Regional climate and local-scale biotic acceptance explain native–exotic richness relationships in Australian annual plant communities

Abstract

Native and exotic species richness is expected to be negatively related at small spatial scales where individuals interact, and positive at larger spatial scales as a greater variety of habitats are sampled. However, a range of native–exotic richness relationships (NERRs) have been reported, including positive at small scales and negative at larger scales. We present a hierarchical metacommunity framework to explain how contrasting NERRs may emerge across scales and study systems, and then apply this framework to NERRs in an invaded winter annual plant system in southwest Western Australia. We analysed NERRs at increasing spatial scales from neighbourhoods (0.09 m²) to communities (225 m²) to metacommunities (greater than 10 ha) within a multilevel structural equation model. In contrast to many previous studies, native and exotic richness were positively related at the neighbourhood scale and were not significantly associated at larger scales. Heterogeneity in soil surface properties was weakly, but positively, associated with native and exotic richness at the community scale. Metacommunity exotic richness increased strongly with regional temperature and moisture availability, but relationships for native richness were negative and much weaker. Thus, we show that neutral NERRs can emerge at larger scales owing to differential climatic filtering of native and exotic species pools.

1. Introduction

Early theories in invasion ecology proposed that highly diverse communities should have greater biotic resistance to invasion because niche space is more completely occupied, increasing the likelihood that invaders will be competitively excluded [1,2]. Thus, areas with high native diversity are expected to have low exotic diversity and vice versa. Indeed, many fine-scale experimental studies have reported negative relationships between native and exotic species richness in a variety of environments [3–5]. However, observational studies at larger spatial scales tend to reveal positive native–exotic richness relationships (NERRs; [6]). This apparent contradiction has been termed the ‘invasion paradox’ [6], and attempts to explain it have led to many studies explicitly investigating NERRs at multiple scales.

Two main hypotheses have been erected to explain positive NERRs: the environmental heterogeneity and environmental favourability hypotheses. The first states that positive NERRs emerge because native and exotic diversity both positively covary with increasing variation in environmental conditions [7]. The second hypothesis, also referred to as ‘biotic acceptance’, posits that favourable conditions for native species are also favourable for exotic species [8]. While both hypotheses have received empirical support, deeper insights can probably be gained by explicitly considering the processes and mechanisms underlying NERRs at different scales [6].

Metacommunity theory [9] provides an ideal framework to examine NERRs at different scales [10]. The smallest scale is the interaction neighbourhood (or patch), where individuals interact and conditions are assumed to be spatially homogeneous [9]. Dispersal is high and interaction outcomes are largely determined by resource partitioning [6]. At the next scale, communities comprise multiple interaction neighbourhoods that differ in environmental conditions but are linked by dispersal. At the largest scale, metacommunities comprise multiple communities that are also linked by dispersal, but dispersal is much lower. Dispersal among metacommunities (e.g. across regions) is expected to be lower again.

Using this framework, environmental heterogeneity among neighbourhoods can drive both negative and positive NERRs within communities depending on how native and exotic species respond to, and interact along, local-scale gradients. Negative NERRs can emerge if native and exotic species sort differentially along local gradients (figure 1a). Mechanistically, such sorting requires a specific combination of niche and fitness differences among native and exotic species [11], such as exotics filling empty spatial niches, or natives and exotics having fitness advantages in different neighbourhoods. Alternatively, heterogeneity can drive positive NERRs if some neighbourhoods are more favourable than others for both native and exotic species (figure 1a). This could emerge if exotic species facilitate native species [6], or if native and exotic species are able to partition resources within neighbourhoods.

Figure 1.

A hierarchical metacommunity framework for understanding NERRs within and across scales. In this example, soil depth is assumed to be the most important structuring variable within and among communities, although this variable is not included in our analysis. Within communities (a), negative NERRs emerge if native species (or just those remaining in the wake of invasion) and exotic species consistently occupy different spatial niches (different soil depths), whereas positive relationships emerge via local-scale biotic acceptance (more native and exotic species on deeper soils). Among communities (b), more favourable average conditions (deeper soils) and/or high environmental heterogeneity are theorized to maintain higher diversity of both native and exotic species, though if conditions become too favourable (i.e. highly productive) diversity is expected to decline in many systems. At the metacommunity scale (c), climate and other broad-scale environmental gradients become important. NERRs will be positive if native and exotic species respond similarly to a climate gradient, but if they respond differently to one or more gradients, then NERRs could be neutral or positive. Letters in (a) represent different species and colours indicate native and exotic species as indicated. We refer the reader to the main text for specific mechanisms that are probably involved in particular scenarios.

At the community scale, environmental heterogeneity is theorized to increase both native and exotic diversity (figure 1b) via heterogeneity-dependent coexistence mechanisms, including the spatial storage effect and fitness-density covariance [12]. But again, outcomes for native and exotic species (and community-scale NERRs) will depend on the balance of niche and fitness differences in each spatial context [11]. Mean environmental conditions may also influence community-scale NERRs. For example, if many neighbourhoods in a community are stressful for most species (e.g. very shallow soil, figure 1b), then strong abiotic filtering will reduce the diversity of both native and exotic species compared to communities with a lower proportion of stressful neighbourhoods. On the other hand, if conditions are productive and competition is mainly for light, invasion by resource-acquisitive exotic species can drive competitive exclusion and declines in both native and exotic species richness [13–15].

At the metacommunity scale (e.g. across regions), climate and other broad-scale factors (e.g. dispersal barriers) are likely to shape the composition and diversity of native and exotic species pools [16]. NERRs will be positive if native and exotic species are filtered similarly along climate gradients, but if they are differentially filtered along one or more gradients, then NERRs could be neutral or negative (figure 1c).

This study uses a metacommunity framework to examine NERRs at neighbourhood, community and metacommunity scales in winter annual plant communities growing in the York gum–jam woodlands in Western Australia. These woodlands are characterized by a patchy overstorey of Eucalyptus loxophleba (York gum) and Acacia acuminata (jam), and now persist as discrete remnants within a matrix of cropping and sheep pasture [17]. Native winter annuals and geophytes continue to dominate the understorey away from agricultural influences, but even under these conditions exotic species have invaded [18,19]. These mixed communities provide an opportunity to test hypotheses about invasion in the absence of direct anthropogenic influences. We used multilevel structural equation modelling to examine how relationships between native and exotic richness change with increasing scale, from neighbourhoods to metacommunities. Specifically, we ask three questions: (i) how do NERRs change across scales? (ii) how important are environmental conditions for driving native and exotic richness at each scale? and (iii) after controlling for environmental drivers of native and exotic richness, are NERRs still evident?

2. Material and methods

(a). Study region

This study uses data collected previously in the Avon Wheatbelt subregion of southwest Western Australia. Soils in this region are associated with the weathering and erosion of the underlying gneiss and granite block and are low in phosphorus [17,20]. The climate is mesic Mediterranean in the west and semi-arid in the east. During the winter wet season (June–October), mean daily temperatures range from 16°C in the south to 19°C in the north.

(b). Field survey

Field surveys were undertaken between 25 August and 10 October 2011. Full details of this survey can be found in Dwyer et al. [21]. In summary, within the natural distribution of the York gum–jam woodlands, 12 remnant patches were selected, spanning more than 3° latitude and 3° longitude, to capture variation in moisture availability and temperature in this region (electronic supplementary material, table S1). Three 15 m × 15 m (225 m²) sites were established within each remnant at least 80 m apart and more than 80 m from remnant edges in herb-dominated understorey vegetation. Sites were non-randomly located within remnants to capture variation in tree (E. loxophleba and A. acuminata) cover. Within each site, plant assemblages were sampled in 15 randomly located quadrats (0.3 m × 0.3 m).

Within each quadrat, the identity and abundance (counts of individuals) of all species was recorded. A soil sample (0–70 mm depth excluding litter) was collected from the centre of each quadrat following the inventory. Soil samples were air-dried and later analysed for extractable phosphorous (P), potassium (K), ammonium and nitrate. Ammonium and nitrate values were summed to approximate total plant-available nitrogen (N). In addition, soil hardness was estimated using a penetrometer (Lang Penetrometer, Inc.) and expressed as a proportion of the maximum value observed in each remnant on a given day [21] to account for soil moisture differences during the survey. Woody canopy cover was estimated using a spherical crown densitometer (Forestry Supplies Inc., Jackson, MS, USA). Sclerophyllous leaf litter was estimated by overlaying a 100-point grid on quadrat photos and counting ‘hits’. The presence of biological soil crusts (biocrusts) and coarse woody debris (CWD; all branches or logs greater than 5 cm diameter within or immediately adjacent to quadrats), were also recorded as binary variables. Interpolated climate data were extracted for each remnant [22] and used to calculate two orthogonal climate variables: moisture availability, expressed as the ratio of mean growing season precipitation to mean growing season potential evapotranspiration, and mean minimum temperature during the growing season. Plant density (number of individuals in each quadrat), and plant-available N, P and K were positively skewed and were natural log- or square root-transformed prior to statistical analysis.

(c). Community and neighbourhood definitions

Sites (15 m × 15 m) and quadrats (0.3 m × 0.3 m) were chosen in the initial survey design to correspond to community and interaction neighbourhood (herein ‘neighbourhood’) scales, respectively. While our communities are small compared to previous studies of NERRs [7], they spanned considerable environmental and compositional variation (figure 2) and were therefore sufficiently large to investigate environment–richness relationships. The spatial scale of neighbourhoods was selected to minimize within-unit environmental heterogeneity. We refer to remnants as ‘metacommunities’, because dispersal between them is low and they therefore correspond to regional metacommunities as described by Melbourne et al. [10].

Figure 2.

The three scales examined in this study and the multivariate hypotheses tested in the gmSEMs. Three 15 m × 15 m communities were established in each metacommunity. Within each community, 15 0.3 m × 0.3 m neighbourhoods were randomly located. In the gmSEM, single-headed arrows indicate the direction of causality of hypothesized relationships, while double-headed arrows indicate that a relationship is hypothesized, but there are mixed a priori expectations about the direction of causality. Refer to Ecological justification for the gmSEM structure for a detailed explanation of the hypothesized relationships between variables. Photographs by John. M. Dwyer.

Species richness at the neighbourhood scale was the number of species observed in each quadrat for both native and exotic species. Because of incomplete sampling at the community and metacommunity scales, we extrapolated native and exotic richness using Jackknife estimators, which have performed well in other species-rich ecosystems in southwest Western Australia [23,24]. In particular, we used a first-order Jackknife estimator because estimates produced by the second-order estimator were sometimes lower than the observed richness. To extrapolate richness at the community and metacommunity scales, we used species occurrence data from the constituent 15 and 45 neighbourhoods, respectively. For one metacommunity no exotic species were recorded in two communities, so we assigned the exotic richness estimate (1 species) from the third community. Sample-based rarefaction curves indicated that metacommunity-scale exotic richness was close to asymptotic in almost all cases, and while native richness was still rising, all curves extended well past the region of most rapid species accumulation (electronic supplementary material, figure S1). Herein, we refer to estimated richness values simply as richness. Because the spacing of communities within remnants was variable, we calculated the average distance between communities as a potential predictor of metacommunity richness (see below).

(d). Statistical analyses

We used a series of simple models to address our first question regarding ‘raw’ NERRs. We fitted these relationships in both directions, as we did to address questions 2 and 3 below. In the neighbourhood-scale model, we included metacommunity and community as nested random effects, allowing both the intercepts and slopes of NERRs to vary. We did the same in the community-scale model, but without the community random effect. Finally, the metacommunity model was a simple linear regression with no random effects. Neighbourhood exotic richness included zeros, so we used a negative binomial mixed-effects model when treating this as the response variable.

Generalized mixed-effects structural equation modelling (gmSEM) was used to address questions 2 and 3 [25]. We first developed a directed acyclic graph to represent the causal structure and direction of hypothesized relationships [26] and then assessed support for this structure using directed separation (d-sep) tests based on Fisher's C statistic [27]. Small C values, and consequently large p-values, indicate that the model fits the data well, whereas large C values, and small p-values (in this case p < 0.05), instead indicate that there are significant causal paths missing from the model and that the model fits the data poorly.

(i). Calculation of community-scale environmental predictors

To create variables that describe environmental heterogeneity within communities, standardized neighbourhood-scale environmental variables were expressed as multivariate dispersion variables using the betadisper function in the vegan package. Soil fertility heterogeneity included plant-available P, K and N, while physical heterogeneity included sclerophyll litter cover, soil hardness, biocrust and CWD. We also calculated the univariate dispersion of canopy cover because of its known influence on compositional turnover in the study system [28]. In addition, the distance of each community from the edge of the remnant was calculated as a proxy for exotic species' propagule pressure.

(ii). Ecological justification for the generalized mixed-effects structural equation modelling structure

At the metacommunity-scale, stress imposed by climate may influence native and exotic richness by reducing the number of species with strategies to maintain viable populations (figure 2; [21,29]). Metacommunity-scale native and exotic richness should then positively influence community-scale native and exotic richness, because diversity at the larger scale indicates the diversity of propagules available to colonize communities [30,31]. In turn, the same positive effects of diversity are expected from the community scale to the neighbourhood scale [9,32].

Beyond these ‘trickle-down’ diversity effects, environmental heterogeneity is hypothesized to positively influence both native and exotic richness at the community scale [10]. We did not include mean environmental variables at the community scale (i.e. means of 15 neighbourhood values) or the metacommunity scale (i.e. means of 45 neighbourhood values) because these were tested as missing paths in the gmSEMs. In addition, if propagule pressure from the agricultural matrix drives invasion in this system, then the distance from the remnant edge should be negatively related to community exotic richness (figure 2; [33]).

At the neighbourhood scale, patch conditions and climate may influence local plant density [34,35] which could, in turn, influence the number of species (native and exotic) that are present in a given neighbourhood. Patch conditions may also directly influence neighbourhood diversity through local-scale habitat filtering [36]. Native and exotic richness may then be correlated at each of the three spatial scales measured after taking these other factors into consideration (question 3).

(iii). Generalized mixed-effects structural equation modelling fitting and selection procedures

All response and explanatory variables other than neighbourhood exotic richness (discussed below) were standardized to permit comparison of coefficient estimates (mean = 0, s.d. = 1). The various paths were fitted as separate models and then analysed holistically within the gmSEM framework (table 1). Simple linear models were fitted to the metacommunity-scale data (12 observations), whereas linear mixed-effects models were used to model community and neighbourhood response variables, except for neighbourhood exotic richness which included zeros and was modelled using a mixed-effects negative binomial model. Metacommunity was included as a random effect in the community-scale models, and metacommunity and community were included as nested random effects in the neighbourhood scale models (540 observations in 36 communities in 12 metacommunities).

Table 1.

Predictor variables included in gmSEM component models for each response variable. (T_min stands for minimum temperature.)

scale	response	predictors
metacommunity (n = 12)	native richness	growing season Tmin + moisture availability + mean distance between communities + metacommunity exotic richnesse
	exotic richness	growing season Tmin + moisture availability + mean distance between communities + metacommunity native richnessn
community (n = 36)	native richness	Metacommunity native richness + soil fertility heterogeneity + physical heterogenetiy + canopy cover heterogeneity + community exotic richnesse
	exotic richness	Metacommunity exotic richness + soil fertility heterogeneity + physical heterogenetiy + canopy cover heterogeneity + distance from remnant edge + community native richnessn
neighbourhood (n = 540)	native richness	community native richness + neighbourhood plant density + soil potassium + soil phosphorous + soil nitrogen + CWD + biocrust + canopy cover + sclerophyll litter cover + soil hardness + exotic richnesse
	exotic richness	community native richness + neighbourhood plant density + soil K + soil P + soil N + CWD + biocrust + canopy cover + sclerophyll litter cover + soil hardness + native richnessn
	plant density	soil K + soil P + soil N + CWD + biocrust + canopy cover + sclerophyll litter cover + soil hardness + growing season Tmin + moisture availability

^eDenotes diversity variables that were only included in the ‘exotic-driven’ gmSEM.

ⁿDenotes variables that were only included in the ‘native-driven’ gmSEM. n indicates the number of observations at each scale.

Because native richness may influence exotic richness and vice versa, we fitted two gmSEMs, one with native richness predicting exotic richness (the ‘native-driven gmSEM’) and the other with exotic richness predicting native richness (the ‘exotic-driven gmSEM). Native-driven and exotic-driven gmSEMs were compared using Akaike information criterion corrected for small sample size (AICc) values [37].

Basis-set construction, model validation and parameter estimation were conducted using the piecewiseSEM package [38] in the R statistical program [39]. Component linear mixed-effects models were fitted using the nlme package [40] and mixed-effects negative binomial regressions were fitted using the lme4 package [41]. Diagnostic inspection of component models was performed using the DHARMa package [42].

3. Results

In total, 141 native species and 38 exotic species were recorded across the study region. At the metacommunity scale, respective estimates for native and exotic richness averaged 63.58 (s.e. = 3.29) and 14.52 (2.34), at the community scale they averaged 41.72 (1.38) and 9.84 (0.99), and at the neighbourhood scale they averaged 10.59 (0.15) and 1.65 (0.07), respectively.

(a). Question 1: how do native–exotic richness relationships change across scales?

We present results for models with native richness as response variables, but results were consistent when exotic richness variables were used (electronic supplementary material, figure S2). The overall NERR was positive at the neighbourhood scale, as were the separate slopes fitted for each community (figure 3a). The estimated NERR at the community scale was negative but the slope was not significant (figure 3b). At the metacommunity scale the NERR was completely flat (figure 3c).

Figure 3.

‘Raw’ NERRs at the (a) neighbourhood, (b) community and (c) metacommunity scales in winter annual plant communities growing in York gum – jam woodlands in Western Australia. Thick black lines are fitted fixed-effects relationships from linear models and light grey envelopes are 95% confidence intervals (CIs). Thin dark grey lines in (a) and (b) are within-community and among-community NERRs, respectively, and were estimated using mixed-effects models with varying intercepts and slopes.

(b). Question 2: how important are environmental conditions for driving native and exotic richness at each scale?

d-sep tests indicated that both final gmSEMs fitted the data adequately (p > 0.05). However, the ‘exotic-driven’ gmSEM (figure 4; electronic supplementary material, table S2; AICc = 138.05; Fischer's C = 166.53; d.f. = 156; p = 0.27) including direct paths from exotic richness to native richness was better supported by the data (greater than 2.5 AICc units lower) than the ‘native-driven’ gmSEM (electronic supplementary material figure S3 and table S3; AICc = 150.04; Fischer's C = 178.51; d.f. = 156; p = 0.10). Thus, the results presented herein refer to the exotic-driven model unless otherwise stated.

Figure 4.

The ‘exotic-driven’ gmSEM including directed paths from exotic to native richness. Black and red paths indicate positive and negative relationships, respectively. Numerical values adjacent to paths indicate the associated standardized regression coefficients. Italicized numerals indicate that the regression coefficient is in log(units). Alphanumeric values adjacent to selected paths correspond to the partial plots included in figure 6. Line thickness of each path is proportional to the respective regression coefficient. The explained variation (marginal) is included for each response variable (R²). T_min stands for minimum temperature and biocrust stands for biological soil crust. In total, there were 540 observations at the neighbourhood scale, 36 observations at the community scale and 12 observations at the metacommunity scale. The directed separation test indicated that the gmSEMs fit the data adequately (AICc = 138.05; Fisher's C = 166.53; d.f. = 156; p = 0.27).

(i). Metacommunity and community scales

Metacommunity-scale native richness was not significantly related to either climate variable (figure 5a,b), but the negative relationship with minimum growing season temperature explained a considerable amount of variance (R² = 0.34 for this pairwise relationship; electronic supplementary material, table S4). By contrast, both climate variables were strongly and positively related to metacommunity-scale exotic richness (R² = 0.67; figure 5c,d). In turn, metacommunity-scale native and exotic richness were positively related to community-scale native and exotic richness. Physical heterogeneity of the soil surface was positively related to native community richness, but not to exotic community richness (figure 4). Significant missing paths for mean environmental variables indicated a negative effect of average canopy cover on community exotic richness, and a positive effect of average soil hardness on metacommunity exotic richness (electronic supplementary material, table S5). The mean distance between communities was a poor predictor of metacommunity richness variables, and distance to agricultural fields was a poor predictor of exotic richness at the community scale.

Figure 5.

Partial plots of metacommunity-scale relationships from the exotic-driven gmSEM: (a) mean growing season T_min and (b) mean growing season moisture availability versus metacommunity native richness, and (c) mean growing season T_min and (d) mean growing season moisture availability versus metacommunity exotic richness. Solid lines are fitted relationships from the component models of the gmSEM and grey envelopes are the associated 95% confidence intervals (CIs). (Others) indicates that the plotted variable is the residuals of the original variable after being regressed on all other independent variables.

(ii). Neighbourhood scale

Neighbourhood plant density declined with increasing soil hardness and sclerophyll litter cover but increased with minimum growing season temperature (figure 4). However, a large amount of variation in plant density remained unexplained (R² = 0.17). In turn, neighbourhood plant density had a strong positive effect on both native and exotic richness. As expected, community native and exotic richness also had strong positive effects on neighbourhood native and exotic richness (figure 4). Plant-available N and the presence of biocrusts had positive and negative effects on exotic richness, respectively.

(iii). Comparison with the ‘native-driven’ model

All significant paths in the exotic-driven model were also significant in the native-driven model (electronic supplementary material figure S3 and table S2), except for the relationship between physical heterogeneity and community native richness which was marginally non-significant. Physical heterogeneity instead had a marginally significant positive relationship with community exotic richness. In addition, the presence of CWD had a significant positive effect on neighbourhood native richness. Missing paths from the native-driven model are presented in the electronic supplementary material, table S6.

(c). Question 3: after controlling for environmental drivers of native and exotic richness, are native–exotic richness relationships still evident?

Consistent with the ‘raw’ NERRs (figure 3), native and exotic richness were not significantly related to each other at either the metacommunity or community scales. Neighbourhood native and exotic richness remained positively related, but the partial slopes were weaker than the raw slopes from our original analysis (cf. figures 3a and 6a).

Figure 6.

Partial NERRs from the native-driven gmSEM at the (a) neighbourhood, (b) community and (c) metacommunity scales provided as a comparison with the ‘raw’ NERRs shown in figure 4. Solid lines are fitted relationships from the component models of the gmSEM, and grey envelopes are the associated 95% confidence intervals (CIs). (Others) indicates that the plotted variable is the residuals of the original variable after being regressed on all other independent variables.

4. Discussion

This study analysed NERRs within a hierarchical metacommunity framework to examine invasion in a natural annual plant system. Native and exotic richness were positively related at the neighbourhood scale, but not significantly associated at larger spatial scales, and these relationships remained after accounting for environmental drivers of native and exotic richness. We found support for both the environmental heterogeneity and environmental favourability hypotheses at the community scale, while climate appeared to have a stronger influence on metacommunity exotic richness than on native richness.

(a). Native–exotic richness relationships in interaction neighbourhoods

Neighbourhoods supporting more exotic species typically had greater native richness, consistent with local-scale biotic acceptance (figure 1a; [8]) and a growing body of literature suggesting that NERRs need not be negative at small spatial scales [6]. Positive associations between native and exotic diversity may arise for a number of reasons.

First, low-productivity soils may permit greater resource partitioning among native and exotic species owing to high within-neighbourhood niche dimensionality [13]. Consistent with this prediction, Davies et al. [36] found positive NERRs at small spatial scales in low-productivity sites in Californian serpentine plant communities, which have similar Mediterranean climates and low-productivity soils to our study system. Low productivity may also prevent neighbourhoods from being saturated with individuals of resident native species [36], which is supported by seed addition experiments in our study system (M.L. Raymundo, J.M. Dwyer and M.M. Mayfield 2016–2017, unpublished data). Second, positive relationships may reflect facilitation of native species by exotic species or vice versa [43]. For example, in experimental mixtures of our study species, the native forb Waitzia nitida experienced higher survival when planted with the exotic feather grass Pentameris airoides than when planted at the same density without the exotic, possibly owing to reduced evaporation at the soil surface [19]. Invasive species may also be facilitated in neighbourhoods with high native richness as they have a greater probability of encountering specific pollinators or mycorrhizal symbionts [44]. Finally, neighbourhood native diversity may positively covary with the supply rate of exotic species' propagules [45], but this is perhaps the least likely explanation in our system.

Even though we found positive neighbourhood-scale NERRs that are indicative of biotic acceptance (figure 1a), the unique responses of exotic richness to soil N and biocrusts suggest at least some segregation of spatial niches between exotic and native species. The positive effect of soil N on exotic richness is consistent with nutrient addition experiments demonstrating a competitive advantage for exotics under enrichment [46,47] and biocrusts are known to differentially inhibit germination of natives and exotics [48].

(b). Native–exotic richness relationships at larger scales

We found that NERRs at the metacommunity and community scales were both neutral (figure 3b,c), and these relationships persisted even after external drivers of diversity were controlled for statistically (figure 6b,c). In one of the few studies to report a similarly neutral NERR at broad scales, Symonds et al. [49] attributed the result to a focus on a single ecosystem type. Although our study was also conducted in a single ecosystem, sampling occurred across a large geographical region and captured substantial variation in environmental conditions. Instead, it is likely that neutral associations reflect semi-independent responses of native and exotic diversity to different abiotic drivers, most probably climate at the metacommunity scale (figure 1c; [50]).

Broad-scale relationships between climate variables and richness suggest strong but differential climatic filtering of native and exotic species. Although the negative relationship with growing season minimum temperature was not significant for native metacommunity richness, it seems that fewer native species are able to maintain viable populations in the warmer, northern parts of the study region. By contrast, greater exotic richness in warmer metacommunities may reflect apparency, whereby lower seed dormancy in warmer regions increases the likelihood of exotic species being detected [35,51]. The positive influence of metacommunity-scale diversity on community diversity suggests that communities in the York gum–jam woodlands are strongly limited by the diversity of available propagules [30]. This was particularly evident for exotic species, which implies that metacommunity-scale processes may be the predominant driver of exotic species distributions in this system, consistent with a global analysis of grassland invasions [50].

(c). Environmental heterogeneity and mean environmental conditions

In one of the major studies evaluating the role of environmental heterogeneity in driving NERRs at large scales, Davies et al. [7] found that increased variability in aspect significantly increased native and exotic richness in Californian grassland communities (defined as areas greater than 2156 m²). Despite the spatial scale of our communities being an order of magnitude smaller (225 m²), we found that variability in physical properties of the soil surface was positively related to community native richness and exotic richness. However, these effects were only marginally significant, and sensitive to the direction of causality in the model. It may also be possible that heterogeneity between our communities influenced native and exotic richness at the metacommunity scale, but we could not reliably test this with only three communities per metacommunity. Overall, our results are consistent with studies revealing a positive relationship between environmental heterogeneity and species diversity [52], potentially via the spatial storage effect [12], but this remains to be tested in our study system.

Mean environmental conditions also influence diversity to some extent. Significant missing paths in the gmSEMs indicate that average canopy cover was positively and negatively associated with native and exotic community richness, respectively. This suggests that native and exotic species are differentially sorting across communities, and may explain the slightly negative NERR at the community scale (figure 3b). At the metacommunity scale, exotic richness increased with average soil hardness. It is unlikely that harder soils support more exotic species, and instead this relationship may reflect a response to an underlying geological factor. Once again though, these relationships were weak and dependent on the causal direction assumed in the model.

5. Conclusion

Although there is emerging empirical support for positive NERRs at the neighbourhood scale, neutral NERRs at the community and metacommunity scales are at odds with the overwhelmingly positive NERRs reported in the literature [6]. It may be that our sampling scale was insufficient to capture the processes that drive positive relationships in other study systems. However, our larger study scales captured a large number of non-interacting individuals across considerable environmental variation, which is considered adequate to test the invasion paradox [6]. It is also important to consider that we limited our sampling to areas far from agricultural influences. Previous research in our system showed that P enrichment associated with agriculture favours a suite of exotic species with resource-acquisitive strategies [18], and that these exotics can competitively exclude native species in high-P neighbourhoods [14,19]. Thus, had we included P-enriched areas, we may have observed productivity-driven switches in neighbourhood-scale NERRs [36], and possibly more positive relationships at larger scales.

The role of environmental heterogeneity in regulating invasion into the York gum–jam woodlands deserves further enquiry [10]. Future research should experimentally examine the performance of native and exotic species along natural and anthropogenic environmental gradients. Understanding the mechanisms that regulate invasion into natural systems can improve our ability to predict future invasions and also reveal the processes by which species coexist, regardless of their origin.

Data accessibility

The datasets and R code supporting this article have been uploaded as part of the electronic supplementary material. The data and R code are also available on Dryad Digital Repository: http://dx.doi.org/10.5061/dryad.j7028s5 [53].

Authors' contributions

I.R.T. conducted the statistical analysis and drafted the manuscript. J.M.D. designed the study, collected the field data and drafted the manuscript. Both the authors gave their final approval for publication.

Competing interests

We have no competing interests.

Funding

Data collection were originally funded by an Australian Research Council grant (DP1094413) awarded to Margie Mayfield and Richard Hobbs.