Shifts of forest species along an elevational gradient in Southeast France: climate change or stand maturation?




Recent vegetation changes in mountain areas are often explained by climate warming. However, effects of land-use changes, such as recolonization of abandoned pastures by forest, are difficult to separate from those of climate change. Even within forest belts, changes in stand structure due to forest management and stand maturation could confound the climate signal. Here, we evaluate the direction and rate of plant species elevation shifts in mountain forests, considering the role of stand dynamics.


Forests in the plains and mountains of Southeast France.


We compared floristic data from the French National Forest Inventory collected in the 1980s and 1990s. They provided a large-scale (30 985 plots) and representative sample of vegetation between 0 and 2500 m a.s.l. Species response curves along the elevation and exposure gradients were fitted with a logistic regression model. In order to assess the effect of changes in successional stages of the forest stands, we compared plant species shifts in the whole set of stands with those solely in closed stands.


A total of 62 species shifted downward, whereas 113 shifted upward, resulting in a significant upward mean shift of 17.9 m. Upward shifting species were preferentially woody and heliophilous, suggesting a role for forest closure and maturation in the observed changes. Excluding all open forest stages from analyses, the upward trend became weaker (−3.0 m) and was not significant. Forests of the study area have undergone closure and maturation, more strongly at lower altitudes than at higher ones, producing an apparent shift of species.


In the mountain relief of Southeast France, changes in the successional stages of stands appear as the main cause of the apparent upslope movement of forest species. Since a similar trend of forest maturation exists in large areas throughout Europe, forest dynamics should be better taken into account among the causes of vegetation changes before inferring any climate change effect.

Flora Europaea

(Tutin et al. 1968)


Mountain regions have been identified as crucial areas to test ecological and evolutionary responses of biota to geophysical influences (Körner 2007), and thus to study the impact of global warming on vegetation. Compared to lowlands, mountains are characterized by a steep gradient of temperature and fewer obstacles to plant migration, such as infrastructure or agricultural fields. Thus, species should have to move over smaller distances and in a given habitat to follow their temperature range, allowing mountain plants to migrate more rapidly when the climate changes. Indeed, mountain areas were amongst the earliest systems where effects of recent climate change were detected (Grabherr et al. 1994). Because of this high sensitivity to temperature increase, and the high diversity and rate of endemism of mountain ecosystems, the potential impact of climate change on biodiversity in mountains is particularly alarming, as underlined by the IPCC (Fischlin et al. 2007). For instance, bioclimatic models project extreme migrations and serious risks of extinctions in montane areas (Thomas et al. 2004).

Observed impacts of climate change on species distribution ranges suggest that a species response is already obvious (Walther et al. 2002; Parmesan 2006; Chen et al. 2011). In high mountains, most studies have focused on small areas, such as the flora of summits in the Swiss and Austrian Alps. Grabherr et al. (1994), Walther et al. (2005) and Holzinger et al. (2008) observed an increase in species richness on these summits. Pauli et al. (2007) monitored an alpine–nival ecotone in the Austrian Alps and showed an expansion of alpine species at their upper limit, while some nival species decreased in cover at their lower limit. Community changes towards species requiring a higher temperature were shown by Keller et al. (2000) and Walther & Grundmann (2001). Similarly, the re-inventory of a Swiss nunatak revealed that species from the lower vegetation belt have migrated to a glacier island (Vittoz et al.2008). Few studies have focused on the whole distribution range to measure shifts in the observed mean altitude of a species (Kelly & Goulden 2008) or in its modelled elevation optimum (Lenoir et al. 2008).

The drivers and causes of these vegetation shifts are still debated; recent climate change is often considered as the main factor. According to Kammer et al. (2007), upward shifts on mountain summits could alternatively be explained by re-colonization since the Little Ice Age (i.e., about 1850). Until now, most of vegetation responses attributed to climate warming were observed above forest belts, where direct anthropogenic effects in relation to land-use changes could also play an important role. Vegetation shifts in sub-alpine grasslands could be promoted by climate warming, but in addition, should be linked to pasture abandonments that have widely occurred during the last century (Vittoz et al. 2009). Many studies have focused on forest or treeline ecotones, showing an upward shift during the last decades (Kullman 2002; Beckage et al. 2008). However, original climatic treelines in the Alps were often largely lowered through historical human activities (Holtmeier & Broll 2007). Consequently, the current treeline advance in the Alps is at least accelerated, if not entirely controlled, by the decrease in management and grazing intensity during the last decades (Gehrig-Fasel et al. 2007; Albert et al. 2008; Vittoz et al. 2009).

Forest belts below the treeline offer several advantages for the detection of climate change impacts on plant distributions. They should be less prone to the confounding effects of a decrease in pasturing and subsequent forest recolonization. In addition, studies at the treeline have often concentrated only on changes in tree species, not considering understorey species. Yet, the lower strata of forest ecosystems are less influenced by changes in forest management than the tree stratum itself, of which the composition is strongly controlled by the choice of target species. Thus, understorey species could be more reliable indicators of climatic change than species of the tree canopy alone.

However, even within the forest belts, human activities and natural dynamics not linked to climate change could induce vegetation changes. Agricultural abandonment at lower altitudes widely occurred in European mountains during the last centuries (MacDonald et al. 2000). As a consequence, lower elevation forests are still biased towards early stages of the forest succession. Moreover, after having been overexploited until the 19th century, Western European forests show a general trend of increasing wood stock, due to an increasing lag between forest growth and wood uptake (Dupouey et al. 2009). This is especially relevant in mountain forests, where poor accessibility to the stands limits their exploitation. This phenomenon could involve a slow maturation and closure of forest, leading to long-term changes in vegetation towards more mesic and shady communities (Walther & Grundmann 2001), including more species of northern biogeographic affinity (Harrison et al. 2010).

Limited sets of permanent plots, which are mostly used in long-term vegetation studies, are especially prone to bias because they are often not representative of the entire range of forest successional stages. In these datasets, it appears difficult to disentangle the effects of stand ageing from those of other environmental changes, as they are confounded, both developing over time (Thimonier et al. 1994). Large resurveys of independent vegetation plots allow a better account of these forest stand changes, under the condition that they are controlled as part of the sampling scheme at each sampling period. National forest inventories (NFI), when they collect vegetation data, offer such advantages. Samples are drawn in order to be statistically representative of large areas, and methods of vegetation sampling are well documented and remain consistent over space and time.

Objectives of the study

Here, we analysed plant species shifts between two successive samples of the French NFI covering the plains and mountains of Southeast France, collected between the 1980s and 1990s during a period of strong warming (0.5 °C·decade−1) and using a stratified sampling protocol.

Our first objective was to determine the direction and rate of forest species shifts along the elevation gradient in the mountains of Southeast France. For that purpose, species presence was modelled according to elevation, exposure and sampling period. We then tried to identify the causes of the observed shifts. Two main causes were explored, climate change and trends in forest succession. Because the study region has undergone elevation-dependent changes in forest successional stages, leading to forest closure and changes in dendrometrical characteristics, we looked for their impact on species shifts.

First, we analysed plant species traits associated with ascending or descending species. Climate warming should favour the upward movement of light seed, evergreen and short life-cycle (herbaceous) species because of their expected faster response and migration potential (Walther et al. 2002; Lenoir et al. 2008). On the other hand, closure and maturation of forest stands should promote a decrease in light-demanding, shrubby, pioneer species, and an increase in shade-tolerant, more water-demanding, ancient woodland species, often hemicryptophytes or geophytes (Hermy et al. 1999).

Second, we compared the observed species shifts in the whole forest area, including a variety of successional stages, with those measured in closed forests only, where the impact of forest dynamics should be lessened and climate change effects highlighted.


Study area

The study region covers an area of 60 000 km2 in Southeast France (2°32′37″–7°42′05″ E, 42°59′03″–45°21′37″ N) and in the southern half of Corsica Island (8°37′09″–9°23′41″ E,41°23′37″–42°20′22″ N; Fig. 1), from sea level to 4100 m a.s.l. We chose such a large region because even small truncations of the altitudinal gradient can lead to strong bias in the characterization of species niche parameters.

Figure 1.

Study area with relief and vegetation type. Bioclimatic limits according to the Council of Europe & Commission of the European Communities (1987).

A strong and concomitant elevation gradient of climate and forest vegetation belts (Quézel & Médail 2003)was early recognised and characterized by biogeographers: thermomediterranean climate is restricted to the eastern part of the study area, near sealevel, with a mean annual temperature above 15 °C and Olea europaea var. sylvestris, Pistacia lentiscus and Ceratonia siliqua as marker species; mesomediterranean and supramediterranean belts occur at intermediate elevations, with mean annual temperature above 8 °C, and are dominated by Quercus ilex, Q. suber, Q. pubescens, Castanea sativa, Pinus halepensis, P. pinaster or P. sylvestris; mountain and sub-alpine belts occur at the highest elevations, dominated by Fagus sylvatica, Abies alba, P. sylvestris, Picea abies, Larix decidua, P. uncinata and P. cembra.

The observed thermal gradient is 0.56 °C·100 m−1 elevation (decrease in mean annual temperature), according to Aimé & Sarrailh (1972). The warming trend in the study region over the 1971–2000 period was 0.5 °C·decade−1 for mean annual temperature. A careful study of 71 meteorological stations in our study region, of which only 32 were retained after exclusion of dubious series following the filtering procedure of Caussinus & Mestre (2004), showed a homogeneous warming trend along elevation (Fig. 2) between sea level and 1675 m a.s.l. This confirms on a local scale the similarity of warming trends in mountain and plain regions of the Alps (Auer et al. 2007). Minimum and maximum temperatures displayed the same uniform trend with elevation.

Figure 2.

Mean annual temperature increase over the 1971–2000 period as a function of elevation at 32 meteorological stations located in the study area. The median 0.5 °C per decade temperature increase is marked with a horizontal line.

Forests of the study area were inventoried twice in the French NFI between 1981 and 2004. The first inventory was conducted between 1981 and 1989 (mean inventory year, weighted per area sampled each month: 1984.7). The second inventory cycle took place between 1992 and 2004 (mean weighted inventory year: 1998.8).

Forest cover was 33% of the region at the first inventory and 36% at the second one, and extended from sea level to 2500 m a.s.l. Indeed, forest cover had started to increase since the 19th century; it was estimated to be only 19% in 1878 (Douguédroit 1981). In the study area, forest structure also revealed heterogeneous land-use history along the elevation gradient, illustrated by a more recent forest closure at lower than at upper elevations.

Vegetation dataset

At each of the two successive inventories, an independent sample was drawn. The sample was stratified by combining ecological regions, forest types, administrative divisions and ownership types:

  1. Ecological regions: the area was divided into 66 small forest ecological regions, delineated according to geological and climate criteria (see Appendix S1).
  2. Forest types: based on an aerial photograph analysis, the zone was stratified according to the main tree species and forest structure (14 initial types, belonging to the following main categories: coppice, high stand, coppice with standards, young plantation or open stand, including woody heathland, scattered and sparse woodlands).
  3. Administrative divisions: inventories were conducted independently in each of the 11 French departments, and stratified as a function of ownership type (private, communal and state forest).

In each sampling stratum, plots were randomly selected on a regular grid in order to reach a theoretical sampling effort of one plot per 130 ha of forest on average. This objective was globally reached, with an observed sampling intensity of one plot every 127.3 ha at the first inventory (15 754 plots spread over 2 005 998 ha) and 141.2 ha at the second one (15 231 plots spread over 2 150 261 ha). Sample stratification was entirely rebuilt at each of the two inventories, resulting in independent and distinct plot locations. In the following analyses, each of the 30 985 plots was given a weight equal to the total area of the stratum it belonged to, divided by the total number of plots belonging to this stratum. This improved the representativeness of plots. But, in practice, due to the large number of plots, this weighting process had little effect on the final results. The plots were distributed between 0 and 2250 m a.s.l. (99.9 weighted percentile of the elevation distribution) at the first inventory and 0 and 2240 m a.s.l. at the second one. The distribution of forest cover along the elevational gradient was similar among the two sampling periods. Weighted average elevations of the plots were not significantly different between the two inventories (693.8 m a.s.l. and 701.6 m a.s.l. at the first and second inventories, respectively) according to a Student's t-test (= 0.14). However, at the second inventory, sampling pressure had decreased in high-altitude forests leading to larger weights for the corresponding plots.

At each plot, several dendrometric and ecological variables were recorded. Among these, we used for the present study: geographical coordinates, date of sampling, elevation, aspect, age of dominant trees obtained by coring and botanical observations consisting of a list of all vascular species and terricolous bryophytes. Vegetation was recorded in a 6-m radius circle at the first inventory and in a 15-m radius circle at the second one. A total of 1945 taxa were observed in the whole sample (1536 and 1546 taxa at the first and second sampling inventories, respectively), among which 1137 were common to both samples. In total, 88.9% of the overall occurrences of taxa were identified at the species level or more precisely at the first inventory, and 90.2% at the second one. All taxa were assigned an identification reliability code of high, intermediate or low by NFI observers (respectively 45%, 23% and 32% of the taxa inventoried at the second inventory). The mean number of species per plot was 16.6 in 1985, increasing to 25.3 in 1999 because of a higher sampling plot area and, probably, a higher exhaustiveness of plant censuses due to better training of observers.

Data analyses

Data selection

Variations of plant census exhaustiveness between observers has been recognized as a recurrent bias in vegetation monitoring (Vittoz & Guisan 2007; Archaux et al. 2008). Here, several strategies were used in order to minimize the effects of such a bias between the two sampling periods. First, we only retained the taxa that had been noted in at least 50 plots at both sampling periods (386 taxa). This threshold frequency has also been suggested as a general minimum value to derive acceptable ecological response curves with logistic regression (Coudun & Gégout 2006). Among these 386 taxa, we only selected those identified at least at the species level (341 species) and noted by NFI observers with a ‘high’ identification reliability code on both dates (252 species). Finally, we studied only changes in the position of the elevation and aspect optimum of species, not dealing with changes in global frequency or width of species response curves, which both depend on plot area and census exhaustiveness.

Estimation of species shifts

Contrary to studies conducted in permanent plot networks, where the variation in mean elevation of a species can be used as an indicator of its altitudinal shift, the observed mean elevation cannot be used for independent plots because samples do not have exactly the same altitudinal distribution at each inventory (Lenoir et al. 2008; Crimmins et al. 2011). Thus, in a first step, we modelled the species response to the elevational and exposure gradients and, in a second step, we studied the shift between the two sampling periods of the elevation optimum provided by the model.

For each of the 252 selected species, a quadratic logistic regression model was fitted to presence–absence data. The sampling weight previously defined was used to adjust for representativeness of each plot:

display math

where P, probability of species presence; elevation, elevation of the plot where the species was recorded; expo, exposure (cosine of the aspect of the plot); period, first or second inventory period; ε, error.

After elevation, aspect is an important topographic factor of the local climate in mountains, influencing presence–absence of species (Stage & Salas 2007; Ashcroft et al. 2008). We introduced the north–south exposure gradient in our model as the cosine of the aspect measured from north in the field (north: expo = 1, east and west: expo = 0, south: expo = −1) to take into account its effect on species distribution. Below a 6% slope, the aspect was not recorded in the field. In this case, a more spatially extended aspect value was extracted from a digital elevation model using ArcView 9.2, based on the eight closest neighbours in a 50 × 50 m grid.

The binary variable ‘period’ was introduced in the model, both in interaction with elevation and exposure, in order to test for changes in the optimum of each species, and alone, in order to control for variations of the global frequency of species from one inventory to another.

Among the 252 species whose distribution was modelled, we only retained those responding to the three following criteria:

  • Criterion 1: species displaying a significant global model (< 0.001). The global significance of the model was assessed by the likelihood ratio test. A low threshold value of probability was used because of the large number of tests made and the large sample size.

  • Criterion 2: species responding to elevation with a bell-shaped curve, i.e., with parameter a for square elevation both negative and significantly different from zero at the 0.05 level (likelihood-based confidence interval).

  • Criterion 3: species having an elevation optimum in the altitudinal range of our study area (0–2200 m a.s.l.) at both inventories. This constraint avoided extrapolating optimum positions outside the sampled range. For each species, the elevation optimum of the bell-shaped curve was calculated as:

display math

Then, for each retained species, we calculated the altitudinal shift as the difference between the optimum elevation at the two successive inventories (second minus first). Species displaying a positive or negative shift are hereafter called ascending and descending species, respectively. We tested significance of this shift in a type III test of the c parameter, comparing the full model with the same model without the elevation*period interaction (< 0.05).

Finally, the global rate of species shift along the altitudinal gradient was estimated as the mean elevation shift of all species selected in steps 1–3, divided by the average time elapsed between the two sampling periods (14.1 yr). The difference of the mean from zero was tested using a t-test. This average rate was also calculated over only those species displaying a significant shift.

The same procedure (steps 2 and 3) was applied to select species responding to exposure with a bell-shaped curve (< 0, < 0.05) and an optimum within the −1 to +1 range. We tested significance of the exposure shift in a type III test of the f parameter (< 0.05).

Species shifts and plant species traits

We looked for plant species traits associated with species shifts, testing which species traits were differently distributed between descending and ascending species. We studied Landolt's indicator values (Landolt 1977) for light (L), temperature (T), continentality (K), soil reaction (R), nitrogen availability (N), soil moisture (F), soil dispersion (D) and humus content (H). Landolt's indicator values were preferred among others because they were established for a region (Switzerland) closer to our study area. Then we analysed Raunkiaer's life forms, divided in deciduous phanerophytes, evergreen phanerophytes, deciduous nanophanerophytes, evergreen nanophanerophytes, woody chamaephytes, herbaceous chamaephytes, hemicryptophytes, geophytes and therophytes. These growth forms were also grouped into simpler categories: woody vs herbaceous and evergreen vs deciduous. Woody species were classified into tall trees (more than 12 m in height) or low trees and shrubs (less than 12 m), based on the maximum height to which they can grow. We extracted the following traits from the French Mediterranean forest flora (Rameau et al. 2008): (1) pioneer species or not; (2) species that can be found in immature stages of the vegetation succession or not, where immature stages included fallows, heaths and scrubs, thickets; (3) seed dispersal mode (by gravity, seed projection, wind, ants, birds or other animals); and (4) biogeographic affinity of the species, distinguishing between exclusively mountainous and ubiquitous species, the latter being found both in plains and mountains. We also used species light requirement from the same source. As this gave the same results as Landolt's L indicator value, this is not presented here. Mean seed masses were downloaded from the online Leda Traitbase (Kleyer et al. 2008) and log-transformed.

For quantitative traits, i.e., Landolt indicator values and seed mass, we tested the difference in the means between descending and ascending species using a t-test. For categorical traits (Raunkiaer's life forms, woodiness, deciduousness, pioneer habit, successional habit, dispersal mode and biogeographic affinity), we tested whether the trait was randomly distributed or not among descending and ascending species using Fisher's exact test of association. For each category of a trait, we calculated the mean elevation shift and tested its departure from zero with a t-test. Moreover, we looked at which categories were more linked to species shifts by calculating their odds ratio:

display math

where, u: proportion, among ascending species, of species belonging to the given trait category, d: proportion, among descending species, of species belonging to the given trait category.

Species with missing values for a given trait were excluded from the calculation of u and d for all categories of this trait. An odds ratio > 1 for a given trait category means that this category is composed of more ascending than descending species relative to the other categories of the same trait, and conversely for an odds ratio smaller than 1. Significance of departure of the odds ratios from 1 was tested using a chi-square test. Mean elevation shifts and odds ratios are two complementary statistics for each trait category. The first gives an estimated value of the shift amplitude for all species belonging to a given category, but depends strongly on the global shift. The second provides information on the distribution of ascending and descending species in this category, relative to the global number of ascending and descending species.

Species shifts and successional dynamics

Because French forests are undergoing both an expansion due to previous agricultural abandonment and a maturation due to a delay in forest renewal, the balance between forest succession stages has changed between the two inventories. We characterized these changes by calculating the percentage area of open forest, the mean basal area and the mean forest age per 100-m elevation class at each inventory. Open forests had been defined during the aerial photography analysis (see above). They included woody heath land and scattered or sparse woodlands, whereas coppices, high stands, coppices with standards and young plantations were classified as closed forests. Stand age was available in even-aged forests only, i.e., for 78% of the total sample. All calculations were made using area-weighted sums over the sample plots.

The overall percentage area of open forests decreased from 37% to 19% of the region (Fig. 3). The percentage area of open forests decreased at all elevations, but the decrease was much more pronounced at lower elevations. Below 1300 m a.s.l., it decreased from 39% to 20% between the two inventories, whereas it only decreased from 21% to 16% above 1300 m a.s.l. A parallel trend was observed for forest age, which showed a homogeneous and significant ageing of Mediterranean forests at lower elevations, where forests were youngest (Fig. 4). Below 1300 m a.s.l., mean age increased significantly from 52.4 to 56.4 yr between 1984.7 and 1998.8 (n = 20 390, < 0.001), indicating less harvesting and natural mortality than recruitment over the studied period. Above 1300 m a.s.l., the age did not increase significantly, from 102.0 to 104.8 yr.

Figure 3.

Percentage area of open forests by 100-m elevation class at the two inventories (1985: dotted line, 1999: full line).

Figure 4.

Average stand age (left) and basal area (right) by 100-m elevation classes at each inventory (1985: dotted line, 1999: full line), in the whole sample (up) and in closed forests only (down). Asterisks designate elevation classes with values significantly different between 1985 and 1999 according to a Student t-test. The meaning of the asterisks is the same as in Table 2.

Because these changes in forest openness and age were not uniformly distributed along the elevation gradient, the response of vegetation in closed and open forests could be different from that observed previously in the entire sample. To analyse the interaction between successional dynamics and observed altitudinal species shifts, calculation of species optima and analyses of plant traits were separately run on two different datasets: the whole sample including all forest types sampled by NFI, and the subset of 24 368 plots only in closed forests, considered as established forest habitat. Open forests, considered as pioneer stage, were not separately analysed because they constituted only a minor part (6617 plots) of the total sample. The species responses after open forests were removed from the analysis indicate the role of open forests in the species shifts. In the case of no effect of changes in the altitudinal distribution of successional stages of forest stands, the same shifts should be observed in both the entire sample and closed forests alone. All statistical calculations were made using the SAS software, v. 9.1 (SAS Inc., Cary, NC, USA).


Estimation of species shifts

Out of the 252 species with an occurrence above 50 at each inventory and with good identification reliability, 251 displayed a significant global model of distribution (< 0.001). Among these, 175 had both a unimodal (bell-shaped) response curve along the altitudinal gradient (a different from 0 at the < 0.05 level and a negative) and a calculated optimum, Optelev, between 0 and 2200 m a.s.l. Seventy-four out of these 175 species showed a significant shift of the optimum elevation (Table 1, and see Appendix S2). For illustration, two examples of observed and modelled data are presented in Fig. 5, one of an ascending species and the other of a descending species.

Figure 5.

Observed frequency distribution (dotted line) and modelled probability of presence (full line) at each inventory (1985 in grey and 1999 in black) along the elevation gradient for two selected species: one shifting downward (Quercus petraea, left) and one shifting upward (Teucrium chamaedrys, right). The modelled probability of presence was calculated using the observed average value of exposure for each species at each inventory.

Table 1. Number of species responding to elevation in a quadratic logistic model and displaying an altitudinal shift between the last two national forest inventories (1985–1999) in all forest types and in closed forests alone
 Number of species displaying a global significant responseNo response to elevation or optimum outside the sampled rangeUnimodal response curve to elevation and optimum within the sampled range
Descending species all species (sign. shift)Ascending species all species (sign. shift)
All forests2517662 (18)113 (56)
Closed forests only2316763 (23)101 (39)

The total number of ascending species was 1.8 times higher than that of descending species (113 and 62 species, respectively) and this difference was significant (Fisher's exact test, < 0.001). When considering significant shifts only, three times more species displayed an upward shift than a downward shift and this difference was also significant (56 and 18 species, respectively, Fisher's exact test, < 0.001; Table 1, Appendix S3).

Over the 175 species responding to elevation in a unimodal way, the mean altitudinal shift was positive and significantly different from zero (+17.9 m, < 0.05). Interestingly, species with an optimum below 1300 m a.s.l. displayed, on average, a much larger shift (+35.4 m, n = 113 species, < 0.001; Appendix S3) than those having their optimum above 1300 m a.s.l. (n = 62, mean shift not significant). The shift even tended to reverse for the few species having an optimum elevation above 1800 m a.s.l. When only considering the 74 species shifting significantly, the mean shift was +47.3 m (significantly different from zero, < 0.05)

Along the exposure gradient, only 45 species displayed both a bell-shaped distribution (d different from 0 at the < 0.05 level and d negative) and a calculated optimum, Optexpo, between −1 and 1, of which 21 shifted northward and 24 southward. Among these, only nine showed a significant shift of their exposure optimum, with three species shifting significantly northward and six southward. The mean exposure shift was not significantly different from zero. The analysis will not be developed further on exposure shift since the response appears negligible.

Species shifts and plant species traits

We analysed differences in species traits according to their shift in elevation. Odds ratio and mean elevation shifts for categorical traits are listed by trait category in Table 2. Mean values of quantitative traits for descending and ascending species are presented in Table 3. Values of the main traits are given in Appendix S2 for species displaying both a unimodal response curve to elevation and an optimum within the sampled range. Species optimum elevation in 1999 as a function of the optimum elevation in 1985 are presented for different traits categories in Appendix S3.

Table 2. Number of descending and ascending species (desc./asc.) between 1985 and 1999, odds ratio (o.r.) and mean species shifts for different plant trait categories, in all forest types (left) and only in closed forests (right)
 Species shift in all forestsSpecies shift in closed forests only
Number of species (desc./asc.)o.r.Mean (m)Number of species (desc./asc.)o.r.Mean (m)
  1. Significance of departure of the odds ratios from one was tested using a chi-square test. Whether the mean shifts were significantly different from zero was tested with a t-test (***≤ 0.001; **≤ 0.01; *≤ 0.05). Only species having available information for the given category are included in the calculations. Plant species traits from Rameau et al. (2008) and Leda Traitbase (Kleyer et al. 2008).

Raunkiaer's life form
 Phanerophytes48 (14/34)1.48+44.5**47 (18/29)1.01+1.2
Evergreen phanerophytes18 (9/9)0.51+1.318 (12/6)0.27**−9.0
Deciduous phanerophytes30 (5/25)3.24*+70.4**29 (6/23)2.80*+7.6
 Nanophanerophytes (incl. ev. and dec.)42 (9/33)2.43*+25.938 (12/26)1.47+2.2
 Chamaephytes32 (13/19)0.76+40.030 (7/23)2.36+17.1
Woody chamaephytes25 (9/16)0.97+50.1*24 (3/21)5.25**+50.2**
Herbaceous chamaephytes7 (4/3)0.39+1.86 (4/2)0.30−115.1
 Hemicryptophytes33 (15/18)0.59+7.031 (17/14)0.43*+3.5
 Geophytes19 (11/8)0.35*−88.0*17 (9/8)0.52−85.1*
 Therophytes1 (0/1)/+65.71 (0/1)/+186.0
 Herbaceous species60 (30/30)0.39**−22.755 (30/25)0.36**−33.5
 Woody species115 (32/83)2.59**+39.0***109 (33/76)2.76**+12.3
Tall trees32 (10/22)1.26+54.2**31 (12/19)0.98−9.0
Low trees and shrubs83 (22/61)2.13*+33.278 (21/57)2.59**+20.8
 Evergreen48 (19/29)0.44*+19.045 (19/26)0.47−1.1
 Deciduous71 (16/55)2.25*+49.3***67 (17/50)2.15+8.4
Dynamic stages
 Found in immature stages128 (42/86)1.52+22.2*119 (41/78)1.82+5.2
 Not found in immature stages47 (20/27)0.66+6.045 (22/23)0.55−24.9
Pioneer habit
 Non-pioneer species34 (11/23)0.48+38.3*33 (11/22)1.78−1.7
 Pioneer species16 (3/13)2.07+49.4*17 (8/9)0.56+2.7
Dispersal mode
 Gravity/autodispersed21 (6/15)1.26+21.920 (4/16)2.56+32.3
Gravity18 (6/12)0.98+20.117 (2/15)4.93*+37.0*
 Wind42 (14/28)0.97+53.4**37 (15/22)0.78+11.6
 Ants12 (6/6)0.46−25.49 (5/4)0.43−48.2
 Birds29 (8/21)1.36+30.426 (10/16)0.89+16.6
 Other animals45 (15/30)0.97+4.646 (16/30)1.10−25.3
 Ubiquitous142 (48/94)1.44+17.6135 (50/85)1.38−5.6
 Exclusively mountainous33 (14/19)0.69+18.829 (13/16)0.72+9.0
All species 175 (62/113) +17.9*164 (63/101) −3.0
Table 3. Mean value of quantitative traits for descending and ascending species, in all forests and only in closed forests
 Mean value of quantitative traits for descending/ascending species
In all forestsIn closed forests only
  1. Temperature, light and moisture indicator values from Landolt (1977), seed mass from Leda traitbase (Kleyer et al. 2008). Significance of the difference between means of descending and ascending species was tested using a t-test (*≤ 0.05; **≤ 0.01).

Light (Landolt)2.7/3.1* (n = 150)2.7/3.1** (n = 141)
Temperature (Landolt)3.4/3.9** (n = 150)3.6/3.9 (n = 141)
Moisture (Landolt)2.5/2.2 (n = 147)2.5/2.2* (n = 138)
Seed Mass (mg, Leda)179.8/187.5* (n = 107)327.0/106.0 (n = 98)

Landolt's indicator values were available for 150 species among the 175 studied (Table 3). Ascending species had a significantly (n = 150, < 0.01 in a Student's t-test) higher mean Landolt's indicator value for temperature (T = 3.9) than descending species (T = 3.4), meaning that they were more thermophilous. The difference in Landolt's indicator values for light was also significant (n = 150, < 0.05), with again a higher mean value for ascending species (L = 3.1) than for descending species (L = 2.7), meaning that ascending species were more light demanding. The two indicator values for temperature and light were significantly, but not highly, correlated among the set of studied species (Spearman rank correlation r = 0.33, n = 150, < 0.001; see also Appendix S4). There were no other significant differences in Landolt's indicator values.

Raunkiaer's life forms were available for all the 175 studied species (see details in Table 2). They were significantly not randomly distributed between ascending and descending species (Fisher's exact test of association: < 0.05). Relative to other categories, there were more ascending species belonging to deciduous phanerophytes (o.r. = 3.24), deciduous nanophanerophytes (o.r. = 2.35) and evergreen nanophanerophytes (o.r.  = 1.98), whereas there were relatively more descending species among evergreen phanerophytes (o.r. = 0.51), herbaceous chamaephytes (o.r. = 0.39), hemicryptophytes (o.r. = 0.59) and geophytes (o.r. = 0.35). Woody chamaephytes did not display any departure from the general trend. In terms of average elevation shifts, only geophytes showed a significant downward shift (−88.0 m) whereas deciduous phanerophytes (+70.4 m) and woody chamaephytes (+50.1 m) showed a significant upward shift. The upward response of nanophanerophytes was large (mean shift = +25.9 m, o.r. = 2.43), mainly due to the shift of deciduous nanophanerophytes (+27.2 m).

When comparing all woody species pooled together (phanerophytes, nanophanerophytes and woody chamaephytes) against all herbaceous species (herbaceous chamaephytes, hemicryptophytes and geophytes), the odds ratios were highly different from one (2.59 for woody species), indicating a strong tendency for woody species to shift upward and the reverse for herbaceous species (see Appendix S3). Woody species displayed an average altitudinal shift of +39.0 m, significantly different from zero.

Both low and tall woody species were significantly shifting upward. But only low trees had an odds ratio significantly different from one, i.e., a higher relative proportion of ascending species in comparison with all other herbaceous or tall tree species (Table 2).

The information about deciduousness was mostly available for woody species. Deciduous species were significantly more shifting upward (o.r. = 2.25) than evergreen species, with a mean shift of +49.3 m, significantly different from zero (n = 71). The difference became even more significant when considering phanerophytes only, with an odds ratio of 2.92 for deciduous species (51 ascending and 12 descending among deciduous vs 16 ascending and 11 descending among evergreens).

Seed masses were available in the Leda Traitbase for 107 out of the 175 studied species. Seed mass was significantly higher among ascending species (mean = 187.5 mg) than among those descending (mean = 179.8 mg) when considering all 107 species (t-test of difference in means of the logarithms, n = 107, < 0.05). However, it must be pointed out that a significant positive relationship exists between seed mass and woodiness (n = 107, < 0.001, r² = 0.15 in an analysis of variance of woodiness effect on seed mass logarithm; see also Appendix S4).

Considering the species according to their presence in stages of forest development, there were relatively more ascending species among species found in early stages (o.r. = 1.52) than among other species (o.r. = 0.66), and the former displayed a larger shift (mean = +22.2 m) than the latter (mean = +6.0 m). The same trend was observed for pioneer vs non-pioneer species. However, because this latter information was only available for woody species, both categories displayed a significant upward shift.

Species dispersed by ants were less ascending than those with other dispersal modes (o.r. = 0.46) and presented a negative average shift, although not significant. In contrast, species dispersed by birds were slightly more ascending than other species (o.r. = 1.36, not significant).

There were no marked differences between species restricted to mountain belts and ubiquitous species. However, there were less ascending species among exclusively mountain species (o.r. = 0.69) than among ubiquitous species (o.r. = 1.44).

Species shifts and successional dynamics

When considering closed forests only, the sampling area was restricted to 1 266 803 ha at the first inventory (11 745 plots) and 1 733 310 ha (12 623 plots) at the second one. Closed forests still showed a significant ageing trend at most of the lowest elevations classes (Fig. 4). This was also accompanied by a homogeneous increase of basal area throughout the elevation gradient (Fig. 4).

When fitting the logistic model to data from closed forests alone, 164 species (Table 1) showed a bell-shaped response to elevation (compared to 175 with all the data), among which 63 were descending and 101 ascending (ratio = 1.60, vs 1.82 in all forests). Although the number of species shifting upward was higher than that of descending species, the mean elevation shift was no longer significantly different from zero and even became negative (−3.0 m) (Table 2). Out of these 164 species, 62 species presented a shift statistically significantly different from zero, among which 23 were descending and 39 ascending (odds ratio of 1.70, mean of −8.4 m).

Ascending species still had a significantly higher mean Landolt's indicator value for light (mean L = 3.1) than descending species (mean L = 2.7, n = 141, < 0.01). The difference was no longer significant for temperature (= 0.09). A difference appeared for the moisture indicator value, with more hygrophilous species among descending species (mean indicator value for moisture F = 2.5) than ascending species (F = 2.2, n = 134, < 0.05).

The most prominent difference from previous results was that average shifts were lower, more often negative and no longer significantly different from zero for most of the plant trait categories (light-demanding species, ubiquitous species, phanerophytes, woody species, tall and low trees, deciduous species, species found in immature stages of forest dynamics, pioneer and non-pioneer species, species dispersed by wind), although the relative number of species shifting upward (o.r.) was still significant for some of those categories (Table 2). Evergreen phanerophytes and hemicryptophytes displayed a significant odds ratio for ascending vs descending behaviour in closed forests, whereas it was not significant in the entire sample, although in the same direction (relatively more descending species). Woody chamaephytes comprised many more ascending species (o.r. = 5.25), whereas they did not show any pattern in the entire sample. For the pioneering habit, the pattern was reversed in closed forests in comparison to that in all forests: pioneer species comprised relatively less ascending species. Species dispersed by gravity, which displayed no trend in the entire sample, were now composed of more ascending than descending species (o.r. = 4.93), and their mean shift was significantly positive (+37.0 m).


The observed upward shift of species (+12.6 m·decade−1) in the mountains of Southeast France, corroborated by the observation that more species were ascending than descending, confirms previous findings that forest vegetation is currently responding to some environmental changes in French mountains (Lenoir et al. 2008). But the amplitude of the observed shift decreased drastically when only considering closed forests. Moreover, contrary to previous observations that most shifting species were herbaceous, mountainous and fast-migrating, we found here that they were mainly woody species, low or tall, light-demanding with a pioneer habit.

Several explanations can be proposed for the altitudinal shift of plant species. First, it could be partly the effect of climate warming. The last episode of continuous and steep warming in the Southern Alps started at the turn of the 1970s, one decade before our first inventory, and developed at a rate of 0.5 °C·decade−1 until 2000 (Moisselin et al. 2002), by the end of the second inventory. This should have pushed species habitats into higher elevations.

Lag between species shifts and climate warming

The amplitude of the species shifts we calculated lagged largely behind the rate of climate warming observed in the same area. If vegetation had exactly tracked the 0.5 °C· decade−1 warming rate, the theoretical shift should have been +89 m·decade−1, because of an environmental lapse rate of 0.56 °C·100 m−1 in the studied area. We observed a shift of only +12.6 m·decade−1 in all forests, seven times less than expected in the equilibrium model. This lag is one of the highest observed among plants (Bertrand et al. 2011; Chen et al. 2011). In closed forests alone, the shift even decreased to an insignificant value of −2.1 m·decade−1. The lag in response to climate warming was observed in almost all previous studies of plant elevation shifts (Chen et al. 2011). This could be due to a combination of low dispersal capacity (Svenning & Skov 2004) and an ability to acclimate rather than suffer extinction or migration (Hirzel & Le Lay 2008). In the forest ecosystems we studied, nearly all species are perennials, a species trait that probably favours resistance and acclimatization to periods of extreme conditions. It might also be that the gradual canopy closure due to forest maturation counteracted climate warming effects on microclimate in the understorey.

Cause of the observed shifts

Most of our results point to an explanation other than climate warming. A strong relation appeared between the shifting of species and successional changes i.e., forest closure and ageing. First, plant species traits more often associated with the shifting behaviour were those related to forest successional dynamics: pioneer habit, potential presence in immature forest stages and potential height of the species. Second, the area proportion of open forests shrank between the two inventories at the lower elevations. This forest closure was accompanied by a parallel ageing of forest stands, more pronounced at low altitudes. Lastly, the global upward trend of species became insignificant when calculated solely for closed forests, in spite of a still very large sample of plots.

Therefore, the significant upward shift of species in the entire sample could rather be interpreted as a global change in forest development stage, varying with altitude. The loss of open forests at lower elevation led to a decrease in frequency of species from open forest habitats and an increase of species from closed forests. Thus, the apparent upslope movement was the result of changing dominance or frequency within communities at lower elevations, due to stand maturation, rather than an actual shift of species to higher elevations, as illustrated in Fig. 6. A similar role of changing dominance, but due to climate change, on the observed altitudinal shifts of species as been already observed by Kelly & Goulden (2008); see also Breshears et al. (2008).

Figure 6.

Illustration of the change in probability of presence depending on a forest closure mainly occurring at lower elevation. The example is given for a species found in immature stages. Retraction of the distribution range due to forest closure at the lower end of the species distribution leads to a shift in the calculated optimum.

Low trees and shrubs are characteristic of immature stages of forest succession in temperate and Mediterranean zones, giving way to tall trees as a forest ages. This explains the large odds ratios (apparent ascending habit) observed for these plant traits. The high cover and heavy shade of woody shrubs in open forests is responsible for a low herbaceous species richness (Shachak et al. 2008). Thus, forest dynamics at lower elevations could also explain the small odds ratio (apparent descending habit) of the herbaceous plant trait.

The closure of forests at lower elevations also explains both the apparent downward shift of geophytes, shade-tolerant and ant-dispersed species, these three plant traits being associated with late-successional stages (Hermy et al. 1999), and the apparent upward shift of light-demanding species, which are progressively eliminated from maturing forests due to habitat loss at lower elevations (Fig. 6). Because the Landolt's indicator values for temperature and light are correlated in our sample of species, even the upward shift of thermophilous species could be explained by the same phenomenon. Moreover, the difference in Landolt's indicator values for temperature between descending and ascending species became insignificant in closed forests. Crimmins et al. (2011) observed a downward shift of plant species optimum in Californian mountains but explained it as a regional decrease in climatic water deficit. However, this explanation cannot apply in our study region, where no trend was observed in the precipitation pattern (Moisselin et al. 2002).

Forest closure in Mediterranean mountains causes an attenuation of the Mediterranean character and a transition of mesomediterranean vegetation to supramediterranean and mountain vegetation (Simon & Tamru 1998). Woody species are more frequent in Mediterranean than temperate ecosystems. Quézel & Médail (2003) indicate a decrease in woody species richness from 15 species· 100 m−2 on average in mesomediterranean forests, corresponding to our lower elevation plots, to 10 species· 100 m−2 at higher elevations. Thus, such a loss of Mediterranean character in the lower elevation belts explains the large odds ratio (upward shift) for woody species. This apparent upslope movement of woody species also explains the heavier seed mass of ascending species, because these two plant traits are correlated. Abies alba, which was descending in our study, has been shown to colonize the understorey of Mediterranean mountain forests at lower elevation when they mature (Quézel & Médail 2003; Chauchard et al. 2007). Evergreens, broad-leaved or not, did not shift upward, contrary to what was expected in the case of climate warming impact (Berger et al. 2007; Walther et al. 2007).

Our interpretation of a major effect of forest successional changes at lower elevations was confirmed by another observation: species with an optimum above 1300 m a.s.l. did not display any significant directional trend (Appendix S3). Above 1800 m a.s.l., the tendency was even reversed, towards a slight downward shift. In addition, the discrepancy between the altitudinal homogeneity of the warming trend (Fig. 2) and the heterogeneous nature of the species upward shift we observed is again not in agreement with a uniform effect of climate change.

Forest closure, ageing and maturation

The closure and ageing of forests in the Mediterranean zone during the last decades is the result of several factors. It is the natural legacy of agricultural land abandonment earlier in the 20th century, which occurred first in mountain belts then, during more recent periods, in the Mediterranean lowlands. It is also the result of a continuous decrease in forest pasturing. Finally, it reflects a persistent imbalance between increasing forest growth and stable or decreasing biomass removals (Dupouey et al. 2009). This latter cause could also play a role in the dynamics of closed, mature forests (see also Walther & Grundmann 2001). In our study, closed and open forests were separated in the photo-interpretation. But photo-interpretation cannot discriminate between the more subtle successional stages linked with the slow wood accumulation in already closed forest stands. Thus, it should be possible that some of the residual changes that we observed between the two inventories only in closed forests, such as the apparent descent of geophytes, were due to faster rates of stand maturation at lower elevations in our study region.

Bias in resampling studies

Resampling studies, especially when they are based on historical and bibliographical data, are prone to bias due to lack of homogeneity in sampling schemes and field methods over time. Although the first inventory we used here was not initially planned for later resampling, the NFI protocols allowed a better control in space and time of data quality and homogeneity. The ecological and forest type stratification of the sample at each date ensured a good and constant representativeness of the entire area under study. Knowledge of the reliability of species identification permitted an efficient filtering of taxa before analysis. However, the increase in plot area prevented us from studying changes in species niche amplitude or absolute frequency of presence. We could only consider shifts in species optima, which are not dependent on census exhaustiveness and observed species frequency.


Tracking the impacts of climate change, we have found that forest belts are subject to the confounding influence of land-use changes, similar to what has been observed above the treeline. The main factor for the apparent upward shift of vegetation was the closure and maturation of forest stands, varying in intensity with elevation, masking the response to climate change. The influence of such changes in successional status of forest stands is difficult to detect in forest belts because stand closure, ageing and maturation in already established stands are more difficult processes to characterize than the woody encroachment of abandoned pastures above the treeline. Measuring basal area, characterizing openness of forest stands, and more importantly, counting the age needs specialized observers. Such data are not usually available in classical plant censuses. Our study points out the need for care in interpreting long-term forest vegetation changes when this information is lacking.

We conclude that land-cover and land-use changes belong to the few major processes, together with climate change, atmospheric pollution and CO2 fertilization that are able to reshape biodiversity patterns at regional scales within a short time span. Such a major impact of land-use change could be expected in the Mediterranean context we studied, where human pressure on wooded lands was stronger and persisted until more recent times than elsewhere in Europe. However, since large areas in Western Europe are still undergoing similar trends of forest recolonization and maturation at lower elevations, the interpretation of upward shifts of vegetation in mountains should be made with care.


We are grateful to Thomas Wohlgemuth who provided a numerical version of the Landolt's indicator values and Raunkiaer life forms, to Elisabeth Bienaimé and EvelyneGranier for their help in GIS mapping, and to Thierry Tatoni and two anonymous reviewers whose constructive comments helped to improve the manuscript. We thank Nabila Hamza who provided useful explanations of the NFI data. This work was financially supported in the frame of the French IFB-GICC program ‘Biodiversity and Global change’.