Ken Thompson, Department of Animal and Plant Sciences, The University, Sheffield S10 2TN, UK (tel. 0114 2224314; fax 0114 2220015; e-mail email@example.com).
1A long-term experiment was established in 1990 in which seeds of 54 native species, not originally present at the site, were sown into a fertility × disturbance matrix established in unproductive limestone grassland at the Buxton Climate Change Impacts Laboratory (BCCIL). The objective was to examine the roles of productivity and disturbance as major factors controlling the invasibility of plant communities, and to identify the functional characteristics of successful invaders in response to different types of invasion opportunity. The results of the first 2 years of the study have already been published.
2After 2 years, invasion was strongly promoted by disturbance and less so by increased fertility. Three years later the cover of invaders had declined over most of the matrix, and the greatest cover of sown species was where the highest levels of fertility and disturbance coincided. However, no part of the fertility–disturbance matrix was immune to invasion and the area of the matrix occupied by each of the sown species that successfully established was unique. Abundance of invaders was reduced by low soil pH.
3The identity and distribution of the successful invaders changed as the early stages of invasion gave way to a later stage of consolidation. After 2 years regenerative traits (seed mass and germination characteristics) were the best predictors of success. After 5 years these traits were unrelated to success of the invaders, the most successful invaders were perennial grasses, and no single trait was a good predictor of invasiveness.
4Our results are consistent with the hypothesis that invasions are promoted by an increase in the availability of resources, either through addition of extra resources or a reduction in their use by the resident vegetation.
The past 10–15 years have seen an unprecedented interest in biological invasions, partly deriving from an increasing realization that, against a background of increasing human population and global climate change, invasive species are likely to become one of the key environmental problems of the coming century. Much of this work has centred on attempts to identify the key ecological characteristics of invasive species and potentially invasible ecosystems (Crawley 1986, 1987; Usher 1988; Roy 1990; Lodge 1993; Williamson 1996; Levine & D’Antonio 1999; Lonsdale 1999). However, almost all this effort has been expended on observations of invasions in natural systems (e.g. case studies in de Waal et al. 1994; Pyšek et al. 1995; Brock et al. 1997). There have been few attempts to study invasion experimentally, by manipulating either the characteristics of the invaded community or the identity of potential invaders (Peart & Foin 1985; Robinson et al. 1995; Bastl et al. 1997; Tilman 1997; Crawley et al. 1999; Knops et al. 1999; Levine 2000). Still more rarely have such experimental manipulations been conducted for relatively long periods of time. Experimental studies, including the one reported here, have focused on grasslands and other herbaceous communities, since the time scales involved in woody plant communities are generally too long. Although important insights have begun to emerge from the analysis of the large body of accumulated observational data (Lonsdale 1999; Levine & D’Antonio 1999), important questions remain unanswered and perhaps unanswerable by such an approach. It seems likely that further progress will come from experimental manipulation of disturbance, fertility, climatic variables, herbivory and other factors that might determine the outcome of invasions.
In September 1990, we established an experiment with the objective of examining the roles of productivity and disturbance as major controlling factors of the invasibility of plant communities, and to identify the functional characteristics of successful invaders in response to different types of invasion opportunity. The results of the first 2 years of this study have already been published (Burke & Grime 1996); here we report on the changes that occurred in the next 3 years. The experiment continues.
The site and methods are described in detail in Burke & Grime (1996). Briefly, the experiment was located at the Buxton Climate Change Impacts Laboratory (BCCIL), near Buxton, Derbyshire, UK. The vegetation was unproductive species-rich grassland, previously sheep pasture but now derelict. For more on the site, see http://www.shef.ac.uk/~nuocpe/bccil/.
In September 1990, 36 2 × 2 m plots were marked out within the established vegetation, in a 6 by 6 square with 1-m paths between plots. Within each plot, orthogonally opposed five-step gradients of soil fertility and physical disturbance resulted in a matrix within each plot of 25 subplot combinations of productivity and disturbance, each 30 × 30 cm. The plot matrix was randomly allocated to one of eight possible orientations. The central 1.5 × 1.5 m area was marked as the vegetation to be monitored, surrounded by an unmonitored 25-cm ‘buffer’ zone that received the same treatment as the adjacent subplot. Samples of the upper 5 cm of soil were collected from every subplot in October 1992 only and the pH determined, after mixing with deionized water, with a Kent EIL 7055 pH meter calibrated with buffers at pH 4 and 7.
Agricultural grade granular 20 : 10 : 10 NPK fertiliser was applied at 0, 60, 120, 180 and 240 kg ha−1 in October 1990, April 1991 and October 1991, and subsequently in every March or April. The highest rate therefore corresponds to 48 kg N, 10.5 kg P and 20 kg K ha−1, or approximately one-third of the average rate applied to British agricultural grasslands in 1999 (Anonymous 2000). Two methods were chosen to create the disturbance gradient: cutting gaps in the turf, and mowing the vegetation at different heights, although only the 18 plots from the gap-cutting treatment are considered here. Circular gaps were created within the turf and the soil disturbed to a depth of approximately 50 mm. On the first occasion, all shoot and root material was removed from the gaps, but material was left in situ on subsequent occasions. Two gap sizes were used; a ‘small’ gap that measures 49 mm in diameter (area 1885 mm2), and a ‘large’ gap, six times the area of the small gap measuring 119 mm in diameter (area 11 122 mm2). These gaps mimic the small natural disturbances caused by the hooves of grazing animals, or the slightly larger gaps created by mole hills, that this grassland would normally experience. It would not normally experience any larger-scale disturbance.
The disturbance gradient was achieved by increasing gap density within the subplots, from zero to 48 small (small gap treatment), 8 large (large gap treatment) or 24 small and 4 large gaps (mixed gaps treatment) (six plots per treatment). In each case, the highest gap treatment had the potential to create 100% bare ground, but as gaps were allowed to overlap, the maximum bare ground created was usually rather less than this. Levels of bare ground across the disturbance gradient were identical in small, large and mixed gap treatments. Gaps were placed randomly within each subplot, and were allowed to overlap each other and the borders of the subplot. All gap treatments were applied for the first time in October 1990 and have been repeated in October in every subsequent year.
The advantages of the matrix approach described above are: (i) the elimination of edge effects (even the edge of the matrix is bounded by a border receiving identical treatment), as any treatment combination is adjacent to similar treatments; (ii) the continuous gradients allow plants to expand quickly from their point of establishment to the part of the matrix most conducive to their survival, thus reducing the chance of random establishment failure determining the outcome; and (iii) the opportunity to create a wide range of environments in a very compact space. A disadvantage is that disturbance and fertilizer treatments are not statistically independent. For further discussion see Campbell & Grime (1992).
APPLICATION OF THE SEED MIXTURE
A standard inoculum of 54 species, none originally present at the site, was sown into all plots in October 1990. All species were common, widespread native British species found in the Sheffield area, in habitats with climatic and altitudinal characteristics similar to those of the experimental site. The choice included a broad range of ecological types with respect to both regenerative and established stages of the life cycle. Large seeds were applied at lower densities than small ones. Nomenclature follows Stace (1997).
ANNUAL CENSUS OF THE PLOTS
Plots were censused by placing a 1.5 × 1.5 m quadrat over the centre of the plot. This quadrat was subdivided into a 5 × 5 grid to mark the positions of the 25 subplots. At the first census in 1991, recording was limited to presence of all species within each subplot, regardless of where it was rooted. In subsequent censuses, in 1992, 1993 and 1995, the percentage cover of each species within each subplot was recorded, using the following scale: absent, 0; 1–5%, 1; 5–25%, 2; 25–50%, 3; 50–75%, 4; 76 + %, 5. All censuses took place at the time of maximum plant biomass, in July, August or September.
Since the disturbance and fertility gradients are not independent, their effects cannot be analysed statistically. Therefore we present maps of the distribution of individual species and groups of species for illustrative purposes only. However, visual assessment confirmed that these distributions were highly consistent across replicates. At the whole-plot level, we analysed the total abundance of sown and resident species in the three gap-size treatments by one-way analysis of variance, with mean soil pH as a covariate. Summed sown and resident abundances were suitable for analysis without transformation.
In examining the relationship between invasion success and plant traits, values of some plant traits were log or arcsine transformed as necessary before analysis to reduce heterogeneity of variances.
Total cover of invaders showed a significant plot effect. Although the fertile disturbed corner was the most highly invaded in every case, plots varied a lot in the absolute level of invasion. Analysis of variance of the 18 gap-disturbed plots, at the whole plot level and with pH as a covariate, showed that invader abundance was not dependent on gap size (F2,12 = 1.5, NS), but was very strongly influenced by pH (F1,12 = 26.9, P < 0.001). The gap size–pH interaction was also not significant (F2,12 = 1.5, NS). Soil pH obviously limited invasion at a very early stage, as the pH effect was well established during the first season of the experiment (Fig. 1). Total resident species abundance in 1995 was not significantly affected by gap size (F2,12 = 0.05, NS), pH (F1,12 = 3.0, NS) or their interaction (F2,12 = 0.09, NS).
Although biomass was not measured, visual inspection of the plots revealed that the treatments caused pronounced gradients of biomass, from very high in the undisturbed, fertilized corner to very low in the disturbed, unfertilized corner. As gap size did not significantly affect total abundance of sown or resident species, we show mean abundance of species groups (and later, individual species) across all 18 gap-disturbed plots (Figs 2 and 3). Total cover of both introduced (in 1992 and 1995) and resident (in 1995) species was strongly influenced by the experimental treatments (Fig. 2). In 1992 invasion was strongly promoted by disturbance, with fertility having relatively little effect, particularly at low intensities of disturbance (Burke & Grime 1996; Fig. 2a). By 1995, disturbance and fertility appeared to have equally large positive effects on cover of sown species (Fig.2b). Cover of resident species, on the other hand, was lowest at high fertility (Fig. 2c).
The overall pattern in Fig. 2(b) conceals considerable variation between individual sown species. The diversity of patterns of invasion of the sown species is illustrated in Fig. 3. Some species, generally of rather fertile habitats, appeared to respond positively to both increased fertility and disturbance, including the most abundant invader, Holcus lanatus. Others responded positively to disturbance (e.g. Cynosurus cristatus) but appeared to be indifferent to fertility, while others (e.g. Arrhenatherum elatius) responded only to fertility. A few sown species responded negatively to disturbance, fertility or both (e.g. Vicia cracca and Brachypodium pinnatum). Finally Taraxacum officinale appeared to be relatively indifferent to both treatments and was present throughout the matrix, with a broad peak under conditions of low to moderate disturbance and fertility. Note the similarity of these distributions to those in Buckland et al. (2001). The strong association between total cover of invaders and high levels of fertility and disturbance (Fig. 2b) was the net effect of these individual distributions.
TRAITS OF SUCCESSFUL INVADERS
The dominant indigenous species in 1992 remained the most abundant in 1995. In contrast, ranking of introduced species in 1992 was a poor predictor of ranking in 1995 (r2 = 0.4). If the criterion of success of an invader is an ability not only to persist, but also to increase in cover over a period of years, then most of the introduced species were failures, at least in the medium term. Over most of the matrix, the abundance of introduced species declined between 1992 and 1995 (Figs 1 and 2), and only eight species increased in mean cover during this period. All were perennial grasses, most notably Dactylis glomerata and Alopecurus pratensis.
Correlation of change in cover of introduced species between 1992 and 1995, across all treatments, with a range of plant traits involving seed size, current status, life history, ecological strategy, habitat, germination, morphology and growth, showed no significant relationships (Table 1). One of the most striking results of the analysis of the first 2 years of the experiment was clear evidence of the role of large seed size in facilitating invasion of dense turf, i.e. in the fertile, undisturbed quadrant (Burke & Grime 1996). Ability to increase in abundance under these conditions between 1992 and 1995 was no longer related to seed size (r2 = 0.018, P = 0.27).
Table 1. Pearson correlation coefficients between change in cover of introduced species between 1992 and 1995 and selected plant traits
A measure of proximity to the S (stress-tolerant) corner of the CSR strategic triangle, calculated according to the protocol in Hodgson et al. (1999).
A dominance index derived from extensive vegetation surveys, calculated as described in Hodgson et al. (1999).
A measure of the range of habitats utilized by a species (high values = high specialization), calculated as described in Thompson et al. (1998).
An index of palatability to the generalist mollusc herbivore Helix aspersa. Data from Grime et al. (1996). None of these correlations is significant at the Bonferroni-adjusted P-value of 0.05/18 = 0.003.
One of the principal obstacles to increased understanding of invasions is that success is influenced by three factors: the number of propagules (propagule pressure); the characteristics of the invading species; and the susceptibility of the environment to invasion (Crawley et al. 1996; Williamson & Fitter 1996; Williamson 1996; Lonsdale 1999). In this case, all species were sown in large numbers on a single occasion, and therefore propagule pressure was uniform across all treatments. We can also discount differences in the identities of the invaders available to different parts of the matrix, as all parts of every matrix were sown with the same wide taxonomic and functional range of species. The observed pattern of invasion is therefore a direct quantification of susceptibility to invasion (invasibility). Furthermore, many post hoc investigations of natural invasions are hampered by lack of knowledge of the identities of the unsuccessful invaders, i.e. species that were introduced but never became established, and thus remained undetected. Here we had complete knowledge of all the sown species, and so we were able to attempt to relate invasion success to plant traits.
There are those who would argue that invasions by exotic species are a distinct ecological phenomenon, quite separate from the normal processes of regeneration, colonization and succession by native species (Elton 1958; Dukes & Mooney 1999). However, detailed study of the traits of exotic invaders and native colonizers in NW Europe showed that the two groups were essentially indistinguishable (Thompson et al. 1995). We would therefore argue that the processes that facilitate invasion by exotic plant species and colonizations by native species are fundamentally the same. This view has also been adopted by other workers (Huston 1994; Levine & D’Antonio 1999; Davis et al. 2000), including many of those who have undertaken experimental studies of invasions (Robinson et al. 1995; Tilman 1997; Crawley et al. 1999; Knops et al. 1999).
PLANT COMMUNITY INVASIBILITY
By 1992, at the time of the earlier analysis (Burke & Grime 1996), most individuals of the introduced species were still juveniles. Their distribution was still predominantly a function of the response of the initial inoculum to the opportunities for establishment afforded by the experimental treatments. Not surprisingly, the distribution of invaders was mainly dependent on disturbance, i.e. the creation of bare ground and debilitation of the pre-existing vegetation (Fig. 2a), and success or failure were largely explicable in terms of regenerative traits, e.g. seed size and germination characteristics (Burke & Grime 1996).
Over the following 3 years, the focus of invasion shifted towards the more fertile parts of the matrix and disturbance alone appeared to become less important (Fig. 2b). This seems to mark a transition from the early stages of invasion (seed germination and seedling establishment) to a later stage of consolidation. As many of the species successful in this second stage were robust, perennial grasses of fertile habitats, it is not surprising that high fertility became more important. The increasing importance of high fertility is probably an amalgamation of two effects. First, the time required for large perennial invaders to accumulate a large biomass. Secondly, the sustained accumulation of mineral nutrients in the initially infertile soil. The early finding that invasion was promoted more by small than by large gaps (Burke & Grime 1996) was no longer supported after 5 years.
Overall, the results are consistent with the hypothesis, proposed by Davis et al. (2000), that invasibility is correlated with the availability of unused resources. Both disturbance and fertilizer addition increase the availability of resources, and invasibility was clearly greatest where both were combined. Bastl et al. (1997) found that, among a variety of treatments applied to an infertile wet meadow, only large additions of manure promoted the successful establishment of the invasive annual Impatiens glandulifera. Nevertheless, invasibility is predictable only in a statistical sense, as although increased fertility and disturbance promoted invasion overall, no part of the matrix was immune to invasion, and a minority of species (e.g. Vicia cracca and Brachypodium pinnatum, Fig. 3) did not conform to the general trend.
Although the experimental plots were established on limestone, the high rainfall and low temperatures have led to surface leaching and thus the pH of surface soil was variable and sometimes very low. This had a major impact on invasion, with low pH clearly reducing the overall level of invasion (Fig. 1). The effect of pH is explicable in terms of the known tolerances of the potential invaders; only one of the sown species (Deschampsia flexuosa) was a plant of acidic soils, and the great majority had pH optima ≥ 6 (Grime et al. 1988). Increased soil acidity is often employed in habitat restoration to inhibit invasion by weeds of neutral soils (e.g. Owen & Marrs 2000). pH clearly had a major impact on invasion at an early stage; later, the impact appeared to decrease (Fig. 1). This decrease could be due to the pH being altered by fertiliser addition, or by the later stages of invasion being less sensitive to pH than seedling establishment. As pH was measured only once, near the beginning of the experiment, it is not possible to distinguish between these possibilities.
FUNCTIONAL CHARACTERISTICS OF INVADING SPECIES
Although the factors promoting invasibility were clear (Fig. 2), our results confirm the well-known difficulty of predicting the identity of successful invaders (Crawley 1986, 1987; Crawley et al. 1996; Williamson & Fitter 1996; Williamson 1996), despite notable successes in some taxa, e.g. Pinus (Rejmánek & Richardson 1996). Among all the sown species, no single trait was significantly associated with success (Table 1). Perhaps this is not too surprising, given the variety of opportunities offered by the experimental matrix and the individualistic way in which those opportunities were exploited by different species (Fig. 3). No doubt the distribution of the less-competitive invaders was strongly influenced by the more-competitive species. Of all the traits examined in Table 1, we suggest that palatability is worthy of further investigation, in the light of other evidence that herbivory is important in preventing invasion by palatable species (e.g. Urtica dioica) at this site (Fraser 1998).
RELATIONSHIP TO OTHER WORK
The recent review by Lonsdale (1999) illustrates the difficulty of drawing unambiguous conclusions from the study of natural invasions. Successful invasions are influenced by the interaction between propagule availability, ecosystem invasibility and the characteristics of the potential invaders. Usually all these are known incompletely, if at all. The results reported here cannot be compared with most of the generalizations discovered by Lonsdale that concern invasion at the large scale, i.e. the effects of latitude, biome and islands vs. mainland. Lonsdale and others (Planty-Tabacchi et al. 1996; Stohlgren et al. 1999) found that species-rich communities were more invaded, but our plots were all similar in this respect, at least initially. If the treatments induced changes in the distribution and diversity of indigenous species in different parts of the matrix, the impacts of these changes on potential invaders were inseparable from the effects of the treatments themselves. It is also possible that the success of some invaders was inhibited by the growth of other, more competitive invaders, but again such effects cannot be distinguished from direct treatment effects.
Other experimental studies of invasion are few and have tended to focus on the role of species richness of the invaded community (Peart & Foin 1985; Robinson et al. 1995; Tilman 1997; Levine 2000). The results of such studies have been contradictory; Peart & Foin (1985) and Robinson et al. (1995) added seeds of five grass species and Eschscholzia californica, respectively, to Californian grasslands and found in each case that the most diverse patches were most invasible. The least diverse plots at Cedar Creek are the most invasible (Tilman 1997; Knops et al. 1999), but Crawley et al. (1999) have attributed a similar result to the ‘sampling effect’. The most likely explanation for such contradictions is that invaders are responding to environmental variables that often covary with species richness, rather than to species richness itself (Levine & D’Antonio 1999; Levine 2000). At variance with our results, and consistent with the hypothesis that the identity of successful invaders depends strongly on the invaded environment, Tilman (1997) found that perennial grasses were the least successful invaders in the strongly nitrogen-limited system at Cedar Creek. Here N-fixing legumes were the most successful invaders.
This work was supported by the Natural Environment Research Council. Thanks to past and present members of UCPE, Peter Mitchell for statistical advice, Jerry Tallowin for fertiliser data and to Mark Williamson and several referees for much helpful advice.