Factors associated with alien plant richness, cover and composition differ in tropical island forests

To examine how native plant native communities, environment and geography are associated with alien plant species invasion in tropical island forests.


| INTRODUC TI ON
The invasion of native communities by alien (exotic or non-native) species is one of the main threats to biodiversity. Island biodiversity is particularly threatened and vulnerable to invasion by alien species (Caujapé-Castells et al., 2010;Denslow, 2003;Moser et al., 2018).
Compared to mainland, islands have more alien species with small tropical islands exhibiting a disproportionately high number of alien species relative to their size (Dawson et al., 2017;Lonsdale, 1999;Turbelin, Malamud, & Francis, 2017;Westphal, Browne, MacKinnon, & Noble, 2008). Understanding how alien species invasions vary between and within these small tropical islands is of critical importance to managing this threat.
To establish and spread in native island communities, alien plant species have to overcome several filters. First, they must reach the island (i.e., the dispersal filter), then grow, survive and reproduce under local environmental conditions (i.e., the abiotic filter), while facing competition and other interactions with native communities (i.e., the biotic filter) (Blackburn et al., 2011;Pearson, Ortega, Eren, & Hierro, 2018;Richardson & Pyšek, 2006;Theoharides & Dukes, 2007). The dispersal filter is often overcome with the help of anthropogenic transportation supported by intentional (e.g., for horticulture, agriculture or botanical gardens) or unintentional importation to islands (Turbelin et al., 2017). The dispersal of alien species is indeed better correlated to socio-economic factors (e.g., island population, gross domestic product or degree of international trade) than geographical factors (Blackburn, Delean, Pyšek, & Cassey, 2016;Dawson et al., 2017;Kueffer et al., 2010;Moser et al., 2018;Westphal et al., 2008).
On a global scale, the abiotic filter seems to be of low importance to determining the degree of alien plant species invasion. Most studies have found no effect of climate Moser et al., 2018) or a weak positive effect of temperature (Blackburn et al., 2016;Dawson et al., 2017) on differences in alien species richness between islands. Within islands, alien species richness tends to decrease at higher elevations (e.g., Arévalo et al., 2005;Guo et al., 2018;Jakobs, Kueffer, & Daehler, 2010;Ohlemüller, Walker, & Wilson, 2006;Tanaka & Sato, 2016). This decrease is due to the effects of several confounding factors, including decreasing temperature, human disturbance, propagule pressure and greater biotic resistance (Pauchard et al., 2009). Nutrient and water availability are additional abiotic factors likely to affect the success of alien plant species invasion within islands (Alpert, Bone, & Holzapfel, 2000). On volcanic islands, nutrient availability varies with soil age (e.g., Hughes & Denslow, 2005;Vitousek, Aplet, Turner, & Lockwood, 1992), and low nutrient availability on young soils may constitute an important filter to alien species invasion (e.g., Ostertag & Verville, 2002;Zimmerman et al., 2008). Surprisingly, the effect of water availability on alien species invasion, for example along precipitation gradients, has been hardly explored (Alpert et al., 2000;Walther et al., 2009).
The biotic filter has raised considerable interest in invasion ecology. The hypothesis that the biotic resistance of native communities to alien species invasion increases with diversity (Elton, 1958) is one of the most popular and controversial hypotheses to explain the higher degree of invasion on islands compared to mainland (Denslow, 2003;Levine & D'Antonio, 1999;Richardson & Pyšek, 2006, 2008Simberloff, 1995;Vitousek, D'Antonio, Loope, Rejmanek, & Westbrooks, 1997). Indeed, because the factors and processes driving the invasibility of native communities vary with context (e.g., propagule pressure, environment or disturbance regime) and across spatio-temporal scales, both positive and negative relationships between native and alien species richness can be observed (Clark & Johnston, 2011;Davies, Harrison, Safford, & Viers, 2007;Fridley et al., 2007;Gurevitch, Fox, Wardle, & Taub, 2011). For instance, on a regional to global scale, more favourable or diverse environmental conditions may support higher native and alien species richness (leading to positive relationships), while on smaller scales (where biotic interactions occur and environmental conditions can be considered homogeneous), higher native species richness leaves fewer resources and niche opportunities to alien species, leading to negative relationships (Davies et al., 2007;Fridley et al., 2007;Shea & Chesson, 2002;Stohlgren et al., 1999).
Invasive alien species tend to be fast-growing and light-demanding (van Kleunen, Weber, & Fischer, 2010;Leishman, Haslehurst, Ares, & Baruch, 2007;Rejmanek & Richardson, 1996). As a result, the low level of light availability in the understorey of undisturbed forests tends to make them relatively resistant to alien species invasion (Fine, 2002). Higher light availability is also one of the factors explaining the higher degree of alien species invasion at forest edges or canopy gaps in comparison with forest interiors (e.g., Arellano-Cataldo & Smith-Ramírez, 2016;Green, Lake, & O'Dowd, 2004).
Lastly, several studies have tested whether the geophysical parameters of islands or archipelagos affect the degree of invasion on islands. The theory of island biogeography (MacArthur & Wilson, 1967) predicts that the number of species on an island results from an equilibrium between species immigration and extinction which in turn are affected by the size and isolation of islands. The general dynamic theory of oceanic island biogeography (Whittaker, Kostas, & Richard, 2008) supplemented this theory by adding the effect of island age and speciation. It predicts that the number of species on an island peaks at intermediate age, when high elevation and complex topography offer the highest diversity of environmental niches.
All studies support the prediction that native and alien plant species richness increases with island size, elevation or habitat diversity (Blackburn et al., 2016;Dawson et al., 2017;Denslow, Space, & Thomas, 2009;Kueffer et al., 2010;Moser et al., 2018). However, contrary to native species, alien plant species richness tends to increase with island isolation (Denslow et al., 2009;Moser et al., 2018) although some studies have found weak  or no effects (Traveset, Kueffer, & Daehler, 2014). Blackburn et al., (2016), suggested that isolation indirectly affects alien plant species richness through its effect on native species richness. The effect of island age on alien species richness has received less attention. Kueffer et al., (2010) found that within an archipelago, older islands tend to be less invaded and suggested that on younger islands, less diverse flora may leave more niche opportunities favouring invasions.
Here, we use plot data from forests located in Hawai'i and American Samoa, two archipelagos belonging to the Polynesia-Micronesia biodiversity hotspot (Mittermeier et al., 2005), to examine how native plant communities, environment and geography are associated with alien plant species invasion. We test the prediction that tree density, canopy height and the taxonomic richness of native plant communities play an important role in determining the resistance of native forests to invasion by alien species. We expect that (a) forests with high native tree density and high canopies prevent the establishment (i.e., lower richness) and growth (i.e., lower coverage) of alien species by decreasing light availability, and (b) the species richness and composition of native communities affect the species richness and composition of alien communities by determining available niches and levels of competition. We also expect that environment and geography should play an important role in determining the degrees of alien plant invasion. In particular, (c) the richness and coverage of alien plant species should be higher at lower elevations (higher temperature, disturbance and propagule pressure), on older soils (higher nutrient availability), in areas receiving higher precipitation (higher water availability), and on small and isolated islands (lower native plant diversity).

| Study sites
We analysed spatial variation in alien plant species richness, coverage and composition across 204 plots located in four United States National Parks. Plots were located in two archipelagos (Hawai'i and American Samoa) and five islands (Table 1). In the National Park of American Samoa, plots were located on two different islands: Tutuila and Ta'ū. Each island is volcanic but differed in age, size and elevation On the Island of Hawai'i, plots were located on the active Kīlauea and Mauna Loa volcanoes. This activity results in a mosaic of soils with different ages: the oldest ones originate from lava flows that occurred ~50 ka ago and the youngest ones from lava flows that occurred in 2018. No detailed map of soil ages are available for Ta'ū island, but the island is ~20-70 ka (McDougall, 2010). The other islands (Maui, Molokai and Tutuila) being older (0.75-1.90 Ma), we assumed that within islands variations in soil age is of minor importance. The effect of soil age on alien species richness and coverage was therefore only explored for plots located in the Island of Hawai'i.

| Vegetation surveys
Vegetation surveys were conducted between 2010 and 2016 in relatively undisturbed wet forests (Ainsworth, Berkowitz, Jacobi, Loh, & Kozar, 2011). In each 0.1-ha plots (20 m × 50 m), all native and alien vascular plant species were inventoried. Trees and woody plants (i.e., excluding tree ferns and palms) with a stem diameter ≥10 cm at ~1.3 m above the base (DBH, diameter at breast height) were recorded. We computed native tree density as the number of native trees per plot and canopy height as the average height of three trees representative of the canopy. In most plots (>70%), the canopy was exclusively comprised of native trees. When present in the canopy, alien trees made up only 13% of trees, on average. The coverage of alien species was measured using the point-intercept method (Elzinga, Salzer, Willoughby, & Gibbs, 2001) along three 50-m transects located along the two long edges and middle of the plot. Along these transects, the presence of species was recorded every 0.5 m if one or more species intercepted a 2-m height pole. The coverage of alien species was computed as the ratio between the number of measurement points where an alien species was intercepted and the total number of measurement points (i.e., 300 points). Highly invasive alien species were identified using the "100 of the world's worst invasive alien species" species list, which encompasses 33 terrestrial plant species (Global Invasive Species Database 2018, http://www. iucng isd.org/gisd/100_worst.php).

| Environmental features
Mean annual precipitation (MAP), minimal (T min ) and maximal tem-

| Statistical analyses
All analyses were performed using R 3.4.4 (R Core Team, 2018). We used generalized linear mixed-effect models to examine the fixed effects of the richness and structure (tree density and canopy height) of native communities, elevation, mean annual precipitation, and the random effects of island and archipelago identity on alien species richness and coverage (see Bunnefeld & Phillimore, 2012). Minimal and maximal temperatures were not included as explanatory variable as they strongly correlated with elevation ( Figure S1).
A Poisson distribution was used for alien species richness and a Binomial distribution for alien species coverage using the glmer function from the lme4 R package (Bates, Mächler, Bolker, & Walker, 2015). This analysis was also performed separately for the island of Hawai'i to assess the effects of soil age on alien species communities. The log-transformed soil age was then added as a fixed effect and the identity of sampling frames as random effects.
We used the MuMin R package (Bartoń, 2016) and the dredge function to generate different sets of models representing all possible combinations of subsets of fixed effects. We then selected the best models based on their corrected Akaike information criterion (AIC c ), which express the quality of a model as a function of the goodness of fit (maximum likelihood) and the number of parameters (ΔAIC c < 2 from the best models, Bunnefeld & Phillimore, 2012). We used conditional R 2 (with random effects) and marginal R 2 (without random effects) to assess the relative importance of fixed and random effects in GLMMs (Nakagawa & Schielzeth, 2013).
Given that some studied plots are located close to each other ( Figure 2), spatial autocorrelation in a models' residuals may violate the assumptions of the models and compromise the interpretation of the results (Dormann et al., 2007). To avoid spatial autocorrelation, we subsampled the data set before fitting the models. For each island, we built a dendrogram (hclust function) based on the geographical distance between plots and Ward's grouping method. We then cut the dendrogram at 1,000 m height (cutree function) to group plots and randomly sampled one plot per group. As a result, no plots were located closer than 300 m from another plot in the subsamples. Random subsampling was done 100 times and the parameters of the best models were averaged using the model.avg function. We The proportion of variance explained by environment and geographical distance between plots was estimated by comparing the variance explained by different models computed with both environment and geographical distance as predictors (full models) and with only environment or geographical distance as predictors (Legendre, 2008). Finally, to assess whether the composition of alien communities may be affected by the composition of native communities, we tested the correlation between the residuals of the full GDMs, that is, the variance in alien and native species turnover unexplained by environment and geographical distance.

| State of the invasion
Alien species were found in >90% of the studied plots ( Figure 2).
On average, four alien species covering about 15% of the understorey were found in each plot. The average alien species richness did not vary significantly between islands but their average coverage was lower on Tutuila (2.6%) and higher on the Island of Hawai'i (Myrtaceae) and Rubus ellipticus (Rosaceae). At least one of these highly invasive species was present in 55% of the plots (Figure 2).

| Species richness and coverage
Elevation and native communities together with the identity of the island and archipelago explained ~70% and 65% of the variation in alien species richness and coverage among plots, respectively

| Species turnover
Together, environmental and geographical distance between pairs of plots explained 87.0% of the variance in turnover of native species but only 39.5% of the variance in turnover of alien species.
Moreover, contrary to turnover in alien species, turnover in native species was primarily explained by geographical distance (Figures   6 and 7). Indeed, geographical distance only affected alien species turnover at short distances (within park) while it affected native species turnover from short (within park) to long distances (between archipelago, Figure S4). Although differences in precipitation significantly affected alien species turnover, differences in elevation were the main driver of the observed turnover (Figure 7). The unexplained variance in alien species turnover was significantly but poorly correlated with the unexplained variance in native species turnover. The strength and the significance of this correlation did not change when we considered all pairs of plots or only pairs located on the same archipelago (Pearson's R = .11, p value <.001).

| D ISCUSS I ON
The high degree of invasion of Pacific islands by alien plant species (Dawson et al., 2017) was observed in relatively undisturbed tropical wet forests of Hawai'i and American Samoa. At least one alien plant species was found in the vast majority of the studied plots (>90%) and at least one of the world's worst invasive alien species selected by the IUCN invasive species specialist group (Lowe, Browne, F I G U R E 3 Mean fixed (mean annual precipitation, elevation, native richness, canopy height and native tree density) and random effects (island and archipelago identity) on alien species richness and coverage. Vertical segments represent 95% confidence intervals around the mean, and numbers above bars represent the percentage of iterations (among 100 iterations, N = 78 plots for each iteration) that retained the variable within the best model sets. R 2 m and R 2 c refer to marginal (without random effects) and conditional (with random effects) R 2 . For random effects, black and grey bars correspond to islands located in Hawaii and American Samoa, respectively. Note that within archipelagos islands are sorted in order of age (the youngest on the left hand side). Spatial autocorrelation in models' residuals was significant for only 6% and 4% of the 100 iterations ( Figure S2) F I G U R E 4 Elevational patterns of alien species richness and coverage. Lines represent fitted generalized linear models. "***" = p values <.001, "**" = p values <.01, "*" = p values <.05, ns. = p values >.05 Boudjelas, De, & Poorter, 2000) was observed in more than half of the plots. This high degree of invasion is particularly alarming since it is likely to be much higher outside the national park sites included in this study (see Lonsdale, 1999).
Our best models were able to explain 65%-70% of the variance in the richness and coverage of alien species across islands. In both cases, the identity of archipelagos and to a lesser extent the identity of islands explained the most variance, followed by elevation and finally native community richness or structure for alien species richness or coverage, respectively. Most of the variance in alien species composition remained unexplained, with most of the explained variance shared by environment and geographical distance.

| Archipelago and island effects
Alien species richness and coverage tended to be higher in plots located in Hawai'i compared to those located in American Samoa.
This archipelago effect may have resulted from the greater isolation of the Hawaiian archipelago, one of the most isolated land masses on earth. Indeed, as suggested by several studies (Blackburn et al., 2016;Denslow et al., 2009;Moser et al., 2018), isolation is likely to F I G U R E 5 Mean fixed (mean annual precipitation, elevation, soil age, native richness, canopy height, and native tree density) and random effects (sampling frame) on alien species richness and coverage in Hawai'i Volcanoes National Park (HAVO, Hawaii). Vertical segments represent 95% confidence intervals around the mean, and numbers above bars represent the percentage of iterations (among 100 iterations, N = 44 plots for each iteration) that retained the variable within the best model sets. R 2 m and R 2 c refer to marginal (without random effects) and conditional (with random effects) R 2 . Spatial autocorrelation in models' residuals was not significant for all of the 100 iterations ( Figure S3) F I G U R E 6 Variation partitioning (%) of alien and native species turnover. The full generalized dissimilarity models (GDM) are represented in Figure 7 indirectly affect alien species invasion success through its effect on the diversity of native communities. Notably, due to its isolation the native Hawaiian flora originates from relatively few long-distance colonization events followed by local radiations (e.g., Gemmill, Allan, Wagner, & Zimmer, 2002). This results in a highly original flora (~90% of the flowering plants are endemic compared to ~30% in American Samoa) but with reduced genetic and functional diversity (e.g., the Hawaiian flora encompasses more species but less genera than the American Samoa), leaving more niche opportunities to invaders (see Denslow, 2003;Simberloff, 1995).
Within the archipelagos, plots located on the younger islands (the Island of Hawai'i in Hawai'i and Ta'ū in American Samoa) tended to be more invaded. Kueffer et al. (2010) compared alien plant species richness between 30 island groups and also found that in younger island groups, the number of dominant invaders (i.e., those reaching maximal coverage >25% in natural areas) tended to be higher. They suggested that on younger islands, a less diverse flora may leave more niche opportunities (as supported by the general dynamic theory of oceanic island biogeography, see Whittaker et al., 2008) favouring invasion. In our case, an alternative hypothesis may be that the higher degree of invasion in Hawai'i Volcanoes National Park may result from higher accessibility and number of visitors. Lonsdale (1999) found that the number of alien plant species in protected areas increases with the number of visitors. However, McKinney (2002) found no effect of the number of visitors on alien species richness in US parks (including National Parks). The alternative hypothesis that a higher degree of alien plant species invasion is due to higher number of visitors was not verified in this study for American Samoa. Indeed, the degree of alien species invasion was higher on the poorly accessible and less visited island of Ta'ū than on the main island of Tutuila.
The effect of age also held within Hawai'i Volcanoes National Park, where plots were located on soils resulting from lava flows that occurred 200-30,000 years ago. We found that, in this range of ages, plots located on younger soils tend also to exhibit higher alien species richness and coverage. Conversely, Zimmerman et al. (2008), analysing forest growing on 50-1,500 years old lava flows, found higher degree of invasion on older soils. Low N availability on young soil may indeed limit the establishment and spread of alien species (Hughes & Denslow, 2005;Ostertag & Verville, 2002;Turner & Vitousek, 1987). Combining our results with those from Zimmerman et al., (2008), we suggest that the invasibility of these forests may rather follow a humped curve. Alien plant species may be limited by nutrient availability on young soils and competition with more diverse native communities on older soils.
F I G U R E 7 Generalized dissimilarity models (GDMs) transformation functions for alien (full lines) and native (dotted) species turnover (β sim ). Only significant effects are shown (p value <.05). MAP = mean annual precipitation

| Forest structure affects alien species coverage
The vast majority of studies explore alien species invasion at large scales (e.g., between protected areas, regions or countries).  (Aplet, Hughes, & Vitousek, 1998). This threat is all the more im-

| Turnover in alien species composition
Contrary to our expectation, native species composition did not affect alien species composition. Similar to Kueffer et al. (2010), we did not find clear patterns in alien species composition. Although environment and geographical distance had significant effects, most of the variance in alien species turnover remained unexplained. While large-scale geographical distances (between archipelagos and islands) had a strong effect on native species turnover, it did not affect alien species turnover (Leihy, Duffy, & Chown, 2018). This suggests that alien species easily reach those islands. While on a global scale climate may be a major driver of alien plant species composition, on smaller scales it may rather rely on introduction, land use history, and stochastic colonization or disturbance events (e.g., Burns, 2016;Kueffer et al., 2010;Seipel et al., 2012).

| CON CLUS ION
Most studies on invasions by alien plant species have focused on alien species richness, while the capacity for established alien populations to expand within natural communities has retained less interest. We show that abiotic and biotic factors associated with alien species richness differed from those associated with alien species cover. Elevation was the factor that explained the most variance in the degree of invasion by alien species with higher elevations being relatively spared so far. Our results also suggest that elevation has larger effects on alien species coverage than on alien species richness. Canopy height and tree density were not associated with lower alien species richness but rather with reduced alien species coverage. Finally, our results support that this threat may be particularly high on young and isolated islands. Weisenborn, Anthony Wyberski, Corie Yanger and to all those that have helped with data collection and management. We thank parkbased staff for assistance with all logistical efforts. We thank three anonymous referees for helpful comments on the manuscript.

DATA AVA I L A B I L I T Y S TAT E M E N T
All data are available on the Integrated Resource Management Application (IRMA) portal of the National Parks Service (https :// irma.nps.gov/Porta l/).