1There has been much debate about how losses of species and functional groups may affect the invasibility of vegetation, but little is understood about how invasibility differs across ecosystems or is driven by environmental context.
2We studied the invasibility of field plots in two ongoing removal experiments set up across thirty lake islands in northern Sweden. These islands differ in size, and therefore soil fertility and productivity. One experiment involves full factorial removal of three functional groups (dwarf shrubs, mosses and tree roots), and the other involves full factorial removal of three species of dwarf shrub (Vaccinium myrtillus, V. vitis-idaea and Empetrum hermaphroditum).
3We investigated the effects of removal treatments in both experiments on the invasibility of each of three species (Betula pubescens, Pinus sylvestris and Picea abies). This included a seed sowing study, and a seedling planting study, for each of the three species.
4For the functional group experiment, removal of shrubs promoted invasibility by all species, and removal of mosses also had positive effects. For the species removal experiment, the two Vaccinium species exerted the strongest effects against invasibility. The floristic components that had the greatest effects represented only a small proportion of total plant biomass.
5The effects of the removal of shrubs (or of either Vaccinium species) on invasibility often varied across island size classes. In these cases, removals usually had the greatest positive effects on the largest and most productive islands. In contrast, the effects of moss removals on P. sylvestris seedling survival were greatest on small islands.
6These results show clearly that the effects of loss of components of the resident flora (at either the functional group or species level) on invasibility at the plot scale are context dependent, and can vary greatly across ecosystems.
7Synthesis. Our results contribute to the ongoing debate about how loss of species and functional groups influences community-level processes, by showing that the effects of loss of resident biota on invasion of new species depends on the attributes of the biota that are lost and the ecosystems that they are lost from.
Plant community ecologists have long been interested in how community-level characteristics and neighbourhood conditions influence plant colonization and growth (Grubb 1977; Grime 1979; Goldberg 1987; Howard & Goldberg 2001). One aspect of this issue that has attracted considerable attention over the past decade focuses on whether and if so how species and functional group diversity of plant communities (i.e., their composition and richness; Hooper et al. 2005) influences community invasibility (Levine & D’Antonio 1999; Fridley et al. 2007). This topic has often been studied in the context of consequences of human-induced biodiversity loss in plant communities for the ingress of new species (Kennedy et al. 2002; Hooper et al. 2005). Several short-term experimental studies that have utilized randomly assembled plant communities have found invasibility to be negatively associated with community diversity, suggesting that loss of species or functional groups from these communities may promote invasibility (e.g., Levine 2000; Kennedy et al. 2002; Fargione & Tilman 2005). This is consistent with the predictions of long-standing theory (Elton 1958; MacArthur 1970). However, these findings are not consistent with much observational evidence, or with experiments performed on non-randomly assembled plant communities, that point to a positive association between plant diversity and success of invasion by new species (e.g., Wiser et al. 1998; Stohlgren et al. 1999; Cleland et al. 2004; Emery & Gross 2006). Further, several studies suggest that species identity can be an important driver of invasibility (Crawley et al. 1999; Davis et al. 2000; Emery & Gross 2006), meaning that the effects of species loss from the community on invasibility should depend on the attributes of those plant species that are lost. For example, the ‘mass ratio hypothesis’ predicts that the greatest effects of species loss should result from loss of those species that occupy the greatest biomass in the community (Grime 1998).
There is increasing recognition that the ecological impact of plant diversity can be influenced by environmental conditions (Fridley 2002; Wardle & Zackrisson 2005), and this may explain the discrepancy of results across studies. Given that the effects of plant diversity on resource availability and community competitiveness can depend on environmental context (Hooper et al. 2005; Wardle & Zackrisson 2005), and that resources and community competitiveness affect the invasibility of the community (Davis et al. 2000; Davis & Pelsor 2001; Wardle 2001), it is plausible that the effects of plant diversity on invasibility may also be context dependent (Davies et al. 2007; Maron and Marler 2007). Direct empirical evidence of the context dependency of the relationship between community diversity and invasibility in natural ecosystems is scarce, but a better understanding of this will inform on how the effects of species or functional group loss on invasibility may vary among ecosystems.
Our study system consists of a series of 30 lake islands in northern Sweden. These islands vary greatly in disturbance history; larger islands are struck by lightning more often than are smaller ones and have therefore burned more frequently (Wardle et al. 1997, 2003). As such, some islands have burned in the last century, while others have not burned for over 5000 years (Wardle et al. 2003). It has previously been shown that these islands collectively form a retrogressive chronosequence, with reductions in nutrient availability and plant productivity with increasing time since fire and decreasing island size (Wardle et al. 2003, 2004). The system consists of three plant functional groups (Wardle & Zackrisson 2005), namely trees (dominated by Pinus sylvestris, Picea abies and Betula pubescens), dwarf shrubs (dominated by Vaccinium myrtillus, V. vitis-idaea and Empetrum hermaphroditum), and feather mosses (dominated by Pleurozium schreberi and Hylocomium splendens).
In the present study, we utilize long-term experimental plots (initiated in 1996) in which combinations of the three functional groups, and of the three species within one functional group (dwarf shrubs) have been experimentally removed on each of the 30 islands (Wardle & Zackrisson 2005). We use these plots to determine whether and how the loss of plant species and functional groups affects the invasibility by plant species that are not present in the plots, by addressing the following hypotheses: (i) Effects of removal of components of the vegetation (species and functional groups) on invasibility are positive, and greatest for those components that occupy the most biomass as predicted by the ‘mass ratio hypothesis’; (ii) There are important interactive effects between removal treatments and island size, with the effects of species or functional group removals on invasibility being greater on larger islands that are more nutrient rich and support a more productive vegetation; and (iii) Increasing island size in its own right influences invasibility, either positively in the absence of vegetation (because more soil resources are available) or negatively in the presence of vegetation (because existing vegetation is more productive and provides a more competitive environment). By addressing these hypotheses we will be able to assess the extent to which the effect of loss of plant species and functional groups is dependent upon environmental context.
The study system consists of 30 forested lake islands in lakes Hornavan and Uddjaure in the boreal zone of northern Sweden (65°55′ N to 66°09′ N; 17°43′ E to 17°55′ E). The mean annual precipitation is 750 mm, and the mean temperature is +13 °C in July and –14 °C in January. The islands were divided into three size classes with 10 islands per class: large (> 1.0 ha), medium (0.1–1.0 ha) and small (< 0.1 ha), with a mean time since last major fire of 585, 2180 and 3250 years, respectively (Wardle et al. 2003).
We utilized a ‘removal experiment’ approach, because removal experiments are powerful tools for studying the effects of local, non-random losses of biotic components in natural ecosystems (Díaz et al. 2003). In August 1996, we established 14 experimental plots on each of the 30 islands, each representing a different removal treatment of functional groups or species (Wardle & Zackrisson 2005). The study was conducted as two experiments. For the ‘functional group removal’ experiment, this consisted of a full factorial combination of tree root removal, ericaceous dwarf shrub removal and moss removal (3 factors, 8 treatments in total). For the ‘species removal’ experiment, this consisted of a full factorial of the removal of V. myrtillus, removal of V. vitis-idaea and removal of E. hermaphroditum (3 factors, 8 treatments in total). These shrubs dominate the ericaceous shrub layer in large, medium and small islands respectively in terms of both biomass and productivity (Wardle et al. 1997, 2003; see Table S1 in Supporting Information). Two treatment plots (removal of all shrubs, and no removals) were common to both experiments, yielding 14 plots per island or 420 plots in total. The three manipulated functional groups represent 99% of the plant biomass present, and the three manipulated shrub species represent 98% of total biomass in the dwarf shrub layer. All plots were 55 × 55 cm, but only the inner 45 × 45 cm were used. All plots were located at similar distances from the shore for each island regardless of island size, to prevent edge and microclimatic effects from confounding the results (Wardle et al. 1997, 2003). To our knowledge it is the longest running biodiversity manipulation study established across an environmental gradient, and the longest running in a non-grassland setting. Further details about the experiments are given in Wardle & Zackrisson (2005).
The experiment has been maintained annually ever since set-up in 1996 (Wardle & Zackrisson 2005). Tree removals have been performed only below-ground, by annual root trenching to below the tree rooting zone around the plot borders (Coomes & Grubb 2000). Dwarf shrub and moss manipulation treatments have been conducted through annual physical removal of vegetation (Díaz et al. 2003). It is recognised that both root trenching and vegetation removals impose initial disturbance effects, but these are likely to be transient (Coomes & Grubb 2000; Díaz et al. 2003) and of minimal importance after the first few years; in this light it is relevant that the relative effects of different treatments on vegetation measurements performed in these plots have varied little after 2000 (Wardle & Zackrisson 2005). Physical removal of shrubs and mosses each year is likely to represent a negligible disturbance effect, because the shrubs are clonal and only slowly grow into the plots from outside them, and because the mosses that are removed do not penetrate deeply into the soil substrate.
Over 4–15 August 2007, the total cover of each of the three dwarf shrub species in each plot was assessed by point quadrat analysis, through determining the total number of times the vegetation of that species was intercepted by a total of 100 downwardly projecting points (Wardle et al. 2003). Calibration equations derived for each shrub species (Wardle et al. 2003) were used to convert total numbers of point intercepts of that species to biomass per unit area. The depth (mm) of the two dominant moss species was also determined at this time for 10 positions in each plot. Calibration equations derived for each moss species (Lagerström et al. 2007) were used to convert mean moss depth in each plot to biomass per unit area.
sowing and seedling studies
We used seed sowing and seedling planting studies to study the invasibility of each plot in both experiments. The invaders that we used were seeds and seedlings of three boreal forest tree species, that is, B. pubescens, P. sylvestris and P. abies. These species were chosen because they are not part of the vegetative ground cover of any of the plots used in the experiments, and because they vary in their responses to nutrient availability and soil fertility (P. sylvestris is the most adapted of the three to low fertility while B. pubescens is the least; Dehlin et al. 2004). Note that the ‘invaders’ in our study are not alien species, and our use of the term ‘invader’ for organisms that are in the local species pool but excluded from our treatment plots is consistent with previous studies on biodiversity–invasibility relationships (e.g., Fargione & Tilman 2005; Crutsinger et al. 2008). For all experiments we used seeds supplied by Svenska Skogsplantor AB collected from northern Sweden and Finland, and the viability of seeds was 68.3%, 99.5% and 99.0% for B. pubescens, P. sylvestris and P. abies respectively.
We set-up a seed sowing study on all plots over 20–30 June 2004, using the approach described by Zackrisson et al. (1997) and Nilsson et al. (2000) for boreal communities. This involved identifying four positions in each plot, each approximately 10 × 10 cm. Three of these positions were each sown with 100 seeds (one position for each tree species) lightly sprinkled over the ground surface. The fourth was left unplanted to serve as a control. A plastic mesh cage (40 × 40 × 10 cm high, with 1 × 1 cm holes) was placed over each plot and anchored to the ground surface on each side with plastic sticks, to prevent birds from consuming the seeds. All seedlings that emerged in each of the four positions per plot were counted over 7–20 August 2004 (T1), with islands visited over this period in the same order as they were sown. Subsequent counts of the numbers of seedlings remaining at each position on each plot were performed over 6–16 June 2005 (T2), 28 July to 9 August 2005 (T3), 11–15 June 2006 (T4), 6–17 August 2006 (T5), 14–17 June 2007 (T6) and 4–15 August 2007 (T7). The growing season at this latitude is short, and June and August represent the early and late portions of the growing season. Occurrence of seedlings in the control portion of the plot was negligible, with two or fewer seedlings being present across all 420 plots each year.
We set up a seedling planting study on all plots over 6–16 June 2005. This involved germinating seeds of each of the tree species in trays of moist coarse sterile sand; these were transplanted into plots when they had reached the first true leaf stage. For each plot, three positions were chosen, each around 10 × 10 cm, and at least 10 cm from the edge of any of the positions used for the seed sowing experiment in that plot. Each of these positions was planted with 10 seedlings (one position per tree species), with adjacent seedlings at least 1 cm apart. All seedlings were planted carefully into a small hole made by a plastic stick. The number of planted seedlings that were alive were counted during 28 July to 9 August 2005 (T1), 11–15 June 2006 (T2), 6–17 August 2006 (T3), 14–17 June 2007 (T4) and 4–15 August 2007 (T5).
Because sufficient numbers of seedlings of P. sylvestris and P. abies had survived by August 2007 in the seedling sowing study, the height (mm) of the largest surviving seedling of each of these species in each plot was recorded. The largest seedling rather than the average of all seedlings in each plot were used because of high variation in numbers of surviving seedlings across plots. For each of 30 seedlings of each species, dry mass (80 oC, 24 h) was also determined by using a representative subset of the seedlings within the plots, to develop calibration relationships to convert mass to height. These relationships were:
P. sylvestris: B = 1.51H (R2 = 0.8555, P < 0.001) P. abies: B = 0.109H + 0.0358H2 (R2 = 0.921, P < 0.001)
where H is seedling height (mm) and B is seedling biomass (mg). All seedling heights were converted to biomass using these calibrations.
All data was analyzed by anova, as previously used in this removal experiment by Wardle and Zackrisson (2005), and with individual islands serving as the units of replication. Data for the numbers of seedlings present of each tree species in both the seed sowing and seedling planting studies was analyzed using Repeated Measures anova. For the ‘functional group removal’ experiment, between-subject factors used in the anova were island size class (small, medium or large), tree roots (removed or not removed), mosses (removed or not removed) and dwarf shrubs (removed or not removed), with the separate measurement times serving as the within-subject factor. For the ‘species removal’ experiment, between-subject factors used in the anova were island size class, V. myrtillus (removed or not removed), V. vitis-idaea (removed or not removed) and E. hermaphroditum (removed or not removed), with the separate measurement times serving as the within-subject factor. Following each Repeated Measures anova, univariate four-way anovas were performed separately for each of the measurement times, with P-values for treatment effects adjusted by Bonferroni corrections to correct for the number of measurement dates. Seedling biomass data for P. sylvestris and P. abies was also analyzed using four-way univariate anovas testing for the effects of island size class and removal treatments as described above. Whenever there was a significant interactive effect between a given removal treatment and island size class on a response variable, this was interpreted as a context-dependent effect of that removal treatment (Wardle & Zackrisson 2005). Data was transformed as necessary (by ln(X + 1)) to satisfy assumptions of normality and homogeneity of variances required for anova.
shrub and moss biomass
For the functional group removal experiment, moss biomass in 2007 was not significantly affected by tree root removal (F1,108 = 1.17, P = 0.281), shrub removal (F1,108 = 1.81, P = 0.188), or island size class (F2,108 = 1.15, P = 0.320); mean moss biomass in all non-moss-removal plots was 318 g/m2. Dwarf shrub biomass in 2007 did not respond to tree root removal (F1,108 = 1.08, P = 0.098) or moss removal (F1,108 = 0.30, P = 0.584), but did respond to island size class (F2,108 = 6.21, P = 0.003). Across those treatments with dwarf shrubs present, shrub biomass was greater on large (223 g/m2) and small (243 g/m2) than medium (198 g/m2) islands. In the plots with no removals performed, dwarf shrub biomass was unaffected by island size (although shrub productivity is known to be less on the small islands) (Table S1); in these plots V. myrtillus, V. vitis-idaea and E. hermaphroditum had the greatest biomass on the large, medium and small islands respectively (Table S1). For the species removal experiment, no treatment or treatment combination affected moss biomass. However, there were important effects of all plant species removal treatments on total shrub biomass, and significant (P = 0.05) interactions between island size and removal of E. hermaphroditum or V. myrtillus (Fig. S1). The interaction between island size and V. vitis-idaea removal was also almost significant (Fig. S1). For small islands, all two-species removal combinations reduced total shrub biomass (Fig. S1). For medium islands all treatments that involved removal of V. myrtillus and/or V. vitis-idaea reduced shrub biomass, while for large islands all treatments that involved removal of V. myrtillus reduced shrub biomass (Fig. S1).
functional group removal experiment
For the seed sowing study performed in the functional group removal experiment, the number of B. pubescens seedlings present was low (< 2% of sown seeds) in all treatment combinations. However, B. pubescens seedlings were significantly promoted by removal of dwarf shrubs and mosses, especially during the initial measurement dates (Table S2, Fig. 1). During the first three of the four measurement times, B. pubescens seedlings were also significantly positively influenced by island size (Table S2, Fig. 1). There were no consistent interactive effects among removal treatments, or between removal treatments and island size class, on B. pubescens seedling numbers. From August 2006 (T5) onwards, seedling numbers in all treatments was negligible, that is, < 0.1% of total number of seeds initially sown. Meanwhile, P. sylvestris and P. abies seedling numbers were promoted by shrub removal, and both showed significantly greater numbers on small than on large islands (Table S2, Fig. 2). Moss removal had no effect on either species, although there was a weak positive interactive effect between moss removal and shrub removal on P. sylvestris seedlings at one measurement date (Table S2). There were significant interactive effects between island size class and shrub removal on both P. sylvestris and P. abies seedling numbers during the last few measurement dates (Table S2). For both species, shrub removals initially had significantly greater positive effects on seedling numbers for large than for small islands, though at the last sampling date there were no seedlings remaining in the plots with shrubs present for any island size class (Fig. 2).
For the seedling planting study performed in the functional group experiment, all three planted species showed significant positive responses to both moss removal and shrub removal; root removal had no effect (Table S3, Fig. 3). Further, there were strong interactive effects between moss removal and island size class on both B. pubescens and P. abies seedlings, which became particularly strong later in the study (Table S3). Over time, seedlings of both species became increasingly scarce unless both shrubs and mosses were removed (Fig. 3). Only P. sylvestris seedling numbers showed a significant relationship with island size (Table S3), with most seedlings on small islands (Fig. 4). There was also a significant interactive effect between moss removal and island size on P. sylvestris seedlings (Table S3), with moss removal stimulating seedling survival on small and (for some measurement occasions) medium islands, but not on large islands (Fig. 4).
For the seedling sowing study in the functional group experiment, there were sufficient numbers of planted seedlings of P. sylvestris and P. abies by T5 to determine the effects of each experimental factor on seedling biomass, but not the effects of interactions among these factors. Seedling biomass of both species was promoted by removal of mosses and shrubs, but not by removal of tree roots (Table 1). For both species, seedling biomass was least on small islands (Table 2).
Table 1. Effects of experimental removal treatments on biomass (mg) of the largest surviving seedling present in each plot in August 2007 (T5), for seedlings planted in June 2005
Species (S) or Functional Group (FG) treatment
Seedling biomass (mg)
S or FG removed
S or FG present
Degrees of freedom (d.f.) for removal factors are 1 in all cases, and residual d.fs are 121 (P. sylvestris, functional group removal), 52 (P. abies, functional group removal) and 74 (P. sylvestris, species removal).
Note: insufficient plots with live seedlings of P. abies (species removal experiment) and B. pubescens (both experiments) at the time of harvest (August 2007) to perform data analysis.
Values in bold are significant at P = 0.05.
Functional group removal
Table 2. Effects of island size class on biomass (mg) of the largest surviving seedling present in each plot in August 2007 (T5), for seedlings planted in June 2005
Island size class
Within each row, numbers followed by the same letter are not significantly different at P = 0.05 (Tukey's test).
Degrees of freedom (d.f.) for island size class are 2 in all cases, and residual d.fs are 121 (P. sylvestris, functional group removal), 52 (P. abies, functional group removal) and 74 (P. sylvestris, species removal).
Note: insufficient plots with live seedlings of P. abies (species removal experiment) or B. pubescens (both experiments) at the time of harvest (August 2007) to perform data analysis.
Values in bold are significant at P = 0.05.
Functional group removal
species removal experiment
For the seed sowing study performed in the species removal experiment, all species removal treatments influenced B. pubescens seedling numbers, but univariate anovas revealed no significant treatment effects after T2 (Table S4). Seedling numbers were significantly greater in the all shrub removal treatments than in those treatments that contained both E. hermaphroditum and V. vitis-idaea (Fig. 5). There was also an interactive effect between island size and V. vitis-idaea removal, with a significantly greater influence of removals on large than on small islands (Fig. 5). For P. sylvestris, removal of V. vitis-idaea and of V. myrtillus both promoted seedling numbers for the first few measurement dates (Table S4, Fig. 6). Further, there were significant interactive effects between island size and removal of either V. myrtillus or V. vitis-idaea at T1 (Table S4). Removal of V. myrtillus had significant effects only on large islands, while removal of V. vitis-idaea had significant effects only on medium islands (Fig. 6). The only removal treatment that influenced P. abies seedling numbers was removal of V. vitis-idaea, and this was significant only until T4 (Table S4). As such, at T1 the mean seedling numbers per plot were 1.54 when V. vitis-idaea was absent and 3.55 when it was present; at T4 the corresponding values were 0.15 and 0.51. Seedling densities of all three sown species showed similar relationships with island size to what was found for the seed sowing study in the functional group removal experiment, with seedling numbers of B. pubescens being greatest on large islands, and numbers of P. sylvestris and P. abies being greatest on small islands (data not presented).
For the seedling planting study performed in the species removal experiment, B. pubescens seedling numbers showed significant responses at T1 and T2 to removals of all three shrub species, as well as to the two way interactions between these removals (Table S5). This was because seedling survival was negligible in all plots unless all three shrub species were also absent. At T1 the mean seedling number present for plots with all shrubs removed was 0.533 and the mean number for plots with any shrubs present was 0.019. At T2 the corresponding values were 0.267 and 0.004. Seedling numbers of B. pubescens were unaffected by island size (Table S5). Seedling numbers of P. sylvestris responded positively to removals of both Vaccinium species throughout the study (Table S5, Fig. 7). Seedling densities were also initially influenced by island size (being least on small islands) (Fig. 7), as well as by interactive effects of both Vaccinium species with island size and with each other (Table S5, Fig. 7). Initially, effects of removals of both Vaccinium species were greater on medium and large islands than on small islands (Fig. 7). Seedling numbers of P. abies showed very similar responses to P. sylvestris to all factors (Table S5, Fig. 8), except that the effect of V. myrtillus removal was significant only until T3, and that there were no significant two-way interactive effects of species removals (Table S5).
Sufficient numbers of planted seedlings were present for P. sylvestris in the species removal experiment at T5 for data analysis to be performed. Here, seedling biomass was promoted by removal of V. myrtillus and V. vitis-idaea but not by removal of E. hermaphroditum (Table 1), and was least on small islands (Table 2).
Our study enabled us to study the effects of loss of plant functional groups and species on three components of the success of the invaders, that is, seedling emergence (seed sowing study), survival of established seedlings (planted seedling study) and seedling growth. For the functional group removal experiment, dwarf shrubs had the strongest effects of the three groups present, and when significant these effects were always negative. Mosses also had negative effects on seedling survival and growth of all species, but adversely influenced emergence of only B. pubescens. Shrubs had greater effects than did mosses, despite mosses on average having a higher biomass in the plots. Meanwhile, tree roots had no detectable effect. This suggests that trees as a functional group are unimportant in influencing plot colonization at least by below-ground mechanisms, despite occupying the majority of the biomass on the islands (Wardle et al. 1997). It is also relevant in this context that light interception by trees on the islands in August (when leaf canopy cover is maximal) is characteristically < 70% (Wardle et al. 2003; Wardle and Zackrisson 2005), and that this is less than the degree of light interception usually required to impair seedling performance of the three invader species (Dehlin et al. 2004).
These results pointing to strong competitive suppression of the invaders by understorey vegetation are consistent with other studies showing strong negative effects of dwarf shrubs (Nilsson & Wardle 2005) and mosses (Zackrisson et al. 1997; Spackova & Leps 2004; Gough 2006) on the recruitment of new plant species. There was never any evidence of facilitation of the invaders by understorey plants, in contrast to other studies providing some evidence of facilitation under comparable climatic regimes (Choler et al. 2001; Callaway et al. 2002). Instead our results are partially consistent with our first hypothesis predicting positive effects of removals on invaders, at least for mosses and dwarf shrubs. However, as dwarf shrubs occupy less than 10% of the total plant biomass on the islands (Wardle et al. 2003), the fact that they emerged as the plant functional group that was most effective at reducing plot invasibility is inconsistent with both the ‘mass ratio hypothesis’ (Grime 1998), as well as our first hypothesis.
The species removal experiment (which focused on the dwarf shrub functional group) revealed that two of the three dominant shrub species (i.e., V. myrtillus and V. vitis-idaea) were largely responsible for impairing success of the invaders. Both species suppressed seedling emergence, survival and growth in all cases, except that V. myrtillus did not affect P. abies in the seed sowing experiment. Despite their low standing biomass, previous work on these islands shows that in comparison with E. hermaphroditum, these two shrub species are often productive (Table S1), have high turnover rates of tissues (62% and 39% of above-ground tissues turn over annually for V. myrtillus and V. vitis-idaea respectively), are highly competitive, and often greatly deplete levels of available nitrogen present in the soil (Wardle & Zackrisson 2005). Further, Vaccinium species have considerably larger leaves than does E. hermaphroditum, which may allow significantly lower light transmission (Karlsson 1987) and therefore the amount of light that reaches the invaders. These attributes presumably contribute to the role of two Vaccinium species as the most important components of the flora for reducing plot invasibility. Empetrum hermaphroditum had generally weak effects on the invaders, despite previous studies pointing to its potential to inhibit establishment of other plant species through allelochemical production (Nilsson 1994; Nilsson et al. 2000). However, the greatest negative effects of E. hermaphroditum that we found were against B. pubescens; seedlings of Betula species are known to be more sensitive to allelochemicals produced by dwarf shrubs than are those of conifers (Wardle et al. 1998). While each of the three species serve as the dominant shrub biomass component on at least some islands (Table S1), the fact that they differ overall in their effects relative to each other is inconsistent with the first hypothesis.
A primary focus for this study was whether the effects of functional groups or species loss on invasibility are context-dependent, that is, affected by island size class, and therefore nutrient availability and vegetation properties. There were several instances in which context dependency occurred, both for the seed sowing and seedling planting studies. Most of these context dependent effects involved the response of invasibility to removal of all shrubs, or to removal of V. myrtillus and/or V. vitis-idaea, varying across island size classes (Tables S2–S5). Removal of all shrubs had a greater effect on seedling emergence of both P. sylvestris and P. abies on large than small islands. This is despite total shrub biomass (and therefore shrub mass removed) being constant across the three island size classes, but is consistent with greater total shrub productivity on the larger islands (Table S1; Wardle et al. 2003). Vaccinium myrtillus and V. vitis-idaea often exerted large effects on invasibility for the island size class in which they had the greatest biomass, that is, large islands for V. myrtillus and medium islands for V. vitis-idaea. This means that effects of removal of a species is probably important in ecosystems in which it is the dominant biomass component (Wardle and Zackrisson 2005). However, V. myrtillus often had much greater effects on invasibility on medium islands than on small islands, despite having a similar biomass on both (Table S1). Similarly, although V. vitis-idaea had a similar biomass on large and small islands, it often had much greater effects on large islands. As such, our results show that there must be factors other than simply the amount of biomass present that determine how the two Vaccinium species affect invasibility, and highlight that both species are less competitive against invaders in environments that have lower productivity and nutrient availability, regardless of their biomass. These results are consistent with earlier work on these plots which have shown that both species have their greatest effects on soil properties and nutrient availability on large and medium islands (Wardle & Zackrisson 2005). In any case, these results are consistent with our second hypothesis, that removal effects are generally greater in more productive environments that have greater nutrient availability.
In contrast, E. hermaphroditum had few effects on invasibility even on the small islands in which it dominates the shrub layer. This is despite E. hermaphroditum having as much biomass on small islands as V. myrtillus had on large islands (Table S1). These results suggest that species that dominate the shrub layer in productive and relatively nutrient rich ecosystems (in this case, Vaccinium species on medium and large islands) have negative consequences for invaders, while species that dominate this layer in unproductive ecosystems (in this case, E. hermaphroditum) have minor effects even when they have a comparable biomass. As such, the effect of biodiversity loss on invasibility should depend upon the species or functional groups that are lost, and the ecosystems that they are lost from. Our results in general are consistent with literature predicting greater competitive effects of resident vegetation as soil fertility and ecosystem productivity increases (e.g., Grime 1979; Rajaniemi 2002; Wilson 2007).
Mosses also exerted a context-dependent effect, but only on the survival of planted P. sylvestris seedlings. In contrast to many of the shrub removal effects, moss removals had their greatest positive effects on small islands (Fig. 4). Moss biomass was independent of island size, so this result is not explicable simply in terms of the amount of moss removed. The precise mechanism for this result is unclear. However, mosses can potentially serve as effective competitors and scavengers for soil nitrogen (Oechel & Van Cleve 1986; Zackrisson et al. 1997; Turetsky 2003), and this effect could be particularly important where nitrogen availability is low, such as on the small islands. The physical effects of a dense moss layer, for example through crowding and smothering (Ohlsson and Zackrisson 1992), might also be more harmful for slower growing seedlings on small islands than for more vigorous seedlings on large islands (see Table 2). As such, we would expect P. sylvestris seedlings to show the greatest negative response to mosses on small islands because the seedlings are less vigorous and less able to grow above the moss layer, and to show the greatest negative response to shrubs on larger islands because of greater competition for resources. Furthermore, P. sylvestris seedlings are more likely to show a context dependent response to moss removal than are seedlings of the other two invader species, simply because they grow faster and are therefore likely to outgrow the moss layer on at least a subset of the islands. In any case, our finding that mosses have a greater inhibitory effect on the less productive small islands works in the opposite direction to theories predicting greater competitive effects in more productive ecosystems (Grime 1979; Grace 2001), and is inconsistent with our second hypothesis.
Island size in its own right also had significant effects on seedling emergence, survival and growth. Seedling emergence (seed sowing study) of B. pubescens was greatest on the relatively fertile large islands, especially when ground layer vegetation was absent. This is consistent with our third hypothesis, as well as with previous studies pointing to the responsiveness of Betula seedlings to soil fertility (Dehlin et al. 2004). In contrast, P. sylvestris and P. abies seedling emergence (sowing study) and P. sylvestris seedling survival (planting study) was greatest on small islands even in the absence of ground layer vegetation, which is inconsistent with our third hypothesis. The reason for this is unclear, but coniferous seedlings tolerate higher concentrations of polyphenolics in the soil (such as is the case for small islands; Wardle et al. 1997) better than do seedlings of deciduous species (Nilsson et al. 2000). Further, small islands have much deeper humus layers than do the large islands (Wardle et al. 2003); deep humus serves as a significant store of soil moisture. It is conceivable that greater retention of moisture on small islands during the dry summer period (notably July) could help reduce desiccation of seedlings. In contrast, growth of those P. sylvestris and P. abies seedlings that had survived establishment was greatest on large islands (independent of ground layer vegetation) (Table 2), probably as a result of greater soil fertility and nutrient availability.
Our study has several implications. First, consistent with our first hypothesis, it shows that loss of functional groups and species can enhance the invasibility of plant communities, but that this depends on the identity of the species and functional groups that are lost. The floristic components that were most effective at suppressing invasibility comprised a relatively a small proportion of the total plant biomass on the islands. This is inconsistent with our first hypothesis, as well as predictions of the ‘mass ratio hypothesis’ that the species with the most mass also have the greatest effects on community and ecosystem processes (Grime 1998; Vile et al. 2006). Second, it highlights the importance of environmental context (in this case, island size and therefore soil and vegetation properties) in determining the effects of biodiversity loss on invasibility. Removals of shrubs and of two shrub species often had their greatest effects on larger and more productive islands which is consistent with our second hypothesis, while removals of mosses sometimes had their greatest effect on small islands which is inconsistent with this hypothesis. Further, in the absence of vegetation the invader species only sometimes performed best on large islands, providing only partial support for our third hypothesis. Third, because replicated island ecosystems were used, it is relevant to understanding how the competitiveness of the resident island biota (and the loss of components of this biota) may allow ingress of new species. This is consistent with recognition of the need to incorporate biotic interactions into island biogeography theory (Burns 2007). Finally, it highlights the utility of removal experiments for better understanding the ecological consequences of loss of functional groups and species on community and ecosystem properties. Our results are consistent with a growing body of literature that has used approaches other than randomly assembled planted communities for understanding the effects of biodiversity loss, and that point to interactions between plant functional groups or species, and environmental context, as key drivers of community and ecosystem processes (Wardle & Zackrisson 2005; Emery & Gross 2006; Grace et al. 2007).
Authors thank G. Crutsinger and two anonymous referees for helpful comments on the manuscript, and Vetenskaprådet for funding the study.