Introduction

In areas surrounding the Mediterranean Basin, forests are an important element of the established vegetation, covering around 20–30% of the total land areas in the northern part and reaching up to 50% in Greece (Archibold 1995; Scarascia-Mugnozza et al. 2000). These forests are expected to experience warmer and drier conditions in the near future due to global warming (Giorgi and Lionello 2008; Gualdi et al. 2013), as well as potential shifts in fire frequency driven by both climatic and anthropogenic forcing (Barbero et al. 1990; Pausas 2004; Moriondo et al. 2006). Observational evidence of shifts in forest structure and function over the last century has started to accumulate and are frequently attributed to climatic changes. Tree growth changes related to temperature increase (Jump et al. 2006; Linares and Tíscar 2010), decline in precipitation (Sarris et al. 2011) and/or CO2 fertilisation (Martínez-Vilalta et al. 2008) have been documented at both lowland and upland Mediterranean regions. Forest dieback has been attributed to drought (Van Mantgem et al. 2009; Allen et al. 2010), and/or to pathogens outbreaks (Desprez-Loustau et al. 2006; Chrysopolitou et al. 2013), while drought-induced changes in species composition have also been reported (Allen and Breshears 1998; Peñuelas and Boada 2003). Furthermore, some studies report an increase in fire frequency associated with the recent warming, both in the north-western (Pausas and Fernández-Muñoz 2012) and the north-eastern (Koutsias et al. 2013) part of the Mediterranean Basin. The way these shifts will progress under future global change conditions is important in terms of nature conservation and climate change adaptation (Bonan 2008).

Mediterranean plants have evolved under low water availability and thus have developed a number of morphological and physiological adaptations that enable them to withstand prolonged drought periods (Sardans and Peñuelas 2013). Recent studies suggest that in a Mediterranean climate, there is a continuum of plant drought performance, ranging from fast-growing deciduous species with a high resource use and high drought vulnerability to conservative slow-growing evergreen species with low water use and high drought tolerance (Lopez-Iglesias et al. 2014). The two ends of the spectrum reflect a drought tolerance versus a drought avoidance strategy, and functional traits such as rooting depth per leaf area, relative growth rate and net carbon assimilation rate were found to be good predictors of seedling drought survival time (Lopez-Iglesias et al. 2014). Seedlings of fast-growing species were less drought-tolerant in contrast to slow-growing species that exhibited a higher drought tolerance. A similar (weak) trade-off between growth and survival has been reported for mature trees with wood density being a good (negative) predictor of relative growth rate (Martínez-Vilalta et al. 2010). The above suggest that based on their functional configuration, Mediterranean species are responding individualistically to drought and thus their distribution could be controlled by water availability (Piedallu et al. 2013).

Simulations from both local- and global-scale vegetation models suggest that forests established under a Mediterranean climate are particularly vulnerable to climate change (Morales et al. 2007; Fyllas and Troumbis 2009; Hickler et al. 2012), although the large climatic stochasticity of Mediterranean ecosystems (Blondel and Aronson 1999) could increase the uncertainty in such modelling exercises. Under climate change conditions, some typical Mediterranean forests dominated by species such as Pinus halepensis are projected to be more resilient than others that are mainly found in the temperate zone (Keenan et al. 2011). One of the key drivers of vegetation and/or productivity shifts in these predictions is the increased soil moisture deficits, following an increase in temperature and a decrease in precipitation under climate change (Morales et al. 2007). However, other factors, such as fire frequency and CO2 fertilisation, could also interact with water limitation, leading to complex ecosystem responses (Fyllas and Troumbis 2009; Keenan et al. 2011). Disentangling the role of water limitation, fire and CO2 in forest ecosystem processes is important in order for their current dynamics and future risks to be better understood. Field studies, specifically designed to constrain the way such processes are simulated in vegetation dynamics models, could increase our understanding of forest function under current conditions and enhance our confidence in the projections of their future state.

In addition to the above, the role of inter- and intra-specific functional trait variation has been recently highlighted as an important component that needs to be incorporated into vegetation dynamics model (Scheiter et al. 2013; Fyllas et al. 2014; van Bodegom et al. 2014; Sakschewski et al. 2015). Traditionally, parameterisation of species and/or plant functional types (PFTs) is based on the use of some “average” or “appropriate” mean trait value, for characteristics that have a direct effect on the regeneration, the growth and the mortality of the simulated species. For example, specific leaf area has been used as a parameter to differentiate the turnover rate of leaf biomass between PFTs (Sitch et al. 2008), or as a parameter to describe the architecture (in terms of foliage area/biomass) of different tree species (Bugmann 2001; Fyllas et al. 2007). The selection of one “average” trait value could lead to “static” model behaviour as the population variability in the response of species/PFTs is a priori restricted, just because of the constant value given to some key functional characters (Fyllas et al. 2012, 2014). Ignoring this intraspecific variability is not in agreement with what is observed in real plant communities and comprises a key element of natural selection and evolution. In addition, depending on the way vegetation dynamics models are built, variability in some functional characters could affect more than one simulated process, through either direct or indirect routes. For example, given that most vegetation dynamics models include a “carbon starvation” mortality term, the influence of a specific growth parameter on model behaviour could be manifested directly through growth and indirectly through mortality. Ignoring functional variation in simulated plant communities could be an important bias, especially when projecting vegetation dynamics under climate change conditions, where alternative “functional configurations” could lead to viable life strategies.

Individual-based modelling is a tool widely used to simulate vegetation dynamics (Grimm et al. 2006). Forest gap dynamics models are a special group of individual-based models that follow the life of each tree in a stand and simulate key processes of interest such as regeneration, competition and mortality (Bugmann 2001). Gap dynamics models have a long history in modelling forest ecosystems structure and function with applications all over the world (Shugart 1984; Bugmann 2001; Fyllas et al. 2007; Ngugi et al. 2013). Because these models are based on empirical equations of growth and mortality, they provide reasonable approximations of stand growth, succession and disturbance patterns. Furthermore, as these models focus on individual tree performance, they provide an excellent framework to consider intra-specific trait variability and explore the potential shifts in species- or community-level trait variation under changing environmental conditions. However, to our knowledge there is no study that incorporates trait variability in the widely used forest-gap dynamics modelling framework.

In this study, we combine the GREFOS forest gap dynamics model (Fyllas et al. 2007; Fyllas and Troumbis 2009) with a novel tree ring width and wood density data set to account for growth and mortality intraspecific variability in simulations of forest dynamics and to explore for potential shifts in species composition and functional traits under warmer and drier conditions. In particular, using the tree ring width data set, we initially derive species-specific diameter growth curves. We also use the observed inter- and intravariability in wood density to infer individual tree longevity. In addition, by combining the tree ring width and wood density data sets, we identify a growth versus longevity trade-off that is subsequently incorporated in the model by hardwiring a relationship between two key functional traits, i.e. the maximum diameter growth (G m) and the wood density (D W). We then apply the model across three 500-m-long elevation gradients in three mountainous areas in Greece to:

  1. 1.

    Evaluate the predictive ability of the model and compare the “static” (single trait values) versus the “plastic” (varying trait values) model set-up under current climatic conditions.

  2. 2.

    Explore how the species composition and functional variation of these forests will respond to a gradually warmer and drier climate.

Materials and methods

Study sites and dominant tree species

Three study areas (Fig. 1) were selected to parameterise the model and validate its predictive ability. In each study area, three 30 × 30 m2 plots have been established as part of the Mediterranean Forests in Transition (MEDIT) project, where a suite of plant functional traits and tree ring width data have systematically been measured. Soil texture and depth are also available for each plot. In each plot, all trees above 1 m have been identified and the diameter at breast height has been measured for all trees above 1.3 m. In all areas, the plots are found across an elevation range of ca 500 m (Table 1). The first study area is located in the southern part of Mount Taygetos, Peloponnese. Pinus nigra and Abies cephalonica are the dominant tree species in this area, with the pine dominating the lower elevations and the more disturbed sites of the region. Soils are rather shallow with a high sand content (sandy loam). Mean annual temperatures range between 9.2 and 13.1 °C and annual precipitation between 850 and 950 mm. The second study area, the driest of the three, is located in the Agrafa region, Southern Pindos, and it is dominated by Quercus frainetto at lower elevations and Abies borisii-regis at higher elevations. At lower elevations, soil is sandy clay loam shifting to clay loam at higher elevation. Across this elevation, gradient mean annual temperature ranges from 10.3 to 14.2 °C and annual precipitation from 775 to 864 mm. The last study area, the wettest of the three, is found in the northern part of the Pindos range, and the dominant species are Pinus nigra and Fagus sylvatica, with the beech restricted at higher elevations. Soils are deeper here with higher silt content (sandy loam). Temperature ranges from 7.6 to 9.7 °C and precipitation from 926 to 962 mm per year. At Mt Taygetos in P. nigra-dominated stands, 11 surface fires have been recorded over the last 165 years (Christopoulou et al. 2013), while in P. nigra-dominated stands at Pindos North (Valia Kalda), fire frequency is likely lower with 8 fires recorded over a period of 815 years (Touchan et al. 2012).

Fig. 1
figure 1

Location of the study sites

Full size image
Table 1 Study site description
Full size table

Model description

A detailed description of GREFOS model is provided elsewhere (Fyllas et al. 2007). The model has been developed, parameterised and used for forest species found in the north-eastern part of the Mediterranean area (Fyllas et al. 2007, 2010; Fyllas and Troumbis 2009; Kint et al. 2014). GREFOS takes into account the discrete life history strategies (LHS) of Mediterranean tree and shrub species (Pausas 1999), by assigning a distinct recruitment density and resprouting capacity to each LHS. Regeneration in the model is based on empirical relationships between stand-level LAI and recruitment density (Fyllas et al. 2008, 2010), where a maximum threshold of LAI “ceases” the establishment of saplings due to light limitation. Individuals are competing for light through a height-based hierarchy with taller trees shading all smaller ones. A daily soil water balance model is used to calculate relative water content (θ) and subsequently the annual drought duration in order to adjust growth (Granier et al. 1999; Fyllas and Troumbis 2009). Evaporation is estimated following the Priestley and Taylor (1972) method, while the pedotransfer functions of Wösten et al. (1999), along with site-specific soil texture and depth measurements, are used to calculate soil water retention and release parameters.

As in most forest gap dynamics models, annual tree growth is estimated through the concept of optimum diameter increment, i.e. the growth that an individual of a certain species and size can reach under no-resource limitation or competition (Moore 1989; Bragg 2001; Risch et al. 2005). The “actual” diameter increment is subsequently estimated by adjusting the optimum diameter growth, as a function of the abiotic (temperature and water availability) and biotic (shading) conditions that prevailed for a given time for each tree in the stand. Here we use a novel ring width data to estimate the parameters (and their variation) of a commonly used optimum growth equation (Zeide 1993), as described in “Optimum growth curve and intraspecific plasticity” section.

In GREFOS mortality has three components. The growth-related component (“carbon starvation”) estimated as a function of a tree’s past growth, the background (“intrinsic”) mortality representing species longevity and the fire-related mortality, which is linked to species LHS. Background mortality is usually estimated in forest gap models through species longevity or maximum size. Here we use wood density as a proxy for background mortality (Martínez-Vilalta et al. 2010). Intra- and inter-specific variation in wood density is incorporated in the model based on a novel wood density data set, as described in the “Background mortality” section. Species with a higher wood density generally have lower growth and mortality rates (Reich 2014). This growth versus survival trade-off has been incorporated in the model by hardwiring a relationship between maximum diameter growth rate and wood density based on the combined analysis of the tree ring width and wood density data sets (“Growth–longevity trade-off” section).

Optimum growth curve and intraspecific plasticity

In order to parameterise the diameter growth curve, a minimum of 30 cores were taken from trees of each dominant species at each study area (Fyllas et al. 2017). All cores were collected at breast height with a 5-mm increment borer. In the laboratory, the cores were glued on channelled wood, dried at room temperature and sanded with progressively finer-grade abrasive paper until cells were clearly visible under magnification. All samples were visually cross-dated using visual recognition of tree ring patterns and lists of marker years (those with narrow rings) (Yamaguchi 1991). Tree ring widths were measured to 0.01 mm using Time Series Analysis and Presentation (TSAP) software package and LINTAB measuring table. Raw ring width series were synchronised according to their Gleichläufigkeit score, which represents the overall accordance of two series t values, which are sensitive to extreme values such as marker years and the cross-date index (CDI), which is a combination of both (Rinn 2003). Finally, the COFECHA software was used to perform a data quality control and to evaluate the cross-dating (Grissino-Mayer 2001).

These data were subsequently used to estimate the parameters of an optimum growth curve (Zeide 1993). We considered optimum growth to be species specific, and thus, we estimated the parameters of the curve for each species using data from all available sites. As in Bragg (2001), we assume that individuals growing at the highest rate for a given diameter class provide an adequate estimate of size-specific optimal growth. In this version of the model, the optimum growth of an individual is described by the equation (Zeide 1993):

$$g = G_{\rm m} e^{{- 0.5\left({\frac{{\log \frac{D}{{D_{o}}}}}{{D_{b}}}} \right)^{2}}}$$
(1)

where G m is the maximum radial growth rate (mm a−1) at the peak of the log-normal growth curve, D o is the diameter at breast height (D) associated with the maximum growth rate, and D b determines the width of the curve.

We fitted nonlinear least square regression models using the R programming language and the nls library (R Development Core Team 2015) to estimate the species-specific G m, D o and D b along with their confidence intervals (Table 2). In order to account for intraspecific growth variability in the model (“plastic set-up”), a normal distribution is used to randomly assign the growth parameters for each tree of a certain species using the parameter estimates in Table 2. The first generation of simulated trees is randomly initialised based a normal distribution that follows our observations. Subsequent generations inherit growth characteristics from a normal distribution that is updated each year based on the parameters of the surviving trees.

Table 2 Growth curve parameter estimates along with their confidence intervals for each tree species
Full size table

Background mortality

The background mortality component (Π R ) was parameterised based on the equation reported in Martínez-Vilalta et al. (2010), for tree species found in Spain under a similar range of climatic conditions:

$$\Pi _{R} = 0.51e^{{( - 3.56xD_{\text{W}} )}}$$
(2)

where wood density [D W (g cm−3)] is considered the main predictor of annual background mortality rate.

We collected stem wood samples from individuals within our study sites to estimate species-specific wood density values along with their confidence intervals. D W was calculated for each sample using the water displacement method. A container was filled with water and placed on a digital balance. A dried wood sample (48 h at 60 °C) that was weighted beforehand was then sunk into the container until completely immersed. The volume of the wood sample was estimated from the water displacement. Similar to growth, the inclusion of intraspecific variability for mortality in the model was applied through a species-specific normal distribution with mean equal to D W and standard deviation equal to D Wsd (Table 3).

Table 3 Mean wood density estimates along with their confidence intervals for each tree species
Full size table

Growth–longevity trade-off

Species with higher wood density generally have lower growth and mortality rates (Reich 2014). This growth versus survival trade-off has been incorporated in the model by hardwiring a relationship between maximum radial growth rate and wood density, based on the analysis of the entire MEDIT data set (Fyllas et al. 2017). To derive the G m = f(D W) relationship, we analysed tree ring width and wood density data for nine species in 35 plots across Greece. In this analysis, the optimum growth curve (Eq. 1) was fit with nonlinear least square regression models for all individual of a tree species at site. Then, the species- and site-specific G m, D o and D b estimates were regressed against the average D W of each species at each site. A statistically significant relationship was identified only for G m and D W (Fig. 2), with G m decreasing with D W and supporting the hypothesis of a growth–longevity trade-off. It should be noted that this relationship holds only across species, as the number of samples available was not adequate to validate it within species.

Fig. 2
figure 2

Relationship between wood density (D W) and the growth curve parameters for nine forest tree species found at different sites in Greece. A statistically significant relationship was only identified between maximum radial growth (G m) and D W (G m = 9.68–5.833D W, p = 0.004, R 2 = 0.20)

Full size image

Simulation set-up

At each study area, the model was set up to simulate stand dynamics along an elevation (500 m long) gradient, with an step of 50 m. Soil depth was set to 2.0 m in all study sites. All simulations started from bare ground and lasted for a 1000-year-long simulation period. In this study, the fire component of the model has been disabled in order to explore for merely the effects of drought. At each study site, only the two species known to occur abundantly are allowed to establish. It should be noted that these two species are the dominant elements of vegetation accounting for more than 80% of the total basal area in our study plots. Two climate scenarios were used, namely: (1) the baseline (BL) climate representing the current climatic conditions with the climate of the twentieth century at each study area randomly repeated for the simulation period and (2) the IPCC A1B climate change (CC) scenario with an approximately 3 °C increase in temperature and 20% reduction in precipitation taken as one of the intermediate projections cases from an ensemble regional climate model projections for the Mediterranean area (Gualdi et al. 2013). The baseline climate was extracted from the E-OBS gridded climatology (Haylock et al. 2008) for the time period between 1950 and 2013. Across the elevation gradients, temperature was corrected with a lapse rate of 6.5 °C km−1. Precipitation was assumed not to change with elevation. In both cases, a spin-up period of 500 years, during which the observed 1950–2013 climate was randomly replicated, was used until vegetation reached an equilibrium with climate. For the CC scenario during the spin-up period, climate was assumed to be similar to BL conditions, followed by a transient period of 100 years during which temperature and precipitation anomalies were linearly applied until climate stabilised after simulation year 600.

The model was applied following a “static” (no intraspecific variability) and “plastic” (with intraspecific variability) set-up, and 30 iterations were performed for each elevation and climate scenario. The steady-state (simulation year 600–1000) average basal area of each species under baseline conditions was estimated and compared with local observations. Changes in the species average values of the two key traits, G m and D W, with time and climate scenario were also explored. Potential shifts in these traits indicate the way species could adapt to warmer and drier conditions, by adjusting the two traits that are directly linked to their growth and mortality.

Results

Simulations under baseline conditions adequately captured the range of species distribution across the elevational gradient of the study sites. At Mt. Taygetos, P. nigra is more abundant at lower elevations, while A. cephalonica increases its contribution with elevation (Fig. 3—left panel). Simulated basal area was reasonably well estimated for P. nigra but underestimated for A. cephalonica in both the “static” and the “plastic” model set-up (Fig. SM1). The “static” model set-up yielded a higher P. nigra abundance compared to the “plastic” set-up across all elevations, while A. cephalonica achieved a higher basal area under the “plastic” set-up (Fig. SM1). Following the climate change scenario, the model simulated an uphill shift of P. nigra across the whole elevation gradient and a significant decrease in A. cephalonica using the plastic model set-up (Fig. 3—right panel). The decrease in A. cephalonica was stronger under the static model parameterisation (Fig. SM2).

Fig. 3
figure 3

Simulated average basal area of the dominant forest species across the elevation gradient of Mt Taygetos under baseline climate conditions (left panel) and climate change conditions (right panel). The bars show the average basal area of 30 simulations, and the lines show the 95% confidence intervals. Triangles indicate P. nigra and diamonds A. cephalonica basal area field measurements at different elevations. The graph summarises simulations made using the plastic model set-up

Full size image

At Agrafa in the southern part of Pindos, under BL conditions Q. frainetto is more abundant at lower elevations, but with increasing elevation, A. borisii-regis becomes the dominant element of vegetation (Fig. 4—left panel). Simulated basal area was reasonably estimated for both species especially under the “plastic” model set-up (Fig. SM3). The applied climate change scenario leads to a total replacement of the A. borisii-regis by Q. frainetto across the elevation gradient of this area using the plastic (Fig. 4—right panel) traits parameterisation. The inclusion of trait plasticity did not yield simulation of better A. borisii-regis performance under CC condition, and a complete replacement by Q. frainetto was also simulated (Fig. SM4).

Fig. 4
figure 4

Simulated average basal area of the dominant forest species across the elevation gradient of Agrafa (Pindos South) under current climate conditions (left panel) and warmer and drier conditions (right panel). The bars show the average basal area of 30 simulations, and the lines show the 95% confidence intervals. Triangles indicate Q. frainetto and diamonds A. borisii-regis basal area field measurements at different elevations. The graph summarises simulations made using the plastic model set-up

Full size image

At Valia Kalda in the northern part of Pindos, the observed vegetation transition with elevation was realistically replicated under BL conditions. P. nigra dominates the stands at lower elevations, while F. sylvatica is more abundant at elevations above 1350 m (Fig. 5). The steady-state standing basal area was accurately simulated for both species, with small differences between the static and the plastic parameterisation (Fig. SM5). At this region, CC simulations suggest a complete replacement of F. sylvatica by P. nigra across the simulated elevation gradient using both the plastic (Fig. 5) and the static model set-up (Fig. SM6).

Fig. 5
figure 5

Simulated average basal area of the dominant forest species across the elevation gradient of Valia Kalda (Pindos North) under current climate conditions (left panel) and warmer and drier conditions (right panel). The bars show the average basal area of 30 simulations, and the lines show the 95% confidence intervals. Triangles indicate P. nigra and diamonds F. sylvatica basal area field measurements at different elevations. The graph summarises simulations made using the plastic model set-up

Full size image

We additionally explored for potential shifts in the key traits that were used to predict growth and mortality of the four tree species, i.e. maximum growth rate (G m) and wood density (D W). The long-term trends of the average G m and D W for each study area across elevation is presented in Supplementary Material (Figs. SM7–SM12). In general, under current conditions, average species D W decreased and G m increased with time, suggesting that individuals with greater growth are performing better and thus drive the average trait values of the stand. The same trend was also observed in CC simulations with some shifts in the average species trait values. The steady-state average values of G m and D W under BL and CC conditions are summarised in Figs. 6, 7 and 8.

Fig. 6
figure 6

Average maximum growth rate and wood density across the Mt Taygetos elevation gradient under current (black symbols) and climate change conditions (grey symbols). Continuous lines indicate smoothing splines under current climate conditions, and broken lines indicate smoothing splines under warmer and drier conditions. Grey areas indicate 95% confidence intervals

Full size image
Fig. 7
figure 7

Average maximum growth rate and wood density across Agrafa (Pindos South) elevation gradient under current (black symbols) and climate change conditions (grey symbols). Continuous lines indicate smoothing splines under current climate conditions, and broken lines indicate smoothing splines under warmer and drier conditions. Grey areas indicate 95% confidence intervals

Full size image
Fig. 8
figure 8

Average maximum growth rate and wood density across Valia Kalda (Pindos North) elevation gradient under current (black symbols) and climate change conditions (grey symbols). Continuous lines indicate smoothing splines under current climate conditions, and broken lines indicate smoothing splines under warmer and drier conditions. Grey areas indicate 95% confidence intervals

Full size image

At Mt. Taygetos, P. nigra’s average wood density and maximum growth showed no clear trend with elevation (Fig. 6). Under climate change conditions, an overall small decrease in D W, associated with an increase in G m for P. nigra, was simulated. For A. cephalonica, a small decrease in D W with elevation was simulated, especially under BL conditions. At Agrafa, D W increased and G m decreased with elevation for the more drought-tolerant Q. frainetto under BL conditions. This trend was not as strong under CC conditions (Fig. 7). No clear trend in D W or G m was simulated for A. borisii-regis although at higher elevations, the average D W value was smaller compared to lower altitudes. At Valia Kalda (Fig. 8), the elevational shifts of simulated D W and G m were more pronounced. A lower G m and a higher D W were achieved at higher elevation for the more drought-tolerant P. nigra, with the opposite trend found for F. sylvatica. Overall, G m increased and D W decreased under CC conditions for the more drought-tolerant species, while the opposite was true (decreasing G m and increasing D W) for the less drought-tolerant species across all sites.

Discussion

This study presents the parameterisation of a forest gap model, with novel tree ring width and wood density measurements, aiming to better understand and to project the dynamics of typical mountainous forests in Greece. The inclusion of species-specific parameters leads to a realistic prediction of the distribution patterns and the basal area of the dominant tree species, observed in three study sites under current climate. In general, the predicted steady-state average basal area is accurately simulated for lower elevations and early successional species, but was slightly underestimated at higher elevation and late successional species. By accounting for intraspecific variability in growth and mortality through two functional traits (maximum growth rate and wood density), the model simulates a higher contribution of slow-growing species across all elevations. We suggest that vegetation dynamics models, and particularly those that are implemented at local scales, should be constrained with site-specific information that takes into account the variability of key functional traits.

The inclusion of growth and mortality plasticity is an important aspect of this work. Currently, there is great effort to include functional trait variation into vegetation dynamics models, in order to account for the potential plasticity in the ecosystems’ response to climate change (Scheiter et al. 2013; Fyllas et al. 2014; van Bodegom et al. 2014). In this study, we present a simple method to implement trait variation within existing forest gap dynamics models, based on the measured intraspecific variability in functional traits associated with tree growth and longevity. Our approach is based on the inclusion of a fundamental trade-off between growth and survival. Wood density is selected as the core trait to represent this trade-off. Lower wood density is related to higher growth and mortality rates both within and between the species of interest. On the other hand, higher wood density is associated with smaller maximum growth and mortality rates. By parameterising this theoretical trade-off with species-specific measurements of tree growth and wood density, the model adequately captures the vegetation dynamics in our three study sites. Additional axes of functional trait variation that represent ecophysiological trade-offs could be potentially integrated in this modelling framework in future studies. These could relate to species hydraulic properties and/or response to disturbance (Pausas 1999; Sánchez-Gómez et al. 2006).

In all study sites, CC simulations suggest an upward shift in the dominance patterns, with more drought-tolerant species (Pinus nigra and Quercus frainetto) increasing their contribution with elevation. The inclusion of trait plasticity does not significantly alter the CC simulations outcomes, with the only exception at Mt Taygetos, where the A. cephalonica was more persistent at higher elevations when trait plasticity was included. The overall trend of shifting dominance patterns with CC is related to the drought tolerance of the studied species. Compared to A. borisii-regis and F. sylvatica, the higher elevation dominant species at South and North Pindos, A. cephalonica is considered more drought tolerant, and thus, it maintains its contribution to the stand’s basal area under CC conditions. It should be noted that species response to drought is simulated in this version of the model by counting the number of days with water stress that feeds into a species-specific drought response function (Granier et al. 1999; Fyllas and Troumbis 2009). Hence, there is no direct effect of trait plasticity to a species and/or individual’s drought tolerance. Various studies suggest a relationship between cavitation resistance (Hacke et al. 2001) and drought tolerance (Poorter and Markesteijn 2008; Preston et al. 2006, but see Hoffmann et al. 2011) with D W, and thus, such a relationship could be implemented in future versions of this modelling framework.

In this study, simulated shifts in species dominance were also associated with changes in the two functional traits used to describe variation in growth and mortality. In all three study sites, the CC scenario resulted in higher G m of the drought-tolerant species (P. nigra and Q. frainetto) and lower D W. Shifts in G m and D W were not as clear for the less drought-tolerant species, with the exception of F. sylvatica at North Pindos, where a systematic decrease in G m and increase in D W were simulated. These outputs suggest that across the studied elevation gradients, species that are able to tolerate longer dry periods could increase their growth rate in contrast to less drought-tolerant species. It should be noted that our modelling framework not only is taking into account the uncertainty of vegetation processes (Reyer et al. 2016) by including variation in growth and mortality, but also enables tree populations to shift their functional characteristics. This is particularly important in terms of adaptation to changing biotic and abiotic conditions.

Empirical evidence of species upward shifts (Peñuelas and Boada 2003; Moser et al. 2010; Pauli et al. 2012) has been documented, and modelling exercises identify the vulnerability of Mediterranean ecosystems to climate change (Morales et al. 2007; Fyllas and Troumbis 2009; Hickler et al. 2012). Simulations in this study suggest that climate change could lead to significant shifts of species distribution in mountainous Mediterranean forests. The importance of considering intraspecific variability for modelling purposes is highlighted here. Although in two of the study sites incorporating trait variability did not “enhance” the ability of the least drought-tolerant species to adapt to a changing climate, in Mt Taygetos the response of A. cephalonica to drier conditions was more gradual under the plastic model set-up. Such responses will be dependent on local environmental conditions as well as the ecophysiological ranges of species performance. Field studies that quantify these ranges as well as the way functional traits coordinate and interact to form fundamental ecological strategies could help us better parameterise models of vegetation dynamics.

Conclusions

This study presents the application of a forest gap dynamics model to explore the potential effects of drought on the dynamics of mountainous Mediterranean forests in Greece. Emphasis was given on incorporating intraspecific variability in growth and mortality. Simulations under climate change conditions suggest an upward shift of the more drought-tolerant species. These changes are also accompanied by intraspecific shifts in two key functional traits that express the growth and mortality patterns of the tree species. In general, populations of more drought-tolerant species increase their maximum radial growth and decreased their wood density in contrast to less drought-tolerant species. Incorporating trait variability and accounting for fundamental ecological trade-offs in vegetation dynamics models could increase the realism in projecting the fate of forest ecosystems under global change conditions.