Weak correspondence between landscape classifications and stream invertebrate assemblages: implications for bioassessment

We examined the ability of 5 a priori landscape classifications (3 sizes of catchments, ecoregions, and life zones) to partition observed variation in the composition of stream invertebrate assemblages. Analyses were based on 254 unimpaired sites from montane streams in California and 1912 sites from several regions of the United States. Estimates of classification strength (CS) were used to measure how well classifications partitioned variation in composition among sites. CS was measured as the difference between overall weighted mean within-class compositional similarity ((W) over bar) and mean between-class similarity ((B) over bar). We compared CSs of the 5 a priori classifications with post hoc classification of sites based on cluster analysis of the pair-wise similarities among sites. The latter, a posteriori, classification was assumed to represent a near-optimal partitioning of (W) over bar and (B) over bar, and thus the maximum CS possible. Additional analyses were conducted to determine if level of taxonomic resolution (genus and family) and type of abundance data (presence/absence and log(10) density) affected CSs. All CSs, including those based on the post hoc cluster analyses, were weak with the best classification resulting in only a 14% difference between (W) over bar and (B) over bar. Classifications in which sites within a class were located relatively close to one another were generally stronger than classifications in which sites were located in geographically noncontiguous classes. Although the type of abundance measure used had little effect on CSs, classifications based on genus-level identifications were slightly stronger than those based on families. In general, these results imply that no currently used landscape classification system can predict stream invertebrate faunas with much precision. We suspect weak classifications occur because stream invertebrate taxa appear to vary independently and continuously along environmental gradients, and sites therefore exhibit little tendency to cluster into discrete groups. Methods of bioassessment that classify sites into a few discrete groups are therefore likely to be fundamentally limited in how precisely they can specify expected conditions, and thus the degree of biological impairment that they can detect. Methods that explicitly recognize that streams are biological continua may avoid this limitation.