Do non-native species invasions lead to biotic homogenization at small scales? The similarity and functional diversity of habitats compared for alien and native components of Mediterranean floras


  • Philip W. Lambdon,

    Corresponding author
    1. NERC Centre for Ecology and Hydrology, Hill of Brathens, Banchory, Aberdeenshire AB31 4BW, UK,
    Search for more papers by this author
    • Present address: Global Programmes Department, Royal Society for the Protection of Birds, The Lodge, Sandy, Bedfordshire SG19 2DL, UK

  • Francisco Lloret,

    1. Centre for Ecological Research and Forestry Applications, Unit of Ecology, Department of Animal and Plant Biology, Universitat Autònoma de Barcelona, E-08193 Bellaterra, Spain,
    Search for more papers by this author
  • Philip E. Hulme

    1. National Center for Advanced Bio-Protection Technologies, PO Box 84, Lincoln University, Canterbury, New Zealand
    Search for more papers by this author

*Correspondence: Philip W. Lambdon, Kew Herbarium, Royal Botanic Gardens Kew, Richmond, Surrey TW9 3AB, UK. E-mail:


Although a number of recent studies have demonstrated biotic homogenization, these have mainly focused on larger spatial scales. Homogenizing effects are equally important at finer resolutions, e.g. through increasing similarity between habitats, which may result in a simplification of ecosystem structure and function. One major cause of homogenization is the expanding ranges of alien species, although it is not clear whether they are inherently homogenizing at smaller scales. We therefore assessed whether the alien flora is less complex across habitats than the resident native flora of Mediterranean Islands. From a regional data base, we examined floristic lists for between-habitat taxonomic and functional similarity, and within-habitat functional diversity, using resampled data sets to control for sample size biases. Aliens and natives showed equivalent complexity in most respects. At the taxonomic level, between-island and between-habitat similarities were almost identical, and when ecosystem function was measured by a functional group classification system, this was also true of between-habitat similarities and within-habitat diversities. When ecosystem function was measured using Grime's CSR classification, aliens were found to be more functionally homogenous between-habitats and less functionally diverse within habitats. However, since the CSR profiles of aliens and natives differed, simplification is not inevitable due to ecological segregation of the two floras (aliens tend to be recruited to disturbed habitats rather than displacing natives). One deficiency is a lack of large scale species abundance data. A simple simulation exercise indicated that this is likely to lead to substantial overestimation of true levels of similarity, although would only influence the comparison between aliens and natives if they have different abundance distribution curves. The results indicate that alien floras are not intrinsically more simple than natives, but a higher proportion of competitive strategists among aliens may still cause small-scale homogenization as these include many strong competitors that are likely to dominate communities.


Over the past few centuries, rapid and often uncontrolled expansions in anthropogenic activities have been associated with the erosion of global biodiversity (Mack et al., 2000; Sala et al., 2000; Penuelas & Boada, 2003; Hobbs et al., 2006). Biotic homogenization, the process of increasing similarity between the floras and faunas of different global regions, is one mechanism by which this occurs. It is driven by two major factors: first, by local extinctions of rarer species through habitat loss, pollution or predation, and second, by the spread of alien species into regions where they are not native and may never have reached without the aid of humans. Often, the latter problem may exacerbate the former if the invaders out-compete or prey on natives, or modify their habitat to render it less suitable for the original occupants (Olden & Poff, 2003). While the two processes need not necessarily lead to homogenization (and may even increase diversity at a local level), a number of studies from various parts of the world have shown that the net result is often an enhanced dominance of widespread species, leading to greater similarity between the taxonomic inventories of disparate regions (Collins et al., 2002; McKinney, 2005; Qian & Ricklefs, 2006; Smart et al., 2006).

Although the idea of biotic homogenization has been in existence for some time (Elton, 1958), the phenomenon has only been a topic of wide discussion in the scientific literature since the turn of the millennium. Despite a rush of recent of papers, most have focused on the species level and relatively large spatial scales, and have yet to address in detail further small-scale facets that are important to ecological health of the environment. Biodiversity is essential in many ways. For example, it increases the resilience of communities to catastrophic events, enhances the buffering capacity of ecosystems against climatic change, and accommodates a wealth of resources that are valued by humans (McGrady-Steed et al., 1997; Naem & Li, 1997; Olden et al., 2005). These benefits arise not only from the variety of species but also from more subtle factors. Within species, genetic diversity is now widely considered to be important (Myers & Knoll, 2001; Olden & Poff, 2004), whereas across species, there is increasing recognition that the number and diversity of ecological interactions may be critical to the productivity and adaptability of ecosystems (Lyons & Schwartz, 2001; Montoya et al., 2006). This ‘ecological complexity’ is difficult to quantify, although it is likely to be equally affected by biotic homogenization processes.

Whereas we cannot attempt to address ecological complexity in its entirety within this study, it is important to start with those aspects of it which can be described and evaluated by basic measures. The processes involved operate both within and between communities:

  • (1) At the within-community level, one measure widely considered to be useful is functional diversity: the variety of ecological roles fulfilled within an ecosystem (Olden et al., 2004). High functional diversity is likely to be a critical parameter in the resilience and resource use efficiency of ecosystems (Tilman et al., 2004), and may also help to increase resistance against invasion by alien species (Levine & D’Antonio, 1999; Pokorny et al., 2005).

  • (2) At the between-community level, we propose that the taxonomic and functional dissimilarity of species assemblages in different habitats is likely to be a meaningful indicator of environmental quality. Important benefits arise from maintaining a diverse and distinctive complement of interacting ecosystems within any local region. Ecosystems interact indirectly, via successional processes and abiotic neighbourhood effects (e.g. protection from erosion, interception and storage of excess rain water, etc.), and maintenance of their specific functions relies on distinctive structural composition. They also interact directly through animal species that move between habitats. The services these provide (e.g. pollination, pest control, creation of disturbance) may be beneficial to each of the communities visited, and their survival may depend on a heterogeneous matrix of resources. Some species exhibit complementary survival advantages in different habitats, so diversity is essential for long-term population maintenance. For example, the bittercress Cardamine cordifolia A. Gray grows faster in full sunlight but finds a refuge from herbivory in shaded woodland (Louda & Rodman, 1996).

A further consideration largely neglected in homogenization/differentiation studies is the need to account for differences in species abundance, rather than simply the presence or absence of taxa. This is an important consideration in population dynamics (Denslow & Hughes, 2004), and is likely to have a large influence on our perception of similarity. Harrelson & Cantino (2006) showed that although the number of alien species in an Ohio state park declined between 1957 and 2003, their total abundance increased greatly during this time. Similarly, in a recent meta-study, Lundholm & Larson (2004) showed that alien species richness did not predict the degree to which aliens had become dominant relative to natives.

In this paper, we examine basic evidence for biotic homogenization among the plant assemblages of Mediterranean islands. As far as we are aware, this is the first such examination of floristic homogenization for the Mediterranean region, despite its high global conservation value. Not only is the native flora very diverse, but there is a high rate of local and regional endemism and many species are already threatened by environmental degradation and climate change (Blondel & Aronson, 1999; Gritti et al., 2006). Alien plant species are spreading rapidly and becoming an increasing threat to Mediterranean seminatural environments (Hulme, 2004; Lambdon & Hulme, 2006). Reductions of ecosystem complexity could make these more vulnerable to further invasions and other environmental problems (e.g. they may have negative impacts on the diverse invertebrate communities which the habitats support). Furthermore, McKinney & la Sorte (2007) contended that ‘invasive’ alien plant species have a particularly strong homogenizing effect – that is, they are more likely to occur across multiple regions than natives. The authors acknowledged that, to some extent, such a finding is quite likely: wide range size is indirectly implied in many definitions of invasiveness. But does the same trend apply to aliens as a whole – i.e. is their contribution to communities generally floristically less complex and more uniform than that of native species?

Although a few recent studies of biotic homogenization have started to address such questions on a functional level (Olden et al., 2004; Smart et al., 2006), none have yet, as far as we are aware, addressed between-community interrelationships. Although they have often examined a range of spatial scales, the majority of studies have compared distinct geographical regions and not considered the diversity between habitats occurring within the same area. We assess evidence for between habitat homogenization or differentiation in the Mediterranean according to a range of criteria: using species checklists, we examine this question in terms of within-habitat diversity, and between-habitat similarity at both the taxonomic (species) level and the community functional level. Unfortunately, for large data sets like that covering the Mediterranean, it is currently almost impossible to obtain the information which would be necessary to evaluate abundance parameters. Very few studies record abundance data for specific habitats, and even where this has been achieved, they are unlikely to include assessments for all native and alien species, or deal with more than a very local area. Even at much coarser levels, distribution maps such as those in Atlas Flora Europaea (Jalas & Suominen, 1972–2004) have yet to be published for many of the region's species. Therefore, we were unable to incorporate abundance adequately in the study, but conducted a simulation exercise using hypothetical abundance data to assess how they might influence the findings. From the results of this exercise, we discuss how the patterns obtained from empirical data are likely to be biased.


Data set

From 94 floras, we compiled a list of 212 non-native higher plant species considered to be fully naturalized (sensu Richardson et al., 2000) on islands of the Mediterranean basin, together with full habitat descriptions. Similar information was also compiled for a representative sample of 1982 fully native species, based on the floras of the Balearic, Maltese, and Cretan island groups. Published literature from these three regions contains reasonably detailed habitat descriptions (Haslam et al., 1977; Beckett, 1993; Turland et al., 1993), and Flora Europaea (Tutin et al., 1964–1980) was also used to supplement them. The three island groups cover a varied representation spanning west to east of the region with differing patterns of land cover and use, climate, and soil types. We omitted species that were considered to be native in only some parts of their Mediterranean range because we were unable to determine accurate statuses for each island group.

The habitat descriptions were scored into 45 categories, based initially on a subset of those defined in the EUNIS classification ( These were merged into 12 broader categories using an oblique cluster analysis (the VARCLUS procedure, SAS/STAT 9.1, SAS Institute inc., 2002), which explained 40% of the total data set variance. Very similar patterns were obtained for both the native and the alien species data sets. Three of the categories contained very few species and were not used in latter analyses. Three of the remaining groups comprised strongly anthropogenic habitats, and the rest were predominantly natural or seminatural (Table 1).

Table 1.  Summary of habitat categories and ecological functional groups used in the analyses.
Category nameDescription
 Ruderal (Anthropogenic)Waste or disturbed ground
 Agricultural (Anthropogenic)Cultivated land
 Urban (Anthropogenic)Towns, around buildings, parks, and gardens
 Woodland (seminatural)Broadleaved, coniferous, sclerophyllous, or mixed woodland
 Grassland (seminatural)Rangeland or improved pasture
 Xeric (seminatural)Dry shrublands (garrigue, phrygana), semisteppe
 Montane (seminatural)Cliffs, gorges, mountain rocks, shrublands, or grasslands
 Wetland (seminatural)Rivers, permanent streams, lakes (+ margins), marshes, temporary pools
 Coastal (seminatural)Dunes, beaches, coastal rocks, or cliffs
Ecological functional groups
 Alpine shrubsWoody species adapted to alpine regions
 Alpine herbsHerbaceous species adapted to alpine regions
 Chasmophytic pteridophytesPteridophytes growing on cliffs or other rock surfaces
 HydrophytesAquatic species normally growing fully submerged
 HelophytesAquatic species normally growing as emergents
 Bulbaceous dicotsDicotyledonous species perennating as bulbils
 Wind-pollinated treesWind-pollinated trees, often dominant in mature forests
 Clonal graminoidsGraminoids spreading extensively by rhizomes or tubers
 Other clonalNon-monoctyledonous species exhibiting rapid clonal propagation
 Halophytic shrubsSalt-tolerant shrubs
 Halophytic herbsSalt-tolerant herbaceous species
 Non-clonal graminoidsGraminoids that do not rely extensively on clonal propagation
 HeterotrophsParasitic, saprophytic, or strongly hemi-parasitic species
 Geophytic monocotsOther monocotylenous species perennating by bulbs or tubers
 Wind-dispersed perennialsWind-dispersed chamaephytes and hemicryptophytes
 Other shrubsShrubs not included in any other category
 Perennials with simple pollinationPerennial herbs pollinated by wind/with radial flowers accessible to most insects
 Perennials with complex pollinationPerennial herbs pollinated by vertebrates, bees, or long-tongued insects
 Shrubs/trees with complex pollinationShrubs or trees pollinated by vertebrates, bees, or long-tongued insects
 Shrubby succulentsNon-halophytic succulent shrubs
 Wind-dispersed therophytesWind-dispersed therophytes, often pioneer colonists
 Non-mobile therophytesTherophytes without specialist dispersal or pollination
 Animal-dispersed therophytesTherophytes with fleshy berries or spiny seeds dispersed by animals
 Other therophytesTherophytes not included in any other category
 Shrubs/trees with fleshy fruitsShrubs and trees with fleshy berries dispersed by animals
 Wind-pollinated perennial herbsOther wind-pollinated chamaephytes and hemicryptophytes

We used two categorization systems to identify functional groupings among the species. First, we defined a series of 28 ecological functional groups (henceforth, the ‘EFG system’) from a range of descriptive traits covering modes of reproduction, dispersal mechanisms, Raunkaier growth forms, and some obvious life-history specializations that were important in qualifying the niche occupied (e.g. alpines and subalpines adapted to montane regions, halophytes to salt tolerance, and chasmophytes adapted to life on cliff faces with little or no soil). These data were obtained from a wide range of literature sources including several of the source floras and some reliable professional websites, such as the Ecological Flora data base ( and the USDA Plants data base ( Presence or absence of each trait was encoded in a series of binary variables.

A cluster analysis (the FASTCLUS procedure in SAS/STAT 9.1; SAS Institute Inc., 2002) provided the initial basis for the groupings, using a hierarchical stepwise method, in which the two groups with the minimum squared Euclidean distance between cluster means were merged at each step. Since the method does not prioritize the more important traits, some categorizations proved to be artificial. Ultimately, the classification system was partially modified by the authors, although it proved to be almost identical in explanatory power (35.8% of the Euclidean variance of the explanatory variable data set), and facilitated more direct interpretation (Table 1). All except one of the groups contained at least 20 native species.

The second form of functional grouping was based on Grime's CSR strategy (Grime, 1974), which defines responses to gradients of competition (C), stress (S), and ruderality (R). Since there has been little work on this model in the Mediterranean, the preliminary characterizations were subjective, based on a combination of the descriptive information available and the personal experiences of the authors (see Lambdon et al., 2008 for further details). Grime et al. (1988) list 20 criteria that can be used to distinguish C, S, and R species (although we never had sufficient data to test more than 11). To these we added two new ones based on dispersal type and habitat preference. The criteria each have values characteristically associated with the C, S, or R strategies. For each species, presence or absence of any characteristic was given a preliminary score of 1 or 0, respectively (the habitat criterion was weighted as 2, as it was deemed to be particularly important), and values of the characters associated with each strategy were averaged independently. The final CSR scores were simplified to a three-point scale: 0 = low (averaging < 0.4 on the preliminary evaluations), 0.5 = moderate (averaging 0.4–0.6), and 1 = strong (averaging > 0.6).

Analysis of similarity and diversity

Of the various similarity indices reviewed by Wolda (1981), we chose to use the association index, which is analogous to Simpson's index used by Kühn & Klotz (2006). The main advantage is that it is expressed as a simple ratio (the proportion of the maximum possible number of comparisons which are shared between two habitats), making it easy to interpret and also easily modified for translation between the taxonomic and the functional levels. The disadvantages are that it is sensitive to sample size and diversity (cf. McKinney, 2004), although these were not major concerns since we used randomization tests to generate our null hypotheses, which provided an adequate control. At the taxonomic level, the association index, S(t), was calculated as:


where X is the status group (A = alien, N = native), ni=1 is the number of species occurring in habitat i, nj=1 is the number of species occurring in habitat j, and ni=1,j=1 is the number of species ‘shared’, i.e. occurring in both habitats i and j. At the functional level, the modified association index, S(f), was calculated as:


where for functional group g, ng(i = 1) is the number of species occurring in habitat i and ng(j = 1) is the number of species occurring in habitat j. S(f) is equivalent to S(t) except that any two species in the same functional group can count as a successful ‘sharing’, with the result that for any given comparison, S(f) ≥ S(t). Since the similarity indices were proportions, all data were arcsine-transformed in order to achieve a normal distribution.

We tested five null hypotheses:

1. At the regional level (i.e. across the entire Mediterranean), between-habitat taxonomic similarities are, on average, the same for alien and native species

To assess this hypothesis, we first calculated S(t)A between each pair of habitats. In order to generate an expected distribution from the S(t)N-values, we conducted 1000 random samplings of the native species data set. In each run, 212 species were selected (the same as the number of aliens). The means of the resampled S(t)N-values conformed to a normal distribution (r2 = 0.997). The significance of the mean S(t)A was tested by solving a general linear model of the form:

S(t)X(i,j) = kX·X + k(i,j)(3)

where kX and k(i,j) are constants. For X = A, there was only one set of observed values, but X = N was replicated 1000 times with run number treated as a random factor. Although sample sizes were therefore highly unequal, this form of analysis allowed the variance due to the (i,j) pairing to be modelled, thus making the overall test slightly more sensitive than in a simple comparison of means. The solution was obtained via maximum-likelihood iteration in the MIXED procedure of SAS/STAT 9.1. Model estimates were obtained from a contrast in which X = N was treated as the baseline category and set to zero. Significance was evaluated using a single-sample t-test, with the error degrees of freedom adjusted to the number of model runs.

2. Between-island taxonomic similarities are, on average, the same for alien and native species

This was a test of geographical floristic similarity, as has been presented for several global regions in other studies. The comparisons were generated between each pair of island groups. Expected distributions were obtained from the native species using the same random sampling technique as for hypothesis 1, but only 107 species were sampled in each run (equal to the number of aliens across the three island groups).

3. Between-island taxonomic similarities are, on average, the same for exotic alien and native species across all habitat categories

The methodology from hypothesis 2 was repeated for the each habitat category separately, with the number of resampled species adjusted as appropriate to each case.

4. At the regional level, between-habitat functional similarities are, on average, the same for alien and native species

We calculated S(f)A between each pair of habitats, basing our functional categories either on the 28 EFG groups or on the 26 possible combinations of the C, S, and R scores (each with levels 0, 0.5, or 1). To examine any differences in more detail, we also compared the numbers of alien and native species in each functional group, testing for significance with χ2 tests. Obtaining an expected distribution from the S(f)N-values was more complicated than for taxonomic similarity due to the grouped structure required. It was achieved by producing 1000 simulated data sets, each containing 212 species, which were generated randomly from observed characteristics of the native data. As a check, we used the same method to generate an expected distribution for natives, using the larger sample size of 1982 species. The means of the simulated data coincided almost exactly with that of the natives.

First, we assigned each species at random to a functional category, and randomly determined its habitat breadth (the number of habitat categories occupied), using the relative probabilities of occurrence among the natives. Mean habitat breadths were found to be approximately equal between the functional categories. Next, we assigned the species at random to the required number of habitats, again based on the observed probability matrix. When summarized into a functional category × habitat frequency table, it was found that our method consistently underestimated the number of species in cells where the frequency was expected to be high and overestimated cells where the frequency was expected to be low. This is because species with narrow habitat breadths are more likely to occur in common habitats than rare ones. Since our overall sample size was too small to accurately estimate a functional category × habitat breadth habitat × probability matrix, we applied a simple correction function of f1.2, to the frequency, f, of each cell, which compensated for the discrepancy very effectively. Finally, mean S(f)N was calculated for each run, and the significance of the mean S(t)A was tested against a normal distribution fitted to the values (r2 > 0.99 for both the CSR and the ecological functional groups).

5.  At the regional level, habitat functional diversities are the same for alien and native species.

The Shannon–Weiner diversity index, D (expressed as a proportion of its maximum value), was used to measure functional diversity within each habitat (Krebs, 1985):


where ng is the number of species in functional category g, X1 = alien or native and X2 = observed or expected. D(f)A was calculated from the observed data, and expected distributions were obtained from the simulated data sets described in hypothesis 4. To estimate differences in the native and alien patterns, we analysed the following general linear model:


This allowed us to identify inline image, which is the component indicating deviation between observed and expected diversity after their joint variation had been taken into account. Note that for natives, the observed and expected values were expected to be identical since they were derived from the same data. Their inclusion merely serves as a baseline, against which we could compare the alien pattern.

Overview of the effect of abundance-weighting on similarity measures

Thus far, similarities have been calculated based on presence or absence in a given habitat. However, if a species is, for example, twice as common in one habitat as another, the difference should be better reflected by allocating a lower similarity score than if the species were equally abundant in both habitats. To accommodate this, Equation 1 can be modified to give an abundance-weighted taxonomic similarity index, S(tw):


where, for species k and habitats i and j, aik and ajk are abundance weightings between 0 and 1 and N is the total number of species that occur in either i or j.

We were unable to incorporate abundance adequately in the study. Instead, we conducted a simulation exercise using hypothetical abundance data, testing the magnitude of influence by repeating the analysis for null hypothesis 1 for different levels of distributional evenness between habitats. The input data set was the same, except that for each habitat and species, we allocated a randomly determined abundance weighting, a, ranging between 0 and 1. The weights were calculated in two steps. First, a relative abundance score, ak was generated for each species, k, according to a simple logarithmic distribution:

ak = −4ln(0.77ρ)(7)

where ρ is a random number between 0 and 1. The scores were standardized so that the mean was 1, as in the original (unweighted) analysis for hypothesis 1. Second, a habitat abundance score, ai was given to each habitat, i, in which the species occurred. This was generated from a reciprocal function:

ai = Er−1(8)

where E is a distributional evenness constant ranging from 0 (= maximum inequality) to 1 (= complete equality) and r is the rank order of the habitat in terms of abundance (1 = the habitat with the highest abundance), which was also allocated to the habitats at random. The final weighting, aik was the product of ai and ak. The two functions used in Equations 7 and 8 are purely hypothetical, and since the results of this analysis are only intended to provide a broad indication of the importance of abundance, their parameterizations do not matter greatly. However, they are intended to approximate the types of curves found in real data scenarios and describe similar relationships to those from Lambdon (2008). For simplicity, we only examined the effect of abundance weighting on native species. At six values of distributional evenness (E = 0.17, 0.33, 0.5, 0.67, 0.83, and 1), we conducted 1000 simulated runs, and following the methodology for hypothesis 1, obtained a relationship between E and taxonomic similarity, given by the mean S(t)N scores.


Taxonomic similarity between habitats

From the observed data, mean alien similarity between habitats, S(t)A, was 0.263. For natives, S(t)N, based on the mean of the resampled data, was 0.262 ± 0.0001 standard error (SE) (reconverted from the normalized arcsine-transformed values). The observed alien value was not significantly different from expectation based on the native resampled distribution (t991 = 0.03, P = 0.9759), and thus, null hypothesis 1 is supported: alien floras are no more or less similar between habitats than natives (Fig. 1). The taxonomic similarity index was moderately influenced by sample size, and without correcting for this the contrast would have appeared slightly stronger. The overall S(t)N, based on the full native species data set was 0.227, which was also not significantly different from the mean of the resampled distribution (t991 = 0.19, P = 0.8503), but indicates that there is a small tendency for similarity to decline in large samples.

Figure 1.

Summary of statistical tests used to evaluate null hypotheses 1, 2, and 4. The series indicate: similarity indices obtained for the alien data set (dark shading); similarity indices obtained by resampling of the native data set, with 95% confidence intervals (pale shading); the observed indices obtained for the full native data set without resampling to control for sample size bias (unshaded). Note that this bias underestimates taxonomic similarities but overestimates functional similarities. *Indicates that the alien and resampled native statistics differed significantly.

Taxonomic similarity between islands

The between-island similarity for aliens was greater than that for natives (S(t)A = 0.492, mean resampled S(t)N = 0.424 ± 0.002 SE). However, as above, these values did not differ significantly (t991 = 1.15, P = 0.2515). Therefore, null hypothesis 2 is supported: island floras are not significantly more likely to share similarities amongst alien than native species (Fig. 1). However, the patterns differed geographically, with the Cretan and Maltese islands being the most similar among the aliens (S(t)A(i,j) = 0.769) and the least similar amongst natives (mean S(t)N(i,j) = 0.316 ± 0.002 SE). These comparisons were both significant (based on 95% confidence limits of the parameter estimates), and indicate a degree of local floristic homogenization.

When between-island similarities were examined for each habitat category individually, no significant differences were obtained (Table 2). Although some moderate differences between S(t)A and S(t)N were suggested in a few cases, these always involved a particularly low number of species and the results cannot be considered reliable. In general, anthropogenic habitats were likely to be more similar than (semi)natural ones for both status groups. The finding supports null hypothesis 3: in all habitats, aliens and natives were equally likely to share species across the three island groups.

Table 2.  Between-island taxonomic similarities for each habitat category. S(t)A is the mean of the three interisland comparisons, whereas S(t)N is the mean (± standard error) of these values across 1000 resampling runs. For montane species, sample sizes were too low to obtain a meaningful t-test. The number of alien species occurring in the three island groups was approximately half of that for the entire Mediterranean region.
HabitatNumber of alien speciesMean S(t)AMean S(t)Nt1,994P
Ruderal700.6140.657 (± 0.004)−0.740.4614
Agricultural570.5690.611 (± 0.005)−0.580.5637
Urban700.6260.715 (± 0.003)−1.550.1221
Woodland190.5020.525 (± 0.008)−0.150.8835
Grassland130.6800.481 (± 0.006) 1.590.1151
Xeric250.4790.735 (± 0.009)−1.290.1968
Montane 50.6210.215 (± 0.018)
Wetland320.5800.589 (± 0.004)−0.100.9173
Coastal360.6000.621 (± 0.005)−0.240.8075

Functional similarity between habitats

When ecosystem function was measured by the EFG classification, null hypothesis 4 was supported (Fig. 1). Functional similarity among aliens was slightly higher than among natives (S(f)A = 0.465, mean S(f)N = 0.458 ± 0.0008 SE), but the alien value did not differ significantly from the expected distribution (t991 = 0.28, P = 0.7769). Therefore, aliens share no greater or lesser functional similarity between habitats than natives. However, without the simulations on which to base our expectation, the result would have appeared considerably different. The overall S(f)N using all 1982 species was 0.548, which is significantly higher than the corrected expectation (t991 = 3.68, P = 0.0002). Due to this much increased sample-size effect, the erroneous conclusion from an uncorrected comparison would have been that native floras are substantially more homogeneous.

Stronger differences emerged when ecosystem function was measured by the CSR classification (Fig. 1). Alien functional similarity (S(f)A = 0.635) was significantly greater than among natives (mean S(f)N = 0.530 ± 0.0008 SE; t991 = 4.30, P < 0.0001). The CSR strategy profiles of the two status groups were also rather different (Fig. 2), with the alien flora dominated by ruderals and competitors and the natives flora shifted strongly towards stress tolerators. This would suggest that aliens could potentially change the composition of invaded communities in addition to increasing the functional homogeneity of neighbouring habitats.

Figure 2.

CSR strategy profiles of the alien and native flora, based on species frequencies. Since many of the 26 strategy groupings possessed very few members, we have aggregated them to the seven fundamental strategies: species with a score of 0.5 or 1 are pooled. Aliens and natives differed significantly for all strategies except C/S/R (χ2 test of independence, P < 0.05).

Functional diversity among habitats

Natives tended to have a lower functional diversity than aliens, although this was consistent with expectation due to the many more species involved. For the EFG categorizations and with all habitats pooled, neither alien or native diversity was significantly different from the expected values (Table 3, Fig. 3). When each habitat was examined individually, D(f)A was found to be significantly lower than expected in four cases and significantly higher than expected in two cases (based on tests of the inline image interaction term, P < 0.05). Reduced diversity occurred in 2 of the three anthropogenic habitats (Fig. 3a). Increased diversity occurred in the Xeric and Montane categories, both (semi)natural habitats with low levels of disturbance and invasibility. Since these were also very species-poor in aliens, the findings could easily be low sample-size artefacts, and should be considered with caution. Consequently, we conclude only that null hypothesis 5 is supported at a general (pooled-habitat) level, but there is a tendency for alien floras to be less functionally diverse than natives in some, predominantly disturbed, habitats.

Table 3.  Analyses to determine whether observed functional diversity deviated from expectation. The comparison is based on t-tests of the inline image term in Equation 5. Diversity indices inline image are also indicated, and standard errors are given in parentheses where appropriate.
Functional categorizationData set (X1)Observed inline imageExpected inline imageinline imaget1989P
  1. CSR, competition, stress and ruderality gradients; EFG, ecological functional group.

EFGAliens0.6360.634 (± 0.025) 0.002 (± 0.026) 0.08 0.9382
Natives0.4330.434 (± 0.026)−0.0004 (± 0.026)−0.02 0.9872
CSRAliens0.4030.482 (± 0.019)−0.080 (± 0.020)−4.04<0.0001
Natives0.3600.333 (± 0.019) 0.029 (± 0.020) 1.44 0.1502
Figure 3.

Alien functional diversity D(f)A, assessed for each habitat. Expected means (± standard errors) were derived from the simulations in which species were allocated to habitats and functional groups according to native frequency characteristics. *Denotes that the observed and expected values differed significantly (test of the inline image interaction from Equation 5, P < 0.05).

For the CSR categorizations, alien functional diversity was significantly lower than expectation (Table 3), and this pattern was repeated in most of the of the individual habitat categories although it was only significant in four cases (Fig. 3b). Again, these included all of the anthropogenic habitats. This evidence refutes null hypothesis 5: the alien flora not only displays greater CSR functional similarity between habitats (null hypothesis 4), but also tends to have a reduced functional diversity within habitats.

The effect of abundance weighting on similarity scores

Weighting for abundance has a large effect on the magnitude of the similarity index, S(t), which decreases exponentially with the reciprocal evenness constant, E (Fig. 4). Thus, if in reality the distribution of abundances is highly skewed, with very few species being dominant in very few habitats, and many species being rare in the majority of habitats where they occur, then the use of presence- or absence-based similarity indices will greatly overestimate the true level of similarity. To some extent, this result is inevitable, because the abundance curves express an inherent level of inequality. Assuming that the example curves in Fig. 4(b) are moderately realistic for abundance distributions within a given habitat, the weighted similarity indices are likely to be much smaller than were evaluated in hypothesis 1. However, the comparison between alien and native curves is likely to be more robust to this anomaly unless their abundance distributions are very different. Although E1 and E2 produce markedly different distributions, the weighted values of S(t) they produce differ by less than 0.02.

Figure 4.

(a) The effect of abundance weighting on the value of the taxonomic similarity index, S(tw). Similarity values are means from 1000 simulation runs (± standard error bars are shown, but these cover a very narrow range). Each run simulated a simple abundance distribution for the Mediterranean native species data set, with the degree of evenness in the abundance of species and habitats determined by the distributional evenness constant, E. (b) Two example ranked abundance distributions of plant species in the Mediterranean. The values of the abundance-corrected similarity index that they produce are shown in Fig. 4(a).


Any simplification of biotas is a cause for concern, and this is particularly true at the habitat level where strong ecological interactions occur. We examined three aspects of the phenomenon whose consequences are closely related (Olden, 2006): (1) Increased taxonomic similarity (the aspect which equates most obviously to the recent homogenization literature), important for potentially reducing local pools of genetic diversity and the range of niches available to other trophic levels within a given landscape (Olden et al., 2004). (2) Increased functional similarity – normally, one would expect habitats to be structured differently in order to reflect the abiotic environment, which in turn should generate further niche diversity, and, due to the creation of between-habitat complementarity, the landscape may support more species and individuals across trophic levels. (3) Decreased functional diversity within habitats potentially leads to losses of niche and structural complexity, therefore reducing carrying capacity of species that require a range of conditions to complete their life cycle, and impairing ecosystem resilience (Holway & Suarez, 2006).

We examined whether alien floras carried any of these indications of simplification in comparison to the native Mediterranean flora – i.e. do aliens carry greater ‘simplifying power’. The aim was merely to highlight whether the scientific community should be concerned about the potential for such a threat and to stimulate further research into any possibilities. At present, there are many fewer naturalized alien plants than natives in the Mediterranean, and their overall contribution to taxonomic diversity is low. However, the number and abundance of such species is increasing rapidly, and greater future problems may be expected. In some oceanic island groups (e.g. the Azores, New Zealand), aliens are already locally dominant in taxonomic terms (Wilson et al., 2000; Silva & Smith, 2004).

Our findings suggested that any initial concerns over the ‘simplifying power’ of aliens are largely unfounded. Taxonomic similarity between habitats, and functional similarity and diversity as measured by our Ecological Functional Group classification, were remarkably equal in aliens and natives. This was also true of between-island taxonomic similarities, both overall and for each habitat category individually. At least in the case of functional similarity, this conclusion would not have been obtained without the use of appropriate randomization tests to control for differences in sample size. Few previous homogenization studies using similarity indices have employed such controls, although they only have a strong affect on functional similarity measures, and should therefore be considered in future studies of this phenomenon.

The series of null results may be somewhat surprising, and personal experiences initially lead us to expect that alien floras would indeed be simpler. Direct observations suggest that a number of prominent alien species indeed tend to occur repeatedly in different localities and environments (Groves, 1986; Sobrino et al., 2002; Pino et al., 2005). Also, heavily invaded habitats tend to be species-poor (Vilàet al., 2006). There remain two major potential sources of bias in the use of large-scale data sets to conduct such studies:

(1) Regional level trends only provide an indication of potential floristic complexity assuming that the species are capable of, and likely to, invade their full habitat breadth and geographical range. Regional lists provide no information about whether the species co-occur together or are separated in different localities. They tend to overestimate small-scale diversities, because if the members are spatially segregated then species-richness will be lower within any given landscape. They also tend to underestimate small-scale similarities, because spatially segregated species occupying different habitats across their range are recorded as ‘false’ sharings. If such a bias is operating, it is likely to lead to an underestimation of native relative to alien complexity because, on average, native species occur more widely than aliens in Mediterranean floristic literature. This pattern is in itself potentially biased due to the traditional emphasis of botanists towards indigenous flora (Pyaturalization and invasiek et al., 2004), although since many aliens are recent introductions and have yet to expand across their full potential ranges, it is likely to be at least partially real. At smaller scales, native plant species are therefore more likely to co-occur than aliens and realize a greater proportion of their regional functional diversity. Furthermore, the geometric mean number of habitats occupied (employed due to a left-skewed distribution) was moderately higher in natives (2.26 +0.26/–0.23 SE) than aliens (1.77 ± 0.05 SE), and since a reasonable proportion of these habitats are indeed island-specific (P. Lambdon, unpublished data), one may expect a greater number of false sharings and inflated functional similarities in natives.

Unfortunately, habitat data are normally compiled at much lower resolution than spatial distribution data, leaving such biases inherent in most habitat-level analyses. None of the three focal floras provided sufficiently detailed descriptions of alien habitats that we could compile workable individual lists. Over the whole data set any imbalance is probably reduced by the fact that the native habitat information was compiled from three island groups only, whereas the alien list was derived from a wider range of Mediterranean literature sources. Since natives tend to occur much more abundantly within islands, the number of populations sampled may have been more comparable than if data for the entire Mediterranean region were available for both species groups.

(2) Due to the almost total lack of suitable data regarding local abundance, we were unable to account for its influence on similarity directly: a locally widespread species will have a stronger effect at simplifying community structure than a rare one, because it exerts a greater contribution to the composition of the habitat per unit area (McKinney & La Sorte, 2007). The simulation exercise to examine the effects of abundance weighting suggests that this effect can substantially alter perceptions of similarity. However, although such distortion is undesirable, the finding is not necessarily very relevant to the aims of this study, which is focused on the comparison between levels of between-habitat homogeneity in alien and native species. Abundance will only affect the conclusions if aliens and natives differ in their abundance curves (particularly that described by Equation 7). Although we are unable to cast very much light on whether this is the case, it highlights the fact that approaches to field surveys would benefit greatly from more sophisticated levels of data capture in the future – a goal that is rapidly becoming more attainable thanks to the availability of remote sensing and computerized field recording equipment.

One exception to the general pattern of null trends was that when species were classified according to the CSR functional group classification, aliens showed lower functional diversity and higher between-habitat functional similarities than natives. While this may constitute a ‘threat’, it requires more careful consideration than would be necessary in the other parameters investigated. The Ecological Functional Group categories describe biological function, and one may expect most of them to be represented in most habitats as integral components of a healthy ecosystem. The CSR strategies describe niche affiliation, and are likely to be distributed unevenly (Thompson et al., 1995; Maskell et al., 2006). The greater similarity shown by aliens indicates that they tend to be more restricted to certain parts of the niche space, perhaps leaving the native communities in other microhabitats relatively unaffected. The findings could thus even lead to a net diversification rather than a simplification, increasing the total number of species coexisting in a given region.

Evidence from Fig. 2 indicates that there is indeed a considerable degree of niche partitioning between the native and the alien floras. Natives have a much higher frequency of stress tolerators, which are very successful in the Mediterranean region's widespread arid shrublands (Pons, 1981). As yet, habitats dominated by such species have been invaded by very few aliens (e.g. the Xeric and Montane categories, Table 2), and it may be that the difference in CSR profiles is due to a high failure rate of stress tolerators rather than a preponderance of aggressive weedy species. However, the greater tendency towards competitive strategists in aliens is a potential cause for concern. Competitors are more likely to be community dominants, which tend to invade wide areas and suppress other species. These are most likely to cause small-scale loss of functional and taxonomic diversity, to invade and homogenize multiple habitats, and may ultimately spread rapidly to become widely distributed at a regional level (Prach & Pysek, 1999; Rooney et al., 2004). Furthermore, competitors (and hence alien species in general) are more likely to have higher local abundances, and therefore the abundance-weighted similarity measures are indeed likely to be inflated, suggesting that the previous conclusions have underestimated their homogenizing effect to some degree. Devictor et al. (2007) found that generalist bird species are becoming more common in France at the expense of specialists. They suggested that this has led to increased functional homogeneity of avian communities due to loss of niche distinctiveness in much the same way that we propose for plant competitive strategists.

Ruderal species were particularly common in both the alien and the native floras, and anthropogenic habitats, which are frequently associated with ruderality, were the most homogeneous between islands. These trends are consistent with the growing emphasis on urbanization as a major driver of biotic homogenization (McKinney, 2002; Kühn & Klotz, 2006; McKinney, 2006; Olden et al., 2006; Devictor et al., 2007). Urban land cover is expanding rapidly in the Mediterranean, and one of its major impacts is the destruction or modification of often highly diverse natural areas. Loss of small-scale functional diversity is an additional problem to be considered in this debate.

Ultimately, local extinctions play a major role in driving ecological simplification processes, and the full picture cannot be adequately assessed without taking this into account (Olden & Poff, 2003). The mixing of alien and native floras does not inevitably lead to the resultant communities taking on the characteristics of either component, and many complex interactions determine the resilience of functional structure. Davis (2003) suggests that, globally, extinctions through competition between aliens and natives are extremely rare, but they may be much more important to the composition of local assemblages. If there is indeed a tendency for the most successful aliens to be aggressive dominants, then such small-scale deterioration could increase.


The large-scale floristic homogenization reported widely in other papers is only part of a spectrum of potential negative effects of alien species on native community structure which could be characterized as ecological simplification. In general, this study was able to detect little intrinsically different about the structure of alien floras which suggest impacts at the regional habitat level, but we cannot rule out the possibility of smaller-scale bias or that very small-scale effects (e.g. through the expansion of invasive community dominants) may reveal a different pattern. A higher proportion of competitive strategists suggests that to some extent, this may indeed be the case – alien species may, on average, exhibit higher local abundances than natives, and therefore impose greater ecological uniformity per unit area, even if they do not display greater habitat or niche breadths. While hopefully a stimulant to further investigation, more detailed data are needed in order to make an adequate assessment of the true picture, and the importance of ecological simplification as a major environmental concern.


Thanks to Guiseppe Brundu, Frederic Médail, Anna Travaset, Andreas Trombis, Montse Vilà, Luca Viegi, and their associated research teams for supplying data, and to Louise Ross and Marie Pandolfo for contributing substantially to the data collation. Thanks also to Marten Winter and three anonymous reviewers for comments and discussion, and David Elston for statistical advice. This study was conducted as part of the EU Framework 6 project ALARM (GOCE-CT-2003-506675), and the data base assembled during the Framework 5 project EPIDEMIE (EVK2-CT-2000-00074).