Higher plant colonisation and lower resident diversity in grasslands more recently abandoned from agriculture

Rates of species colonisation and extirpation are increasing in plant communities world‐wide. Colonisation could potentially help compensate for, or compound, resident diversity loss that results from global environmental change. We use a multifactorial seed addition grassland experiment to examine relationships between plant colonisation, resident species diversity and key community assembly factors over 3 years. By manipulating colonist seed rate, imposing disturbance and examining abundance and diversity impacts of 14 formerly absent sown colonists in communities that varied in successional stage and time since agricultural abandonment, we were able to disentangle effects of global change factors (species introduction, novel disturbance and land use change) that are usually confounded. Evidence suggested that cover abundance of sown colonists was most strongly influenced by successional stage of recipient communities, though number of growing seasons was also important for the group of seven colonists with resource conservative ‘slow’ life history traits. Colonist type, seed rate and disturbance had weaker relationships with colonist cover. Factors affecting sown colonist cover were highly conditional. A negative relationship between plot‐level disturbance and colonist cover in early successional communities meant that, despite a positive relationship in late succession, colonisation was negatively related to disturbance overall, defying theoretical expectations. Non‐sown resident diversity was negatively related to colonist cover and positively related to successional stage. Resource acquisitive colonists with ‘fast’ life history traits appeared to limit cover of ‘slow colonists’ when the two groups were sown together, likely reflecting niche pre‐emption. Communities at earlier stages of succession had lower resident diversity and experienced higher levels of colonisation than communities at later stages of succession. Elevated colonisation and lower resident diversity both appeared to be symptoms of human‐induced land use change. However, results suggested that resource competition from plant colonists may also limit resident diversity in grasslands abandoned from agriculture more recently. Synthesis. Our findings point to the importance of resource availability and competition on plant colonisation and colonist impacts on residents. Although colonisation is potentially a source of biodiversity in the short term, our results suggest that plant colonists that reach high abundance may be a further threat to resident plant diversity in secondary grasslands recovering from a recent history of agriculture.


| INTRODUC TI ON
Rates of species colonisation, where species establish new populations in previously unoccupied sites, are increasing in communities across the world (Dornelas et al., 2019;Finderup Nielsen et al., 2019).
Colonisation is integral to population and community dynamics and is required for species persistence and community diversity (Chesson, 2000;Vahsen et al., 2018).With rates of local extinction (extirpation) also increasing, the relative importance of colonisation for biodiversity is rising (Dornelas et al., 2019).Rather than species losses being confined to some communities and species gains to others, biodiversity change research shows that individual communities typically contain both winners and losers (Dornelas et al., 2019;Ladouceur et al., 2022;Pandolfi et al., 2020).Although species wins and losses do not always balance (Blowes et al., 2019;Finderup Nielsen et al., 2019;Jandt et al., 2022;Ladouceur et al., 2022) (and certainly not when considering beta and gamma diversity (Dornelas et al., 2014;Hillebrand et al., 2018;Kortz & Magurran, 2019)), it seems that elevated rates of colonisation and extirpation-including in grasslands (Hillebrand et al., 2018)-are linked, raising questions about this relationship.
But does colonisation directly affect resident diversity too, and does that depend on colonisation co-occurring with other forms of environmental change, like disturbance or land use?Colonisation could reduce the occupancy and abundance of resident species through processes like competition or habitat modification, enabling or exacerbating negative effects of other global change factors (Catford et al., 2020;Kraft et al., 2015).Alternatively, colonists could effectively replace lost resident plants whose loss was independent of colonisation (Ladouceur et al., 2022).Given that the prevalence of colonisation is increasing and given that colonists have the potential to compound or compensate resident species loss, it is critical to increase understanding of factors controlling colonisation and the way colonisation affects resident diversity.
Reflecting drivers of community assembly (Götzenberger et al., 2012), the occupancy and abundance of both colonist and resident plants are influenced by local environmental conditions, resource availability, propagule availability and characteristics of both the colonists and resident community (Catford et al., 2009;Chang & HilleRisLambers, 2016;Pearson et al., 2018;Pichon et al., 2023).Human-induced environmental change can alter these community assembly factors, which could impact colonisation and extirpation in communities world-wide.For example, human activities directly and indirectly: elevate rates of novel disturbances, which can change abiotic conditions, increase resource availability and alter community diversity and composition (Moles et al., 2012); increase the dispersal and introduction of some types of species, and decrease the dispersal of others (Runghen et al., 2021;Tucker et al., 2021); and alter the composition of vegetation communities through land use change like agriculture (Dawson et al., 2017) and nitrogen deposition (Borer & Stevens, 2022;Ke et al., 2023).Unless their effects are disentangled, it is hard to identify which factors are driving increased colonisation, and whether colonisation and resident species loss are both simply outcomes of broader environmental change, or whether colonisation is a driver of resident species loss itself (Catford et al., 2020).
In this study, we use a 3-year seed addition grassland experiment to examine relationships between: key community assembly factors and plant colonisation; and plant colonisation and resident species diversity vis-à-vis other community assembly factors (Figure 1).
We use 'colonisation' to describe the establishment of previously absent plant species that were experimentally sown (colonists).
The processes that we examine may thus relate to numerous types of compositional change that result from the immigration and recruitment of new species, including species added for restoration or translocation, incursion of range-shifting and range-expanding species, human-mediated exotic species invasion, as well as 'natural' resource competition from plant colonists may also limit resident diversity in grasslands abandoned from agriculture more recently.7. Synthesis.Our findings point to the importance of resource availability and competition on plant colonisation and colonist impacts on residents.Although colonisation is potentially a source of biodiversity in the short term, our results suggest that plant colonists that reach high abundance may be a further threat to resident plant diversity in secondary grasslands recovering from a recent history of agriculture.

K E Y W O R D S
Cedar Creek Ecosystem Science Reserve, community assembly and restoration, community biodiversity change, disturbance and agricultural land use change, grassland secondary succession, invasibility and species invasion, plant colonisation and local extinctions, seed addition experiment (stochastic or neutral) turnover (Seabloom et al., 2003).We added two groups of colonist species ('fast' and 'slow' colonists, with group names reflecting species' resource economics and life history characteristics (Reich, 2014;Tilman, 1982)), together and alone, to grassland plots at two seeding rates.We examine effects of smallscale disturbance and three types of communities, which varied in time since agricultural abandonment.Because previous studies in neighbouring grassland experiments have shown that colonisation is higher in communities with lower resident diversity (Catford et al., 2020;Fargione & Tilman, 2005), here we focus on other drivers of colonisation.Specifically, we ask: 1. What are the effects and relative importance of colonist type, colonist seed rate, disturbance, successional stage of recipient community and growing season on the cover abundance (when present) of groups of sown colonist species? 2. What are the effects and relative importance of sown colonist abundance, colonist richness, colonist type, disturbance, community successional stage and growing season on the diversity of non-sown resident species in recipient communities?
Based on empirical and theoretical evidence (Catford et al., 2009;Catford, Daehler, et al., 2012;Hui & Richardson, 2017;Huston & Smith, 1987;Tilman, 2004;Vellend, 2016), we have four main expectations: (i) colonist cover will be positively related to disturbance, seed rate and growing season, and negatively related to community successional stage; (ii) the group of fast colonists will reach higher cover than slow colonists in this 3-year experiment; (iii) resident diversity will be negatively related to cover and richness of sown colonists, and positively related to disturbance, succession and growing season; and (iv) the relationship between slow colonist cover and resident diversity will be more negative than the fast colonistresident diversity relationship because slow colonists are better competitors for limiting resources and will thus exert stronger competitive effects than fast colonists (Figure 1).Despite these expectations about main effects, we expect that many of these relationships will change in sign or magnitude depending on interacting factors (Catford et al., 2022).We consider all two-way interaction effects in this experiment as plausible but refrain from articulating specific hypotheses for these to avoid undue complexity.Cedar Creek has nitrogen-limited sandy soils, annual precipitation of ~770 mm, mean summer temperatures of 27°C and winter lows of −14°C.The three sites are secondary grasslands (Veldman et al., 2015), similar in every aspect (e.g.elevation, aspect, soil type, history) other than successional stage.The sites were used for cropping from at least 1896, with agriculture then abandoned at different points in time giving a gradient of secondary succession (details in Appendix S1).

| Study design
The experiment involved four treatments in a split-plot design: (i) successional stage of recipient community (early, mid, late); (ii) disturbance (disturbed, undisturbed); (iii) colonist type (fast colonists, slow colonists, both fast and slow colonists); and (iv) seed rate of colonist species (none, low, high) (Figures S1 and S2).Colonist type and colonist seed rate were not fully crossed, so only seven out of a possible nine colonist addition treatments were applied (none, low seed rate of fast colonists, high seed rate of fast colonists, low rate of slow colonists, high rate of slow colonists, low rate of both colonists, high rate of both colonists).This resulted in a total of 42 treatment combinations (3 community levels × 2 disturbance levels × 7 colonist levels), replicated four times, giving a total of 168 plots, which were surveyed over three growing seasons.late successional plots will have lower colonist cover but higher resident diversity than early successional plots; Disturbanceundisturbed to disturbed; all other factors-low to high.In our study, we use the term 'colonist' to refer to experimentally sown species only, not self-seeded 'natural' colonists.and dominated by Poa pratensis, Ambrosia coronopifolia, Aristida basiramea and Digitaria cognata; and the early succession site was effectively 0 years old as we killed all standing vegetation with two applications of glyphosate in early spring 2013 (the site was 9 years old before herbicide application).Community composition was distinct among the three sites (Appendix S2-Figure 1; Figures S3   and S4) and species richness and soil fertility varied (Table S7), the latter reflecting the accumulation of organic carbon and total nitrogen through succession (Appendix S1; Clark et al., 2019).To assess whether our sites adequately exemplified their respective successional stages, we compared the taxonomic and functional composition of unmanipulated control plots in our three sites with unmanipulated plots in nearby old fields of comparable ages (Pérez-Navarro et al., 2023) (Appendix S2).These comparisons showed that our grassland communities fell within the range of compositions observed for their respective successional stages and that there was differentiation among the three stages (Appendix S2-Figure 1).
These patterns indicate that our communities provide sound examples of their successional stages.This aligns with findings of Clark et al. (2019), who found that old field succession largely followed consistent and predictable trends at Cedar Creek.Reflecting our split-plot design, we nevertheless treat plots as independent replicates of our target sites, not independent replicates of all grasslands at these three successional stages (Colegrave & Ruxton, 2018).
In each of the three sites, 56 1 m × 1 m plots were permanently marked, separated by 1 m wide walkways (Figure S2).Plots were randomly assigned disturbance (0, 1) and seed sowing treatments, which were enacted in May 2013 (disturbance first).Disturbance involved mowing vegetation to 10 cm followed by soil tilling, with the aim of killing approximately half of the standing vegetation.Litter and dead plants were left in the plots.Walkways were undisturbed.
We selected 23 grassland species for seed addition, all of which were native or naturalised at Cedar Creek but were absent or rare in the sites prior to seed addition.We separated the species into two groups (fast and slow colonists) based on time of colonisation during succession (Table 1; Tilman, 1988) and plant functional traits (Craven et al., 2018).A tendency for high leaf N and P content and high SLA differentiated fast colonists from slow colonists (Figure S5).Such trends are consistent with plant functional characteristics observed along successional gradients in the study region (Appendix S2-Figure 1) and elsewhere (Kelemen et al., 2017).
Within the groups, we aimed to balance functional type, origin and lifespan.We purchased seeds from local suppliers.We added 0.5 g and 4 g of viable seed per species for the low and high seed rate treatments respectively.We added the same mass, rather than the same number, of viable seeds for each species to account for the seed number-seed mass tradeoff and to avoid inadvertently favouring large-seeded species (Turnbull et al., 1999).Of 23 colonist species sown, 15 were observed at least once over 3 years following seed addition (Table S1).There was an insufficient amount of seed of one fast colonist, Strophostyles leiosperma, so it was only added to 70% of the intended plots.To correct for this mistake and avoid an TA B L E 1 Characteristics and occupancy of 14 target colonist species sown in our experiment.Note: Occupancy was calculated as percentage of plots in which each species was recorded relative to the 96 plots where each species was sown.See Table S1 for mean and maximum abundances of the target colonists, and for details of other species that were sown but did not colonise the plots.
unbalanced design in the experiment, S. leiosperma individuals were immediately removed by hand whenever observed.Where present, S. leiosperma had mean cover of 0.3% (maximum 1%).Preliminary analysis showed that its inclusion or exclusion did not affect model outputs, so it was excluded from further analyses.This left seven fast colonists and seven slow colonists (Table 1; Table S1; Appendix S1).

| Data collection
In early August each year, per cent cover of all vascular plants, bare mineral soil, bryophytes and litter were estimated in the central 0.7 m × 0.7 m area of each plot by the same observer (JC).One hundred and six plant taxa were recorded across all plots and years.We measured light availability and volumetric soil moisture content (surface and subsurface) four times throughout each growing season.
We collected and mixed four soil cores (10 cm deep) from each plot in June and July each year, extracted the samples with 0.01 M KCl and measured extractable soil nitrate, the key limiting resource at Cedar Creek (Tilman, 1990).Appendix S1 includes further details.

| Statistical analysis
Addressing our research questions, we used linear mixed-effects models to determine main and interactive effects between our response and explanatory variables, and variance components analysis (using Bayesian hierarchical multilevel models) to determine the relative importance of those relationships.We used measures of resource availability (light at ground level, soil moisture, soil nitrate) and, where relevant, resident species richness to help understand the trends observed.All analysis was completed using R version 4.2.2 (R Core Team, 2022) and JAGS version 4.3.0(Plummer, 2003).Appendix S3 includes additional details.Data and R code are available in Dryad Digital Repository (Catford et al., 2023).

| Four response variables
To address our first question about colonist abundance conditional on presence, we examined three response variables: fast colonist cover; slow colonist cover; and total colonist cover, i.e. fast + slow combined.The group-level covers were not overly dominated by particular species (Figure S6).We use 'colonist' to refer to experimentally sown species only; 'colonist cover' thus does not include non-sown species that have naturally colonised the plots.Our colonist cover analyses only included plots where colonists were sown and had colonised (Appendix S3).All colonist covers were logtransformed for analysis.
To address our second question about resident diversity, we examined one response variable: resident effective species richness (following Lisner et al., 2023;Tilman et al., 2001).Resident effective species richness is defined as the exponential of the Shannon diversity index, H � = − ∑ S i=1 p i ln p i , where p i is the proportional cover of the ith resident (non-sown) species and s is the total number of resident species in each plot.Effective species richness tries to fairly weight species based their relative abundances; it equates to the number of equally abundant species needed to give a value of H′.It should thus be more sensitive to disruption of a community than species richness (i.e.number of species).We define the resident community as all species that were not experimentally sown even if they colonised the plots during the experiment.Our target (i.e.sown) colonists rarely colonised plots where they were not added (Table S2).Resident effective species richness was strongly correlated with resident richness (r = 0.78) and community effective species richness (i.e.calculated using both colonists and residents, r = 0.77), but not with total resident cover (r = 0.28, Table S7).We focus on resident effective species richness (hereon resident diversity) and not resident cover as it corresponds more closely with our interests (Catford, Vesk, et al., 2012).The resident diversity model included all plots where colonists were sown, regardless of whether colonists were present.Resident diversity was square-root transformed for analysis.

| Step 1: Linear mixed-effects models
We used linear mixed-effects models to investigate the relationship between our four response variables and study explanatory variables (details in Appendix S3).For the three colonist cover models, we considered succession stage, disturbance, seed rate, colonist type, growing season and all possible two-way interactions as fixed effects.For the resident diversity model, we considered succession stage, disturbance, colonist type, colonist richness, colonist cover, growing season and all possible two-way interactions as fixed effects.To account for repeat plot measurements and variation in individual plots over the three surveys, we included plot as a random intercept and growing season as an interacting random slope in all models (random effects).Models were implemented through R package 'nlme' (Pinheiro et al., 2022).
For each response variable, different combinations of the study explanatory variables were included in candidate models, ranging from single terms only up to all five terms for colonist cover and all six terms for resident diversity, including all possible two-way interactions.This resulted in fitting 1450 models for each of the three colonist cover response variables and 40,069 models for the resident diversity response variable.We used the Bayesian (Schwarz) information criteria (BIC) to rank models for each response variable and considered all models within 6 BIC points of the best performing model (lowest BIC).For each response variable, the competing models (<6 BIC points) were either subsets of the best model or had a better performing model nested within them.Because the ecological interpretation of the competing models is effectively the same, we concentrate on the best performing models for simplicity (competing models are reported in Tables S3-S6 and S8-S11).
For the best models, we calculated marginal and conditional R 2 values, representing total variance explained by the entire model (fixed and random effects) and fixed effects only (Nakagawa & Schielzeth, 2013).
We included succession stage as a fixed effect in our models despite the split-plot experimental design because succession stage was of particular interest and because we only examined three stages (Bolker, 2015).Succession stage (and its nested structure) was included in every best performing (and competing) model, so this grouping-level factor was always incorporated into our analyses.Residuals for succession stage were uncorrelated, indicating no problems with independence (Schielzeth & Nakagawa, 2013).
We did not include initial resident diversity in the colonist cover models, nor measures of initial resource availability in either the colonist cover and resident diversity models, because of high collinearity between these variables and the experimental treatments (Table S7).Preliminary analyses showed that models incorporating successional stage outperformed models incorporating initial resident diversity and resource availability.By focusing on the experimental treatments, we were therefore able to account for differences in initial conditions while explaining greater variation in colonist abundance and resident diversity.

| Step 2: Variance component analysis
Variance component analysis was used to compare the relative importance of each fixed effect included in the best performing linear mixed-effects models from Step 1.We built a Bayesian hierarchical multilevel linear model for each response variable, implemented through R package 'R2Jags' (Su & Yajima, 2021), following Gelman and Hill (2007), Hector et al. (2011) and Catford et al. (2014).These models were produced by taking the structure of the best performing mixed-effects models for each response variable and replicating this in JAGS (Just Another Gibbs Sampler) multilevel structure, with variable intercepts and slopes applied following Gelman and Hill (2007).Variance components were then calculated from these Bayesian multilevel models and presented on a standard deviation scale to aid comparison between predictors (Gelman & Hill, 2007;Hector et al., 2011).

| Resource availability and resident diversity across treatments
We calculated paired Pearson's correlation coefficients between treatments (excluding colonist type), resources (soil nitrate, soil moisture, light at ground level) and resident community attributes (resident diversity, resident richness, total resident cover).We focus on conditions in the first growing season (2013) as these would have strongly affected colonisation, whereas conditions in the second and third growing seasons (2014 and 2015) would be more likely to also reflect effects of colonisation, complicating likely causal relationships.

| Cover abundance of sown colonists when present
Apart from relationships between sown colonist cover and disturbance, observed main effects between colonisation and the five explanatory variables were largely consistent with expectations (Figures 1-5).However, most relationships involved interactions (Figure 5a-c; Tables S3-S5 and S8 S3 and S5).Colonist type (both groups or target group only) was the only experimental treatment not included in the best model for fast colonists, which had a marginal R 2 of 0.44 and conditional R 2 of 0.80.The high R 2 values indicate that our study captured the key drivers of colonisation for our 14 sown colonist species, especially in the case of slow colonists.Fast and slow colonists differed in their responses to the experimental treatments, especially to disturbance (Figure 2f), so we primarily focus on the results of each of these two groups rather than total colonist cover (i.e. sum of both groups), which generally showed intermediate trends of the two groups (Figures 2-4).
Of the experimental treatments, fast colonist cover was most strongly linked with succession stage (Figure 3a), which itself was strongly linked with extractable soil nitrate (r = −0.86; Figure S7; Table S7), the dominant limiting resource in these grasslands (Tilman, 1990).Succession stage was also correlated with 2013 resident richness (r = 0.68; Table S7).Limited soil nitrate therefore likely explained the low fast colonist cover in late successional communities (Figure 3c), an effect that was partially alleviated when plots were disturbed (Figure 3d), temporarily increasing soil nitrate (Figure S7).Early successional plots had consistently high levels of colonisation across the three growing seasons (Figures 2c,   3c and 4c) despite marked declines in soil nitrate (Figure S7), suggesting that soil nitrate did not limit fast colonist cover in early succession, at least not in undisturbed plots.The non-conditional positive relationship between fast colonist cover and seed rate demonstrated that seed rate was important across all conditions, including in early succession where fast colonists were otherwise unconstrained (Figure 3a-c).Collectively these results suggest that seed availability limited cover of fast colonists in early succession, at least in undisturbed plots (Figure 3e).Fast colonist cover reached intermediate levels in mid succession, where disturbance had a seemingly neutral effect on fast colonists, neither increasing nor decreasing their cover.
Like fast colonist cover, slow colonist cover also appeared to be primarily limited by soil nitrate in late succession, as demonstrated by the positive effect of disturbance on slow colonist cover in late succession combined with trends across the three successional stages (Figure 4c,d).However, unlike fast colonists, there was a strong link between slow colonist cover and growing season (Figures 4a   and 5b,c), with a marked increase in slow colonist cover in early and mid successional plots from growing season 1 to 3 (Figure 4c).
Evidence suggested that fast colonists limited cover of slow colonists (Figure 4b) but co-occurring slow colonists did not affect fast colonist cover (Figure 3; Tables S9 and S10).Summed cover of colonists was similar in plots sown with fast colonists only (seven species, i.e. fast colonist cover) to plots sown with both fast and slow colonists (14 species, i.e. total colonist cover; Figure 2f).Plots with slow colonists only (seven species) had lower cover (Figure 2f).

| Resident diversity
Resident diversity, as indicated by effective species richness of nonsown species, was most strongly related to cover of sown colonists and community successional stage (Figure 6a).The direction of relationships was as expected, except for colonist type and colonist richness (Figure 5d), which were not included in the best model for, and had only negligible relationships with, resident diversity (Tables S6   and S11).Colonist richness was included in one of the competing (high performing) models, but its relationship with resident diversity was negligible (Table S11).Colonist type was not included in any of the best performing models, indicating that the groups of fast and slow colonists examined in our study had similar relationships with resident diversity.S6 and S11).Resident diversity was highest in late succession and lowest in early succession, but these differences declined across the three growing seasons as richness decreased slightly in late succession and increased in mid and, to a lesser extent, early succession (Figure 6c; Figure S8).Resident diversity was higher in disturbed than undisturbed plots (Figure 6d).
Although not definitive, changes in resident diversity across growing seasons in plots with different levels of colonist cover can help to illustrate the direction of causality between colonist abundance and resident diversity.Apart from disturbed late succession plots, resident diversity was similar in plots with high (≥10%) and low (<10%) colonist cover in the first growing season of the experiment (2013, Figure S8).However, by the third growing season (2015), plots with high colonist cover had lower resident diversity than plots with lower colonist cover in both early and mid succession.Resident diversity was similar regardless of colonist cover in undisturbed late succession plots.In disturbed late succession plots, diversity in plots with high colonist cover declined between growing season 1 and 3 (Figure S8).These temporal trends are consistent with effects of colonisation on resident diversity rather than effects of resident diversity on colonisation, though the relationship can go in both directions and is hard to categorically disentangle (Catford et al., 2020;MacDougall et al., 2014).The best performing model for resident diversity had a marginal R 2 of 0.50 and a conditional R 2 of 0.78.

| Resource availability
Based on measurements in 2013 (the first year of the experiment), community successional stage was positively correlated with resident richness (r = 0.679) and negatively correlated with soil nitrate (−0.861;Table S7a).Effects of disturbance on soil nitrate varied with successional stage (Table S7b- S12).+ symbol and red arrow = positive relationship; − symbol and blue circle = negative relationship; ≠ symbol and black line with blunt arrow = main effect that lacks a sign; absence of symbols and a solid line = no discernible relationship found in the top performing models.
successional stage and growing season, and were higher in disturbed than undisturbed plots (r = 0.711), especially in mid and late successional communities (r = 0.90 and r = 0.95 respectively; Table S7; Figure S9).Soil moisture was highest in late successional communities and lowest in mid successional communities (Figure S10).Disturbed plots had lower soil moisture than undisturbed plots, especially in mid successional communities (r = −0.71)where soil moisture was already very low (Figure S10).Apart from soil moisture, these correlations provide evidence that competition generally increased and availability of key resources, namely soil nitrate, decreased with succession, as found in other studies (see summary in Tilman, 1994).

| DISCUSS ION
Using a 3-year grassland seed addition experiment, we find evidence that the total abundance of 14 species of experimentally sown colonising plants, when present, is most strongly influenced by the successional stage of recipient communities, though growing season was also important for colonists with slower life history characteristics (Figure 5).The importance of successional stage likely reflects influences of resident diversity and resource availability, which respectively increase and decrease with succession (Table S7), as well as related differences in community composition (Appendix S2-Figure 1; Figure S4).Colonist type (fast colonists, slow colonists or both), seed rate and disturbance had weaker relationships with colonist abundance.Factors affecting colonist abundance were highly conditional, with changes in both the magnitude and sign of relationships depending on conditions (Figure 5a-c).Particularly noteworthy was the change in sign of the relationship between disturbance and colonist abundance depending on the successional stage of recipient communities.The negative relationship between disturbance and colonist cover in early successional communities meant that, when considering main effects only, disturbance had a negative relationship with colonisation overall, defying expectations (Figures 1a and 5a).
In contrast to colonist abundance, factors affecting resident diversity generally did not depend on the range of conditions examined in our experiment; apart from an interaction between growing season and succession, notable two-way interactions were absent (Figure 5d).There was a negative relationship between colonist abundance and resident diversity.This relationship was as strong as the relationship between community successional stage and resident diversity.The direction of causality between low resident diversity and high colonist abundance is hard to ascertain, and evidence from this and other studies indicate that lower resident diversity enables higher colonisation (Catford et al., 2020;Fargione & Tilman, 2005;MacDougall et al., 2014).However, given that resident diversity effectively declined in communities with high colonist cover relative to communities with low or no colonist cover except in undisturbed late succession plots (Figure S8), the negative colonisation-diversity relationship observed suggests that colonists did reduce or limit resident diversity under most of the conditions we examined.The findings of our 3-year experiment point to a legacy of agricultural land use on grassland communities.Communities at earlier stages of succession, which had been abandoned from agriculture more recently, had lower resident diversity and experienced higher levels of colonisation than communities at later stages of succession.
Although longer terms trends may differ, trends based on this 3-year period suggest that elevated colonisation and lower resident diversity were both facilitated by human-induced environmental change.
Evidence also suggested that plant colonisation may limit or potentially reduce resident diversity further, compounding diversity loss that stems from historical land use change.Even without considering changes in functional community composition, which provide a more sensitive indication of biodiversity change (Hillebrand et al., 2018), results of our experiment suggest that: (i) a rise in colonists will not necessarily compensate for a decline in resident diversity caused by land use change; and (ii) colonists that reach high abundance may be a further threat to resident plant diversity in secondary grasslands recovering from a recent history of agriculture.An extended survey period will be required to see if these short-term trajectories play out long term.

| Community successional stage and disturbance reveal importance of resource availability on colonist abundance
Consistent with our expectations (Figures 1a and 5a-c) and competition and succession theory (Huston & Smith, 1987;Tilman, 2004), the early successional communities experienced higher levels of colonisation than mid or late successional communities, and their levels of colonisation increased over the three growing seasons, especially in the case of slow colonists.Colonisation trends in mid and late successional communities were more complicated, with their ranks changing depending on colonist type and growing season.Collectively, these dynamics seem to reflect colonist life histories and the strength of resource competition in the recipient communities-the latter of which increases with changes in diversity and community composition (Appendix S2-Figure 1; Figure S3), specifically increased dominance of more competitive species at later stages of succession (Figure S4) (Clark et al., 2019;Isbell et al., 2019).Other studies have also found that colonisation levels decline as the diversity of recipient communities increases (Beaury et al., 2020;Hector et al., 2001;Petruzzella et al., 2018) (although not universally, MacDougall et al., 2014) and as limiting resources become scarcer (Catford et al., 2020;Fargione & Tilman, 2005;Seabloom, 2011).
Fast colonists were able to establish quickly, but their cover declined notably in late successional communities, presumably as competition-driven resource limitation took effect (Figures S7 and S9).Slow colonists are more resource competitive and conservative and have slower rates of growth than fast colonists (Lauenroth & Adler, 2008); these attributes presumably helped slow colonists persist in late successional communities and gradually increase in cover across the three growing seasons in midsuccessional communities.
The importance of resource limitation was further highlighted by interactive effects of disturbance and community succession stage on colonist cover (Figures 2d,3d and 4d).
Trends indicated that disturbance facilitated colonisation in late successional communities but inhibited colonisation in early successional communities (Figures 2d,3d and 4d).In late succession, disturbance may have provided an opportunity for colonists to recruit before more competitive resident plants (e.g.perennial C4 grasses S. scoparium, A. gerardii and S. nutans; Figure S4) regrew from rhizomes (rhizomes would have likely been fragmented but not killed by soil tillage).We expected that the faciliatory role of disturbance on colonisation (Figure 1a) would strengthen with succession reflecting greater resource limitation in later stages of succession (Catford, Daehler, et al., 2012).However, the negative disturbance-colonisation relationship in early succession indicates that disturbance can hinder, as well as help, colonisation depending on local conditions (Figures 2d, 3d and 4d).Had we only considered main effects (Figure 5a-c) or only examined colonisation in early successional communities, the effect of disturbance in our study would appear to contradict ecological theory when, in fact, our findings are consistent with hypothesised disturbancecolonisation mechanisms.The succession-dependent effects of disturbance illustrate the importance of explicitly examining interaction effects (Catford et al., 2022), and may partially explain why observed relationships between disturbance colonisation are variable (Moles et al., 2012).
Disturbance is usually hypothesised (and often found; Seabloom, 2011; Kempel et al., 2013) to facilitate colonisation because it decreases the biomass of competing established plants and increases resource availability (Catford et al., 2009;Davis et al., 2000;Lear et al., 2020)-as was the case with soil nitrate in mid and late succession, albeit only for the first growing season (Figure S7).Resource competition tends to be weaker earlier in succession (Catford, Daehler, et al., 2012;Clark et al., 2019;Lohbeck et al., 2014) and in less diverse communities (Catford et al., 2020;Fargione & Tilman, 2005), which may restrict the positive effects of disturbance on colonisation, enabling negative effects to be seen.By destroying standing biomass and disrupting soil structure, physical disturbance (like tillage used in this experiment) can alter microclimates, potentially exposing seedlings to dry and hostile conditions and reducing the number of safe sites for recruitment (Wandrag et al., 2019(Wandrag et al., , 2023)).Indeed, disturbed plots had higher light and lower surface soil moisture than undisturbed plots (Figures S9 and S10).However, disturbance reduced soil nitrate in our early successional communities in the peak growing season (June and July), which could plausibly be the explanation for the negative effect of disturbance on colonist cover in early succession.Disturbance may have facilitated higher rates of seed predation and herbivory too (Korell et al., 2017;MacDougall & Wilson, 2007;Martyn et al., 2023).
We posit that disturbance had a largely neutral effect in mid successional communities because of its opposite effects on soil nitrate and soil moisture (Figures S7 and S10; Table S7c).The benefit of increased nitrate availability may have been annulled by reduced water availability (which was already very low in the mid successional plots), prompting a shift from nitrate to water limitation with disturbance (though water limitation can itself cause N limitation because of reduced access to nutrients in the soil; Bloom et al., 1985).Light is rarely limiting in unfertilised grasslands that occur on the sandy, low-nitrogen soils of Cedar Creek (Tilman, 1990;Wilson & Tilman, 1991), so any benefits of increased light availability with disturbance in this experiment were likely to be secondary to benefits of increased nitrate availability.This explanation accords with findings of Lear et al. (2020) who experimentally showed that disturbance facilitates colonisation of new bacteria morphotypes by increasing resource availability rather than by opening habitat.

| Colonists likely impact resident diversity and other colonists via resource competition and niche pre-emption
The observed decline in non-sown resident diversity with increasing sown colonist cover is consistent with effects of resource competition and resource limitation.The colonisation-diversity relationship did not vary with colonist type, despite slow colonists being better able to persist in the more diverse late successional communities (Figures 3c and 4c) and their potential to reduce soil nitrate and water to lower levels than fast colonists.
Fast colonists appeared to limit cover of slow colonists when the groups were sown together (Figure 4b).This was not one of our a priori expectations, but likely reflects niche pre-emption where the more abundant fast colonists colonised (safe) sites more rapidly than slow colonists (Catford, Daehler, et al., 2012;Wandrag et al., 2019).This may have been particularly the case with Nfixing Lupinus perennis, which could entirely cover some plots (Figure S11), shading other plants, among other potential effects (e.g.plant soil feedbacks; Bakker et al., 2013).No individual slow colonist species were especially abundant (Figure S11), and there was probably insufficient time to detect competitive effects of slow colonists in our 3-year experiment.With more time for population growth, and as their trends across growing seasons suggest (Figures 2c and 3c), slow colonists will likely replace or displace shorter-lived fast colonists (Lauenroth & Adler, 2008) and may have a greater impact on resident diversity than fast colonists, provided that disturbance is rare and no more seeds are sown (Tilman, 1990).
The negative relationship between colonist cover and resident diversity did not depend on colonist type or local conditions (for two-way interactions, at least), suggesting that resource competition was important under all conditions examined here.Our experimental design (including its spatial scale and 3-year time span) meant that we were more likely to detect competition-based effects of colonists on resident diversity than other types of effects (e.g. via enemies, mutualists, abiotic conditions) (Kempel et al., 2013;Levine et al., 2004;Petruzzella et al., 2020).In other situations, colonist impacts may be more variable, reflecting a greater diversity of factors that limit the diversity of resident species and how colonists may affect those limiting factors (Farrior et al., 2013;Harpole et al., 2016).Vegetation in the early succession site was killed immediately before the experimental treatments were applied in 2013, so there was no pre-treatment vegetation in that site.Table S1.Characteristics, occupancy and abundance of 23 species added to the experimental plots, 14 of which are the target colonists.

| CON CLUS IONS
Occupancy was calculated as percentage of plots in which each species was recorded during the vegetation surveys relative to the number of plots where the species was sown.Calculations of mean and maximum cover included all plots involved in the total colonist cover and resident diversity models provided a colonist species was present.Spread (or seed contamination) of seeded colonists between plots was low during the study period (see Table S2), such that different ways to calculate abundance resulted in similar means (and identical maximums).Strophostyles leiosperma colonised some plots but, due to an unbalanced design, germinating individuals were removed and records not included in analysis (see Methods in main text for more detail).Other species that were sown but did not colonise the plots are shown.
Table S2.Occupancy and abundance of seeded colonist species in control plots.Eight species were never observed in any plots, even where directly sown.Strophostyles leiosperma was only added to a fraction of designated plots and any germinating were hand-pulled to avoid an imbalanced design (see Methods in main text for more information).After these exclusions, 14 colonist species remained.Table S3.Total colonist cover model summary.Summary table of the fixed effects in the best performing model examining the effects of the experimental treatments on total colonist cover.The noninteraction model indicates the best performing model where all experimental treatments were considered, but not their interactions.
The interaction model indicates the best performing model where all experimental treatments and their two-way interactions were considered.

Our 3 -
year grassland experiment took place in three old field sites within a 2 km radius at Cedar Creek Ecosystem Science Reserve (hereafter Cedar Creek, 45.4° N, 93.2° W), Minnesota in 2013-2015.
When the treatments were imposed in spring 2013: the late succession site was approximately 80 years old and dominated by Schizachyrium scoparium, Sorghastrum nutans, Andropogon gerardii and Poa pratensis; the mid succession site was 23 years old F I G U R E 1 Hypothesised relationships between explanatory variables investigated in this experimental study and (a) abundance of sown colonists when present, and (b) non-sown resident diversity.Symbols and dashed lines indicate hypothesised relationships (main effects only; interactions are not shown).+ symbol and red arrow = positive relationship; − symbol and blue circle = negative relationship.Variable interpretation: Colonist type-fast to slow, i.e. slow colonists are hypothesised to reach lower abundance (a) but have greater negative impacts on resident diversity (b) than fast colonists; Succession stage-early to late, i.e.
-S10).The most marked interactions for all colonist groups(total, fast, slow; Figures 2a, 3a and   4a)  were between community successional stage and disturbance (Figures2d, 3d and 4d), and succession stage and growing season (Figures2c, 3c and 4c), where the sign of the relationships changed depending on conditions (Figure5a-c).All other interactions involved a change in the magnitude of the relationship.All four experimental treatments plus growing season were included in the best model for total and slow colonist cover, which explained considerable variance (total colonist cover: marginal R 2 = 0.65 and conditional R 2 = 0.85; slow colonist cover: marginal R 2 = 0.83 and conditional R 2 = 0.92, Tables

F I G U R E 2
Relationships between the total cover of sown colonists (fast + slow), when present, and explanatory variables.Left panel (a): Variance components for the best performing model of total colonist cover (log-transformed), produced through Bayesian multilevel modelling, showing relative importance of relationships.Point estimates are set on a standard deviation scale and represent the medians of the posterior distributions.Bars present 95% (wide) and 68% (narrow) intervals.Right panels (b-f): Relationships between total cover of sown colonists (log-transformed) and (b) growing season (separated by seed rate), (c) growing season (separated by succession stage), (d) succession stage (separated by disturbance), (e) growing season (separated by disturbance) and (f) sown colonists (separated by disturbance), fitted from the best performing linear mixed-effects models.Envelopes represent 95% confidence intervals.Points represent partial residuals.Panels with coloured border indicate variable involved in two-way interaction (e.g. between growing season (dark green border) and seed rate (red-toned lines) in (b)), with dark green border indicating growing season, blue border indicating succession stage and pink border indicating colonist type.Relationships between the cover of experimentally sown fast colonists, when present, and explanatory variables.Left panel (a): Variance components for the best performing model of fast colonist cover (log-transformed), produced through Bayesian multilevel modelling.Point estimates are set on a standard deviation scale and represent the medians of the posterior distributions.Bars present 95% (wide) and 68% (narrow) intervals.Right panels (b-e): Fitted relationships between fast colonist cover (log-transformed) and (b) seed rate of colonists, (c) growing season (separated by succession stage), (d) succession stage (separated by disturbance) and (e) growing season (separated by disturbance), fitted from the best performing linear mixed-effects models.Envelopes represent 95% confidence intervals.Points represent partial residuals.Panels with coloured border indicate variable involved in two-way interaction, with dark green border indicating growing season and blue border indicating succession stage.Relationships between the cover of experimentally sown slow colonists, when present, and explanatory variables.Left panel (a): Variance components for the best performing model of slow colonist cover (log-transformed), produced through Bayesian multilevel modelling.Point estimates are set on a standard deviation scale and represent the medians of the posterior distributions.Bars present 95% (wide) and 68% (narrow) intervals.Right panels (b-e): Relationships between slow colonist cover (log-transformed) and (b) sown fast colonists, (c) growing season (separated by succession stage), (d) succession stage (separated by disturbance) and (e) growing season (separated by seed rate), fitted from the best performing linear mixed-effects models.Envelopes represent 95% confidence intervals.Points represent partial residuals.Panels with coloured border indicate variable involved in two-way interaction, with dark green border indicating growing season and blue border indicating succession stage.
negatively related to colonist cover (Figures 5d and 6b).The relationship did not depend on colonist type, colonist richness, succession stage, disturbance or growing season (Figure 6; Tables d; Figure S7); disturbed plots had lower soil nitrate than undisturbed plots in early succession (r = −0.42),but higher soil nitrate in mid and late succession (r = 0.64 and 0.44 respectively).Light levels at ground level declined with F I G U R E 5 Visual summary of results relative to hypotheses for (a) total colonist abundance, (b) fast colonist abundance, (c) slow colonist abundance, when colonists were present, and (d) resident diversity.Dashed lines indicate hypothesised relationships between response and explanatory variables (Figure 1; Section 1); symbols and solid lines indicate observed relationships (Figures 2-4 and 6) with line and arrow thickness scaled by proportion of variance components attributed to each variable (where interactions are present, variance component is divided between the two relevant variables; Table Figure was drawn based on hypotheses shown in Figure1and described in the Introduction, and results from the best models for each response variable, which are presented in Figures 2-4 and 6.Grey nodes show four response variables; coloured nodes show explanatory variables, with colour coding consistent with other figures.Variable interpretation: Colonist type-fast to slow, i.e. slow colonists reach lower cover than fast colonists; Succession stage-early to late, i.e. late successional plots have lower colonist cover but higher resident diversity than early successional plots; Disturbance-undisturbed to disturbed, i.e. disturbed plots have lower colonist cover but higher resident diversity than undisturbed plots; all other factors-low to high.Coloured circles with letters indicate interactions with other variables (circle with outline = interaction involves a change in sign of the relationship; circle without outline = interaction involves a change in relationship magnitude), where G = growing season, S = succession stage, D = disturbance, R = seed rate, C = colonist type.Strength of interactions is not represented.'Colonist' refers to experimentally sown species only and excludes non-sown 'natural' colonists.

F
I G U R E 6 Relationships between non-sown resident diversity, as indicated by resident effective species richness (ESR), and explanatory variables.Left panel (a): Variance components for the best performing model of resident effective species richness (ESR, square-root transformed), produced through Bayesian multilevel modelling.Point estimates are set on a standard deviation scale and represent the medians of the posterior distributions.Bars present 95% (wide) and 68% (narrow) credible intervals.Point estimate scale differs to Figures 2a, 3a and 4a due to different response variable transformations.Right panels (b-d): Relationships between resident diversity (square-root transformed ESR) and (b) total cover of sown colonists (scaled), (c) growing season (separated by succession stage) and (d) disturbance, fitted from the best performing linear mixed-effects model.Envelopes represent 95% confidence intervals.Points represent partial residuals.Panels with coloured border indicate variable involved in two-way interaction, with dark green border indicating growing season.
By manipulating the seed rate of experimentally sown colonists, imposing disturbance and examining temporal trends of colonists with different life history characteristics in communities at different stages of succession, we were able to examine effects of global change factors that are usually confounded.Our findings suggest that variation in successional stage of grasslands caused by time since agricultural abandonment has the greatest potential to affect plant colonisation and resident diversity overall, at least over short time periods immediately following colonist introduction.Land use change that resets succession reduces resident diversity and facilitates colonisation of new species, the latter of which likely puts further pressure on resident diversity, compounding the negative effects of agricultural land use.While potentially a source of biodiversity in the short term, our findings indicate that high levels of plant colonisation could pose an additional threat to resident plant diversity in secondary grasslands that are recovering from recent agricultural land use.

Figure S4 .
Figure S4.Per cent cover of the most abundant resident species observed in seeded plots separated by site successional stage.Top panel: Early successional stage.Middle panel: Mid successional stage.Bottom panel: Late successional stage.Resident species included at each stage are those that were present within a minimum of 10% of all vegetation surveys and had >3% mean cover when present.Note that only two species are among the most common species in more than one site: Leptoloma cognatum in early and mid succession; and Poa pratensis in mid and late succession.

Figure S5 .
Figure S5.Ordination diagram of principal component analysis of seven functional traits of the 14 colonist species.Overlapping species names are slightly jittered for legibility.Fast colonists in red and slow colonists in teal.

Figure S6 .
Figure S6.Violin plots showing percentage of (a) fast and (b) slow colonist group cover made up of their component species when present.

Figure S7 .
Figure S7.Mean summer soil nitrate (NO 3 ) concentrations for each experimental treatment.Panels are split by seed rate (top bar) and colonist type (bottom bar).Colours indicate community succession stage.Solid points and lines are undisturbed, dashed points and lines are disturbed.Error bars represent ± SE.Soil nitrate concentrations were measured twice each year (mid-June and mid-July).

Figure S8 .
Figure S8.Mean resident diversity, measured as resident effective species richness, across growing seasons relative to total cover of seeded colonists in the target plot.Colours indicate total cover of colonists in the target plot, where each plot is classified into colonist cover categories within a given growing season: control, control plots where colonist cover = 0%; low, colonist cover < 1%; medium, 1% ≥ colonist cover < 10%; high, colonist cover ≥ 10%.Panels are split by succession stage (rows) and disturbance treatment (columns).Error bars represent ±1 SE.The figure does not show pre-treatment resident diversity (i.e.before seed addition or disturbance), so trends in season 1 (as well as seasons 2 and 3) may also capture effects of colonisation.

Figure S9 .
Figure S9.Light at ground level (below vegetation and litter, %) for each experimental treatment.Colours indicate succession stage.Solid points and lines are undisturbed, dashed points and lines are disturbed.Panels are split by seed rate (top bar) and colonist type (bottom bar).Error bars represent ± SE.Light was measured four times each summer (approximately every 3 weeks from early June to the end of August).

Figure S10 .
Figure S10.Mean summer soil moisture content (%) for each experimental treatment.Panels are split by seed rate (top bar) and colonist type (bottom bar).Colours indicate succession stage.Solid points and lines are undisturbed, dashed points and lines are disturbed.Error bars represent ± SE.Soil moisture was measured four times each summer (approximately every 3 weeks from early June to the end of August).

Figure S11 .
Figure S11.Species-specific contributions to colonist cover in experimental plots with the highest total (i.e.summed) colonist effects in the best performing model examining the effects of the experimental treatments on fast colonist cover.The non-interaction model indicates the best performing model where all experimental treatments were considered, but not their interactions.The interaction model indicates the best performing model where all experimental treatments and their two-way interactions were considered.
fixed effects in the best performing models examining the effects of the experimental treatments on slow colonisation cover.The noninteraction model indicates the best performing model where all experimental treatments were considered, but not their interactions.The interaction model indicates the best performing model where all experimental treatments and their two-way interactions were considered.

Table S4 .
Fast colonist model summary.Summary table of the fixed

Table S5 .
Slow colonist cover model summary.Summary table of the

Table S6 .
Resident effective species richness (ESR) model summary.Summary table of the fixed effects in the best performing models