Predictive mapping of seabed substrata using high-resolution multibeam sonar data: A case study from a shelf with complex geomorphology

Abstract

Abstract This study explores the full potential of high-resolution multibeam data for the automated and accurate mapping of complex seabed features under a predictive modelling framework. For an area of seabed on the Carnarvon shelf in Western Australia, morphometric variables and textural measures were derived from multibeam bathymetry and backscatter data. Several feature extraction approaches were applied to backscatter angular response curves to obtain new features. These derivatives and new features were used separately and in combination in the predictions. Despite the complex distribution of various hard substrata within the study area, we achieved a nearly perfect prediction of “hard vs soft” seabed types with an AUC (Area Under Curve) close to 1.0. The predictions were also satisfactory for gravel, sand and mud content, with R 2 values that range from 0.55 to 0.73. This study demonstrates that using a full range of derivatives and new features from both multibeam bathymetry and backscatter data optimises the accuracy of seabed mapping. From the modelled relationships between sediment properties and multibeam data, we confirmed that coarser sediment generally generates stronger backscatter return. Importantly, the results again highlight the advantages of applying proper feature extraction approaches over using original backscatter angular response curves.