Acoustic discrimination of relatively homogeneous fine sediments using Bayesian classification on MBES data

Highlights: • Acoustically meaningful clustering correlating with ground truth data • Seascape scale acoustic mapping through classification per beam footprint • Bayesian approach for discriminating sedimentary units with low heterogeneity • Geo-acoustic resolution: a new measure of sedimentary acoustic class separability Modern seafloor mapping is based on high resolution MBES systems that provide detailed bathymetric and acoustic intensity (backscatter) information. We examine and validate the performance of two unsupervised MBES classification techniques for discriminating acoustic classes of sedimentary units with small grain size variability. The first technique, based on a principal components analysis (PCA), is commonly used in literature and has been applied for comparison with the more recent approach of Bayesian statistics. By applying these techniques to a MBES dataset from an estuarine area in The Netherlands, we tested their ability to discriminate fine grained sediments (at least 70% silt) holding small percentages of coarser material such as sand, shell hash or shells. We focus on the Bayesian technique as it outputs acoustically significant classes related to backscatter values. This technique utilizes backscatter values averaged over scatter pixels (projected pulse lengths) inside the footprint of each beam. The originality of our application lies in the fact that, the optimal number of classes is derived by utilizing a number of beams simultaneously. It is assumed that the backscatter values per beam vary relatively to the varying seafloor types. By treating the beams separately, across track variation in the seafloor type can also be accounted for. Thereby the classification is guided by outer, more discriminative beams. Additionally we control the optimal number of classes by employing the quantitative criterion of goodness of fit (χ2). The Bayesian acoustic classes show correlation with grain size parameters such as coarse fraction (>500μm) percentage and mean of the grain size (<500μm) when analyzed with multiple linear regression. In order to examine the relative scale of the acoustic classification results we compare the Bayesian acoustic classes with underwater video interpretation. Our results reveal that the Bayesian approach enhances the sedimentological interpretation of MBES high resolution data, by providing classification on seascape scale (here meters to tens of meters). Hence we suggest that backscatter processing techniques are more commonly applied to produce classes that discriminate sediments with low grain size contrast. To describe this ability we introduce the term geoacoustic resolution. We want to encourage the use of the Bayesian technique also in deep sea applications, based on AUV data, where sediments express low variability but sampling would be time consuming and costly. The advantages of this method would favor mapping of macro-habitats which appear at meter-scale and require datasets of sufficient resolution in order to be quantitatively described.