Quantifying relationships between traits and ... - Wiley Online Library

Journal of Vegetation Science 21: 1014–1024, 2010 DOI: 10.1111/j.1654-1103.2010.01210.x & 2010 International Association for Vegetation Science

Quantifying relationships between traits and explicitly measured gradients of stress and disturbance in early successional plant communities Gre´gory Sonnier, Bill Shipley & Marie-Laure Navas Abstract Questions: How can one explicitly quantify, and separately measure, stress and disturbance gradients? How do these gradients affect functional composition in early successional plant communities and to what extent? Can we accurately predict trait composition from knowledge of these gradients? Location: Southern Quebec, Canada. Methods: Using eight environmental variables measured in 48 early successional plant communities, we estimated stress and disturbance gradients through structural equation modelling. We then measured 10 functional traits on the most abundant species of these 48 communities and calculated their community-level mean and variance weighted by the relative abundance of each species. Finally, we related these community-weighted means and variances to the estimated stress and disturbance gradients using general linear models or generalized additive models. Results: We obtained a well-fitting measurement model of the stress and disturbance gradients existing in our sites. Of the 10 studied traits, only average plant reproductive height was strongly correlated with the stress (r2 5 0.464) and disturbance (r2 5 0.543) gradients. Leaf traits were not significantly related to either the stress or disturbance gradients. Conclusions: The well-fitting measurement model of the stress and disturbance gradients, combined with the generally weak trait–environment linkages, suggests that community assembly in these early successional plant communities is driven primarily by stochastic processes linked to the history of arrival of propagules and not to trait-based environmental filtering. Sonnier, G. (corresponding author, Gregory.Sonnier@ USherbrooke.ca) & Shipley, B. (bill.shipley@USher brooke.ca): De´partement de Biologie, Universite´ de Sherbrooke, Sherbrooke, QC, Canada J1K2R1. Sonnier, G. & Navas, M.-L. ([email protected]): Montpellier SupAgro, Centre d’Ecologie Fonctionnelle et Evolutive (UMR 5175), 1919 Route de Mende, 34293 Montpellier Cedex 5, France.

Keywords: Community weighted mean of trait; Community weighted variance of trait; Environmental filtering; Structural equation model; Trait– environment linkages. Nomenclature: Marie-Victorin (1935–2002). Abbreviations: CWM 5 Community weighted mean; CWV 5 community weighted variance; SEM 5 structural equation modelling or model.

Introduction Stress and disturbance have long been recognized as two major environmental gradients affecting the fitness of plants and therefore the structure of plant communities (Grime 1979; Huston 1979; Tilman 1988). As defined by Grime (1979), ‘‘stress’’ represents the degree to which the abiotic environmental conditions of a site limit the production of plant biomass, while ‘‘disturbance’’ represents the degree to which the environmental conditions of a site remove or destroy pre-existing living plant biomass. According to these definitions, stress and disturbance at a site are modulated by many aspects of the environment. Although it is possible to measure these directly by experimentally preventing biomass removal, while measuring net primary productivity through repeated harvests, these methods are laborious and time-consuming. Because of the difficulty of such direct measurements in the field, most studies attempt indirect measurements of stress and disturbance. One common but indirect approach is first to measure multiple environmental variables related to stress (nutrients and water availability, soil pH) and to disturbance (types, intensity and periodicity), and then to reduce them to one or two major axes of variation using ordination methods (ter Braak 1987; McIntyre & Lavorel 1994; Cingolani et al. 2003; Rusch et al. 2009). One difficulty with this approach is that stress and disturbance are often related in the field through indirect relationships with common variables, and thus produce ordination axes that are complicated

Trait-environment linkages in early successional communities

combinations of both ‘‘stress’’ and ‘‘disturbance’’. When linking such ordination axes to plant functional traits it is therefore impossible to statistically separate inferences about the importance of each process in selecting for different plant traits. In this study, we use structural equation modelling (SEM, Shipley 2000; Grace 2006) to estimate these two primary underlying gradients, even though they are correlated in the field. This is done by representing stress and disturbance as latent (unobserved) variables that are linked to observed, but indirect, measures of each variable, within an explicit and falsifiable multivariate hypothesis concerning the causal links between the indirect measures and the theoretical process of biomass production and loss. In SEM, the numerical values of the latent variable (either ‘‘stress’’ or ‘‘disturbance’’) are predicted after controlling for the effects of all other variables, including the other latent variables. In other words, the estimates associated with the latent ‘‘stress’’ will be conditionally independent of those associated with the latent ‘‘disturbance’’ in a well-fitting SEM, even if the two processes are correlated in the field. A few studies have already used such a statistical framework to quantify disturbance and/or stress gradients and relate them to patterns of species richness in plant communities (Grace & Pugesek 1997; Weiher 2003; Weiher et al. 2004), but the goal of this paper is to quantitatively link our separate measures of stress and disturbance to functional traits measured at the community level. Stress and disturbance gradients are commonly assumed to affect trait composition in plant communities (McIntyre et al. 1995; Cingolani et al. 2007) through trait-based environmental filtering. That is to say, stress and disturbance differentially select species with particular combinations and values of traits, and therefore such traits determine species presence as well as relative abundance in different plant communities. For example, nutrient-poor sites often select for species that have, on average, a low specific leaf area (SLA) and leaf nitrogen content (LNC) but a high leaf dry matter content (LDMC) and long leaf lifespan, while species from nutrientrich sites show the opposite pattern in leaf traits and are generally taller (Grime 1979; Chapin et al. 1993; Poorter & de Jong 1999). Similarly, species experiencing high levels of disturbance are often reported to be shorter, invest more in reproduction, produce small but numerous seeds and have an annual habit (Diaz et al. 2007b). Most of the studies interested in trait variation across stress or disturbance gradients either describe only qualitative patterns or

1015

do not quantify the relationships between traits and these two gradients. Such qualitative statements, although useful, do not allow one to predict trait composition from the position of the community along continuous environmental gradients. In this study, we quantify relationships between the values of community weighted mean traits (CWM, Diaz et al. 2007a) calculated for 10 traits and stress/disturbance gradients measured in 48 early successional herbaceous plant communities. Previous studies suggested that stronger trait– environment linkages are achieved when using this community-weighted metric, which represents the expected dominant trait value in a community (Ackerly et al. 2002; Cingolani et al. 2005). Parallel to the well-documented effects of stress and disturbance on mean values of traits, stress and disturbance may also affect patterns of trait dispersion in plant communities. However, studies interested in trait dispersion patterns are scarce, and most of them do not relate the traits to explicitly measured environmental gradients (e.g. Stubbs & Wilson 2004; Schamp et al. 2008). Grime (2006) suggested that stress constrains species within a community to be similar in their strategies of resource acquisition and conservation, leading to a convergence in traits related to these functions. It follows that the degree of within-community dispersion of resourcerelated traits should decrease with increasing levels of stress. On the other hand, Grime (2006) suggested that disturbance should allow the coexistence of species with different regeneration strategies, but only when disturbance is intermediate and diverse; where, in highly disturbed sites, only specific regeneration strategies will be selected. It follows that the degree of within-community dispersion of regeneration traits should follow a quadratic curve in response to levels of disturbance. To explore this issue, we therefore explored trait dispersion patterns within the 48 herbaceous plant communities and along stress and disturbance gradients using a community weighted variance (CWV). In this paper, our objectives are therefore twofold. First, we develop a measurement model of stress and disturbance (using SEM) based on commonly studied environmental variables in early successional herbaceous vegetation. To do this, we extracted the scores of each stress and disturbance latent variable from the validated structural equation model and related them to the functional composition of each studied community. Second, once we obtain a well-fitting model quantifying stress and disturbance, we quantify relationships between the estimated stress/disturbance values and

1016

Sonnier, G. et al.

CWM/CWV of 10 functional traits, representing different functions of the plants, which have been consistently associated to stress and/or disturbance gradients in previous studies. We hypothesize that dispersion of traits associated to leaf structure (SLA, LDMC) and leaf composition (LNC, LCC) will decrease as stress increases, while within-community variability in traits related to regenerative strategies (seed mass, reproductive allocation and reproductive plant height) will first increase and decrease as disturbance increases.

Methods

Table 1. Studied traits and environmental variables. List of the studied traits and measured environmental variables with their corresponding abbreviations, and range values reported in this study. Discrete variables are indicated with an asterisk (). Traits

Abbreviations

Plant lifespan (annual, biannual, perennial) Specific Leaf Area (m2 kg 1) Leaf Dry Matter Content (mg g 1) Plant reproductive height (cm) Allocation to reproduction (%) Allocation to photosynthesis (%) Allocation to structure (%) Leaf Nitrogen Content (mg g 1) Leaf Carbon Content (mg g 1) Seed mass (mg)

LSPAN

1–3

SLA LDMC HRep RA PA SA LNC LCC Sm

5.49–52.87 100–900 10.8–181.1 0.5–48.2 6.5–82.1 12.0–86.1 10.6–50.7 353.0–492.9 0.01–15.3

Vegetation releve´s Environmental variables

The floristic composition of 48 sites located within a 50-km radius in southern Quebec, Canada (45126 0 5300 N, 71152 0 5500 W) was determined during the summer of 2005 and 2006. Climate of the region is characterized by cold winters (mean daily minimum in January is 181C) and relatively warm summers (mean daily maximum is 251C in July). Mean yearly precipitation is 1144 mm (874 mm rain and 249 cm snow). Sites consisted almost entirely of early successional herbaceous plant communities from roadsides (31 sites), abandoned quarries (seven sites), wetlands (six sites) and recently abandoned agricultural fields (four sites) chosen to cover a large range of above-ground biomass and soil texture. Vegetation of each site was harvested at the period corresponding to peak biomass (which was determined based on previous knowledge and field observations and was the same for all sites) using two randomly placed 0.5 m0.5 m quadrats. Living above-ground biomass was sorted to species and oven-dried to a constant mass at 601C for 72 h. More details on the sites are given in Appendix S1. Trait matrix At each site, we ranked the relative abundance of each species from most abundant to least abundant and selected those comprising the top 85% of the biomass. Every species that was within this top 85% in any one site was included in our list of species for which trait data were sought (Pakeman et al. 2008). This resulted in a total of 80 species on which we measured 10 traits (Table 1) following standardized protocols (Cornelissen et al. 2003). Using the observed biomass proportions (pik) of each species i of the pool (S 5 80 species) in a site k, we then calculated the community-weighted mean,

Species range

Abbreviations

Range

2

Living Above Ground Biomass (g m ) LAGB Litter Litter (g m 2) Bare soil (%) Bare soil Water Holding Capacity (cm 3 cm 3) WHC Mineralisable nitrogen (mg g 1 jour 1) Nitrogen Dist Human disturbance index Humidity (%) Humidity pH PH

30.8–975.2 0–529.3 0–95 4.6–24.6 0.1–2.1 1–3 2.1–34.8 4.8–8

CWM (Diaz et al. 2007a), and variance, CWV, for each trait j as follows: CWMjk ¼ tjk ¼

S X

ð1Þ

pik tij

i¼1

CWVjk ¼

S X

pik ðtij tjk Þ2

i¼1

¼

S X

ð2Þ 2

pik ðtij Þ ðCWMjk Þ

2

i¼1

Where tij is the value of trait j for species i, and tik is the community-weighted mean value of trait j in community k. CWM therefore quantifies the average trait value expressed by the vegetation, while CWV quantifies the variability of this trait value around the average value within the vegetation. Environmental variables To jointly assess stress and disturbance in all 48 sites of our study, we first determined, (i) the total above-ground biomass (g m 2), (ii) the litter mass (g m 2) and (iii) the percentage cover of bare soil (Table 1) observed in the two quadrats. We then took three 10-cm deep soil cores in each quadrat and pooled them per site. From these pooled soil


samples per site, we measured (i) mineralizable nitrogen availability (mg g 1 day 1) using a 12-day anaerobic incubation at 221C (Waring & Bremner 1964); (ii) soil pH; (iii) soil organic matter based on loss on ignition at 4001C during 16 h; and (iv) soil texture following the hydrometer method (Day 1965). To estimate water availability, we derived the water-holding capacity (WHC) of each sample from soil organic matter content and texture data using a modified version of the pedotransfer function of Saxton & Rawls (2006) implemented in SPAW software (version 6.02.74, USDA Agricultural Research Service). We also used a probe (ECH2O Check, Decagon Devices Inc., Pullman) to obtain a separate estimate of the field soil humidity; this was measured in all sites over a 2-day period and 3 to 4 days after a single rain event. Precise information on disturbance, such as time since the last disturbance and the frequency of disturbance was not available. We therefore determined a rough index of human disturbance based on different field observations of human activity, independent of vegetation physiognomy: presence (yes/no) of buildings, excavation, vehicular tracks or agriculture in the immediate area surrounding the community. We then attributed a value of 3 to highly disturbed sites (25 sites, all abandoned quarries were within this group), and the opposite value of 1 to relatively undisturbed sites (10 sites). Moderately disturbed sites were attributed a value of 2 (13 sites). We combined the entire set of environmental variables into a measurement model of stress and disturbance using structural equation modelling. The path diagram in Fig. 1 shows the hypothesized model between our indicator variables and the latent gradients of stress and disturbance. First, since water and nitrogen availability, soil pH and humidity are considered as major limiting factors impacting plant growth in most plant communities (Tilman 1984), we directly related these to our latent ‘‘stress’’ variable. Second, we estimated disturbance through our disturbance index. Third, the amount of living above-ground biomass should decrease with increas-

1017

ing levels of both latent ‘‘stress’’ and ‘‘disturbance’’. Forth, amount of bare soil should increase with both ‘‘stress’’ and ‘‘disturbance’’. Finally, the amount of litter should increase with increasing amount of above-ground biomass and decrease with levels of ‘‘disturbance’’. To fully identify the model, we fixed the variance of the two latent stress and disturbance variables to 1, thus fixing the scale of these variables as standard deviation units. Statistical procedure We tested the SEM model proposed in Fig. 1 using the EQS software (version 6.1 for Windows, Multivariate Software Inc.). Significance tests were based on the Satorra-Bentler robust correction to the maximum likelihood w2 to correct for nonnormality (Satorra & Bentler 1994). If the null probability is smaller than the significance level (0.05), then the tested model is rejected; in which case the model has to be adjusted through an exploratory process leading to a well-fitting but also ecologically meaningful model. We obtained the estimated values of ‘‘stress’’ and ‘‘disturbance’’ for each site using the generalized least squares procedure of Bentler & Yuan (1997) modified from Bartlett (1937) and implemented in EQS. It is important to remember that such scores are not the exact values of latent stress and disturbance (which are unknown since they are not measured directly) but their best unbiased predicted values when holding constant the other latent variables. Finally, we regressed each community-weighted trait mean (CWM) or variance (CWV) to the stress and disturbance scores estimated from our measurement model. A linear model was used to relate the trait values to the stress and disturbance scores, except when nonlinearity was obvious. In this latter case, we used generalized additive models (GAM) to relate traits and environments using the gam function of R in mgcv package (R Foundation for Statistical Computing, version 2.6 for Windows).

Results Environmental gradient and measurement model

Fig. 1. Hypothetical measurement model of stress and disturbance. Abbreviations follow Table 1. Symbols used are: an ellipse for latent variables and an arrow for causal relations, the variances of the two latent variables are fixed to 1.

The hypothetical measurement model proposed earlier in Fig. 1 was clearly rejected (w2 5 60.051, 17df, Po0.00001). We subsequently modified this original model in three ways. First, since the original model assumed independence between stress and disturbance, we added a free covariance (i.e. not

1018

Sonnier, G. et al.

fixed and undirected path) between the disturbance and stress latent variables in order to recognize that these two processes are correlated in our dataset. Indeed, the most disturbed sites (primarily abandoned quarries) were also the most unproductive ones due to historical land-use patterns rather than any direct causal links. This explains our choice of introducing a free covariance between stress and disturbance latent variables rather than any direct causal links. Second, we introduced a third latent variable (‘‘potential above-ground biomass’’) because the actual observed amount of biomass is the difference between production and loss. This latent ‘‘potential above-ground biomass’’ was assumed to decrease with both increasing stress and increasing disturbance. This new latent was estimated using two observed variables, the actual above-ground biomass and the amount of litter. The scale of this latent variable was determined by fixing the path coefficient of actual above-ground biomass to 1; i.e. a unit increase in potential biomass production produces a unit increase in actual biomass when all other variables are held constant. Third, we removed pH from the model since it was only marginally related to the latent stress (r2 5 0.09). In this new model, covariance between stress and disturbance was relatively high (r2 5 0.45) and positive (Table 2), but it should be remembered that construction scores of the latent variable are predicted from a given set of indicator variables after keeping constant all other variables (including other latent variables). We observed that the estimated value of latent ‘‘stress’’ increases as nitrogen availability and water availability decrease, while the percentage of bare soil increases with the latent disturbance. Both latent disturbance and latent stress cause a decrease in the amount of observed litter and biomass. The result was a well-fitting model (Fig. 2, w2 5 5.457, df 5 11, P 5 0.907). All path coefficients Table 2. Standardised structural equations along with r2 values corresponding to the measurement model presented in Fig. 2. Abbreviations as given in Table 1. Standardized solution

r2

LAGB 5 0.91 Potential Biomass10.421.30 Litter 5 0.95 Potential Biomass10.320.20 Bare soil 5 0.78 Disturbance10.63224.91 Dist 5 0.89 Disturbance10.450.13 Nitrogen 5 0.97 Stress10.230.02 Humidity 5 0.80 WHC10.6027.33 WHC 5 0.78 Stress10.628.81 Relation between latent variables Potential Biomass 5 0.43 Disturbance 0.56 Stress10.390.92 Stress 5 0.61 Disturbance

0.82 0.90 0.61 0.79 0.95 0.64 0.61

were significant, and the explained variance (Table 2) of the observed variables ranged from 0.61 (bare soil) to 0.95 (soil nitrogen). Relationships with community-weighted parameters Given the well-fitting measurement model for stress and disturbance, we obtained the scores of each latent variable from the final SEM presented in Fig. 2, and related these scores to the CWM and CWV values of each trait and the 48 sites of our study. We excluded plant lifespan since most of our sites were exclusively composed of perennial species, except in the most disturbed sites. Of the nine remaining traits, the relation between CWM and either stress or disturbance was significant for five values: LDMC, LCC, HRep, RA, Sm. This was reduced to three when considering the CWV of traits LCC, HRep and Sm. However the strength of these relationships was generally low, except for the CWM associated with reproductive plant height (HRep). The CWM and CWV of plant reproductive height were both significantly lower in highly stressed (Fig. 3) or disturbed sites (Fig. 4). Average reproductive allocation (RA) and average leaf dry matter content (LDMC) were also significantly related to each gradient. Mean LDMC decreased while mean RA increased when disturbance or stress increased (Figs 3 and 4). However CWV corresponding to these two traits did not vary along the two gradients. On the other hand, CWM of seed mass (Sm) and leaf carbon content (LCC) were only related to disturbance, both decreasing when disturbance increased (Fig. 4). Variance in seed mass decreased with increasing disturbance. In other words, not only did average seed size decrease in the most disturbed sites, but there was also less variation of seed mass between species in the most disturbed sites. The opposite pattern was found for LCC, since we observed a tendency for greater variance in disturbed sites. Interestingly, neither leaf nitrogen content (LNC) nor specific leaf area (SLA) were related to either of the two gradients in terms of CWM. This result was also seen for traits in relation to allocation of biomass to (i) stems and petioles (SA) and to (ii) photosynthetic tissues (PA).

Discussion The measurement model

0.85 0.45

Combining confirmatory and exploratory approaches of SEM, as already implemented in


1019

Fig. 2. Final and fully parameterized measurement model of stress and disturbance and corresponding standardized solution. Abbreviations follow Table 1. Symbols used are: an ellipse for latent variables and an arrow for causal relations. Measurement errors (unexplained variance/residual variance) are given in italic, while path coefficients relating (i) latent to observed, (ii) each latent, and (iii) each observed variable are given in bold.

community ecology (Grace & Pugesek 1997; Weiher 2003; Weiher et al. 2004), we obtained a simple measurement model of two latent environmental filters (stress and disturbance) that both successfully captured the observed patterns of covariance between the measured environmental variables and that agreed with the original definitions of Grime (1979). As hypothesized, stress increased when both soil nitrogen and water availability decreased, disturbance was positively related to amount of bare ground, and increases in both stress and disturbance caused potential above-ground biomass (reflecting both measured above-ground biomass and litter) to decrease. Certainly other resource and non-resource factors, not measured in our study, can limit plant growth in particular systems (Fonseca et al. 2000; Baer et al. 2003). Furthermore, the latent disturbance variable used here lacks more precise information on the frequency and intensity of disturbance. Such limitations of environmental variables are common in community ecology because of the difficulty and cost of obtaining more precise information on dynamic processes. However, when such additional information is available, it is easy to include in a structural equations model without changing the underlying hypothetical causal structure of the system (Grace & Bollen 2008). This is an important advantage over ordination approaches because SEM still allows for a clear biological interpretation of the underlying latent variables while also allowing for modifications as new indicator variables become available. The stress and disturbance gradients are not independent in our data set, formed of communities

established soon after disturbance, as recorded in other studies on communities that are regularly disturbed. For example, Weiher (2003) identified with SEM a latent ‘‘stress’’ variable associated with soil characteristics that was correlated to a latent ‘‘disturbance’’ variable associated with fire frequency and time since last fire in oak savannas. In our study, the most disturbed sites were quarries established on sandy soils that were characterized by the lowest nitrogen and water availability found in our data set. Despite the correlation between stress and disturbance, SEM allows one to estimate each variable separately because these scores extracted from the model are functions of the path coefficients linking each latent to their indicator variables while holding constant the other parameters in the model. This is different from taking the scores of samples along ordination axes. In fact, a PCA ordination of our data extracted a first axis that was a combination of both stress- and disturbance-related variables, thus preventing one from obtaining measurements of either stress or disturbance independently of the other. CWM–environment linkages After quantifying gradients of stress and disturbance, we then searched for systematic relationships between these underlying gradients and selected plant traits that are expected to co-vary with them. It is important to note that we focused on early successional plant communities and so even the less disturbed sites of our study are comparatively more disturbed than most late successional

Sonnier, G. et al.

800

r2 = 0.13

LDMC (mgg−1)

30 25 20 15

600 400

−2

−1

0

1

−2

−1

HRep (cm)

440 420 r2 = 0.1

0

25 20

100

−2

−1

0

1

−2

−1

0

1

−2

−1

0

1

60 50

50

400

30

1

r2 = 0.46 r2 = 0.12

150

460

35

15

200

10

LCC (mgg−1)

40

r2 = 0.16

PA (%)

SLA (m2 kg−1)

35

LNC (mgg−1)

1020

40 30 20 10

−2

−1

0

1

−2

−1

0

1

80 r2 = 0.13

3.0

70

15 10

60

Sm (mg)

20

SA (%)

RA (%)

25

50 40

2.0 1.0

30

5

20 −2

−1

less stressed

0

1

more stressed

0.0 −2

−1

0

1

less stressed more stressed

less stressed

more stressed

Scores of stress

Fig. 3. Trait patterns along stress gradient. Trends of community-weighted mean (CWM) traits along with communityweighted variance (CWV) with stress gradient. The scores are the estimated values of stress in each site obtained from the final measurement model. Sites with less nitrogen and water are located on the right of the gradient, with nutrient-rich sites on the left of the gradient. Dark lines represent the predicted CWM, while grey lines depict patterns of trait dispersion (here standard deviations) around this CWM. r2 of the relationship between CWM and stress (black) or between CWV and stress (grey) are reported when significant.

plant communities. This could have particular consequences when looking at trait composition of the community. In agreement with other studies, we observed a decrease in average plant reproductive height with both stress and disturbance (Cingolani et al. 2007; Diaz et al. 2007b; Gaucherand & Lavorel 2007). This pattern, as well as the increase in reproductive allocation (RA) and the decrease in seed mass observed along the disturbance gradient, reflects the fundamental competition/colonization trade-off (Tilman 1994). Plants in highly disturbed sites invest more in their reproduction, thus producing smaller but more numerous seeds that allow them to rapidly reach open space created by disturbance. At the opposite extreme, in fertile (low stress) and undisturbed sites, competition for light and other nutrients

increase and species that are able to out-compete others are generally taller (Gaudet & Keddy 1988). By contrast, the patterns of variation in leaf traits reported in our study are surprising. Indeed, average SLA and LNC did not vary significantly along our stress gradient, while average LDMC and LCC decreased very slightly with stress. These results contradict previous studies showing that community-level values of SLA and LNC increase with fertility (Garnier et al. 2007; Gaucherand & Lavorel 2007) and with water availability (Fonseca et al. 2000; Cingolani et al. 2007), while LDMC correspondingly decreases. In a recent study, Rusch et al. (2009) found that SLA was not related to a combined fertility– stress gradient and explained it by the dependency of the SLA index to other morphological attributes, such


25 20

LNC (mgg−1)

LDMC (mgg−1)

SLA (m2 kg−1)

30

40

r2 = 0.17

800

35

600 400

15

35 30 25 20 15

200

10 −1.0

0.0

1.0

2.0

−1.0

460

0.0

420 r2 = 0.26 r2 = 0.11

400

2.0

−1.0

100

0.0

1.0

2.0

−1.0 0.0

1.0

2.0

60 50 PA (%)

440

1.0

r2 = 0.54 r2 = 0.22

150 HRep (cm)

LCC (mgg−1)

1021

50

40 30 20 10

−1.0

0.0

1.0

2.0

−1.0

0.0

1.0

2.0

80 r2 = 0.16 SA (%)

20 RA (%)

r2 = 0.11

3.0

70

15 10

r2 = 0.28

60

Sm (mg)

25

50 40

2.0

1.0

30

5

20 −1.0

0.0

less disturbed

1.0

2.0

more disturbed

0.0 −1.0

0.0

less disturbed

1.0

2.0

more disturbed

−1.0

0.0

less disturbed

1.0

2.0

more disturbed

Scores of disturbance

Fig. 4. Trait patterns along the disturbance gradient. Variation of community-weighted mean (CWM) traits along with community-weighted variance (CWV) with disturbance gradient. The scores are the estimated values of disturbance in each site obtained from the final measurement model. Highly disturbed sites are located on the right of the gradient, with undisturbed sites on the left of the gradient. Dark lines represent the predicted CWM while grey lines depict patterns of trait dispersion (here standard deviation) around this CWM. r2 of the relationship between CWM and disturbance (black) or between CWV and disturbance (grey) are reported when significant.

leaf thickness and LDMC. Considering the lack of response in all leaf traits, we rather suggest that this lack of response in our study arises because, in early successional plant communities, stochastic processes relating to dispersal may be more important than habitat filtering. Trait dispersion along studied gradients Of the nine traits whose dispersion was investigated (using CWV) in our study, only SLA, LCC, HRep and Sm were significantly affected by either stress or disturbance, but the strength of these relationships were all very low. We did not observe any relationships between traits related to acquisi-

tion/conservation of resources and stress, except for SLA. As in our study, Cornwell & Ackerly (2009) did not find any consistent patterns in the distributions of SLA and LNC (per unit mass) in relation to a soil moisture gradient. Our results therefore do not support the hypothesis that stress filters results in the convergence of traits related to acquisition/ conservation of resources traits, such as SLA, LDMC, LNC and LCC, within a community. We did, however, observe a decrease for reproductive plant height with increasing stress, which was also reported by Cornwell & Ackerly (2009), with dispersion in maximum plant height increasing with soil moisture. In our study, the variance in seed mass values decreased with disturbance, suggesting that

1022

Sonnier, G. et al.

there is more convergence of this trait in disturbed sites. This is again the opposite to our initial hypothesis, namely that disturbance will result in a bell-shaped curve in the dispersion of regenerative traits. However, since even the less disturbed sites of our study are comparatively more disturbed than most late successional plant communities, we are presumably only observing the decrease phase of the curve. This could explain why we did not observe a quadratic curve and emphasizes the importance of developing quantitative scales for stress and disturbance gradients.

Why so many weak relationships? Globally, the strength of the relations between community-level properties of traits (either CWM or CWV values) and environmental gradients were weak. Possible explanations are that we missed important aspects of the underlying environmental gradients, or that filtering was based on a large number of individually weak traits rather than being dominated by a few key traits. We have no way of objectively evaluating the first of these possibilities, but we acknowledge that the latent disturbance gradient is missing some important aspects. A more compelling explanation is that dispersal-related processes, rather than habitat filtering, were more important in the assembly of our communities. Researchers have already emphasized the importance of dispersal limitations (Ozinga et al. 2005) and specieslevel priority effects (Fukami et al. 2005) during community assembly. It is likely that these historical contingent factors were important in our study, since the sites were all in the first stages of secondary succession. If so, then community assembly in our sites is only weakly determined by species traits, which could also explain the lack of response of leaf traits to stress gradients. This is in contrast to the successional sites following agricultural abandonment that were studied in Vile et al. (2006) and Shipley et al. (2006), in which the community-weighted means of eight traits were strongly predicted by successional age. The lack of response in the community-weighted variances may also reflect the effect of competition filters that are simultaneously acting during community assembly and whose effects on trait dispersion patterns are difficult to separate from stress and disturbance filters. Indeed, the CWM and CWV observed in this study result from both the convergence of traits due to environment and divergence of traits due to competition between species processes. The recent development of statistical methods for discriminating trait con-

vergence versus trait divergence patterns (Pillar et al. 2009) will allow further insights into the assembly of communities.

Conclusions The results presented in this article show that even though particular attributes of plants are selected by stress and disturbance filters, the strength of the trait–environment linkages was low for any single trait, and a large part of the variation in communityweighted mean and variance of trait values was unexplained in our data set formed of early successional communities. This prevented us from predicting accurately the trait composition of a community from independent knowledge of the environment, as recently proposed by Shipley et al. (2006). Considering this weak trait–environment linkage, we suggest that dispersal limitations and species-level priority effects are more important than trait-based assembly rules in our systems. We, however, acknowledge that this could also arise from focusing only on early successional plant communities, where even the less disturbed sites are still more disturbed than an oldfield site. Future research should therefore apply the SEM approach to further quantify the strength of habitat filtering in many different environments and many different plant communities.

Acknowledgements. This research was financed by the Natural Sciences and Engineering Research Council of Canada. We thank Valerio Pillar, Leandro Duarte and another anonymous referee for their comments on an early version of this manuscript.

References Ackerly, D.D., Knight, C.A., Weiss, S.B., Barton, K. & Starmer, K.P. 2002. Leaf size, specific leaf area and microhabitat distribution of chaparral woody plants: contrasting patterns in species level and community level analyses. Oecologia 130: 449–457. Baer, S.G., Blair, J.M., Collins, S.L. & Knapp, A.K. 2003. Soil resources regulate productivity and diversity in newly established tallgrass prairie. Ecology 84: 724–735. Bartlett, M.S. 1937. The statistical conception of mental factors. British Journal of Psychology 28: 97–104. Bentler, P.M. & Yuan, K.H. 1997. Optimal conditionally unbiased equivariant factor score estimators. In: Berkane, M. (ed.) Latent variable modeling with applications to causality. pp. 259–281. Springer-Verlag, New York, NY, US.


Chapin, F.S., Autumn, K. & Pugnaire, F. 1993. Evolution of suites of traits in response to environmental stress. American Naturalist 142: S78–S92. Cingolani, A.M., Cabido, M.R., Renison, D. & Solis, V.N. 2003. Combined effects of environment and grazing on vegetation structure in Argentine granite grasslands. Journal of Vegetation Science 14: 223–232. Cingolani, A.M., Posse, G. & Collantes, M.B. 2005. Plant functional traits, herbivore selectivity and response to sheep grazing in Patagonian steppe grasslands. Journal of Applied Ecology 42: 50–59. Cingolani, A.M., Cabido, M., Gurvich, D.E., Renison, D. & Diaz, S. 2007. Filtering processes in the assembly of plant communities: are species presence and abundance driven by the same traits? Journal of Vegetation Science 18: 911–920. Cornelissen, J.H.C., Lavorel, S., Garnier, E., Diaz, S., Buchmann, N., Gurvich, D.E., Reich, P.B., ter Steege, H., Morgan, H.D., van der Heijden, M.G.A., Pausas, J.G. & Poorter, H. 2003. A handbook of protocols for standardised and easy measurement of plant functional traits worldwide. Australian Journal of Botany 51: 335–380. Cornwell, W.K. & Ackerly, D.D. 2009. Community assembly and shifts in plant trait distributions across an environmental gradient in coastal California. Ecological Monographs 79: 109–126. Day, P.R. 1965. Particle fractionation and particle size analysis. In: Black, C.A. (ed.) Methods of soil analysis. pp. 545–567. American Society of Agronomy, Madison, WI, US. Diaz, S., Lavorel, S., de Bello, F., Quetier, F., Grigulis, K. & Robson, M. 2007a. Incorporating plant functional diversity effects in ecosystem service assessments. Proceedings of the National Academy of Sciences USA 104: 20684–20689. Diaz, S., Lavorel, S., McIntyre, S., Falczuk, V., Casanoves, F., Milchunas, D.G., Skarpe, C., Rusch, G., Sternberg, M., Noy-Meir, I., Landsberg, J., Zhang, W., Clark, H. & Campbell, B.D. 2007b. Plant trait responses to grazing – a global synthesis. Global Change Biology 13: 313–341. Fonseca, C.R., Overton, J.M., Collins, B. & Westoby, M. 2000. Shifts in trait combinations along rainfall and phosphorus gradients. Journal of Ecology 88: 964–977. Fukami, T., Bezemer, T.M., Mortimer, S.R. & van der Putten, W.H. 2005. Species divergence and trait convergence in experimental plant community assembly. Ecology Letters 8: 1283–1290. Garnier, E., Lavorel, S., Ansquer, P., Castro, H., Cruz, P., Dolezal, J., Eriksson, O., Fortunel, C., Freitas, H., Golodets, C., Grigulis, K., Jouany, C., Kazakou, E., Kigel, J., Kleyer, M., Lehsten, V., Leps, J., Meier, T., Pakeman, R., Papadimitriou, M., Papanastasis, V.P., Quested, H., Quetier, F., Robson, M., Roumet, C., Rusch, G., Skarpe, C., Sternberg, M., Theau, J.P., Thebault, A., Vile, D. & Zarovali, M.P. 2007. Assessing the effects of land-use change on plant traits, communities

1023

and ecosystem functioning in grasslands: a standardized methodology and lessons from an application to 11 European sites. Annals of Botany 99: 967–985. Gaucherand, S. & Lavorel, S. 2007. New method for rapid assessment of the functional composition of herbaceous plant communities. Austral Ecology 32: 927– 936. Gaudet, C.L. & Keddy, P.A. 1988. A comparative approach to predicting competitive ability from plant traits. Nature 334: 242–243. Grace, J.B. 2006. Structural equation modelling and natural systems. 1st edition. Cambridge University Press, Cambridge, UK. Grace, J.B. & Bollen, K.A. 2008. Representing general theoretical concepts in structural equation models: the role of composite variables. Environmental and Ecological Statistics 15: 191–213. Grace, J.B. & Pugesek, B.H. 1997. A structural equation model of plant species richness and its application to a coastal wetland. American Naturalist 149: 436–460. Grime, J.P. 1979. Plant strategies and vegetation processes. 1st ed. John Wiley and Sons, Chichester, UK. Grime, J.P. 2006. Trait convergence and trait divergence in herbaceous plant communities: mechanisms and consequences. Journal of Vegetation Science 17: 255–260. Huston, M.A. 1979. A general hypothesis of species diversity. American Naturalist 113: 81–101. Marie-Victorin, Rouleau, E. & Brouillet, L. 2002. Flore laurentienne. 3rd ed. G. Morin, Boucherville, QC, CN. McIntyre, S. & Lavorel, S. 1994. How environmental and disturbance factors influence species composition in temperate Australian grasslands. Journal of Vegetation Science 5: 373–384. McIntyre, S., Lavorel, S. & Tremont, R.M. 1995. Plant life-history attributes: their relationship to disturbance response in herbaceous vegetation. Journal of Ecology 83: 31–44. Ozinga, W.A., Schaminee, J.H.J., Bekker, R.M., Bonn, S., Poschlod, P., Tackenberg, O., Bakker, J. & van Groenendael, J.M. 2005. Predictability of plant species composition from environmental conditions is constrained by dispersal limitation. Oikos 108: 555–561. Pakeman, R.J., Garnier, E., Lavorel, S., Ansquer, P., Castro, H., Cruz, P., Dolezal, J., Eriksson, O., Freitas, H., Golodets, C., Kigel, J., Kleyer, M., Leps, J., Meier, T., Papadimitriou, M., Papanastasis, V.P., Quested, H., Quetier, F., Rusch, G., Sternberg, M., Theau, J.P., Thebault, A. & Vile, D. 2008. Impact of abundance weighting on the response of seed traits to climate and land use. Journal of Ecology 96: 355–366. Pillar, V.D., Duarte, L., Sosinski, E.E. & Joner, F. 2009. Discriminating trait-convergence and trait-divergence assembly patterns in ecological community gradients. Journal of Vegetation Science 20: 334–348. Poorter, H. & de Jong, R. 1999. A comparison of specific leaf area, chemical composition and leaf construction costs of field plants from 15 habitats differing in productivity. New Phytologist 143: 163–176.

1024

Sonnier, G. et al.

Rusch, G.M., Skarpe, C. & Halley, D.J. 2009. Plant traits link hypothesis about resource-use and response to herbivory. Basic and Applied Ecology 10: 466–474. Satorra, A. & Bentler, P.M. 1994. Corrections to test statistics and standard errors in covariance structure analysis. In: von Eye, A. & Clogg, C.C. (eds.) Latent variables analysis: applications for developmental research. pp. 399–419. Sage, Thousand Oaks, CA, US. Saxton, K.E. & Rawls, W.J. 2006. Soil water characteristic estimates by texture and organic matter for hydrologic solutions. Soil Science Society of America Journal 70: 1569–1578. Schamp, B.S., Chau, J. & Aarssen, L.W. 2008. Dispersion of traits related to competitive ability in an old-field plant community. Journal of Ecology 96: 204–212. Shipley, B. 2000. Cause and correlation in biology: a user’s guide to path analysis, structural equations and causal inference. Cambridge University Press, Cambridge, UK. Shipley, B., Vile, D. & Garnier, E. 2006. From plant traits to plant communities: a statistical mechanistic approach to biodiversity. Science 314: 812–814. Stubbs, W.J. & Wilson, J.B. 2004. Evidence for limiting similarity in a sand dune community. Journal of Ecology 92: 557–567. ter Braak, C.J. 1987. The analysis of vegetation–environment relationships by canonical correspondence analysis. Vegetatio 69: 69–77. Tilman, D. 1984. Plant dominance along an experimental nutrient gradient. Ecology 65: 1445–1453. Tilman, D. 1988. Plant strategies and the dynamics and structure of plant communities. Princeton University Press, NJ, US. Tilman, D. 1994. Competition and biodiversity in spatially structured habitats. Ecology 75: 2–16. Vile, D., Shipley, B. & Garnier, E. 2006. A structural equation model to integrate changes in functional strategies during old-field succession. Ecology 87: 504–517. Waring, S.A. & Bremner, J.M. 1964. Ammonium production in soil under waterlogged conditions as an index of nitrogen availability. Nature 201: 951–952. Weiher, E. 2003. Species richness along multiple gradients: testing a general multivariate model in oak savannas. Oikos 101: 311–316. Weiher, E., Forbes, S., Schauwecker, T. & Grace, J.B. 2004. Multivariate control of plant species richness and community biomass in blackland prairie. Oikos 106: 151–157.

Supporting Information Additional supporting information may be found in the online version of this article: Appendix S1. Additional information on the 48 studied sites giving, for each site, species richness, three most abundant species, living above-ground biomass (LAGB), amount of litter, disturbance index (1 to 3) and the corresponding types of habitat. Appendix S2. Results of linear or generalized additive models between traits and stress and disturbance. Results of linear or generalized additive models between (a) community-weighted mean (CWM) and (b) between community-weighted variance (CWV) of nine traits and scores of each stress or disturbance latent variable obtained from the measurement model for 48 sites. Levels of significance are Po0.001; Po0.01; Po0.05; ns non-significant. Abbreviations of traits follows Table 1 in the main text. Appendix S3. Predicting trait composition from environment. Distribution of r2 and Fj between predicted and observed community-weighted mean traits obtained by cross-validation in 1000 repetitions. This was done in response to disturbance for (A) plant maximum height and (B) LCC, and for (C) LDMC. Fj was obtained from comparison between the RMSE of prediction calculated in the original data sets (using cross-validation) and the median RMSE of prediction determined after 1000 random permutations of stress or disturbance values. A value greater than 0 means that the predictive model under consideration performs better than a null model. Please note: Wiley-Blackwell is not responsible for the content or functionality of any supporting materials supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author of the article. Received 14 May 2009; Accepted 10 June 2010. Co-ordinating Editor: Dr. Valerio Pillar.