Explaining invasiveness from the extent of native range: new insights from plant atlases and herbarium specimens

Abstract

Aim

We tested the relationship between the extent of the native range and the success (number of occurrences) in the introduced range of European vascular plant species naturalized in the province of Québec (Canada). We hypothesized that the performance of models linking native range size and species invasiveness can be improved if residence time and climate tolerance are taken into account.

Methods

The extent of the native range (Europe, Asia) was estimated using plant atlases. The number of occurrences in the introduced range (Québec) was estimated using the number of herbarium specimens stored in herbaria. Herbarium specimens were also used to obtain residence time. Plant hardiness was used as an indicator of the suitability of a species to the climate of the introduced range. Multiple linear regression models, corrected to take into account phylogenetic biases, were used to calculate correlations between the extent of the native range and the number of occurrences in the introduced range.

Results

The larger the native distribution area in Eurasia, the greater the number of occurrences (herbarium specimens) in Québec. The shorter the residence time and the less hardy the plant, the fewer the number of occurrences. In all models tested, the phylogenetic structure explained a significant proportion of the variance, but its influence decreased as the number of species or area studied (Europe versus Eurasia) increased.

Main conclusions

The extent of the native range is a good explanatory variable for the invasion success of vascular plants, especially once other factors (residence time, climate tolerance, phylogeny) are taken into account. Thus, a model using these variables could be used by environmental managers to flag species warranting further investigation. With the emergence of online databases, gathering the required information is becoming easier and cheaper. As online databases continue to improve and new analytical tools are developed, this approach will become even more powerful.

Introduction

Several highly sophisticated models have recently been proposed to explain or predict the invasive potential of exotic plant species (see Pyšek & Richardson, 2007 and van Kleunen et al., 2010 for reviews). The performance of these models is sometimes impressive, especially those constructed using life history traits, although there is debate as to whether these traits are really associated with an exotic status or whether invasive species, regardless of their status, are simply better adapted to human-disturbed environments than other species (van Kleunen et al., 2011; Thompson & Davis, 2011). However, the success of these models is often based on traits for which data are rarely available for a high number of species, or on large datasets requiring a huge collection effort. These problems preclude widespread application of the trait-based models, especially in developing countries with limited botanical expertise. Decision-making tools requiring relatively little botanical expertise, such as the Weed risk assessment, may represent an alternative and can perform well in identifying weed species, but can also have a high rate of false positives (McClay et al., 2010; Hulme, 2012).

Shah et al. (2012) recently proposed that simpler models, based on the native range size of a species, could be used as low-cost early warning tools for the management of invasive plants. Species with a large native range are likely to become invasive because (1) some traits allow a species to have a large range, whether native or exotic, (2) broad native distribution is reflective of wide environmental tolerance, which is often correlated with invasiveness, and (3) wide-ranging species are more likely to be dispersed through different propagule transportation vectors (Pyšek et al., 2004, 2009; Hui et al., 2011; Dawson et al., 2012; Knapp & Kühn, 2012; Shah et al., 2012). Life history traits would only play an indirect role in invasion success by determining the size of the native range (Pyšek et al., 2009). To our knowledge, this hypothesis – not new but rarely tested (see Pyšek et al., 2004) – is essentially supported by some statistical models (Goodwin et al., 1999; Gravuer et al., 2008; Pyšek et al., 2009; Hui et al., 2011; Proches et al., 2012; see Bucharova & Kleunen, 2009; for other models). However, we hypothesized that the performance of models linking native range and invasiveness can be significantly improved if residence time in the exotic range is taken into account: a newcomer, whatever the extent of its native range, will not be as ‘successful’ as a species that has been present for centuries (Castro et al., 2005; Wilson et al., 2007; Gravuer et al., 2008). For instance, exotic plants require on average 150 years to reach their maximum range in Europe (Gassó et al., 2010). We also hypothesized that the climatic tolerance of plants is an important factor to consider: if a species is not adapted to the climatic conditions of the introduced area, it will not be able to spread over a wide range, regardless of the extent of its native range (Gravuer et al., 2008).

In this study, we tested the relationship between the extent of the native range and the ‘success’ (or number of occurrences) in the introduced range for 456 European vascular plant species naturalized in the province of Québec (Canada). The extent of the native range was estimated using plant atlases. The number of occurrences in the introduced range was estimated using the number of herbarium specimens stored in herbaria. Herbarium specimens were also used to obtain the residence time. Plant hardiness (cold tolerance) was used as an indicator of the suitability of a species to the climate of the introduced range. Models were corrected to take into account phylogenetic biases (Felsenstein, 1985). Notwithstanding the biases associated with our data sources, we show here that the number of occurrences of plant species in their introduced range may be explained using these sources.

Methods

Selected species

We used the recent checklist of exotic vascular plants naturalized in Québec (about 900 species; see Lavoie et al., 2012; for details) to test the relationship between the extent of the native range and the number of occurrences in the introduced range. Only species naturalized in Québec with a Eurasian native range, and with at least part of their native range located in Europe (west of Ural Mountains, including Turkey), were considered. Some of the selected species were also present in northern Africa along the Mediterranean coast, but they were nevertheless retained for analysis because their native range was largely Eurasian. Furthermore, only plant species introduced and naturalized in Québec after 1860 were considered, because the naturalization dates of plants introduced in the province before the mid-19th century are too imprecise (Lavoie et al., 2012). Range and residence time considerations thus restricted our analysis to 456 species.

Extent of the native range

The area occupied by species in Europe was first reconstructed using the Atlas Flora Europaeae, which provided relatively precise mapping (presence/absence) of many European vascular plants in 50 × 50 km grid cells (Finnish Museum of Natural History, 2012). The Atlas Flora Europaeae also specified whether mapped species were native or introduced to each cell. The number of cells (CELL EUROPE) was counted for each species; cells indicating that the species was introduced in a specific part of the European range were discarded. Numbers, electronically calculated, were provided by the Finnish Museum of Natural History (A. Sennikov, personal communication).

The Atlas Flora Europaeae is an ongoing project: to date, only 138 of the 456 European species naturalized in Québec and analysed in this study have been mapped. To expand our analysis, we used the information provided by the Euro+Med PlantBase (for 80% of species; Euro+Med PlantBase, 2012) and by the Flora Europaea database (for 20% of species; Royal Botanic Garden of Edinburgh, 2012). Both databases list the countries – or regions for the specific case of the former USSR – in which a species is present, specifying whether the species is native or introduced in a specific part of the European range. The Euro+Med PlantBase is an expanded and updated version of the Flora Europaea database and was thus our primary information source. However, not all European species are listed in this database; in these cases, the Flora Europaea database was used. For each species, we summed the respective area (km²) of the countries (or regions) where the species was native. The resulting total area (AREA EUROPE) was used as a surrogate for precise mapping.

We hypothesized that the area occupied by a species in Europe was a good indicator of the total area of the species in Eurasia. This hypothesis was tested by estimating the total area occupied by a species in Eurasia for all species of the database (n = 293) that were also mapped in the Atlas of North European vascular plants (Hultén & Fries, 1986). Maps were scanned and incorporated into a geographical information system (ArcGIS; ESRI, 2012). The area of the zone corresponding to the range where a species was ‘common or fairly common’ (sensu Hultén & Fries, 1986) was delineated and then calculated using the ArcMap tool in ArcGIS (ESRI, 2012). This provided an additional variable (AREA EURASIA) that was used to test the hypothesis of a relationship between the extent of the native range and the ‘success’ (or number of occurrences) in the introduced range.

Number of occurrences in the introduced range

No plant atlas was available for Québec (the introduced range). To estimate the ‘success’ of introduced species, we used the number of herbarium specimens (SPECIMEN) stored in the two main herbaria of Québec, MT and QFA (see New York Botanical Garden, 2012; for herbarium codes), as a surrogate measure of the number of occurrences of exotic vascular plants naturalized in the province. MT and QFA harbour about 80% of the 1,800,000 vascular plant specimens stored in Québec herbaria (New York Botanical Garden, 2012). To what extent the number of herbarium specimens in Québec is an indicator of population size or of the area occupied by a species remains to be substantiated; it is likely a combination of both. Several studies have shown that the number of herbarium specimens is a good indicator of the size of a plant population in the field (MacDougall et al., 1998; Vetaas, 2000; Puyravaud et al., 2003; Wu et al., 2005; Phillips et al., 2011), although common species and rare species are usually under- or over-represented in herbaria, respectively (Garcillán et al., 2008; Garcillán & Ezcurra, 2011). The collections of MT and QFA are not computerized, so the number of specimens for each species was manually counted (total: 35,574 specimens). Herbarium specimens from the six largest herbaria containing Québec specimens (CAN, DAO, MT, MTMG, QFA and QUE) were also checked to obtain the oldest proof (collection year of specimen, or YEAR) of naturalization of the species in the province (see Lavoie et al., 2012, for details).

Climate matching

Plant hardiness (HARDINESS) was used as an indicator of the suitability of a species to the climate of the introduced range. Gardeners and plant growers regularly use plant hardiness zone maps to identify the plants that are more likely to thrive at a particular location. Plant hardiness maps are based on the average annual minimum winter temperature, divided into 10 °F (5.5 °C) zones (United States Department of Agriculture, 2012a). For example, a plant classified as Zone 3 is theoretically able to resist winter temperatures down to those corresponding to this zone (−40 to −30 °F; −40 to −34 °C). Québec has five plant hardiness zones, from 1 (−60 to −50 °F; −51 to −46 °C) to 5 (−20 to −10 °F; −29 to −23 °C). We hypothesized that in Québec, two species with a similar area (native range) but with a different plant hardiness will not have the same ‘success’ (or SPECIMEN): a species classified as Zone 1 will perform better than a species classified as Zone 5, because it is better adapted to the climatic conditions of the whole province.

Collecting plant hardiness data for a high number of species is a difficult task, and the information provided by American and European plant growers (catalogues) is not very reliable for Québec, according to local professional horticulturists (Fédération interdisciplinaire de l'horticulture ornementale du Québec, personal communication). Because all the species we used in our model were naturalized in Québec, they were at least classified as Zone 5. To determine whether they could be classified as other zones (1–4), we examined the geographical distribution of the species in Europe (Euro+Med PlantBase, Flora Europaea) and in North America (PLANTS; United States Department of Agriculture, 2012b), and we compared this distribution to the plant hardiness zone maps available for Europe (BackyardGardener, 2012), Canada (Agriculture and Agri-Food Canada, 2012) and the United States (United States Department of Agriculture, 2012a). If part of the geographical distribution corresponded to a plant hardiness zone colder than 5, this plant hardiness zone was attributed to the species.

Phylogenetic and statistical analyses

To verify whether our results were taxonomically or phylogenetically biased, we constructed the phylogeny of the three sets of species that were generated for this study (i.e. with CELL EUROPE data, with AREA EUROPE data and with AREA EURASIA data) using the online tool Phylomatic (Webb & Donoghue, 2005). The topology of the trees generated by Phylomatic was in accordance with the APG III classification system (Angiosperm Phylogeny Group, 2009). The accuracy of the trees was increased by adjusting the length of the branches. Age of the nodes was calibrated with data from Wikström et al. (2001), and branch lengths were adjusted using the bladj tool in Phylocom (Webb et al., 2008). Trees obtained were then used to calculate three matrices of phylogenetic distances between species. The matrices were analysed using a principal coordinates analysis with the Cailliez correction (Gower & Legendre, 1986) in order to obtain three phylogenetic structures that can be used in linear regression models (Desdevises et al., 2003). The principal coordinates analyses were used to extract orthogonal eigenvectors representing the phylogenetic component associated with each set of species. The R software (R Development Core Team, 2010) was used for calculations.

We evaluated whether AREA EUROPE was a good surrogate for precise mapping. We assumed that the data from the Atlas Flora Europaeae (CELL EUROPE) provided the most reliable picture of the distribution of the plants in Europe, and we calculated the Pearson correlation coefficient between CELL EUROPE and AREA EUROPE for the 138 species for which both sets of data were available. We also evaluated whether AREA EUROPE was a good predictor of the total area occupied by a species in Eurasia by calculating the Pearson correlation coefficient between AREA EUROPE and AREA EURASIA for the 293 species for which both sets of data were available. The significance of each coefficient was verified with a two-tailed test.

Three other different models were built, one with the species with CELL EUROPE data, one with the species with AREA EUROPE data and the other with the species with AREA EURASIA data. In each model, SPECIMEN data (the explained variable) were log-transformed for normalization, and all explanatory variables were standardized (μ = 0; σ = 1) to allow the comparison of regression coefficients between models. The distribution of residuals from each model indicated that OLS (ordinary least-squares) assumptions of linearity, normality and homoskedasticity were met. Also, VIF (variance inflation factors; Sokal & Rohlf, 1995) were below two for all explanatory variables in the various models, thus indicating negligible multicollinearity.

The first step in the modelling was to identify phylogenetic eigenvectors that had a significant influence on SPECIMEN data. Because all eigenvectors produced were orthogonal to each other, it was possible to identify significant (α < 0.05) eigenvectors on the basis of their simple correlation coefficient with SPECIMEN. The second step was to introduce in different multiple regression models (1) the significant phylogenetic eigenvectors and the biogeographical variables (CELL EUROPE or AREA EUROPE or AREA EURASIA + HARDINESS + YEAR) or (2) only the phylogenetic eigenvectors or (3) only the biogeographical variables, to test their respective influence on SPECIMEN. The possibility of significant interaction terms among the biogeographical variables was tested using a backward elimination procedure to select the variables (Tabachnick & Fidell, 2007). This yielded statistically non-significant results for the interaction terms in the most parsimonious models. The last step was to use the adjusted R square resulting from these analyses to partition the variance into four distinct parts associated with (1) phylogeny and biogeography, (2) phylogeny only, (3) biogeography only and (4) other unknown factors, respectively (Desdevises et al., 2003). The SPSS software (SPSS Inc, 2004) was used for calculations.

Results

In Europe, the total area of the countries (or regions) where a plant species is native can be used for estimating the extent of the native range: the Pearson correlation coefficient between CELL EUROPE and AREA EUROPE was highly significant (P < 0.001), with a value of 0.796 (Fig. 1). AREA EUROPE is also a good predictor of the total area occupied by a species in Eurasia: the correlation coefficient between AREA EUROPE and AREA EURASIA (0.550) was highly significant (P < 0.001). However, the correlation between CELL EUROPE and AREA EURASIA (0.692; P < 0.001) was better, which can easily be explained considering the approximate nature of AREA EUROPE data. In all cases, linear models produced the best results.

Figure 1. Relationship between different estimations of the range occupied by European vascular plant species in Europe or Eurasia. CELL EUROPE: area (Europe only) estimated from the Atlas Flora Europaeae (number of grid cells 50 × 50 km; Finnish Museum of Natural History, 2012); AREA EUROPE: area (Europe only) estimated from the Euro+Med PlantBase (2012) and the Flora Europaea database (Royal Botanic Garden of Edinburgh, 2012); AREA EURASIA: area (Eurasia) estimated from the Atlas of North European vascular plants (Hultén & Fries, 1986).

In the set of species with CELL EUROPE, AREA EUROPE and AREA EURASIA data, 6, 18 and 9 phylogenetic eigenvectors had a significant influence on SPECIMEN data, respectively. The multiple linear regression models incorporating these vectors and biogeographical variables were all highly significant (P < 0.001), with R²_a values ranging from 0.399 to 0.491 (Table 1). All variables were significant, apart from one (CELL EUROPE) and six (AREA EUROPE) of the phylogenetic eigenvectors. In all models, the larger the native area in Europe, the greater the number of occurrences (herbarium specimens) in Québec. On the other hand, the later the naturalization and the less hardy the plant (hardiness zone with a higher number), the fewer herbarium specimens collected in Québec. The regression weights (β values) of YEAR and HARDINESS suggest that these factors were as important as CELL EUROPE as explanatory variables. Variance partitioning indicated an influence of phylogeny on results, but this influence remained below that of biogeographical variables (Fig. 2).

Figure 2. Proportion of the variance in the number of herbarium specimens of Eurasian vascular plant species that were collected in Québec (Canada) and that is explained by different sets of explanatory variables. Three different multiple linear regression models were tested, each with a particular set of species according to the information source for the extent of the native range (CELL EUROPE, AREA EUROPE, AREA EURASIA; see text for details). Explanatory variables were grouped in two categories, that is, (1) associated with biogeography (CELL EUROPE or AREA EUROPE or AREA EURASIA + YEAR + HARDINESS; see text for details) and (2) associated with phylogeny (phylogenetic structure of each set of species). The b+p component is the proportion of the variance explained by biogeographical and phylogenetic variables.

Table 1. Multiple linear regression models calculated between the extent of the native range (CELL or AREA: explanatory variables) and the number of occurrences (herbarium specimens) in the introduced range (SPECIMEN: explained variable) for European vascular plants introduced and naturalized in the province of Québec, Canada, after 1860. Models were tested with the oldest proof (year) of naturalization of the plants (YEAR), plant hardiness (HARDINESS) and phylogenetic eigenvectors as covariables. Only scores associated with CELL, AREA, YEAR and HARDINESS are shown

Model and explanatory variable	β	t	P
Explaining SPECIMEN with CELL EUROPE(n = 138;  = 0.399; P < 0.001)
CELL EUROPE	0.198	2.736	0.007
YEAR	−0.290	−3.909	< 0.001
HARDINESS	−0.192	−2.820	0.006
Explaining SPECIMEN with AREA EUROPE (n = 456;  = 0.484; P < 0.001)
AREA EUROPE	0.208	5.798	< 0.001
YEAR	−0.350	−9.531	< 0.001
HARDINESS	−0.234	−6.400	< 0.001
Explaining SPECIMEN with AREA EURASIA (n = 293;  = 0.491; P < 0.001)
AREA EURASIA	0.159	3.475	0.001
YEAR	−0.408	−8.749	< 0.001
HARDINESS	−0.215	−4.721	< 0.001

Discussion

Explaining the number of occurrences of an exotic vascular plant from the extent of its native range is possible, although the results should be nuanced. The strength of relationships ( value) of the models that included the area of the native range, the residence time and the plant hardiness as explanatory variables is high, despite the fact that all the information sources we used (plant atlases, herbarium specimens) suffer from biases. Biases are especially of concern for herbarium specimens, which are not necessarily sampled in proportion with the size of the plant population in the field (Garcillán et al., 2008; Garcillán & Ezcurra, 2011), nor with a constant collecting effort over time (Prather et al., 2004; Rich, 2006; Hofmann et al., 2007; Lavoie et al., 2012). Plant atlases also have several spatial and temporal sampling biases (Robertson et al., 2010). Notwithstanding all these biases, the fact that the values are high and that all models are consistent suggests that the link between the extent of the native range and the number of occurrences in the introduced range is real, and not a statistical fluke.

In all models tested, the phylogenetic structure explained a significant proportion of the variance, but its influence (the phylogeny and b+p components of Fig. 2) decreased as number of species or area studied (Europe versus Eurasia) increased. This is probably because species from a larger dataset and a larger area are less likely to be taxonomically linked. This supports the use of the Eurasian range to construct models linking the size of the native range with invasiveness, although European data are good surrogates for the total range of Eurasian species.

Our models suggest that residence time and plant hardiness should be taken into account in studies investigating the link between the extent of the native range and the invasion ‘success’ of plants. Models that do not incorporate these variables could potentially come to misleading conclusions (see Castro et al., 2005; Wilson et al., 2007; and Gravuer et al., 2008; for other examples). Of course, we assumed that the climatic niche requirements of exotic species were conserved between their native and invaded ranges. Recent large-scale tests of niche conservatism for terrestrial plant invaders between Eurasia, North America and Australia indicated that substantial niche shifts are very rare, providing support for the use of climate niche models in explaining biological invasions (Petitpierre et al., 2012).

Other studies have specifically investigated the statistical link between the extent of the native range and the invasion ‘success’ of plants. For instance, Goodwin et al. (1999), also using data from Flora Europaea and multiple regression analyses, linked the size of the European range with invasiveness (more precisely, successful naturalization in human-disturbed habitats) for 240 exotic vascular plant species established in New Brunswick (Canada). Using regression trees and a set of 1218 species, Pyšek et al. (2009) explained 45% of the variability in the success (occurrence in floras) of exotic plants by using plant distribution and climate tolerance as explanatory variables. Three other models focussed on a specific group of plants. Using principal component or cluster analyses as statistical tools, Gravuer et al. (2008) and Hui et al. (2011) found significant relationships between the extent of the native range and plant invasion success for clovers (Trifolium spp.) and acacias (Acacia spp.), respectively. Rapidly spreading clover species are not only characterized by a large native range, but also by early naturalization dates and a climatic tolerance that matches the climate of the introduced range – precisely what we found for plants introduced in Québec. Acacias with large native ranges are also more likely to become invasive, but this link is only strong at the early stages of invasion, that is, introduction and naturalization. Acacias' distributional characteristics are apparently not useful for predicting invasiveness, once the plants are naturalized. Proches et al. (2012), using generalized linear or general additive models, found a significant correlation between indigenous and exotic range size for pine species (Pinus spp.), but other factors, such as the importance of the species to forestry, were better able to explain the variation in the extent of the exotic range.

All these studies (including the present one), and some others (see Bucharova & Kleunen, 2009, for additional information), point towards native range as a good explanatory variable for the invasion success of vascular plants, once other confounding factors such as residence time, climate tolerance and phylogeny are taken into account. In a decision-making process, residence time is irrelevant for species that have not yet been introduced. Thus, a model using two variables, extent of the native range and plant hardiness, could be used by environmental managers to flag species warranting further investigation. With the emergence of online databases, gathering the needed information is becoming easier and cheaper. As online databases continue to improve and new analytical tools are developed, this approach will only become more powerful.

Acknowledgements

This research was financially supported by the Natural Sciences and Engineering Research Council of Canada, Université Laval, Université Paul Sabatier and Institut Hydro-Québec environnement, développement et société (grants to Claude Lavoie). The research fellowship awarded to Manzoor Shah by the Department of Foreign Affairs and International Trade, Canada, under the Canadian Commonwealth Exchange Programme (Asia-Pacific), is acknowledged. We are also grateful to Stefano Biondo, Noémie Blanchette-Forget, Elisabeth Groeneveld and Geneviève Guay for assistance in data collection, to Alexander Sennikov for providing data from Atlas Flora Europaeae and to Mark van Kleunen and three anonymous reviewers for thorough comments on an earlier draft.

Biosketches

Claude Lavoie is biologist and Professor of environmental management at Université Laval. He specializes in exotic plant species, particularly the historical reconstruction of the spread of aquatic invasive species and of plant invaders using rivers and roads as dispersal corridors. He commonly uses herbarium specimens in his historical studies and has developed methods to account for biases associated with herbarium data.

Manzoor A. Shah is a plant ecologist, working as Senior Assistant Professor in the Department of Botany at the University of Kashmir. Plant invasions are his primary research focus. He specifically studies the biogeographical basis of plant invasions, the dispersal patterns of invasive plants in terrestrial and aquatic ecosystems and the molecular ecology of invasive species, including plant and soil microbe interactions.

Alexandre Bergeron is a plant ecologist and Ph.D. student at Université de Montréal. He specializes in the use of spatial quantitative analyses to better understand biodiversity patterns, especially in urban environments.

Paul Villeneuve is a geographer and Emeritus Professor of land management at Université Laval. He specializes in the application of quantitative analyses to better understand the spatial patterns associated with social and environmental problems.

Author contributions: C.L. and M.A.S. involved in data collecting and preparing the concept for paper; C.L. first drafted the manuscript; A.B. and P.V. involved in phylogenetic and statistical analyses. All authors contributed to the writing.