Classifying the grain size of seabed sediments based on multibeam backscatter data—A case study in Joseph Bonaparte gulf, Australia

The accurate information of subaqueous topography and seabed substrata are of great significant for marine engineering construction, benthic habitat mapping, and management of marine protected areas (MPAs). The bathymetric and backscatter data of 880 km2 in the Joseph Bonaparte Gulf, Northern Australia were collected by using a multi-beam echo-sounder system (Kongsberg’s 300 kHz EM3002), and 54 samples of seabed sediments were collected simultaneously. The Random Forest Decision Tree (RFDT) was chosen as the modelling method for prediction. The results show that: (1) Improvement of the predicted accuracy for bed sediment classification is made when the parameters of RDFT are set as “number of trees” 200, “minimum size node to split” 2 and the “maximum splitting levels” 5 in this paper. (2) The highest accuracy of 83.3% is predicted from the incidence angle (backscatter) of 13° and 37°, and the coarse sediment, such as sandy gravel and gravelly sand are mainly distributed in the area with stronger backscatter intensity, but the fine sediment, such as gravelly muddy sand and (gravelly) muddy sand are distributed in the shallow area. However, it is noteworthy that the predicted accuracy of sediment classification may decrease when bathymetry data is chosen as the characteristic variable with the back-scatter.