Replication and mitigation of effects of confounding variables in environmental impact assessment: effect of marinas on rocky-shore assemblages

Design- and model-based approaches provide 2 distinct frameworks in which to analyse environmental impacts. While both approaches have been criticized to some extent, the possibility of modeling environmental variability with covariates in the analysis of structured sampling designs seems to have eluded the attention of environmental ecologists. We applied a beyond-BACI (before, after, control, impact) sampling design to a post-impact study on the effects of marinas on assemblages of algae and invertebrates on rocky seashores at 2 locations in the NW Mediterranean. The 2 locations were on the opposite (northern and southern) sides of a promontory, about 30 km apart. A shore next to the marina and 3 reference shores were sampled at each location to obtain independent estimates of differences between disturbed and undisturbed conditions. An additional set of 3 shores were sampled at a location on the western coast to test whether structured spatial patterns around the promontory might confound the analysis of impact. Midshore and lowshore habitats were sampled on all shores. Multivariate and univariate analyses revealed differences between disturbed and reference assemblages mostly at one location in the midshore habitat. Analyses conducted on the lowshore data indicated that assemblages of marinas followed different temporal trajectories than reference shores consistently at the 2 locations. Spatial variability around the promontory was large, as indicated by significant linear and quadratic contrasts among all locations. When impacts were examined using either the geographic position of shores or their rank position around the promontory as a covariate, results did not differ from those of the original analyses. The present study highlights possible ways to overcome the dichotomy between design- and model-based approaches, namely that studies of impact can and should be examined at replicated locations whenever possible and that covariables can be included in the analysis of structured sampling designs to increase precision and accuracy in assessment of environmental impacts.