Relating sets of variables in environmental studies: The sediment quality triad as a paradigm

Abstract

Many environmental studies generate multivariate data, often consisting of two or more conceptually different sets of variables. For example one may wish to relate species abundances to environmental variables. A recent concept gaining popularity in marine benthic impact studies is the "sediment quality triad", which relates benthic community composition as one set of variables to sediment chemistry (contaminents) as a second set, and to toxicities of sediment to test organisms in the laboratory as a third set. This is conceptually appealing, but the original applications of the concept rely heavily on indices (e.g. diversity indices, toxicity indices) rather than on multivariate statistics. Procedures for testing and describing relationships among variables are reviewed, with emphasis on environmental applications. We describe and exemplify several multivariate statistical approaches to effective application of the sediment quality triad concept, including Mantel's tests and related descriptive analyses; ordination on each set of variables followed by methods to relate ordination axes among the sets; canonical correlation analysis; fitting a general linear model; and cluster analysis on each set of variables followed by a three-way contingency table analysis.

Presentation and interpretation of Sediment Quality Triad data

“…those which retain the multivariate properties of such data) have been conducted by various authors (e.g. Green, 1993;Green et al, 1993Green et al, , 1995Chapman et al, 1996;Kennicutt, 1995). The following comprises a general example of how Triad data might be analysed by multivariate techniques (adapted from Chapman et al, 1996 and Paine et al, in press).…”

“…Carr et al, in press). Green (1993) and Green et al (1993) also describe multivariate methods for analysing Triad data (including physical sediment characteristics), and similarly favour Mantel's test, which they follow with ordination. Kennicutt (1995) and Green et al (1995) describe application of these methods to document chronic sublethal effects due to oil and gas operations in the Gulf of Mexico.…”

“…Canonical correlation analysis (CCA) is another possibility. However, Green et al (1993) caution that CCA may not be appropriate due mainly to the sensitivity of this analysis to non-linear between-sets relationships (within-set relationships are assumed to be linear). Canonical correspondence analysis (CANOCO) could be more appropriate as it can deal with non-linear responses, but Green (personal communication, 1995) cautions that CANOCO has algorithmic and theoretical problems and does not advise using the detrending option.…”

Evaluation of Heavy Metal Sediment Toxicity in Littoral Ecosystems Using Juveniles of the FishSparus aurata

“…Nevertheless, the objective in this study was to provide site-specific SQVs for the area studied by means of chemical and biological data obtained by simultaneous analysis (synoptic sampling) and use of a multivariate approach to link both data sets. In this sense, the authors note that instead of using large data sets from samples collected and/or analyses performed at different times (coincident), they used this small data set based on samples collected and analyses carried out at once (synoptically), as recommended by different authors when integrative assessment is performed (Chapman, 1995(Chapman, , 1996Chapman et al, 1991b;Green et al, 1993).…”

Relationship of parasites and pathologies to contaminant body burden in sentinel bivalves: NOAA Status and Trends ‘Mussel Watch’ Program

“…Correlations were common among parasites for both variables and occurred commonly as well among the contaminants (see also Ringwood et al, 1999;Green et al, 1983Green et al, , 1993. Accordingly, to evaluate the potential relationships between parasites/pathologies and contaminant body burden in sentinel bivalves, principal components analysis (PCA) was used to reduce the many correlated variables in each dataset down to a few orthogonal variables for further analysis.…”