Geology & Geophysics Group
Seabed imagery & characterisation
Contact: Dr Tim Le Bas
Characterisation of the seafloor is achieved through a variety of acoustic methods. Sidescan sonar has traditionally been the most indicative of seabed types and processes taking place. Multibeam bathymetry data is now much more accessible, affordable and understandable and usually has the added advantage of providing backscatter intensity information. However sidescan systems often provide much better resolution and quality backscatter data.
Some key areas of research include:
- Reduction of systematic and acoustically generated features which may mask more subtle seabed imagery features
- Texture analysis of sonar imagery
- Habitat mapping
- Marine GIS - correlation of different frequency and resolution datasets (acoustic and non-acoustic)
- Geomorphologic analysis of bathymetric data
Above: Pseudo-colour GLORIA imagery with false shaded relief in the Bering Sea Basin just north of the Aleutian Islands at 175ºW 53ºN. The image shows the Umnak Fan-lobe system. It is a channel-lobe transition running easterly, downslope. There is zonation from broad channel mouth (weak backscatter - light brown) to braid-like bedform bars (strong backscatter- blue).
To assist the processing of sonar imagery, a number of software packages are used - GMT scripts for basic mapping, to commercially-available software packages for specialised hardware-specific data. However, our principal processing package is PRISM (Processing of Remotely-Sensed Imagery for Seafloor Mapping) which is our own in-house open-source highly flexible software suite. It is currently available free of charge to academic instotutions and comes with documentation and user training.
Above: Multibeam backscatter and bathymetry shaded relief imagery South West of Barra (Outer Hebrides). Water depths vary from 60 to 160metres (red to purple respectively). Tabular sand bodies showing gaps to the deeper gravel below are invisible in the bathymetry imagery (purple) but easily stand out in backscatter imagery (top right corner).
The main purpose of digital image processing is to remove as many characteristics of the specific sonar vehicle as possible, leaving an image of the geological features of the area. This means altering and correcting the imagery by, for example, removing system noise, geometric and radiometric distortions and providing as near as possible a geomorphological map of the seafloor on which it is irrelevant which imaging system was used. This therefore allows easier interpretation and increases data confidence, as the imagery does not require inherent knowledge of a particular imagery system. This is imperative on presentation to the wider scientific community.
Above: An example of the across-track beam pattern of the TOBI sidescan sonar system – which must be removed from the imagery when performing the radiometric processing.
The G&G Group is fortunate to be able to process sonar data from a great variety of sources. Many digital processing techniques have been developed for sidescan sonar in terms of geometric corrections, radiometric corrections and imagery manipulation, together with the calcuation of the errors involved. The geometric corrections change the relative positions of pixels in the imagery, such corrections include slant-range, anamorphosing and georeferencing. The radiometric corrections alter the original pixel values so that the relative values of neighbouring pixels are changed. Techniques include removal of surface reflections, errors in time varied gain, line dropout, beam illumination shading, filtering for speckle noise and deblurring. Imagery manipulation is defined as the whole process of image processing together with the automation of techniques processing, geographic information systems (GIS), error analysis and data confidence.
Any type of acoustic image can be described by its texture. We take advantage of this fact by using texture analysis techniques, e.g. Grey Level Co-occurrence Matrices, combined with supervised or unsupervised classification, to describe the sidescan sonar data. This method calculates statistical indices that quantify the distribution of grey levels and their spatial relationship within the image, and classifies each image pixel according to the combination of its resulting textural indices. This is being done in moving windows using TexAn, and the indices used most often to classify sidescan sonar imagery are Entropy and Homogeneity (Blondel, 1996; Huvenne et al., 2002; Cochrane, 2002).
Right: Supervised classification on High resolution Sidescan Sonar imagery showing classes in texture analysis. The method has correctly identified the training areas (white boxes). Imagery near-nadir is unreliable and thus has been greyed out.
