Rapid Communication / Communication RapideAcoustic seabed classification: improved statistical method

Abstract

Huge amounts of money will be spent by industrialized nations during the next decades to obtain detailed maps of continental shelf seabeds. These maps, which will allow a more rational exploitation of the sea floor, are needed to assess the impact of anthropic activities. The statistical method of analysis of echosounder backscatter data described in this paper presents several improvements over existing techniques. The steps are as follows. (i) The backscatter data are decomposed mathematically into a number of quantitative variables, which are subjected to principal component analysis (PCA). (ii) Principal components representing 95–99% of the variation are used in a K-means partitioning procedure. A statistical criterion indicates what the number of groups is that best reflects the variability of the data. (iii) The groups are then plotted on maps of the survey area. Insofar as the mathematical decomposition produces variables that reflect the variations of the physical nature and composition of the seabed, the classes of the partition will correspond to different seabed types. Free software (The Q Package) implementing this method is available at http://www.fas.umontreal.ca/biol/legendre/.