Factors explaining alien plant invasion success in a tropical ecosystem differ at each stage of invasion

Authors


*Correspondence author. E-mail: wda@ceh.ac.uk

Summary

  • 1Understanding why some alien plant species become invasive when others fail is a fundamental goal in invasion ecology. We used detailed historical planting records of alien plant species introduced to Amani Botanical Garden, Tanzania and contemporary surveys of their invasion status to assess the relative ability of phylogeny, propagule pressure, residence time, plant traits and other factors to explain the success of alien plant species at different stages of the invasion process.
  • 2Species with native ranges centred in the tropics and with larger seeds were more likely to regenerate, whereas naturalization success was explained by longer residence time, faster growth rate, fewer seeds per fruit, smaller seed mass and shade tolerance.
  • 3Naturalized species spreading greater distances from original plantings tended to have more seeds per fruit, whereas species dispersed by canopy-feeding animals and with native ranges centred on the tropics tended to have spread more widely in the botanical garden. Species dispersed by canopy-feeding animals and with greater seed mass were more likely to be established in closed forest.
  • 4Phylogeny alone made a relatively minor contribution to the explanatory power of statistical models, but a greater proportion of variation in spread within the botanical garden and in forest establishment was explained by phylogeny alone than for other models. Phylogeny jointly with variables also explained a greater proportion of variation in forest establishment than in other models. Phylogenetic correction weakened the importance of dispersal syndrome in explaining compartmental spread, seed mass in the forest establishment model, and all factors except for growth rate and residence time in the naturalization model.
  • 5Synthesis. This study demonstrates that it matters considerably how invasive species are defined when trying to understand the relative ability of multiple variables to explain invasion success. By disentangling different invasion stages and using relatively objective criteria to assess species status, this study highlights that relatively simple models can help to explain why some alien plants are able to naturalize, spread and even establish in closed tropical forests.

Introduction

Understanding why some alien plant species become invasive while others fail has long been a central research theme in invasion ecology (Rejmánek et al. 2005). Attempts to discriminate between invasive and non-invasive alien species often involve comparative analyses of native geography and plant traits (Herron et al. 2007; Kuster et al. 2008; Milbau & Stout 2008; Sol et al. 2008). Although some generalizations have begun to emerge from such studies (Pyšek & Richardson 2007), the explanatory power of comparative analyses may be limited in cases where propagule pressure, residence time and phylogeny are not taken into account (Hayes & Barry 2008). Intuitively, greater propagule pressure should increase the probability that a species will overcome stochastic survivorship and Allee effects known to impact small founder populations (Cassey et al. 2004; Lockwood et al. 2005). Similarly, the time since a species was first introduced into a region often corresponds well with its probability of becoming naturalized (Rejmánek 2000; Wilson et al. 2007; Lambdon et al. 2008). However, data on these two aspects of introduction history are rarely available for most species introduced to a particular location. Furthermore, although taxonomic patterns among invasive plant species are recognized (e.g. Daehler 1998; Pysek 1998), there is an increasing awareness that the phylogenetic structure of the species pool itself needs to be taken into account in comparative studies (Lambdon 2008; Procheş et al. 2008). Although recent comparative studies have attempted to address phylogeny simultaneously with ecological traits and residence time (Kuster et al. 2008; Milbau & Stout 2008; Gasso et al. 2009), no study to date has additionally considered propagule pressure. Finally, because the relative importance of these factors may vary at different stages of the transition from initial plant introduction to subsequent invasion (Lloret et al. 2005; Dietz & Edwards 2006; Diez et al. 2008), analyses should examine the regeneration, naturalization and spread stages separately since there is no reason to expect factors to be of equal importance at all stages.

The existence of detailed records of planting history for Amani Botanical Garden (hereafter ABG) in Tanzania offers an opportunity to assess the relative importance of a potential surrogate for propagule pressure, residence time and plant traits in explaining alien plant invasion success, whilst accounting for phylogeny. For a significant subset of extant species introduced that includes some of the most problematic woody weeds in the tropics, the following key questions were addressed:

  • 1What is the relative importance of propagule pressure, residence time, phylogeny, plant traits and other factors in the transitions from survival to regeneration and regeneration to naturalization?
  • 2Once a species is naturalized, what factors are associated with establishment in anthropogenically undisturbed rain forest habitats?
  • 3Once a species is naturalized, what is the relative importance of propagule pressure, residence time, phylogeny and plant traits in explaining the degree of species spread?

Methods

study site

The ABG (5°05′30″ S, 38°38′10″ E) is situated in the lowland and submontane rainforests of the East Usambara Mountains in northeast Tanzania. The ABG was formally established in 1902 (Iversen 1991), and more than 500 species (mostly woody) were planted over a 28-year period from 1902 to 1930 in a series of trial plantations, with the majority of species (80%) being introduced between 1902 and 1914 (Greenway 1934). The ABG spreads over some 300 hectares, ranging in altitude from 400 to1100 m a.s.l. (Iversen 1991). It originally consisted of 20 plantation blocks, divided into 141 compartments, varying in shape and size from 0.1 to 7 hectares, and containing almost 2000 plots planted with individual species (Greenway 1934; Sandy et al. 1997). Today, only one-third of the species (214) remain (partly as a result of compartments having been cleared for human settlement) and a number of these are now known to have spread into the surrounding tropical rain forest of Amani Nature Reserve, a site of high conservation importance (Newmark 2002).

designating invasion status

Following detailed field surveys in ABG (Dawson et al. 2008), extant species with data available for all explanatory variables (Table 1) were classified into one of three categories:

Table 1.  Explanatory variables and their measures considered in the study (see supporting information for sources)
VariableMeasure/LevelsCriteria
Propagule pressureLog (number of plots)Number of plots planted
Minimum distance between species plot and forest edge (for forest establishment only)√ (metres)Smallest distance between a plot planted and forest edge, measured according to GPS co-ordinates and map locations of plots
Residence time2005 minus year first planted (years)Earliest year a species recorded as planted was used
Climate in native rangeSqrt (midpoint native latitudinal range)Native latitudinal ranges extracted from data bases, or from species range descriptions
Dispersal syndromeBird/primate; other animal; wind; other (autochory, hydrochory)Species recorded as having dispersal syndrome in native or introduced range, or obvious morphological adaptations (e.g. presence of wings or parachutes for wind dispersal); Arboreal includes canopy-feeding animals (frugivorous birds and primates, which are present at ABG)
Vegetative reproductionYes; NoSpecies recorded as being able to spread vegetatively, without human intervention (e.g. not layering or cuttings)
Growth rateSlowerSlower = growing < 1 m year−1 or described as slow/moderate growing
FasterFaster = growing > 1 m year−1 or described as fast growing
Shade toleranceYes; No 
(i) At any stage of life cycle (i) Shade tolerance at any stage of life cycle described in information sources, including ability to grow under canopies.
(ii) Beyond seedling stage (For forest establishment only) (ii) Shade tolerance when adult/reproductive described in information sources, including ability to grow as crops in understorey
Seed size< 100 mg; > 100 mgDry or air-dry seed mass; average of congeneric masses taken when species seed mass unavailable
Number of seeds per fruit< 10; > 10Midpoint taken of a range of number of seeds per fruit given; images of fruits/direct observations used in absence of fruit descriptions
Breeding systemDioecious; Monoecious/ HermaphroditeDioecious includes species that have hermaphroditic/monoecious flowers, but have sterile sexual organs for one sex.
Presence/absence of native congenersPresent; absentPresent = At least 1 species in the native East Usambara flora in the same genus as the alien species
Number of native confamilial speciesLog (number + 1)Number of species in the native East Usambara flora that are in the same family as the alien species
  • 1Surviving (n = 74). Original plants present, but seedlings or saplings were not found. Although these species showed evidence of having the capacity to reproduce in the form of flowering and fruiting bodies, no seedlings or saplings were observed either beneath original canopies or elsewhere. These species are either unable to produce viable seeds, or germination of viable seeds is prevented under the environmental conditions prevalent at this site
  • 2Regenerating (n = 24). Producing seedlings and saplings, or vegetative shoots, but no new adults have been recruited (i.e. the species has the capacity to produce viable propagules and to replace itself, but has not yet naturalized)
  • 3Naturalized (n = 44). The species has recruited new adults into the population that are capable of reproducing themselves. Persistence of the species is now less dependent on propagules produced by the originally planted individuals
  • 4Forest Establishment (n = 13). A subset of naturalized species with adults found > 150 m away from forest edges adjacent to ABG, in semi-natural or natural forest identified as having late successional tree species at high density, a developing or developed subcanopy and understorey, and minimal signs of human disturbance.

These four categories represent different invasion stages and are mutually exclusive except for species establishing in forest (4). The categories of species correspond well with invasion history elsewhere and weed risk assessment scores of invasion risk (Dawson et al. 2009). Using these survey data, we were able to calculate two metrics of spread for naturalized and spreading species: the number of newly occupied compartments and the maximum linear distance a species has spread to a newly occupied plot from the nearest originally planted plot. The distance metric was calculated to the nearest 10 m using a combination of accurate historical maps (Greenway 1934) and a global positioning system (GPS) device (GPSMap 76CSx, Garmin Ltd., Kansas, USA).

explanatory variables

The likelihood of a species successfully transiting each of the three invasion stages was explored in relation to seven traits, two components of introduction history and three other factors. Five traits were believed a priori to be important in plant establishment and survival: vegetative reproduction, growth rate, seed size, shade tolerance and breeding system (Table 1). Two further traits, dispersal syndrome and number of seeds per fruit, were assumed to correlate with dispersal success. Additional factors included the midpoint of the native latitudinal range as a measure of climate match, and both the occurrence of native congeners and number of native confamilial species in the East Usambara Mountains, to address the theory that likelihood of invasion is influenced by taxonomic relatedness to the native flora (Strauss et al. 2006; Diez et al. 2008). The number of plots originally planted and the date of planting were used as measures of propagule pressure and residence time, respectively, and were extracted from a historical survey of ABG (Greenway 1934; Table 1). Two additional variables were included in the analysis of spread into closed forest: the minimum distance from a species plot to the forest edge, and shade tolerance of saplings and/or adults (i.e. after a seedling is established, can it survive as a sapling and/or an adult under shade?) (Table 1). All other variables were considered at each stage of invasion. Seed size was recorded as seed mass (mg dry weight or air-dry weight), and was largely obtained from the electronic Plant Information Centre (Royal Botanic Gardens, Kew 2008) and primary literature sources. When seed mass for species was unavailable, a mean seed mass was taken of congeneric species as an estimate. The midpoint of native latitudinal range was obtained from actual ranges in the Forestry Compendium (CAB International 2005), or from ranges derived from descriptions of geographical distribution recorded in various sources. The minimum distance between a planted plot and a forest edge (to nearest 10 m) was measured using a GPS and the historical maps of ABG. All other variables were obtained from primary literature, online and CD-Rom data bases and species fact sheets.

analysis

The explanatory variables formed the basis of binomial generalized linear models (GLMs) that were used to model the likelihood of species progressing through two transition stages: regeneration (distinguishing regenerating from surviving species) and naturalization (distinguishing naturalized from only regenerating species). Species were scored as either succeeding (1) or failing (0) to make the transition. These explanatory variables were also used to model the number of compartments colonized by naturalized species using a GLM with a quasi-Poisson error distribution. A linear model was also performed with linear distance spread per species (square-root transformed) as the response variable. A binomial GLM was used to model successful establishment in closed forest by naturalized species, forming a third stage of invasion. Variables retained in the minimum adequate binomial models were identified by non-sequential backwards elimination using Akaike's Information Criterion and the likelihood ratio test. The minimum models for linear and compartmental spread were obtained using F- and quasi-F tests, respectively. Once minimum models were obtained, we checked for the presence of significant interactions between retained variables. Odds ratios (the ratio of the probability that the alien is successful to the probability of failure) were calculated for significant variables retained in binomial GLMs, which can take values of 0 to infinity; values < 1 represent a decrease in success probability and those > 1 represent an increase in success probability per unit increase of the variable (Milbau & Stout 2008). Greater departure from 1 indicates a stronger relationship between the variable and the response (Milbau & Stout 2008).

The variation partitioning procedure was used to calculate the percentage of deviance explained by variables alone, phylogeny alone, and jointly by phylogeny and variables (Desdevises et al. 2003). Initially a phylogenetic distance matrix was constructed from a phylogenetic tree obtained from the Phylomatic website (Webb & Donoghue 2007) for the species pool at each transition, and for naturalized species. Adjusted branch lengths (millions of years) were included in the phylogenies, which are obtained by calibrating the tree topology against the root node and other branch lengths with known fixed age estimates (Wilkstrom et al. 2001; Webb & Donoghue 2007). Although there is some debate regarding the utility of including estimated branch lengths in phylogenies (Strauss et al. 2006; Diez et al. 2008), they have been included in this study to account for the strongly divergent and phylogenetically distant clades present in the data sets (e.g. gymnosperms and Arecales (monocotyledonous) versus dicotyledonous angiosperms). Knowledge of plant evolutionary history remains incomplete and no claim is made that the estimates of evolutionary distances provided by Phylomatic are accurate (Webb & Donoghue 2007). However, their inclusion in working phylogenies should constitute a more realistic representation of the true phylogeny than if all branch lengths were set to 1 (Lambdon 2008).

The phylogenetic distance matrices were subjected to classical multidimensional scaling analysis, and the first 10 axes describing variation in phylogenetic structure were extracted. These axes were then used as explanatory variables in a second set of models with species success at the corresponding invasion stage as the response variable. Backwards selection was then used to select the subset of phylogenetic axes best explaining the response (P < 0.05 for the retained axes from the likelihood ratio, F and quasi-F tests for binomial, linear and quasi-Poisson models respectively). A third set of models was constructed to correct for phylogeny. In this set the variables entered into the model only included the phylogeny axes that significantly explained species success and the significant variables from the minimum adequate models. Finally, variation partitioning was performed using the explained deviances or adjusted R2 of these three sets of models algebraically as follows: a = abc – bc, c = abc – ab, and b = (ab + bc) – abc, where a = the proportion of variation explained by variables alone, c = the proportion of variation explained by phylogeny alone, b = variation explained jointly by variables and phylogeny, ab = explained deviance from the minimum variable model, bc = explained deviance from the minimum phylogeny model, and abc = explained deviance from the model combining variables and phylogeny axes (Desdevises et al. 2003). To determine the predictive power of binomial GLM models, cross-validation was not used, as removal of species from a data set during the procedure would inevitably change the phylogenetic structure. Instead, minimum adequate models were refitted to corresponding species data sets, and we calculated sensitivity and specificity of model predictions (Altman & Bland 1994), with a cut-off probability of 0.5 demarcating predicted species success from predicted failure (Williamson 2006). Cohen's κ statistic was calculated with the same cut-off probability (Landis & Koch 1977) to summarize goodness-of-fit. The κ statistic can range from +1 (perfect agreement) to −1 (complete disagreement) with 0 indicating no agreement. For a positive κ value to be significantly greater than that expected by chance, the 95% confidence interval should not include 0, and the κ value should ideally be above 0.5.

Results

regeneration and naturalization

Before phylogenetic correction, only the midpoint of a species’ native latitudinal range and its seed mass significantly explained regeneration (22% deviance explained) and the odds ratios indicated that seed mass was more important (Table 2). However, model sensitivity was low at 0.21 (i.e. only 21% of regenerating species were correctly predicted as regenerating), but specificity was high at 0.89 (89% of species not regenerating were correctly predicted). There was no significant interaction between midpoint of native range and seed mass. Phylogenetic correction did not alter the significance of the variables included in the minimal adequate model, and did not greatly improve specificity or sensitivity or help explain much more variation (23% of total deviance explained; Table 2). Variation partitioning revealed that variables alone and phylogeny jointly with variables were responsible for most of the explained deviance (Fig. 1). κ suggested that the model was not able to distinguish regenerating from non-regenerating species above the level expected by chance, whether or not phylogenetic correction was applied (Table 2).

Table 2.  Coefficients of significant variables and prediction statistics of minimum adequate models explaining successful regeneration and naturalization before and after phylogenetic correction. Figures in parentheses represent 95% confidence intervals of κ statistic; OR = odds ratio
VariableBefore phylogenetic correctionAfter phylogenetic correction
EstimateSEZPOREstimateSEZPOR
Regeneration
 Intercept−0.7200.649−1.1090.267 −0.8480.714−1.1880.235 
 Midpoint of native latitudinal range−0.0920.033−2.7900.0050.912−0.0880.368−2.3960.017 0.916
 Seed mass (> 100 mg)1.4110.5922.3850.0174.1001.4130.6022.3460.019 4.108
 Axis 1     −0.0020.002−0.7860.432 
 Axis 5     −0.0010.007−0.2120.832 
 Model predictionSensitivity = 0.21; Specificity = 0.89; Sensitivity = 0.29; Specificity = 0.91;
κ = 0.119 (−0.089–0.326)κ = 0.229 (0.012–0.447)
Naturalization
 Intercept−8.5463.955−2.1610.031 −10.4454.287−2.4360.014 
 Residence time0.0830.0392.1260.0341.0860.0970.0432.2880.022 1.102
 Growth rate (faster)3.1771.0802.9410.00323.9743.5341.2672.7900.00534.261
 Shade tolerant2.3281.0932.1290.03310.2571.6161.2161.3290.184 
 > 10 seeds per fruit −1.9701.031−1.9110.0550.139−2.0251.210−1.6740.094 
 Seed mass (> 100 mg)−1.7861.013−1.7620.0780.168−0.6231.156−0.5390.590 
 Axis 2     −0.0400.020−2.0490.040 
 Axis 3     0.0350.0162.1950.028 
 Model predictionSensitivity = 0.94; Specificity = 0.58Sensitivity = 0.90; Specificity = 0.75
κ = 0.539 (0.321–0.758)κ = 0.663 (0.463–0.863)
Figure 1.

Summary of the transition of species between invasion stages and corresponding analyses in the study. Dashed lines separate species in the mutually exclusive surviving, regenerating and naturalized categories, and the subset of naturalized species establishing in forest. The connected boxes summarize the variables significantly distinguishing between species at each transition, and the direction of their effects on likelihood of species ‘success’. The significant variables (and direction of effects) explaining compartmental and linear spread of naturalized species are also shown in boxes. The percentages of total variation explained by variables, phylogeny, and by variables and phylogeny jointly are shown for each analysis.

The minimum adequate model for naturalization retained (in order of importance according to odds ratios) growth rate, shade tolerance, number of seeds per fruit, seed mass and residence time (32.5% deviance explained; Table 2). There were no significant interactions between variables. Sensitivity and κ were much higher in the naturalization model than in the regeneration model (Table 2). Phylogenetic correction changed the significance of variables in the naturalization model, such that only residence time and growth rate remained significant (Table 2). In addition, specificity increased after phylogenetic correction, whilst sensitivity and κ did not (Table 2). The model including phylogeny explained 43.1% of total deviance, and variation partitioning revealed that variables alone were responsible for the majority of explained deviance (Fig. 1). Phylogeny alone was responsible for a greater proportion of explained deviance in the naturalization model than in the regeneration model (Fig. 1).

establishment in forest

Species with larger seeds and seeds dispersed by birds and/or primates had a greater likelihood of establishing in forest than other naturalized species (31.9% deviance explained), and the odds ratios suggested that dispersal was more strongly related to forest establishment success (Table 3). There was no significant interaction between dispersal syndrome and seed mass. Phylogenetic correction weakened the importance of seed mass and increased sensitivity, but did not increase specificity and κ (Table 3). The model including phylogeny explained 43.5% of total deviance, and variation partitioning revealed that only a small majority was explained by traits alone; a greater proportion was explained by phylogeny and jointly by phylogeny and variables than in regeneration and naturalization models (Fig. 1).

Table 3.  Coefficients from significant variables of minimum adequate models explaining forest establishment (binomial), colonization of new compartments (quasi−Poisson) and linear distance spread (linear model), before and after phylogenetic correction. SE = standard error; figures in parentheses represent 95% confidence intervals of κ statistic; OR = odds ratio
 Before phylogenetic correctionAfter phylogenetic correction
EstimateSEZPOREstimateSEZPOR
Forest establishment
 Intercept−3.6011.436−2.5060.012 −4.5671.993−2.2960.022 
 Seed mass > 100 mg1.8090.9471.9100.056 6.1041.5951.0841.4710.141 
 Dispersal
  Other−16.3581935.7−0.0080.993 −16.3323166.42.0590.996 
  Bird/Primate2.8531.2832.2250.02617.3403.3081.6071.3290.04027.330
  Wind2.0761.5771.3170.188 2.6652.0041.4710.184 
 Axis 6     −0.0600.032−1.8830.060 
 Axis 7     0.0470.0301.5690.117 
 Model predictionSensitivity = 0.62; Specificity = 0.90Sensitivity = 0.77; Specificity = 0.90
Kappa = 0.543 (0.269–0.82)Kappa = 0.672 (0.431–0.913)
Number of compartments  t    t  
 Intercept2.1120.5933.562< 0.001 2.2300.5803.847< 0.001 
 √(midpoint of native latitudinal range)−0.2790.099−2.8110.008 −0.2550.097−2.6250.012 
 Dispersal
  Other−0.6400.850−0.7530.456 −0.8120.826−0.9840.331 
  Arboreal1.1090.6011.8470.072 0.6590.6161.0710.291 
  Wind1.0670.6331.6850.099 1.0250.6141.6710.103 
 Axis 7     0.0110.0052.2770.028 
Linear spread  t    t  
 Intercept17.7351.42912.409< 0.001 18.0011.40612.799< 0.001 
 > 10 seeds per fruit11.1402.7374.071< 0.001 10.1632.7373.713< 0.001 
 Axis     0.0590.0351.7060.096 

spread

The number of compartments colonized by naturalized species was significantly related to the midpoint of native latitudinal range and dispersal syndrome (36.7% deviance explained; Table 3), and there was no significant interaction between these two variables. Naturalized species dispersed by birds and primates and those with range midpoints closer to the equator had spread to more compartments than other species (Fig. 2). The significance of bird and primate dispersal was weakened after phylogenetic correction and 45% of total deviance was explained (Table 3). Variation partitioning showed that variables alone were responsible for more than two-thirds of explained deviance, whereas phylogeny alone was responsible for only one-sixth (Fig. 1). Only the number of seeds per fruit could significantly explain linear spread (Adjusted R2 = 0.266, F1,42 = 16.57, P < 0.001; Table 3). Species producing more than 10 seeds per fruit spread further (95% confidence interval = 636–1171 m) than species with fewer seeds per fruit (95% confidence interval = 275–474 m). This model remained significant after phylogenetic correction (Adjusted R2 = 0.30, F3,41 = 10.12, P < 0.001; Table 3). Variation partitioning revealed that the number of seeds per fruit was responsible for the majority of deviance explained, whereas phylogeny alone contributed relatively little to explained deviance (Fig. 1).

Figure 2.

Number of compartments colonized by naturalized species with differing dispersal syndromes and in relation to midpoint of native latitudinal range in degrees. Lines represent predicted number of compartments colonized by species with differing dispersal syndromes, according to the fitted model.

Discussion

factors explaining species success at each invasion stage

Recognizing that causal factors behind alien plant success at different stages of invasion may change throughout the invasion process should help improve our ability to explain and predict differential success among species (Lloret et al. 2005; Dietz & Edwards 2006). We found in this study that variables explaining species success did indeed change at each invasion stage, but we also found that plant traits relating to reproduction and dispersal (seed mass, number of seeds per fruit and bird and primate dispersal) were important at several stages (Fig. 1).

Species with larger seeds were more likely to be regenerating than species with smaller seeds. More than a third of the species failing to regenerate were eucalypts and conifers (Cupressaceae) that typically had small seeds (Eucalyptus mean = 3.14 mg; Cupressaceae mean = 17.1 mg). Species with very small seeds are often light-demanding and require bare substrates in order to germinate and survive as seedlings and saplings (Rejmánek & Richardson 1996). The variation in regeneration success explained by phylogeny jointly with variables may at least partially result from the failure of eucalypts and conifers to regenerate. This phylogenetic constraint on regeneration may result in a possible bias toward lower perceived likelihoods of regeneration for species with smaller seeds and with native ranges further away from the tropics. The minimum model explaining regeneration success had relatively weak predictive power, suggesting that regeneration may be affected by more complex and as yet unidentified drivers. Establishment of naturalized species in closed forest was also more likely for larger seeded species, possibly because establishment in a forest may favour large seeds with high energy reserves to ensure survival and growth in a low-light environment (Westoby et al. 1992; Osunkoya et al. 1994; Baraloto & Forget 2007).

Naturalization was marginally more likely for species producing fewer seeds per fruit but also for species with smaller seeds, whereas species with more seeds per fruit tended to have spread over a greater linear distance. This result may reflect that species producing smaller, more numerous seeds per fruit may have greater powers of colonization and spread (Leishman 2001). However, evidence for increased colonization ability of smaller seeded species is equivocal (Leishman 2001), and it is not entirely clear why number of seeds per fruit and seed mass are both negatively related to naturalization success, as species producing smaller seeds might be expected to have more seeds per fruit than larger seeded species.

The midpoint of the native latitudinal range was retained in models describing regeneration and spread among compartments. Species were more likely to regenerate and spread widely if they had native ranges centred on the tropics. Climate matching between the native and introduced ranges can form a key component of models seeking to predict invasion risk (Scott & Panetta 1993; Pheloung et al. 1999), at least at a broad scale (Rejmánek 2000). Species adapted for dispersal by birds and primates, once naturalized, were also marginally more likely to spread to more compartments and to establish in forest than species with other dispersal syndromes (Fig. 1). The weakening of importance of dispersal for compartmental spread when phylogeny was considered probably reflects phylogenetically conservative dispersal syndromes. Dispersal by frugivorous birds is often considered as an important attribute contributing to plant invasion success (Buckley et al. 2006). Greater spread and forest establishment by species adapted for bird and primate dispersal reflects known differences in maximum dispersal distances and dispersal kernels for species with differing dispersal syndromes (Greene & Calogeropoulos 2002). However, dispersal syndrome was not important in explaining linear spread of a species. This suggests that the linear distance a species has spread is similar for each dispersal syndrome on average, and distance measurement may be constrained by the localized spatial scale of this study. Evidence for the role of animals in dispersing alien plant species in the East Usambaras is limited, but avian frugivore diversity is high and includes silvery-cheeked hornbills (Bycanistes brevis) that are known to be efficient dispersers of the relatively large-fruited invasive tree Maesopsis eminii (Cordeiro et al. 2004).

It is intuitive that shade-tolerant species would be more likely to naturalize, since this trait would facilitate survival of seedlings and saplings under plantation canopies. Naturalized species tended also to be faster growing and fast growth rates might be advantageous in order to compete for light. However, these results seem inconsistent as shade tolerance is often negatively correlated with relative growth rate (Wright et al. 2004; Poorter & Bongers 2006). Ideally, in situ measurements of growth rate and shade tolerance are required in order to elucidate the strength and direction of influence of these two variables on invasion success in this system. As recorded elsewhere (Krivánek et al. 2006), species planted earlier were marginally more likely to be naturalized, despite the narrow 28-year timeframe within which species were introduced to ABG. For regenerating species, only 34% were planted before 1905, compared to 58% of naturalized species. Finally, we found no relationship between alien species success and the number of native relatives in the East Usambaras. This result does not support the notion that relatedness to native flora affects the likelihood of alien plant species invasion (Rejmánek et al. 2005; Strauss et al. 2006; Diez et al. 2008). More experimental studies are needed to test explicitly whether taxonomic isolation from a native flora can facilitate or impede alien plant invasions, and to understand the mechanisms driving its effect (Diez et al. 2008).

propagule pressure and phylogeny

Planting effort, the surrogate for propagule pressure used in this study, did not explain species success at any invasion stage in this study. The propagule pressure experienced by rain forest neighbouring ABG will not only be a function of the planting effort and reproductive output per species, but also the distance of plantings from the forest (Edward et al. 2009). However, we found that naturalized species with original plots closer to existing forest edges were no more likely to be established in closed forest than species planted further away. These results suggest that the ability of alien species to naturalize, spread and invade natural vegetation will be enhanced for species possessing certain traits, even if introduction effort is low. However, an alternative explanation is that the number of plots planted may not be an accurate surrogate for propagule pressure.

As in other studies (Kuster et al. 2008; Lambdon 2008; Milbau & Stout 2008), phylogeny played a relatively minor role in explaining species success at each stage of invasion compared to explanatory variables (Fig. 1), suggesting that any ‘inherent invasiveness’ among species may be largely independent of evolutionary history. However, a relatively large proportion of deviance in forest establishment was explained by phylogeny (Fig. 1), which may indicate greater phylogenetic constraints on the ability to invade forest, compared to the ability to regenerate, naturalize and spread. A similarly large proportion of deviance was explained by phylogeny and traits jointly, which probably reflects phylogenetically conserved seed sizes and dispersal syndromes. In addition, a number of plant traits in the naturalization and forest establishment models were no longer significant after phylogenetic correction, including shade tolerance and number of seeds per fruit (naturalization), seed mass (naturalization and forest establishment) and dispersal (forest establishment). This may suggest that phylogeny is more strongly related to naturalization and forest establishment success, but also that the traits involved are largely phylogenetically conserved.

concluding remarks

This study demonstrates the importance of viewing the invasion process as a sequence of stages in order to understand the relative ability of multiple factors to explain species success. We found that the suite of factors explaining species success did change at each transition. A clear message from this work is that it matters considerably how ‘invasive’ species are defined, as the resulting factors associated with invasiveness are likely to differ depending on the definition. Relatively objective and repeatable measures of invasion success are scarce, and comparative studies such as this one vary in terms of scale and the range of residence times for species introduced. This variety among studies undoubtedly limits our ability to make comparisons and to obtain a general set of clear traits associated with invasions. However, by disentangling the different invasion stages and using objective criteria to assess species status, this study highlights that relatively simple models containing a handful of variables can help to explain why some alien plants are able to naturalize, spread and even establish in closed tropical forests.

Acknowledgements

This work is part of the Darwin initiative project ‘Combating alien invasive plants threatening the East Usambara Mountains in Tanzania’ (162/13/033) and authors are grateful for financial support from Defra and NERC (W.D.). The authors would also like to thank the staff of the Amani Nature Reserve, especially Mr. Corodius T. Sawe, and the Tropical Biological Association, especially Dr. Rosie Trevelyan, for logistical support. Thanks also to Mr. Ahmed S. Mndolwa and Iddy Rajabu (Botanists, Tanzanian Forestry Research Institute), and Mr. Abduel B. Kajiru, for fieldwork assistance.

Ancillary