Object-based classification of sub-bottom profiling data for benthic habitat mapping. Comparison with sidescan and RoxAnn in a Greek shallow-water habitat

Abstract

Although Ground Discrimination Systems and Multi-beam Echo Sounders tend to be the preferred tools towards gaining knowledge about the seafloor, numerous surveys have been and are still being performed by simultaneously using conventional sidescan Sonars (SSSs) and Sub-Bottom Profilers (SBPs). This is in respect of gaining fast, wide-scale, three-dimensional imaging of the seafloor and its substrate, extracting maximum value from a single, time-limited survey. The combination of these two systems offers good knowledge of both the stratigraphy and the habitat extents on the seabed, aspects often linked to each other. However, a basic drawback is the inconsistency between their mapping scales, as SSS produces high-resolution backscatter maps and SBP produces substrata information of high vertical but very low horizontal density. In this work, 100-kHz SSS and 3.5-kHz SBP data, collected simultaneously during a geophysical survey at a 5.5 × 3.3 km, shallow (10–50 m depth) Marine Protected Area in Lourdas Gulf, Kefalonia Island, Greece, underwent post-processing and analysis, to extract numerous statistical features from both the seafloor and its substrate, towards automated seafloor classification. The SSS mosaic image was subjected to segmentation, coupling the mean-shift algorithm and unsupervised classification. The SBP images were processed using edge enhancement/detection techniques and numerous features were extracted regarding the acoustic appearance, transparency and density of the automatically detected seismic reflectors. Supervised classification of the SBP derivatives exhibited their high discrimination ability between different sea-bed types, making them suitable for ground discrimination. Linking of the SBP classes to the SSS segments (object-based analysis) led to full coverage, high detailed benthic habitat maps. Training of the classifiers was carried out using data points in proximity to underwater video inspected areas, while validation was performed using a manual habitat map made by expert interpreters. Three marine habitat classification schemes have been examined separately, namely NATURA, EUNIS and a custom one, successively including more Habitat types in a hierarchical manner, as to validate the method under different complexity levels. The SSS mosaic, as well as an independent data-set acquired using RoxAnn acoustic ground discrimination system, were also used, each individually, to classify Lourdas Gulf in yet again an object-based manner. The latter showed that SBP derivatives led to an equally significant and in most cases better classification of benthic habitats, compared to those produced by the widely tested and documented SSS and RoxAnn systems, implying that they deserve further exploitation.