Climatic and biogeographical drivers of functional diversity in the flora of the Canary Islands

Aim: Functional traits can help us to elucidate biogeographical and ecological processes driving assemblage structure. We analysed the functional diversity of plant species of different evolutionary origins across an island archipelago, along environmental gradients and across geological age, to assess functional aspects of island bio-geographical theory.


| INTRODUC TI ON
Functional traits determine the responses of species to the environment and can be used to assess meaningful variation of species assemblages through time and space (Díaz & Cabido, 2001;Kattge et al., 2020).Functional traits not only help us to understand how environmental processes have influenced the ecological strategies of species, but also can inform about how evolutionary conditions have led to adaptation and specialization of species, especially in oceanic archipelagos (Whittaker et al., 2014).However, studies on the functional diversity of oceanic island floras that apply island biogeographical theory (e.g., MacArthur & Wilson, 1967;Whittaker et al., 2008) are lacking, but hold the promise to answer fundamental questions on how spatial and ecological processes drive functional diversity patterns within insular systems and how functional diversity patterns compare between islands (Ottaviani et al., 2020;Patiño et al., 2017;Schrader et al., 2021).
Functional diversity (i.e., the trait variability in a species assemblage) can be quantified as the multidimensional trait volume that is occupied by a species assemblage (Blonder, 2018;Blonder et al., 2014).On the one hand, expansion of the functional trait volume of an assemblage with increasing species richness might indicate the exploitation of novel regions of niche space (MacArthur, 1965).
On the other hand, if the niche space of an assemblage becomes more densely packed as species richness increases, it suggests finer specialization or greater overlap of ecological niches (Klopfer & MacArthur, 1961;Pigot et al., 2016).Thus, measuring functional diversity might help us to gain a better understanding of fundamental ecological strategies of species of different origins (e.g., endemic, non-endemic native, and alien).
The floras of oceanic islands are comparatively species poor, disharmonic and rich in endemics (Kier et al., 2009;Kreft et al., 2008;Taylor et al., 2019).Oceanic islands harbour species that have colonized from the mainland, overcoming strong dispersal filters, and eventually evolved or even diversified into endemic species (Stuessy et al., 2006).This evolution in isolation might have led to a high trait differentiation in native island species, making some islands hotspots of functional diversity (García-Verdugo et al., 2020).
However, an isolated evolution and small distributions have left endemic species on islands particularly vulnerable to habitat loss, climate change and biological invasions through alien species that have been introduced intentionally or unintentionally by humans (Fernández-Palacios, Kreft, et al., 2021;Macinnis-Ng et al., 2021;Veron et al., 2019).Given the high, possibly unique, trait diversity on oceanic islands and its inherent vulnerability, it is important to understand how endemic species, non-endemic native species and alien species are distributed in space and the nature of their adaptations to island environments.
Oceanic islands, and in particular islands of volcanic origin, are often distinguished by highly heterogeneous environments with distinct orographic precipitation regimes over short geographical distances (Kier et al., 2009;Weigelt et al., 2013).Water availability is a well-known driver of functional diversity and species richness (Poorter et al., 2010;Wright et al., 2007).In arid conditions, multidimensional functional hypervolumes and related the resulting patterns to cli- functional diversity tends to be low, because functional traits are subject to strong environmental filtering through physiological constraints (Spasojevic et al., 2014).Hence, only species with specific functional traits might be adapted to survive strong environmental pressure (Cornwell & Ackerly, 2009).In particular, leaves and flowers are energetically costly for plants when water availability is limited (Roddy et al., 2019).Thus, dry environments should select species with water-efficient traits, such as smaller statures, leaves and flowers (Kuppler & Kotowska, 2021;Moles et al., 2009;Wright et al., 2017).However, other mechanisms, such as the limiting similarity of coexisting species (MacArthur & Levins, 1967), can shape the distribution of trait values.In contrast, the distribution of alien species on oceanic islands is determined mainly by their pathway of introduction (Pauchard et al., 2009).Alien species decline from low elevations, where they have been introduced predominantly, to high elevations owing to directional ecological filtering (i.e., progressive dropping out of species with narrow ecological niches; Alexander et al., 2011) via water availability (Irl et al., 2021).On many highly elevated trade wind islands, water availability increases from the coast up to the orographic cloud layer (Garzón-Machado et al., 2014) and, consequently, alien species with broad climatic niches should be more prevalent in arid conditions, because species with narrow climatic niches tend to be filtered out selectively along a humidity gradient (Alexander et al., 2011;Irl et al., 2021).
Island biogeographical models predict that diversity of species assemblages on oceanic islands varies in relationship to isolation, area and island age as a result of immigration, speciation and extinction (MacArthur & Wilson, 1967;Schrader et al., 2021;Whittaker et al., 2008).The models suggest that isolated areas are less diverse in species owing to reduced dispersal probability with greater distance to the mainland, influencing the chance of colonization (MacArthur & Wilson, 1967;Weigelt & Kreft, 2013).In contrast, less-isolated areas receive continuous arrival of propagules, increasing assemblage diversity (Brown & Kodric-Brown, 1977).Moreover, dispersal probability depends on plant height and diaspore size, because tall and small-seeded species can disperse their diaspores further owing to their lower settling velocity (Greene & Johnson, 1993;Thomson et al., 2011;Venable & Brown, 1988).Hence, it can be expected that plant height and diaspore size of colonizers (i.e., non-endemic native species) are more similar in isolated habitats, because primarily species with similar trait combinations might reach those habitats.
Functional diversity of endemic species should increase with isolation relative to species richness, because speciation rates should be higher where colonizing assemblages are disharmonic (König et al., 2020).Geographically isolated areas on oceanic islands can limit gene flow with mainland populations and offer greater ecological opportunity for exploitation of resources; this tends to enhance endemic species diversity (Steinbauer et al., 2012;Whittaker & Fernández-Palacios, 2007).Hence, adaptive speciation might be more important in isolated habitats than non-adaptive speciation (e.g., genetic drift; Losos & Ricklefs, 2009;Marques et al., 2019).Potentially, alien species could also profit from unoccupied niches, but unidirectional expansion from the sources of anthropogenic introduction might prevent them from reaching more isolated sites through directional ecological filtering (Alexander et al., 2011;Irl et al., 2021).
Likewise, topography can act as a dispersal barrier between populations, potentially leading to in situ speciation (Irl et al., 2015;Otto et al., 2016).Hence, topographic complexity might favour local adaptations by reducing genetic swamping (i.e., genetic homogenization through hybridization; Herrera & Bazaga, 2008), possibly leading to an increase of functional diversity.Moreover, it has been shown that topography predicts habitat diversity (Barajas-Barbosa et al., 2020;Keppel et al., 2016), which can have vital implications for species richness-area relationships on islands (Hortal et al., 2009).However, at small spatial scales an area-heterogeneity trade-off could prevent native assemblages from diversifying in very complex environments because the effective area of individual habitats is reduced, especially for species with narrow niches (Allouche et al., 2012).Given that flat environments have low habitat diversity and topographically complex environments have limited effective area per habitat, functional diversity of native assemblages might have a unimodal relationship with topographic complexity.Alien species are more likely to be subject to directional ecological filtering, because human introduction to easily accessible areas might so far have impeded colonization of topographically complex areas on oceanic islands (Alexander et al., 2011;Steinbauer et al., 2017).
We also expect plant species to be more functionally diverse relative to species richness in environments with a heterogeneous geological structure.For instance, it has been shown that geological heterogeneity is highest at an intermediate geological age, when soils have already formed and erosion does not yet prevail (Lambers et al., 2008;Mueller-Dombois & Boehmer, 2013).Moreover, both poorly developed, young soils from recent volcanic eruptions and highly weathered, older soils, which have been above sea level and have not been glaciated for millions of years, are poor in nutrients and can act as an ecological filter on functional properties of plants (Laliberté et al., 2013;Lambers et al., 2008).Hence, we expect species assemblages in early and late successional stages to be functionally less diverse relative to species richness owing to lower geological heterogeneity and nutrient depletion on very young and old sites.
In this study, we analyse how the functional diversity of endemic, non-endemic native and alien assemblages changes across ecological gradients, using a 500 m × 500 m gridded distribution dataset for the flora of the Canary Islands (Figure 1a-d).First, we analyse how the functional diversity of these assemblages, relative to species richness, changes with humidity.We test the prediction that environmental filtering results in an increase of functional diversity with increasing humidity for endemic and non-endemic native assemblages (Figure 1a).For alien assemblages, we expect the pattern to be the opposite, owing to the introduction of alien species to arid coastal environments and the subsequent directional ecological filtering (Figure 1a).Second, we analyse whether the mechanisms invoked by classical island biogeographical theory predict variation in functional diversity relative to species richness.Based on theory, we expect a positive relationship between functional diversity and geographical isolation for endemic assemblages and a negative relationship for non-endemic native and alien assemblages (Figure 1b).We also expect a hump-shaped relationship with topographic complexity for endemic and non-endemic native assemblages and a negative relationship for alien assemblages (Figure 1c).Finally, we expect a hump-shaped relationship between functional diversity and geological age for endemic, non-endemic native and alien assemblages (Figure 1d).Table 1).The high-elevation alpine ecosystems of La Palma and Tenerife are the most environmentally isolated, because the closest area with comparable climate is in the Atlas Mountains of Morocco, whereas the coastal climate across the archipelago is similar to the climate of the nearby north-western coast of Africa (Table 1).
Owing to high volcanic activity, each island is composed of older eroded terrain units and younger parts, leading to great variation F I G U R E 1 (a-d) Hypothesized and (e-h) observed relationships of standardized effect sizes of functional diversity (FD SES ) for endemic (green), non-endemic native (blue) and alien (violet) plant species assemblages, based on 500 m × 500 m grid cells (n = 3,065) across the Canary Islands, along humidity, geographical isolation, topographic complexity and geological age gradients.Functional diversity is based on four functional traits (plant height, leaf length, flower length and fruit length) of 347 endemic, 306 non-endemic native and 240 alien plant species.Lines show the trends of the models given in Table 2 and Figure 4 as La Palma and Tenerife (Table 1).Recent volcanic activity on El Hierro, La Palma and Tenerife further underlines the hotspot origin of the Canary Islands, where the islands have emerged successively and allow the observation of different phases of island ontogeny (Fernández-Palacios et al., 2011; Table 1).
The native flora of the Canary Islands is considered to be mostly of Mediterranean origin (Carine et al., 2010).Since the prehistoric human settlement, and especially since the arrival of European settlers in the 15th century, the Canary Islands have been subject to intense landscape alterations and introduction of alien plants (del Arco Aguilar et al., 2010;de Nascimento et al., 2020).Human pressure decreases with elevation, because large villages and agricultural settings are located mainly near the coast, and human population density becomes low above 1,000 m a.s.l.(Arévalo et al., 2005).
Currently, the Canary Island flora encompasses c. 2,000 species, of which roughly a third of the species are endemic, non-endemic native and alien species, respectively (Acebes Ginovés et al., 2009).

| Species occurrences and species distribution models
We collected occurrence data from Atlantis 3.3, an open-access database for all seed plant species, in a raster of 500 m × 500 m grid cells covering El Hierro, La Palma, La Gomera, Tenerife and Gran Canaria (www.biodiversi dadca narias.es/biota).Atlantis provides presence-only information, with better coverage for endemic species and flowering plants than other plant taxa (Steinbauer et al., 2016).Therefore, we also interpolated species occurrences using species distribution models (SDMs; Calabrese et al., 2014), which were parameterized following Irl et al. (2020) (see Supporting Information Appendix S1).To check whether using modelled species distributions created an artificial relationship between the predictor variables and functional diversity, we performed all statistical analyses separately with the modelled data and the original occurrence data.Furthermore, we excluded every grid cell that had less than half of the cell covered by land mass to account for low species occurrences based on grid cell area.

| Morphological trait data
We collected data on four functional morphological traits that reflect fundamental ecological strategies of plants and relate to the different axes of the leaf-height-seed scheme defined by Westoby (1998): plant height, leaf length, flower length and fruit length.
Morphological trait data of the respective plant species were collected from literature sources, measured on digitized plant specimens and extracted from descriptions of the species (a list of the data sources is given in Appendix 1; for further information, see Supporting Information Appendix S2).We collected data on maximum trait values from floras, because the expected maximum at maturity reflects responses of the fitness of species to environmental gradients (Violle et al., 2007) and has better data coverage throughout the literature.Recent work shows that trait data from plant identification books are comparable to measured trait data in the Canary Islands (Cutts et al., 2021).
For our analyses, we focused on the following traits.
1. Plant height is a crucial part of plant ecological strategy because it affects the ability of a plant to compete for light resources (Falster & Westoby, 2003) and is correlated with several life-history traits, such as seed mass and longevity (Moles & Leishman, 2008).Plant height is limited by water availability, because leaves need to be supplied with water without risking a xylem embolism (hydraulic limitation hypothesis; Moles et al., 2009;Ryan et al., 2006).Furthermore, plant height is more positively associated than seed mass with the seed dispersal distance of species (Thomson et al., 2011) and relates to the competitive ability of species (Gaudet & Keddy, 1988).
2. Leaf length is highly correlated with leaf area, independent of leaf shape (Cutts et al., 2021;Shi et al., 2019).Leaf size has important consequences for the energy and water balance of plants  Note: Humidity was quantified as the mean annual precipitation in relationship to mean annual potential evapotranspiration per grid cell.Geographical isolation was quantified as the distance of a grid cell to the nearest terrestrial area on the continent that has a mean annual temperature within 1°C.Topographic complexity per grid cell was estimated by calculating the ratio between three-and two-dimensional surface area.We derived the mean geological age per grid cell from a continuous digital geological map of Spain (scale 1:50,000; Bellido Mulas et al., 2020).Also given is the number of species analysed per island and the respective percentage of endemic, non-endemic native and alien plant species.(Parkhurst & Loucks, 1972).Smaller leaves are more water efficient, because they cool down more quickly and maintain lower leaf temperatures (Leigh et al., 2016).Larger leaves with a less efficient energy exchange capacity are more beneficial in humid habitats because they have higher photosynthetic capacities while being able to provide enough water for sufficient transpirational cooling (Parkhurst & Loucks, 1972).For stem-photosynthesizing succulents (e.g., Opuntia robusta), the ephemeral leaf length was collected.
3. Flower length is also important for the energy and water balance of plants, because flower petals can transpire significant amounts of water (Roddy, 2019;Roddy et al., 2016).Hence, flower length, which can affect pollinator attraction (Thompson, 2001), declines on average with water deficit (Paušič et al., 2019;Kuppler & Kotowska, 2021).For Poaceae, spikelet length was considered as the functional analogue to single flower length.For Asteraceae, ligule length was considered as the functional analogue to single flower length.
4. Fruit length is relevant for the dispersal strategy of plants.Fruit size is positively correlated with seed size (Muñoz et al., 2017;Rossetto & Kooyman, 2005;Wheelwright, 1993) and can inform about the dispersal probability and seedling survival of the plant (Greene & Johnson, 1993;Moles & Westoby, 2004;Thompson et al., 1993;Venable & Brown, 1988).In long-term isolated habitats, plants tend to have larger seeds to increase their establishment and persistence, but also to decrease their dispersal ability (Kavanagh & Burns, 2014;Rossetto & Kooyman, 2005).
We were able to collect complete trait and occurrence data for 893 species in total (347 endemics, 306 non-endemic natives and 240 aliens).Plant height, leaf length, flower length and fruit length were standardized by subtracting the column means from their corresponding columns and by dividing the (centred) columns by their standard deviations before functional diversity calculations.The four traits were only slightly correlated (Pearson's r < .33 in all cases).

| Environmental and biogeographical variables
For each grid cell, monthly mean temperature was used to calculate potential evapotranspiration according to the Thornthwaite equation (Thornthwaite, 1948).Subsequently, we calculated the humidity index after UNEP (1992).The humidity index is useful because it classifies the type of climate in relationship to water availability by considering temperature, precipitation, sunshine hours and relative humidity (UNEP, 1992).Hence, the humidity index is considered a reliable source of potential water availability at various scales (Zarch et al., 2015).Humidity index ranged approximately from 0.07 (arid) to 2.05 (humid) across the five islands (Table 1; see Supporting Information Appendix S3, Figure S3.1a).
We also calculated geographical isolation per grid cell based on the distance to climatically similar land mass (Weigelt & Kreft, 2013).
Specifically, geographical isolation was quantified as the distance of a grid cell to the nearest terrestrial area on the continent that had a mean annual temperature within 1°C, following Steinbauer et al. (2016).Geographical isolation of the island grid cells ranged from 196.5 to 885 km to the closest continental grid cell with similar environmental conditions (Table 1; see Supporting Information Appendix S3, Figure S3.1b).
We calculated topographic complexity per grid cell because it relates to the rate of elevational change in response to changes in location (Amatulli et al., 2018) and can therefore act as a surrogate for habitat heterogeneity (Irl et al., 2015).We estimated topographic complexity per grid cell by using a moving window approach that calculates the surface area for a cell based on slope information from a specified set of smaller grid cells (after Jenness, 2004).Topographic complexity ranged from 1 (flat) to 2.17 (high complexity; Table 1; see Supporting Information Appendix S3, Figure S3.1c).
Furthermore, we calculated geological age per grid cell because it represents a proxy for plant nutrient availability (Laliberté et al., 2013;Lambers et al., 2008), which is known to affect functional diversity (Lambers et al., 2011) For further analyses, we ln-transformed the topographic complexity index to approximate normality and subsequently centred and scaled all environmental variables, yielding estimates in standard deviation units per grid cell.After standardization, we calculated correlation coefficients (Pearson's r) between the environmental variables, in addition to elevation (see Supporting Information Appendix S3, Table S3.1).Humidity was highly correlated with geographical isolation (Pearson's r = .80,p < .001)and elevation (Pearson's r = .72,p < .001).

| Functional diversity of oceanic island floras
We calculated functional diversity using all traits to determine how ecological strategies vary between endemics, non-endemic natives and aliens (here also referred to as floristic groups) and to establish whether climate gradients and evolutionary processes can explain patterns of functional diversity across the Canary Islands.We used a hypervolume algorithm for calculating the overall functional diversity for each floristic group using the "hypervolume" R package (Blonder et al., 2014).We prefer this to other functional diversity metrics (e.g., functional richness) because it recognizes clusters or holes in occurrence datasets within trait space (Blonder et al., 2014).
Finally, we calculated the pairwise overlap between functional diversity values of floristic groups (2× shared volume/summed volume).Using the same methodological approach, we also calculated occurrence-based and SDM-derived functional diversity per grid cell for each floristic group separately.To ensure that we did not overor underestimate functional diversity, we excluded grid cells with <10 species (Blonder et al., 2014) for occurrence-based and SDMderived assemblage models.We constructed functional diversity by building a Gaussian kernel density estimate on an adaptive grid of 100 random points wrapping around each original data point.We used a fixed kernel density estimate bandwidth of 0.5 standard deviations to make functional diversity calculations comparable across analyses (Lamanna et al., 2014).We used a quantile threshold, which ensured that 95% of the estimated probability density was enclosed by the chosen boundary (Blonder et al., 2014).To ensure that we did not estimate hypervolumes into negative trait space, we calculated the intersection between a hypothetical box hypervolume that we defined by the range of our trait data and the calculated hypervolumes with the function hypervolume_set ("hypervolume" R package; Blonder et al., 2014).
To ensure that our results were not sensitive to species richness, we compared the functional diversity values of every floristic group with null-model expectations (for relationships between functional diversity and species richness, see Supporting Information Appendix S4).Therefore, we calculated values from 10 sets of randomized assemblages created with the "quasiswapcount" algorithm ("vegan" package; Oksanen et al., 2017), based on the species pool within each island (i.e., all species observed in any grid cell in each island).This algorithm retains row and column sums and thus constrains the species richness of grid cells.We computed the standardized effect size of functional diversity (FD SES ) as follows: FD SES = (FD − mean randomized FD)/SD of randomized FD.Values of FD SES lower than zero indicate functional clustering (FD lower than expected for a given species richness), whereas values of FD SES higher than zero suggest functional overdispersion (FD higher than expected for a given species richness), and values of FD SES approximately equal to zero indicate a random functional structure.

| Phylogenetic signal at the genus and assemblage levels
Given that the floristic groups could be structured phylogenetically (e.g., a high proportion of Crassulaceae in the endemic group), we tested for phylogenetic non-independence in our data.Ignoring a phylogenetic signal and treating species as independent could result in pseudoreplication (Felsenstein, 1985).Hence, we extracted phylogenies of the species in the dataset of a recently published super-tree of seed plants (Smith & Brown, 2018).Given that the phylogenetic tree contains many polytomies at the species level, we pruned the rooted tree to the genus level for each floristic group using the "ape" package in R (Paradis & Schliep, 2019; see Supporting Information Appendix S5, Figure S5.4).Subsequently, we measured Blomberg's K statistic of phylogenetic signal for each analysed trait within every floristic group using the "picante" package in R (Kembel et al., 2010).The K statistic is a measure of phylogenetic signal that compares the observed signal in a trait with the signal under a Brownian motion model of trait evolution on a phylogeny (Blomberg et al., 2003).Values of K closer to zero correspond to a random or convergent pattern of evolution, whereas values of K greater than one indicate strong phylogenetic signal and conservatism of traits (Kembel et al., 2010).
To analyse the influence of phylogenetic history on assemblage functional diversity, we calculated phylogenetic diversity for endemic, non-endemic native and alien assemblages based on occurrence and modelled data in the analysed grid cells.This approach allowed us to assess the degree of correlation between phylogenetic and functional diversity (termed phylogenetic signal at the metacommunity level sensu Pillar & Duarte, 2010).Phylogenetic diversity was calculated as Faith's PD, which is defined as the total branch length of a tree including all species in a grid cell ("picante" R package; Kembel et al., 2010).To test whether functional diversity was correlated with phylogenetic diversity, we computed correlation coefficients (Pearson's r) for each floristic group using occurrence-based and modelled data.A strong correlation is expected when assemblages that are more similar in terms of phylogenetic structure are also similar regarding their average trait values.

| Statistical analyses
We ran generalized linear regression models (GLMs) to test the relationship between FD SES and environmental variables for occurrence-based and modelled endemic, non-endemic native and alien assemblages across all grid cells.We tested for quadratic relationships of all explanatory variables, because traits do not necessarily change in a linear manner along environmental gradients.We ran an initial full model including all four environmental variables but excluding elevation owing to collinearity with humidity and geographical isolation.We ran additional models that excluded either humidity or geographical isolation owing to collinearity.Subsequently, we chose a model selection procedure, based on minimizing the Akaike information criterion (AIC; Burnham & Anderson, 2002).We performed AIC model selection using the function dredge in the R package MuMIn (Barton, 2022) to obtain the overall best model.Given that spatial autocorrelation was present in the model residuals analysed here, we performed spatial autoregressive (SAR) models using the same explanatory variables selected for the GLMs.We implemented error-dependence models with weighted neighbourhood structure because they accounted best for the spatial structure in the analysed dataset (Dormann et al., 2007).We selected a threshold distance of 500 m by examining correlograms.Spatial statistics were performed using the "spatialreg" package in R (Bivand et al., 2013).The explained variation of the models was quantified using Nagelkerke's R 2 .Spatial Moran's I correlograms for the response variable and for the GLM and SAR residuals are provided in the Supporting Information (Appendix S6, Figure S6.5).All analyses were performed using R v.3.6.2 (R Core Team, 2020).

| RE SULTS
Some alien species had trait values outside the range of traits of endemic and non-endemic native species and expanded the trait space of the Canary Islands within all trait dimensions.Around 27% of the functional trait space of alien species was shared by endemic species, and c. 28% of the functional trait space of aliens was shared by non-endemic native species.Functional diversity of endemic and non-endemic native species showed a great overlap in all trait dimensions (hypervolume overlap = 72%; Figure 3).
On a grid cell scale, occurrence-based FD SES of endemic and nonendemic assemblages had a unimodal relationship with humidity, whereas occurrence-based FD SES of alien assemblages decreased with increasing humidity (Table 2; Figures 1e and 4).With increasing geographical isolation, occurrence-based FD SES of endemic assemblages increased.Occurrence-based FD SES of non-endemic native assemblages had a non-significant hump-shaped relationship with geographical isolation, and occurrence-based FD SES of alien assemblages had a unimodal relationship with geographical isolation (Table 2; Figures 1f and 4).Occurrence-based FD SES of endemic assemblages decreased with topographic complexity, and non-endemic native assemblages showed a non-significant decrease with topographic complexity.Occurrence-based FD SES of alien assemblages had a non-significant U-shaped relationship with topographic complexity (Table 2; Figures 1g and 4).Geological age was not included in the most parsimonious model of endemic assemblages.With increasing geological age, occurrence-based FD SES of non-endemic native assemblages increased linearly, whereas occurrence-based FD SES of alien assemblages decreased non-significantly (Table 2; Figures 1h and 4).Hence, the functional diversity of endemic assemblages was highest on the more humid and isolated north-eastern slopes in less topographically complex habitats.Likewise, functional diversity of non-endemic native assemblages was highest in habitats with intermediate humidity and isolation in the older islands.In contrast, functional diversity of alien assemblages was highest at arid sites at the coasts across all islands.Humidity had the overall strongest explanatory power for functional diversity (Figure 4).
In the models excluding humidity, geographical isolation had a stronger explanatory power for functional diversity of endemic, nonendemic native and alien assemblages (see Supporting Information Appendix S6, Table S6.2; Figure S6.6).However, overall model fit according to AIC and explained variance was better for the models that included all environmental variables in comparison to models that accounted for collinearity (ΔAIC endemic = 5.9/87.3;ΔAIC nonendemic native = 29.7/41.8;ΔAIC alien = 6.3/7.9).
The FD SES patterns of the modelled data were broadly consistent with the occurrence-based FD SES models (see Supporting Information Appendix S6, Table S6 .3;Figures S6.7 and S6.8).
However, contrary to the occurrence-based data, modelled FD SES of alien assemblages showed a convex relationship with geographical isolation.This indicates that our results using species distribution models might have created an artificial relationship between functional diversity of alien assemblages and the analysed predictor variables.Therefore, modelled FD SES values were used only for graphical illustration of endemic and non-endemic assemblages (see Figure 2) and not discussed further in this study.
Spatial autoregressive models were selected over GLMs because they improved model fit consistently.According to the R 2 , accounting for spatial structure increased the explained variation between 41% (for alien assemblages) and 26% (for endemic assemblages; Table 2; see Supporting Information Appendix S6, Table S6.3 for modelled FD SES ).
Blomberg's K statistic of phylogenetic signal revealed that endemic and alien genera showed a significant phylogenetic signal for plant height (endemic: K = 2.57, p < .001;alien: K = 1.90, p < .001; see Supporting Information Appendix S6, Table S6.4).This indicates that within endemic and alien genera, close relatives were more likely to have a similar plant height than would be expected by chance.
There was a significant random (to intermediate) pattern in trait evo-

| DISCUSS ION
Our study demonstrates that the functional traits of endemic and non-endemic native plant species on the Canary Islands are very similar to each other, whereas alien species possess novel trait combinations, suggesting differential effects of environmental filtering and human introduction pathways on these different floristic groups.
On a 0.25 km 2 grid cell scale, the functional diversity of endemic assemblages is highest in humid and isolated conditions, and functional diversity of non-endemic native assemblages is highest in humid conditions.In contrast, functional diversity of alien assemblages is highest in dry environments, where most alien species have been introduced.These findings highlight the importance of functional traits for understanding the composition of island assemblages.

| Functional trait space of floristic groups
The high overlap of trait space between endemic and non-endemic native species suggests a packing of niche space rather than an exploitation of new functional strategies.This is consistent with studies that have shown that high species richness is associated with denser occupation of functional trait space, which can arise either through very fine morphological specializations or through overlap in resource use (Klopfer & MacArthur, 1961;Pigot et al., 2016).
However, there are also species that do not follow this pattern, such as Phoenix canariensis, a Canary Island endemic palm tree.It clearly extends the occupied trait space through its large plant height and long fronds.Furthermore, we show that alien species both expanded the overall trait space and overlapped in trait composition with endemic and non-endemic native species.This suggests that alien species were able to exploit novel regions of functional niche space (MacArthur, 1965) and to use similar resources to native species (Klopfer & MacArthur, 1961).However, given that the observed

| Functional diversity of endemic and nonendemic native plant assemblages
Our results indicate that humidity and geographical isolation might be important factors driving functional diversity patterns in native assemblages on the Canary Islands.However, owing to their high collinearity with each other and with elevation, we cannot disentangle their effects on functional diversity clearly.On the Canary Islands, humidity is highest at intermediate elevation on more isolated islands (i.e., in the laurel forest on La Palma) and lowest at the coast on islands closer to the continent (i.e., in the succulent shrub on Gran Canaria).We found overlap in explained variation between humidity and geographical isolation, especially in endemic assemblages.Hence, the functional clustering in coastal habitats could be attributed to environmental filtering via humidity.According to the "physiological tolerance hypothesis", environments with higher water availability permit a wider range of functional strategies (Spasojevic et al., 2014), whereas species that occur in habitats with extreme environmental conditions often tend to have specific and similar adaptations (Cornwell & Ackerly, 2009).
However, competition could have led to functional divergence, hence increased functional diversity, in humid habitats as well (MacArthur & Levins, 1967;Spasojevic & Suding, 2012).Taller plants with larger leaves, flowers and fruits have a higher competitive vigour and could be able to increase partitioning within the assemblages.Moreover, functional diversity could even be facilitated by tall, large-leaved and large-flowered species through a reduction of evaporative losses for smaller, functionally distinct species through light interception (Callaway et al., 2002;Spasojevic & Suding, 2012).
Functional diversity increased with increasing geographical isolation in endemic assemblages and had a hump-shaped relationship with non-endemic native assemblages.In endemic plant species, isolation can select against dispersal ability by increasing fruit and seed size to reduce dispersal into the sea (Kavanagh & Burns, 2014;Rossetto & Kooyman, 2005).Larger seeds have increased nutrient reserves and thus a highly competitive vigour during establishment (Moles & Westoby, 2004;Thompson et al., 1993).To decrease competitive interaction with large-seeded species, endemic species might diverge in trait values (MacArthur & Levins, 1967), leading to greater functional diversity with greater isolation.Moreover, isolation can limit gene flow between species, which drives ecological opportunity and might foster adaptive speciation in endemic summit scrub and rock communities of the Canary Islands (Fernández-Palacios, Otto, et al., 2021).Responses of non-endemic native functional diversity to isolation could be influenced by the dispersal mode of species.Species with wind and unassisted dispersal have smaller diaspores than species with animal dispersers (Leishmann et al., 1995).Moreover, dispersal limitation via plant height could have driven the decrease of functional diversity in highly isolated habitats in non-endemic native species (Thomson et al., 2011).
Hence, establishment of functionally diverse, non-endemic native assemblages at intermediate isolation appears to have been dependent on disharmonic diaspore size and plant height.However, our results need to be interpreted with caution, because our metric of isolation considers the distance to climatically suitable mainland and ignores other factors, such as wind and ocean currents, in addition to climatically similar areas on other islands, as possible modifiers to these distances (Muñoz et al., 2004;Price, 2004).
Moreover, studies on edaphic islands have shown that target effect (i.e., indicating the increased probability of a larger island to be colonized than a smaller island given the same isolation) might play a key role in determining plant diversity patterns (Conti et al., 2022;Mendez-Castro et al., 2021).Ignoring these additional isolation metrics might lead to an under-or overestimation of geographical isolation and its relationship with functional diversity.Nonetheless, despite its limitations, climatic similarity has previously been shown to be an adequate measure of geographical isolation (Weigelt & Kreft, 2013).Note: The standardized effect size of functional diversity (FD SES ) is based on four functional traits (plant height, leaf length, flower length and fruit length) and was calculated for grid cells with occurrence-based data (n = 3,065).The percentage of total deviance explained (% dev.), Akaike information criterion (AIC) and Moran's I are given.Model coefficients of SAR models are given in Figure 4.
TA B L E 2 Explained variation of generalized linear models (GLMs) and spatial autoregressive (SAR) models analysing the combined effect of humidity, geographical isolation, topographic complexity and geological age on the standardized effect size of functional diversity for endemic, non-endemic native and alien seed plant assemblages based on 500 m × 500 m grid cells across the Canary Islands Functional diversity of endemic and non-endemic assemblages decreased with topographic complexity.The observed pattern could be an indication of a trade-off between topographic complexity and the amount of area available for species (Allouche et al., 2012).
With very high topographic complexity, the amount of effective area available for species should decrease, thereby increasing the likelihood of stochastic extinctions in species with a narrow niche (Allouche et al., 2012).The area-heterogeneity trade-off seems to be more important for endemic species than for non-endemic species, suggesting that non-endemic species might have wider niches in topographically complex habitats.Moreover, very high topographic complexity facilitates the isolation of small populations that might diverge to new endemic allopatric species, augmenting the number of species but not the functional diversity (Badgley et al., 2017).
However, we must be careful with the interpretation of these results.Species that occur in very steep sites are rarely reported; therefore, information on species assemblages in topographically complex sites is limited and could possibly over-or underestimate the functional diversity of plant assemblages.
Our results indicate that geological age has no relationship with the functional diversity of endemic assemblages.This indicates that soil age does not act as an ecological filter on the functional diversity of endemic assemblages.Surprisingly, non-endemic native species assemblages seem to be functionally different from endemic assemblages in nutrient-deficient sites.This suggests that non-endemic native species have functional properties to grow on older volcanic sites, where destructive processes, which cause nutrient depletion and soil homogenisation, dominate (Laliberté et al., 2013;Lambers et al., 2008;Mueller-Dombois & Boehmer, 2013).However, the explained variation of the models is low, and the results should be interpreted with caution.Furthermore, our measurement of geological age was approximate and might not be precise enough to reflect mechanisms of soil development on plant functional diversity.

| Functional diversity of alien plant assemblages
Alien assemblages show a decrease of functional diversity with increasing humidity, geographical isolation and geological age, in addition to an increase of functional diversity with increasing topographic complexity.This mostly contrasts with the relationships of endemic and non-endemic native assemblages.This could be attributable to a direct negative effect of native functional diversity through biotic resistance.If native assemblages are functionally diverse, it is more likely that they have overlapping resource requirements with alien species and are therefore able to outcompete them (Funk et al., 2018; but see Galland et al., 2019).For instance, the laurel forest is thought to be ecologically resistant to invasion through light interception in the understorey, highly competitive pressure for nutrients from tree roots and the effect of allelopathy from laurel leaf-litter on alien species (Bermúdez et al., 2007).Nevertheless, there is evidence that even this stable ecosystem can be invaded by an alien tree (Devkota et al., 2020).We also expect the analysed predictor variables to be negatively correlated with anthropogenic activity (for patterns of alien species richness on Tenerife and La Palma, see Irl et al., 2021).Hence, the observed relationship for alien species might also be influenced by a subsequent directional filtering proceeding from sources of anthropogenic introduction (Irl et al., 2021).
This indicates that, so far, only a functional subset of alien species with wide environmental tolerances has been able to invade habitats separated from the source pool by a steep environmental gradient.

| Spatial structure and phylogenetic nonindependence
Although we find relationships with environment and isolation, we need to consider that spatial autocorrelation and phylogenetic non-independence have influenced the observed patterns.Spatial structure accounted for a large part of the explained variation in some of the regression models.For alien assemblages, this could indicate a non-equilibrium distribution attributable to their recent island-specific introductions.Furthermore, the analysed environmental variables are spatially aggregated on the Canary Islands.For example, very humid environments exist on La Palma, but we do not find these conditions on other islands.Therefore, we can analyse the relationship between functional diversity and highly humid environments only within La Palma, and the relationship might be influenced by the phylogenetic non-independence of (single-island) endemic species.
At the assemblage level, we could detect a significant but low to moderate correlation between functional and phylogenetic diversity for the three floristic groups.These results imply that the differences in functional diversity between the floristic groups reflect, at least in part, trait conservatism (Blomberg et al., 2003).However, application of a correction for non-independence has been shown to lead to an unintended "over-correction" (Ricklefs & Starck, 1996; but see Rohle, 2006) and is likely not to be feasible using a pruned tree with polytomies, as is the case for the tree used in the present study.We conclude that although our results might be influenced, in part, by phylogenetic non-independence, they nevertheless point to important relationships between functional diversity and the environment that inform us about the drivers of species assemblages on oceanic islands.

| Conclusion
Our study indicates that endemic species do not differ substantially, in a functional sense, from the more widespread non-endemic native species group.Both groups have adapted to the island environment over long time periods, although much longer for endemic than for non-endemic native species.However, the functional diversity of endemic and non-endemic native plant assemblages is distributed differently along environmental and island biogeographical gradients, indicating the importance of climate and ecological opportunity for the evolution of plant form and function.In contrast, alien species tend to be dissimilar, in a functional sense, to endemic and non-endemic native species, and their assemblages show great functional diversity in semi-arid and less isolated ecosystems, which are the main gateway for anthropogenic introduction.Furthermore, the strong connection between the functional diversity of both endemic and non-endemic native assemblages and climatic conditions suggests that current ongoing climate change might alter patterns of plant functional diversity on the Canary Islands.However, we are aware that the choice of traits and environment are highly relevant for the results of a trait-based study (Bernard-Verdier et al., 2012;Carvajal et al., 2019;Ottaviani et al., 2020).Our study did not analyse leaf or flower economics data (e.g., ratio of leaf or flower area to dry mass), regenerative traits (e.g., seed mass) or below-ground traits (e.g., specific root length), which should be considered in future studies (if the data become available) to provide a better link between trait functions and hypotheses concerning water-use efficiency.In particular, functional diversity patterns in different types growth forms should be analysed in future studies, because growth form can have important implications for community assembly across climatic and biogeographical gradients (Irl et al., 2020;Schrader et al., 2020).Moreover, further analyses with entire floras from other islands, archipelagos or mountainous regions on continents, and with other taxonomic groups, are needed to test the wider generality of our results in the context of functional island biogeography and to understand the importance of functional diversity for the establishment and diversification of native and alien plant species in space and time.
Study areaWe tested our hypotheses using the Flora of the Canary Islands (excluding Fuerteventura and Lanzarote owing to underrepresented species occurrence data from very limited sampling; Figure2).The Canary Islands are an active volcanic archipelago characterized by a subtropical-mediterranean climate regime(del Arco Aguilar et al., 2010).The islands are under the influence of the NE trade winds, and rain shadow effects cause humidity to differ greatly between northern and southern parts of the islands, with the north-eastern flank of the islands receiving the highest precipitation.Within islands, humidity varies greatly, from a dry and warm coast, through moderately temperate mountainous zones up to dry and cool conditions at high elevations(del Arco Aguilar et al., 2010; ).The mechanisms on which we based our hypotheses are indicated on the left-hand side of the figure.Triangles indicate either a hypothesized increase or decrease of FD SES along the respective environmental gradient.Circles indicate a hypothesized unimodal relationship of FD SES along the respective environmental gradient F I G U R E 2 Map of the spatial distribution of modelled standardized effect size of functional diversity (FD SES ) values per 500 m × 500 m across the five studied Canary Islands for (a) endemic and (b) non-endemic native assemblages.Source of general map: Esri, DigitalGlobe, GeoEye, Earthstar Geographics, CNES/Airbus DS, USDA, USGS, AeroGRID, IGN and the GIS User Community in topographic complexity, especially on emerging islands, such

F I G U R E 3
Estimated four-dimensional hypervolumes for endemic (green; n = 347), non-endemic native (blue; n = 306) and alien (violet; n = 240) seed plant species on the Canary Islands.The coloured points represent the different plant species, and the coloured lines reflect the areas filled by random points sampled from the inferred hypervolume.Endemic species overlap non-endemic native species by 72%.Endemic species overlap alien species by 27%, and non-endemic species overlap alien species by 28% c. 43% of the Canary Island flora, they might be dependent on the species considered.

F
Multipredictor spatial autoregressive models of humidity (HUM), geographical isolation (ISO), topographic complexity index (TCI) and geological age (GA) on the occurrence-based standardized effect size of functional diversity (FD SES ) of endemic (n = 347 species), non-endemic native (n = 306 species) and alien (n = 240 species) plant assemblages on the Canary Islands.Functional diversity is based on four functional traits (plant height, leaf length, flower length and fruit length) and was calculated for occurrence-based data (n = 3,065) based on 500 m × 500 m grid cells.Shown are the respective best models according to a model comparison approach.Standardized estimates are provided for each predictor.Error bars represent confidence intervals of the coefficient estimates.Asterisks denote statistical significance (***p < .001;**p < .01;*p < .05)