1Changes in the frequency and intensity of disturbances are expected to occur during the coming decades as a consequence of climatic changes. Mountain forests are sensitive to climate variability, disturbances and changes in human activities; this is particularly true for subalpine forests located close to the limits of tree-growth. Here we test the role of surface fires on the structure and the dynamics of a subalpine forest compared to a control stand not affected by fire events for at least two centuries.
2The fire events are deduced from fire scars, the age-structure from tree-ring counting, regeneration from sapling and seedling counts, necromass from the volume of woody debris, and the understorey structure from shrub and herb cover, height and biomass. Land-use history is assessed from livestock and human inventories.
3Four surface fires occurred during the last 200 years in the burned stand. Tree density is greater in the burned stand, and, even 50 years after the last fire, burning could still be having a positive effect on regeneration density. Land-use history, mainly grazing by cattle and sheep, explains the modern forest structure through its promotion of Larix decidua. The understorey composition is the same within the two stands, suggesting a rapid recovery of dwarf shrubs (Rhododendron ferrugineum and Vaccinium myrtillus) after the latest fire, which coincided with the final change in land-use. The exception is V. vitis-idaea cover, which is higher in the control stand. The removal of livestock in the 1940s indirectly triggered a decline in L. decidua regeneration, which was formerly promoted by grazing and trampling. Conversely, Pinus cembra seedlings have begun to establish during the last 70 years.
4Synthesis. Expansion of forests dominated by P. cembra is expected during the 21st century, unless current global warming leads to a significant and lethal increase in fire frequency, that is, a reduction of fire-return intervals, which would reduce the number of sexually mature trees. The re-introduction of grazing could maintain the mixed subalpine forests, whereas surface fires would alter the woody debris load, promoting greater regeneration of L. decidua and P. cembra.
In Europe, although the upper subalpine forests dominated by Pinus cembra L., Pinus mugo Turra or Larix decidua Mill. are classified as key habitats (European Commission DG Environment 2003), little is known about their functioning, in terms of structure and dynamics. However, since the middle of the 19th century, forest landscapes have been transformed as a result of changes in forestry practices, and the cessation of traditional agro-pastoral activities; this has included land abandonment (Jail 1969) and grazing cessation (Barbero et al. 1990; Didier 2001; Motta et al. 2006b). The abandoned subalpine pastures were rapidly colonized by L. decidua or P. mugo, followed by massive regeneration of P. cembra (e.g. Piussi 1994; Krüsi & Moser 2000; Didier 2001). Significant decreases in tree and sapling densities of shade-intolerant trees have been observed since 1950s (Risch et al. 2003). Subalpine forests experienced strong modifications although the kinetics of forest dynamics are slow because forests are slow-growing ecosystems. Although world-wide, fire is recognized as having a major effect on terrestrial ecosystems (Wright 1974; Bond et al. 2005), little is known about the consequences of fire in mountain ecosystems, because for centuries these forests were cut and fire was indirectly suppressed for agricultural purposes. However, several studies have shown that fires did occur during the last millennia in the European mountains (Carcaillet 1998; Tinner et al. 1999), and that their occurrence could increase during the 21st century as a result of global warming (Beniston 2000).
In the French Alps, during the Holocene, fires were asynchronous, with similar frequency patterns between 9000 and 4500 cal. years BP but not after this period (Carcaillet et al. 2009). Paleoecological studies have shown that wildfires occurred at moderate frequencies within the P. cembra-dominated forests during the early- and middle-Holocene (Genries et al. 2009b); when fires were more frequent, plant composition (Genries et al. 2009a) and soil quality (Mourier 2008) were affected for several centuries. Nevertheless, the impact of fires on forest structure and species composition of these systems still needs to be deciphered. It is especially crucial to understand how the stand structure and development of subalpine forests may change in response to environmental modifications, such as a disruptive fire regime (disturbance) or the cessation of grazing (stress). Two complementary approaches were taken.
First, we hypothesized that fires could affect the accumulation of woody debris (WD) or shrub biomass (and thus the fuel-load), altering the risk of new fires spreading and promoting tree regeneration. To test this hypothesis, we compared a burned stand and a neighbouring control. We quantified the ecological consequences of fires on forest structure in terms of regeneration, living biomass, WD and thickness of the litter and humus layers. Fire scars on P. cembra trunks were used as fire proxies (Fig. 1).
Second, we differentiated between the impact of fires and land-use change on the establishment and the age-structure of tree populations, based on tree-ring data and inventories of the human population and livestock.
The ultimate aim of our study was to identify possible approaches to manage the mountain forests, taking into account fire as a process resulting from both land-use and climatic changes.
The forest is located at 45°17′20″N, 6°57′07″E, approximately 2150 m a.s.l. within the district of Lanslevillard in the upper Maurienne Valley (Savoy, central French Alps; Fig. 2). This valley is located along the French–Italian border, at the northern limit of any Mediterranean climatic influence and is one of the driest areas of the Alps. The mean annual precipitation and air temperature are 884 ± 167 mm and 3.5 ± 0.6 °C, respectively (Bessans meteorological station, 1715 m a.s.l.). The mean temperatures for the coldest (January) and the warmest (July) months are −6.5 ± 1.8 and 13.7 ± 1.3 °C, respectively.
At the turn of the 18th century, the upper Maurienne Valley was entirely used for agriculture (Delcros 1994). Former land uses include cattle, mule and sheep husbandry and cultivation for haymaking; large areas were therefore meadows. Land abandonment began at the end of the 19th century (Jail 1969) and resulted in significant vegetation change leading to forest establishment and subsequently to the introduction of a fire-suppression policy. Today, the lower subalpine belt (c. 1700–1900 m a.s.l.) is covered by open coniferous woodlands dominated by Pinus sylvestris L., L. decidua and Picea abies (L.) H. Karst and an understorey characterized by Vaccinium spp., Arctostaphylos uva-ursi (L.) Sprengel, Juniperus communis L. and Ononis rotundifolia L., or by shrublands composed of tall and dwarf shrubs (Betulaceae, Ericaceae, Salicaceae). Above 1900 m a.s.l., the upper subalpine forests are dominated by P. cembra, P. mugo, L. decidua, and P. abies, with understoreys of Juniperus sibirica Loddiges ex Burgsdorff, Vaccinium spp. and Rhododendron ferrugineum. The upper tree line reaches 2350–2400 m a.s.l. (Bartoli 1966). The alpine belt (> 2300 m a.s.l.) is covered with boulders and species-rich short-grass meadows characterized by Carex curvula All. and Nardus stricta L.
sampling design for floristic and biomass inventories
In June and July 2006, we sampled two contiguous stands of 8000 m2 each (80 × 100 m), one burned and one control. The stands are separated by a trekking path and the burned stand is located above the control one. They have the same soil type (podzolic), aspect (N–NW) and slope (23°). Scattered trees with fire scars allowed us to delimit the burned stand.
Each stand was analysed at the stand and plot scales. Each stand contained three transects of 25 m and 10 plots measuring 100 m2 each, evenly distributed across the area. Each plot contained four 3-m-radius circles, five quadrats measuring 2 × 5 m and one quadrat of 1 m2 (Fig. 3).
1.a. At the stand scale (8000-m2), for trees with ø ≥ 5 cm at a height of 30 cm, within both stands, we described canopy characteristics at the 8000-m2 stand scale, that is, number of trees, tree species and diameter at 30 cm, in order to calculate density and basal area per stand and per species (McRae et al. 1979). These trees were cored with an increment borer at around 30 cm above the ground to determine their age by counting the annual growth rings; this provided us with data on the age structure of the populations, which mirrors forest dynamics.
1.b. The necromass volume was estimated by tallying the WD (method adapted from Motta et al. 2006a). WD was classified into small (diameter (ø) < 7 cm) and large (ø > 7 cm) logs (fallen stems or branches), snags (dead standing trees, ø > 7 cm and height ≥ 1 m) and stumps (short, vertical trunks created by cutting or wind throw, top-ø > 7 cm and height ≤ 1 m). Snags were distinguished from logs by their angle: where the angle with the ground was more than 45° the object was classified as a snag. The volume of snags and large logs was estimated as a frustum of a double cone from the radius at ground level (avoiding roots or irregular shapes) and the length:
where VSLL is the volume of the snags and large logs (m3), R is the radius of the larger end (m) and L is the length of the snag or large log (m). The volume of stumps was estimated as a frustum of a cone from the diameter at the top, the diameter at ground level (avoiding roots or irregular shapes) and height, estimating a plane cutting off the apex, parallel to the base (Motta et al. 2006a):
where Vs is the volume of the stump (m3), St is the area of the smaller end (m2), S0 is the area of the larger end (m2) and L the length (m).
1.c. Fine WD was quantified by the line-intersect method (adapted from Van Wagner 1982), applied to three series of transects. Woody fragments were classified into the five diameter classes recommended by McRae et al. (1979); the original classification was developed for boreal forest, a close analogue of subalpine forest. Different intersect-line lengths were used for each diameter class: class I, [0–0.5] cm along a 5-m transect; class II, [0.5–1] cm along a 10-m transect; class III, [1–3] cm along a 15-m transect; class IV, [3–5] cm along a 20-m transect; class V, [5–7] cm along a 25-m transect (Fig. 3; Hély et al. 2000a,b). The equation to calculate the volume is adapted from Brown (1974) modified by Van Wagner (1982), as follows:
where VSWD is the volume of fine WD (m3), k is a constant with a value of 1.234 (Van Wagner 1982), L is the length of the transect (m), d is the fragment diameter at the intersection with the transect line (cm), a is a correction factor for non-horizontal debris, that is a value of 1.13 (Brown 1974), and c is a slope correction factor (Brown 1974) computed from the slope angles of each stand.
2.a. At the plot scale (100-m2, Fig. 3), within the four 3-m-radius circles, all tree species seedlings (height < 0.3 m) and saplings (height ≥ 0.3 m, ø < 5 cm at 30 cm high) were counted to quantify recent regeneration.
2.b. Within each plot, five 10 m2-quadrats (2 × 5 m) were used to quantify the height, the species-richness and the cover of shrubs. The cover was estimated on the basis of the six categories of Van der Maarel's cover-index (Van der Maarel 1979).
2.c. At the centre of the plot, one quadrat of 1 m2 (1 × 1 m) was delineated (Fig. 3) in order to estimate the cover of herbs. Furthermore, exhaustive sampling of shrub and herb species was conducted to characterize the understorey biomass. Finally, the thickness of the litter and humus layers was measured in a 0.4 × 0.4 m quadrat included in the previous one, and samples of both layers were collected for dry mass measurements.
To predict shrub biomass using a non-destructive method, allometric equations were derived from linear and power models, taking into account plant cover and above-ground height, using the data from the 1 m2 quadrats. Parameter estimates were derived by simulated annealing using the Metropolis algorithm (Metropolis et al. 1953). This algorithm allows exploration of the parameter space to ensure convergence towards the absolute least-square minimum (details in Chuine et al. 1998). These models were constructed using Delphi 6.0 software (BorlandTM). Different models were tested, including linear and power relationships, and a power model was finally chosen for each species. Finally, the allometric equations were applied to all measurements of height and cover over the larger area of 500 m2 for each stand, allowing estimation and comparison of the shrub biomass in all quadrats.
tree age determination
Cores were mounted, air-dried and smoothed using progressively finer sand paper in order to make the annual growth rings readily visible. To determine tree age, tree rings were counted and cross-dated by visual observation under a dissecting microscope (40× or 60×). For incomplete cores where the pith was not sampled, we estimated the length of the missing radius by matching the curvature of the innermost rings to concentric circles drawn on paper (Applequist 1958; Arno & Sneck 1977). The number of years to the pith was then estimated by calculating the mean number of tree-rings on radii of corresponding lengths from 739, 239 and 14 complete cores of P. cembra, L. decidua and P. abies, respectively; these data were taken from Beilhe et al. (2009), who collected their data in the same area. To estimate total age, the number of years taken to reach coring height has to be added to the age previously obtained. This number of years was derived from Beilhe et al. (2009), and was assessed by determining the height–age relationship of seedlings and saplings (Mast et al. 1998; Kajimoto et al. 2002; Chauchard et al. 2007). Heights (ln-transformed) were plotted against ages and, a linear regression model was used for modelling the height–age relationship. Total age was obtained by adding our estimates of age at coring height to the estimated number of years required to reach 30 cm, as determined using the equations in Table 1.
Table 1. Allometric equations derived from Beilhe et al. (2009) to calibrate the missing years when estimating tree-ages from a tree-ring core at 30 cm height, where x is the tree-age and y is the ln-transformed height value. n is the sapling number used to produce the model
Sampling numbers (n)
Age at 30 cm (± SD)
y = 0.19x + 4.7
10 ± 5
y = 0.15x + 8.22
13 ± 9
y = 0.15x + 12.34
17 ± 7
The method provides estimated tree ages closer to real ages and assumes that the growing conditions of extant seedlings and samplings are similar to those experienced by canopy trees. Whatever the accuracy of this assumption, the age estimates obtained by this method are more appropriate than simply using age at coring height to compare the tree age-structure with the human or livestock inventories.
statistics and modelling
To reduce errors introduced by the procedure for estimating tree age, age-structure was plotted using 10-year classes. Because our data did not exhibit a normal distribution (χ2 test, P-value < 0.01), we used the nonparametric Kolmogorov–Smirnov test (KS-test) to examine differences in age distributions between burned and control stands. This test was first applied to raw data and then to data standardized as follows (zsi):
where xsi is the number of individuals of species s in the ith age-class, and x̄s and σs are the mean number and the SD of s over the whole sequence, respectively.
The nonparametric Mann–Whitney test allows data series to be compared when they do not adhere to the criteria for normality; we used it to test differences between the burned and control stands with respect to: i) the number of P. cembra seedlings (h < 10 cm); ii) the number of P. cembra saplings (10 cm < h < 30 cm); iii) the number of L. decidua saplings; iv) the percentage cover and the height of R. ferrugineum, V. myrtillus and V. vitis-idaea L.; v) the number of shrub species; vi) the number of herb species; vii) the mass of herbs; and viii) the mass and the depth of litter and humus layers.
data inventories of human populations and livestock
As a first approximation, we assumed that human populations and livestock numbers can be considered as proxies for land-use pressures on vegetation (Chauchard et al. 2007). The cumulated number of permanent inhabitants and transient shepherds in the three villages closest to the study site (Lanslebourg, Lanslevillard, Bessans) were obtained from http://cassini.ehess.fr and from the Archives Départementales de Savoie (archive number C580–C606–608, Département de Modane) combined with data from Jail (1969). The number of cattle and sheep in the upper Maurienne Valley was obtained from Jail (1969). The trajectories for the cumulated number of trees on the two stands were compared with the number of inhabitants and transient shepherds in the three villages mentioned above and with the number of sheep and cattle in the upper Maurienne Valley.
fire dates, tree-stand characteristics and age structures
Within the burned stand, six P. cembra have fire-scars (Fig. 1). Four fire events were identified and dated by tree-ring counting (Fig. 4). The fires occurred in 1833, 1887, 1894 and 1954 (Fig. 4). Of the six trees, two have scars corresponding to two fire events. No scars were visible on old-growth L. decidua or P. abies.
In total, 534 trees were cored, majority were in the burned stand, that is 67% of the total (Table 2). Of the 356 trees cored from the burned stand and the 178 from the control stand, 28 and 17, respectively, were not dated because they had rotten centres. In both burned and control stands, the majority of trees were P. cembra or L. decidua; the most efficient tree recruitment occurred between 1920 and 1940 (Fig. 4). The only other tree species present was P. abies (Table 2), for which recruitment never exceeded three trees per 10-year class (Fig. 4). The oldest dated tree was a 419-year-old L. decidua tree of and the youngest trees, that is, < 60 years old, were mostly P. cembra (Fig. 4). Very few L. decidua were < 60 years old.
Table 2. Stand characteristics. Only cored individuals are reported (ø > 5 cm at 30 cm high)
Number of individuals
Density of individuals
The age structures based on total tree numbers indicate a significant difference between the burned and control stand (KS-test, P-value < 0.01), but there is no difference when the tree numbers are standardized (KS-test, P-value > 0.10). The standardized age structures show the trends in tree population dynamics irrespective of the unit of tree abundance and without effect on the variation profile. In both the burned and control stand, similar phases of tree recruitment were found (Fig. 4). Between 1660 and 1730, there was a slight increase in tree number, from 4 to 18 individuals in total. This was followed by a decrease over the subsequent 50 years, with only a total of three trees dated from 1775. Recruitment then increased again until 1835, before the number of trees in both stands stabilized. The increase in numbers from 1890 to 1940 was large, with a maximum of 52 trees per 10 years for the two stands combined; this was recorded for the period 1920–1930. Since 1940, the recruitment in both stands has dropped continuously, with only one tree recorded during the last decade.
seedling and sapling regeneration patterns
In both stands, the only seedlings recorded were P. cembra: 840 seedlings ha−1 in the burned stand and 150 seedlings ha−1 in the control stand (Table 3). Most of the saplings were P. cembra in both the burned (292 saplings ha−1) and control (80 saplings ha−1) stand, but there were a few L. decidua and Abies alba L. saplings in the burned stand (35 and 9 saplings ha−1, respectively) and L. decidua and P. abies saplings in the control stand (27 and 9 saplings ha−1, respectively) (Table 3). The mean numbers of P. cembra seedlings and saplings and L. decidua saplings were not statistically different in the two stands because of large variances (Fig. 5a,c,e). However, there were significantly more P. cembra seedlings and saplings per hectare in the burned stand than the control (Fig. 5b,d). These data indicate that P. cembra regeneration is more spatially heterogeneous in the burned stand.
Table 3. Mann-Whitney tests on the differences between the burned stand and the control stand. The bold P-values indicate significant differences (P < 0.05); asterisks indicate the two parameters that were not tested because there were too few seedlings or saplings (only one spruce and one fir sapling recorded)
Seedlings (number ha−1)
840 ± 1855
150 ± 348
Saplings (number ha−1)
292 ± 431
80 ± 233
35 ± 108
27 ± 94
9 ± 56
9 ± 56
Richness (number of species)
3.53 ± 0.91
3.41 ± 0.73
5.40 ± 2.59
5.10 ± 1.97
Cover (Van der Maarel index)
2.51 ± 1.62
2.59 ± 1.40
2.88 ± 1.61
3.18 ± 0.95
0.76 ± 1.13
1.47 ± 1.04
46.07 ± 12.95
43.14 ± 9.88
29.77 ± 7.31
31.94 ± 9.51
10.89 ± 2.45
10.80 ± 6.88
2.20 ± 0.95
1.72 ± 0.91
3.50 ± 1.37
3.50 ± 1.32
Mass (g m−2)
294 ± 385
220 ± 139
120 ± 72
143 ± 112
51 ± 28
51 ± 36
271 ± 254
159 ± 59
861 ± 767
1353 ± 790
194 ± 124
149 ± 77
understorey structure and composition
The mean cover of R. ferrugineum and V. myrtillus was similar in the two stands, but there were significant differences in the cover of V. vitis-idaea between the two (Table 3; Fig. 6a,c,e). In fact, approximately 30% of the 10-m2 quadrats did not contain V. vitis-idaea in the burned stand, and approximately 30% of plots had < 5% cover in the control (Table 3; Fig. 6f). Rhododendron ferrugineum and V. myrtillus were the only species with cover values exceeding 50% of the sampled areas (Table 3; Fig. 6b,d). The above-ground biomass and height of the three main dwarf shrub species did not differ between the stands (Table 3; Fig. 7a–c). Similarly, the mean number of shrub and herb species and the mean mass of herbs were not significantly different between the stands (Table 3; Fig. 7d). Finally, both litter and humus layers had similar mean thicknesses and masses in the two stands (Table 3; Fig. 7e,f).
For the two main shrubs, the allometric equations to estimate plant biomass from plant cover and height are as follows; R. ferrugineum:
M = (0.1034 + 3.0995 × (C × H)1.2419) × 20(eqn 5)
and V. myrtillus:
M = (0.0237 + 0.5508 × (C × H)0.5545) × 20(eqn 6)
where M is the above-ground dry mass (kg ha−1), C is the cover (m2) and H is the height (m2). The relationships between the observed and the modelled dry masses of R. ferrugineum and V. myrtillus are significant with r2 values of 0.94 and 0.57, respectively (Fig. 8). These results demonstrate that accurate estimations of above-ground shrub biomass are possible on the basis of just plant cover and height. The allometric eqns 5 and 6, applied to all measurements of height and cover, thus allow us to estimate shrub biomass over the larger 500 m2 area for each stand. The estimated dry biomass of R. ferrugineum was greater on the burned stand (3614 kg ha−1) than the control stand (2783 kg ha−1); there was little difference between the stands for V. myrtillus (389 and 447 kg ha−1, respectively).
necromass and woody debris
Figure 9 shows the volumes of WD in the two stands. Although these results cannot be tested statistically, they indicate that the volume of large downed logs, stumps and small downed logs is greater in the control stand than in the burned stand, whereas the volume of snags is almost the same in the two stands. The greatest difference appears to be associated with the large-downed logs, with 6.3 m3 ha−1 in the burned stand and 87.7 m3 ha−1 in the control stand. Table 4 shows that about two-thirds of all the stumps in the control stand and about a quarter in the burned stand were created by human activities.
Table 4. The proportion of cultural (sawn) and natural (wind throw) stumps in the burned and control stands
data inventories: human population and livestock
Between 1790 and 1850, the numbers of sheep remained constantly high, and the numbers of human inhabitants and cattle increased, as did the cumulated number of trees on the burned and control stands (Fig. 10). Between 1850 and 1940, the number of livestock dropped (from 7700 to 2200 sheep and from 3800 to 2200 cattle), as did the numbers of permanent inhabitants and transient shepherds (from 3300 to 2000 people). Consistently, over the same period (1850–1940), the cumulated tree number increased. Since 1940, the recruitment of trees (i.e. individuals with ø ≥ 5 cm at 30 cm high) and the numbers of inhabitants and cattle have decreased; in contrast, the number of sheep increased as a result of new agricultural practices during the 1990s, supported by French and EU policies aimed at sustaining mountain agriculture, especially sheep husbandry.
the role of fire in forest dynamics and composition
Judging from the fire scars found on trees in a formerly grazed mixed subalpine forest, fires were probably surface fires and probably enhanced tree regeneration in the burned stand. Fire scars (Fig. 1) on living P. cembra trees indicate that this species can survive several surface fires. Although we have no evidence relating to the origin of these fires, they were probably anthropogenic: stubble-burning has long been used by farmers to clear pastures.
Although individuals of L. decidua and P. cembra were found to be of the same age, no scars were found on L. decidua. This finding suggests that adult individuals of this species are less sensitive to fire, probably because their thick (> 5 cm), non-flammable bark composed of parenchyma rich in water (Hare 1965; Hood et al. 2007) provides efficient fire protection. There are differences in the age structure of the burned and control stands (Fig. 4), with more trees in the burned stand, although the total basal area per stand is similar (Table 2). Table 4 highlights that markedly more trees were cut by humans in the control stand. This difference between the numbers of anthropogenic stumps cannot be explained simply by the greater number of trees in the burned stand. The small number of sawn stumps (54) in both stands compared to the cumulated number of logs (253) and snags (17) also suggests that if cutting activity influenced age structure, it was clearly not the main factor.
The observed tree densities indicate that tree recruitment was more intense in the burned stand. This difference in regeneration could be a result of chemical properties of the soil, which change immediately after burning due to the charcoal load (Nilsson & Wardle 2005). Indeed, charcoal produced by fires alters the soil nitrogen cycle (Berglund et al. 2004; DeLuca & Sala 2006) and absorbs phenolic compounds produced by Ericaceous species (Wardle et al. 1998; MacKenzie & DeLuca 2006). Phenols have numerous functions in soil processes, including limiting nitrification (Hättenschwiler & Vitousek 2000). By adsorbing phenols, charcoal could have a positive effect on nitrogen availability to young plants, resulting in more regeneration. This ecosystem determinant function of charcoal with respect to the regeneration of L. decidua and P. cembra needs to be tested in light of the specific phenols produced by subalpine Ericaceae, i.e. R. ferrugineum and V. myrtillus.
Larix decidua and P. cembra each constituted half of the tree populations, thus both have recruited more or less equally since the 17th century, and their recruitment was not affected by fires. Furthermore, based on the tree age structure, the comparison of the plant biomass (shrubs, herbs) and the necromass (WD), the four fires did not significantly alter the forest community except for WD, which is more abundant in the control stand (Fig. 9).
Surface fires had no effect on P. cembra and L. decidua population dynamics over the two centuries. However, a positive effect of fire on tree density is indicated by studies in boreal cold conifer forests, where fire represents the main disturbance (e.g. Johnson 1992). It has been shown that with increasing severity of a fire (severe fires involve organic soil combustion), recruitment of trees becomes more important (Johnstone & Chapin 2006). In our study, this efficient post-fire recruitment is obvious: the density of P. cembra seedlings and saplings, accounting for more than 82% of the total young plants, is higher in the burned than in the control stand (Table 3), although the mean densities of seedlings and saplings are not statistically different between the stands because of high variances (Fig. 5a,b). Pinus cembra seed dispersal is related to the hoarding activity of the European nutcracker (Nucifraga caryocatactes), a bird which caches seeds in open areas (Crocq 1990). The high levels of variability in regeneration may thus illustrate the patchy vegetation cover within the stand after fires, with well-burned patches of a few tens or hundreds of square meters being more attractive to nutcrackers. Moreover, except for the cover values for V. vitis-idaea, the cover of the other Ericaceae, the masses of shrubs, herbs, litter layer and humus were not significantly different between the two stands (Table 3; Figs 6 and 7). Finally, the modelled biomass values of R. ferrugineum indicate that this species would be more important in the burned stand. This suggests that the poorer recent tree regeneration in the control stand is not the result of competition with shrubs or herbs, neither is it related to the litter and humus depth, but pleads for an effect of repetitive surfaces fires, possibly as a result of their effect on soil chemistry. Furthermore, we did not find evidence of any effect of fire on the species richness of shrubs or herbs (Table 3), although, frequently, disturbances are crucial driving forces for ecosystem processes and biodiversity (White 1979; Frelich 2002) and they have often been found to increase diversity (e.g. Hiura 1995; Glitzenstein et al. 2003). The last fire occurred in 1954, that is, 52 years before our sampling. The effects of fire on understorey species richness are probably not visible after 50 years, which could explain why we were unable to find any difference in species richness between the stands. However, richness and diversity are not necessarily related to the length of time after the fire (Capitanio & Carcaillet 2008).
Conversely, the impact of the fires on the WD is still visible. The volume of stumps, and large and small downed logs is smaller on the burned stand (Fig. 9), and the proportion of stumps resulting from cutting is insufficient to explain this difference (Table 4). This is probably due to the fact that fires burned the logs and stumps in the forest.
potential effects of climate variability and land use on the tree-community
The two stands have followed the same dynamics, as indicated by standardized age-structure analysis (KS-test P > 0.05; Fig. 4d), although their total tree abundances differ. The similar trends in the population dynamics of L. decidua and P. cembra in the two stands during the 18th and 19th centuries indicate that fire was not the main factor controlling the forest structure and composition; this underlines the role of more regional influences such as land-use or climate variability (Fig. 4).
Climate data for our study area are very limited, covering the last 50 years at best. Thus, any climatic analysis of tree regeneration is impossible. However, climate reconstructions covering the last 130 years have been undertaken using tree-rings (Rolland et al. 1998); our age structure data, however, covers more than 300 years (Fig. 4). Although we cannot rule out an effect of climate on tree recruitment, the absence of fine temporal climatic records or reconstructions close to the study site prevents us from analysing the potential effect of climate on recruitment. Climate, however, is a key driver of tree recruitment, even when there are frequent fires (Brown & Wu 2005; Brown 2006).
Mountain forests have long been exploited by human population, notably through forestry (direct impact) and livestock grazing (indirect impact). In our study area, 60% of the tree stumps were created by cutting (Table 4), revealing that timber harvesting occurred recently. In addition, ungulates greatly affect the survival rates of seedlings and saplings in mountain forests through browsing, grazing and trampling (e.g. Ammer 1996; Motta 1996). The Maurienne valley is known to have had an economy based on livestock-breeding activities for centuries (Jail 1969). Livestock could thus have been one factor influencing the vegetation. During the period between 1850 and 1940, tree numbers increased while the number of cattle and sheep decreased, along with numbers of permanent inhabitants and transient shepherds. This could illustrate the severe impact of grazing on tree seedling and sapling survival. Conversely, between 1790 and 1850, and since 1940, all proxies roughly follow the same trends. The decrease in tree abundance since 1950 (Fig. 10a) could be an artefact resulting from the sampling strategy of dating trees with an increment corer: only individuals with a diameter > 5 cm at 30 cm-high were cored. The mean age of the smallest cored trees indicates that they reach a diameter of 5 cm at 30 cm-high when they are almost 55 years old. Consequently, this end of the chronology cannot be discussed in the light of land-use changes.
effect of pasture abandonment on regeneration
The analysis of the age-class distribution of trees (Fig. 4) shows that, until the 1960s, P. cembra and L. decidua each represented almost half of the recruited trees, with very few P. abies. Since then, with the exception of one P. abies, only P. cembra is recorded and it continues to regenerate well-enough to be the sole seedling species (Fig. 5).
The reason for this current change in forest composition could be the cessation of grazing almost 50 years ago in the Maurienne Valley, farming having begun to decline as early as 1850 (Fig. 10b,c). Indeed, L. decidua recruitment is affected by thick litter layers, and this species achieves better seedling establishment than P. cembra on the exposed mineral soil created by severe trampling in overgrazed pastures (Risch et al. 2003; Motta et al. 2006b). Although both limited by grazing and forest suppressing, L. decidua establishment would thus have been possible during periods of grazing because P. cembra establishment would have been more limited (Motta et al. 2006b; present study). However, since the 1960s, massive P. cembra regeneration has occurred, with severely restricted L. decidua recruitment (Figs 4 and 5). The great reduction in grazing pressure after the World War II has resulted in WD accumulation, favouring P. cembra; this has occurred in both the French (Eynard-Machet 1994; Wlérick 1994) and the Italian Alps (Motta & Dotta 1994; Lingua et al. 2008). Decreased livestock grazing in our study area over the last 50 years probably caused the decline in L. decidua establishment on abandoned subalpine pastures because of its low competitive ability.
It has been suggested in many recent studies that pure P. cembra communities will expand across formerly pure L. decidua or mixed L. decidua × P. cembra forests (Motta & Lingua 2005; Schulze et al. 2007; Lingua et al. 2008). These studies suggest that intensive land use for grazing was crucial for L. decidua regeneration, and that pure stands of L. decidua will disappear in the near future as a result of large-scale land-use changes in the European mountains unless management is once more intensified to create mineral soil suitable for L. decidua regeneration.
what future landscapes?
The abundant recent regeneration of P. cembra, together with the lack of L. decidua recruitment, could lead to a secondary successional pathway characterized by the progressive replacement of a mixed L. decidua × P. cembra community in favour of a pure P. cembra community (Motta & Lingua 2005). Furthermore, one paleoecological study shows that moderate fire regimes did not endanger subalpine ecosystems dominated by P. cembra over millennia (Genries et al. 2009b), thus corresponding to our results covering two centuries. Nevertheless, ongoing climatic changes may lead to an increase in disturbance frequencies (Schär et al. 2004; Schröter et al. 2005; Schumacher & Bugmann 2006); it has also been shown that high-frequency fires, that is, repetitive fire intervals < 80 years, could disturb P. cembra populations without favouring L. decidua populations (Genries et al. 2009a). Finally, L. decidua seedlings need larger patches caused by severe disturbances, such as clear-cuts, avalanches or flooding, to establish successfully. Studies have also shown that an increase in summer temperature could give L. decidua a competitive advantage over P. cembra in terms of radial growth if everything else is equal (Anfodillo et al. 1998; Carrer et al. 1999), although all species exhibit increased growth under warmer conditions (Rolland et al. 1998; Carrer & Urbinati 2004; Carrer et al. 2007). The predicted decrease in precipitation in the Alps (Sheffield & Wood 2008) could also alter the current trend. It is questionable, therefore, whether the cessation of grazing could confound the eventual effects of climate change.
The study of a formerly grazed and partially burned forest, affected by surface fires, has provided novel information concerning the role of fire in the dynamics and structure of subalpine forests. The four surface fires in the study area resulted in increased tree density and regeneration. This positive effect could be due to the impact of charcoal on the soil chemistry, a factor that should be examined in the future. Nevertheless, our interpretation argues in favour of multiple factors affecting forest structure, including climate variability and, more probably, human land use. Fire probably plays a complementary role in the dynamics of mixed subalpine forests, without altering the tree composition. Conversely, pasture abandonment, which intensified during the 20th century, has a wider impact. Indeed, a decline in L. decidua regeneration is obvious, and pure P. cembra stands are expected to be common in the near future.
Nevertheless, pasture abandonment and the lack of current fires favour deadwood accumulation, which could lead to more severe fires in the future. Furthermore, current climatic change could modify this trend by increasing disturbance frequency, with a threshold of fire-intervals that could be lethal for P. cembra communities (Genries et al. 2009a). Here, we could not find evidence of any negative effect of fire on the P. cembra population in the studied forest, but the four surface fires occurred during a period of only 120 years and mature P. cembra trees probably produce cones over longer periods. We therefore question the ability of such populations to maintain themselves if there were more severe or frequent fires over periods longer than 200 years. The transformation of mixed to pure P. cembra forests could be prevented if grazing is restored. This fact highlights the difficulties in assessing the sustainable development of mountain forests and shows that studies are needed to elucidate the relative impacts of climate, disturbances and human practices on long-term vegetation dynamics.
Financial support (to C.C.) was provided by the Institut National des Sciences de l’Univers (INSU-CNRS, France), through the national program ECCO, and by a grant (to A.G.) from the Ministère de l’Enseignement Supérieur et de la Recherche (France). We thank François Bailly, Loïc Bircker, Fanny Combet and Nicole Denis for field or laboratory assistance.