Biological Invasions

, Volume 12, Issue 7, pp 2105–2116

Diversity–invasibility relationships across multiple scales in disturbed forest understoreys

Authors

    • Department of BiologyYork University
    • Department of Plant SciencesUniversity of Cambridge
  • Dawn R. Bazely
    • Department of BiologyYork University
  • Raffaele Lafortezza
    • Department of Plant SciencesUniversity of Cambridge
    • Department of Scienze delle Produzioni VegetaliUniversity of Bari
Original Paper

DOI: 10.1007/s10530-009-9612-3

Cite this article as:
Tanentzap, A.J., Bazely, D.R. & Lafortezza, R. Biol Invasions (2010) 12: 2105. doi:10.1007/s10530-009-9612-3

Abstract

Non-native plant species richness may be either negatively or positively correlated with native species due to differences in resource availability, propagule pressure or the scale of vegetation sampling. We investigated the relationships between these factors and both native and non-native plant species at 12 mainland and island forested sites in southeastern Ontario, Canada. In general, the presence of non-native species was limited: <20% of all species at a site were non-native and non-native species cover was <4% m−2 at 11 of the 12 sites. Non-native species were always positively correlated with native species, regardless of spatial scale and whether islands were sampled. Additionally, islands had a greater abundance of non-native species. Non-native species richness across mainland sites was significantly negatively correlated with mean shape index, a measure of the ratio of forest edge to area, and positively correlated with the mean distance to the nearest forest patch. Other factors associated with disturbance and propagule pressure in northeastern North America forests, including human land use, white-tailed deer populations, understorey light, and soil nitrogen, did not explain non-native richness nor cover better than the null models. Our results suggest that management strategies for controlling non-native plant invasions should aim to reduce the propagule pressure associated with human activities, and maximize the connectivity of forest habitats to benefit more poorly dispersed native species.

Keywords

Biotic resistanceDiversity–invasibility paradoxFragmentationInvasive non-indigenous species

Introduction

Land managers tasked with managing for conservation and biodiversity are increasingly called upon to address the issue of invasive non-native species, and their potential impacts on native species. However, the ongoing debate about the general pattern of the relationship between native and non-native species richness makes developing general management approaches problematic. Native species richness may either be negatively or positively correlated with non-native species richness because of differences in niche exploitation, resource availability, competitive abilities, and/or propagule pressure (Fridley et al. 2007). These contrasting trends may also be due to scale effects (Levine and D’Antonio 1999; Fridley et al. 2007). Emerging evidence suggests that patterns of invasion are dependent on spatial scale, both theoretically (Shea and Chesson 2002; Byers and Noonburg 2003; Fridley et al. 2004) and empirically (Levine 2000; Brown and Peet 2003; Knight and Reich 2005). Species diversity may also vary along productivity gradients that determine whether the relationships between native and non-native species are either positive or negative, via coexistence or competitive exclusion mechanisms, respectively (Davies et al. 2007).

Shea and Chesson’s (2002) theoretical framework for examining correlations between native and non-native species richness can be applied at multiple scales. At the micro-scale, physical space and competitive exclusion are likely to restrict the number of individuals able to survive and colonize (Levine 2000; Davies et al. 2005). Non-native species will be unable to become established because existing native species occupy space, and native and non-native species richness should be negatively correlated (Shea and Chesson 2002). However, this relationship may be reversed at low productivity sites that are more environmentally heterogeneous, and thus, provide a greater range of available niches (Davies et al. 2005, 2007). The shared responses of species to environmental factors such as resources may also overwhelm relatively weaker competitive exclusion at low productivity sites, resulting in positive relationships between native and non-native species (Davies et al. 2007). Across micro-sites (at the community-scale), environmental heterogeneity, which may function through processes such as gap creation, should provide opportunities for the establishment of invasive species, and overwhelm inter-specific competition that excludes species at smaller scales (Davies et al. 2005). The increased propensity for disturbance at larger scales should also contribute to a positive relationship between diversity and invasibility at the landscape level (across communities).

Comparisons between island and mainland sites at the landscape level can provide insight into the factors implicated in species invasions. In addition to environmental heterogeneity, dispersal barriers, such as isolation, that act equally on native and non-native species can lead to positive relationships between species at the landscape level (Levine 2000). Islands that occur at a range of distances from the mainland can provide a natural experiment for comparing the roles of propagule pressure versus environmental heterogeneity (Long et al. 2009), particularly since islands that vary in their degree of isolation may have large gradients in native species richness (MacArthur and Wilson 1963). In general, islands may be more susceptible to invasion because isolation limits potential native colonizing species (Herben 2005), and this suggests that the dynamics of regional species pools may overwhelm local-level determinants of plant richness in determining patterns of non-native species (Freestone and Harrison 2006). Propagule pressure has been regularly treated as equal and/or more important than resource availability in determining community invasibility (Levine 2000; Lockwood et al. 2005; Tanentzap and Bazely 2009), and relatively high historical rates of species introductions on islands provide a contrast to mainland sites (Lonsdale 1999). Islands may also be prone to invasion because of greater resource and niche availability for invading species arising from less competitive native flora and a lack of natural enemies (Lonsdale 1999).

Two common disturbances in northeastern North American forests that are positively associated with invasive non-native species are: (1) human land use patterns (Pyšek 1998; McKinney 2001, 2002); and (2) high populations of white-tailed deer (Odocoileus virginianus) (Koh et al. 1996; Baiser et al. 2008). Forest patches persisting in highly fragmented, human-modified landscapes appear particularly susceptible to invasion by non-native species (Borgmann and Rodewald 2005; Duguay et al. 2007; McDonald et al. 2008), and as these patches decrease in size, a greater ratio of forest edge to area may enhance their invasibility (Cadenasso and Pickett 2001; With 2002). Increases in open-canopy habitats, due to fragmentation, may promote seed production, and subsequently, a high propagule pressure, from many invasive non-native species (Boutin and Jobin 1998; Charbonneau and Fahrig 2004; McDonald and Urban 2006). Recent experimental evidence suggests that the establishment of non-native species within forest interiors is determined more by propagule pressure from adjacent human land uses than by resource availability (Tanentzap and Bazely 2009). This finding is of particular concern, because heavily disturbed lands subject to intensive human use are increasingly situated adjacent to protected areas (Pauchard et al. 2003; Pauchard and Alaback 2004). Non-native species are also commonly introduced via horticulture, agriculture and other activities associated with human settlement (Mack and Lonsdale 2001; Reichard and White 2001). Additionally, irruptive populations of herbivores, such as white-tailed deer, which are themselves, a direct consequence of human-mediated landscape changes (Côté et al. 2004), can disperse non-native plants (Myers et al. 2004).

In this study, we examined patterns in native and non-native plant species richness in relation to various indicators of disturbance and propagule pressure across the landscape of southeastern Ontario, Canada, including at four island sites. The natural habitat cover across much of this landscape is highly fragmented (Riley and Mohr 1994), and we included sites located in a variety of human land-use types with varying levels of disturbances, such as deer herbivory. In order to test the relative importance of the roles of both resource availability and propagule pressure in determining the establishment of non-native species across our sites, we asked the following questions:
  1. 1.

    What is the correlation between native and non-native plant species richness and the percent vegetation cover of these two groups across southeastern Ontario?

     
  2. 2.

    Do diversity–invasibility relationships differ between island and mainland sites?

     
  3. 3.

    Do human land use (specifically, fragmentation patterns), deer density, and environmental variables (photosynthetically active radiation and soil nitrogen concentrations) explain patterns of non-native species occurrence at the landscape level?

     

Methods

Study sites

Vegetation surveys were conducted at 12 forest locations in the Great Lakes-St. Lawrence Forest Region of southern Ontario (Fig. 1). The typical canopy vegetation was comprised of mixed coniferous and deciduous trees, including Acer saccharum, Ostrya virginiana, Fraxinus americana, Pinus strobus, Thuja occidentalis, and Tsuga canadensis. Understorey vegetation was dominated by spring ephemerals, e.g. Trillium grandiflorum, Claytonia virginica, Maianthemum canadense, ferns, sedges (i.e. Carex pennsylvanica), and other herbaceous vegetation such as Rubus spp. and Solidago spp. Five sites were in St. Lawrence Islands National Park: Hill Island (HLL), Gordon Island (GOR), Grenadier Island (GRE), Camelot Island (CAM), and Jones’ Creek (JCR), two at Frontenac Provincial Park (FRN and FRS), one each at Murphy’s Point Provincial Park (MPP) and Charleston Lake Provincial Park (CPP), and three in Northumberland and Peterborough Counties: Bentlage (BEN), Linton (LIN), and lands of the Otonabee Region Conservation Authority (ORC). All of the forests were located near agriculture and two of these sites, BEN and LIN, were privately owned woodlots. Precambrian and igneous rocks underlay all study locations except for sites that were situated off the Frontenac Arch region of the Canadian Shield: BEN, LIN, and ORC. Soils varied from well-drained sand and sandy loams to more poorly drained clay loams, although there was frequently as much variation in soil type and drainage within a location as among locations (Hoffman et al. 1964). The climate of the region is humid continental, with precipitation of between 700 and 1,000 mm year−1, and January and July temperatures of −7.7 and 20.3°C, respectively, recorded at Kingston Airport (Environment Canada 2008).
https://static-content.springer.com/image/art%3A10.1007%2Fs10530-009-9612-3/MediaObjects/10530_2009_9612_Fig1_HTML.gif
Fig. 1

Study sites in southeastern Ontario and forested area within corresponding counties (gray shading). Study area in relation to northeastern North America denoted within inset (black shading)

Vegetation sampling

The area covered by each plant species less than 1.5 m in height was visually estimated to the nearest percent in 1 m × 1 m quadrats in May, July, and September 2006, at all sites. The maximum percent cover attained by each species in each plot was extracted from the three sampling dates for use in analyses. Quadrats were located along two to three randomly located transects within each site, and the number of quadrats per transect varied from between 10 and 25, depending on site area and availability of habitat. The distance between quadrats along transects was randomly selected, varying from 4 to 8 m. Consequently, total transect length could vary between 36 and 192 m. Plants were identified to species as often as possible, and unidentifiable (i.e. small, non-flowering) plants were removed from the analysis (11 of 161 unique species in our dataset, Supplementary Table 1). Grasses and sedges were also excluded since these species were rarely observed flowering and difficult to identify accurately to species based solely on their vegetative characteristics. Since none of the unidentifiable species occupied a substantial area of the quadrats, all dominant species were identified (Supplementary Table 2).

Patterns of forest fragmentation and deer density estimates

We calculated five indicators of fragmentation in order to assess how human land use affected patterns of invasive non-native species. The fragmentation metrics calculated for each mainland site were: (1) the percentage of forested area, (2) the density of forest patches, (3) the density of forest edges, (4) the mean Euclidean distance to the nearest neighboring forest patch, and (5) mean shape index (see Supplementary Table 3 for details). Mean shape index typically ranges from 1.0 to 3.0, with larger values describing less circular shapes that have a greater edge to area ratio (Laurance and Yensen 1991). Metrics were calculated from rasterised forest cover data (1:10000 Ontario Base Maps; cell = 25 m × 25 m) using FRAGSTATS v.3.3 (McGarigal et al. 2002) within a circular area centered on each site and with a neighborhood radius of 500 m. We selected a maximum buffer zone of 500 m, because small seeded, non-native species are unlikely to disperse over greater distances and plant invasions are likely to be most strongly influenced by the surrounding landscape at this scale (Charbonneau and Fahrig 2004). Extremely rare long-distance dispersal events could also be captured within our buffer (Cain et al. 2000; Myers et al. 2004). We did not calculate fragmentation measures for island sites since our buffer size would have included portions of the mainland and confound our analysis with differences in dispersal dynamics between the aquatic and terrestrial matrices. We also calculated the distance from each site to the nearest urban area, defined as areas with population densities of ≥400 persons km−2.

At each site, deer densities were extrapolated from the mean heights of Trillium grandiflorum plants, using the regression developed by Koh et al. (2009) for 10 sites across southern Ontario. The nearest T. grandiflorum within 5 m of the centre of each quadrat was measured from the soil surface to the base of the leaf whorl at all sites in the spring of 2006 and per site means were used in regression calculations. Deer preferentially select taller plants that grow less than ungrazed plants in the subsequent year, and as such, heights of this widely distributed herb are a more reliable indicator of local-scale deer densities and browsing pressures than estimates based on hunter returns across regional Wildlife Management Units (Koh et al. 2009).

Environmental variables: understorey light and soil nitrogen

Photosynthetically active radiation (PAR; 400–700 nm) above understorey vegetation (0.5 m height) was measured with a Licor line quantum sensor (LI-191SA, LI-COR Biosciences, Lincoln, NE) at every quadrat during maximum canopy cover in July 2006. A Licor point sensor (LI-190SA, LI-COR Biosciences, Lincoln, NE) simultaneously recorded PAR in the nearest open area. Measurements were averaged over a period of 15 s to account for sun flecks, and % PAR was calculated by dividing the understorey values by those in the open area. All measurements were recorded between 11:00 a.m. and 3:00 p.m. on cloudless days. Due to cloud cover limitations, measurements were not obtained for Gordon and Camelot Islands.

Soil blocks, 6 cm × 6 cm × 6 cm, were sampled from the first, middle, and last quadrats of each transect for each site in May and September 2006. Samples were immediately frozen and later air dried for a minimum of 3 days. Total Kjeldahl Nitrogen (TKN) was measured colorimetrically with a Technicon Autoanalyzer following acid digestion on a Technicon block digester (McGill and Figueiredo 1993).

Statistical analysis

We used generalized linear models with Poisson error structures to compare native and non-native species richness at different spatial scales (i.e. across quadrats, transects, and sites) and model significances were tested with Chi-square tests (R version 2.8, R Foundation for Statistical Computing, Vienna). We also compared the percent cover of non-native species with both the percent cover and richness of native species at the site-level. Cover data were over-dispersed and as such, models were fit with a quasi-Poisson error structure (ver Hoef and Boveng 2007) for which F-tests are a more appropriate test of model significance (Hastie and Pregibon 1992). We also reported pseudo r2 values for our models, calculated as 1-(residual deviance/null deviance), given that we employed a Poisson error structure (Myers and Montgomery 1997). To compare patterns of native and non-native species richness and percent cover between island and mainland sites, and the percent of species in a quadrat that were non-native, we used t-tests or Mann–Whitney U-tests, where data were not normally distributed.

We used a series of generalized linear models to identify the importance of the following factors: human land uses, deer densities, PAR, and soil TKN, in explaining the percentage of species at a site that were non-native and the mean non-native species cover per site. We removed Gordon and Camelot Islands from our analyses, since these sites lacked PAR data, and the remaining two island sites in our dataset (Grenadier and Hill Islands) were also removed to avoid the effects of island sites biasing our results. We also excluded TKN measured in September, since this was a poorer model predictor than May measurements. We tested the following models: (a) each factor individually, (b) only abiotic factors (light and nitrogen), (c) only human land use factors (fragmentation measures and distance to nearest urban settlement); and (d) a model containing all of the fragmentation measures, removing inter-correlated factors. Models were fitted using maximum likelihood methods (Burnham and Anderson 2002), and assessed using the small sample unbiased Akaike Information Criterion (AICc) and the Akaike weight (wi) (Crawley 2005). The statistical significance of all models with lower AICc scores than the null model was compared against both the null model and the best fitting model using log likelihood ratio tests (denoted as L with subscript referring to degrees of freedom in the model comparison, Crawley 2005).

Results

Diversity–invasibility relationships at multiple scales across southern Ontario

Native and non-native species richness were positively and negatively correlated at all spatial scales, although, only positive relationships were statistically significant (Table 1). Only two transects had significant correlations between the number of native and non-native species per quadrat (Hill Is. Transect 2 and Murphys’ Point Transect 2, Table 1), and this positive relationship was also evident across transects at the site-level for Camelot Is., Hill Is., Jones’ Creek, and Murphys’ Point (Table 1). Across all sites surveyed in the greater landscape of southeastern Ontario, there was also a positive relationship between native and non-native species richness per quadrat (P < 0.001, slope = 0.10, n = 317). Averaging species data for each site, the coefficient of correlation for the regression between native and non-native species richness was much higher than that based on individual quadrats (Fig. 2; r2 = 0.59, P < 0.001, n = 12). The percent cover of non-native species in quadrats across all sites also showed a significant positive relationship with native percent cover (r2 = 0.06, P = 0.002, slope = 2.41, n = 317) but not with native species richness (r2 = 0.01, P = 0.362, slope = 2.76, n = 317).
Table 1

Correlations between non-native and native species richness per quadrat across quadrats along individual transects and across all quadrats at each of 12 sites in southern Ontario, Canada

Site

Scale

Slope

r2

P

n

Bentlage

Transect 1

0.20

< 0.01

> 0.999

15

Transect 2

0.76

0.04

0.666

10

All transects

0.95

< 0.01

0.829

25

Camelot Is.

Transect 1

0.89

0.01

0.875

10

Transect 2

1.16

0.27

0.160

10

All transects

1.27

0.20

0.048

20

Charleston L.

Transect 1

0.99

< 0.01

0.974

20

Transect 2

1.12

0.02

0.716

11

All transects

1.05

< 0.01

0.731

31

Frontenac Central

Only 1 transect

n/a—0 non-native species in all quadrats

Frontenac South

Only 1 transect

2.05

0.23

0.239

20

Grenadier Is.

Only 1 transect

1.01

< 0.01

0.878

20

Gordon Is.

Transect 1

0.88

0.01

0.821

10

Transect 2

1.08

0.01

0.723

20

All transects

0.97

< 0.01

0.895

30

Hill Is.

Transect 1

1.13

0.07

0.229

20

Transect 2

1.23

0.20

0.028

20

Transect 3

1.39

0.26

0.153

11

All transects

1.16

0.14

0.007

51

Jones’ Creek

Transect 1

1.12

0.18

0.057

15

Transect 2

1.05

< 0.01

0.909

15

All transects

1.21

0.34

< 0.001

30

Linton

Only 1 transect

1.14

0.02

0.598

20

Murphy’s Point

Transect 1

n/a—0 non-native species in all quadrats

Transect 2

1.90

0.40

0.017

20

All transects

1.93

0.27

0.033

30

Otonabee

Only 1 transect

1.14

0.04

0.611

20

Bold values are significant at α = 0.05

https://static-content.springer.com/image/art%3A10.1007%2Fs10530-009-9612-3/MediaObjects/10530_2009_9612_Fig2_HTML.gif
Fig. 2

The relationship between mean non-native and native species richness at the site level for four island (filled symbols) and eight mainland (open symbols) forest sites in southern Ontario: non-native richness = 0.30e(0.06 × native richness)

Forest understorey plant communities of islands were more likely to be affected by non-native plant species than those in mainland sites. Non-native species richness per quadrat was significantly greater on islands than on mainland sites, although there was still less than one non-native species per square metre (U = 15 609, P < 0.001, n = 317; Fig. 3). However, the percent cover of non-native species per quadrat was approximately 10-times greater on islands than on mainland sites, where it approached 1% (U = 15 952, P < 0.001, n = 317; Fig. 3). Non-native species also constituted a greater percentage of the flora on island sites compared to the mainland (U = 15 717, P < 0.001, n = 317; Fig. 3). These relationships were also significant at the site level for the percent cover of non-native species (U = 30, P = 0.022, n = 12) and percent of all species that were non-native (t10 = 2.83, P = 0.018), but not for total non-native species richness (U = 26, P = 0.112, n = 12). Slopes for the relationship between non-native and native species richness per quadrat at the site-level on islands also did not differ from mainland sites (mean slope for islands = 1.10; mean slope for mainland sites = 1.35; U = 11, P = 0.636, n = 11).
https://static-content.springer.com/image/art%3A10.1007%2Fs10530-009-9612-3/MediaObjects/10530_2009_9612_Fig3_HTML.gif
Fig. 3

Non-native species on island (filled bars) and mainland (open bars) sites. a Mean non-native species richness per quadrat. b Mean percent cover of all non-native species present per quadrat, and the mean percentage of all species in a quadrat that were non-native. Mean ± SE

Models explaining landscape-level patterns of non-native species

Mean shape index and the mean distance to the nearest forest patch were the only factors that provided a significantly better fit than the null model in predicting the percentage of species that were non-native at our sites (L1 = 7.22, P = 0.007 and L1 = 4.55, P = 0.033, respectively; Table 2). Sites that were surrounded by forest patches with a high ratio of edge to area contained a smaller percentage of non-native species relative to native species (Fig. 4a). These sites were also closer to neighboring forest patches (Fig. 4b). However, mean distance to the nearest forest patch and mean shape index were not correlated with each other (r2 = 0.31, P = 0.153). Among the other models, TKN, distance to the nearest urban settlement, the density of forest patches, and the distance to the nearest forest edge, were as strongly supported as the null model based on AICc scores but were not significantly different based on log likelihood ratio tests (P > 0.05; Table 2). No model explained mean non-native species cover per site better than the null model based on log-likelihood tests (P > 0.160; Supplementary Table 4).
Table 2

Candidate models for explaining the percentage of species at a site that were non-native

Candidate model

K

AICc

Δi

wi

Null model

2

47.28

4.64

0.04

Mean shape index

3

42.64

0.00

0.45

Mean distance to nearest forest patch

3

45.31

2.66

0.12

TKN

3

46.09

3.45

0.08

Mean distance to nearest forest edge

3

46.73

4.08

0.06

Distance to nearest urban settlement

3

46.85

4.20

0.06

Density of forest patches

3

47.01

4.37

0.05

Multiple fragmentation measures (Edge density, mean shape index, percent forested area)

3

47.71

5.07

0.04

PAR + TKN

5

48.16

5.51

0.03

PAR

4

48.54

5.89

0.02

Deer density

3

48.84

6.20

0.02

Percent forested area

3

49.39

6.74

0.02

Edge density

3

49.79

7.15

0.01

The number of estimated parameters in each model (K), the AICc value for each model, the difference in AICc between each model and the best fit model (Δi), and the Akaike weight (wi) are reported. Bolded text refers to the most significant candidate model compared to the null model using log-likelihood ratio tests (at α = 0.05)

https://static-content.springer.com/image/art%3A10.1007%2Fs10530-009-9612-3/MediaObjects/10530_2009_9612_Fig4_HTML.gif
Fig. 4

Relationship between the percent of all species that were non-native, at each of eight mainland forested sites across southern Ontario, Canada, and a mean shape index and b mean distance to the nearest forest patch (m), calculated within a 500 m buffer. Equations of the lines: non-native species = 22.93–10.63(mean shape index), r2 = 0.59 P = 0.025; and non-native species = 0.08(distance to nearest patch) + 0.36, r2 = 0.43 P = 0.076

Discussion

The richness and proportions of non-native plant species in southeastern Ontario forests were low compared with other studies in protected areas situated in landscapes with similar vegetation and disturbances, and covering as broad a geographic extent (~175 km) as our study (McKinney 2002; Clemants and Moore 2003; Howard et al. 2004). Generally, the same non-native species were recorded across our sites (i.e. Alliaria petiolata, Rhamnus cathartica, Leonurus cardiaca), and these species frequently covered less than 5% of the area sampled. Although non-native species may exhibit a “lag” phase of slow population growth before exponential increases in the amount of area invaded (Myers and Bazely 2003), most of the non-native species that we recorded were introduced between 1800 and the early 1900s (White et al. 1993), and our results suggest that several factors may influence this relationship.

Factors explaining diversity–invasibility relationships

Spatial heterogeneity (Davies et al. 2005) and species turnover (Stohlgren et al. 2005) will increase niche availability (Simberloff 1995) and may explain the positive relationship between native and non-native species richness at our sites. The similarity in species composition among all of our sites suggests that the differential establishment of non-native species was unlikely to be driven by competitive interactions between natives and non-natives (Daehler 2003), and this would explain the lack of a negative relationship between native and non-native species richness at smaller spatial scales (Davies et al. 2005). Results from a nearby, forested site in the St. Lawrence Valley have also suggested competitive interactions to be unimportant in determining the positive relationship between native and non-native species richness (Gilbert and Lechowicz 2005). Local variation in soil characteristics is often the primary factor explaining patterns of non-native species richness in similar forest types in the northeastern U. S. (Howard et al. 2004; McDonald et al. 2008) and non-native earthworms play an important role in structuring understorey plant communities in this region (Hale et al. 2006). Accelerated litter removal in northern hardwood forests by non-native earthworms uproots native plants, exposes seeds and seedlings to herbivory, and alters soil nutrient cycling and chemistry (Bohlen et al. 2004), and these changes may facilitate non-native plant invasions (Nuzzo et al. 2009). Earthworm grazing also disrupts mycorrhizal fungi (McLean et al. 2006), with which many native species are associated (Brundrett and Kendrick 1988). Although we did not quantify earthworm abundances, earthworm distributions are likely to differ across the large geographic area covered by our vegetation surveys and this factor warrants further investigation. One possibility is that since non-native earthworm introductions are strongly associated with recreational fishing (Hale et al. 2006), island sites may have higher levels than privately maintained, mainland woodlots that are surrounded by agriculture.

The positive relationship between native and non-native species richness suggests that islands in this ecoregion are not more invasible than mainland sites because of lower native species diversity (Elton 1958). Rather, islands may have a greater propagule pressure of non-native species compared to mainland sites (Lonsdale 1999) because of higher rates of human visitation than privately owned woodlots (BEN and LIN) and relatively inaccessible forest interiors (FRN and FRS) and/or traits of non-native species that confer greater dispersal abilities than native species (Long et al. 2009), e.g. fleshy fruits dispersed by birds. Islands may also have lower levels of native species propagule pressure than larger and less-isolated mainland sites (MacArthur and Wilson 1963). Larger islands or mainland sites may be better able to mitigate species turnover from disturbances such as human visitation (Lonsdale 1999), deer browsing (Gaston et al. 2006) or earthworm invasions, and we are not aware of any other large-scale biotic disturbances that would have varied across both mainland and island sites.

Influence of fragmentation on relationships between native and non-native species

Sites with a higher shape index (more edge to area) and that were located closer to other forest patches had a smaller proportion of non-native species. This result contrasts with many other studies that have found habitats with large amounts of edge to be highly invaded by non-native species (Brothers and Spingarn 1992; Parendes and Jones 2000; With 2002). However, fragmented forests with high shape indices may not always be prone to greater levels of invasion (Yates et al. 2004), particularly where propagule pressure is more important than resources in determining species establishment (Tanentzap and Bazely 2009). High edge to area ratios and closely situated forest patches may increase levels of connectivity for native species from a meta-population perspective, thereby benefiting native species that may be slower dispersers (e.g. myrmecochory or gravity) than non-native species, and this relationship can determine the recovery of forest understoreys from human-disturbances in northeastern North American forests (McLachlan and Bazely 2003). Similarly, Long et al. (2009) hypothesized that non-native species may be better dispersed than native species and therefore, better able to colonize isolated sites. Sites close to other forest patches may also be less invaded due to there being less open habitat adjacent to forests that would promote the establishment and dispersal of non-native species into forest interiors (McDonald et al. 2008). Greater connectivity may also allow greater movement by deer, which can decrease the abundance of those non-native species that are less adapted to browsing than native plants (Maron and Vilà 2001; Parker et al. 2006).

Forest patches with high mean shape values are more likely to be associated with agricultural modifications to the landscape (Corry and Lafortezza 2007; Saura and Carballal 2004), and these sites may contain significantly fewer proportions of non-native species than those in areas that are more frequented by the humans (Duguay et al. 2007). At the large protected areas in our study, which have higher human visitation rates than the private woodlots located within agricultural landscapes (e.g. HLL, CPP and MPP), human disturbances such as forest trails can introduce non-native species and create conditions favorable for their establishment (Dickens et al. 2005), e.g. exposed soil (Nuzzo et al. 2009). Non-native agricultural weeds commonly produce small, wind-dispersed seeds (Baskin and Baskin 2006), and intact edge vegetation can function as a physical barrier to these species (Cadenasso and Pickett 2001). The edge vegetation across our sites was comprised of shrubs (e.g. Rubus), saplings, and lateral branches from canopy trees, and may explain why only 25% of the non-natives we recorded were wind-dispersed (Supplementary Table 1). Within forested landscapes (patches with less edge to area), species that are less associated with agriculture may rely on other dispersal strategies (e.g. endozoochory: Rhamnus cathartica) or human transport along trails (Dickens et al. 2005), and hence, may be more likely to invade forest interiors successfully.

Conclusions

Our results suggest that the proportions of non-native species within forests are limited by patch connectivity that may disproportionately benefit native species richness. Although this may seem to contradict the finding that increases in native species coincide with increases in non-native species, changes in the proportion of non-native species arising from greater native species richness will be negligible. Sites with high native species diversity may be more prone to invasion than less diverse sites because these sites are more attractive to human visitation than less diverse sites, and hence associated with greater non-native species propagule pressure (McKinney 2002). While we did not experimentally test whether greater connectivity benefits native species, our correlative study is supported by other studies that suggest that regional species pools influence local-level species richness (Herben 2005; Freestone and Harrison 2006). Since propagule pressure is a primary determinant of species invasion (Levine 2000; Lockwood et al. 2005; Colautti et al. 2006), we predict that where regional species pools are dominated by native species, native species propagule pressure may limit non-native species establishment and abundance (Levine et al. 2004). Overall, management efforts should aim to isolate the propagule pressure of non-native species, in addition to maintaining the connectivity of habitat for native species.

Acknowledgments

We thank two anonymous reviewers for comments that substantially improved our article. We also thank J. van Wieren, E. Reid, and C. Brdar for logistic and field support, and P. Hertz, K. Turcotte, J. Labonne, and J. Langat for field and lab work. Frank Bentlage and Terry Linton kindly permitted access to their properties. Funding was provided by Parks Canada, the Ontario Federation of Anglers and Hunters, the Silverhill Institute of Environmental Research and Conservation, and a Toronto Regional Conservation Authority B. Harper Bull Fellowship.

Supplementary material

10530_2009_9612_MOESM1_ESM.doc (104 kb)
(DOC 104 kb)

Copyright information

© Springer Science+Business Media B.V. 2009