to as TsuolbmajavriW) is almost as close to the study site (15 km as opposed to 9.6 km) and has a more representative altitude and climate. For this reason, we decided to perform separate model simulations for these two grid cells and average the output in all simulation results presented below.

For the Holocene simulations, we combined a long-term trend of Holocene climate variability with the CRU data. The long-term trend was constructed using output from the Hadley Centre's Unified Model (UM). The version of the UM used in this study is HadSM3, which is an AGCM coupled to simple mixed-layer ocean and sea ice models (Pope et al. 2000). The AGCM has a spatial resolution of 2.5° × 3.75°. These data are available from the Bristol Research Initiative for the Dynamic Global Environment (BRIDGE, http://www.bridge.bris.ac.uk/), for most 1000-year time periods from 18 kyr ago to 0 kyr ago (i.e. pre-industrial times).

We calculated anomalies for each month of the year relative to the UM output for 0 kyr ago, adapted them to the CRU grid cells containing the study sites for each monthly climate variable and time slice, and then linearly interpolated them between time slices. This gave a separate trend for each month of the year and interval between time slices, on which we superimposed the observed interannual variability for the relevant month between 1901 and 1950, as given by the CRU data set (Mitchell & Jones 2005) in a continuously repeated sequence. Amalgamating these data gave a 10 000-year climate forcing record that varies on both millennial and annual timescales. This procedure assumes that the interannual climate variability for temperature, precipitation and cloudiness did not change over the Holocene. We plot three reconstructed bioclimatic variables in Fig. 2. The GDD5 and Tc_min bioclimatic limits are of key importance to the modelled vegetation (Sykes et al. 1996), and represent the Holocene trend in summer warmth and minimum winter temperatures, respectively. Annual precipitation shows less variability at each site after 8 kyr ago.

Figure 2. (a).

100-year mean growing degree-days (GDD5, °C day), (b) minimum temperature of the coldest month (Tc_min,°C), and (c) annual precipitation (mm) at the study sites through the Holocene, as derived from the UM AGCM model anomalies and modern CRU TS 2.1 climate data. Dashed lines: Laihalampi; solid lines: Holtjärnen; dotted lines: Abborrtjärnen; dash-dot-dot lines: Tsuolbmajavri; dash-dot lines: Tsuolbmajavri W.

Yearly varying atmospheric CO2 concentrations were obtained by linear interpolation between the values used as boundary conditions in the UM simulations. Soil texture was assumed to be static and taken from Sitch et al. (2003) based on the FAO (1991) soil data set.

To simulate species biomass for each study site, each of which was ice-free by 10 kyr ago, we first ‘spun-up’ the model for 500 years using repeated 1901–50 CRU data for the nearest grid cell corresponding to the location of the site, with climate anomalies for 10 kyr ago imposed. The purpose of this spin-up was to establish an initial vegetation whose composition, structure and biomass was at equilibrium with the climate at the beginning of the study period. Thereafter the model was forced with the climate forcing described above until 0 kyr ago. All species within their prescribed bioclimatic limits were allowed to occur, except that Picea was only allowed to establish when the pollen data indicated its presence near the sites.

description of sensitivity experiments

To aid in the determination of some of the main factors driving simulated vegetation change at the study sites, we carried out a range of simulation experiments with different model configurations, listed in Table 3.

Table 3.  Description of model experiments
Experiment nameDescription
Standard runPicea is allowed to establish at the time its presence can be inferred from the pollen data
–DisturbFrequency of patch-destroying disturbances halved
–FireNo fires allowed
–DroughtNo drought-limited establishment
VarX2Monthly deviations from the 1901–50 climatology were doubled
VarX0.5Monthly deviations from the 1901–50 climatology were halved
PiceaPicea is never allowed to establish

The –Disturb experiment examined the impact on Holocene vegetation dynamics of a restriction of the disturbances deemed essential to the observed succession in pristine forests. –Picea examined the influence on simulated trends of the absence of a strong competitor. –Fire and –Drought examined the impacts of fire suppression and the removal of species-specific drought-limited establishment thresholds (Table 2), respectively. Experiments VarX2 and VarX0.5, examined the impact on modelled vegetation of relaxing our assumption that the interannual variability over the last 10 000 years had the same magnitude as observed climate between 1901 and 1950.

comparison of simulated and observed vegetation

LPJ-GUESS predicts absolute plant biomass. In order to use this information for characterizing past vegetation changes, we decided to compare it to pollen accumulation rates (PAR). The utilization of PAR circumvents the interdependency of percentage pollen data and shows a direct relationship to species abundance and biomass (Seppä & Hicks 2006; Giesecke & Fontana et al. 2008). However, differences in pollen productivity and dispersal between species complicate comparisons between biomass and PAR. Despite early attempts (Davis et al. 1973) and recent efforts to calibrate PAR against biomass, such conversion factors are not yet available. In order to adjust PAR from different species to broadly resemble species biomass, we devised conversion factors (Rpoll, Table 2), which are inspired by studies on the relationship between pollen percentages and plant abundance (Andersen 1970; Prentice 1978; Prentice et al. 1987; Sugita et al. 1999). LPJ-GUESS computes plant biomass on an annual basis, which, for comparison with adjusted PAR data, was averaged over 50 years (100 years for the sensitivity tests). A three-term running mean was applied in order to reduce the noise in the PAR data.

Due to the relatively low diversity of the arboreal flora in Fennoscandia, it is often possible to link pollen types directly to a plant species. In this respect we assume that Quercus pollen originates mainly from Quercus robur, Tilia pollen from Tilia cordata and Ulmus pollen from Ulmus glabra. Within the comparisons in this study, Alnus pollen may derive from both Alnus incana and Alnus glutinosa. At present the northern limit of Alnus glutinosa is not far north of Holtjärnen and Laihalampi, whereas 8 kyr ago Alnus glutinosa may have grown near Abborrtjärnen (Giesecke 2005b), but it never reached Tsuolbmajavri. Because Alnus glutinosa is most common on waterlogged soils that were not separately modelled we decided not to model this species and assume that most of the Alnus pollen at the sites derives from Alnus incana. Betula pollen cannot be assigned to a single species either. During pollen analyses, pollen from Betula nana were not separated and we assume here that Betula pollen derives from Betula pendula and Betula pubescens. Both species were modelled and the biomass values aggregated for comparison to the pollen data.


simulated holocene vegetation – standard run

There is generally a good correspondence between modelled biomass and adjusted PAR estimates for all four sites during the Holocene (Fig. 3). Reflecting the total adjusted PAR, total modelled biomass is larger for the southernmost sites, and also increases rapidly as a result of rising winter and summer temperatures from 10 kyr to 9 kyr ago (Fig. 2). In the case of Picea abies, the predicted distribution is broader than observed in the pollen record, confirming that factors other than the prescribed bioclimatic limits, the climate forcing data and the simulated dynamics have restricted the actual occurrence of this species at the study sites. For this reason, the establishment of this species was artificially restricted in subsequent simulations (see the Methods).

Figure 3.

Adjusted pollen accumulation rate (PAR) (left column), and modelled biomass (right column), for each species, at Tsuolbmajavri (top row), Abborrtjärnen, Holtjärnen and Laihalampi (bottom row).

The marked decline in pollen accumulation over the last 1000 years due to the opening of the surrounding forests for agriculture and forestry at Holtjärnen, Abborrtjärnen and Laihalampi (Heikkilä & Seppä 2003; Giesecke 2005a; Giesecke 2005b) is not reflected in the modelled biomass since land use and human influence were not considered in this study.

Pinus and Betula dominated the modelled vegetation at Tsuolbmajavri throughout the Holocene, though Alnus was also more abundant prior to 4 kyr ago compared with the present day. These patterns are in general agreement with the vegetation reconstructed using the PAR data gathered at the sites. At present, the lake lies 30 km north of the Pinus-dominated forest, and Alnus is only present in very small numbers. Individual comparisons of adjusted PAR and biomass for Pinus and Alnus (Fig. 4a) show that the model largely captures the observed trends, with a rise before 9 kyr ago, maximum between 8 kyr and 6 kyr ago, and decline thereafter (Seppä & Birks 2001; Seppä & Hicks 2006). These trends are driven by summer temperatures for the grid cell, as reflected by GDD5 values (Fig. 2). The GDD5 local maximum 1 kyr ago in the forcing data results in an increase in modelled biomass that does not agree with the adjusted PAR data, which may indicate that summer temperatures are too high in the UM simulations of the late Holocene.

Figure 4.

Direct comparisons of adjusted PAR (solid lines throughout) and modelled biomass (dashed lines) for various species and sites: (a) Pinus sylvestris (top) and Alnus incana (bottom) at Tsuolbmajavri; (b) Alnus incana (top) and Ulmus glabra (bottom) at Abborrtjärnen; (c) Ulmus glabra at Holtjärnen; and (d) Corylus avellana at Laihalampi.

Further south, summer and winter temperatures around Abborrtjärnen (Fig. 2) were high enough to simulate the presence of both Corylus and Ulmus during the Holocene. Both species were interpreted to have occurred near the lake, although Corylus only in small abundances (Giesecke 2005b). While the long-term Holocene climate trend (Fig. 2) never exceeded the GDD5 limit for the establishment of Ulmus, the applied interannual temperature variability allowed establishment in some summers during the early to mid-Holocene. The modelled biomass for Ulmus is too small compared to adjusted PAR (Fig. 3), although some similarity is visible between the simulated and reconstructed trends (Fig. 4b). The peak in modelled Ulmus biomass occurs approximately 1000 years earlier than the reconstructed peak, which may indicate that the warmest summers were later in the Holocene, and not at 8 kyr ago, as assumed here (Fig. 2).

The climate of Abborrtjärnen is more favourable for Alnus than that of Tsuolbmajavri (Table 2, Fig. 2). This is reflected in a higher adjusted PAR value for Alnus at this site than at Tsuolbmajavri, and in a greater modelled biomass (Fig. 4a,b). The reasons for the modelled Holocene trend in this taxon are complex. Summer temperatures before 9 kyr ago were too cold for Alnus establishment (Fig. 2), but thereafter its biomass reflects the summer temperatures in the forcing scenario. Disallowing the establishment of Picea abies at Abborrtjärnen (–Picea sensitivity experiment, not shown) causes Alnus biomass to increase after 3 kyr ago in response to the slight rise in GDD5 at that time (Fig. 2). The continued decline over the last 3 kyr could therefore have been caused by the competition with Picea abies.

Holocene summer warmth at Holtjärnen exceeds the GDD5 requirement for Ulmus (Table 2) over much of the last 10 kyr, while the limits for Tilia and Quercus are only surpassed due to the imposed interannual variability. With little competition from Tilia and Quercus, Ulmus is modelled as the most abundant temperate species. The modelled and reconstructed Ulmus abundances agree well for the timing of maximum abundances and the declining trend (Fig. 4c). The latter agreement is surprising, considering the effect that competitive replacement of Ulmus by Tilia may have had near Holtjärnen (Giesecke 2005a; Fig. 3), which is not modelled since the abundance of Tilia is limited by its GDD5 requirements. The modelled, declining trends of Ulmus, as well as of Alnus, Populus and Quercus (Fig. 3) are partly a response to the decreasing summer warmth until 4 kyr ago (Fig. 2a) and partly caused by competitive replacement by Picea. The overall effect the introduction of Picea has on the reduction in biomass of temperate species is illustrated in Fig. 5, demonstrating that the expansion of a strong competitor such as Picea abies can outweigh the effect that climate warming has on the abundance of temperate species.

Figure 5.

Adjusted PAR (solid line) and aggregated percentage of modelled temperate species (Corylus, Populus, Quercus, Ulmus and Tilia) biomass at Holtjärnen, both with (long dashed) and without (short dashed) Picea abies in the simulation.

Compared to the other sites, the simulation for Laihalampi least resembles the vegetation reconstructed using the PAR data (Fig. 3), though the model does capture the observed trend for Corylus (Fig. 4d). Interestingly, the modelled shift from Tilia to Quercus may help to understand a similar shift in the adjusted PAR at Holtjärnen (Fig. 6). Compared to Quercus, Tilia may occur in regions with lower winter temperatures, which is reflected in its lower Tc_min limit in Table 2. The Holocene climate forcing for Laihalampi depicts rising January temperatures through the Holocene. Average minimum temperatures between –10 °C and –9 °C during the early Holocene allow Tilia to grow, but limit the populations of Quercus and also Ulmus, while GDD5 values are not limiting. Average minimum temperatures above –9 °C during the late Holocene permit larger populations of Quercus and Ulmus, which in turn reduce the abundance of Tilia. Thus the late expansion of Quercus adjusted PAR at Holtjärnen could be interpreted as a late Holocene increase in minimum temperatures, while minimum temperatures at Laihalampi may never have been high enough for Quercus to be abundant here.

Figure 6.

(a) Tilia (solid line) and Quercus (dashed line) adjusted PAR at Holtjärnen; (b) Modelled Tilia (solid line) and Quercus (dashed line) biomass at Laihalampi; (c) UM AGCM January (dashed line) and July (solid line) temperatures at Laihalampi.

sensitivity experiments

The –Disturb, –Fire and –Drought sensitivity experiments (Table 3) were found to have had little effect on the overall modelled vegetation trends at the study sites on millennial timescales. Results from these experiments are described in more detail in Appendix S2.

The VarX2 and VarX0.5 (Table 3) sensitivity experiments, however, show that changes to the amplitude of interannual variability in meteorological parameters can influence the biomass proportion, year of establishment and even the very presence of a species near its bioclimatic limits. By doubling the amplitude of interannual variability in VarX2, summer temperatures are more likely to exceed a species’ lower GDD5 limit for establishment, but temperatures of the coldest month are also more likely to drop below species’ Tc_min limits (Table 2).

Conversely, by halving the amplitude of interannual variability in VarX0.5, summer temperatures are less likely to exceed a species’ GDD5 limit to allow establishment, but temperatures of the coldest month are also less likely to drop below species’ Tc_min limits, and more likely to stay above the minimum temperature for survival.

We examine these processes in detail at Holtjärnen (Fig. 7). The abundance of Quercus is reduced in both experiments compared to the standard run (Fig. 3), because sufficiently warm summers occur less frequently in VarX0.5, and cold winters become more frequent in VarX2. Similarly, Ulmus is limited by more frequent cold winters in VarX2, and its biomass is sharply reduced. The overall biomass is also significantly reduced in the VarX2 scenario. Increased tree mortality in the VarX2 experiment is due to more frequent cold winters, which leads to shorter lifetimes for most trees, especially the temperate species. Thus the reduction in the biomass of broadleaved trees is similar to that caused by increased disturbance frequency. The establishment and survival of Tilia and Quercus in the VarX0.5 experiment is limited to the time period around the GDD5 maximum, when establishment outweighs mortality.

Figure 7.

Illustrative results from the VarX2 and VarX0.5 sensitivity experiments: modelled biomass at Holtjärnen from the (a) VarX0.5 and (b) VarX2 experiments; (c) Tilia adjusted PAR (solid line) and biomass (dashed line) at Laihalampi, overlain with Tilia biomass from the VarX2 (dotted line) and VarX0.5 (dashed double-dotted line) experiments.

At Laihalampi, frequent cold winters become the limiting climate variable for the establishment and survival of Tilia in the VarX2 experiment (Fig. 7). Its brief period of establishment at 6 kyr ago marks the first time that sufficiently high Tc_min values are reached. Furthermore, its larger abundance in the late Holocene is caused by the lack of competition from Quercus, which is absent in this experiment due to cold winters (results not shown). Conversely, halving variability in VarX0.5 leads to earlier sustained establishment for Tilia at Laihalampi (Fig. 7).


Comparison of quantitative palaeoecological data with results from the model shows that a combination of different abiotic and biotic factors are necessary to explain past vegetation changes. Changes in interannual variability may be as important as millennial scale trends, and different climate variables may become limiting at different times during the Holocene.

Earlier site-specific data model comparisons (e.g. Cowling et al. 2001; Lischke et al. 2002; Heiri et al. 2006) were generally conducted with more simplified models and often only used a small number of variable forcing parameters. The focus in these earlier studies was on the effects of change in average climate, taking little account of the presumptively critical effect of variability and extremes (e.g. low temperatures) on vegetation distributions (Woodward 1987). Using monthly climate anomalies derived from AGCM output, which is completely independent from climate reconstructions using biological proxy data, and superimposing upon it modern climate variability, makes possible the study of the effects of a larger range of parameters. This allows us, for example, to assess the effects of independent changes in summer and winter temperatures as well as interannual variability. For this study AGCM output was only available at 1000-year intervals. Centennial or decadal climate variability could therefore not be considered, although Cowling et al. (2001) showed it to be important in shaping species composition in Scandinavia during the late Holocene.

It was not the aim of this study to evaluate the UM AGCM output used to force LPJ-GUESS, but rather to use it as a possible Holocene scenario of climate variability and investigate its impact on the modelled vegetation. Similar AGCMs were used to simulate mid-Holocene climate in phase one of the Palaeoclimate Modelling Intercomparison Project (PMIP, http://pmip.lsce.ipsl.fr/). In a comparison of 16 models of varying complexity and resolution with pollen-based reconstructions of bioclimatic variables, Masson et al. (1999) found that models could not capture all aspects of the reconstructed mid-Holocene climate in Europe (Cheddadi et al. 1997). Climate reconstructions based on pollen data indicating warmer than present summers for northern Europe during the mid-Holocene were well reproduced by AGCMs due to higher insolation forcing. However, reconstructions indicating northern European winters being milder than present during the mid-Holocene are generally not captured by AGCMs, as winter temperatures in Europe are primarily due to circulation patterns (Bonfils et al. 2004), particularly changes in the mean state or variability of the North Atlantic Oscillation (NAO), which is difficult to predict (Gladstone et al. 2005). Holocene climate variability at the four sites may also have been influenced by changes in the extent of the Baltic Sea, which covered nearly twice its present area during the early Holocene (Björck 1995). This increased water mass, which is not considered by the AGCM's, would buffer the annual temperature range and cause milder winters. For these reasons, the winter temperature anomalies used here are questionable. Even though this forcing mechanism results in interesting modelled vegetation change (Fig. 6), it should be viewed merely as an example of how changes in winter temperature can drive vegetation change.

Using these climate scenarios, the modelled Holocene vegetation composition in Fennoscandia corresponds in general terms with vegetation reconstructions from pollen data. Good agreement is found near the tree line and the northern limit of temperate trees: changes that are mainly driven by millennial changes in temperature. However, the climate forcing parameters cannot explain the absence or insignificant abundance of Picea abies in Fennoscandia during the early to mid-Holocene. Numerous palaeoecological reconstructions show that Picea abies populations expanded in Fennoscandia thousands of years after all other boreal species had established (Giesecke & Bennett 2004; Latalowa & van der Knaap 2006). As a strongly shade-bearing, cold-tolerant and reasonably long-lived species, Picea abies is a strong competitor in Scandinavia, and significantly affects the modelled vegetation dynamics if it is allowed to dominate when it was not observed to have been abundant. This necessitated a restriction on its establishment in the simulations prior to the period when the species became abundant in the pollen records. However, the model does capture the effect that this strong competitor has on the abundance of temperate taxa near their climatic limits (Fig. 5) and shows how competition can influence species range limits (Giesecke 2005a). Similar findings were made by Heiri et al. (2006), who simulated the occurrence of Picea abies in the early Holocene in the Alps although reconstructions indicate a late Holocene expansion of the tree there. Bradshaw et al. (2000), on the other hand, successfully modelled the abundance of Picea abies over the last 1000 years. The latter study focused on changes near the range limits in southern Sweden and winter warmth was identified as the limiting parameter. Considering the large uncertainties of the winter temperature scenarios and the large internal variability of winter temperatures in Fennoscandia today, it seems conceivable that mild and variable early Holocene winters could have limited the abundance of Picea abies during the early Holocene (Giesecke 2004). However, more information on the evolution of Holocene winter temperature and variability is needed to test this hypothesis.

With the exception of Picea, we did not limit the establishment of the modelled tree species based on their historical biogeography. As the agreement between model output and independent vegetation reconstructions using PAR data is generally good, this shows that a time lag due to slow range expansion need not be invoked for most species, which contrasts with the conclusions of Lischke et al. (2002). The reason for a strong climatic control on species dynamics may be due to the close proximity of the sites to the physiological climatic limits of the taxa.

By adopting a regional approach, the study provides insight into mechanisms of climate-driven vegetation change and shows how millennial-scale variations of different climate variables might influence the regional presence and abundance of tree species. For example, whereas minimum GDD5 appears to be the most important limiting variable for Tilia and Quercus at Holtjärnen, winter temperatures are found to limit the growth of these species under the climate scenario for Laihalampi. However, for this study the climate affinities of the tree species were characterized by a relatively small set of general rules and limits (Table 2). Introducing further species-specific climate limitations to the model, such as the warmth requirement for the pollen tube growth in Tilia cordata (Pigott 1991), could potentially improve simulations of individual species responses to complex climate forcing.

The sensitivity tests VarX0.5 and VarX2 for Holtjärnen (Fig. 7) illustrate the strong effect that interannual variability has on forest biomass and composition. In common with the experiments of Bugmann & Pfister (2000), our simulations indicate (Fig. 6) that biomass decreases in response to increased interannual variability, particularly for Ulmus in our study. This is due to the overriding negative effect of exceptionally cold years. This effect needs to be considered in predictions of future species distribution and abundance under climate change scenarios, which are likely to include changes in interannual variability. Similarly, interannual variability is likely to have varied in the past, as indicated in the Holocene climate simulations of Renssen et al. (2005). Although this has so far not been the subject of much discussion as a driving factor for past vegetation change, our sensitivity studies indicate that it deserves more attention. Presently, there is little information available on how interannual variability has changed in the past, and the sensitivity studies conducted here may therefore be considered as indicators of the effect of this uncertainty.

The results presented indicate that the AGCM climate anomalies used do adequately describe some patterns of Holocene climate history. The modelled increase in biomass for many species between 10 and 9 kyr ago, due to the strong increase in temperature, agrees well with the vegetation reconstructed using the PAR data. Furthermore, the early thermal maximum starting at 9 kyr ago and lasting until about 6 kyr ago in the climate scenario for Tsuolbmajavri (Fig. 2) gives rise to a modelled vegetation change that agrees with adjusted PAR data. Climate scenarios for Abborrtjärnen and Holtjärnen exhibit GDD5 and July temperature maxima at 8 kyr, which result in a maximum simulated abundance of temperate species after 8 kyr. The high adjusted PAR estimates for Ulmus at Abborrtjärnen and for Tilia at both Holtjärnen and Laihalampi between 7 kyr and 6 kyr ago could have been due to maximum summer temperatures at that time (Heikkilä & Seppä 2003; Kaplan & Wolfe 2006). However, maximum adjusted PAR for Corylus and Ulmus at Holtjärnen between 8 and 7 kyr ago may indicate that GDD5 had already reached high values by 8 kyr ago at this site. The climate scenario for Laihalampi has its highest July temperatures 8 kyr ago (Fig. 6) but its highest GDD5 7 kyr ago (Fig. 2). Although this scenario has little effect on the modelled vegetation, the concept of separating maximum July temperatures from maximum GDD5 holds explanatory power for some features of vegetation change. The high summer insolation at northern latitudes during the early Holocene (Berger 1978; Berger & Loutre 1991) would alone have caused early Holocene summers to be warmest with declining temperatures thereafter (Renssen et al. 2005). However, the melting of ice-sheets during the early Holocene complicated the climatic response to the insolation forcing and may have altered general circulation patterns. A climate scenario with an initial rise in GDD5 but a delayed maximum in July temperature could explain the late expansion of Tilia reconstructed for Holtjärnen and Laihalampi (Heikkilä & Seppä 2003).

The climate scenarios for Abborrtjärnen and Holtjärnen depict a minimum summer warmth at 4 kyr and higher values between 3 and 1 kyr, when, in the absence of Picea, the model simulates increased biomass of Populus and Alnus at Abborrtjärnen, and Alnus and Ulmus at Holtjärnen (Fig. 5). However, these trends are suppressed by Picea abies when it is allowed to establish during that period. Similarly, the climate scenario for Tsuolbmajavri has a GDD5 minimum at 3 kyr and a secondary maximum at 1 kyr. Here the forcing results in a secondary Pinus biomass peak that is not seen in the adjusted PAR data. The late Holocene temperature scenarios for the sites disagree with some other studies, such as the GCM simulations of Renssen et al. (2005) or general trends reconstructed from a large number of pollen diagrams (e.g. Davis et al. 2003; Kaplan & Wolfe 2006). However, it is interesting to note how the expected response to increased summer warmth can be reversed by the expansion of a strong, shade-tolerant competitor. This points to a potential bias in climate reconstructions from pollen data.


This study demonstrates that combining quantitative vegetation reconstructions using pollen accumulation rate (PAR) data with process-based vegetation modelling can yield valuable insights into the potential drivers, both biotic and abiotic, of Holocene vegetation change in Fennoscandia. Millennial-scale climate change determines in large part the observed changes in tree species composition in the region, as hypothesized. As indicated by the adjusted PAR data, the total modelled biomass in Fennoscandia decreases gradually towards the north, and the early Holocene increase in biomass can be explained by rapidly rising temperatures at that time. The north–south movement of Pinus sylvestris and Alnus incana at the latitudinal tree line are largely driven by summer temperature changes. However, the occurrence of temperate trees near their northern distributional limits is likely to have been determined by different climate parameters at different locations and times. These limiting parameters include summer warmth as recorded in growing degree-days, minimum winter temperatures and maximum mid-summer temperatures. The applied, seasonally variable climate forcing from the UM AGCM was crucial in enabling these effects to be captured.

Even though the modelled occurrence of Picea abies at the study sites did not agree with the periods at which it was abundant in the pollen record, biotic effects of the population expansion of Picea on other tree populations were adequately modelled. Furthermore, we found that the expansion of a strong competitor such as Picea can outweigh the effect that climate warming has on the abundance of temperate species.

Natural disturbances were shown to have an effect on overall biomass, but a changing disturbance regime would probably have had little effect on Holocene vegetation trends at the study sites. Similarly, neither fire nor drought-limited establishment were found to strongly affect the modelled tree species composition on these timescales.

In addition to millennial-scale climate variability, the magnitude of interannual variability was also found to be important in determining the biomass, establishment and occurrence of tree species near their bioclimatic limits. Holocene changes in interannual variability could have significantly altered species composition in Fennoscandia, and potentially elsewhere.

These findings highlight the importance of the distinguishing characteristics of modelled species, for example shade tolerance, longevity and bioclimatic limits. Further advances may be expected with the introduction of improved parameterizations and species knowledge. Climate scenarios derived from the UM AGCM adequately describe some pattern of Holocene climate history, and data model comparisons like the present study can be used to evaluate their output. The combined use of palaeoecological data and dynamic vegetation models helps in understanding past vegetation changes, which in turn will enable us to evaluate the impacts of future climate change on vegetation.


The authors would like to express appreciation to Prof. Colin Prentice and an anonymous referee for helpful comments upon an earlier draft of this manuscript. Authors Miller, Giesecke, Bradshaw and Sykes are grateful for financial support for project DECVEG from the European Science Foundation (ESF) under the EUROCORES Programme EuroCLIMATE, through contract no. ERAS-CT-2003-980409 of the European Commission, DG Research, FP6. T. Hickler received financial support from the ALARM (Assessing LArge scale environmental Risks for biodiversity with tested Methods) Integrated Project (IP) (contract no. GOCE-CT-2003-506675), funded by the European Commission within the 6th Framework Programme.