Functional diversity in a large river floodplain: anticipating the response of native and alien macroinvertebrates to the restoration of hydrological connectivity

Authors

  • Amael Paillex,

    Corresponding author
    1. Institut F.A. Forel, Laboratoire d'Ecologie et de Biologie Aquatique, Université de Genève, Carouge, Switzerland
    • Aquatic Ecology Group, Department of Zoology, University of Cambridge, Cambridge, UK
    Search for more papers by this author
  • Sylvain Dolédec,

    1. Laboratoire d'Ecologie des Hydrosystèmes Naturels et Anthropisés, CNRS, UMR 5023, Université Lyon1, Villeurbanne Cedex, France
    Search for more papers by this author
  • Emmanuel Castella,

    1. Institut F.A. Forel, Laboratoire d'Ecologie et de Biologie Aquatique, Université de Genève, Carouge, Switzerland
    Search for more papers by this author
  • Sylvie Mérigoux,

    1. Laboratoire d'Ecologie des Hydrosystèmes Naturels et Anthropisés, CNRS, UMR 5023, Université Lyon1, Villeurbanne Cedex, France
    Search for more papers by this author
  • David C. Aldridge

    1. Aquatic Ecology Group, Department of Zoology, University of Cambridge, Cambridge, UK
    Search for more papers by this author

Correspondence author. E-mail: adp49@cam.ac.uk

Summary

  1. Rivers and floodplains are among the most threatened ecosystems. Hydroelectric power plants and embankments have reduced the hydrological connectivity between rivers and their floodplain channels, leading to loss of freshwater habitats and biological communities. To prevent and reverse such loss, numerous restoration programmes have aimed at rejuvenating the lateral hydrological connectivity between rivers and floodplain channels. Despite considerable global attention, we know remarkably little about the ecological benefits of floodplain restoration programmes.
  2. We analysed the functional diversity of different macroinvertebrate groups (natives and aliens) along a gradient of lateral hydrological connectivity on the Rhône river in France. We used 36 sampling sites to describe the functional diversity (Rao's quadratic entropy) before and after the enhancement of the lateral hydrological connectivity by restoration. The effects of restoration on functional diversity were tested for each macroinvertebrate group and at multiple spatial levels (alpha and beta).
  3. Before restoration, alpha functional diversity of the entire macroinvertebrate community peaked in sites with a high lateral connectivity. The contribution of the native groups to functional diversity was higher than that of the alien group. The latter was not constrained by high values of lateral hydrological connectivity and reached a maximum in highly connected sites.
  4. After restoration, within-site functional diversity (alpha FD) declined linearly following the enhancement of lateral hydrological connectivity. The restoration operations increased the contribution of the aliens to functional diversity and reduced the contribution of a group of native taxa. In addition, among-sites functional diversity (beta FD) was successfully enlarged by restoration.
  5. Synthesis and applications. The lateral hydrological connectivity (LHC) represents a key parameter for explaining the functional diversity (FD) of macroinvertebrates in a floodplain ecosystem. Our results demonstrate that restoration-induced changes to functional diversity can be predicted. Controversially, restoration-induced enhancement of lateral hydrological connectivity increased the functional diversity of the alien macroinvertebrates. However, these species contributed only to a small part of the total macroinvertebrate functional diversity. We recommend that restoration programmes diversify the levels of lateral hydrological connectivity among the channels to ensure an optimal functional diversity at the floodplain scale.

Introduction

Large river floodplains are deterministic ecosystems, in which the lateral hydrological connectivity (LHC) describes the exchanges by water of energy, material and organisms from the river to secondary floodplain channels. LHC is known to influence the species composition, the species richness and the functional response of aquatic communities (Ward, Tockner & Schiemer 1999; Amoros & Bornette 2002; Paillex et al. 2009). Evidence of floodplain degradation has resulted in an increasing number of costly restoration programmes (Bernhardt et al. 2005). However, despite this global attention, the ecological benefits of such programmes remain poorly documented (Bernhardt et al. 2005; Palmer et al. 2005).

LHC is essential to floodplain functioning as it sustains a mosaic of habitat conditions and diverse stages of ecological successions from early to late communities (Amoros & Bornette 2002). In natural large river floodplains, the secondary channels can be arranged along a wide continuum of LHC from totally disconnected to permanently connected with the river channel (Ward, Tockner & Schiemer 1999). However, worldwide human activities, such as construction of hydroelectric power plants and embankments have caused dramatic reductions to LHC, which has accelerated terrestrialization of the floodplain channels and resulted in a loss of aquatic habitats and native species (Tockner & Stanford 2002; Nilsson et al. 2005). Furthermore, human activities have favoured the colonization of these habitats by invasive species (Strayer 2010). The resulting changes in biodiversity, driven by land use and biotic invasion, are expected to have further deleterious impacts on ecosystem functioning (Chapin et al. 2000). Growing awareness of the rapid degradation of running waters has therefore promoted river restoration schemes to encourage communities of organisms that optimize ecological functions and processes (Palmer et al. 2005).

In this context, estimating functional diversity (FD) in large rivers is of growing interest to monitor biodiversity (Bady et al. 2005) and to assess the effect and success of restoration programmes (Cadotte, Carscadden & Mirotchnick 2011). FD can be defined as the functional trait differences among species in a given community, which is suggested to reflect some aspects of ecosystem function (e.g. stability or resilience in response to disturbance) (Mayfield et al. 2010; Cadotte, Carscadden & Mirotchnick 2011). Numerous measures of FD have been developed and used to describe biodiversity (Mouchet et al. 2010), with some including quantitative measures such as abundance to weight the contribution of each species to overall diversity (e.g. Rao's quadratic entropy; Rao 1982). Such measurement has proved useful for the monitoring of freshwater ecosystems (e.g. Bady et al. 2005; Péru & Dolédec 2010). However, in large river floodplain, little attention has been given to the response of FD along a gradient of LHC. Moreover, to date no studies have considered how restoration of LHC might affect FD at the floodplain scale. Monitoring the effects of large river floodplain restoration provides a good opportunity to understand the role of LHC and its restoration on the FD of aquatic communities.

In 1998, a large programme was designed to restore the ecological functioning of the Rhône river and its floodplain. The project aimed at enhancing the fluvial dynamic in several sections of the river by (i) increasing the minimum river flow in the by-passed sections and (ii) rehabilitating the connections between the river and secondary floodplain channels (Olivier et al. 2009). Both measures have led to an increase in LHC within secondary channels and a rejuvenation of macroinvertebrate communities to early successional stages (Paillex et al. 2009). However, given the limited knowledge to date about FD in floodplains and the effect of restoration, we aimed (i) to model the response of macroinvertebrate FD to LHC, (ii) to study the contribution of different macroinvertebrate groups (aliens, lentic and lotic natives) to the total FD and (iii) to assess the effect of LHC restoration on macroinvertebrate FD at the local and floodplain scale (i.e. alpha and beta functional diversities).

A series of hypotheses were developed to test the response of FD to LHC before and after its restoration. First, according to landscape and niche-filtering theories, LHC should select traits that allow species to persist and live in a given area (Poff 1997; Statzner, Dolédec & Hugueny 2004; Mouchet et al. 2010). Therefore, we expected that a high LHC in permanently or frequently connected channels would select only few traits allowing species to maintain and develop communities, which would result in more similarity among species (Fig. 1). Conversely, we expected that a low LHC would favour higher competitors yielding dissimilarity among the species to avoid competition (Hardin 1960) (Fig. 1). Second, according to compositional changes along LHC (Paillex, Castella & Carron 2007; Leigh & Sheldon 2009), we hypothesized that different groups of species would contribute differently to the total FD. We expected that a lentic group (i.e. lenitophilic species) would have a maximal contribution to FD in the most disconnected channels, while lotic ones (i.e. rheophilic species) would make a maximal contribution to FD in highly connected channels. Alien species being mostly lotic species (Statzner, Bonada & Dolédec 2008) were expected to contribute most to FD in highly connected channels. Third, according to the known impact of LHC restoration on macroinvertebrate characteristics (Reckendorfer et al. 2006; Paillex et al. 2009), we hypothesized that the enhancement of LHC by restoration would intensify niche filtering and reduce the dissimilarity among macroinvertebrate species, resulting in a reduction in FD within the restored sites.

Figure 1.

Conceptual framework for the expression of functional diversity (FD) within sites of a floodplain ecosystem (alpha FD). (a) The lateral hydrological connectivity (LHC) is defined as a key factor influencing functional trait diversity in a floodplain. (b) The gradient of lateral connectivity is composed of a succession of channels from permanently connected to totally disconnected channels (a. permanently connected, b. upstream disconnected, c. totally disconnected but frequently reconnected during flood, d. totally disconnected and rarely reconnected; adapted from Richardot-Coulet & Greenwood 1993). (c) FD is related to assembly rules in response to the LHC (1. Competitive exclusion and limiting similarity, 2. Niche filtering). (d) The effect of assembly rules on the species pool is summarized using three different assemblages for clarity. (e) The expected effect of LHC changes on FD is a reduction in highly connected channels and stability in the less connected and totally disconnected channels.

Materials and methods

Study area and sites

The study was carried out in two sections of the Rhône river in France (I: Belley, II: Brégnier-Cordon) (Olivier et al. 2009). A low-fall hydroelectric power plant was located in each studied section for which a head-race canal deflects the river flow up to the power plant (see Fig. S1, Supporting information). The main river channel supplies the floodplain with a minimum discharge established for each month of a year and which is exceeded during flood periods. Eighteen secondary channels were surveyed before and after restoration, seven in Belley and eleven in Brégnier-Cordon. Two sampling sites were selected in each channel: one upstream and one downstream, making a total of 36 studied sites (see Fig. S1). The two sites per channel represent the extremes of the habitat conditions currently used for describing disconnected floodplain channels (Paillex, Castella & Carron 2007).

Two years after restoration, 16 of the 36 sampling sites remained unrestored and were used as control, whereas 20 sites encompassed different types of restoration works. We classified the restored sites into three categories according to their main restoration types: (i) Flow increase sites: for the sites (n = 6) restored only by an increase of the minimum discharge imposed to the natural floodplain. These sites were situated in the Belley section for which the minimum monthly discharge increased from 25–60 m3 sec−1 before restoration to 60–100 m3 sec−1 after restoration. (ii) Dredging sites: for the sites (n = 6) additionally deepened by dredging but without direct reconnection to the river channel. Most of the sites were situated in the upstream part of the channels, and we considered the flow increase as less influential on the short term for the biodiversity of these sites than the dredging of the bed sediment. (iii) Reconnection sites: for the sites (n = 8) directly reconnected to the main river channel by dredging or for the sites reconnected after the removal of an alluvial plug in the upstream part of the channel (see Table S1 for detailed information on sites, Supporting information).

Macroinvertebrate sampling

Within each site, four sampling quadrats (area: 0·25 m2 per quadrat) were randomly selected along a 30-m stretch. Macroinvertebrates were sampled in each quadrat with a hand-net (mesh size: 500 μm). To account for seasonal variations, samples were taken in spring and summer. Sampling was performed 1 year before restoration (Belley: summer 2003 and spring 2004, Brégnier-Cordon: summer 2004 and spring 2005) and repeated 2 years after restoration (Belley: spring and summer 2007, Brégnier-Cordon: spring and summer 2008). Samples were preserved in alcohol and sorted in the laboratory. Macroinvertebrates were identified to the finest taxonomic resolution possible. Oligochaeta were omitted from further analyses due to their fragmentation by the sampling method and the resulting over-count.

Biological traits

To assess FD, 63 categories distributed among 11 biological traits were considered (Table S2, Supporting information). The association between a taxon and a trait category was described using an affinity score based on a fuzzy coding technique (Chevenet, Dolédec & Chessel 1994). The affinity of a taxon to each category was coded from 0 (no affinity) to 3 or 10 (maximum affinity depending on the source code). Affinity scores were standardized, so that their sum for a given taxon and a given trait was equal to 1. This procedure ensured the same contribution of each trait to the calculation of FD (Péru & Dolédec 2010). We computed the frequency of each category within a trait for each taxon, and we replaced missing data by the average contribution of all taxa to the corresponding trait.

Macroinvertebrate classification

The macroinvertebrate densities (individuals m−2) were log(x + 1)-transformed prior to analyses, and only those species present in more than one site were retained. We used correspondence analysis (COA) to study the taxonomic composition of the entire community before restoration (e.g. Hill 1973). The taxonomic distance among species was assessed using Euclidean distance calculated from the scores of the first two COA axes. We further applied Ward's clustering method to identify different groups of taxa (Ward 1963). According to the species identity in the groups and their changes in density along LHC before restoration, we assigned them to one of two categories: lotic or lentic. Alien species were identified according to reference lists (Devin et al. 2005; DAISIE 2009). Three groups of macroinvertebrates resulted from these analyses: (i) aliens, (ii) lotic native and (iii) lentic native.

Functional diversity measurement

Fuzzy correspondence analysis (FCA; Chevenet et al. 1994) was performed on the species-trait matrix, which yielded species scores accounting for the trait dissimilarity among species. We considered the first two FCA axes scores and computed a Euclidean distance between species. FD was further calculated for each site using Rao's quadratic entropy (Rao 1982). Rao's index weights the trait dissimilarity among species by their relative abundance (Botta-Dukat 2005) and allows a decomposition of the information in alpha, beta and gamma levels (Villéger & Mouillot 2008; Mouchet et al. 2010). A high FD value measured with Rao's index reflects a high trait diversity among species weighted by their abundances (Botta-Dukat 2005). FD was calculated for the entire macroinvertebrate community and for the three groups previously defined.

The contribution of each macroinvertebrate group to the total FD and the among-sites FD (i.e. beta diversity) were both calculated using a double principal coordinates analysis (i.e. DPCOA) (Pavoine, Dufour & Chessel 2004). We considered the first two DPCOA axes for computing the contribution of each taxon to FD within the three previously defined groups of macroinvertebrates. We considered the site-to-site distances pre- and post-restoration generated from the DPCOA to calculate the changes of FD among-sites after restoration (i.e. change in beta diversity post-restoration).

Lateral hydrological connectivity

Following Paillex, Castella and Carron (2007), we sampled five environmental variables known to be representative of the LHC within each site, 1 year before restoration and 2 years after restoration. Three variables were expected to decline with increasing LHC: (i) water conductivity, (ii) organic content of the sediment and (iii) aquatic vegetation density. Two variables were expected to increase accordingly: (iv) diversity of the mineral grain size and (v) NH3-N concentration of the water. These five variables were analysed by a principal component analysis (PCA), and the score of the sites obtained along the first PCA axis was used as a surrogate of LHC for each site (Paillex, Castella & Carron 2007; Paillex et al. 2009). A value of LHC was derived for each site 1 year before restoration (i.e. 36 values pre-restoration), and the post-restoration sites were plotted as supplementary individuals on the pre-restoration PCA providing a post-restoration value of LHC for each site (i.e. 36 values post-restoration).

Models and predictions

A rarefaction was applied to the total richness per site using the lowest abundance found in all sites as the target number of individuals (Oksanen et al. 2011). The relationships between rarefied richness, FD and LHC were analysed using polynomial regressions. Significant improvement of the higher-order regression coefficient was assessed by an analysis of variance (anova). The explained variance by a given higher-order model was compared by anova to a smaller-order model. The higher-order model was rejected when its explained variance was not significantly improved compared with the smaller-order model. Influential outliers were removed from analyses according to the cut-off for Cook's distance defined as D = 4/(n-k-1), where n is the number of observations (i.e. 36 sites), and k is the number of parameters in the model (i.e. k = 1; LHC) (Cook & Weisberg 1982; Chatterjee & Hadi 1988). An analysis of covariance (ancova) was applied to test whether the intercept of the relationship between FD and rarefied richness differed among restoration periods (i.e. restoration treatment) or the slope of the relationship differed between periods (i.e. LHC × restoration treatment). To study the effect of LHC changes on FD, we calculated the deviations in LHC and FD after restoration for each site. The relationship between both deviations was tested by simple linear regression.

We studied the effect of restoration by examining the difference between predictions based on pre-restoration models with post-restoration measurement of FD. We computed the expected FD for each sampling site using pre-restoration models and post-restoration LHC values. Therefore, we obtained one predicted value of FD for each site (i.e. 36 predicted values in total) that we compared with the observed values of FD for each site after restoration (i.e. 36 observed values). To study the effect of restoration works, we grouped post-restoration sites according to their main restoration types [(i) flow increase, (ii) dredging, (iii) reconnection and (iv) control] and compared post-restoration predictions of FD with observations. Wilcoxon signed-rank tests were performed to compare predictions and post-restoration observations, and to test the effect of restoration on FD within and among-sites. To account for multiple comparisons and increase rate of Type I errors, we applied a Bonferroni-Holm procedure (Holm 1979).

Multivariate statistics were computed with the ade4 package (Chessel, Dufour & Thioulouse 2004) implemented in R 2.13.0 freeware (R Development Core Team 2011). Function divc() of the ade4 package was used to calculate FD within samples, and function dpcoa() was used to calculate FD among samples. For rarefaction, we used function rarefy() of the vegan package (Oksanen et al. 2011). Other statistics for linear models, Wilcoxon signed-rank tests and ancova, were computed with R 2.13.0 (R Development Core Team 2011).

Results

Lateral hydrological connectivity and rarefied richness

The first two axes of the PCA performed on environmental variables before restoration explained 71·8% of the variability. Sites before restoration were ordered along the first PCA axis from the most disconnected to the frequently and permanently connected sites with the main river channel (see Fig. S2, Supporting information). Given the high proportion of the variance explained by the first PCA axis (43%) and its strong correlation with three of the environmental variables (vegetation cover, diversity of the mineral grain size and organic matter content in sediment), the scores of the sites along the first PCA axis were used as a surrogate for the relative LHC of each site. The average rarefied richness within-sites before restoration equalled 26 taxa for 500 individuals. No relationship occurred between the rarefied richness per site and LHC, either using a simple linear regression (R2 = 0·03, P = 0·31) or a quadratic one (R2 = 0·14, P = 0·075).

Macroinvertebrate groups

The overall macroinvertebrate richness in the 36 sites comprised 186 taxa before restoration. Two clusters that grouped 51 native taxa, encompassing species living in running waters, were combined to form the lotic group (1st and 2nd clusters, Fig. S3, Supporting information). A further cluster comprised 95 native taxa, including those associated with stagnant water conditions, and was defined as the lentic group (3rd cluster; Fig. S3). The density of the lotic group significantly increased along LHC before restoration, while the lentic group significantly decreased along LHC (see Fig. S4, Supporting information). Thirty-one native taxa were unique (i.e. occurred in only one site of the 36). Among the 186 taxa, nine species were identified as alien: seven species were present in more than one site (species marked with an asterisk in Fig. S3), and two species were present in only one site before restoration (Dikerogammarus villosus Sowinski and Orconectes limosus Rafinesque). The density of the alien macroinvertebrates significantly increased with LHC before restoration (Fig. S4). One supplementary alien species arrived after restoration: Hypania invalida Grube.

Functional diversity

Before restoration, FD of the entire macroinvertebrate community peaked in frequently connected sites and was lowest in highly connected and disconnected sites (R2 = 0·48, P < 0·001, y = −211·1x4 + 435·3x3 − 316·3x2+ 96·2x − 8·8, Fig. 2a). FD of the alien group increased with LHC (R2 = 0·46, < 0·0001, y = 2·5x − 0·8, Fig. 2b), but most sites were scarcely invaded by alien species. FD of the lotic group reached a maximum in frequently connected sites (R2 = 0·67, P < 0·0001, y = −142·6x4 + 267·2x3  167·6x2 + 43·2x − 3·5, Fig. 2c). Finally, FD of the lentic group declined with LHC (R2 = 0·47, P < 0·0001, y = −1·6x + 2·0, Fig. 2d). The alien, lotic and lentic groups accounted, respectively, for 2·4%, 25·5% and 72% of the total FD (Table 1).

Table 1. Contribution of macroinvertebrate groups to the total functional diversity (FD) before (PRE) and after (POST) restoration. Wilcoxon signed-rank test statistics (W) and significance of the change in contribution after restoration are given
GroupsContribution to total FD (%)
PREPOSTW P
Alien2·44·400·002
Lotic25·530·55840·46
Lentic72·064·43443< 0·0001
Figure 2.

Pre-restoration response of within-site functional diversity (FD) to lateral hydrological connectivity (LHC). The LHC was scaled between 0 (disconnected sites) and 1 (permanently connected sites). FD was calculated for (a) the entire community of macroinvertebrates, (b) the alien, (c) the lotic and (d) the lentic groups. Each dot represents a site with all seasons combined. 95% confidence intervals are given by dashed lines.

After restoration, similar FD patterns were observed along LHC (see Fig. S5, Supporting information). FD response of the entire community and lotic group to LHC followed a cubic regression with a low FD in highly connected channels (R2 = 0·33, P = 0·006 and R2 = 0·53, P < 0·0001, respectively). The FD of the alien group increased with LHC, whereas the lentic group ones declined (R2 = 0·60, P < 0·0001 and R2 = 0·55, P < 0·0001, respectively). The contribution of alien species to the total FD significantly increased, whereas the contribution of the lentic group decreased (Table 1). Finally, the contribution of the aliens (4·4%) to the total FD remained lower than the contributions of the lentic and the lotic groups (Table 1).

Impacts of restoration works

The total FD was positively related to the rarefied richness per site before restoration (Fig. 3). The y-intercept of the relationship was similar after restoration (ancova, restoration treatment, F = 0·26, d.f. = 1, P = 0·61), and the slope of the relationship did not differ after restoration (ancova, F = 2·12, d.f. = 1, P = 0·15, Fig. 3). The different types of restoration works did not induce a change in total FD, while dredging and reconnection increased significantly the FD of the alien group (Table 2), and reconnection decreased significantly the FD of the lentic group (Table 2). FD measured in control sites did not demonstrate any significant change either for the entire community or for the different groups of taxa (Table 2).

Table 2. Post-restoration changes in functional diversity (FD) for the entire community of macroinvertebrates and different groups identified in the floodplain. Sites are grouped according to their main restoration types. Wilcoxon signed-rank tests statistics (W) and the significance of the FD changes after restoration are given
FDFlow increaseDredgingReconnectionControl
ChangeW P ChangeW P ChangeW P ChangeW P
  1. a

    significant under P < 0·05 after Bonferroni-Holm correction

Total0·2330·95−0·17150·22−0·57310·04−0·04810·26
Alien0·7110·030·7500·015a0·8500·004a0·12130·15
Lotic−0·12130·72−0·12130·72−0·03190·58−0·21980·94
Lentic−0·0680·72−0·43180·08−0·58360·004a−0·02800·28
Figure 3.

Restoration effect on the relationship between species rarefied richness and functional diversity (FD). Only the restored sites are represented in their pre-restoration states (closed symbols and solid line, y = 0·037x + 0·57, R2 = 0·22, P = 0·037) and post-restoration states (open symbols and dashed line, y = 0·044x + 0·28, R2 = 0·51, P < 0·001).

Deviations in FD were significantly and negatively related to the deviations in LHC, all sites combined (R2 = 0·28, P = 0·001, y = −1·4x − 0·1, Fig. 4a). The model showed a great variability in FD response for low and high LHC deviations (Fig. 4a). Based on the pre-restoration model established for the entire community (cf. Fig. 2a), we computed the expected FD after restoration works (Fig. 4b). Post-restoration predictions conformed to the observations because we found no significant differences between them (Wilcoxon signed-rank test, flow increase P = 0·68, dredging P 0·063, reconnection P = 1·0, control P = 0·54, Fig. 4b). Overall, all types of restoration works combined did not induce a significant decrease in FD within-sites (i.e. alpha diversity), whereas it increased the among-sites diversity (i.e. beta diversity, Table 3). This beta diversity rise was statistically significant for the restored sites but not for the control (Table 3).

Table 3. Alpha and beta functional diversity (FD) before (PRE) and after (POST) restoration. The diversities were calculated separately for the restored and the unrestored sites (control). Wilcoxon signed-rank tests statistics (W) and the significance of the pre- and post-restoration difference of FD within (alpha diversity) and among-sites (beta diversity) are given
Sitesα-diversityβ-diversity
PREPOSTW P PREPOSTW P
Restored1·4151·3201370·250·2750·5074201<0·0001
Control1·4321·383540·950·4050·30040390·28
Figure 4.

Assessment of the restoration effects on functional diversity (FD). (a) Pre-post-restoration deviations in FD and lateral hydrological connectivity (LHC). Open symbols stand for restored sites and closed symbols for unrestored sites. Regression line is drawn with 95% confidence interval (dashed lines). (b) Comparisons between the distributions of FD values expected and observed 2 years after restoration. (ns, not significant differences between predictions and observations).

Discussion

How does functional diversity vary with lateral hydrological connectivity?

Our results emphasize that LHC is a key parameter influencing the macroinvertebrate FD in a large floodplain river. On one hand, FD of macroinvertebrate communities followed a hump-shaped response peaking in frequently connected channels and declining rapidly in permanently connected channels. This response is in agreement with a previous study in a Mediterranean river-floodplain where the main river channel encompassed a lower FD than secondary channels (Gallardo et al. 2009). In our study, the rapid decline in total FD highlights the impact of niche filtering on communities when the abiotic constraints are high (i.e. high LHC). This effect in permanently connected channels was also supported for the lentic and the lotic groups of species, underlining the role of LHC as an abiotic filter for macroinvertebrate communities. This pattern is in agreement with previous research using multiple traits at several spatial scales, which showed that abiotic filters acted significantly and independently of the taxonomic richness on the traits of European stream macroinvertebrates, whereas biotic filters had no significant effects (Statzner, Dolédec & Hugueny 2004). On the other hand, the total FD in disconnected channels was stable and higher than in permanently connected channels. By definition, disconnected channels are less frequently disturbed by floods than connected channels, and we expected the competition for resources to be higher in disconnected channels (Ward, Tockner & Schiemer 1999). The high value of FD in the disconnected channels represents a high diversity of biological characteristics and a high dissimilarity among species. This result corroborates the expectation that coexisting species will tend to differ in a competitive environment, because species using the same resources in the same way could lead to competitive exclusion (Szabo & Meszena 2006; Mouchet et al. 2010) or would coexist only if they partition resources in time and/or space (Belyea & Lancaster 1999).

Sites in frequently connected channels (Fig. 1 channel type b.) had higher FD than either completely connected channels or less connected channels (Fig. 2a). The location of this maximum may be explained by the coexistence of taxa with different environmental optima due to (i) a low effect of niche filtering and (ii) a low effect of the competitive exclusion (as illustrated by the midpoint in Fig. 1 panel c). Our results show that at these sites both alien and native macroinvertebrates groups and a group of highly functionally diversified lotic macroinvertebrates coexist. Two years after restoration, the maximum FD in frequently connected channels was less pronounced but was always greater than in disconnected channels. As a result, monitoring changes over time is of prime importance to confirm the persistence of functionally diversified communities in frequently connected channels. Nevertheless, the trends in the FD responses along the gradient of LHC before and after restoration suggest that the assembly rules are relevant for the study of FD in large river floodplains (cf. Fig. 1). As expected, the LHC had a major influence on the FD of macroinvertebrate groups and communities. It corroborates the findings of previous studies that the hydrological connectivity is a key factor explaining the macroinvertebrate diversity in semi-natural and natural large river floodplains (Reckendorfer et al. 2006; Gallardo et al. 2009; Leigh & Sheldon 2009).

How do aliens contribute to the total functional diversity?

Alien species represent a growing problem for the management and restoration of ecosystems (Strayer 2010). In our study, FD of the alien group was maximum in permanently connected sites. This pattern was even stronger after restoration, with more sites invaded and a stronger relationship between FD and LHC. Such finding suggests that the LHC did not impose any abiotic filter to the alien group and in contrast the restoration of the LHC promoted the development of alien populations in restored zones. The increase of human disturbance in freshwater ecosystems is a well-accepted cause for promoting alien invasion (Strayer 2010). In this context, an increase in lateral connectivity by human activities is a source of new disturbance for the biota that will not constrain the development of alien species populations but rather favour them. Indeed, our results show that any type of restoration (dredging or reconnection) can increase the FD among alien species. While the low contribution (<4·5%) of the alien group to the total FD moderates this controversial finding, the significant increase in their contribution to the total FD 2 years after restoration (from 2,4 to 4,4%) calls for caution on the conflicting ecological benefits associated with hydrological connectivity enhancement (see e.g. Jackson & Pringle 2010). Moreover, the contribution of aliens to FD should increase with time because a continued increase in colonization of alien taxa combined with an increase in their abundance is generally expected (Strayer 2010). To date, three among the 100 most invasive alien species in Europe were identified in this study (Corbicula fluminea, Dikerogammarus villosus and Dreissena polymorpha) (DAISIE 2009). Long-term post-restoration monitoring of the Upper Rhône river will help assess whether these invasive species will establish important populations. Such increases are of considerable concern because invasive species are known to have deleterious impacts on ecosystems processes (Chapin et al. 2000) and rising economic costs for society (Pimentel, Zuniga & Morrison 2005).

How does restoration affect functional diversity?

We observed a negative relationship between the pre- and post-restoration deviation of both FD and LHC (cf. Fig. 4a), which corroborates the assumption that LHC imposes an abiotic filter to the FD of macroinvertebrates. However, for the same change in LHC (x-axis of Fig. 4a), FD showed a high variability that can be explained by the hump-shaped relationship between FD and LHC before restoration. Indeed in our study, for the same level of increase in LHC, FD showed either an increase or a decrease depending on the initial level of hydrological connectivity (cf. Fig. 2a). The antagonistic changes in FD if the competitive exclusion is decreased (i.e. FD is expected to increase), and the niche filtering increased after restoration (i.e. FD is expected to decline) may also explain the observed variability (Mayfield et al. 2010). Therefore, the antagonistic trajectories for small changes in LHC after restoration may explain the observed FD variability for all sites combined or the absence of FD changes when considering the restoration types. However, restoration types had different influences on native and alien groups. Dredging and reconnection increased FD among alien species, while reconnection decreased FD among lentic species. In this context, the reduction in FD for the lentic group after restoration corroborates that niche filtering is exacerbated by an increase in LHC. Conversely, the increase of FD among alien species following an increase of LHC after restoration may underline a reduction in competitive exclusion for those species rather than an increase of niche filtering.

In addition, we observed that FD and species richness were significantly and positively linked both before and after restoration. However, these relationships were not significantly different before and after restoration suggesting that restoration did not promote homogenization. The relationship between FD and richness observed after restoration yielded a higher explained deviance, which could be due to an increase in the abiotic filter (i.e. LHC). Moreover, the absence of homogenization provided by restoration is also sustained by the good concordance between the predictions of FD and their observations made post-restoration (cf. Fig. 4b). We considered the values of FD post-restoration as not inflated or reduced by restoration but rather underlining the fast response and recovery of the macroinvertebrate community to restoration operations, possibly due to the increase of factors permitting dispersal of individuals in the system (Mackay 1992).

A successful aspect of restoration concerns the increase of FD among the channels (i.e. beta diversity). Higher beta diversity implies an increase in the ecological distance among-sites (Amoros & Bornette 2002; Pavoine, Dufour & Chessel 2004). Thereby, a diversification of FD among channels may compensate the local decrease of FD within the channels and avoid a strong loss of FD at the scale of the floodplain. The absence of short-term homogenization of FD within and among-sites after only 2 years of monitoring provides an overall positive assessment of the restoration. However, post-restoration monitoring should last over a decade to assess the durability of a diversification of the FD within the floodplain and to quantify the impact of new alien arrivals (Lake, Bond & Reich 2007).

Restoration of the LHC in many large rivers, including the Rhône, is considered as a necessity to rejuvenate the floodplain, mitigate habitat loss and enhance resilience capacities. It is broadly viewed that lack of intervention leads to increased terrestrialization and a progressive disappearance of aquatic ecosystems in floodplains affecting in turn aquatic biodiversity. In this context, we showed that homogenization of the LHC among the channels may have deleterious effects on FD and thus potentially on ecosystem functioning. Therefore, we encourage environmental managers to promote a high diversity of LHC among channels to avoid biotic homogenization and ensure a high beta FD. A higher FD at the scale of the floodplain is a desirable aim for restoration strategies as it potentially ensures a better ecosystem functioning (Cadotte, Carscadden & Mirotchnick 2011).

Acknowledgements

We thank the Swiss National Science Foundation (project PBGEP3 – 136309), the RMC Water Agency, the DIREN, the CNR for financial support; J.M. Olivier and N. Lamouroux for the coordination of the scientific monitoring of the Rhône restoration project; the Editor, the Associate Editor, two anonymous referees and P. zu Ermgassen for their constructive comments; D. McCrae, N. Peru, A.L. Besacier-Monbertrand, O. Béguin and all the persons involved in collecting the data.

Ancillary