Abstract
Multibeam bathymetric sonar technology and benthic habitat research require the systematic characterization of the seafloor, necessitating reliable and accurate sea floor descriptors in combination with a robust means to statistically assess descriptor associations. Historically, geoscientific sea floor characterisation involves identifying the spatial extent and relationship of geological units, broadly following litho– or chronostratigraphic criteria, but these conventions may not be meaningful biologically because they incorporate temporal elements that stem from a geochronological qualifier. Textural properties of geological facies are typically given in terms of distribution-dependent statistics, which have been shown to be inappropriate with multimodal marine sediments, such as on glaciated shelves. As habitat classification is aimed at boundary definition, the boundaries between groups in such cases could be arbitrary, or based on very subtle differences, or noise (e.g., sampling bias). This study uses an independent statistical approach pioneered by Calinski and Harabasz (C–H) which offers significant advantages in determining the appropriate number of groups that might exist in any sample population. Used in conjunction with a multivariate extension to information-entropy, grain size populations can be clustered into statistically validated groups. This study utilizes a 30-yr legacy of 4-class grain size data collected from the Scotian Shelf, Canadian Atlantic continental margin, we show that a traditional stratigraphic approach does not provide clear discrimination between basic textural types, and hence, basic benthic habitats. Considerable improvements in textural zonation are obtained using a combination of information entropy-clustering and C–H technique. Two high resolution, 32-class particle-size data sets yield a solution where no obvious textural groups exist, contrary to published field-based studies. Comparison of sediment grab samples to bottom photographs from other shelf sites show that photos capture (sample) a wider range of textural variability, particularly the coarsest-gravel component that is sometimes absent from grabs, and therefore, classification from photos creates more groups. This study emphasizes that data resolution and sea-floor sampling strategies should be intimately linked, and to fully unravel high-resolution textural data might require in excess of a four order of magnitude increase in the number of bottom sediment samples. Therefore, data should be collected at the highest practical resolution but be reduced to a resolution meaningful for statistical analysis, in accordance with the total sample population.