rationale of plot selection
Field work was done between 2004 and 2007. For each of the 13 invasive neophytes (Table 1), the impact of invasion was studied in various regions of the Czech Republic (See Appendix S1 in Supporting Information). For each species, 10 pairs of adjacent 4 × 4-m vegetation plots were sampled. The plots were chosen so as to cover a range of site conditions and vegetation types in which the invader achieves dominance in the invaded communities (Appendix S1). In each vegetation type, one plot of the pair was placed in heavily invaded vegetation (‘invaded plots’) where the invader was dominant and had a high cover (Table 1) and the second plot in a neighbouring vegetation, where the invader had no cover (‘uninvaded plots’). The uninvaded plot was chosen so as to have as similar site conditions as possible to the invaded plot, which could have been assumed with reasonable certainty due to the uninvaded plot being located in close proximity to the invaded plot. In a few cases, very low amounts of the invader occurred in the uninvaded plot, which could not have induced any changes to vegetation structure and species composition. In total, 260 vegetation plots were sampled in regions ranging from lowland alluvial meadows to sub-alpine meadows at the highest altitudes of the Czech Republic (Appendix S1).
Table 1. The invasive alien species studied, their life form (rp – rhizomatous perennial, mp – monocarpic perennial, a – annual; all species are dicots), region of origin (NA – North America, A – Asia, E – Europe) and impact on community characteristics. The range of covers of the invading species in invaded plots and the number of species (mean ± SD, n = 10) in invaded (S inv) and uninvaded plots is shown (S uninv). At the plot scale, the impact on species richness S is expressed as the mean percentage reduction of species number in invaded plots compared to uninvaded (100%). Positive value indicates a higher species number in uninvaded, negative in invaded vegetation. At the larger scale, the impact is expressed as the percentage reduction of the total number of species recorded in invaded (Stot inv) plots and related to that recorded in uninvaded plots (Stot uninv = 100%). Mean Sørensen similarity index, calculated as an average value for 10 pairs of plots, indicates the impact of invasion on species composition; the lower the similarity the less similar is the invaded and uninvaded vegetation. Species are ranked according to the decreasing reduction in S. Significant differences in species richness S between invaded and uninvaded plots, corrected for multiple comparisons using the Bonferroni method, are shown: *P < 0.05, **P < 0.01 and ***P < 0.001
|Species||Life form||Origin||Cover range (%)||S uninv†||S inv||Impact on S (%)||Stot uninv||Stot inv||Impact on Stot (%)||Sørensen similarity|
|Fallopia sachalinensis (F. Schmidt) Ronse Decraene||rp||A|| 70–100||13.3 ± 4.9||1.8 ± 1.6||86.4***||70||10||85.7||0.17|
|F. japonica (Houtt.) Ronse Decraene ||rp||A||100||12.1 ± 3.5||3.3 ± 2.8||73.0**||76||21||72.4||0.23|
|F. × bohemica (Chrtek & Chrtková) J. P. Bailey||rp||A|| 40–100||14.8 ± 7.3||5.4 ± 5.0||65.9*||75||37||50.7||0.36|
|Heracleum mantegazzianum Sommier & Levier||mp||A|| 90–100||16.7 ± 4.5||7.4 ± 3.1||52.6**||91||40||56.0||0.33|
|Rumex alpinus L.||rp||E|| 75–100||12.6 ± 2.5||7.7 ± 2.4||39.1***||51||34||33.3||0.42|
|Aster novi-belgii L. agg.||rp||NA|| 60–90||14.1 ± 4.8||8.9 ± 6.3||38.7||80||54||32.5||0.34|
|Helianthus tuberosus L.||rp||NA|| 50–100||12.7 ± 6.5||8.0 ± 4.9||33.7||57||39||31.6||0.59|
|Rudbeckia laciniata L.||rp||NA|| 80–100||10.6 ± 2.6||6.9 ± 3.0||29.8||45||34||24.4||0.60|
|Solidago gigantea Aiton||rp||NA|| 70–100||16.4 ± 6.7||12.0 ± 6.3||25.5||92||62||32.6||0.51|
|Imperatoria ostruthium L.||rp||E|| 50–80||14.3 ± 5.6||9.9 ± 2.6||21.4||61||39||36.1||0.63|
|Lupinus polyphyllus Lindl.||rp||NA|| 60–95||21.1 ± 2.3||16.4 ± 3.8||21.2||93||71||23.7||0.62|
|Impatiens glandulifera Royle||a||AS|| 60–90||10.9 ± 1.8||9.5 ± 2.6||12.3||49||46|| 6.1||0.75|
|Mimulus guttatus DC||rp||NA|| 30–40||17.2 ± 7.4||17.1 ± 7.6||−6.3||93||90|| 3.2||0.61|
sampling and measuring impact
In each plot, all species of vascular plants were recorded and their covers (%) estimated. Species covers were used as importance values for calculating the Shannon diversity index H′ and evenness J. Evenness was calculated as H′/ln S, where S is the species richness expressed as the number of species (Magurran 1983). Differences in species richness S, Shannon index H′ and evenness J between invaded and uninvaded plots were used to measure the effect of invasion on these community characteristics. To assess the impact of invasion on species composition and relative covers of resident species, we calculated the Sørensen index of similarity between each plot pair, based on species covers (Chao et al. 2005). The invading neophyte was excluded from the calculation of community characteristics (Hejda & Pyšek 2006) and so were species in the shrub and tree layers, which were only rarely present.
The species richness S was taken as a measure of diversity at the plot scale. In addition, for each invasive species studied, the total numbers of species recorded in all plots with invaded and uninvaded vegetation (Stot) were used as a measure of the impact of invasion on diversity at the landscape scale.
To assess the effects of population characteristics of the invading species on species richness at the plot scale, the invader's height (cm) and cover (%) were measured in each invaded plot. To compare the absolute performance of the invader with its relative performance in comparison with a dominant native species, height and cover of the dominant native species were also measured and differences in both characteristics between the invading and native dominant species calculated.
Differences in species richness S were tested by paired t-tests of invaded and uninvaded plots on square rooted data (e.g. Sokal & Rohlf 1995, pp. 352–356), using the correction for multiple comparisons based on the Bonferroni method, where the achieved significance levels are multiplied by the number of inferences to obtain conservative tests that eliminate type I error inflation (Daalgard 2002, p. 116). The correlation between reduction in species richness S and Sørensen index of similarity was tested by Pearson's correlation.
Differences in impacts among the invasive species were assessed by one-way anovas. Differences in species richness S, Shannon's diversity H′and evenness J between each pair of invaded and uninvaded plot were the response variables, and the individual invasive species a factor. Differences among the influence of the species were then compared by a posteriori multiple comparisons among means using SNK tests (Underwood 1997, pp. 234–242).
Population characteristics determining the impact were first evaluated by ancovas. Response variables were the differences within pairs in species richness S, Shannon's diversity H′ and evenness J between uninvaded and invaded plots. The explanatory variables were (i) individual invasive species as a factor, and (ii) height and cover of the invasive species and (iii) differences in height and cover between the invasive and native dominant species as covariates. The modelling of ancovas started with fitting maximal models that included the interaction of each covariate with each species and all one-level interactions among the covariates. The aim of the analyses was to determine the minimal adequate model (MAM), in which all explanatory variables are significantly different from zero and from one another, and all non-significant explanatory variables are removed (e.g. Crawley 1993). This was achieved by a step-wise process of model simplification, beginning with the maximal model and then proceeding by eliminating non-significant terms (using deletion tests), and retaining significant terms (e.g. Pyšek et al. 2005).
Invaders’ heights were ln-transformed, their proportional covers angular-transformed (e.g. Sokal & Rohlf 1995) and all covariates standardized to zero mean and unit variance to achieve their comparable influence. Using these standardized values, collinearity was checked by a matrix of correlation coefficients, and then by calculating tolerance values (Quinn & Keough 2002, p. 128). All fitted models were checked by plotting standardized residuals against fitted values, and by normal probability plots (e.g. Crawley 1993). Their explained variance was expressed both as R2 based on sum of squares, and as . based on mean squares, following Quinn & Keough (2002, p. 139). In the latter case, larger values of indicated better fits taking into account sample sizes and number of predictors.
To provide understandable and generally interpretable results of the interactions between explanatory variables, regression trees (Breiman et al. 1984, De’ath & Fabricius 2000, Chytrýet al. 2008) were constructed by repeatedly splitting the response variables using binary recursive partitioning in CART® v. 6.0 (Steinberg & Colla 1995). To find the best tree, a sequence of nested trees of decreasing size, each being the best of all trees of its size, was grown, and their re-substitution relative errors, corresponding to residual sums of squares, were estimated. Tenfold cross-validation was used to obtain estimates of cross-validated relative errors of these trees. These estimates were then plotted against tree size, and the minimum cost tree was selected as the best tree (Steinberg & Colla 1995). Following De’ath & Fabricius (2000), a series of 50 cross-validations was run, and the modal (most likely) single tree chosen for description. Total variance explained by the best single tree was calculated as R2 = 1 − re-substitution relative error. The best trees were represented graphically, with the root Node 1 standing for undivided data at the top, and the terminal nodes, describing the homogeneous groups of data, at the bottom of the hierarchy. The quality of each split was assessed by improvement, corresponding to the proportion of the total sum of squares explained by the tree at each node. To reduce the splitting power of the high categorical variable species (13 factor levels, corresponding to the individual species), the species were adjusted to have no inherent advantage over continuous variables, following penalization rules of Steinberg & Colla (1995).
To reveal net effects of invaders, independent of species identity, the effect of individual invasive species was removed from the analyses. This was done by refitting the individual invasive species in the anovas, and calculating Pearson's standardized residuals of these models (Hastie & Pregibon 1993, p. 205). Residuals from these models were then examined as the response variables (Lonsdale 1999, Chytrýet al. 2008) by the step-wise backward procedures aimed to determine the MAMs, beginning with the maximal model which contained all possible interactions among the explanatory variables. Significant interactions of the differences between the cover of invading and dominant native species with the cover of invading species were then examined by regressing simple slopes on increasing differences between the cover of invading and dominant native species at varying values of the cover of the invading species: its mean, and mean plus and minus its sample standard deviation (Quinn & Keough 2002, pp. 131–133). Analysis of these interactions was made using centred variables (Quinn & Keough 2002, p. 131).