Acoustic mapping and classification of benthic habitat using unsupervised learning in artificial reef water

Abstract

a b s t r a c tArtificial reefs (ARs) are effective means to maintain fishery resources and to restore ecological environment in coastal waters. ARs have been widely constructed along the Chinese coast. However, understanding of benthic habitats in the vicinity of ARs is limited, hindering effective fisheries and aquacultural management. Multibeam echosounder (MBES) is an advanced acoustic instrument capable of efficiently generating large-scale maps of benthic environments at fine resolutions. The objective of this study is to develop a technical approach to characterize, classify, and map shallow coastal areas with ARs using an MBES. An automated classification method is designed and tested to process bathymetric and backscatter data from MBES and transform the variables into simple, easily visualized maps. To reduce the redundancy in acoustic variables, a principal component analysis (PCA) is used to condense the highly collinear dataset. An acoustic benthic map of bottom sediments is classified using an iterative self-organizing data analysis technique (ISODATA). The approach is tested with MBES surveys in a 1.15 km 2 fish farm with a high density of ARs off the Yantai coast in northern China. Using this method, 3 basic benthic habitats (sandy bottom, muddy sediments, and ARs) are distinguished. The results of the classification are validated using sediment samples and underwater surveys. Our study shows that the use of MBES is an effective method for acoustic mapping and classification of ARs.

Automated Stone Detection on Side-Scan Sonar Mosaics Using Haar-Like Features

“…In environments with high sediment mobility, e.g., shallow shelf seas affected by waves and tides, hard substrates might become temporarily buried while previously buried ones might become exposed [9,10]. As yet, the spatial detection of underwater objects can only be achieved using hydroacoustic remote-sensing devices such as side-scan sonars (SSS), multibeam echo sounders (MBES), and parametric sediment echo sounders (pSES) (e.g., [11][12][13]). SSS data are usually analyzed by means of automated and semi-automated methods for the detection of objects such as ship wrecks or mines (e.g., [14,15]).…”

Assessing the Repeatability of Automated Seafloor Classification Algorithms, with Application in Marine Protected Area Monitoring

“…An initial list of terrain features, including BPI, curvature (planar and profile), VRM, eastness, northness and slope was assessed, considering their successful application in previous studies [7,13,51,52,70]. These terrain features are so popular that they have been incorporated in an ArcGIS toolbox named Benthic Terrain Modeler [29,30].…”

Unified Geomorphological Analysis Workflows with Benthic Terrain Modeler

“…The software can also be used to generate covariates at multiple scales to aid in understanding the effects of scale on prediction models like Maximum Entropy (MaxEnt) model outputs as in Miyamoto et al [68]. In use cases like Li et al [69], the outputs from BTM are used in an unsupervised classification using the ISODATA algorithm to build a validated benthic classification map. While BTM has been providing new algorithms for rugosity for many years, it is still common for users to rely on earlier methods like SAPA [70], likely because of its well-known association with capturing rugosity.…”