Predictive mapping of seabed substrata using high-resolution multibeam sonar data: A case study from a shelf with complex geomorphology
Introduction
Accurate information on seabed substrata is key for effective benthic habitat mapping (e.g., Lanier et al., 2007, Erdey-Heydorn, 2008, Brown et al., 2011), benthic biodiversity/species prediction (e.g., Brown et al., 2002, Holmes et al., 2008, McGonigle et al., 2009, McGonigle et al., 2011), and management of marine protected areas (MPAs) (e.g., Huang et al., 2011). Traditionally, seabed information was only available for a limited number of points collected during marine surveys. However, the rapid development of remote sensing technologies provides great potential for automated and accurate mapping of the seabed across large areas. In particular, (active) acoustic remote sensing techniques such as sidescan and multibeam sonar utilise the propagation of acoustic signals through the water column and their return from the seabed interface to map large areas of seabed in water depths up to several thousand metres (e.g., Dartnell and Gardner, 2004, McGonigle et al., 2009, Brown et al., 2011, Huang et al., 2012a). In the last two decades, multibeam sonar (echo-sounder) has become the preferred seabed mapping tool because it can collect simultaneous and co-registered bathymetry and backscatter data (Hewitt et al., 2010, Brown et al., 2011, Micallef et al., 2012). Modern high-frequency multibeam sonar systems transmit pulses of sound and receive backscatter signals from hundreds of narrow-angle beams that generate small footprints on the seabed. Therefore, they can produce bathymetry and acoustic backscatter data with a spatial resolution similar to airborne remotely sensed data (e.g., less than 3 m at a depth up to 150 m).
The overall aim of this study was to map (and classify) an area of geomorphically complex seabed from high-resolution multibeam data using a robust predictive mapping approach. Both unsupervised and supervised classification techniques, as well as hybrid approaches, have previously been used for seabed mapping. Two examples of unsupervised techniques used to classify acoustic data are QTC-Multiview (McGonigle et al., 2009, Preston, 2009) and the CLARA clustering algorithm (Hamilton and Parnum, 2011). In these approaches, data are classified into seabed acoustic classes and ground-truth samples are used to attribute the acoustic classes into meaningful seabed substrata. In contrast, predictive modelling methods such as classification trees, Neural Networks and rule-based approaches have been used to classify acoustic data in a supervised manner (e.g., Dartnell and Gardner, 2004, Zhou and Chen, 2005, Lathrop et al., 2006, Rooper and Zimmermann, 2007, Huang et al., 2012a, Huang et al., 2013). These supervised approaches used ground-truth data to develop a predictive model which was then used to predict the whole study area. In this study, we use the supervised method for its ability to provide reliable accuracy assessments and for investigating modelled relationships between explanatory variables and target variables.
One yet to be fully realised advantage of multibeam data is that we can derive a large number of additional variables from both bathymetry and backscatter data (Huang et al., 2012a, Huang et al., 2013). Only a few studies have used both multibeam bathymetry and backscatter data for seabed mapping (Mitchell and Hughes Clarke, 1994, Dartnell and Gardner, 2004, Huang et al., 2012a), but not to their full potential. For example, Mitchell and Hughes Clarke (1994) used the backscatter data and two derivatives of the bathymetry data. Dartnell and Gardner (2004) used the backscatter data and three derivatives of the backscatter and bathymetry data. Huang et al. (2012a) used a larger number of derivatives of the backscatter and bathymetry data. None of them, however, used the backscatter angular response curves (see Section 2 for details), which have been demonstrated to be very useful for seabed mapping (e.g., Hughes Clarke, 1994, Fonseca and Mayer, 2007, Hamilton and Parnum, 2011, Huang et al., 2013).
The main objective of this study is to explore the full potential of multibeam data for the automated and accurate mapping of seabed substrata by using a large number of derivatives of bathymetry and backscatter data, and new features obtained from backscatter angular response curves. Although these datasets do not exhaust all possible variables, they represent a full range of information from both primary sources of multibeam data (i.e. bathymetry and backscatter). We hypothesise that, with additional input variables of different sources, the accuracy of the seabed mapping will be improved. Another objective of this study is to demonstrate the value of using this robust modelling technique to quantitatively investigate the underlying relationships between multibeam data and seabed properties. We chose a study area that has complex seabed geomorphology ranging from hard substrate to various soft sediment types to investigate the above two objectives.
Section snippets
Background — multibeam data and seabed mapping theory and applications
Multibeam bathymetry and backscatter data, and new variables derived from them provide complementary information for accurate seabed mapping. Their chief utility lies in the capacity to describe seafloor morphology and seabed texture that are proxies of oceanographic processes and seabed physical properties.
Materials and methods
In this section, we first describe the study area, and the acquisition of the multibeam data (Section 3.1). We then detail the methods used to process the raw multibeam data, obtain derivatives of the bathymetry and backscatter data, and derive new features from the backscatter angular response curves (3.2 Processing of raw multibeam data, 3.3 Bathymetry data and its derivatives, 3.4 Backscatter mosaics, segmentation and textural measures, 3.5 Backscatter angular response curves and feature
Results for the prediction of “hard vs soft” seabed
The statistical results of the seven experiments in the prediction of “hard vs soft” classes show a progressive improvement in the area under the curve (AUC) value with the use of combined sets of explanatory variables (Fig. 6). Thus, the lowest ARC value of 0.92 resulted from the experiment that used only the backscatter angular response curves (BARCs), whereas the highest AUC value of > 0.99 resulted from the combination of bathymetry and backscatter information (BATHYs–BARCs and
Summary and discussion
This study explored the full potential of high-resolution multibeam data for the automated and accurate mapping of complex seabeds by using a large number of derivatives of bathymetry and backscatter data, and new features obtained from backscatter angular response curves. Despite the complex spatial distribution of hard substrata in the study area we achieved a nearly perfect prediction of “hard vs soft” classification with an AUC close to 1.0 (Fig. 5). The predictions were also satisfactory
Conclusions
This study demonstrates the advantages of using high-resolution multibeam sonar data and derivatives for predictive modelling of seabed substrate types across large complex areas. For a given study area, the predictive modelling approach using the Random Forest Decision Tree methodology can also highlight the underlying relationships between sediment grain size properties and the backscatter and bathymetry data.
In summary, the major findings of this study are:
- •
Co-registered bathymetry and
Acknowledgements
This work was undertaken for the Marine Biodiversity Hub, funded through the Commonwealth Environment Research Facilities (CERF) programme, an Australian Government initiative supporting world class, public good research. We thank two anonymous reviewers for their constructive reviews. The useful comments from Kim Pichard and Wenping Jiang from Geoscience Australia are also acknowledged. Published with permission of the Chief Executive Officer, Geoscience Australia.
References (72)
- et al.
Multi-resolution, object-oriented fuzzy analysis of remote sensing data for GIS-ready information
ISPRS Journal of Photogrammetry and Remote Sensing
(2004) - et al.
Small-scale mapping of sea-bed assemblages in the eastern English channel using sidescan sonar and remote sampling techniques
Estuarine, Coastal and Shelf Science
(2002) - et al.
Benthic habitat mapping: a review of progress towards improved understanding of the spatial ecology of the seafloor using acoustic techniques
Estuarine, Coastal and Shelf Science
(2011) - et al.
Use of acoustic classification of sidescan sonar data for mapping benthic habitat in the Northern Channel Islands, California
Continental Shelf Research
(2002) - et al.
Estimating the foliar biochemical concentration of leaves with reflectance spectrometry — testing the Kolaly and Clark methodologies
Remote Sensing of Environment
(2001) - et al.
Relationships between multibeam backscatter, sediment grain size and Posidonia oceanica seagrass distribution
Continental Shelf Research
(2010) - et al.
The effects of fine-scale surface roughness and grain size on 300 kHz multibeam backscatter intensity in sandy marine sedimentary environments
Marine Geology
(2006) - et al.
Angular range analysis of acoustic themes from Stanton Banks Ireland: a link between visual interpretation and multibeam echosounder angular signatures
Applied Acoustics
(2009) - et al.
Detailed investigation of continental shelf morphology using a high-resolution swath sonar survey: the Eel margin, northern California
Marine Geology
(1999) - et al.
Seabed characterization on the New Jersey middle and outer shelf: correlatability and spatial variability of seafloor sediment properties
Marine Geology
(2004)
Distribution of benthic macrofaunal communities in the western Baltic Sea with regard to near-bottom environmental parameters. 2. Modelling and prediction
Journal of Marine Systems
Acoustic seabed segmentation from direct statistical clustering of entire multibeam sonar backscatter curves
Continental Shelf Research
Modelling distribution of marine benthos from hydroacoustics and underwater video
Continental Shelf Research
Estimating foliage nitrogen concentration from HYMAP data using continuum removal analysis
Remote Sensing of Environment
A new approach to mapping marine benthic habitats using physical environmental data
Continental Shelf Research
Predictive mapping of seabed cover types using angular response curves of multibeam backscatter data: testing different feature analysis approaches
Continental Shelf Research
Seafloor habitat mapping of the New York Bight incorporating sidescan sonar data
Estuarine, Coastal and Shelf Science
Evaluation of image-based multibeam sonar backscatter classification for benthic habitat discrimination and mapping at Stanton Banks, UK
Estuarine, Coastal and Shelf Science
Detection of deep water benthic macroalgae using image-based classification techniques on multibeam backscatter at Cashes Ledge, Gulf of Maine, USA
Estuarine, Coastal and Shelf Science
A multi-method approach for benthic habitat mapping of shallow coastal areas with high-resolution multibeam data
Continental Shelf Research
Classification of seafloor geology using multibeam sonar data from the Scotian Shelf
Marine Geology
Shelf habitat distribution as a legacy of Late Quaternary marine transgressions: a case study from a tropical carbonate province
Continental Shelf Research
Oceanographic processes and the preservation of sedimentary structure in Eckernforde Bay, Baltic Sea
Continental Shelf Research
Automated acoustic seabed classification of multibeam images of Stanton Banks
Applied Acoustics
A bottom-up methodology for integrating underwater video and acoustic mapping for seafloor substrate classification
Continental Shelf Research
Mapping reef features from multibeam sonar data using multiscale morphometric analysis
Marine Geology
Geometric and radiometric correction of multibeam backscatter derived from Reson 8101 systems
Random forests
Machine Learning
Carnarvon Shelf Survey Post-Survey Report
Predicting seafloor facies from multibeam bathymetry and backscatter data
Photogrammetric Engineering & Remote Sensing
Acoustic backscatter and sediment textural properties of inner shelf sands, northeastern Gulf of Mexico
Geo-Marine Letters
Seafloor acoustic remote sensing with multibeam echo-sounders and bathymetric sidescan sonar systems
Marine Geophysical Researches
Rugosity-based regional modeling of hard-bottom habitat
Marine Ecology Progress Series
An ArcGIS seabed characterization toolbox developed for investigating benthic habitats
Marine Geodesy
A review of methods for the assessment of prediction errors in conservation presence/absence models
Environmental Conservation
Petrology of Sedimentary Rocks
Cited by (52)
DNN-based seabed classification using differently weighted MBES multifeatures
2021, Marine GeologySeafloor habitat mapping using multibeam bathymetric and backscatter intensity multi-features SVM classification framework
2021, Applied AcousticsCitation Excerpt :A series of shallow submarine sonar detection equipment such as multi-beam, side-scan and sub-bottom profiler sonar system can acquire multi-source and multi-dimensional seafloor acoustic information, and the significant progress they have made allows the habitat maps to be displayed with more detail and spatial accuracy [46]. In particular, multi-beam sounding system (MBES), as the mainstream detection device, can acquire backscatter intensity data that characterize the sediment properties while acquiring bathymetry data [5,47]. Depth and bathymetry-derived features such as slope, aspect and roughness, together with backscatter intensity, form a valuable set of information for mapping seafloor substrate, habitats, and biota [9].
Deep learning model for seabed sediment classification based on fuzzy ranking feature optimization
2021, Marine GeologyCitation Excerpt :In addition, this technique is tolerant of noise and has linear computation complexity, which provides credible theoretical support and solutions for the selection and optimization of high-dimensional feature spaces. To date, the supervised classification methods commonly used in seabed sediment prediction mainly include neural networks (Marsh and Brown, 2009; Huang et al., 2012; Stephens and Diesing, 2015), support vector machines (Che et al., 2012; Huang et al., 2012; Stephens and Diesing, 2015), random forests (Lucieer et al., 2013; Diesing et al., 2014; Huang et al., 2014; Stephens and Diesing, 2015), the maximum likelihood method (Buhl-Mortensen et al., 2009; Ierodiaconou et al., 2011; Che et al., 2012) and decision trees (Simons and Snellen, 2009; Stephens and Diesing, 2015). As the core step of seabed sediment classification, its purpose is to establish a function-mapping model between survey data and output categories (Michaels, 2007).
Characterisation of seafloor substrate using advanced processing of multibeam bathymetry, backscatter, and sidescan sonar in Table Bay, South Africa
2020, Marine GeologyCitation Excerpt :Multibeam echosounder backscatter data were used in conjunction to segment the seafloor into different classes; this information was then ground-truthed (using sediment samples and towed camera video logs of the seafloor) in order for each identified class to be differentiated as separate seafloor types (Innangi et al., 2015). Studies have shown that acoustic backscatter strength correlates with sediment mean grain size (e.g. Goff et al., 2004; Ferrini and Flood, 2006; Sutherland et al., 2007; De Falco et al., 2010; Zhi et al., 2014). According to Mazel (1985), the sonar properties directly result from the composition of the seafloor.