Functional diversity indices reveal the impacts of land use intensification on plant community assembly


Correspondence author. E-mail:


1. Functional diversity has been suggested as an important descriptor of an assemblage and as an indicator of ecosystem function. However, there has been little testing of recent functional diversity measures on field data and across environmental gradients of disturbance and productivity. This study tested how three recently developed measures of functional diversity behave in practice and what could be deduced about assembly rules operating in the systems studied.

2. Data from 30 plant communities in a small area, comprising arable fields, mown and grazed grasslands, moorland and woodland, were analysed in terms of these diversity indices. Functional Divergence, Evenness and Richness were independent of each other, but Richness correlated to species diversity through the sampling effect.

3. As evidenced by reduced Functional Richness, habitat filtering operated at a significant proportion of the sites sampled. Functional Richness was also reduced against expectations at high levels of both productivity and disturbance. Functional Divergence showed no clear patterns, but Functional Evenness showed a clear promotion by disturbance and a reduction at high standing biomass.

4.Synthesis. Knowledge of how functional diversity is related to the environment provides indications of the processes governing community assembly, and the possibility of developing them as measures of ecosystem function.


Functional diversity has become an important descriptor of species assemblages in two research areas. Firstly, it has been seen as an indicator of processes governing community assembly (Cornwell, Schwilk & Ackerly 2006), and one that can show the impact of perturbations (Villéger, Mason & Mouillot 2008) or identify ecological gradients (Mouillot, Dumay & Tomasini 2007). It has been predicted that increased productivity will lead to convergence in traits, whilst disturbance may increase functional diversity (Grime 2006). There is evidence for convergence of some traits, especially at high productivities (Cornwell & Ackerly 2009; Pakeman, Lennon & Brooker in press), but there is also limited evidence for divergence in others (Cornwell & Ackerly 2009) and no evidence of either convergence or divergence in other studies (Schamp, Chau & Aarssen 2008).

Secondly, functional diversity has also been suggested as an indicator of ecosystem function (Díaz & Cabido 2001; Petchey, Hector & Gaston 2004). For instance, it has been shown to be correlated to productivity, though it was outperformed by other parameters (Cadotte et al. 2009). Also, in a study of the relationship between ecosystem processes and functional diversity, community-weighted mean traits and abiotic factors, functional diversity was often well correlated to the processes examined (Díaz et al. 2007). However, in no case was functional diversity kept in the best predictive model, indicating that variation in this metric often covaries with variation in other metrics or the environment.

However, there have been many different metrics of functional diversity that have been developed. The earliest were simple measures of functional group number, but these have progressively been replaced by the ones that take into account functional traits and species abundance together in a robust theoretical framework (e.g. Mason et al. 2005; Cornwell, Schwilk & Ackerly 2006; Villéger, Mason & Mouillot 2008). This is because ideal measures of functional diversity take into account abundance and use multiple traits to acknowledge linkages between them. They are also not trivially related to species richness and rely on original data rather than data transformed through classification or ordination processes (Villéger, Mason & Mouillot 2008). Functional diversity has now been decomposed into three parts, Functional Richness (FRic), Functional Evenness (FEve) and Functional Divergence (FDiv), that each measure different aspects of the diversity of functional traits within a community (Mason et al. 2005; Villéger, Mason & Mouillot 2008). The utility of these measures has been assessed by Mouchet et al. (2010) against other candidate diversity indices and each performed best in measuring their individual aspects of functional diversity: richness, evenness and divergence.

Despite the theoretical basis behind these measures of functional diversity being sound, there has been little testing of them with field data and so there is little knowledge of how they may respond to different environmental gradients. This knowledge of how they relate to the environment is necessary to assess their utility as measures of both community assembly and ecosystem function. Theoretical expectations include negative departures from expectation revealing habitat filtering whilst positive ones reveal neutral assembly rules (Mouchet et al. 2010). Also, the dynamics of functional diversity against species diversity have been used to set up a framework to assess how land use change is impacting on plant communities (Mayfield et al. 2010). Field assessments of functional diversity reveal a number of patterns. Plots of tropical trees appear to be more functionally diverse (larger range in traits) than expected (Kraft, Valencia & Ackerly 2008), although measures of diversity were not tested against environmental gradients. Expectations covering FRic are more common than for the other measures; early work suggested it should increase linearly with species richness (Díaz & Cabido 2001), but other predictions include a monotonic increase with species richness (Mayfield et al. 2005). In a study of lake fish assemblages (Mason et al. 2008) there was, however, a ‘hump-backed’ relationship between both FRic and FDiv versus species richness, and a power relationship between FEve and species richness. A similar relationship was demonstrated for estuarine fishes (Villéger et al. 2010) – it appears that if conditions are conducive for high species richness then assemblages are characterised by many redundant species. Similarly, a study aimed specifically at analysing the response of the functional diversity of plants to disturbance showed peaks in richness at intermediate levels of disturbance (Biswas & Mallik 2010). The Mason et al. (2008) study of lake fish assemblages also showed hump-backed relationships between both FRic and FDiv with temperature, and a linear one between FEve and temperature. Their interpretation was that increased temperature may have permitted increased species richness by allowing increased niche specialization.

To further develop the utility of these metrics this paper tests the following questions for a set of plant assemblages monitored in the field: (i) Are the chosen measures of functional diversity independent of one another, as indicated by tests with artificial data, and of species richness? (ii) Can the functional diversity indices identify the operation of ecological processes, such as the operation of assembly rules, and whether these processes are modulated by environmental disturbance and productivity?

Materials and methods

Data collection

Thirty sites were chosen within an area on the west coast of Scotland. The area was selected to have a large range of land uses and intensities within a small area (5 × 7 km, centred on latitude 57°18′36″ N, longitude 5°39′35″ W) sharing the same geology and climate. A range of management regimes including arable, fallow areas cropped for hay, winter-grazed rough grassland and tall herb communities, unimproved silage fields, unimproved pasture, moorland and woodland were in place on the samples sites (Table S1 in Supporting Information).

Vegetation surveys, carried out in July 2007, recorded relative abundance of all higher plant species, bryophytes and litter estimated in four randomly placed 1 × 1 m quadrats per site. A total of 148 species were recorded across all sites. Annual net primary productivity (ANPP) was estimated as the difference between live biomass present in four 50 × 50 cm quadrats at the start of the growing season and four different 50 × 50 cm quadrats areas at the time of peak biomass (Garnier et al. 2007). On sites with grazing animals, growth was protected by 80 × 80 × 60 cm high wire cages. Summer samples used to estimate ANPP were analysed for carbon and nitrogen (Flash EA112 Elemental Analyser; ThermoFinnegan, Milan, Italy). Soil nitrogen release was quantified using resin bags (Hobbie 2005) buried on 23 April 2007 and retrieved on 9 July 2007. Eight 8 × 10 cm nylon bags containing 10 g of Dowex Marathon MR-3 mixed ion exchange resin (Supelco, Bellefonte, PA, USA) were buried at 10 cm depth at locations spread across each site. Ammonium and nitrate ions were extracted from the retrieved bags by shaking in 400 mL of 2 M KCl for 1 h. The resulting solutions were then analysed for nitrate by ion chromatography (Dionex, Sunnyvale, CA, USA) and for ammonium colourimetrically using a Konelab Discrete Analyser (Konelab Corporation, Espoo, Finland). Disturbance was quantified using the metrics developed from those of Kühner & Kleyer (2008) for frequency of disturbance, biomass loss, below-ground disturbance and start of disturbance in weeks (Table 1).

Table 1.   Disturbance parameters and categories. Derived from Kühner & Kleyer (2008)
Disturbance measureCategorisationDetails
1Extensive grazing
2Moderate winter grazing
3Mowing + moderate winter grazing
4Year round intensive grazing
Biomass loss0–100Percentage of biomass removed each year. 100%
Below-ground disturbance0No below-ground disturbance in previous 5 years
0.5Ploughing 1–5 years previously
1Ploughing in year of sampling
Start of disturbance1–53First disturbance event of year measured from start of growing season. Week 53 equates to undisturbed.
Vegetation heightLinearMeasured height of community (m)

As the chosen methods of computing functional diversity (see Data analysis) are dependent on there being more species than traits and on traits being quantitative (Villéger, Mason & Mouillot 2008), trait data were limited to eight traits. Traits chosen for the analysis covered the vegetative phase and the reproductive phase of a plant’s life cycle. The former comprised canopy height (log-transformed), leaf size (log-transformed), leaf dry matter content (LDMC) and specific leaf area (SLA), all sourced from the LEDA trait data base (Kleyer et al. 2008), which together carry information about growth rate and competitive ability. The traits concerned with the reproductive phase were seed longevity (computed from Thompson, Bakker & Bekker 1997), seed mass (log-transformed from BiolFlor, Klotz, Kühn & Durka 2002), terminal velocity (Kleyer et al. 2008) and variance between length, width and height of the seed (Thompson, Band & Hodgson 1993; data from Klotz, Kühn & Durka 2002). These traits cover aspects of dispersal in space and time. Where gaps were present (c. 2% of cases), information was supplemented with data from floras and by averaging data from congeners.

Data analysis

Six metrics of diversity were computed for each site. Three were measures of taxonomic diversity: the total number of species recorded at each site, the mean species richness per 1 × 1 m quadrat and the mean Shannon diversity index per 1 × 1 m quadrat. The other three were measures of function diversity: the FRic, FEve and FDiv measures of Villéger, Mason & Mouillot (2008). FRic represents the convex hull volume of functional space occupied by the community (Cornwell, Schwilk & Ackerly 2006), FEve or FEve represents the regularity of distribution in abundance in this volume and FDiv represents the divergence in the distribution of the species traits within the trait volume occupied. They were calculated using the methods and scripts of Villéger, Mason & Mouillot (2008). This approach was possible as all traits were quantitative and fewer traits were selected than the minimum species richness. Qualitative and semi-quantitative traits, and more traits than species, could be accommodated with an initial multivariate step to reduce dimensionality (Laliberté & Legendre 2010) though that approach does not capture all the variation in the initial trait data and significant inter-trait correlations can be lost.

To test if the functional diversity indices were independent of each other, the three measures (FRic, FEve and FDiv) as well as the three measures of species richness for each site were subject to correlation (Pearson). To identify how unusual the measured values of Functional Diversity were, the value for each site was compared against those from 999 simulated assemblages. These simulated assemblages were constructed using all the species recorded across the sites, and species were randomly drawn from this list without replacement and allocated an abundance from the list of abundances of the species at that site (species not recorded at a site had, therefore, an abundance of zero). This method of simulated assemblage construction maintained the species richness and the pattern of abundance within each site – a matrix swap randomisation (Manly 1995). In addition, a null model weighted by the frequency of occurrence of each species across the 30 sites was also tried following the method of Kraft, Valencia & Ackerly (2008).

There is much a priori evidence to expect that FRic should be lower than random through the operation of selective filters (Cornwell, Schwilk & Ackerly 2006). However, competition or neutral patterns cannot be excluded. Hence all measures were assessed with a two-tailed test – an index was deemed significantly different from random if it was ranked less than 975th in the rankings of the values from the simulated communities or higher than or equal to 25th. In addition, the index of variance (IV) of each of the functional diversity measures was calculated, where IV = 2 × (Obs/(Obs + Exp)) − 1. This provides a stable index of how observed values of the test statistics vary against expected (Mason et al. 2008). They were also correlated with the measures of species richness.

Finally, the indices and their IVs were subject to linear and quadratic regression against the indices of disturbance (Table 1) and the measures of ecosystem productivity. A quadratic relationship between functional diversity and the environmental parameters was tested for, as well-known quadratic relationships exist between species diversity and both disturbance (Grime 1973; Connell 1978) and productivity (Lepŝ 2005). Quadratic relationships were only reported if they were a significant improvement on the linear regression and if the maximum or minimum occurred within the bounds of the data. This was tested using the Mitchell-Olds and Shaw test (Mitchell-Olds & Shaw 1987), implemented in R using vegan (Oksanen et al. 2011). As there are similar relationships between diversity and standing biomass (Al-Mufti et al. 1977), live, dead and total standing biomass from the samples used to estimate ANPP were also included in the modelling. The quadratic model was chosen over the linear if an analysis of variance between the two models was significant at P < 0.05.

All analyses were carried out in R version 2.10.0 (R Development Core Team 2009) with the calculation of the functional diversity indices done using the F_RED function of Villéger, Mason & Mouillot (2008).


Independence of diversity measures

Correlation analysis, as expected, showed a strong correlation between site (total richness across four quadrats) and quadrat-level species richness (Table 2). It also showed strong correlations between Shannon diversity and both measures of species richness.

Table 2.   Correlation matrix between species richness at a site level (Number of species) and quadrat level (Sp. m−2), species diversity (Shannon) and functional diversity indices: FRic = Functional Richness; FEve = Functional Evenness; FDiv = Functional Divergence; IV = index of variance. *0.05 ≥ P > 0.01, **0.01 ≥ P > 0.001, ***P < 0.001
 Number of speciesSp. m−2ShannonFRicFEveIV FRicIV FEve
Sp. Density0.89***      
IV FRic−0.31−0.47**−0.40*    
IV FEve−0.12−0.01−0.02  0.32 
IV FDiv0.240.09−0.04  −0.130.10

Of the three possible correlations between the different measures of functional diversity, none were significant or even approached significance. This was the same for their indices of variance. Also, of the nine correlations between the measures of species diversity and functional diversity, only one was significant. This was a positive correlation between FRic and number of species recorded at a site level. However, there were also significant negative relationships between FRic’s IV and quadrat species richness and Shannon diversity. FRic was much lower than expected at high levels of species richness and diversity. Using the occurrence-weighted null model had no appreciable effect on the correlation between the measures of species diversity and the indices of variance (Table S2).

Difference of functional diversity indices from random

It is clear from the comparison of the measured FRic values with calculated values from random communities that 13 (of 30) sites show values ranked in the lowest 2.5% of those from simulated communities (Fig. 1a). The communities that showed this feature included those on arable land, fallows, silage fields, pastures and abandoned croft land. Communities with lower management intensities had ranks that were not significantly different from those expected. No community had a ranking of FRic in the top 2.5% of random draws for that abundance and richness pattern. The abundance-weighted null model produced a similar result (Fig. S1a).

Figure 1.

 Histograms showing the distribution of rankings of the measured functional diversity indices for each of the 30 sites as compared with the 999 calculated ranks; (a) FRic = Functional Richness; (b) FEve = Functional Evenness; (c) FDiv = Functional Divergence.

The pattern was very different for FEve (Fig. 1b). Here, the rankings of the individual communities showed a spread across all possibilities. However, one heathland and one silage field showed higher than expected levels of evenness, and one abandoned silage field a lower than expected FEve. Again, a similar distribution of higher and lower than expected evenness values occurred as would be expected by chance from 30 tests. A similar pattern was also observed for FDiv (Fig. 1c), although here, only one site, an arable plot, showed an extreme value for this measure that put it in the highest 1% of the communities created by random draws. The abundance-weighted null models for FEve and FDiv produced similar results (Fig. S1b,c).

Relationship between functional diversity and the environment

The species-based measures of diversity showed some of the expected hump-backed relationship with productivity (Table 3, all reported quadratic relationships passed the Mitchell-Old & Shaws test). In particular, there was evidence of a peak in site species richness at intermediate levels of dead biomass (187 g m−2, R2 = 0.218, P = 0.036), as well as weak evidence for peaks at intermediate levels of live biomass (407 g m−2, R2 = 0.183, P = 0.065) and total standing biomass (582 g m−2, R2 = 0.182, P = 0.066). Similarly, species richness per quadrat peaked at 172 g m−2 dead biomass (R2 = 0.251, P = 0.018, Fig. 2a), Shannon diversity at 354 g m−2 live biomass (R2 = 0.208, P = 0.043) and 579 g m−2 total standing mass (R2 = 0.221, P = 0.035). There was weak evidence that quadrat richness peaked at 538 g m−2 total standing biomass (R2 = 0.181, P = 0.068). There was also a peak in site species richness (R2 = 0.283, P = 0.011) and weak evidence for a peak involving species richness per quadrat (R2 = 0.195, P = 0.053), for intermediate levels of soil nitrogen release.

Table 3.   Significance of the fitted curves between the species and functional diversity indices and the selected environmental parameters. The results for the functional diversity indices and their index of variance are shown in the same column separated by ‘/’. The shape of the relationship is indicated by: + positive linear trend; − negative linear trend; U quadratic trend with a minimum within the data range; and ∩ quadratic trend with a maximum within the data range. The strength of the relationship is indicated by the number of symbols, for example: (+) 0.10 ≥ P > 0.05; + 0.05 ≥ P > 0.01; ++ 0.01 ≥ P > 0.001; +++ 0.001 ≥ P. FRic = Functional Richness; FEve = Functional Evenness; FDiv = Functional Divergence; IV = index of variance
 Number of speciesSp. m−2ShannonFRic/IV FRicFEve/IV FEveFDiv/IV FDiv
ANPP   ///
Live Biomass(∩) ///
Dead Biomass /−−−/−−/
Standing mass(∩)(∩)/−/(−)/
Vegetation C:N   /+//
log soil N release(∩) (−)/−−UU/U/
Disturbance frequency ∩∩(∩)/−−+//
Biomass loss(∩)∩∩∩/−−+//
Below-ground disturbance∩∩ /+//
Start of disturbance(−)−−/++(−)//
Vegetation height   /+//
Figure 2.

 Example relationships between diversity indices and the environment: species richness per quadrat against (a) dead biomass and (b) biomass loss, index of variance (IV) of Functional Richness (FRic) against (c) loss soil nitrogen release and (d) the timing of the start of disturbance, and IV of Functional Evenness (FEve) against (e) dead biomass and (f) log soil nitrogen release. Fitted curves and their respective R2 and P-values are shown.

The patterns in the functional diversity indices were very different. FDiv showed no significant relationships with any of the productivity measures used, nor did its IV. FRic was weakly negatively related to soil nitrogen release (R2 = 0.264, P = 0.087). However, its IV was significantly positively related to vegetation C : N (R2 = 0.145, P = 0.038) and to vegetation height (R2 = 0.264, P = 0.087) and negatively related to soil nitrogen release (R2 = 0.276, P = 0.003, Fig. 2c). As an illustration, two sites of contrasting FRic are shown in Fig. 3. The winter-grazed site, D2R, had much lower levels of soil nitrogen and higher vegetation C : N ratio than the arable site, D6CA, and considerably higher FRic for both the two traits illustrated and overall. FEve was strongly negatively related to both dead biomass (R2 = 0.378, P = 0.0003) and total standing biomass (R2 = 0.183, P = 0.018). It also showed a minimum at intermediate levels of soil nitrogen release (R2 = 0.331, P = 0.004). Its IV was also negatively related to dead biomass (R2 = 0.253, P = 0.005, Fig. 2e) and weakly to standing mass (R2 = 0.129, P = 0.051), and also showed a minimum at intermediate levels of soil nitrogen (R2 = 0.264, P = 0.016, Fig. 2f).

Figure 3.

 Functional Richness (FRic) (equal to the total area of the convex hull) of two plots showing extreme values of FRic for just two traits, specific leaf area (SLA) (mm2 mg−1) and seed mass (mg). •: D2R, a winter-grazed site and ○: D6CA, an arable site.

In general, there were also strong, quadratic relationships between the species diversity measures and the measures of the degree of disturbance (Table 3). Of the nine possible relationships between the degree of disturbance and species diversity, five were significant at P < 0.05 (range of R2: 0.200–0.414), and a further two relationships had P-values < 0.1. As an example, the relationship between quadrat species richness and biomass loss is shown in Fig. 2b. There were also significant, negative relationships between the start of disturbance and both species richness per quadrat (R2 = 0.277, P = 0.003) and Shannon diversity (R2 = 0.147, P = 0.036), as well as a weak relationship with site species richness (R2 = 0.101, P = 0.086). This indicated that species diversity was higher where disturbance occurred earlier in the year.

Neither FRic nor FDiv showed any significant relationships with the chosen measures of disturbance. However, the IV of FRic was strongly negatively correlated with disturbance frequency (R2 = 0.222, P = 0.009) and biomass loss (R2 = 0.228, P = 0.008), and positively with the start of disturbance (R2 = 0.271, P = 0.003, Fig. 2d); the observed declined in relation to the expected value as disturbance increased in intensity or arrived earlier. FEve showed a positive relationship with the frequency of disturbance (R2 = 0.264, P = 0.003), the severity of disturbance (R2 = 0.184, P = 0.010) and the time since below-ground disturbance (R2 = 0.213, P = 0.020). It also showed a weak positive relationship with time of disturbance (R2 = 0.095, P = 0.097). Its IV was unrelated to disturbance. Indices of variance calculated with the occurrence-weighted null model produced similar patterns (data not shown).


Independence of diversity measures

There was no correlation between the three functional diversity measures which confirmed the expectation from artificial communities shown by Villéger, Mason & Mouillot (2008) and Mouchet et al. (2010). Of the correlations between the functional and species diversity measures, only the correlation between FRic and site species richness was significant and showed the expected positive correlation. However, it must be stated that the power of these tests is not high, as n = 30. This followed from the sampling effect where more species increase the convex hull volume (Cornwell, Schwilk & Ackerly 2006). However, it contrasted with the pattern seen in estuarine fishes, where FRic saturated at intermediate levels of species richness (Villéger et al. 2010). It also contrasted with the pattern observed for lake fishes where the IV of FRic peaked at intermediate species richness (Mason et al. 2008), whilst in this study IVFric declined with species richness per square metre (Table 2). This indicated that species added within the species-rich assemblages of this study were not completely redundant, and that more trait space was exploited within these species-rich communities. Possibly, the species pool within this study could not supply many species able to survive at the highly disturbed sites to produce the same effect. The independence of the IV of FEve and FDiv (as well as the original indices) from species richness in this study also contrasted with the pattern in Mason et al. (2008) where IV FEve continued to increase as species richness increased and IV FDiv peaked or saturated at moderate levels of species richness. This indicates that assembly rules were relatively independent of species richness, as suggested by Mouchet et al. (2010).

The patterns in the simulated data showed different trends to those from Villéger, Mason & Mouillot (2008) and Mouchet et al. (2010). Mean FRic of the 999 simulations increased in an exponential fashion (Fig. 4a) against species richness over the range investigated, whilst in Villéger et al.’s simulated data the relationship appeared to saturate, as it did in Mouchet et al.’s when limiting similarity was included in the simulation. This was due to the large and diverse species pool across the sites which would likely saturate at higher species richness levels than those tested. Simulated FEve was unrelated to species richness (Fig. 4b), whereas in the simulated data of Villéger, Mason & Mouillot (2008) and Mouchet et al. (2010) it declined non-significantly with species richness. Simulated FDiv decreased, albeit very slowly, as species richness increased (Fig. 4c), whilst the simulated values in Villéger, Mason & Mouillot (2008) and Mouchet et al. (2010) were unrelated to species richness. Similar patterns were seen with the weighted null model (Fig. S2).

Figure 4.

 Patterns of measured (•) and mean simulated (○) diversity indices against species richness for (a) Functional Richness; (b) Functional Evenness; and (c) Functional Divergence.

Difference of functional diversity indices from random

Around half of the plots (13/30) showed FRic values lower than expected from random (Fig. 1a). This highly skewed pattern suggests a high degree of habitat filtering within the system as assemblages occupied less trait space than expected (Cornwell, Schwilk & Ackerly 2006), although this was only significant at a site level for a proportion of sites. This replicated the findings of Cornwell, Schwilk & Ackerly (2006) in Californian woody vegetation where 40 of 43 plots had lower than expected values of this measure. Expressing the data here on the same basis showed that 23 of 30 plots were lower than expected, i.e. ranked below 500, and a two-tailed Wilcoxon signed-rank test gave P < 0.001 that the measured values and the calculated ones were from the same distribution. This indicates a general prevalence of habitat filtering across the sites studied.

The two other measures of functional diversity showed no difference from a random pattern. Similar Wilcoxon signed-rank tests between measured and mean calculated values for both indices indicated there was no significant relationship between them (Evenness P = 0.612, Divergence P = 0.129). There are no values in the literature against which these results could be compared, but theoretical expectations from Mouchet et al. (2010) indicate that lower than expected values should again indicate habitat filtering. There was no strong evidence of general patterns in this data.

Relationship between functional diversity and the environment

Functional Richness indicates the degree of habitat filtering in operation (Cornwell, Schwilk & Ackerly 2006; Mouchet et al. 2010); the latter study in particular showed that FRic was very sensitive to environmental filtering in simulated communities despite its dependence on species richness. As well as strong evidence for overall filtering effects (see Difference of functional diversity indices from random), examination of the IV of FRic clearly showed that FRic is reduced against expectations as both productivity and disturbance increase (Table 3, Fig. 2a,b). This suggests that the degree of habitat filtering increased as the intensity of land use increased, as plots subject to cropping, silage making, grazing or hay making had the lowest values of the IV of FRic. The decline in FRic with productivity and disturbance was in contrast to the general patterns for species diversity, which indicated peaks at intermediate levels of productivity and disturbance [c.f. the hump-backed model (Al-Mufti et al. 1977) and the intermediate disturbance hypothesis (Grime 1973)]. This indicates that species packing within functional trait space is highest at intermediate levels of disturbance and productivity, indicating a higher degree in functional redundancy at high species richness. There is little functional redundancy at low levels of productivity and disturbance. This pattern was also in contrast to the findings of Biswas & Mallik (2010) who demonstrated higher FRic at intermediate levels of disturbance. Potentially their highly disturbed habitats – forest clearcuts – represented much higher levels of disturbance than those of the sites in this study, and their least disturbed habitats were mature forests where competition for light may impose strong habitat filtering, so the results may not be contradictory with this study. However, the results of this study agree with the findings of Lalibertéet al. (2010) and Flynn et al. (2009) who both showed that the intensification of land use reduced functional diversity and redundancy. The increased filtering in this study arose largely from reduced ranges in canopy height, LDMC and seed mass as both productivity and disturbance increased, and variance in seed dimensions and seed terminal velocity as only disturbance increased (Pakeman, Lennon and Brooker in press). The range in leaf size, SLA and seed longevity appeared relatively insensitive to the environmental factors investigated.

Functional Evenness is an indicator of how functional space is occupied within a community, and it decreases with less even or less regularly spaced species in trait space (Villéger, Mason & Mouillot 2008) and was sensitive to environmental filtering in simulated communities (Mouchet et al. 2010). In this study, FEve (and the IV of FEve) was reduced in sites with high levels of litter and standing biomass (Table 3, Fig. 4e), indicating a concentration of species abundances along a small part of the functional trait gradient: the dominant species then were similar in trait values, possibly indicating a high degree of habitat filtering (Mouchet et al. 2010). This pattern contrasted with the hump-backed pattern displayed by species diversity. It also showed a minimum at intermediate levels of soil nitrogen release (IV of FEve shown in Fig. 4f). This was the opposite pattern to species diversity, although the negative relationship between FEve and species richness was not significant. However, a clear pattern emerged between FEve and disturbance; evenness was promoted by increased levels of disturbance and an earlier timing of disturbance. However, there was no increased departure of FEve from expected, IV FEve did not increase with increased disturbance (Table 3). This provides some evidence that low levels of FEve are indicative of sites with little disturbance, and that more disturbed communities are characterised by species that are more spread out along functional trait gradients and have similar abundances. This suggests that in habitats where competition is of low importance in structuring communities, such as highly disturbed ones, FEve can be high (even though FRic is low), whereas in less disturbed habitats, competition is possibly more important in structuring the community. This explanation based on patterns in evenness reflecting the importance of competition would fall if FDiv decreased as FEve decreased. This would indicate further habitat filtering as FDiv decreases if abundant species are close to centre of the functional volume. However, FDiv was unrelated to the measures of productivity and disturbance used in this study and thus it appears competition was structuring the communities at low levels of disturbance.

The study clearly showed that habitat filtering was operating within the study area, even with the relatively small number of sites available. There was a strong reduction in FRic compared with expectations at almost half the sites, and this reduction was correlated with increased productivity and disturbance. However, in contrast, FEve increased with disturbance and this was complemented by evidence of a decline at sites with high biomasses present. This indicated that at low levels of productivity and disturbance, these sites displayed high FRic and low FEve (little habitat filtering but strong structuring by competition), that at higher levels of productivity FRic declined but there was no change in FEve (increased habitat filtering), but that at higher levels of disturbance the FRic declined as FEve increased (increased habitat filtering and reduced importance of competition). This indicates that assembly rules concerning competition and habitat filtering were operating in different ways along gradients of productivity and disturbance. The study also confirmed that the three measures of functional diversity were orthogonal to each other in practice (Villéger, Mason & Mouillot 2008; Mouchet et al. 2010). It also confirmed that one measure, FRic, was clearly linked to species richness through the sampling effect, but that the other two were not.

There is now a need to increase the number of studies like this to see how these indices behave in practice. This would enable their use to become established as indicators of processes governing community assembly. Once this is accomplished, there is the possibility of developing them as measures of ecosystem function that may help explain further variance in relationships between ecosystem processes and functional traits (Díaz et al. 2007).


I thank Iain Turnbull of the National Trust for Scotland for all his help in arranging access. I also thank the many crofters and the grazing clerks of Balmacara, Drumbuie, Duirinish and Plockton for their help in this work. Rob Brooker, Jenni Stockan, Antonia Eastwood, Roger Cummins and Jelle van Rijmenant all helped with the fieldwork. Comments from Norman Mason and two anonymous referees substantially improved the paper.