Acoustic seabed classification

Surficial seabed classification is concerned primarily with distinguishing marine benthic habitat characteristics (e.g. hard, soft, rough, smooth, mud, sand, clay, cobble) of the surveyed area. Multibeam echosounders, sidescan sonar systems and acoustic ground discrimination systems (AGDS) are the most commonly used technologies. The use of optical sensors has been restricted to depths less than 40 m due to absorption of electromagnetic radiation by water. Despite this limitation, processing tools have been developed to classify data acquired using airborne bathymetric LiDAR systems.[1] Nevertheless, acoustics remain the preferred method of imaging the seafloor because data can be acquired over a much larger area (than in-situ sampling) from almost any depth.

Multibeam systems acquire both bathymetry (depth) and backscatter (intensity) data. Multibeam backscatter was previously considered to be a by-product of a multibeam survey, with bathymetry being the primary information. Recent advances in multibeam backscatter acquisition, processing and analysis methods have increased the range of applications for which multibeam systems can be used and now allow the collection of spatially and temporally coincident multispectral multibeam backscatter.[2] [3] [4] New methods of analyzing backscatter data, have increased its potential for seabed characterization. Backscatter data resolution has also increased significantly with the introduction of snippet data. Snippet data is raw backscatter time-series data for each beam footprint and each ping (Lockhart et. al., 2007). These advances have allowed some multibeam backscatter data to achieve a quality comparable to that of sidescan sonar imagery.

Different classification approaches and algorithms can yield different results. These approaches include image-based seabed classification methods such as texture analysis, artificial neural networks (ANN); and other methods, such as angular response characterization (Hughes-Clarke et al., 1997). Image processing methods traditionally used in satellite remote sensing are often adapted to quantitatively analyze multibeam backscatter intensity data. After image segmentation and classification, acoustic imagery can be used to discriminate between areas with different morphological properties. No classification method produces a map that is 100% accurate and some attempt must always be made to assess the accuracy of classification results (e.g. confusion matrix).

Classification maps are subject to ground-verification in order to identify the compositions and bottom type that characterize each class. The functionality of Geographic Information Systems (GIS) can be used to integrate data from different sources, including ground truth data. Such data may come from in-situ sediment grab sampling, the use of a dredge, trawl net, visual imagery or surveys using Remotely Operated Vehicles (ROVs). The seabed classification map can be combined with other information about the area, such as fish distribution and abundance or vegetation characteristics, to establish habitat groups based on associations. This process allows classification maps derived from multibeam data to help characterize the seabed and more effectively manage its use.