Classification of various algae canopy, algae turf, and barren seafloor types using a scientific echosounder and machine learning analysis

Abstract Diverse algae form algae canopies and turfs with various community structures, which play an important ecological role in coastal waters. Acoustic methods have been suggested and applied as effective quantitative methods for some algae canopy measurements across a large-scale area. However, these approaches face difficulties in accurately classifying turfs from barren seafloor due to weak backscattering strength. Thus, to estimate the community structure of various algae assemblages, we developed a classification method using a combination of acoustic-derived physical distance and backscattering strength parameters using a scientific echosounder. The prediction accuracy for algae or barren seafloor using four machine learning methods based on seven parameters was higher than that for the manual classification results based only on the acoustic physical distance. The classification accuracies of six types of algae canopy, turf, and barren seafloor were also higher than those obtained based only on commonly used seafloor parameters. Hence, machine learning methods based on the seven derived parameters from acoustic data are suggested to be effective for the classification. Applications in the classification and distribution estimations of various types of algae canopies, turfs, and potential algae habitat areas are promising.