Multi-frequency Acoustic Mapping of Marine Benthos: Data-driven Multibeam Classification in the Dutch North Sea

With increasing offshore human activities and accelerating climate change, regular seabed habitat monitoring is essential for marine conservation and sustainable coastal development. Compared to destructive bottom sampling that is labor-intensive and optical remote sensing with limited penetration in seawater, the multibeam echosounder (MBES) provides a cost-effective solution for high-resolution, large-scale seabed mapping by simultaneously acquiring bathymetry and acoustic backscatter. In recent years, multi-spectral MBES has become a state-of-the-art acoustic mapping technique, providing nearly co-located multi-frequency measurements and largely enriching seabed characterization. Despite difficulties in obtaining calibrated backscatter, data-driven methods, especially machine learning techniques, still allow for linking MBES measurements to seabed geophysical and biological properties.
Nevertheless, MBES-based benthic habitat mapping remains challenging. Limited seabed ground truth hinders model construction and evaluation. Lack of absolute calibration poses challenges when comparing or combining MBES backscatter across surveys. Backscatter angular dependency and large volume of multi-frequency measurements further complicate data processing. This thesis addresses these challenges by exploiting the multi-spectral MBES, making optimal use of limited ground truth, and improving the MBES data processing workflow....