Choice of field and laboratory methods affects the detection of anthropogenic disturbances using stream macroinvertebrate assemblages.

Accurate and precise detection of anthropogenic impacts on stream ecosystems using macroinvertebrates as biological indicators depends on the use of appropriate field and laboratory methods. We assessed the responsiveness to anthropogenic disturbances of assemblage metrics and composition by comparing commonly employed alternative combinations of field sampling and individuals counting methods. Four datasets were derived by, in the field 1) conducting multihabitat sampling (MH) or 2) targeting samples in leaf packs (single-habitat sampling - SH) and, in the laboratory A) counting all individuals of the samples, or B) simulating subsampling of 300 individuals per sample. We collected our data from 39 headwater stream sites in a drainage basin located in the Brazilian Cerrado. We used a previously published quantitative integrated disturbance index (IDI), based on both local and catchment disturbance measurements, to characterize the intensity of anthropogenic alterations at each site. Family richness and % Ephemeroptera, Plecoptera and Trichoptera (% EPT) individuals obtained from each dataset were tested against the IDI through simple linear regressions, and the differences in assemblage composition between least- and most-disturbed sites was tested using Permutational Multivariate Analysis of Variance (PERMANOVA). When counting all individuals, differences in taxonomic richness and assemblage composition of macroinvertebrate assemblages between least- and most-disturbed sites were more pronounced in the MH than in the SH sampling method. Leaf packs seemed to concentrate high abundance and diversity of macroinvertebrates in highly disturbed sites, acting as 'biodiversity hotbeds' in these situations, which likely reduced the response of the assemblages to the disturbance gradient when this substrate was targeted. However, MH sampling produced weaker results than SH when subsampling was performed. The % EPT individuals responded better to the disturbance gradient when SH was employed, and its efficiency was not affected by the subsampling procedure. We conclude that no single method was the best in all situations, and the efficiency of a sampling protocol depends on the combination of field and laboratory methods being used. Although the total count of individuals with multihabitat sampling obtained the best results for most of the evaluated variables, the decision of which procedures to use depends on the amount of time and resources available, on the variables of interest, on the availability of habitat types in the sites sampled, and on the other methods being employed in the sampling protocol.