Sampling effort affects multivariate comparisons of stream assemblages

Abstract

Multivariate analyses are used widely for determining patterns of assemblage structure, inferring species-environment relationships and assessing human impacts on ecosystems. The estimation of ecological patterns often depends on sampling effort, so the degree to which sampling effort affects the outcome of multivariate analyses is a concern. We examined the effect of sampling effort on site and group separation, which was measured using a mean similarity method. Two similarity measures, the Jaccard Coefficient and Bray-Curtis Index were investigated with 1 benthic macroinvertebrate and 2 fish data sets. Site separation was significantly improved with increased sampling effort because the similarity between replicate samples of a site increased more rapidly than between sites. Similarly, the faster increase in similarity between sites of the same group than between sites of different groups caused clearer separation between groups. The strength of site and group separation completely stabilized only when the mean similarity between replicates reached 1. These results are applicable to commonly used multivariate techniques such as cluster analysis and ordination because these multivariate techniques start with a similarity matrix. Completely stable outcomes of multivariate analyses are not feasible. Instead, we suggest 2 criteria for estimating the stability of multivariate analyses of assemblage data: 1) mean within-site similarity across all sites compared, indicating sample representativeness, and 2) the SD of within-site similarity across sites, measuring sample comparability.

Reliable sample sizes for estimating similarity among macroinvertebrate assemblages in tropical streams

“…However, at increased sample size, this dependence tends to disappear and attain a constant similarity value. Similar results were found by Wolda (1981) and Cao et al (1997Cao et al ( , 2002. This result shows the importance of determining adequate sampling efforts, because small sample sizes may underestimate similarities among samples of macroinvertebrate assemblages.…”

“…This results in high autosimilarity values and early curve stabilization. Similarly, Cao et al (2002) found that individual-based autosimilarity curves of stream macroinvertebrates stabilized earlier than sample-based ones, for both presence-absence and relative abundance data. However, in contrast to our results, they observed similar autosimilarity values for individual-and samplebased curves.…”

“…They observed asymptotes for presence-absence data (Jaccard index) but not for relative abundance data (Bray-Curtis index), with the latter being strongly dependent on sample size. Likewise, Cao et al (2002) using macroinvertebrates, and Kennard et al (2006) studying stream fish assemblages in Australia, found that smaller sample sizes are required to estimate resemblance using species composition than relative abundances. As in our study, Cao et al (2002) and Schmera and Eros (2006) observed lower autosimilarity values for presence-absence data than for relative abundance data.…”

“…The procedure is repeated for different sample sizes up to the maximum possible sample size (half of the full data set). Such autosimilarity curves (Cao et al, 2002) usually attain an asymptote and thus an adequate or representative sample size is obtained when increases in sample sizes do not result in higher similarity values.…”

Estimating sample representativeness in a survey of stream caddisfly fauna

“…In contrast, precision can be expressed by comparing repeated measurements of the same quality. It is the basic idea in the development of the term sample representativeness (Cao et al 2002a(Cao et al , 2002b, a term used to describe how well a sample represents the community. In order to measure representativeness, two samples are collected from the same community and a measure of the similarity (or autosimilarity Cao et al 2003) determined.…”