Bioassessment V: Analytical Approaches
Presiding: M F Bowman, Department of Zoology, University of Toronto; G Pond, USEPA Region 3
NB43A-01 13:30h
Incorporating Hypothesis Testing and Biologically Meaningful Effect Sizes into Bioassessments Using Test Site Analysis (TSA)
We use various benthic macroinvertebrate (BMI) datasets to illustrate the utility of a novel test site analysis (TSA) method. As in other approaches, the degree and nature of anthropogenic impairments are identified by comparing communities in potentially impacted test sites to communities in minimally impacted reference sites. The first step in TSA is to use traditional (e.g., % chironomids) and, or multivariate (e.g., ordination axes scores) metrics to describe the communities. Using all metrics simultaneously (i.e., overall biological distance), TSA calculations provide a single probability that a test community is different from reference communities. Any redundancies among metrics are accounted for in the biological distance calculation. If a test community is not statistically different from reference, it is further categorized as in reference condition or potentially impaired (i.e., more data required). If a test community is statistically impaired, the metric(s) important in distinguishing the test community from reference condition are identified. All TSA calculations can easily be done using an ExcelŽ spreadsheet. Thus, our use of a variety of summary metrics to obtain a single statistical test of significance within the context of the reference-condition approach provides a simple and unambiguous framework for evaluating the biological condition of test sites.
NB43A-02 13:45h
Comparison of a Multimetric Index and a Multivariate Predictive Model for Assessing the Biological Condition of Kentucky Streams
There is still debate on the strength of various data analysis tools for assessing biological condition in streams. This study compared two popular assessment approaches (multimetric index and RIVPACS-type O/E model) using macroinvertebrates from Kentucky streams. Data from 557 targeted and randomly selected sites (212 reference, 345 non-reference) sampled between 2000 and 2004 were used in this analysis. The Kentucky Macroinvertebrate Bioassessment Index (MBI) combines seven metrics (total generic richness, EPT generic richness, modified HBI, %Ephemeroptera, %EPT minus Cheumatopsyche, %midges+worms, and %clingers) that are scored by standardizing to the 95th or 5th percentile of the reference distribution and averaged. For comparison, three separate genus-level RIVPACS-type models were constructed (high-, low-, and mixed gradient streams) using four predictive variables (area, latitude, longitude, and week number) and taxa from reference sites. All 3 models preformed well but the low gradient model had the lowest precision. Assessments of non-reference sites based on MBI and O/E scores yielded similar results in terms of discrimination efficiency but the model based on mixed-gradient streams was the least sensitive. Using a subset of data from 84 headwater streams in the Appalachian region, MBI and O/E scores responded almost identically to stressors such as conductivity and habitat degradation.
NB43A-03 14:00h
Scaling by Reference Conditions for Ecological Assessment
Reference sites provide important information about the range of biological, physical and chemical measurements. Using reference information to scale data from other sites is useful for evaluating the status of sites and establishing impairment. A common approach is to use all the data on the reference conditions to standardize measurements for a new site. Rather than using all available reference sites to scale the observed value of a particular metric at a non-reference site, we present an alternative approach uses only the k closest (in terms of selected predictors) reference sites. These selected predictors may include only natural variables (for instance latitude and longitude), only potential stressors, or a combination of both and these variables may be continuous or categorical. We believe this nearest neighbors based distribution of the metric at a non-reference site is closer to the proper reference distribution for that site. Using benthic data from the Mid-Atlantic Highlands, we show that the nearest neighbor method improved on the ability of the existing approach to classify non-reference sites correctly and performed comparably on reference sites. We discuss issues related to the choice of k and present results suggesting an optimal k in this application.
NB43A-04 14:15h
Biological classification of Pennsylvania streams and rivers
The Pennsylvania Aquatic Community Classification (ACC) project is developing a classification of lotic systems by their biological characteristics using comprehensive macroinvertebrate, fish, and mussel community data. The ACC project seeks to identify primary aquatic system types using communities, as an alternative to ecoregions and other terrestrially-derived landscape classifications often applied to aquatic systems. Community class types identified through non-metric multidimensional scaling and cluster analysis revealed distinct communities across lotic systems. Landscape, physical, and chemistry variables were also associated with community types. Fish and mussel communities were primarily stratified by drainage basin; watershed drainage area, watershed land cover, slope, and water temperature secondarily classified fish community types. Mussel community classes were related to land cover and water chemistry. Geology, longitudinal stream gradients, and watershed land cover, rather than basins, were the most important variables in structuring macroinvertebrate community types. Classification strength analysis revealed that mussel community types have the highest class strength, following in descending class strength by fish and macroinvertebrates community types. Analysis of geographic distribution of community types, their relatively rarity, and threats have implications for stream conservation and restoration.
NB43A-05 14:30h
A Method for Adapting Ecological Distance Matrices to Asymmetrical Autocorrelation Within Dendritic Stream Networks.
Ecological relationships within dendritic stream networks are inherently asymmetrical: upstream sites (U) are effectively independent, while downstream sites (D) are not. In expressing a stream network as a distance matrix, D may be closer to U than U is to D. (Or: you can't describe a stream network with a symmetrical matrix!) These kinds of asymmetrical distance matrices are not amenable to common methods of accounting for autocorrelation, such as partial Mantel tests. A method is presented to account for asymmetrical dendritic flow relationships that produces a newly modified, yet still symmetrical, ecological distance matrix that may then be used in ordinations. The method is applied to data relating macroinvetebrate community composition within the streams of the semi-urbanised environs of Brisbane, Australia, to one environmental driver, catchment imperviousness. The resultant descriptive model is considerably improved after modifying the ecological distance matrix to account for the stream network relationships, thereby supporting the efficacy of this approach. A further benefit of the method is in allowing a direct comparison between asymmetrical, or flow-directed effects, and symmetrical, or non-directed, effects of stream networks, giving a measure of the relative importance of flow in determining ecological composition.
NB43A-06 14:45h
Order, Family, Genus, Species: How Does Choice of Taxonomic Resolution Affect Multivariate Statistical Outcome?
Taxonomic resolution is an important component of community analyses and may affect statistical recognition of biological responses to environmental conditions. A 60-site macroinvertebrate dataset collected from New York City water supply watersheds was examined for taxonomic resolution effects on indirect gradient analyses, specifically correspondence analysis (CA) and non-metric multidimensional scaling (NMDS). Environmental gradients were significantly correlated with species assemblages in the two study regions, East (EOH) and West of Hudson River (WOH). WOH sites contained 464 taxonomic units (mostly species, including chironomids) from 87 families in 22 orders. EOH sites contained 435 "species" from 80 families in 24 orders. CA clearly separated EOH from WOH sites at all taxonomic levels but order using raw densities, relative abundance, and presence/absence datasets. This indicated strong biogeographic differences at family but not order level. Within each region, NMDS clearly separated impacted sites at species and genera levels, but family and order level ordinations only separated severely impacted EOH sites and did not separate WOH sites along established environmental gradients. Taxonomic differences reflecting biogeography were apparent at gross and fine taxonomic levels but genus/species level contributed to the identification of mild impact gradients and differentiation of severe impact within regions.
http://www.stroudcenter.org/research/nyproject