Ecography 37: 001–010, 2014 doi: 10.1111/ecog.00963 © 2014 The Authors. Ecography © 2014 Nordic Society Oikos Subject Editor: Reagan Early. Editor-in-Chief: Miguel Araújo. Accepted 7 July 2014

Widespread native and alien plant species occupy different habitats Robin Pouteau, Philip E. Hulme and Richard P. Duncan R. Pouteau ([email protected]), Conservatoire des Espaces Naturels, Presqu’île de Foué, 98 860 Koné, New Caledonia. – P. E. Hulme, R. P. Duncan and RP, Bio-Protection Research Centre, PO Box 84, Lincoln Univ. 7647, Canterbury, New Zealand. RP also at: Research Inst. for Development (IRD), UMR AMAP, Laboratory of Applied Botany and Plant Ecology, Herbarium NOU, 98848 Noumea, New Caledonia. RPD also at: Inst. for Applied Ecology, Univ. of Canberra, Canberra, ACT 2601, Australia.

Theories to explain the success of alien species often assume that they are inherently different from native species. Although there is an increasing body of evidence showing that alien plants tend to dominate in highly human-modified environments, the underlying reasons why widespread natives might differ in their habitat distribution have rarely been addressed. We used species distribution models to quantify the dominant environmental axes shaping the habitat of 95 widespread native and alien herbaceous species in a highly modified grassland-dominated landscape in New Zealand. For each species, support vector machines were used to determine 1) the environmental variables that most strongly determined a species’ distribution; 2) the affinity towards a particular position along environmental axes; and 3) tolerance to environmental variation. These three measures were compared among native perennials (n ⫽ 31), alien perennials (30) and alien annuals (34). Independent of their origin, species’ distributions were defined by similar environmental variables. Nevertheless, native and alien species occupied different regions of the dominant environmental axes. Perennial natives occupied regions associated with lower human disturbance, while perennial aliens were associated with habitats that had been modified by vegetation clearance, pasture development and livestock grazing. Annual aliens differed from perennials and were associated with both semi-natural and more intensively managed vegetation. No evidence was found that aliens had broader environmental tolerances than natives that might facilitate invasion into a wider range of environments. Thus, widespread native and alien species differ in the degree to which environmental factors shape their distribution as a result of anthropogenic perturbations to which they respond differently as well as the introduction of functional groups that are capable of exploiting novel environments.

The search for fundamental differences between alien and native plant species that might explain biological invasions remains a major focus in ecology (van Kleunen et al. 2010). While numerous studies have identified one or more plant attributes that differ between co-occurring alien and native species (Lambdon et al. 2008a, Hulme 2011, Knapp and Kühn 2012), the ecological attributes that distinguish successful aliens and natives are often strongly habitatdependent (Thompson et al. 1995). Thus, rather than comparing plant attributes we might gain a clearer insight into the character of plant invasions by asking whether prevalent native and alien species differ in their distributions and if these are shaped by similar environmental factors (MacDougall et al. 2009). Alien plants often occupy human-dominated sites associated with disturbance and high propagule pressure, but the extent to which co-occurring widespread natives behave differently has rarely been assessed (Gassó et al. 2009, Pyšek et al. 2010, Aikio et al. 2012). The few studies addressing the environmental associations of both native and alien species mostly focus on patterns in native and alien species richness rather than the distribution of individual species (Roy et al. 1999, Boughton et al. 2010, Marini et al. 2012, Tomasetto

et al. 2013). However, models of species richness do not always mirror the drivers of the underlying individual species distributions and tend to overpredict richness in species-poor sites while underestimating it in species-rich sites (Bennett 2014, Calabrese et al. 2014). The few attempts to compare the individual habitat of a large number of native and alien species have used differences in occupancy across a range of environmental categories such as habitat types (Prinzing et al. 2002, Lambdon et al. 2008b, Knapp and Kühn 2012). The use of environmental categories on their own often represents only a coarse classification that fails to capture the subtleties influencing species distributions and disregards the considerable co-variation among environmental gradients that shapes a plant’s habitat. In contrast, species distribution models (SDM) are designed to model habitats by linking the spatial distribution of a species to underlying environmental variables (Fielding and Bell 1997, Guisan and Thuiller 2005). These models identify a multidimensional hypervolume that describes the environment a species occupies, and which has been interpreted as closely mirroring the Hutchinsonian definition of the realised niche (Warren 2012, 2013 but see McInerny and Etienne 2013 for a contrary view). SDM can thus quantify Early View (EV): 1-EV

the ecological importance of different environmental factors associated with species distributions, including providing objective measures of environmental preference and tolerance (Hirzel et al. 2002). Yet SDM have never been used to compare the environmental associations of co-occurring natives and aliens (Steiner et al. 2008, Wolmarans et al. 2010). Here we use SDM to assess whether native and alien species are fundamentally different in how their current distributions are related to environmental factors, irrespective of whether this arises through biotic or abiotic constraints. New Zealand provides an opportunity to examine whether native and alien plants occupy different regions of the environmental space due to 1) the high proportion of naturalised alien species in the flora (Diez et al. 2008); 2) the marked under-representation of particular life-forms in the native but not alien flora (e.g. annuals, nitrogen-fixing legumes, rhizomatous grasses), potentially leading to vacant niches (Allen et al. 2006, Wilson and Lee 2012); and 3) recent and dramatic anthropogenic transformation of the natural landscape including extensive conversion of native forests to grasslands, leading to new sets of environmental conditions to which alien plant species may be pre-adapted but to which native species have yet to adapt (Rose et al. 1995, King and Wilson 2006). If habitat differences do exist between native and alien plants, we expect them to be especially pronounced in grasslands due to their high susceptibility to invasion and the potential role of niche dissimilarities in shaping herbaceous communities (Tilman 1997, Seastedt and Pyšek 2011). We constructed individual SDM for 95 widespread native and alien plant species that co-occur in the grasslands of Banks Peninsula, New Zealand, and examined whether their distributions were shaped by different sets of environmental factors and if so, why.

Methods Study area Banks Peninsula (ca 1000 km2) on the east coast of New Zealand’s South Island (43°33′–43°54′S, 172°37′–173°07′E; Fig. 1A) comprises the eroded remnants of two large and extinct shield volcanoes, leading to a complex topography ranging from 0 to 920 m a.s.l. The climate is temperate with annual mean temperature ranging from 9°C to 14°C and annual rainfall from 600 to 2000 mm with greater rainfall at higher elevations. Soils derived from volcanic bedrock and loess are moderately to very fertile (Wilson 2009). Prior to human colonization, Banks Peninsula was almost completely forested but Polynesian burning (commencing about 1300 AD) and then European forest clearance (beginning in the 1800s) converted the Peninsula to a grassland-dominated landscape (Wilson 2009). The present day landscape comprises a mosaic of about 10% original or regenerating native forest, 5% native scrub, 10% alien shrubland and forest plantations, and 75% grassland. The latter ranges from less-modified and lessproductive areas dominated by native tussock grasses including Festuca spp., Poa spp. and Chionochloa spp., to areas highly modified by fire, fertilization, ploughing, irrigation, oversowing and grazing, resulting in productive pas2-EV

tures dominated by alien grasses and legumes e.g. Dactylis glomerata, Lolium perenne and Trifolium spp. Species data and sampling design We used data from a comprehensive floristic survey conducted between 1983 and 1988 in which 1227 plots measuring 6 ⫻ 6 m were located systematically at 920 m (i.e. 1000 yards) intervals across the entire Banks Peninsula (Fig. 1B; Wilson 1992). The plot size (36 m2) was consistent with the minimum area sufficient to describe herbaceous communities (Mueller-Dombois and Ellenberg 1974). A total of 1021 vascular species were recorded and each species was classified according to whether they were native or alien. While many SDM studies are based on presenceonly data derived from opportunistic sampling schemes such as herbarium data, our botanical survey provided a systematic presence/absence dataset for a large and representative sample of the vegetation drawn across a varied landscape. Since our aim was to compare the environmental factors shaping the distribution of common native and alien plants, we controlled for differences in life-form and confined our analyses to herbaceous species. To compare natives and aliens, we selected the 20% most prevalent herbaceous species (the mean number of plots a species in this group occupied was 230), which ensured that we had sufficient presence points to accurately model distributions (Stockwell and Peterson 2002). This method selected 33 native species, including 31 perennials and two annuals (Cotula australis and Helichrysum bellidioides) and 64 alien species including 30 perennials, two biennials and 32 annuals (Supplementary material Appendix 1). The two native annual species were not included in the analysis because the small sample size would prevent generalisations regarding this life-form. Alien annuals and biennials were combined into a single annual category to facilitate analysis. We therefore focused on three groups for comparison: native perennials, alien perennials and alien annuals. The majority (94%) of the aliens were recorded as naturalised in New Zealand in the 1800s (Supplementary material Appendix 1) and their widespread distribution across Banks Peninsula suggests that these species had sufficient time to colonise most environmentally suitable sites. We thus aimed to minimise the problem that alien species with expanding distributions will not be accurately modelled using SDM. Environmental variables We described the environment of the 1227 plots using a geographic information system (GIS) managed under ArcGIS 10.0 (ESRI 2010) and including variables chosen a priori to capture the most critical dimensions of the environmental space (Supplementary material Appendix 2). To enable perpixel computation of the multi-format and multi-scale GIS layers, they were preprocessed in two steps. First, shapefiles were rasterised at 10 m resolution to match the resolution of the digital elevation model (DEM) and each pixel was assigned to the polygon that overlapped its centre. Second, raster resolution was homogenised to 10 m by either assigning each 10 m pixel the value of the original, larger raster in

Figure 1. (A) The inset shows the location of Banks Peninsula within New Zealand. (B) Digital elevation model of Banks Peninsula and location of the 1227 plots. (C, D and E) Potential species richness reconstructed from the individual species distribution models of each of the 31 perennial natives (C), 30 perennial aliens (D) and 34 annual aliens (E). The shading gradient in (C) is common to the three groups of species.

which it was nested or combining small pixels into larger 10 m-resolution pixels which were assigned the mean value of the smaller aggregated pixels. Each 6 ⫻ 6 m vegetation plot was then allocated to the appropriate 10 ⫻ 10 m pixel on which it was centred. We chose four climatic variables as surrogates for regional gradients in temperature and moisture that are likely to affect plant productivity and survival across the landscape: mean annual air temperature, mean annual rainfall, total solar radiation (Tecco et al. 2010, Marini et al. 2012, Tomasetto et al. 2013) and, following Box (1981), calculated a moisture index derived from a combination of the first three variables. Annual mean temperature, annual rainfall and total solar radiation were averaged for the period 1971–2000 and provided at an initial resolution of 500 m by the National Inst. of Water and Atmospheric Research (NIWA) (⬍www.niwa. co.nz/our-science/climate/our-services/mapping⬎). Five physiographic variables with 10 m initial resolution were used to capture the topoclimate and resource gradients driven by topography. Elevation was incorporated as a proxy for air temperature (lapse-rate) and rainfall. Slope influences soil moisture through runoff rates and its effect on local radiation and temperature. Aspect also strongly influences local radiation and hence soil moisture, and in New Zealand livestock grazing pressure also tends to be higher on warmer

north facing-slopes (Scott et al. 2001). Aspect was extracted from the DEM then decomposed into both ‘northness’ and ‘eastness’, but eastness was not considered further since it did not appear to be an important environmental axis in preliminary tests. The distance of each plot from the nearest coastline was used as a surrogate for factors such as salt spray exposure, while distance to the nearest river reflected environmental conditions associated with riparian systems. Three variables linked to anthropogenic factors were also extracted. First, we used a land-cover map (⬍www.mfe.govt. nz/issues/land/land-cover-dbase/⬎) from 1996–1997 to classify plots into six land-cover categories (improved grassland, unimproved grassland, broadleaved hardwood, pine forest, manuka and kanuka, and gorse and broom). These categories reflect differences in land-cover linked to the degree of human disturbance through factors such as forest clearance, pasture development and livestock grazing (Tecco et al. 2010). Land-cover data were produced eight years after the vegetation surveys were completed but changes in landuse across Banks Peninsula are likely to have been minor over this period with little urban expansion or change in agricultural practices (Winterbourn et al. 2008). The remaining two variables were distance to the nearest road and distance to the nearest built-up area, reflecting proximity to human development potentially associated with increased 3-EV

disturbance and propagule pressure for alien species (Aikio et al. 2012, Tomasetto et al. 2013).

model predictions over chance) as an additional measure of fit (Fielding and Bell 1997).

Species distribution modelling method

Comparison of native and alien species distributions

We used support vector machines (SVM) which overcome most of the practical difficulties and data limitations that face classical SDM (Drake et al. 2006). SVM make no assumptions regarding the distribution of training data, tolerate non-independence of input data which enables the use of collinear variables with unshared information (e.g. northness and total solar radiation) (Dormann et al. 2013), can deal with both continuous and categorical environmental variables (Guo and Liu 2010) and perform well, typically outcompeting traditional statistical approaches using maximum entropy or classification and regression trees (Bedia et al. 2011). The environmental variables we measured form a 12dimensional space describing the environment available for species occupancy on Banks Peninsula. SVM aim to circumscribe the region of this environmental space that is occupied by a species from the unoccupied region, and uses the occupied region to identify which set of environmental conditions characterises species presence and thus shapes species distributions. SVM contour species environmental distributions by maximising the distance of presence points at the edge of the occupied environmental space from absence points at the edge of the unoccupied space. The shape of the occupied region fitted by SVM is controlled by a kernel function and a regularisation parameter. The kernel function is used to discriminate presence and absence points when they are not linearly separable and we used the Gaussian radial basis function which relies on tuning only one kernel-specific hyperparameter, γ, while outputting accurate results. The regularisation parameter, C, controls the trade-off between generalisation of the occupied region (by maximising the distance between the occupied and unoccupied portions of the available environmental space) and minimisation of misclassification errors. We used the two-class SVM implementation in the R package ‘e1071’ (Meyer et al. 2012). To avoid misclassification problems due to unbalanced training datasets (Batuwita and Palade 2013), we randomly excluded presence points of the most widespread species prevalent in more than half the plots and excluded absence points of the rarest species found in fewer than half the plots so as to obtain an equal number of presence and absence points for each species. The ‘tune.svm()’ function was used to find the optimal γ and C in the range [2⫺10, 2⫺9, …, 210] that minimised classification error (proportion of misclassified plots) after 10-fold cross-validation to prevent overfitting. Assessment of SVM performance was based on the area under the curve (AUC) of the receiver operating characteristics (ROC) plot calculated from a randomly selected 25% hold-out dataset (Fielding and Bell 1997). After assessing AUC, continuous maps were converted to presence/absence maps by selecting thresholds where sensitivity was equal to specificity (Liu et al. 2005). We computed Cohen’s Kappa (which expresses the proportion of specific agreement between observed and predicted presence/absence scores and assesses the improvement of 4-EV

A SVM model was built for each species from data recorded in the 1227 plots to classify all 10 ⫻ 10 m pixels covering the landscape (n ⫽ 10 125 390) as predicted to be occupied or unoccupied, and the occupied pixels were then used to describe the species’ environmental associations across Banks Peninsula. Three metrics were calculated for each species and each environmental variable: 1) the relative contribution to the overall model; 2) the species Preference; and 3) the species Specialisation. The relative contribution to the overall model of each environmental variable allows ranking the variables that most clearly separate the occupied from the unoccupied environmental space, and hence most strongly determine the species distributions. It was based on a jackknife approach that evaluates the change in classification accuracy between a full model and one with that particular environmental variable omitted. The mean relative contribution (MRC) averaged over all species within each group (native perennials, perennial aliens and annual aliens) provides a measure of the importance of each environmental variable in determining species distributions within these groupings (Gormley et al. 2011). Preference expresses the affinity of each species towards a particular location within the available environmental space. For each variable, a species’ Preference was defined as the standardised difference between the centroid of the occupied portion (the mean value of the environmental variable for all pixels where the species was predicted to be present, mP) and the centroid of the unoccupied portion of the available environmental space (the mean for all pixels where it was predicted to be absent, mA). This difference was standardised by dividing by 1.96 standard deviations of the total available environmental space σT (1). Preference ⫽ (mP – mA)/(1.96 ⫻ σT)

(1)

Preference was described by both its magnitude and sign. The magnitude of the Preference reflects the extent to which a species is found in regions of the environmental space that are distinct from the background availability with larger values indicating greater distinctiveness. The sign captures the association with a particular environmental variable, with a positive Preference value indicating a species is more likely to be present on pixels with high values of that variable. Preference for the categorical land-cover variable was computed as the difference between the frequencies of presence and absence in each land-cover category, such that a positive sign means that the species is more frequently present than absent in that category. Preference for land-cover was not standardised: thus while its magnitude gives a relative ranking within land-cover categories it is not comparable to the other environmental variables. Specialisation expresses the extent to which a species is distributed across the entire variation in the available

environmental space. For each species and each environmental variable, Specialisation was defined as the standard deviation of the environmental variable across all pixels of the available environmental space σT divided by the standard deviation when measured only across those pixels where the species is present σP (2) (Hirzel et al. 2002). Specialisation ⫽ σT / σP

(2)

A value close to one reflects a ubiquitous species while a value above one denotes increased specialisation. Specialisation was only computed for each non-categorical environmental variable. Linear mixed-effect models (LME) were used in order to compare MRC, Preference and Specialisation of each group (perennial natives, perennial aliens and annual aliens) while controlling for any taxonomic bias due to the unbalanced number of species within different plant families (Supplementary material Appendix 3). LME analyses were performed in R using the ‘lme4’ package (Bates et al. 2012). Groups were defined as a fixed effect and plant family as a random effect. The significance of LME statistics was estimated by comparison to a probability distribution obtained by 10 000 Markov chain Monte Carlo (MCMC) simulations using the ‘pvals.fnc()’ function of the ‘languageR’ package (Baayen 2008).

Results Native perennial species occurred on average in 10% of plots and were three times less prevalent than perennial alien species and just over half as prevalent as annual alien species across Banks Peninsula (Supplementary material Appendix 3). Not surprisingly, perennial natives were most prevalent in native broadleaved hardwood and native scrub (dominated by manuka Leptospermum scoparium and kanuka Kunzea ericoides) land-cover categories, and perennial aliens were less frequent than perennial natives or annual aliens in unimproved grassland (pairwise comparison t-test with Bonferroni post-hoc correction; p-value ⬍ 0.001). Although SVM for native species were derived from a smaller number of presence records, all SVM performed well at characterising species’ environmental space (AUC ⫽ 0.983 ⫾ 0.008 and Kappa ⫽ 0.886 ⫾ 0.043 for perennial natives, AUC ⫽ 0.976 ⫾ 0.009 and Kappa ⫽ 0.898 ⫾ 0.051 for perennial aliens; AUC ⫽ 0.977 ⫾ 0.012 and Kappa ⫽ 0.899 ⫾ 0.039 for alien annuals). The MRC rankings of the different environmental variables were similar for each species group (Fig. 2) reflected in strong correlations between each of the species groups (r2 ⬎ 0.97, p-value ⬍ 0.001; n ⫽ 12 in all cases). The only statistical difference in MRC values was found between perennial natives and aliens for the climatic cluster of variables, the former having a significantly lower MRC (LME; p-value ⬍ 0.001). The prevalence of all three groups was most strongly defined by similar physiographic and human-related variables, while climatic variables were of less importance. Northness was the environmental variable that most clearly distinguished the occupied from unoccupied space and thus most clearly defined species’ prevalence,

irrespective of species origin. Northness was then followed by two anthropogenic variables: the distance to the nearest built-up area and land-cover. In contrast to the similarity in MRC values, native and alien species had very different Preference patterns (Table 1, 2). This is clearly evident from their contrasting spatial distributions on Banks Peninsula (Fig. 1). Perennial natives occupied an environmental space that differed the most from the remainder of the available environmental space and thus showed the strongest Preference. These species tended to occur in cooler, steeper locations at higher elevation further from built-up areas and roads, which were often classified as broadleaved hardwood forest or unimproved grassland (Fig. 1C). Perennial aliens had lower specificity in Preference but there was a tendency for them to favour locations with higher moisture index, higher annual rainfall and proximity to rivers, and to be strongly associated with improved grasslands (Fig. 1D). In contrast to perennial aliens, annuals showed marked Preference for warm, dry, low rainfall areas at low elevation usually close to roadsides in both improved and unimproved grasslands (Fig. 1E). Perennial natives and aliens differed consistently in the magnitude of their Preference, the former always showing greater Preference when LME analyses indicated a significant difference between these two groups (Table 1). The same trend occurred for all categories of land-cover other than pine forest (Table 2). In contrast, perennial natives and annual aliens always differed in the sign of their Preference with distance to the nearest built-up area the only exception (Table 1). Perennial natives and annual aliens also differed in the sign of their Preference for land-cover (Table 2). While perennial aliens and natives both had a contrasting Preference for unimproved grasslands at high elevation, alien annuals showed no difference in Preference for these land-cover categories. Perennial natives also appeared to prefer habitats under a woody canopy. There was no consistent trend in the breadth of the environmental space occupied by the three groups as described by Specialisation (Table 3). For each group some environmental axes were broader or narrower than others and, as a result, mean Specialisation was similar across the groups (one way ANOVA F(1,30) ⫽ 4.01, p ⫽ 0.673). Although Specialisation was relatively consistent across environmental variables for perennial species, it appeared much more variable for annual aliens, with this group having both the highest (total solar radiation) and lowest (distance to built-up area) levels of Specialisation. Alien annuals had the greatest Specialisation for climatic variables and the least for human related variables. Differences were less marked for the other two groups with perennial aliens having the greatest Specialisation for human-related variables, especially proximity to roads, and perennial natives with physiographic variables. Pairwise comparisons show that perennial natives had a greater Specialisation than perennial aliens in the dimensions described by northness and the distance to the nearest river whereas perennial aliens had a greater Specialisation for the distance to the nearest road. Perennial natives appeared more specialised than annual aliens for physiographic variables (namely slope steepness, elevation and distance to the nearest river) but less specialised for climatic variables (especially annual mean temperature and total solar 5-EV

Figure 2. Comparison of environmental variables’ mean relative contribution (MRC). MRC scores are calculated as the difference in AUC between a full SVM model and one with each environmental variable omitted one by one. Clusters of variables are computed by omitting all environmental variables included in each cluster (Supplementary material Appendix 2). The gradients specifies in which cluster the environmental variables were grouped.

radiation). Compared to perennial aliens, annual aliens tolerated a wider range of human-related conditions (distance to the nearest built-up area and to the nearest road), often exhibiting no specialisation (Specialisation close to one), but they were much more restricted in terms of climatic conditions (annual mean temperature and global solar radiation). Table 1. Comparison of the dominant environmental axes characterised by the Preference metric. We indicate whether species Preference is significantly different from zero using bold figures (t-test if the assumption of a normal density was met, otherwise Wilcoxon signed-rank test) and superscripts indicate whether Preference is significantly different between perennial natives, perennial aliens and annual aliens after pairwise comparison (LME). Environmental variables are sorted by the order of their MRC value. Test significance levels are given in Supplementary material Appendix 4. Environmental variables Northness Distance to the nearest built-up area Distance to the nearest shore Slope steepness Elevation Distance to the nearest road Annual mean temperature Total solar radiation Distance to the nearest river Annual rainfall Moisture

6-EV

Perennial natives

Perennial aliens

Annual aliens

ⴚ0.106A 0.148A

⫺0.022B ⴚ0.062B

0.049C 0.035C

0.058A 0.159A 0.185A 0.109A ⴚ0.171A ⴚ0.107A ⴚ0.057A 0.046A 0.091A

0.004A ⴚ0.007B 0.036B ⴚ0.022B ⴚ0.026B 0.004B ⴚ0.063A 0.114A 0.099A

ⴚ0.103B 0.104A ⴚ0.138C 0.055A 0.148C 0.170C 0.052B ⴚ0.173B ⴚ0.199B

Discussion Independently of origin, the spatial distribution of plants is expected to be strongly shaped by their ecological requirements for water, nutrients, light and temperature, and their sensitivity to disturbance (Schulze et al. 2005, Gurevitch et al. 2006). Our results highlight that a similar set of environmental variables shape the distribution of widespread alien and native plant species on Banks Peninsula, and that the relative importance of these environmental variables is similar, even for annual and perennial life-histories. The primary difference among species of different origin is that native and Table 2. Comparison of the dominant environmental axes characterised by the Preference metric in the dimension described by land-cover. We indicate whether species Preference is significantly different from zero using bold figures (t-test if the assumption of a normal density was met, otherwise Wilcoxon signed-rank test) and superscripts indicate whether Preference is significantly different between perennial natives, perennial aliens and annual aliens after pairwise comparison (LME). Test significance levels are given in Supplementary material Appendix 5. Land-cover categories

% study area

Improved grassland Broadleaved hardwood Unimproved grassland Pine forest Manuka and kanuka Gorse and broom

68 8 7 7 6 4

Perennial natives ⴚ12.32A 3.41A 3.71A ⫺0.41A 2.96A 2.65A

Perennial Annual aliens aliens 9.77B ⴚ2.72B ⴚ2.04B ⴚ2.87B ⫺1.41B ⫺0.74B

1.17C ⫺3.03B 2.23A ⫺0.58A 0.19B 0.01B

Table 3. Comparison of the dominant environmental axes characterised by the Specialisation metric. The score of the clustered variables was obtained by averaging scores of all environmental variables included in each cluster (Supplementary material Appendix 2). We indicate whether species Specialisation is significantly different from one using bold figures (t-test if the assumption of a normal density was met, otherwise Wilcoxon signed-rank test) and superscripts indicate whether Specialisation is significantly different between perennial natives, perennial aliens and annual aliens after pairwise comparison (LME). Environmental variables are sorted by order of their MRC value. Test significance levels are given in Supplementary material Appendix 6. Environmental variables Physiographic cluster Human-related cluster Climatic cluster Northness Distance to the nearest built-up area Distance to the nearest shore Slope steepness Elevation Distance to the nearest road Annual mean temperature Total solar radiation Distance to the nearest river Annual rainfall Moisture

Perennial natives

Perennial aliens

Annual aliens

1.088A 1.025AB 1.064A 1.092A 1.043A

1.050AB 1.076A 1.085A 1.052B 1.069A

1.048B 0.985B 1.256B 1.073B 0.973B

1.031A 1.068A 1.056A 1.008A 1.040A 1.019A 1.295A 1.103A 1.093A

1.023A 1.050A 1.088A 1.082B 1.075A 1.138A 1.139B 1.060AB 1.067A

1.040A 1.012B 1.223B 0.997A 1.166B 1.727B 0.991C 1.036B 1.094A

alien species tend to separate out into different regions of the space defined by these variables, reflected in marked differences in the Preference of alien and native species to the same set of environmental variables. A key finding is that differences in Preference among perennial and annual aliens were as, if not more, marked than between either of these groups and perennial natives. Indeed, it is likely that individual lifeforms, whether alien or native, will respond differently to environmental gradients (Marini et al. 2012). Such information is often lost when examining trends in species richness alone and thus while our results for native species corroborate those found when modelling species richness on Banks Peninsula, the different environmental correlates of alien annual and perennial taxa are not captured by a single richness metric (Tomasetto et al. 2013). There was little evidence that broader environmental tolerances (as estimated by Specialisation) play a strong role in shaping the differences in the prevalence of native and alien species, and thus does not support the idea that alien species benefit from a wider niche breadth (Kitayama and MuellerDombois 1995). However, there was evidence that patterns of environmental tolerance differed between annual and perennial plants. Overall, our results support the view that the environmental correlates of widespread alien and native plants differ, with especially marked differences between native perennials and alien annuals. The question then arises as to what is driving these differences. Alien plant species benefit from anthropogenic change Slope aspect (northness) was the variable that most strongly defined the distributions of native and alien plant species on Banks Peninsula. While slope aspect is a physiographic variable, it is also a proxy for anthropogenic disturbance.

Drier north-facing slopes would have been more easily cleared of native vegetation by fire (Perry et al. 2012) and these warmer aspects would have been more favourable for pasture development and livestock grazing (Scott et al. 2001). Plant species on Banks Peninsula therefore appear to separate out along a gradient related to aspect that is linked to historic and current land-use associated with fire, intensity of pasture development and livestock grazing. How do native and alien plant species partition this gradient? Native species were most prevalent on south-facing aspects, and in semi-natural vegetation of intermediate and advanced successional stage including tussock grasslands and native scrub, usually on cooler, less disturbed areas distant from built-up areas and roads. Semi-natural grasslands on Banks Peninsula are largely a product of forest clearance by burning and have been colonised by native species that originally occurred in forest gaps, coastal cliffs and rocky outcrops (Wilson 2009). In contrast, perennial aliens occur on north-facing slopes and in improved grasslands that have been subject to more intensive development including fertiliation and higher levels of livestock grazing. Perennial aliens showed no strong association with roads or elevation which we might expect if dispersal was limiting their distribution as a consequence of spread from roadside corridors or lower elevation sites of introduction. Rather they appeared more closely associated with available moisture in high rainfall areas and/or close to rivers. The overall dominance of grassland cover on Banks Peninsula combined with historical sowing of grasses across the landscape (and associated weed seed contaminants) might explain the limited evidence for dispersal limitation of alien plant species in this study. Given this marked difference in Preference between native and alien perennials, what accounts for the higher prevalence of the latter? A few widespread native herbaceous species appear to be pre-adapted to the conditions associated with the major landcover changes that accompanied European arrival on Banks Peninsula, and some native species (e.g. Aceana anserinifolia, Oxalis exilis) are recorded as invasive alien plants elsewhere in the world (Yeates and Williams 2006). Nevertheless, landuse change following European arrival has overwhelmingly favoured alien species in New Zealand and the distribution of native and alien perennial plant species mirrors this trend. Grazing mammals were absent from New Zealand until European colonisation and as a result native plant species are generally poorly adapted to ungulate herbivory (Mark 1993, Antonelli et al. 2011), whereas the most prevalent alien perennial species (e.g. Dactylis glomerata, Holcus lanatus, Lolium perenne) have coevolved with grazers and are able to readily replace grazed meristems (Seastedt and Pyšek 2011). Similarly, repeated burning and fertiliser addition tends to favour faster growing and more productive alien plant species over natives in New Zealand (Craine and Lee 2003). Under this scenario it appears logical that alien species, which are better pre-adapted to the contemporary environment, achieve higher prevalence than natives. Alien plant species benefit from vacant niches Annuals might be expected to be closely associated with human-modified environments due to their ability to tolerate high levels of disturbance (Grime 2002). However, 7-EV

on Banks Peninsula, while alien annual species were found close to roadsides their distribution was distinct from alien perennials, being less associated with built-up areas. The distribution of alien annuals also differed from native perennials and appeared strongly associated with sites that have low moisture and rainfall as well as high solar radiation and temperature, in both improved and unimproved grasslands. Such conditions would be typical of sites that undergo seasonal drought over summer to which annuals would be well adapted (Grime 2002). In addition to having few native species pre-adapted to mammalian grazing or fire, the flora of New Zealand is underrepresented in key functional groups including having few nitrogen fixers or annual species (Allen et al. 2006, Wilson and Lee 2012). While there are 54 species of native nitrogen-fixers in seven genera Coriaria (Coriariaceae), Carmichaelia, Chordospartium, Clianthus, Sophora (Fabaceae), Gunnera (Gunneraceae) and Discaria (Rhamnaceae), over 40 genera of legumes have been introduced to New Zealand (McQueen et al. 2006). Introduced legumes amount to several hundred species including many herbaceous taxa such as Trifolium spp. and Vicia spp. (Fabaceae). Similarly, only about 2% of the native flora are classed as annuals (Wilson and Lee 2012). Thus in addition to transforming the landscape, European settlers introduced plant functional groups that were previously rare in New Zealand. The relatively high prevalence of alien nitrogen-fixers and annual species has been attributed to their ability to exploit both the new environments created by European land-use changes and the presence of other disturbed environments that were not fully exploited by native species (Wilson and Lee 2012). Most of the Banks Peninsula landscape has been transformed by human land-use. In the face of intensive pasture development and livestock grazing, alien perennials have replaced native perennial vegetation. In more productive grasslands, competition between native and alien plants may be an important structuring force but competition may be less important in low productive grasslands (Gross et al. 2013). At higher elevations, evidence of direct competition between aliens and natives is weak (Wiser et al. 1998, Meffin et al. 2010, Gross et al. 2013), suggesting that alien species may be the passengers of land-use change rather than driving out native species (Rose et al. 1995). Indeed, it appears competition may be more marked among perennial aliens (Makepeace et al. 1985, Radford et al. 2007). In contrast, annuals appear to opportunistically benefit from environmental space that is unoccupied by either native or alien perennial species.

Conclusion Although the patterns observed in this study reflect features associated with the Banks Peninsula landscape, our SDM approach appears useful in describing the differences in the environmental factors that shape the distribution of native and alien species. While our results corroborate those from studies based on species attributes that indicate species prevalence is dependent on origin (Prinzing et al. 2002, Lambdon et al. 2008b, Knapp and Kühn 2012), they offer a clearer perspective as to the potential mechanisms that 8-EV

underpin differences in distributions between widespread alien and native species. Such differences appear primarily attributable to anthropogenic changes to which natives and aliens respond differently, as well as the introduction of novel functional groups that are capable of exploiting anthropogenic disturbance in semi-natural grasslands. This situation is clearly distinct from the view drawn from the northern hemisphere (Thompson et al. 1995, Roy et al. 1999, Kühn et al. 2004) where widespread native and alien species both appear to benefit from anthropogenic disturbance as a result of human transformation of the environment. The longer history of anthropogenic disturbance in much of the northern hemisphere has undoubtedly led to a subset of native species that are adapted to intensively grazed, disturbed, eutrophic conditions and thus share many of the traits of successful alien species. This does not appear to be the case for New Zealand and suggests that our SDM approach would be a valuable extension to studies in Europe and North America. Acknowledgements – The authors are most grateful to Federico Tomasetto for setting up part of the GIS used in this study and Hugh Wilson (Hineway Reserve, Akaroa, New Zealand) for sharing the botanical data of his inventory on Banks Peninsula. This work was funded through a New Zealand’s Tertiary Education Commission grant to The Bio-Protection Research Centre.

References Aikio, S. et al. 2012. The vulnerability of habitats to plant invasion: disentangling the roles of propagule pressure, time and sampling effort. – Global Ecol. Biogeogr. 21: 778–786. Allen, R. B. et al. 2006. Updated perspective on biological invasions in New Zealand. – In: Allen, R. B. and Lee, W. G. (eds), Biological invasions in New Zealand. Springer, pp. 435–451. Antonelli, A. et al. 2011. Absence of mammals and the evolution of New Zealand grasses. – Proc. R. Soc. B 278: 695–701. Baayen, R. H. 2008. languageR: data sets and functions with ‘Analyzing linguistic data: a practical introduction to statistics’. – R package ver. 1.4, ⬍http://cran.r-project.org/web/packages/ languageR/⬎ accessed 5 February 2014. Bates, D. et al. 2012. lme4: linear mixed-effects models using S4 classes. – R package ver. 0.999999-0, ⬍http://cran.r-project. org/web/packages/lme4/⬎ accessed 5 February 2014. Batuwita, R. and Palade, V. 2013. Class imbalance learning methods for support vector machines. – In: He, H. and Ma, Y. (eds), Imbalanced learning: foundations, algorithms, and applications. Wiley, pp. 83–96. Bedia, J. et al. 2011. Predicting plant species distribution across an alpine rangeland in northern Spain. A comparison of probabilistic methods. – Appl. Veg. Sci. 14: 415–432. Bennett, J. R. 2014. Comparison of native and exotic distribution and richness models across scales reveals essential conservation lessons. – Ecography 37: 120–129. Boughton, E. H. et al. 2010. Land-use and isolation interact to affect wetland plant assemblages. – Ecography 33: 461–470. Box, E. O. 1981. Macroclimate and plant forms: an introduction to predictive modeling in phytogeography. – Junk. Calabrese, J. M. et al. 2014. Stacking species distribution models and adjusting bias by linking them to macroecological models. – Global Ecol. Biogeogr. 23: 99–112. Craine, J. and Lee, W. 2003. Covariation in leaf and root traits for native and non-native grasses along an altitudinal gradient in New Zealand. – Oecologia 134: 471–478.

Diez, J. M. et al. 2008. Darwin’s naturalization conundrum: dissecting taxonomic patterns of species invasions. – Ecol. Lett. 11: 674–681. Dormann, C. F. et al. 2013. Collinearity: a review of methods to deal with it and a simulation study evaluating their performance. – Ecography 36: 27–46. Drake, J. M. et al. 2006. Modelling ecological niches with support vector machines. – J. Appl. Ecol. 43: 424–432. ESRI 2010. ArcMap-ArcInfo version 10.0. Software and manual. – Ecological Systems Research Inst., Redlands, CA, USA. Fielding, A. H. and Bell, J. F. 1997. A review of methods for the assessment of prediction errors in conservation presence/ absence models. – Environ. Conserv. 24: 38–49. Gassó, N. et al. 2009. Exploring species attributes and site characteristics to assess plant invasions in Spain. – Divers. Distrib. 15: 50–58. Gormley, A. M. et al. 2011. Using presence-only and presence– absence data to estimate the current and potential distributions of established invasive species. – J. Appl. Ecol. 48: 25–34. Grime, J. P. 2002. Plant strategies, vegetation processes, and ecosystem properties. – Wiley. Gross, N. et al. 2013. Functional differences between alien and native species: do biotic interactions determine the functional structure of highly invaded grasslands. – Funct. Ecol. 27: 1262–1272. Guisan, A. and Thuiller, W. 2005. Predicting species distribution: offering more than simple habitats models. – Ecol. Lett. 8: 993–1009. Guo, Q. and Liu, Y. 2010. ModEco: an integrated software package for ecological niche modeling. – Ecography 33: 637–642. Gurevitch, J. et al. 2006. The ecology of plants. – Sinauer. Hirzel, A. H. et al. 2002. Ecological-niche factor analysis: how to compute habitat-suitability maps without absence data? – Ecology 83: 2027–2036. Hulme, P. E. 2011. Contrasting impacts of climate driven flowering phenology on changes in alien and native plant species distributions. – New Phytol. 189: 272–281. King, W. M. and Wilson, J. B. 2006. Differentiation between native and exotic plant species in a dry grassland: realized responses to perturbation, and comparison with fundamental responses. – Austral Ecol. 31: 984–995. Kitayama, K. and Mueller-Dombois, D. 1995. Biological invasion on an oceanic island mountain: do alien plant species have wider ecological ranges than native species? – J. Veg. Sci. 6: 667–674. Knapp, S. and Kühn, I. 2012. Origin matters: widely distributed native and non-native species benefit from different functional traits. – Ecol. Lett. 15: 689–703. Kühn, I. et al. 2004. The flora of German cities is naturally species rich. – Evol. Ecol. Res. 6: 749–764. Lambdon, P. W. et al. 2008a. Do alien plants on Mediterranean islands tend to invade different niches from native species? – Biol. Invasions 10: 703–716. Lambdon, P. W. et al. 2008b. Do non-native species invasions lead to biotic homogenization at small-scales? Similarity and functional diversity of habitats compared for the alien and native components of Mediterranean floras. – Divers. Distrib. 14: 774–785. Liu, C. R. et al. 2005. Selecting thresholds of occurrence in the prediction of species distributions. – Ecography 28: 385–393. MacDougall, A. S. et al. 2009. Plant invasions and the niche. – J. Ecol. 97: 609–615. Makepeace, W. et al. 1985. Interference phenomena due to mouseear and king devil hawkweed. – N. Z. J. Bot. 23: 79–90. Marini, L. et al. 2012. Alien and native plant life-forms respond differently to human and climate pressures. – Global Ecol. Biogeogr. 21: 534–544.

Mark, A. F. 1993. Indigenous grasslands of New Zealand. – In: Coupland, R. T. (ed.), Ecosystems of the world. Volume 8b. Natural grasslands – Eastern Hemisphere. Elsevier, pp. 361–410. McInerny, G. J. and Etienne, R. S. 2013. ‘Niche’ or ‘distribution’ modelling? – Trends Ecol. Evol. 28: 191–192. McQueen, J. C. et al. 2006. Consequences of alien N2-fixers on vegetation succession in New Zealand. – In: Allen, R. B. and Lee, W. G. (eds), Biological invasions in New Zealand. Springer, pp. 295–306. Meffin, R. et al. 2010. Experimental introduction of the alien weed Hieracium lepidulum reveals no significant impact on montane plant communities in New Zealand. – Divers. Distrib. 16: 804–815. Meyer, D. et al. 2012. Misc functions of the Department of Statistics (e1071), TU Wien. – R package ver. 1.6-1, ⬍http:// cran.r-project.org/web/packages/e1071/⬎ accessed 5 February 2014. Mueller-Dombois, D. and Ellenberg, H. 1974. Aims and methods of vegetation ecology. – Wiley. Perry, G. L. W. et al. 2012. Reconstructing spatial vulnerability to forest loss by fire in pre-historic New Zealand. – Global Ecol. Biogeogr. 21: 1029–1041. Prinzing, A. et al. 2002. Which species become alien? – Evol. Ecol. Res. 4: 385–405. Pyšek, P. et al. 2010. Contrasting patterns in the invasions of European terrestrial and freshwater habitats by alien plants, insects and vertebrates. – Global Ecol. Biogeogr. 19: 317–331. Radford, I. J. et al. 2007. Functional and performance comparisons of invasive Hieracium lepidulum and co-occurring species in New Zealand. – Austral Ecol. 32: 338–354. Rose, A. B. et al. 1995. Vegetation change over 25 years in a New Zealand short-tussock grassland: effects of sheep grazing and exotic invasions. – N. Z. J. Ecol. 19: 163–174. Roy, D. B. et al. 1999. Effects of urban land cover on the local species pool in Britain. – Ecography 22: 507–515. Schulze, E. D. et al. 2005. Plant ecology. – Springer. Scott, N. A. et al. 2001. Biogeochemical impact of Hieracium invasion in New Zealand’s grazed tussock grasslands: sustainability implications. – Ecol. Appl. 11: 1311–1322. Seastedt, T. R. and Pyšek, P. 2011. Mechanisms of plant invasions of North American and European grasslands. – Annu. Rev. Ecol. Evol. Syst. 42: 133–153. Steiner, F. M. et al. 2008. Combined modelling of distribution and niche in invasion biology: a case study of two invasive Tetramorium ant species. – Divers. Distrib. 14: 538–545. Stockwell, D. R. B. and Peterson, A. T. 2002. Effects of samplesize on accuracy of species distribution models. – Ecol. Model. 148: 1–13. Tecco, P. A. et al. 2010. Functional traits of alien plants across contrasting climatic and land-use regimes: do aliens join the locals or try harder than them? – J. Ecol. 98: 17–27. Thompson, K. et al. 1995. Native and alien invasive plants: more of the same? – Ecography 18: 390–402. Tilman, D. 1997. Community invisibility, recruitment limitation, and grassland biodiversity. – Ecology 78: 81–92. Tomasetto, F. et al. 2013. Environmental gradients shift the direction of the relationship between native and alien plant species richness. – Divers. Distrib. 19: 49–59. van Kleunen, M. et al. 2010. Are invaders different? A conceptual framework of comparative approaches for assessing determinants of invasiveness. – Ecol. Lett. 13: 947–958. Warren, D. L. 2012. In defense of ‘niche modeling’. – Trends Ecol. Evol. 27: 497–500. Warren, D. L. 2013. ‘Niche modeling’: that uncomfortable sensation means it’s working. A reply to McInerny and Etienne. – Trends Ecol. Evol. 28: 193–194.

9-EV

Wilson, H. D. 1992. Banks Peninsula ecological region. – Report for the New Zealand Protected Natural Areas Programme, Dept of Conservation, Christchurch. Wilson, H. D. 2009. Natural history of Banks Peninsula. – Canterbury Univ. Press. Wilson, J. B. and Lee, W. G. 2012. Is New Zealand vegetation really ‘problematic’? Dansereau’s puzzles revisited. – Biol. Rev. 87: 367–389. Winterbourn, M. et al. 2008. The natural history of Canterbury. – Canterbury Univ. Press.

Supplementary material (Appendix ECOG-00963 at ⬍www. ecography.org/readers/appendix⬎). Appendix 1–6.

10-EV

Wiser, S. K. et al. 1998. Community structure and forest invasion by an exotic herb over 23 years. – Ecology 79: 2071–2081. Wolmarans, R. et al. 2010. Predicting invasive alien plant distributions: how geographical bias in occurrence records influences model performance. – J. Biogeogr. 37: 1797–1810. Yeates, G. W. and Williams, P. A. 2006. Export of plant and animal species from an insular biota. – In: Allen, R. B. and Lee, W. G. (eds), Biological invasions in New Zealand. Springer, pp. 85–100.