Mapping seabed assemblages using comparative top-down and bottom-up classification approaches
Acoustic technologies yield many benefits for mapping the physical structure of seabed environments but are not ideally suited to classifying associated biological assemblages. We tested this assumption using benthic infauna data collected off the south coast of England by applying top-down (supervised) and bottom-up (unsupervised) classification approaches. The top-down approach was based on an a priori acoustic classification of the seabed followed by characterization of the acoustic regions using ground-truth biological samples. By contrast, measures of similarity between the ground-truth infaunal community data formed the basis of the bottom-up approach to assemblage classification. For both approaches, individual assemblages were mapped by first computing Bayesian conditional probabilities for ground-truth stations to estimate the probability of each station belonging to an assemblage. Assemblage distributions were then interpolated over a regular grid and characterized using an indicator value index. While the two methods of classification yielded assemblages and output maps that were broadly comparable, the bottom-up approach arrived at a slightly better defined set of biological assemblages. This suggests that acoustically derived seabed data are not ideally suited to class ifying biological assemblages over unconsolidated sediments, despite offering considerable advantages in providing rapid and low-cost assessments of seabed physical structure.