We studied the three alien tree species that are most frequent in European woodlands: Prunus serotina Ehrh., Quercus rubra L. and Robinia pseudoacacia L. (Wagner et al. 2017). These species came from eastern North America and have been introduced via forestry, in the 18^th, 19^th and 17^th centuries, respectively (Muys et al. 1992; Cierjacks et al. 2013; Woziwoda et al. 2014). All the species strongly modify invaded habitats by influencing leaf litter decomposition (Dobrylovská 2001; Horodecki and Jagodziński 2017), nutrient cycling (Rice et al. 2004; Aerts et al. 2017) and light availability (Knight et al. 2008; Chmura 2013; Jagodziński et al. 2018). R. pseudoacacia is dispersed by wind (Cierjacks et al. 2013; Vítková et al. 2017), P. serotina by birds (Pairon et al. 2006; Jagodziński et al. 2015; Dylewski et al. 2017) and Q. rubra by birds and rodents (Myczko et al. 2014; Bieberich et al. 2016). P. serotina and R. pseudoacacia are mid-successional species while Q. rubra – late-successional, which is manifested in their abundance and disturbance responses (Dyderski et al. 2018). For comparison we chose the four native species most frequent within study plots: Acer pseudoplatanus L., A. platanoides L., Quercus petraea (Matt.) Liebl. and Fagus sylvatica L. In a preliminary inventory of natural regeneration in 2015 these species were found in 80.1%, 46.5%, 68.5% and 34.9% of 372 study plots, respectively, in comparison with P. serotina (59.9%), Q. rubra (44.6%) and R. pseudoacacia (23.6%). The native species chosen were not phylogenetically related to the invasive species, but are their most frequent competitors in the regeneration layer. A. platanoides and A. pseudoplatanus are mid-successional species dispersed by wind while F. sylvatica and Q. petraea are late-successional species dispersed by birds and rodents. Despite the affiliation of the species being studied in successional stages, the pattern of species co-occurrence did not follow the division into early-, mid- and late-successional species (Suppl. material 1: Fig. S1).

We conducted our study in the Wielkopolski National Park (WNP; W Poland; 52°16'N, 16°48'E; 7584 ha). The main aim of conservation in WNP is to preserve a valuable post-glacial landscape, including valleys, moraine hills and lakes. The climate in WPN is temperate, transitional between oceanic and continental. Mean annual temperature in Poznań (c.a. 15 km from WNP) was 8.4 °C and mean annual precipitation was 521 mm for the years 1951–2010. Dominant soil types in the study area are luvisols (47%) and brunic and haplic soils (30%), while podzols constitute only 7% (Nowak et al. 2000). The most frequent plant associations are Galio sylvatici-Carpinetum (fertile broadleaved forest dominated by Quercus, Carpinus and Tilia), Calamagrostio arundinaceae-Quercetum (acidophilous forest with Q. petraea) and Querco roboris-Pinetum (mixed Quercus-Pinus forest). However, there are small vegetation patches of both fertile riparian forests (Fraxino-Alnetum, Querco-Ulmetum minoris) and poor coniferous forests (Leucobryo-Pinetum). Most of the fertile sites were planted with Pinus sylvestris before the national park establishment in 1957 (Nowak et al. 2000). Moreover, WNP is the national park with the highest richness of alien woody taxa in Poland (158; Purcel 2009; Gazda and Szwagrzyk 2016). There are pure stands of alien species surrounded by large areas of forest ecosystems with different levels of disturbances and human impacts. This makes WNP an excellent place to test hypotheses about the impact of disturbances on the ecological success of invasive species.

The study design covers a set of 378 plots (100 m²) arranged in 21 blocks: nine for Q. rubra and six for P. serotina and R. pseudoacacia with the central part of each block located in a monoculture stand of invasive species (Dyderski and Jagodziński 2018a, b). Because P. serotina occurs only as an admixture in tree stands, as central parts we used tree stands with high densities of fruiting P. serotina trees in the shrub layer. In the center of each block we established a pair of study plots. Then, we established additional pairs of study plots along each of the four sides (N, S, E, and W) nearly outside the stand, at the invasion edge (the border between crowns of the invasive species monoculture and the surrounding trees; Rodríguez et al. 2017). The next four pairs were located 30 m from the invasion edge. This design generated 18 study plots within each block. Six study plots were excluded, as due to their systematic distribution, these plots were located outside the forest vegetation. Thus, the final number of plots was n=372. Study plots represented most of the vegetation types occurring in Wielkopolski National Park and wide gradients of resource availability: litter pH ranging from 3.83 to 6.44 and light availability ranging from 0.7 to 25.1% of the open sky (see details of measurements below). The variability of microsites conditions within the plot system (Dyderski and Jagodziński 2018a, b), was accounted for in the study design by including the ‘plot’ term as a random effect in mixed models (Roberts et al. 2017).

In July 2017 we destructively harvested sample trees: seedlings (defined as individuals germinated in a particular year) and saplings (defined as individuals at least one year old and with height < 0.5 m). We divided natural regeneration into seedlings and saplings due to low seedlings survival (Beckage and Clark 2003; Knight et al. 2008; Canham and Murphy 2016), supply of carbohydrates stored in seed (Curt et al. 2005; Ziegenhagen and Kausch 1995) and differences in morphology affecting the functional traits studied (Cierjacks et al. 2013; Annighöfer et al. 2016). We surveyed an area within a 5 m wide buffer around each plot pair and we randomly selected up to five specimens for each species. The number of harvested plants was proportional to the available number of specimens and species densities within plots. We did not collect heavily damaged and browsed plants, unless there were no alternative specimens in the area examined. This accounted for the joint effects of lower growth and resistance to herbivory in suboptimal sites (Dyderski and Jagodziński 2018b). Field study within uncontrolled growth conditions resulted in unequal numbers of sample trees per species: 75 saplings and 48 seedlings of A. platanoides, 91 and 341 of A. pseudoplatanus, 23 and 59 of F. sylvatica, 141 and 191 of P. serotina, 29 and 78 of Q. petraea, 56 and 71 of Q. rubra and 37 and 89 of R. pseudoacacia. This inequality was an effect of unequal distributions of natural regeneration of the species within the study plots (Dyderski and Jagodziński 2018a, b). We dug each sample tree, and then carefully cleaned and divided it into roots, stems with branches and leaves. We did not include acorns which were still attached to Q. rubra and Q. petraea seedlings in the total biomass. We excluded from the dataset 14 P. serotina and 23 R. pseudoacacia saplings which had root suckers.

After separation, all sample tree biomass components were packed into envelopes, unfolded and transported into the laboratory. Leaves which were suitable for scanning, according to Cornelissen et al. (2003), were dried separately in a special press. All material was dried in an oven with forced air circulation at 65 °C (UN 750 and ULE 600, Memmert GmbH+Co.KG, Germany), to a constant mass. All biomass components were weighed using a BP 210 S (Sartorius, Göttingen, Germany) with an accuracy of 0.001 g. Leaves were scanned using WinFOLIA 2013 PRO software (Regent Instruments Inc., Quebec, Canada) to measure their leaf area. From the area and leaf dry biomass of scanned leaves, specific leaf area (SLA) was calculated, as leaf area divided by leaf mass. We also calculated leaf area ratio (LAR) as a ratio of projected leaf area (PLA; SLA multiplied by total leaf biomass) and TB (total biomass), as well as leaf mass fraction (lmf), stem mass fraction (smf) and root mass fraction (rmf). We analyzed traits of biomass allocation (lmf, smf, rmf) expressed as plant investment in particular organs – acquisition of limiting resources (Jagodziński and Oleksyn 2009; Poorter et al. 2015; Lieurance and Landsbergen 2016). TB was used as an overall plant fitness measure reflecting space filled by the species, and is also claimed to be a good measure of plant fitness (Younginger et al. 2017). We chose three leaf traits: SLA reflecting resource use efficiency within foliage (Wright et al. 2004; Díaz et al. 2016), PLA reflecting filling of space by foliage (Jagodziński et al. 2016) and LAR as a measure of whole plant investment in foliage efficiency. These factors are known to vary across soil fertility and light availability gradients.

To characterize environmental gradients we used light availability and litter pH. We measured light availability as canopy openness index (diffuse non-interceptance; DIFN) using an LAI-2200 plant canopy analyzer (Li-Cor Inc., Lincoln, NE, USA). For each plot pair we recorded eight series of ten samples (four for each single plot) in August 2016. Although light availability was not measured in the year of harvest, we found low differences in light availability expressed by Ellenberg’s ecological indicator community-weighted mean values between 2016 and 2017. The differences ranged from 0.00 to 1.37, with an average of 0.23±0.02, in nine-degree scale, which is within the range of interannual species turnover, as the highest changes we found in plots with low number of species. Thus, we assumed that light availability did not change between 2016 and 2017 significantly. In March 2017 we collected four samples of leaf litter from circular plots (0.16 m²; Fig. 1b). Litter pH was assessed using an electronic pH-meter in distilled water solution after 24 hours. We used litter pH as a proxy for nutrient cycling rates and soil acidity in forest ecosystems (Bigelow and Canham 2002). However, as this variable explains only part of the variability in soil fertility, we encourage caution in interpretation, as one component of soil fertility may be masked by another, not related to litter pH, e.g. nitrogen:phosphorus ratio.

Figure 1.

Study design: a scheme for a block of experimental plots in the field (each of 21 blocks is a set of 18 100 m² square plots; Dyderski and Jagodziński 2018a, b) b scheme for litter collection (circles with numbers 1–4 are 0.16 m² litter samples, collected in March 2017).

Our study design covered three invasive species with different biologies and we did not choose native species as phylogenetically-related pairs, but rather the most frequent competitors. For that reason we did not test specific alien-native species pairs but we compared each alien species to each native. For comparison of mean trait values of species we used one-way mixed-effects ANOVA followed by a Tukey post hoc test, implemented in the multcomp::glht() function. In this model we treated species as a fixed effect and plot as a random effect, to account for plot-specific effects, such as microsite variability and other unknown effects. Mixed models were developed using the lmerTest::lmer() function (Bates et al. 2015; Kuznetsova et al. 2017). To assess differences between saplings and seedlings within each species-trait combinations we used t-tests implemented in stats::t.test() function. Due to multi-species comparisons we decided not to apply any restrictions on p-values following Moran (2003).

To compare variability of the traits studied we assessed differences in trait coefficients of variation (CV) between two species using Krishnamoorthy and Lee’s (2014) modified signed-likelihood ratio test (M-SLR test) implemented in the cvequality::mslr_test() function (Marwick and Krishnamoorthy 2016). This test has lower rates of type I error and more power across a range of conditions than the widely used (e.g. Funk 2008; Paquette et al. 2012) asymptotic test of Feltz and Miller (1996). Moreover, the M-SLR test allows for uneven sample numbers, which is necessary for our study design.

To assess the differences among species across resources availability gradients we used random forest algorithm (Breiman 2001). This method has a good performance in case of non-normal distributions of studied parameters, accounts for interaction between correlated predictors and has also a high predictive power. Its potential drawback might be an overfitting, which limits potential model transferability. To stabilize variance and avoid the influence of different units of predictors, prior to analyses we centered and scaled predictors, i.e. we subtracted mean values and divided by SD. To decrease overfitting we used repeated cross-validation by randomly splitting a dataset into training and validation sets within each iteration of model building (10 repeats 10 times) in the caret::train() function (Kuhn 2008). Within each iteration 75% of the dataset was used as training set and 25% – as validation set. To conclude about models output we provided information about variable importance, expressed as drop-out loss, i.e. loss in model RMSE when a particular variable is perturbed within the data table. Drop-out loss, similarly to RMSE, is expressed in the units of dependent variable. To show differences between species studied and interactions between species and resource availability we used partial dependence plots. These plots show output predicted for each observation, assuming constant value of other variables. As a constant value algorithm inputs mean values of parameters. These analyses were conducted using DALEX and ceterisParibus packages (Biecek 2018a, b). All analyses were conducted using R software (R Core Team 2018).

Analysis of mean values of traits revealed statistically significant differences among the species studied (p<0.001; Fig. 2, Suppl. material 1: Table S1). Two species (Q. rubra and Q. petraea) had the highest TB and Q. rubra had the highest PLA. Two invasive species (P. serotina and R. pseudoacacia) had the highest SLA, LAR and leaf mass fraction and the lowest root mass fraction. Comparing seedlings and saplings we found statistically significant (p<0.05) differences in 38 of 47 species-trait pairs studied (Fig. 2, Suppl. material 1: Table S2). We did not find statistically significant differences (p>0.05) in the case of Q. petraea and R. pseudoacacia leaf mass fraction, F. sylvatica and Q. rubraPLA, Q. petraea and Q. rubra root mass fraction, A. platanoides and F. sylvatica seedlings SLA and R. pseudoacaciaTB.

Figure 2.

Mean (+SE) values of species traits of natural regeneration: SLA – specific leaf area (cm² g^-1), TB – total biomass [g], rmf – root mass fraction, lmf – leaf mass fraction, smf – stem mass fraction, PLA – projected leaf area [cm²], LAR – leaf area ratio [cm² g^-1]. Differences were assessed using one-way mixed effects ANOVA and Tukey post hoc tests – species marked by the same letter (lower-case letters for saplings and upper-case letters for seedlings) did not differ significantly statistically (p<0.05). ANOVA details are provided in Suppl. material 1: Table S1. Species are ordered according to successional stages.

Analysis of trait CVs within species pairs in most cases revealed statistically significant differences in CVs between species (Fig. 3). Within native species F. sylvatica usually had higher CVs of functional traits than other native species and did not, statistically, differ significantly in any CV with A. platanoides. P. serotina and R. pseudoacacia had significantly higher CVs statistically of most traits studied than native species. The exceptions were higher CV of smf in A. pseudoplatanus saplings and LAR in A. pseudoplatanus seedlings than P. serotina, and higher TB CVs for Acer spp. than R. pseudoacacia. Moreover, all native species saplings had higher CVs of PLA than R. pseudoacacia saplings.

Figure 3.

Coefficients of variation for age stages and parameters: SLA – specific leaf area (cm² g^-1), TB – total biomass [g], rmf – root mass fraction, lmf – leaf mass fraction, smf – stem mass fraction, PLA – projected leaf area [cm²], LAR – leaf area ratio [cm² g^-1]. The same letters species which did not differ statistically significantly (p>0.05) in pairwise comparisons by modified signed-likelihood ratio (M-SLR) tests of differences. Tests revealed the same division into species groups for both saplings and seedlings.

Random forest models revealed that in all traits except SLA in case of seedlings species identity was the most important factor shaping trait values (Table 1). In general, SLA, LAR and lmf decreased with increasing DIFN, however differences among species studied were higher than differences along resource availability gradients (Figs 4, 5). We noted similar traits reaction to soil pH, while in the highest range of pH (over 6.0), SLA, LAR and lmf decrease was higher than in pH up to 6.0. Root mass fraction increased with increasing DIFN but decreased with increasing pH. TB increased with both increasing DIFN and pH. In terms of leaf investment (SLA, LAR, lmf) R. pseudoacacia presented the highest values across both DIFN and pH gradients, P. serotina had intermediate SLA and LAR, in comparison with native species. Q. rubra had the lowest values, while it had the highest total biomass and PLA, as well as similar rmf as native Q. petraea.

Figure 5.

Seedlings traits variability of the species studied across predictors explained by a random forest model. For explanations see Fig. 4.

Figure 4.

Saplings traits variability of the species studied across predictors explained by a random forest model. Partial dependence plots (ceteris paribus plots) show changes of predicted values when a particular predictor is changed while all remaining predictors are constant (i.e. mean value) – in the middle (B) and right (C) column we showed interactions between species and DIFN and litter pH. In the left column (A) we showed partial group predictions – predicted trait values assuming constant levels of other predictors, boxes represent interquartile range and median, whiskers represent minimum-maximum range, abbreviations of species: Apla – Acer platanoides, Apse – A. pseudoplatanus, Fsyl – Fagus sylvatica, Pser – Prunus serotina, Qpet – Quercus petraea, Qrub – Q. rubra, Rpse – Robinia pseudoacacia; traits: SLA – specific leaf area, lmf – leaf mass fraction, PLA – projected leaf area, LAR – leaf area ratio, rmf – root mass fraction, smf – stem mass fraction, TB – total biomass. For further details see Table 1. Species marked by the same letter did not exhibit significant statistical difference in predicted trait values, according to the Tukey posteriori test at p=0.05.

Our study revealed that alien and native species differed the most in SLA, rmf, lmf and LAR; however the variability was usually not related to the resource gradients. This is connected with high inter-specific variability of functional traits, which results from different morphology and phylogeny of species studied (e.g. Poorter et al. 2015). For example, analysis of seasonal variability of 12 herbaceous species SLA, biomass and TLA(i.e. along DIFN temporal gradient) growing in the same locality also revealed a higher range of interspecific differences than during ontogeny within species (Jagodziński et al. 2016). For that reason previous studies focused on phylogenetically-related pairs of species (e.g. Grotkopp et al. 2010), to reduce phylogenetic structures from analyses (Roberts et al. 2017). However, in natural conditions species are co-occurring regardless of phylogenetic relativity. Thus, the success of some invasive species is claimed to be connected to being different from native species in terms of functional traits, e.g. by colonizing empty niches or by ability to break habitat filtering (Melbourne et al. 2007; Hierro et al. 2005).

The differences among species reflect higher investment in leaves and a more acquisitive strategy of alien species, especially in the cases of P. serotina and R. pseudoacacia. These species used higher investment in foliage and higher SLA as ways of increasing PLA. SLA is a strongly acquisitive trait, correlated with photosynthetic capacity (Wright et al. 2004), as well as with ruderal and competitive life strategies (Westoby 1998). Q. rubra did not differ in these traits from native species but revealed higher biomass allocation to root systems. This is in line with previous observations indicating higher root:shoot ratio of Q. rubra (Kuehne et al. 2014). The investment in leaves and roots is a predictor of total biomass (Enquist and Niklas 2002; Poorter et al. 2015) – the higher these parameters are, the lower the total biomass. However, this assessment covered the whole spectrum of plants. At the stage of saplings higher investment in leaves resulted in higher total leaf area and competitive advantage over native species in limited light availability (te Beest et al. 2015). Nevertheless, previous studies are equivocal in assessing differences in lmf and rmf between alien and native species. Some of these studies indicated a lack of differences in rmf (Grotkopp et al. 2010; Matzek 2011) while some showed higher rmf (Kuehne et al. 2014) and some lower rmf (Paquette et al. 2012; González-Muñoz et al. 2014) of alien species. Biomass allocation to the root systems also depends on root competition (Kawaletz et al. 2014). Investment in acquisitive traits causes a competitive advantage over native species (Blossey and Notzold 1995; te Beest et al. 2015).

Our study revealed differences between seedlings and saplings in case of most traits and species. These differences are mostly connected with different leaves morphology (Cierjacks et al., 2013; Annighöfer et al., 2016). The youngest leaves are usually thinner than older ones, which results in higher SLA and PLA. This may influence low survival of seedlings (Beckage and Clark 2003; Canham and Murphy 2016; Knight et al. 2008). For TB and PLA, seedlings of alien species had lower values than native species, while saplings had higher values. This may show that alien species do not have a biomass advantage in the first year of life. This comparison of alien and native species is biased by the outstandingly high biomass of Q. rubra. However, this species also differs from the other alien species studied in leaf-root allocation. This may indicate that the alien species studied had two ways to achieve the same goal – high leaf area. In the case of smaller plants the goal was realized by changes in biomass allocation and in the case of Q. rubra – by high biomass. Similar differentiation of alien species into acquisitive and conservative was found by Tecco et al. (2013). Higher biomass production and PLA was also mentioned as a factor connected with invasiveness (Grotkopp et al. 2010; Kuehne et al. 2014; Jo et al. 2015).

Our results revealed that P. serotina and R. pseudoacacia mostly had higher CV than native species. However, this was not always connected with higher variability of alien species traits along resources availability gradients, which may indicate high variability of trait values within young generations of invasive species. Most previous studies revealed differences in phenotypic plasticity between alien and native species (Paquette et al. 2012; Lamarque et al. 2013; Hou et al. 2014). Kuehne et al. (2014) did not find differences in phenotypic plasticity of Q. rubra and three co-occurring native species. Our study indicated that for acquisitive traits, variability of Q. rubra traits, expressed by interaction with environmental gradients, also did not differ from native species.

Assuming high propagule pressure, observed high variation in trait values may be the reason for their ecological success. However, most of the previous studies found that this variability was connected to phenotypic plasticity (e.g. Davidson et al. 2011; Paquette et al. 2012; Lamarque et al. 2013). Nevertheless, in our study we cannot distinguish which part of variability is driven by phenotypic plasticity, due to lack of data about each individual provenance. Neutral theory assumes similar probability of mortality, speciation and reproduction within the same functional guild (Hubbell 2001). Application of this theory to invasion ecology highlights the need of accounting for the randomness in modeling community structure and functioning (Daleo et al. 2009). This randomness may be reflected in high variation of traits and high density of alien species seedlings and saplings, which allow for survival of the best fit individuals. This interpretation is strongly in line with previous studies highlighting the role of propagule pressure in ecological success of invasive species (Lockwood et al. 2005; Vanhellemont et al. 2009; Pyšek et al. 2015; Dyderski and Jagodziński 2018a). This speculation may be solved by testing whether survival of alien species seedlings depends on resource availability per plot, or shows a random pattern.

Our study revealed that alien species had a more acquisitive strategy of light acquisition, expressed by higher LAR than native species. The effects of resource availability on the leaf traits studied – lmf, LAR and SLA – were lower than differences between alien and native species. Also Robakowski et al. (2018) found high importance of light availability on lmf, rmf and photosynthesis efficiency in P. serotina and Q. petraea. Previous studies usually considered the effects of resource availability on particular species (Funk 2008; Lamarque et al. 2013; Kuehne et al. 2014) or phylogenetically related pairs of alien and native species (Allison and Vitousek 2004; Grotkopp et al. 2010; Matzek 2011). Only a few studies accounted for ecologically related alien and native species – i.e. co-occurring in one ecosystem type (González-Muñoz et al. 2014; Kuehne et al. 2014; te Beest et al. 2015). This lack of real relationship was raised as an important concern by Hulme (2008). Moreover in our study alien species occur and regenerate spontaneously in the study plot system.

Comparing litter pH and light availability gradients, differences between alien and native species traits were usually clearer along the DIFN gradient. The exception was PLA and (for saplings only), SLA and TB. We would expect that SLA will differ mostly due to light availability, as this factor drives SLA variability (Jagodziński et al. 2016). However, as SLA is nitrogen-dependent (Wright et al. 2004; Díaz et al. 2016), this trait also differed with litter pH, which is a proxy for general soil fertility and nitrogen content. For that reason SLA increases with litter pH and decreases with DIFN (Table 1). In the cases of lmf and LAR effect sizes of pH were lower. This indicates that general investment in foliage – both based on higher allocation and on leaf morphology (SLA) – is a reaction to light availability. Thus, as invasive species studied strongly modify environment, especially light availability and litter chemistry (e.g. Allison and Vitousek 2004; Knight et al. 2008; Horodecki and Jagodziński 2017), traits of natural regeneration might be influenced by invasive species presence in overstory. This feedback may be more important in case of alien species studied, as their abundance is strongly related to the presence of parental trees (Dyderski and Jagodziński 2018a). However, native species are also able to strongly decrease light availability (e.g. Acer spp., Fagus sylvatica) or rate of nutrients cycling by low decomposition rate of leaf litter (Dobrylovská 2001; Horodecki and Jagodziński 2017). Thus, in our study we investigated effect of resources availability rather than alien-native species-specific effects of overstory.

Invasive species studied revealed three different patterns of biomass investment differentiating them from the native species. R. pseudoacacia realized strategy ‘try hard’ in terms of investment in foliage, supporting the suggestion of Tecco et al. (2010) that invasive species invest more in acquisitive traits. In contrast, Q. rubra represents rather conservative strategy (Tecco et al. 2013) – invests in traits connected with persistence – low SLA, and in belowground resources competition. However, this strategy seems to be effective, as this species reaches higher individual biomass and high projected leaf area. Although P. serotina has higher lmf, its SLA and LAR places this species among native ones. This would suggest that P. serotina rather ‘joins the locals’ – represents similar life strategy to its most frequent competitors. All of these strategies lead to high total leaf area, which allows for out-shading native vegetation and reducing competitors (te Beest et al. 2015). The three species studied are the most frequent alien woody species in European forests (Wagner et al. 2017). Therefore, we conclude that both strategies are effective ways to reach ecological success, in contrast to Tecco et al. (2010; 2013), who claimed that an acquisitive strategy is more efficient. However, we found very limited support for variability of traits across interactions of species and resource gradients. Despite high variability and differences among species, this suggest that invasive species success is not related to the species traits variability. Thus, all three species are rather limited by propagule pressure, which is the main factor driving their biomass in the plant communities studied (Dyderski and Jagodziński 2018a). High variability of invasive species traits also suggests randomness in seedling survival, which similarly to neutral theory of invasion, highlights the necessity of including randomness in modelling biological invasions.