Acoustic seabed classification methodology: a user’s statistical comparison

Methods for seabed classification using echosounder data have been available for about 10 years. In this paper, we examine aspects of the statistical processing used by the QTC VIEW acoustic bottom classification system, which characterises bottom types through the shape of the first echo. We focused our investigation on 5 questions: (1) How to filter errors in input data? (2) How is principal component analysis (PCA) computed in the QTC software? How many principal components should be kept for classification analysis? Is PCA the only way of eliminating redundancies and noise in the data? (3) Is the method used by QTC for clustering a good method? (4) How to decide on the optimal number of acoustic classes? (5) How does the QTC classification compare to K-means results? While the decomposition of the acoustic signal into shape variables produced by QTC VIEWTM is empirically useful, the QTC IMPACTTM software implements sub-optimal classification procedures. An alternative method (PCA followed by K-means partitioning) is presented which produces statistically better results; software is freely available.