Interactions among tree-line conifers: differential effects of pine on spruce and larch



    Corresponding author
    1. VINCA – Vienna Institute for Nature Conservation and Analyses, Gießergasse 6/7, A-1090 Vienna, Austria,
    2. Department of Conservation Biology, Vegetation Ecology and Landscape Ecology, University of Vienna, Althanstraße 14, A-1090 Vienna, Austria
    Search for more papers by this author

    1. Federal Environment Agency, Spittelauer Lände 5, A-1090 Vienna, Austria,
    Search for more papers by this author
  • R. KÖCK,

    1. Department of Forest and Soil Sciences, Institute of Silviculture, University of Natural Resources and Applied Life Sciences, Gregor Mendel Straße 33, A-1180 Vienna, Austria,
    Search for more papers by this author

    1. Department of Forest and Soil Sciences, Institute of Silviculture, University of Natural Resources and Applied Life Sciences, Gregor Mendel Straße 33, A-1180 Vienna, Austria,
    Search for more papers by this author

    1. Department of Biogeography, University of Vienna, Rennweg 14, A-1030 Vienna, Austria, and
    Search for more papers by this author

    1. VINCA – Vienna Institute for Nature Conservation and Analyses, Gießergasse 6/7, A-1090 Vienna, Austria,
    Search for more papers by this author

    1. Department of Conservation Biology, Vegetation Ecology and Landscape Ecology, University of Vienna, Althanstraße 14, A-1090 Vienna, Austria
    Search for more papers by this author

Stefan Dullinger (tel. +43 1 4029675; fax +43 1 402967510; e-mail


  • 1Plant–plant interactions are increasingly considered as complex phenomena involving both negative and positive components. Within a community, the relative importance of these components is probably species-specific and may also vary among life-history stages and along environmental gradients.
  • 2We used the tree line of the north-eastern Calcareous Alps of Austria, composed of shrubby Pinus mugo and upright Picea abies and Larix decidua, as a simple system in which to investigate these interactions. We focused on the largely unknown effects of pines on spruce and larch, rather than on P. mugo, which is known to be competitively displaced by the two tree species.
  • 3We used regression models on observational data to analyse the responses of the trees to a gradient of pine cover in terms of recruitment, growth, fecundity and browsing damage, and to determine whether effects involved both competitive and facilitative components, if they depended on the life-history stage of the trees and if they were species-specific.
  • 4We detected a pronounced negative effect of pine cover on recruitment and growth of both spruce and larch, whereas seed production was unaffected. Larch turned out to be more sensitive to pine competition: its recruitment and growth are superior to that of spruce in open habitats but this advantage vanishes in dense pine thickets.
  • 5Contrary to expectations, the effects of pine cover on growth rates of spruce and larch did not depend on the life-history stage of the trees.
  • 6Pine cover is a major determinant of browsing damage for both spruce and larch, indicating that it does have a positive effect in providing shelter against herbivores.
  • 7The differential effects of pines on spruce and larch are likely to favour spruce at the expense of larch in realizing potential habitat expansion as a result of climate change. Disregarding the complex details of plant–plant interactions may thus result in unrealistic predictions of species responses to environmental changes.


Research on interactions among plants has long been biased towards competition (Keddy 1989; Goldberg & Barton 1992; Bruno et al. 2003). However, the number of studies dealing with the positive effects that coexisting species may have on each other has considerably increased in recent years (e.g. Callaway 1998; Choler et al. 2001; Callaway et al. 2002). A general conclusion that has been drawn from this more balanced research is that most interspecific interactions among plants are in fact complex phenomena that involve both negative and positive components (Holmgren et al. 1997; Callaway & Walker 1997; Holzapfel & Mahall 1999; Choler et al. 2001; Stachowicz 2001). Within a community, the relative importance of these components is probably specific to individual species combinations (Callaway 1992; Ryser 1993; Callaway et al. 1996; Takahashi 1997) and, within a specific combination of species, may also differ among life-history stages of the species involved (Aguiar & Sala 1994; Rousset & Lepart 2000) and along environmental gradients (Callaway et al. 2002; Hastwell & Facelli 2003; Pennings et al. 2003). Accordingly, environmental changes may have differential and complex feedbacks on community structure and dynamics. Acquiring more information on species interactions is thus not only of theoretical interest but is also increasingly called for in studies dealing with species or community responses to global changes (Brown et al. 1997; Davis et al. 1998; Jiang & Kulczycki 2004; Klanderud 2005).

Alpine tree-line forests are particularly suitable to study such complex interaction patterns as they are usually simple systems composed of only a few, ecologically similar and often taxonomically related species. Moreover, these forests are considered especially sensitive to climate change (e.g. Theurillat & Guisan 2001; Kullman 2002; Dullinger et al. 2004). Taking the north-eastern Calcareous Alps of Austria as an example, the current tree line is dominated by extensive mono-dominant stands of the obligatory shrubby pine species Pinus mugo Turra. Below this pine–krummholz belt, which covers about 100–200 altitudinal metres, subalpine forests of Norway spruce (Picea abies (L.) Karsten) and European larch (Larix decidua Mill.) prevail. Predicted climate warming is likely to shift range margins of all three conifer species upslope, but only if populations of spruce and larch are able to invade the resident pine belt. Whereas it is well established that the shade-intolerant pine is not able to recruit or even to maintain viable clonal populations in the understorey of spruce or larch forests (Michiels 1993), the effects of P. mugo on Picea abies and L. decidua are much less well known. Based on anecdotal evidence it has been suggested that pines may serve as ‘nurse plants’ for spruce and larch, improving local site conditions by accumulating humus and conveying shelter against frost, strong winds and browsing herbivores (Michiels 1993). However, these facilitative effects will probably be restricted to the early life-history stages of the two tree species, i.e. until they overtop the shrub layer. Additionally, low light conditions and raw humus accumulation below P. mugo have been documented severely to reduce germination and establishment rates of the pines themselves (Hafenscherer & Mayer 1986; Michiels 1993) and may have similar effects on spruce and larch. Moreover, L. decidua is a light-demanding tree the recruitment of which is sensitive to thick litter layers whereas Picea abies is comparatively shade-tolerant and its regeneration benefits from raw humus accumulation (Zukrigl 1973; Mayer 1976). It is thus likely that potentially negative effects of pines on the recruitment of the two trees will affect larch more severely than spruce. As a consequence, pine shrubland may represent a differentially permeable barrier if environmental changes trigger simultaneous range dynamics of Picea abies and L. decidua.

In the current study we focus on the effects pines may have on recruitment, growth rates, fecundity and browsing damage of spruce and larch. Based on the above considerations we derived three hypotheses. (i) Effects of P. mugo on Picea abies and L. decidua will involve both positive and negative components: browsing is likely to be reduced within dense pine shrublands whereas the effects on recruitment, growth and fecundity of the two tree species are difficult to predict a priori. (2) Effects of pines on growth rates of the two tree species will be restricted to the juvenile phase of the trees, i.e. while they are growing within the pine canopy. (3) Effects of pines will vary between the two tree species, at least for growth rates and recruitment. We tested these hypotheses by analysing observational data from 250 sampling plots that cover both the environmental variability of the studied tree-line system and a gradient of pine cover.


study area

The study area covers the subalpine and the lower alpine belt of Mts Hochschwab, Schneealpe, Rax and Schneeberg (north-eastern Calcareous Alps of Austria, 15–16° E and 47°30′−47°50′ N, approximately 150 km2; uppermost summit at 2273 m a.s.l.). These mountain ranges are formed from Mesozoic limestone and dolomite and are characterized by displaced plateaux of different altitudes with meso- and microrelief shaped by a variety of karst landforms such as dolines and karren. Climatic conditions are temperate humid. Mean annual temperature is approximately 6–8 °C in the valleys, decreasing to about 0–2 °C in the summit region. Annual precipitation averages 700 mm (valleys) and 1500–2500 mm (summits), with a distinct peak during the summer.

Summer pasturing in the region – mainly of cattle – dates back at least to the 16th century and has been documented over 30% of the study area. Since the mid 19th century, grazing intensity has decreased and much former pasture land has been abandoned, leaving only about 7.5% of the study area as cattle pasture today (Dullinger et al. 2003b). By contrast, as in most other parts of the European Alps (Breitenmoser 1998), regional populations of wild ungulates (roe deer, Capreolus capreolus L., red deer, Cervus elaphus L. and chamois, Rupicapra rupicapra L.), the main browsers on tree-line conifers, have recovered from extremely low levels during the 19th century. Although exact numbers are lacking, population densities of chamois and red deer, in particular, are known to be fairly high (Krause 1997).

Under natural conditions Norway spruce (Picea abies) or mixed spruce and European larch (Larix decidua) forests cover the lower parts of the subalpine belt, and dense, mono-dominant prostrate pine (Pinus mugo) shrubland (= krummholz) predominates at the tree-line ecotone between around 1700 m and 1900 m a.s.l. However, owing to the long history of summer farming and the rugged topography, the subalpine belt is currently a mosaic of woody and non-woody vegetation with the latter mainly consisting of different kinds of pastures and natural grasslands. Above the tree line, alpine grasslands prevail. Moreover, rock and scree vegetation is widespread from the valley bottoms up to the summits.

environmental data

A suite of spatially explicit data layers for climatic and topographical variables was generated based on a digital elevation model (resolution 20 × 20 m) and some additional data sets (see Dirnböck et al. 2003 for methodological details).

Climatic conditions were represented by annual degree days (= days with a mean daily temperature > 0 °C, DD), solar radiation income at the beginning (15 May, SRM), in the middle (15 July, SRJ) and at the end (15 September, SRS) of the growth period, and site water balance in August (WBA).

Topography was characterized by slope inclination (SLOPE), a wetness index (WET), a soil erosion index (EROS) and an estimate of topographically modified near-surface wind velocity during strong, north-westerly winds (WSP).

Bedrock mineralogy (limestone vs. dolomite) was derived from geological maps (scale 1 : 50 000, Geological Survey of Austria, unpublished; GEO). Information on soil type (either rendzic leptosol or chromic cambisol or a mixture of both; SOIL) was collected directly on the sampling plots (see below).

We used digital vegetation maps (scale 1 : 10 000, Greimler & Dirnböck 1996; Dirnböck & Greimler 1997; Dirnböck et al. 1998, 1999; Dullinger et al. 2001; W. Gatterbauer et al. unpublished data, R. Köck et al. unpublished data) together with infrared (acquired on 23 July 1994, pixel resolution 25 cm) and black-and-white (resolution 1 : 10 000, Austrian Mapping Agency) ortho-photographs to obtain information on distances of each sampling plot (see below) to the nearest stand of Picea abies (DISTPic) or Larix decidua (DISTLar). Distances were measured on-screen on aerial photographs using the distance function of ARC-Edit 8.1. The distance measurements were truncated at 500 m because we assumed seed input from trees further away to be negligible (Müller-Schneider 1986). For some sampling plots (spruce 11 plots, larch 11 plots) we imputed the mean value of the DISTPic or DISTLar variable, respectively, because we could not unambiguously decide if and where single individuals of one tree species were present in the surrounding stands dominated by the other one.

Data on land use history were compiled from various sources and combined into a map representing time since pasture abandonment transformed to a series of discrete time steps [1000 (= never used), 180, 110, 73, 40, 0 (= still in use) years; see Dullinger et al. (2003b) for details].

selection of sampling plots

Data on spruce and larch performance were collected on 250 20 × 20 m plots. One hundred and thirty-nine plots were situated in abandoned pastures with a maximum pine cover of about 40%. Position of these plots was selected by means of a stratified random sampling design (see Dullinger et al. 2003a for details). Overall, they span an altitudinal range of about 800 m. The remaining 111 plots were grouped into altitudinal line-transects of three plots each with individual plots 100 m apart. The position of the transects was preselected in the laboratory. We first stratified the area of the tree line-ecotone according to DD, WBA, SLOPE, WET and GEO. For the purpose of stratification, continuous variables were transformed to discrete classes. From each of the seven largest strata of the study area we then selected five GIS polygons at random. Using the ortho-photographs, we placed transects at the current forest–krummholz margin (= spruce/larch forest changing into pine–krummholz) within these polygons. The transects started where the total cover of upright trees dropped below around 40% and ran upslope into the krummholz belt. However, for some strata there were too few large polygons consistently to retain the line-transect design. In these cases, we determined the position of the first (lowest) plot as described above and placed the consecutive ones upslope as near as possible in polygons of the same stratum (4 of 35 transects). Additionally, we had sampled one transect and two single plots, the positions of which had been determined subjectively in the field, to test and standardize sampling methods among field-workers. Overall, the transects span a range of about 550 altitudinal metres.

All preselected plots were localized in the field by means of a GPS receiver.

data collection

We established the sampling plots using a tape measure and a compass and marked them temporarily by means of coloured ropes. Each plot was subdivided into four quarters with additional ropes tightened along the central cross. Recruitment data for spruce and larch were collected by counting seedlings (< 30 cm tall) and saplings (> 30 cm and < 1.3 m) in strips of 1 m width along this central cross (39 m2 overall). For each seedling and sapling we recorded its vitality (dead/alive), its height and, for those alive, the status of overall browsing damage using a four-level ordinal scale.

For sampling data on trees taller than 1.3 m we used the whole plot area. For each individual, we recorded vitality (dead/alive), diameter at breast height (d.b.h., measured at 130 cm height), height (using a tape measure for small and a SUUNTO© hypsometer for taller individuals), fecundity (number of cones produced) and damage status [due to ‘climatic’ constraints such as frost desiccation or snow-ice abrasion (CLD), browsing and undefined reasons (UND)]. Fecundity and damage were estimated on a four-level ordinal scale (see Table 1). From all individuals > 1.3 m and < 4 m tall we selected a maximum of three trees (smallest, tallest and median) per plot from spruce as well as from larch populations and estimated their age by counting terminal bud scars. A maximum of three individuals > 4 m (smallest, tallest and median) per species and per plot were cored as close to the root collar as possible. Cores were analysed using a digital positiometer. Ring widths were measured to the nearest 0.01 mm and tree age was determined by counting the annual rings. If the core had fallen short of the pith, the number of missing years was estimated: we first calculated the radius of a circle around the missing pith based on the curvature of the innermost ring and then estimated the number of missing years based on the average width of the five innermost rings. If the increment borer had missed the pith completely (n = 30 out of 226), i.e. the innermost rings of the core were apparently uncurved, individuals were excluded from further analyses.

Table 1.  Sample parameters of analysed response variables for Picea abies and Larix decidua. n = number of sampled plots for recruitment (in parentheses is the number of plots with at least one seedling or sapling of the respective species); and number of individuals for growth, fecundity and browsing damage. Growth rates were only calculated for individuals whose age had been determined. Fecundity and browsing damage were estimated on a four-level ordinal scale: fecundity: 0 – no cones, 1 – cones present but sparse, 2 – cones relatively abundant, 3 – cones highly abundant; browsing damage: 0 – no damage, 1 – branches damaged, 2 – treetops damaged, 3 – both branches and treetops damaged. Values represent numbers of individuals in the respective classes
Recruitment (seedlings and saplings/plot)
 Picea abies250 (69) 0.81.9015
 Larix decidua250 (39) 0.62016
Growth (cm yr−1)
 Picea abies312
 Larix decidua13912.27.31.731.4
 Picea abies107410263610 2
 Larix decidua 588 467572638
Browsing damage
 Picea abies1074 867635193
 Larix decidua 588 402771495

For each plot we additionally estimated the percentage cover of pines and their average height. All field data were collected in 2001 (July–October).

data analysis

We examined the effect of percentage pine cover on recruitment, growth, fecundity and browsing damage of both spruce and larch. Recruitment data were analysed at the plot level, and growth, fecundity and browsing damage at the level of individual trees. We used the ratio of tree height to tree age as an approximation for growth rate (= average lifetime growth rate).

For each response, the analysis started with a basic regression model. The type of the regression model was specific to each response, depending on its respective scaling: we used a Generalized Linear Model (GLM) assuming a Poisson distribution for recruitment data, ordinary least-squares regression (OLS, after log-transformation of the response) for growth, and proportional odds regression (POR = logistic regression for ordinal data) for fecundity and browsing damage. The basic regression models were established starting with full models. Full models included all environmental descriptors together with, for individual-level responses (growth, fecundity, browsing damage), damage status (CLD, UND) and an indicator of life-history stage or, for (plot-level) recruitment responses, variables accounting for seed input (see below). As an indicator for life-history stage we used whichever of tree height or tree d.b.h. maximized the variance explained by the respective model (the differences were only minor as the two variables are closely correlated; Pearson r2 = 0.9, n = 1394). The analysis proceeded with a backward selection procedure until all remaining predictors had significant partial effects on the response (P < 0.05). Significance of partial effects was evaluated by Wald tests. Non-linear effects were accounted for by using restricted cubic spline functions with three knots (Harrell 2001) with knots set to the quartiles of the respective predictor.

To evaluate the impact of pines on the performance of spruce and larch, we added percentage pine cover to the final basic models and tested its partial effect on each response, again including non-linearity by means of restricted cubic splines. We also tested for a significant improvement of overall model fit using analysis of deviance for comparison of nested models (e.g. Selvin 1998). For individual-level responses potential differences between spruce and larch were investigated by using species identity as an additional predictor and testing its interaction with pine cover. If this interaction proved to be significant, regression models were run separately for each species and partial effects of pine cover were compared among the two models. If spruce and larch growth is responsive to pine competition only in the juvenile stage, the effect of pine cover on the average lifetime growth rate (tree height : age ratio) should decrease with the size of the trees and we therefore tested the interaction between tree d.b.h. and percentage pine cover.

For the plot-level response (recruitment), all analyses starting from the basic models were performed separately for the two tree species. Apart from environmental predictors, the basic models included the distance to the nearest stand of the respective species and the summed d.b.h. of all trees of the respective species (= DBHSumPic, DBHSumLar) on the plot: in combination, these provide an estimate of seed input.

We used ARC-Info 8.1 for handling and analysing geographical data. All statistical analyses were performed with S-Plus 2000.


Overall, we collected data on 1130 Norway spruce (1074 alive, 66 dead) and 624 European larch (588 alive, 36 dead) individuals, 144 (spruce) and 94 (larch) of them saplings (i.e. between 0.3 and 1.3 m in height) and 48 seedlings (< 0.3 m) of each species on the central crosses of the plots. Sample parameters for the analysed response variables are given in Table 1. Owing to the sampling design, the frequency distribution of percentage pine cover across the plots has two distinct peaks at the very ends of the gradient, i.e. between 0% and 10% and between 90% and 100%.


In general, recruitment is scarce for both spruce and larch. Maxima of pooled seedlings and saplings are 15 (spruce) and 16 (larch) per plot, means are below one per plot for both species and many plots do not have a seedling or sapling of either species (Table 1). Basic regression models are similar for Picea abies and L. decidua in terms of predictors included as well as of goodness of fit (Table 2). Recruitment increases with seed input and with the number of degree days per year, with some additional effects of solar radiation income and topography.

Table 2.  Models for analysing effects of Pinus mugo on recruitment, growth, fecundity and browsing damage of Picea abies and Larix decidua. Basic models comprise site conditions together with descriptors either of life-history stage and damage status due to reasons other than browsing, or of seed input. Pine models are the same as basic models except for the inclusion of percentage pine cover as an additional predictor. TM – type of regression model (PGLM – Generalized Linear Model with log-link for Poisson distributed data; OLS – Ordinary Least-Squares model; POR – Proportional Odds model). d.f. – degrees of freedom in the data set (with those spent for the regression model in parentheses). Deviations from overall sums of sampled individuals listed in Table 1 are due to missing values in single predictors. ‘R2’ is (NullDeviance − ResidualDeviance)/NullDeviance for PGLM, least-squares R2 for OLS, and Nagelkerke's R2 for POR. Somers’Dxy is a measure of a logistic model's predictive discrimination based on the rank correlation between predicted probabilities of response and actually observed responses. Basic and pine models were compared by analysis of deviance for nested models with degrees of freedom equal to the additional parameters to be estimated. For abbreviations of predictors see text and Fig. 1
Basic modelPine modelComparison
Recruitment Picea abiesPGLM 249 (9)DD, SRJ, SLOPE, DBHSumPic, DISTPic0.520.551< 0.0001
Recruitment Larix deciduaPGLM 249 (9)DD, SRJ, WBA, EROS, DBHSumLar, DISTLar0.50.571< 0.0001
GrowthOLS 411 (12)Tree d.b.h., CLD, UND, DD, WET, soil type 0.380.511< 0.0001
FecundityPOR1625 (11)Tree height, DD, SRJ, WSP, WBA, SLOPE, soil type0.530.880.530.881   0.63
Browsing damagePOR1645 (7)Tree height, DD, SRS, WSP, SLOPE0.360.670.460.772< 0.0001

Introducing percentage pine cover as an additional predictor significantly improves goodness of fit for both species (Table 2). Increasing pine cover decreases recruitment of spruce and, to a greater degree, that of larch (Fig. 1). Given a reasonable amount of seed input, larch recruits more intensively than spruce at low pine cover whereas seedling numbers are about the same in dense pine stands (Fig. 2).

Figure 1.

Relative importance of predictors in regression models of recruitment, growth, fecundity and browsing damage of Picea abies and Larix decidua. Effect of cover of Pinus mugo is given in black, species identity in light grey and their interaction term in dark grey. The x-axis represents the Wald chi-squares statistic minus degrees of freedom of the respective predictor within the multiple regression. Predictors significant (P < 0.05) in the final model that includes pine cover, species identity and their interaction term are marked by an asterisk. CLD/UND – damage due to climatic constraints/undefined reasons, DBHSumPic/DBHSumLar – summed diameter at breast height of Picea abies/L. decidua individuals on the sampling plot, DD – degree days, DISTPic/DISTLar – distance of sampling plot to nearest Picea abies/L. decidua stand, EROS – soil erosion index, PPC – percentage pine cover, SLOPE – slope inclination, SOIL – soil type, Species – species identity (Picea abies or L. decidua), SRS/SRJ – solar radiation income in September/July, WBA – water balance in August, WET – soil wetness index, WSP – wind speed. Colons symbolize interactions.

Figure 2.

Partial effects of Pinus mugo cover on recruitment, growth, fecundity and browsing damage of Picea abies and Larix decidua in multiple regression models. Additional predictors were adjusted to the following values: Recruitment – DBHSumPic/DBHSumLar: 50 cm, DD: 225 days, DISTPic/DistLar: 0 m, SLOPE: 20°, SRJ: 26 MJ m−2 day−1; Growth – CLD: 0, d.b.h.: 20 cm, DD: 230 days, SOIL: rendzic leptosol, UND: 0, WET: 5.7; Fecundity – DD: 230 days, SLOPE: 20°, SOIL: rendzic leptosol, SRJ: 25 MJ m−2 day−1, Tree height: 13 m, WBA: 4 mm day−1, WSP: 10 m s−1; Browsing damage – DD: 230 days, SLOPE: 20°, SRS: 15 MJ m−2 day−1, Tree height: 2 m, WSP: 10 m s−1. Dashed lines represent 90% confidence intervals (grey –Picea abies, black –L. decidua). For abbreviations of predictors see text and Fig. 1.


The predominant effect of d.b.h. present in the basic model indicates a strong impact of life-history stage on growth rates of both spruce and larch. Damage, irrespective of its cause, obviously dampens growth rates, whereas climatic and topographic predictors per se have comparatively minor partial effects (Fig. 1).

Pine cover has a pronounced impact on growth rates. Within closed pine shrubland, height growth of the pooled spruce and larch individuals decreases to only half of the rates in open habitats. Goodness of fit increases considerably for a regression model that includes percentage pine cover as an additional predictor (Table 2) and Wald statistics indicate pine cover to be the second most important predictor after d.b.h. in the growth model (Fig. 1).

Not surprisingly, a significant partial effect of species identity (Fig. 1) demonstrates that growth rates differ, with larch growing faster than spruce on average (Table 1). However, similar to recruitment, percentage pine cover has a more severe partial effect on growth rates of larch than of spruce. Thus, growth rate differences among the two tree species decrease with increasing pine cover (Fig. 2).

Contrary to our prediction of an effect of life-history stage on this response, Wald statistics indicate only a marginally non-significant (P = 0.07) interaction between percentage pine cover and tree d.b.h. for larch and no effect for spruce (P = 0.31). Overall goodness of fit of the model is not significantly improved by integration of this interaction term for either spruce (P = 0.84, d.f. = 2) or larch (P = 0.58, d.f. = 2).


Fecundity is, of course, primarily a function of life-history stage, represented by tree height in the basic model. In combination with various environmental descriptors, this model allows for quite accurate prediction of ordinally scaled fecundity values (Table 2).

Adding pine cover to the suite of predictors does not improve the overall model's goodness of fit (Table 2). Accordingly, Wald statistics suggest that percentage cover of P. mugo does not affect the fecundity of the pooled larch and spruce individuals (Fig. 1). By contrast, species identity is a significant predictor, indicating cone crop to be generally more abundant for larch than for spruce in the studied tree-line environment, at least for the non-mast year investigated (Figs 1 and 2). However, the pine-cover – species interaction has no impact on fecundity (Fig. 1), indicating that pine cover is equally unimportant for fruit set of Picea abies and L. decidua.

browsing damage

The basic model for browsing damage has tree height as the most important predictor, reflecting that herbivores preferentially feed on juvenile trees with terminal buds within their reach. Apart from some temperature- and radiation-related trend environmental descriptors have limited predictive value (Fig. 1).

Of all responses analysed, browsing damage of spruce and larch is the one most sensitive to the presence of pines. Adding percentage pine cover to the basic regression model increases the model's goodness of fit significantly (Table 2) with pine cover being the most important single predictor within the extended model (Fig. 1). By contrast, neither species identity nor its interaction with percentage pine cover are significant. Closed pine shrubland thus protects both spruce and larch rather effectively against browsing herbivores with little difference in the facilitative effect on the two tree species (Fig. 2).


In line with our general expectations, the results suggest that the interaction among these three tree-line species involves both negative and positive components and that the effects of pines on spruce and larch are species-specific. Whereas recruitment and growth of Picea abies and L. decidua are a negative function of P. mugo cover, fecundity levels are unaffected by shrub layers in both species. Despite this overall similarity between spruce and larch, the intensity of the effects varies considerably, with recruitment and growth of larch repressed by pine–krummholz more severely than that of spruce. By contrast, dense pine cover has a marked, and similar, facilitative effect on Picea abies and L. decidua, providing shelter against browsing herbivores.

Taken together, our results thus corroborate the recent theoretical emphasis on the complex nature of plant–plant interactions (Callaway & Walker 1997; Holmgren et al. 1997; Stachowicz 2001; Bruno et al. 2003): pines obviously compete with spruce and larch for resources such as light and nutrients but, simultaneously, protect the juvenile trees against browsing herbivores. Similar interaction patterns have also been demonstrated in other environments, such as arid subtropical habitats (Callaway et al. 1996). As both growth and recruitment of larch and spruce are a negative function of pine cover but fecundity is unaffected, the balance of effects is presumably negative, with facilitation simply buffering, to some extent, the competitive effects of P. mugo on indicators of individual performance as well as of population dynamics. This finding somewhat contradicts the hypothesis that the net plant–plant interactions tend to shift towards facilitation under harsh abiotic conditions like those prevailing at the tree line (Callaway 1995; Callaway et al. 2002). In fact, facilitation has primarily been demonstrated in extreme environments such as arid ecosystems (Nobel & Franco 1989; Holzapfel & Mahall 1999), salt marshes (Bertness & Shumway 1993; Callaway & Pennings 2000) and high mountain habitats (Callaway 1998; Choler et al. 2001; Callaway et al. 2002). For conifers at the tree line in particular, advantages of both intraspecific (Srutek et al. 2002) and interspecific (Callaway 1998) spatial clumping have been reported. The contrasting results of our study may, at least in part, be due to the fact that we did not analyse interactions among species of the same but of different growth forms. The predominance of facilitation among different tree species in a Rocky Mountains tree-line ecosystem has primarily been attributed to the reciprocal shelter that individuals – irrespective of their species identity – provide against climatic constraints such as snow-ice abrasion (Callaway 1998). However, these constraints affect trees in high mountain environments most strongly when they are growing above the more favourable climate near the ground (Geiger 1965; Körner 1999). Thus, shrubby pines, which themselves hardly grow higher than 2 m in the studied tree-line system, have rather little potential to protect upright larch and spruce against such stress.

Despite their shrubby growth form, and contrary to our expectations, the negative effects of pines on growth rates of spruce and larch are not restricted to the juvenile phase of the trees: the regression models indicate that both tree d.b.h. and percentage pine cover strongly affect growth of spruce as well as of larch, but their interaction term does not. If competition for light were the dominant component of the interaction between P. mugo and the two tree species, we should observe a competitive release, indicated by sharply increasing growth rates, when trees overtop the shrub layer (e.g. Frelich 2002). As a consequence, the dependence of average lifetime growth rates (tree height : tree age ratios) on pine cover should decrease with the size of the trees. However, the data provide no evidence for such a competitive release for Picea abies and only a marginal indication for L. decidua. There are at least two possible, and not mutually exclusive, explanations for this unexpected finding. On the one hand, when trees grow above the shrub layer and escape the low light conditions below pines, they become exposed to climatic constraints that had been ameliorated within the pine canopy. Overtopping the shrub layers thus involves a trade-off between improved light supply and increased risk of damage due to frost, frost desiccation, snow-ice abrasion and other consequences of the harsh climatic conditions at the tree line. On the other hand, dense pine canopies may have negative effects on growth rates of both juvenile and adult spruce and larch individuals, due to shading, and thus depressed soil temperatures. Low soil temperatures limit root activity with a feedback on shoot development and growth, a mechanism that has recently been suggested to be a main reason for tree-line formation (Körner 1999). In fact, measurements of soil temperatures below dense pine thickets and in nearby grasslands within our study area revealed considerable differences. Soils below pine–krummholz are between 0.5 °C (45 cm depth) and 4 °C (5 cm depth) cooler on average than soils below grasslands during the vegetation period (1 May to 31 October 2000; Köck et al. 2003).

As hypothesized, pines affect spruce and larch differently with respect to some, but not all, of the analysed interaction components. Protection against herbivores is most probably due to the reduced visibility of juvenile trees within dense shrub layers (Callaway 1992; Rousset & Lepart 2000; Garcia & Obeso 2003; Rao et al. 2003; Russel & Fowler 2004) and it is therefore not surprising that this effect does not depend on species identity. Concerning fecundity, comparable studies on the impact of shrub understories on seed production of trees are largely lacking. Our results suggest that this impact may generally be negligible, at least within our study system. However, caution should be taken when interpreting fecundity data from only one year (Clark et al. 1999), especially if a masting species like spruce is involved. Additional data that cover a time series involving at least one mast year of spruce will be necessary to confirm our preliminary results.

In contrast to browsing damage and fecundity, larch is more sensitive to pine competition in terms of recruitment and growth. Its superior regeneration and growth in open habitats such as grasslands is reduced in dense pine thickets. This result is in line with the conventional wisdom of foresters that classifies larch as a light-demanding early to mid-successional and spruce as a shade-tolerant mid- to late-successional species in the subalpine forests of the northern Alps (Zukrigl 1973; Mayer 1976). Although not surprising, these findings may have considerable implications for environmental changes such as climate warming that trigger a simultaneous range expansion of pine shrublands as well as of spruce and larch forests at the expense of subalpine and alpine grasslands (Dullinger et al. 2004). The encroachment of grasslands and other non-forest habitats by pine–krummholz involves a reduction of those parts of the landscape where larch recruits more effectively and grows faster than the late-successional spruce. In other words, the invasibility of the resident vegetation cover decreases more sharply for larch than for spruce, if pine shrublands replace grasslands and other non-forest habitats. Hence, the differential effects of pines on spruce and larch will probably bias the rates of climate-change-triggered range shifts of these two tree species, favouring spruce at the expense of larch. We thus conclude that disregarding the complex details of plant–plant interactions will probably result in unrealistic predictions of species responses to environmental changes.


We are grateful to M. Steinkellner, G. Mandl, G. Bryda and R. Tscheliesnig for meteorological, geological and land-use data, and to N. Zimmermann and A. Bachmann for GIS software. K. Hülber, I. Kleinbauer, D. Moser and two anonymous referees provided helpful comments on earlier versions of the manuscript. Barbara Holzinger and Lindsay Haddon helped with the language. Data collection was funded by the Austrian Federal Ministry for Education, Science and Culture and the Water Department of the Viennese administration.