Biological criteria for defining water quality and the presence of acceptable levels of benthic resources are evaluated for estuarine macrobenthic communities of the lower Chesapeake Bay, USA. Models of expected community values as a function of salinity are presented for community biomass, numbers of individuals, species richness, percent biomass of deep-dwelling species, percent biomass of equilibrium species, and percent biomass of opportunistic species. The models presented may serve as, or be used to develop, biological criteria for estuaries. Tidal freshwater and oligohaline regions had the highest variability in model parameters due to patchily distributed, large-sized bivalve species. In the absence of data from pristine habitats, the models were developed from a 5 year data set (1985-1989) for stations considered to be minimally impacted. The models produced were used to evaluate benthic communities of two regions of the Chesapeake Bay-one exposed to summer low dissolved oxygen events (hypoxia/anoxia) and the other characterized by sediments contaminated with heavy metals and polynuclear aromatic hydrocarbons. Stations exposed to stress from either low dissolved oxygen events or contaminated sediments were characterized by 1. reduced community biomass, 2. reduced species richness, 3. less biomass consisting of deep-dwelling species and equilibrium species and 4. more biomass consisting of opportunistic species. Some unstressed habitats can be highly dominated by shallow-dwelling long-lived species, thus dominance of deep-dwelling species in biomass must be used with caution as a biological criterion. The number of individuals per m2 was highly variable for some stressed stations and this parameter is probably of limited value as a biological criterion characterizing the quality of estuarine habitats. No single method or analysis is likely to produce stress classifications without unacceptable misclassifications. Ecological stress, from any source, is best measured by multiple methods or analyses with different assumptions. The consistency of classification between different approaches would provide the robustness necessary to judge the reliability of a stress classification.
