Multi-angle backscatter classification and sub-bottom profiling for improved seafloor characterization
Alevizos, E.; Snellen, M.; Simons, D.; Siemes, K.; Greinert, J. (2018). Multi-angle backscatter classification and sub-bottom profiling for improved seafloor characterization. Mar. Geophys. Res. 39(1-2): 289-306. https://dx.doi.org/10.1007/s11001-017-9325-4 In: Marine Geophysical Researches. Reidel: Dordrecht. ISSN 0025-3235; e-ISSN 1573-0581, more |  |
| Keyword | | | Author keywords | Acoustic backscatter, Angular response analysis, Bayesian statistics, Sediment classification, Sub-bottom |
| Authors | | Top | - Alevizos, E.
- Snellen, M.
- Simons, D.
| | |
| Abstract | This study applies three classification methods exploiting the angular dependence of acoustic seafloor backscatter along with high resolution sub-bottom profiling for seafloor sediment characterization in the Eckernförde Bay, Baltic Sea Germany. This area is well suited for acoustic backscatter studies due to its shallowness, its smooth bathymetry and the presence of a wide range of sediment types. Backscatter data were acquired using a Seabeam1180 (180 kHz) multibeam echosounder and sub-bottom profiler data were recorded using a SES-2000 parametric sonar transmitting 6 and 12 kHz. The high density of seafloor soundings allowed extracting backscatter layers for five beam angles over a large part of the surveyed area. A Bayesian probability method was employed for sediment classification based on the backscatter variability at a single incidence angle, whereas Maximum Likelihood Classification (MLC) and Principal Components Analysis (PCA) were applied to the multi-angle layers. The Bayesian approach was used for identifying the optimum number of acoustic classes because cluster validation is carried out prior to class assignment and class outputs are ordinal categorical values. The method is based on the principle that backscatter values from a single incidence angle express a normal distribution for a particular sediment type. The resulting Bayesian classes were well correlated to median grain sizes and the percentage of coarse material. The MLC method uses angular response information from five layers of training areas extracted from the Bayesian classification map. The subsequent PCA analysis is based on the transformation of these five layers into two principal components that comprise most of the data variability. These principal components were clustered in five classes after running an external cluster validation test. In general both methods MLC and PCA, separated the various sediment types effectively, showing good agreement (kappa >0.7) with the Bayesian approach which also correlates well with ground truth data (r2 > 0.7). In addition, sub-bottom data were used in conjunction with the Bayesian classification results to characterize acoustic classes with respect to their geological and stratigraphic interpretation. The joined interpretation of seafloor and sub-seafloor data sets proved to be an efficient approach for a better understanding of seafloor backscatter patchiness and to discriminate acoustically similar classes in different geological/bathymetric settings. |
|
