A rapid biodiversity assessment methodology tested on intertidal rocky shores
Post on 06-Jun-2016
AQUATIC CONSERVATION: MARINE AND FRESHWATER ECOSYSTEMS
Aquatic Conserv: Mar. Freshw. Ecosyst. 20: 452463 (2010)
Published online 30 March 2010 in Wiley InterScience(www.interscience.wiley.com). DOI: 10.1002/aqc.1111
A rapid biodiversity assessment methodology tested on intertidalrocky shores
TIMOTHY D. OHARAa,, PRUE F. E. ADDISONa, RUTH GAZZARDa, TRUDY L. COSTAb
and JACQUELINE B. POCKLINGTONb
aMuseum Victoria, GPO Box 666, Melbourne 3001, AustraliabZoology Department, University of Melbourne, Melbourne 3010, Australia
1. Conservation managers require biodiversity assessment tools to estimate the impact of human activities onbiodiversity and to prioritize resources for habitat protection or restoration. Large-scale programs have beendeveloped for freshwater ecosystems which grade sites by comparing measured versus expected species richness.These models have been applied successfully to habitats that suffer from systemic pressures, such as poor waterquality. However, pressures in other habitats, such as rocky intertidal shores, are known to induce more subtlechanges in community composition.2. This paper tests a biodiversity assessment methodology that uses the ANOSIM R statistic to quantify the
biological dissimilarity between a site being assessed and a series of reference sites selected on the basis of theirsimilar environmental prole. Sites with high R values for assemblage composition have an anomalousassemblage for their environmental prole and are potentially disturbed.3. This methodology successfully identied moderate to heavily perturbed sites in a pilot study on 65 rocky
intertidal sites in south-eastern Australia. In general, measures based on percentage cover (ora and sessileinvertebrates) were more sensitive than abundance (fauna). Copyright r 2010 John Wiley & Sons, Ltd.
Received 19 August 2009; Revised 23 December 2009; Accepted 27 January 2010
KEY WORDS: biodiversity assessment; ecosystem health; intertidal; invertebrates; algae; dissimilarity; multivariate; ANOSIM
Coastal marine habitats are affected by various humanactivities including habitat destruction and changedhydrology (e.g. for sea walls, boat ramps, marinas, etc.);water pollution (e.g. nutrients, toxins, salinity, temperature);
trampling or displacement by visitors; direct harvesting oforganisms for food, bait, aquaria or curiosity; introducedspecies; and climate change (Suchanek, 1994; OHara, 2002).
In response, marine managers frequently request standardizedbiodiversity assessment indices of site condition for state ofenvironment reporting and to guide the allocation of scarce
management resources (Bailey et al., 2004). This information isoften required at large spatial scales on a limited budget.Typically managers want to rank site condition across the
entire jurisdiction of their agency.For example, the Natural Heritage Trust of the Australian
Government commissioned Museum Victoria to develop and
test a biodiversity assessment methodology to rank thecondition of at least 60 rocky intertidal sites across 1100 km
of coastline in the State of Victoria in temperate south-easternAustralia. This posed a number of methodological challengesthat would be typical of meso-scale projects in coastal
environments. First, there were no consistent quantitativemeasurements of environmental stressors across the study area.Monitoring of pollutants or nutrient enrichment was spatiallyconcentrated around a few sewerage outfalls and popular
beaches (OHara, unpubl. data). Measurement of recreationaluse was sporadic at best (Addison et al., 2008). Second, thebiological assemblages were potentially inuenced by the range
of strong environmental (e.g. temperature, primary production,wave exposure, geology) and biogeographical gradients thatwere known to occur across the study area (OHara, 2000;
OHara and Poore, 2000; Underwood and Chapman, 2007).Third, the intertidal assemblages were likely to be characterizedby high spatial and temporal variability in species abundance,
*Correspondence to: Timothy D. OHara, Museum Victoria, GPO Box 666, Melbourne 3001, Australia. E-mail: email@example.com
Copyright r 2010 John Wiley & Sons, Ltd.
particularly at small (o1m) and regional (110 km) scales(Archambault and Bourget, 1996; Underwood and Chapman,1998; Fraschetti et al., 2005).
Consequently, without direct measurement of stressors, sitecondition had to be assessed solely from the composition and/or abundance of the fauna and ora, assuming that a change
will occur in the ecosystem in response to exposure to stressors(Bailey et al., 2004; Pinedo et al., 2007). However, the variationinduced by human stressors had to be distinguished from the
high environmental variability. The lack of stressormeasurements precluded partitioning environmental andimpact factors across a multivariate regression frameworksuch as distance-based linear modelling (McArdle and
Anderson, 2001). Moreover, pilot data suggested that asingle comprehensive multivariate analysis of all sites wouldbe unlikely to derive identiable groups of sites with an
affected assemblage because of the presence of many strongenvironmental gradients across the large study area.
An alternative potential methodology was the reference
condition approach (RCA) developed for freshwater systemsthat uses sites as the basic sampling unit, factors outenvironmental variation through the careful selection of
reference (control) sites, and measures the difference inbiological assemblage composition between test and referencesites as the basis for a preliminary assessment of site condition(Bailey et al., 2004). The RCA forms the basis for large-scale
programs to monitor freshwater systems in various countries,including RIVPACS in the United Kingdom (Wright, 1995;Clarke et al., 2003), AUSRIVAS in Australia (Simpson and
Norris, 2000) and BEAST in Canada (Reynoldson et al., 1995).However, there were potential problems in applying these
freshwater methodologies to coastal systems. Existing RCA
implementations develop an expected taxon list for a test sitebased on taxa found at the matching reference sites andcompute an observed over expected taxa (O/E) ratio for the
test site, which is used as an index of site condition. Not onlydoes this probabilistic presence/absence approach requirespecies-rich systems (Marchant et al., 1997), it also assumesthat environmental degradation will cause a loss of taxa at a
site (or at least the loss of the ability to detect them). However,on rocky shores, impacts such as trampling or harvesting areknown experimentally to induce subtle species-specic
responses in abundance rather than result in local extinction(Keough and Quinn, 1998, 2000). Models based on abundanceare likely to be more sensitive than models based on
presenceabsence (Hewitt et al., 2005). Moreover, atmoderately polluted sites, nutrient enrichment can cause anincrease in species richness (Pearson and Rosenberg, 1978;Bishop et al., 2002). These difculties can be avoided by using
a multivariate measure of the difference in assemblagesbetween a test site and a selected group of reference sites.One solution is to use the residuals from a PCA analysis or
multivariate regression as the basis of an index of disturbance(e.g. the CDI method, see Flaten et al., 2007). However,residuals reect model error in addition to natural variability
and human disturbance. PCA in particular is rarelyappropriate for the non-linearities and zero-inated datacommon in ecological studies (Clarke and Warwick, 2001).
Consequently it was decided to test a novel methodologythat used the non-parametric ANOSIM R statistic (Clarke,1993) to quantify the overall difference in assemblagecomposition between a test site and a series of reference
sites, which were quantitatively selected on the basis of theirsimilar environmental prole. ANOSIM R is based onsimilarity coefcients which can utilize abundance or
presence/absence data and make no assumption about thedirection of the assemblage change. Although ANOSIM R wasdeveloped as a test statistic to calculate the signicance level of
multivariate differences between groups (Clarke, 1993), it is auseful comparative measure in its own right because it isderived from ranked rather than absolute similarity measures
and then scaled to lie between 1 and 1 (Clarke and Warwick,2001; Anderson et al., 2008). For example, comparativeANOSIM R measures are used to optimally split groups ofsamples in the linkage-tree procedure (Clarke et al., 2008).
Here we systematically calculate ANOSIM R values betweeneach site and its group of selected reference sites with a similarenvironmental prole, using these values to rank sites with
anomalous assemblages for further investigation.This biodiversity assessment methodology, hereafter called
MAVRIC (monitoring and assessment of Victorias rocky
intertidal coastline), was tested on both macro-invertebratefaunal and oral assemblages from 65 rocky intertidal sites insouth-eastern Australia, one of the largest intertidal surveys
conducted in the southern hemisphere. Sites ranged from beingputatively affected by sewage pollution (adjacent to outfalls) ortrampling (popular recreational sites) to relatively inaccessiblesites with no identied site-specic human impacts.
The MAVRIC biodiversity assessment methodology
The MAVRIC biodiversity assessment procedure has two basicsteps: (1) selection of a set of reference sites for comparisonwith a test site on the basis of their shared environmental
characteristics (termed nearest-neighbour sites after Linkeet al., 2004); and (2) calculation of ANOSIM R statistics forboth biological and environmental data, using the test site as
one group and the selected nearest-neighbour sites as thesecond group. These R values are a comparative measure of themultivariate assemblage difference between the test site and thenearest-neighbour reference set. High biological R values
indicate that the assemblage is clearly distinct from those atthe nearest-neighbour sites. The calculation of environmental Rvalues validates the reference site selection. High environmental
R indicates that the site was not well matched with the availablereference sites (see below). This can be visualized as a series ofordinations (Figure 1). A set of reference sites nearest to the
test site are selected from multi-dimensional environmentalspace (Figure 1(a)). The test site is then compared with this setof nearest-neighbour sites both environmentally (Figure 1(b))
and biologically (Figure 1(c)).ANOSIM R values are calculated by subtracting the mean
ranked similarity between pairs of the nearest-neighbour sitesfrom the mean ranked similarity between the test site and each
nearest-neighbour site, and the result then scaled to lie between1 and 1 (Clarke and Green, 1988). Negative values of Rindicate that the test site cannot be distinguished from the
nearest-neighbour sites. Positive values of R indicate thatthe test site lies outside the range of variation encountered inthe set of nearest-neighbour sites. It is important to note that,
although the magnitude of the positive R values gives an
TESTING A RAPID BIODIVERSITY ASSESSMENT METHODOLOGY 453
Copyright r 2010 John Wiley & Sons, Ltd. Aquatic Conserv: Mar. Freshw. Ecosyst. 20: 452463 (2010)
indication of how distinct the test site is from the nearest-neighbour group, negative values cannot be interpreted in thesame way, as these latter results are inuenced by
compositional heterogeneity (clumpiness) within thenearest-neighbour group itself rather than the relationshipbetween the test and nearest-neighbour sites.
It is more informative to consider the test/nearest-neighbour site relationship as falling into four groups basedon whether the R statistic is positive or negative for
environmental and assemblage data. Negative R values forboth indicate the test site has a suite of environmental factorsand biological composition within the variation found in thegroup of nearest-neighbour sites. Positive R values for
environmental data indicate that although the best-matchednearest-neighbour sites were chosen for comparison, the testsite still has a distinct environmental signature. The test site
may be unique, or inadequate reference sites surveyed. Thesetest sites can either have a positive or negative assemblage Rdepending on whether the unusual environmental conditions
are reected in the biological assemblage. Sites with positiveassemblage R and negative environmental R have an unusualassemblage despite their typical environmental conditions.
These sites, particularly those with relatively high R (e.g.40.5), are the most interesting from a managementperspective and potentially disturbed by human activities.
Selection of reference sites
The accurate matching of test to nearest-neighbour sites isdependent on the selection of environmental predictors and the
number of nearest-neighbour sites chosen. The BIOENV
(Clarke and Ainsworth, 1993) or BEST (Clarke et al., 2008)procedures can be used to nd the combination ofenvironmental variables that best match the assemblage
pattern (using the same non-parametric multivariateframework as ANOSIM) by correlating ranked similaritymatrices generated from both sets of data for all reference sites
using the Spearman coefcient. The use of ranked similaritycoefcients here avoids the requirement in some othermethodologies (RIVPACS and AUSRIVAS) to divide
reference sites into discrete ecological groups, when in realitythey form a continuum along various environmental gradients(Bailey et al., 2004; Dauvin, 2007). It also avoids the need touse ordinations (e.g. the ANNA model, Linke et al., 2004) or
cluster diagrams to match environmental predictors with thefaunal pattern which impose an articial dimensionality on thedata (Clarke and Warwick, 2001).
Determining the number of nearest-neighbour sites to usein the comparison with a test site is more problematic. Smallnumbers of nearest-neighbour sites are likely to generate
variable results reecting the lack of site replication. On theother hand, increasing the number of sites will increase theenvironmental heterogeneity within the nearest-neighbour
group which in turn will alter the likelihood that a test sitewill be considered distinct. This can be investigated empiricallyby treating each reference site in turn as the test site andcalculating the Environmental R for increasing numbers of
nearest-neighbour sites. The optimal solution has minimumstandard deviation (variability) and mean (distinctiveness) ofenvironmental R across all these relatively undisturbed sites.
A similar procedure was used to test the effect of varyingdata transformations of the biological abundance data(binary, fourth-root, log(x11), square root, and none) and
combinations of environmental variables. Finally, tests weremade to determine whether models based on selecting nearest-neighbour sites using the best match of environmental
variables performed better (i.e. lower mean an...