Project Topics
Get Complete Project Material File(s) Now! »
Building functional groups of marine benthic macroinvertebrates on the basis of general community assembly mechanisms
The accurate reproduction of the spatial and temporal dynamics of marine benthic biodiversity requires the development of mechanistic models, based on the processes that shape macroinvertebrate communities. The modelled entities should, accordingly, be able to adequately represent the many functional roles that are performed by benthic organisms. With this goal in mind, we applied the emergent group hypothesis (EGH), which assumes functional equivalence within and functional divergence between groups of species. The first step of the grouping involved the selection of 14 biological traits that describe the role of benthic macroinvertebrates in 7 important community assembly mechanisms. A matrix of trait values for the 240 species that occurred in the Rance estuary (Brittany, France) in 1995 formed the basis for a hierarchical classification that generated 20 functional groups, each with its own trait values. The functional groups were first evaluated based on their ability to represent observed patterns of biodiversity. The two main assumptions of the EGH were then tested, by assessing the preservation of niche attributes among the groups and the neutrality of functional differences within them. The generally positive results give us confidence in the ability of the grouping to recreate functional diversity in the Rance estuary. A first look at the emergent groups provides insights into the potential role of community assembly mechanisms in shaping biodiversity patterns. Our next steps include the derivation of general rules of interaction and their incorporation, along with the functional groups, into mechanistic models of benthic biodiversity.
Biological communities (i.e. sets of co-occurring species) are at the heart of some of the most challenging issues currently raised in the field of ecology. These issues include the degree to which communities are shaped by stochastic versus deterministic processes, the potential for species traits to predict the structure and dynamics of communities and the role of environmental variability in space and time (Sutherland et al., 2013). The elucidation of the mechanisms of community assembly would not only enhance our fundamental understanding of ecological processes. It is also expected to increase our ability to conserve biodiversity and ecosystem function.
Function here refers to the second of the meanings assigned to the term by Jax (2005). It is associated with questions, such as “how is the whole sustained” or “what do specific parts contribute to this”. Answering these questions is important, because we value the services provided by a functioning whole. Yet, in view of the current rate of environmental change and its potential impacts on biodiversity (Bellard et al., 2012), we cannot reliably answer them without first addressing the questions that Jax (2005) linked to the functioning of the specific parts, such as “which processes occur” or “how do organisms interact with each other and with their environment”.
The effort to answer these questions in the marine benthos has been dominated by statistical methods of multivariate analysis (Clarke, 1993; Legendre and Gauthier, 2014). These methods rely on data from temporal and/or spatial sampling schemes, aimed at capturing the species abundance patterns of a system’s macro-, meio- or microbenthic compartment. They often use correlations between environmental variables and community composition with the goal of explaining variations in the latter (ter Braak and Prentice, 2004). With the addition of tools for the analysis of spatial and temporal patterns (Dray et al., 2006; Blanchet et al., 2008) multivariate analysis has become a very efficient exploratory technique. However, its correlative nature, along with its difficulty to account for key ecological phenomena, has restricted its ability to reveal the role of community shaping processes (James and McCulloch, 1990).
In response to the limitations of statistical modelling, efforts have been made to adopt a more mechanistic approach, mostly in the form of dynamic food web models (Yodzis and Innes, 1992) and static trophic network analyses (Ulanowicz, 2004). The amounts of data and knowledge that are typically required by such approaches, along with issues of model complexity and tractability, have set a limit to the number of modelled entities. In spite of efforts to address these issues through the application of tools, such as Ecopath with Ecosim (Ortiz and Wolff, 2002) or the inverse method (Garcia et al., 2011), mechanistic models tend to lack the level of detail that is needed to account for the functioning of benthic communities. The host of biotic interactions that are responsible for shaping these systems is hardly limited to what can be represented by a food web (Menge, 1995). In spite of recent attempts to integrate non-trophic interactions into food web models (Kéfi et al., 2012), the majority of community assembly mechanisms are seldom included in models of marine benthos.
Trait-based approaches have been suggested as an alternative to food web models (Ings et al., 2009). Biological traits have been increasingly employed in the analysis of the functional composition of benthic communities (Bremner, 2008). The emergence of the concept of functional diversity has raised questions, such as “what types of traits”, “which traits” or “how many traits” should be considered. Petchey et al. (2006) argue that the answers depend on the scope of each study, emphasizing the potential for functional classifications of organisms to be nested and the need to treat each classification as a testable hypothesis. Bremner et al. (2006b) suggest including as many traits as possible in biological traits analyses, with recent studies following suit (e.g. Darr et al., 2014; Jimenez et al., 2016). Trait-based modelling approaches have, on the other hand, focused on the most studied processes in the marine benthos: feeding behaviour and substrate modification (Pearson, 2001). The representation of these mechanisms offers valuable information on the contribution of existing communities to the functioning of the system, but it provides very little insight into future trajectories following natural or anthropogenic environmental change.
A variety of ecological theories pertaining to environmental filtering, trophic interactions, resource partitioning, life history trade-offs and response to disturbance have been successfully employed to explain observations of benthic communities. They could be used to generate reliable predictions of benthic biodiversity, if they took the form of mathematical formulations linking a system’s primary functional components. The latter should be generated through a systematic and testable procedure and possess a clear role in various community assembly mechanisms. The framework developed by Boulangeat et al., (2012) for communities of terrestrial vegetation is particularly well-suited for this purpose. It employs the emergent group hypothesis (EGH), which assumes functional equivalence within (neutrality) and functional divergence between (niche differentiation) groups of species (Hérault, 2007). Its application is based on a matrix of species traits that represent their role in important community assembly mechanisms. Group emergence results from correlations among the traits, which are indicative of adaptive responses and evolutionary constraints (Lavorel et al., 1997).
The aggregation of ecosystems through the construction of functional groups is based on the concept of functional redundancy, which is central to theories relating biodiversity variations to ecosystem function (Rosenfeld, 2002). Although the exact nature of this relationship has been subject to debate (Grime, 1997), its existence is beyond dispute (Srivastava and Vellend, 2005). This is why the level of functional redundancy with regard to the assembly of communities, the engines of biodiversity, is particularly important for the conservation of ecosystem function. This level can be demonstrated as the acceptable level of ecological aggregation, i.e. the minimum number of groups that can adequately represent community function. It appears to vary in predictable ways (Hairston and Hairston, 1993), but its accurate assessment requires a good understanding of assembly mechanisms (Walker, 1992).
In spite of recent advances in the quantification of functional redundancy (Muntadas et al., 2016; van der Linden et al., 2016), its assessment remains highly prone to subjectivity, especially with regard to the number of biological traits (Jax, 2005). The framework of Boulangeat et al. (2012) addresses this issue, by defining a specific number of important community assembly mechanisms that need to be explicitly represented. Trait categorization is generally lacking among functional studies of benthic communities. Even when traits are explicitly assigned to a set of general functions (e.g. Törnroos and Bonsdorff, 2012), this is done in order to rather interpret the results of the study than guide the process of biological traits selection. The framework also allows the nesting of finer functional differences within broader ones. This is achieved through the separation of organisms into broad groups with a common resource base, whose consumption is further differentiated based on finer group dissimilarities. Finally, putting the emergent grouping to the test is central to the framework and allows defining the acceptable level of ecological aggregation as the minimum number of groups for which the assumptions of the EGH are supported by observations. Boulangeat et al. (2012) tested the niche constituent of the EGH, by comparing its assumptions with what could be observed in their system. Here, we take their approach one step forward, by investigating the second constituent of the EGH, concerning the neutral behaviour of species within each functional group.
In this study, we revisit a benthic macroinvertebrates abundance data set from the Rance estuary (Brittany, France), previously explored with the use of traditional multivariate analyses (Desroy, 1998). We combine it with a matrix of biological traits, with the goal of aggregating the system through the construction of functional groups. We investigate both niche and neutral attributes of the emergent grouping, gaining insights into the components of functional diversity and redundancy in benthic communities. In doing so, we integrate statistical tools and ecological mechanisms into a quantitative approach toward defining the acceptable level of ecological aggregation. The present study is a first step toward the development of models of benthic community assembly mechanisms, with the generated functional groups as their entities. The application of this generic modelling approach to the Rance estuary is expected to describe the stability characteristics of macroinvertebrate communities as well as their responses to well-documented perturbations, such as the occurrence of particularly cold winters or the introduction of invasive species (Desroy, 1998).
Study site
The framework for the construction of functional groups was applied to the Rance estuary (Brittany, France), in the southern part of the English Channel (Fig. II.1). The site is characterized by the presence of a tidal power plant at its mouth, comprising a lock, the generating station proper, a rock dike and a 115 m wide removable dam made up of 6 sluice gates. The system was fundamentally altered during the construction of the plant (1963-1966), after which it was allowed to gradually return to a more natural state (Kirby and Retière, 2009).
Fig. II.1 Map of the study site. The Rance estuary is situated on the northern coast of Brittany, France. Crosses indicate the location of the 113 stations that were sampled in the spring of 1995. The tidal power plant is located at the mouth of the estuary, south of the city of St-Malo The operating constraints of the installation impose highly specific “tidal” conditions on the estuary: (1) mean water level is elevated by approximately 2.5 m, (2) slack water periods are particularly long (up to 5 h), (3) emersion time may be half that of the open sea and (4) the tidal range varies between 4.0 m and 5.5 m compared to 9.5 m (mean value) in the open sea, depending on which direction the turbines are operating (Retière, 1994). Reduction in tidal range is correlated with a reduction in the surface area of the intertidal zone; the exposed zone accounts today for 50% of the total surface of the Rance estuary, compared to 70% before the construction of the plant. Maximum water depth is 17 m at low tide, but the main part of the basin is 5-6 m deep. Two areas of differing salinities can be identified: the marine reservoir, in which deep-water salinity remains higher than 30, and the upstream estuary of brackish water (Retière, 1994). The junction between brackish and marine waters has moved about 5 km upstream since the scheme was built.
The strong sluice and turbine currents have eroded parts of the riverbed. Sandbanks closest to the dam have shifted and the bed is more or less covered with gravel or pebbles (Retière, 1994). Meanwhile, long periods of slack water have promoted the deposition of fine particles in coves and bays (Bonnot-Courtois and Lafond, 1991). From downstream to upstream of the estuary, pebbles and coarse sands are replaced by medium and fine sands, muddy sands and finally muds, beyond Port-St-Hubert. A similar sequence is observable from the central channel to the banks. Natural silting is presumed to have increased since operation of the tidal power plant started. In the upstream part of the estuary, sedimentation rate increased from 0.5 cm y-1 before the scheme to 2.7 cm y-1 after (Bonnot-Courtois, Ecole Pratique des Hautes Etudes, Dinard, France, personal communication).
Sampling methods
A grid of 113 stations was sampled in April 1995, prior to the spring recruitment (Fig. II.1). Two replicate samples were collected at each of 103 submerged stations using a 0.1 m2 Smith Mac-Intyre grab, while 10 emerged stations were sampled using a hand corer (5 replicates; replicate area of 1/55 m2) to a depth of 20 cm. The number of replicates is assumed to be sufficient to characterize the assemblage of species that can be found at each station. Although densities of organisms were extrapolated to a standard surface area, some bias was unavoidably introduced, due to the different characteristics of the sampling gears. All samples were gently washed in situ through a 1 mm sieve and preserved in 4.5% formalin before being sorted, identified and counted in the laboratory. Macroinvertebrates retained on the mesh were determined at species level when possible. A total of 240 species or higher taxonomic groups belonging to 9 phyla were thus identified.
Design and application
Our approach draws on the work of Boulangeat et al. (2012), who employed the EGH for the classification of terrestrial plant species into groups with similar ecological strategies. Much like their approach, our own is divided into five steps (Fig. II.2), with the respective ecological assumptions and methodological framework presented below.
Step 1: selecting biological traits
The objective of this step was to select species characteristics that describe the role of the average individual of each species in the most important community assembly mechanisms. The list of mechanisms was mostly adopted from the framework of Boulangeat et al. (2012), with a few adjustments, in order to adapt it to the special attributes of estuarine benthic systems. The choice of the traits was made based on both the nature of the community assembly mechanisms and the quality of the data that could be found for each of the traits.
Since community assembly mechanisms include competition for a limited amount of resources, we first identified food and space as the basic resources for which benthic organisms compete. Space was assumed to be two-dimensional, while food was defined with the goal of dividing species into groups with a common resource base. The wide-spread adoption of facultative feeding modes only allowed for a distinction between species that feed on algae and detritus on the one hand and those characterized as predators and scavengers on the other.
Fig. II.2 Schematic representation of the 5 steps that comprise the methodology of functional grouping. In step 1, 14 biological traits were selected, representing 7 community assembly mechanisms. For details, see Table II.1. In step 2, a matrix of species trait values formed the basis for the classification of species into functional groups (solid line). The two matrices, combined with data of species abundance, allowed the assignment of trait values to the functional groups (dashed lines). In step 3, taxonomic diversity and functional divergence were measured for each station at the level of species and functional groups. Measures at the two levels were then compared, in order to evaluate the representation of biodiversity by the functional groups. In step 4, community weighted mean trait values were calculated for each station at the level of species and functional groups. Calculations at the two levels were then compared, in order to assess the preservation of niche attributes by the functional groups. In step 5, the independence between species abundance at each station and their trait values was tested within each functional group and the rejection proportion for every trait was used as an indication of departures from neutrality.
The rest of the biological traits represent seven community assembly mechanisms. The initial goal was for each mechanism to be represented by two traits, so that one set of traits could be used for the species classification and the other for the cross-validation of the resulting grouping. The lack of redundancy in the content of the two sets of traits obliged us to abandon this goal and use all fourteen traits for the classification of the species into groups. The seven community assembly mechanisms are: (1) resistance to perturbation, (2) dispersal potential, (3) environmental filtering, (4) competitive effect, (5) response to competition, (6) population dynamics and (7) biogenic habitat modification. Details about the selected biological traits and the assignment of trait values to the system’s species can be found in Table II.1.
The vast majority of the information that was required for the assignment of trait values to the species of the system was provided by the following online databases: eol.org, genustraithandbook.org.uk, marinespecies.org, marlin.ac.uk and species-identification.org. The remainder was acquired from consultation with experts on the field. Very often the lack of appropriate information for a particular species obliged us to look for data at higher taxonomic levels. The quality of the available information for the ensemble of species and biological traits dictated the resolution of the values that were assigned to them (for details, see Table B.1 in Annex B).
Step 2: building functional groups
This step aims at reducing a community of benthic macroinvertebrates to its principal functional components, by identifying emergent groups of species (Hérault, 2007). It was applied separately for consumers of algae/detritus and predators/scavengers, because the concept of functional equivalence, which is central to the EGH, is defined for trophically similar sympatric species (Hubbell, 2005). The first task involved calculating a distance matrix for both groups of species, based on the rest of the biological traits. Since our list included continuous, ordinal, nominal and binary traits, we opted for the Gower distance (Gower, 1971). These matrices formed the basis for the application of an agglomerative hierarchical clustering technique, the unweighted pair group method with arithmetic mean (Sokal and Michener, 1958). The two generated dendrograms were consecutively pruned at 0.4 and 0.3 distance levels, without, for practical reasons, allowing the formation of groups with only one species.
In order to be able to treat the newly formed groups as independent functional components, we needed to attribute trait values to them. We did that by employing the mass ratio hypothesis (Grime, 1998), which predicts that the functional identity of a group of species is determined by the trait values of the dominant abundance contributors. We measured the abundance contribution of each species in its group, by calculating its median abundance at the stations where it was present. For the ordinal, nominal and binary traits, a group’s trait value was defined as the dominant value, as far as the abundance contribution of its species was concerned. For the continuous traits, a group’s trait value was defined as the mean trait value of all the species in the group, weighted by their abundance contribution. Each group was, finally, assigned a representative species, which was the one with the highest abundance contribution in the group. In case of ties or close calls, the species with the highest body mass was chosen to represent the group.
Step 3: evaluating biodiversity representation
Once functional groups were built, we had to assess their efficiency at representing natural biodiversity patterns. This need stems from the loss of information that is inherent to the process of classifying a number of species into a much smaller number of groups. We in fact wanted to know if this loss of information lay within acceptable limits, or if, instead, it severely impaired the ecological pertinence of the imposed grouping. If we assume that information at the species level provides an adequate representation of biodiversity, we could reach our goal by comparing biodiversity measurements at this level with the same measurements made at the level of functional groups. Since the role of the groups as functional components of the system was what we were especially interested in, we did not want to be limited to measures of taxonomic diversity, but we wanted to include measures of functional diversity as well.
One framework that offers this possibility is Rao’s quadratic entropy, Q (Botta-Dukát, 2005). For an assemblage of T taxa characterized by the relative abundance vector p = (p1, p2, …, pT), it is defined as where dij is the functional distance between the i-th and j-th taxa (dij = dji and dii = 0). Assuming functional equidistance among taxa (dij = 1), it equals the complement of the Simpson dominance index, thus expressing the probability that two individuals taken at random from an assemblage belong to different taxa. When combined with a functional distance matrix for the taxa in question, like the one previously calculated with the help of the Gower distance, the same index becomes a measure of functional divergence. In this case, Rao’s quadratic entropy expresses the average functional distance between two randomly selected individuals of an assemblage. We calculated both versions of the index for all 113 assemblages. If the species-level measures of diversity showed a high correlation with the same measures calculated at the level of functional groups, we could say that the transition from the former level to the latter entailed an acceptable amount of information loss.
Table of contents :
Chapter I: Introduction
I.1 Quantitative models of benthic biodiversity
I.1.1 Overview of modelling approaches
I.1.2 Modelling biotic interactions
I.1.3 A general framework for models of biodiversity
I.2 Modelling biodiversity in the Rance estuary
I.2.1 Objectives of the thesis
I.2.2 General characteristics of the study site
I.2.3 Development of general modelling framework
I.2.4 Work plan
I.3 References
Chapter II: Building functional groups of marine benthic macroinvertebrates on the basis of general community assembly mechanisms
II.1 Introduction
II.2 Methods
II.2.1 Study site
II.2.2 Sampling methods
II.2.3 Design and application
II.3 Results
II.3.1 Building functional groups
II.3.2 Evaluating biodiversity representation
II.3.3 Assessing niche attributes preservation
II.3.4 Detecting neutral behaviour within groups
II.4 Discussion
II.4.1 Selecting biological traits
II.4.2 Building functional groups
II.4.3 Evaluating biodiversity representation
II.4.4 Assessing niche attributes preservation
II.4.5 Detecting neutral behaviour within groups
II.5 Conclusions
II.6 References
II.7 Appendix
Chapter III: Moving from biological traits to functional relationships for use in dynamic models of marine benthic communities
III.1 Introduction
III.2 Methods
III.2.1 Study site
III.2.2 Data collection
III.2.3 Biological traits
III.2.4 Environmental filtering
III.2.5 Life history trade-offs
III.2.6 Signed digraphs
III.2.7 Stability analysis
III.3 Results
III.3.1 Environmental filtering
III.3.2 Life history trade-offs
III.3.3 Signed digraphs
III.3.4 Stability analysis
III.4 Discussion
III.4.1 Environmental filtering
III.4.2 Life history trade-offs
III.4.3 Signed digraphs
III.4.4 Stability analysis
III.5 Conclusions
III.6 References
Chapter IV: Agent-based modelling of the multi-scale dynamics of marine benthic communities
IV.1 Introduction
IV.2 Methods
IV.2.1 Small-scale model
IV.2.1.1 Purpose
IV.2.1.2 Entities, state variables and scales
IV.2.1.3 Process overview and scheduling
IV.2.1.4 Design concepts
IV.2.1.5 Initialization
IV.2.1.6 Input data
IV.2.1.7 Submodels
IV.2.2 Large-scale model
IV.2.3 Parameterization
IV.2.3.1 Settlement probability
IV.2.3.2 Post-settlement mortality
IV.2.3.3 Predation
IV.2.4 Model analysis
IV.2.4.1 Spatial resolution
IV.2.4.2 Sensitivity analysis
IV.2.4.3 Group accumulation
IV.2.4.4 Spatial correlation
IV.2.4.5 Correspondence analysis
IV.2.5 Software
IV.3 Results
IV.3.1 Spatial resolution
IV.3.2 Sensitivity analysis
IV.3.3 Group accumulation
IV.3.4 Spatial correlation
IV.3.5 Correspondence analysis
IV.4 Discussion
IV.4.1 Spatial resolution
IV.4.2 Sensitivity analysis
IV.4.3 Group accumulation
IV.4.4 Spatial correlation
IV.4.5 Correspondence analysis
IV.4.6 Limitations and potential improvements
IV.5 References
IV.6 Appendix
Chapter V: Conclusions
V.1 Study site
V.2 Empirical research
V.2.1 Single scale
V.2.2 Multiple scales
V.3 Modelling macrobenthos
V.3.1 Functional groups
V.3.2 Functional relationships
V.3.3 Qualitative models
V.3.4 Agent-based models
V.3.4.1 Scales of mechanisms
V.3.4.2 Individual-based models
V.3.4.3 Inter-scale modelling
V.3.4.4 Model validation
V.3.4.5 Data sets
V.3.4.6 Minor adjustments
V.3.4.7 Major changes
V.3.4.8 Model generalisation
V.4 References
