Objective automated classification technique for marine landscape mapping in submarine canyons

This study proposes a fully automated and objective technique to map marine landscapes in submarine canyons. The method is suitable for broad and regional scale mapping derived from sonar data using multivariate statistical analysis. The method is divided into two main parts: the terrain analysis and the multivariate statistical analysis. The first part aims to optimise the sonar data and comprises three steps 1) data resampling, 2) determination of length scale, and 3) multiple scale analysis. The second part covers the actual marine landscape classification and consists of 1) principal component analysis (PCA), 2) K-means clustering, and 3) cluster determination. In addition, a confidence map is presented based on cluster membership derived from cluster distance in attribute space.