The reference condition: a comparison of multimetric and multivariate approaches to assess water-quality impairment using benthic macroinvertebrates

Traditional methods of establishing control sites in field-oriented biomonitoring studies of water quality are limited. The reference-condition approach offers a powerful alternative because sites serve as replicates rather than the multiple collections within sites that are the replicates in traditional designs using inferential statistics. With the reference-condition approach, an array of reference sites characterises the biological condition of a region; a test site is then compared to an appropriate subset of the reference sites, or to all the reference sites with probability weightings. This paper compares the procedures for establishing reference conditions, and assesses the strengths and deficiencies of multimetric (as used in the USA) and multivariate methods (as used in the UK, Canada, and Australia) for establishing water-quality status. A data set of environmental measurements and macroinvertebrate collections from the Eraser River, British Columbia, was used in the comparison. Precision and accuracy of the 2 multivariate methods tested (AUStralian RIVer Assessment Scheme: AusRivAS, BEnthic Assessment of SedimenT: BEAST) were consistently higher than for the multimetric assessment. Classification by ecoregion, stream order, and biotic group yielded precisions of 100% for the AusRivAS, 80-100% for the BEAST, and 40-80% for multimetrics; and accuracies of 100%, 100%, and 38-88%, respectively. Multimetrics are attractive because they produce a single score that is comparable to a target value and they include ecological information. However, not all information collected is used, metrics are often redundant in a combination index, errors can be compounded, and it is difficult to acquire current procedures. Multivariate methods are attractive because they require no prior assumptions either in creating groups out of reference sites or in comparing test sites with reference groups. However, potential users may be discouraged by the complexity of initial model construction. The complementary emphases in the multivariate methods examined (presence/absence in AusRivAS cf. abundance in BEAST) lead us to recommend that they be used together, and in conjunction with, multimetric studies.