Abstract Intensive temporal and spatial sampling of benthic infauna have rarely been analysed to improve impact assessment or conservation planning. An impact assessment started in the late 1980s in Jervis Bay (Australia) provided a spatially, temporally and taxonomically comprehensive benthic invertebrate dataset. While the proposed development did not eventuate, the area was later declared a multi‐zone marine park. We specifically used the polychaete data which included nine sampling times (February 1989 to June 1991) in seagrass and unvegetated sediments across multiple sites to identify ecological patterns that could inform current and future conservation management and impact assessment. The polychaete assemblage was diverse with 171 species, many of them rare (40% < 10 individuals). Variation was greatest at small spatial scales (metres) for abundances and diversity, and there was substantial spatio‐temporal variation due to inconsistent site differences. Most species occurred in both unvegetated and seagrass sediments, though several were more abundant or restricted to one habitat (39% and 3% only found in unvegetated or seagrass, respectively), or showed bathymetric differences in abundances (e.g., Eunice cultrifera and Lumbrineris cf. latrelli). The current marine park zoning representatively covered polychaete biodiversity, as all sites were generally similar in terms of abundances and species richness. It also representatively covered seagrass and unvegetated sediments and the bathymetric range of the embayment. These findings reinforce the utility of using habitats as surrogates to representatively zone for biological diversity. The high levels of spatial and spatio‐temporal variation in abundances and diversity only allowed detection of environmental impacts with >70% effect sizes even if numerous sites and times were sampled (i.e., > seven sites and > eight times). This suggests that most univariate measures for polychaetes would only be useful to detect enormous effects, and effective impact assessment might require more sophisticated multivariate approaches.
Read full abstract