Journal Search Engine

ISSN : 2671-9940(Print)
ISSN : 2671-9924(Online)

Journal of the Korean Society of Fisheries and Ocean Technology Vol.55 No.1 pp.39-49
DOI : https://doi.org/10.3796/KSFOT.2019.55.1.039

A preliminary study on seabed classification using a scientific echosounder

Rina FAJARYANTI

, Myounghee KANG^*

Professor, Department of Maritime Police and Production System/Institute of Marine Industry, Gyeongsang National University, Tongyeong 53064, Korea

Corresponding author: mk@gnu.ac.kr, Tel: +82-55-772-9187, Fax: +82-55-772-9189

Received 20181231 Review 20190213 Accepted 20190214

Abstract

Acoustics are increasingly regarded as a remote-sensing tool that provides the basis for classifying and mapping ocean resources including seabed classification. It has long been understood that details about the character of the seabed (roughness, sediment type, grain-size distribution, porosity, and material density) are embedded in the acoustical echoes from the seabed. This study developed a sophisticated yet easy-to-use technique to discriminate seabed characteristics using a split beam echosounder. Acoustic survey was conducted in Tongyeong waters, South Korea in June 2018, and the verification of acoustic seabed classification was made by the Van Veen grab sampler. The acoustic scattering signals extracted the seabed hardness and roughness components as well as various seabed features. The seabed features were selected using the principal component analysis, and the seabed classification was performed by the K-means clustering. As a result, three seabed types such as sand, mud, and shell were discriminated. This preliminary study presented feasible application of a sounder to classify the seabed substrates. It can be further developed for characterizing marine habitats on a variety of spatial scales and studying the ecological characteristic of fishes near the habitats.

Key Words : Seabed classification , Echosounder , Roughness , Hardness , Clustering

초록

키워드 :

This article has been cited by 0 article in crossref

Cited-By

Funding:

National Research Foundation of Korea
2018R1A2B6005666

Introduction

Comprehensive data and information of marine biological resources and their environments are required in order to support sustainable and effective management marine resources. Information of substrate distribution on the seabed is important since seabed is useful as a habitat for aquatic organisms (Fauziyah et al., 2018). Seabed mapping is a proxy to generate habitat mapping. The most commonly used method for determining the substrate type of seabed is grab sampling or coring from a stationary vessel and followed by laboratory analysis. Nevertheless, this method not only fails to obtain undisturbed samples but also extremely slow, expensive, and provides information only at discrete sites. In recent years, considerable effort has been expended on the remote classification of seabed substrates using acoustic tools such as side-scan sonar, underwater video, sub-bottom profilers, and echosounder (Freitas et al., 2003).

Scientific echosounders have been widely used in fish biomass assessment and distributional characteristics of aquatic organisms, and have been applied to marine biological resource mapping. Thus, acoustic seabed classification is an active field of research in various countries, where different methods have been proposed for describing seabed properties. Acoustic seabed classification uses single beam echosounders to obtain information on seabed acoustic hardness (acoustic reflection coefficient) and acoustic roughness (as a backscatter coefficient). Simons et al. (2007) analyzed backscattering strength measurement on seabed and geotechnical parameters, Simons and Snellen (2009) classified the seabed using the backscatter data, Good et al. (2009) examined the effect of the relationship of backscattering values on the composition of substrate particles in New Jersey, and Katsnelson et al. (2015) focused on sound propagation at low frequencies for substrate characteristics in Kinneret Lake. The shape of the seabed echo depends largely on the roughness and hardness of seabed which affects the backscattering strength (Lied et al., 2004). Remote acoustic technique provides a sophisticated technique to discriminate seabed characteristics and to map their distribution at high spatial and temporal resolution (Wienberg and Bartholomä, 2005). In addition, this method is less expensive and less time consuming (Hamilton, 2011).

The study of seabed classification in South Korea involved Chang et al. (1998) using chirp sonar system data, Kim et al. (2002) determined the substrate type using side-scan sonar in Suyoung Bay and Lee et al. (2009) classified the seabed using sub-bottom profiler. The side-scan sonar is a common approach for high resolution seabed mapping due to its capability to cover larger areas at once. Limitation of this system is the dimensions of the resulting image can be very huge (thousands of pixels in width and height respectively) and depend on processed resolution of the recorded data, the size, and the shape of the investigation area (Berthold et al., 2017). Further, the disadvantage is that the measurements are not made with the same insonification geometry (Collier and Brown, 2005). Meanwhile, the single beam echosounder is very commonly used by a number of fisheries scientists. This means that the sounder is installed in not only many research vessels but also training vessels. The single beam echosounder is also used as a portable system in rivers, lakes, and relatively shallow waters. It would be very beneficial for the single beam sounder to be used for researching marine organisms as well as seabed classification. In addition, the single beam echosounder includes the split beam sounder.

In this context, it is necessary to develop the study on seabed classification technique using backscatter data from a scientific echosounder (so called the split beam sounder). Accordingly, this study aims to demonstrate a sophisticated yet easy-to-use technique to discriminate seabed characteristic using a split beam echosounder. This study presents a detail method to determine seabed backscattering strength and to classify seabed type from echosounder records obtained during a survey in Tongyeong waters, South Korea in June 2018. Verification of a particular acoustic seabed classification system is made by ground truthing in the study area.

Material and Methods

The field survey was conducted on June 20, 2018 (from 10:00 am to 17:30 pm) with the ground truth of Van Veen grab sampling at 3 stations in Tongyeong waters, South Korea using the R/V Charmbada (36 tons). Acoustic data were recorded using a scientific split-beam echosounder (Simrad EK60, 120 kHz) and a GPS (Garmin Map 62s) was used for positioning. The specification of transducer is shown in Table 1. The study area including both acoustic track and grab sample points is described in Fig. 1.

Acoustic data analysis

Echo shapes and energies depend on seabed features (Hamilton, 2011). Echoview software (Echoview, Pty. Ltd., ver.9) was used to analyze acoustic data and to export raw acoustic backscatter data into ASCII formats for backscatter analysis. Noise could be generated by the transmitted sound backscattered (surface noise) and background noise. In order to increase signal-to-noise ratio (SNR), surface noise should be excluded from analysis by determining the depth of surface noise (4 m). Background noise level was also removed by estimating the noise and subtracting it. Seabed line was identified using best bottom candidate algorithm and a 0.0 m back-step was used. Within Sv echogram, the seabed line was manually edited to ensure that the line was continuous from the beginning to the end without containing any signals from marine features. The seabed line was important because it delineated the seabed echoes from water column echoes. The seabed line and the echoes below the line were where Echoview applied the seabed classification algorithm.

Depth normalization

Seabed echoes encoded time and energy information for one transmitted pulse incident on the seabed substrate. For a normal incidence the transmitted pulse travels a distance of c_τ/2. However, at the off-axis part of the beam the same pulse travels a distance greater than c_τ/2. As a result, the seabed echo extends over this effective pulse length (Penrose et al., 2005). Some seabed features are depth normalized to account for variation in pulse length with depth. Refer Echoview (2018) for the geometry of the beam showing the off-axis angle offset distance and the differences in pulse length. Thus, the expression for the total distance that the off-axis part of beam needs to travel one whole pulse length in the substrate is:

Effective pulse length = [d_{0} + \frac{c τ}{2}] - d_{0} cos (\frac{θ}{2})

(1)

Where c is sound speed of water (m/s), τ is pulse duration of transducer (s), θ is major axis ‒3 dB beam angle of transducer, d₀ is normal incidence start depth (m) of the first echo at a ping P. The depth normalized coefficient is as follows.

Depth normalized coefficient = (\frac{off axis pulse length (ref)}{off axis pulse length (actual)}) actual​ value

(2)

Where, off-axis pulse length (ref) is off-axis pulse length where the normal incidence start depth is specified by the reference depth, off-axis pulse length (actual) is the off-axis pulse length of the first echo where the normal incidence start depth is given by actual depth for the ping. Echoview determined one seabed point per feature extraction interval. The seabed point data includes time, latitude, longitude, depth, and seabed features. Each seabed features is a mean of the measured ping characteristics in the feature interval (Anderson, 2007;Echoview, 2018). It specified a reference depth to normalize pulse length. The best value for the reference depth was the average depth of the seabed and in this study 20 m was used.

Seabed roughness (E1) and seabed hardness (E2)

Two key factors in seabed classification are seabed roughness (E1) and seabed hardness (E2) parameters. Energy features from the first acoustic seabed returns could be related to roughness parameters. For an incremental area dA1 far from the axis, the first backscatter return becomes incoherent. Total backscatter return is a superposition of all backscatter signals from all areas. Following Heald and Pace (1996), the received acoustic pressure may be expressed as:

P_{b s 1}^{2} = P o^{2} \int_{θ a}^{θ b} \frac{m_{s} (θ_{1}) G^{2} (θ_{1})}{{(R^{'})}^{4}} d A 1

(3)

Where Po² is the source pressure at distance 1 m from the source, $d A 1 = 2 π {R^{'}}^{2} tan θ_{1} d θ_{1}, G (θ_{1})$ is transducer gain, $R^{'} = R \sqrt{(1 + {tan}^{2} θ_{1})}, and m_{s} (θ_{1})$ is the acoustic scattering coefficient. The start point for calculation seabed roughness was given by seabed line depth plus the distance of c_τ/2 plus the off-axis angle offset. The second seabed echo started at twice depth of first seabed echo and the whole second seabed echo is used for calculation of seabed hardness (Anderson et al., 2007). Assuming that the total acoustic pressure reflection coefficient was the best descriptor of seabed hardness and that the second seabed echo reflected up and down twice from the surface is proportional to the 4^th power (rather than the 2^nd) of the acoustic pressure coefficient, integration of the whole second echo could be used to provide an estimate of seabed hardness (Rodríguez-Pérez et al., 2014). Acoustically different seabed types can be discriminated by clustering the backscatter signals by these two parameters E1 and E2. Siwabessy et al. (1999) determined E1 and E2 values using the equation below:

E = 4 π {(1852)}^{2} \frac{\sum_{i = 1}^{n} (\sum_{k = l}^{m} s υ_{(i, k)})}{n}

(4)

Where sv (i, k) is the linear value for Sv for sample k in ping i.

Key seabed features

The next processing stage was calculating the normal deviate for seabed features in the echogram data. Statistical features values were processed by the statistical procedure of Principal Component Analysis (PCA) followed by K-Means clustering. Basically, PCA aimed to compress or simplify datasets by reducing the number of dimensions without much loss of information. PCA was a linear transformation from the axes representing the original variables into a new set of axes called principal components (Pcs), such that greatest variance by projection of the dataset came to lie on the first axis (then called the first principal component), the second greatest variance on the second axis, and so on (Amiri-Simkooei et al., 2011). Under seabed classification, PCA determined principal component as cluster seabed features of dataset. In this study, PCA was used to figure out a relationship between acoustic data and physical substrate parameters thought to provide an overview of seabed characteristics. Echoview used cluster dimensions to characterize each seabed point in cluster dimension space. At this stage, seabed point data includes depth, first bottom skewness, second bottom length normalized, first bottom rise time normalized, bottom roughness normalized, and bottom hardness normalized.

Seabed classification

Seabed classification is the process of partitioning acoustic seabed returns into discrete classes for substrate and seabed types. The classification of the sampled values was performed using clustering analysis. Clustering could be interpreted as process of grouping objects that explains the relationship between objects to maximize the similarity of members in one class and to minimize similarities between classes/clusters. The well-known K-means clustering algorithm was used in this study. Generally, the K-means algorithm aimed to partition n observations into k clusters in which each observation belongs with the cluster with the nearest mean. The result was a set of clusters that were as compact and well-separated as possible (Legendre et al., 2002). The set of cluster was exported to generate a seabed classified map in ArcMap (Esri Inc., ver. 10.2.2).

Data validation

This is a preliminary study on seabed classification using a sounder. Thus, prior knowledge on seabed substrate types confirms the results of acoustic seabed classification. Acoustic classification results should be subjected to ground truthing, which consists in relating the acoustic classes to observed data describing the seabed. The obtained result of seabed classification using acoustic clustering would be validated with the substrate data carried out by grab samples.

Results

Seabed echogram and its property

The echogram of 120 kHz echosounder at the study area after removing background noise and other noises is shown in Fig. 2. The green line was the detected seabed line. First echo and second echo of seabed features could be examined in Fig. 2. The expanded echogram showed the detail features of seabed echo and indicated the different strength of acoustic signal. Sediment which tended to have macroscopic appearance would reflect signal stronger than the softer sediments. It was evidently referred that the sediment at 2^nd station (Fig. 2b) was softer than those were in 1^st station and 3^rd station (Fig. 2a and 2c).

Acoustic seabed classification was generated based on backscatter data (Fig. 3). It could be seen that the seabed material in the study area was mainly composed of sand. In the 1^st station there were 44.7% sand, 44.2% mud, and 11.1% shell, the 2^nd station composed mainly of 68.3% mud, 23.8% sand, and 7.9% shell. Then, in the 3^rd station were also mainly composed of 87.3% sand, 8.7% mud, and 3.9% shell.

Ground truth of acoustic seabed classification

Ground truthing was accomplished in one site point for each station at certain depth. The result of physical survey analysis to the seabed substrate using the Van Veen grab sampler on study area showed that sand dominated in 1^st station at average depth of 10.5 cm, mud dominated in the 2^nd station at average depth of 15.4 cm, and shell dominated the 3^rd station at average depth of 11.1 cm (Fig. 4).

The three acoustic classes were correlated well with the three predominant surface sediment types occurring in the study area (Fig. 3 and Fig. 4), annotated as sediment types sand, mud, and shell. The result of acoustic seabed classification of the whole data set was shown in Fig. 5. It was represented as color coded classified points along the survey line in a seabed classification map. The red code represented sand, the blue code as mud, and the green code as shell. It could be examined the distribution of sediment characteristic along the voyage track in each station.

Backscattering characteristic by seabed type

Backscattering strength could be extracted for further analyzing. The value of backscattering strength obtained as in Table 2 showed the values of E1 (seabed roughness), E2 (seabed hardness), Sv (seabed volume backscattering strength). Sand substrate had an E1 ranged from 6.16 to 7.67 with E2 ranged from 3.74 to 5.75 and Sv ranged from ‒21.79 to ‒4.59 dB. Shell substrate had an E1 ranged from 5.35 to 7.44 with E2 ranged from 3.09 to 5.19 and Sv ranged from ‒30.47 to ‒8.20. Mud substrate had an E1 ranged from 3.95 to 7.03 with E2 ranged from 2.90 to 4.25 and Sv ranged from ‒38.70 to ‒12.31 dB. The mean value of E1 and E2 for sand, shell, and mud in 3^rd station represented the highest value among the other stations. These phenomena occurred due to the seabed composition in 3^rd station that tended to have rougher material which was dominated by sand and had lowest composition of mud. The mean value of Sv for sand, mud and shell in 1^st station represented the highest value among the other station.

Discussion

Underwater acoustic instrument could rather easily measure echoes by reflection of sounding pulses from seabed. It was understood that backscatter echo from the seabed contained information on the characteristics of seabed materials such as mean grain size, seafloor roughness spectrum parameters, sediment volume scattering parameter, density of sediments, sound speed, and so forth (Holliday, 2007). Rough substrate in seabed (stone or coral) would produce high intensity reflection E1 and E2, otherwise the fine substrate would produce low reflection (Hamilton, 2011). Generally, the value of E1 was higher than E2. The first acoustic bottom return was reflected directly from the seabed and the second acoustic bottom return is reflected twice off of the seabed and once off of the sea surface, sub surface bubbles and vessel hull (Anderson, 2007). In addition, the presence of gas bubbles in the water was likely to be a critical source of variability in acoustic bottom return from seabed, particularly the value of E2 (Briggs et al., 2002). The water bubbles would be expected also to change volume, hence acoustic cross section and subject to changes in temperature (MacDonald et al., 2005).

The obtained backscattering strength in this study area showed substrate type of sand had the greatest roughness, hardness, and backscattering strength instead of shell and mud. Sand and shell had much rougher surface than mud. The rough and fine seabed gave an influence to the intensity of the return reflection where the roughness seabed had bigger reflection rather than fine seabed. On the rougher surface, the less energy of acoustic waves would be reflected at the specular angle and the more energy would be scattered in the other directions. This scattering feature was clearly seen in the results of the backscatter analysis for sand, shell, and mud. Both of sand and shell had macroscopic appearance than mud. Shell components in particular could cause unpredictable returns. Evidently, any classification of sediment type based on grain-size data alone was somewhat problematic in this case, because it is known that coarse shell material is a strong and characteristic acoustic reflector. Accordingly, the value of E1 and E2 of shell was slightly higher than sand, however that was not relevant within this study. The backscattering strength of shell was lower than sand, but the difference is not large. Some points of shell above a sandy seabed might fit within any of both classes. Seabed surface roughness also appears to be important in this context. The number of station for ground truth as well as the size of acoustic data for seabed classification should be increased in the next survey. Accordingly, the results can become to be generalized to present the characteristic of seabed substrates.

The spatial distributions of the acoustic classes and main sediment types indicate a close correspondence between the acoustic patterns and the sedimentary assemblages based on ground truthing. The distinct of sediment result between ground truthing and acoustic characterization in 3^rd station was evidently appeared. Sediment on 3^rd station based on ground truth was mainly composed by shell, yet based on acoustic characterization the sediment along 3^rd station was sand. Within the echogram, it could be seen that class result at specific time (16:00 pm) when collected sediment data in 3^rd station was shell. The sediment composition in 3rd station tended to be rougher because the presence of mud (softer sediment) was the lowest among those in 1^st station and 2^nd station. Accordingly, the echogram of 3^rd station (Fig. 2c) represented the more evident color as the stronger acoustic reflected signal. Based on seabed classification map, the sediment on 3^rd station was sand and there were some sample sites that showed the shell and mud sediment. In this case, the drawback of this study was the number of collected sediment data during ground truthing. At certain time, sediment data were only collected at one site point in each station and it was not sufficient to represent the sediment type along track of that station. Consequently, substrates associated with the site were rarely sampled. In the future and in other study areas, this should be ensured that all seabed types identified to be ground truthed more equally.

Conclusion

The acoustic technique presented becomes a reliable tool for seabed classification. Seabed features in study area based on acoustic classification compose of three classes (sand, mud, shell) with respect to ground truthing. The highest backscattering strength of the sediments along study area was sand, then followed by shell and mud. This preliminary study has demonstrated a possible way to use a scientific echosounder for seabed classification. In conjunction with ground-truth information collected with various seabed samplers and underwater imaging techniques it is now possible to derive highly classified seabed maps.

Acknowledgements

This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (No. NRF-2018R1A2B6005666). Authors thank crew members of the R/V Charmbada for supporting us to conduct the field survey and Dr. Gi-beum Kim for allowing us to use the grab sampler.

Figure

Fig. 1.Study area in Tongyeong waters, South Korea. Black circle is acoustic survey track, red triangle is for the Van veen grab point sampling. The number is for the station number.

Fig. 2.Echogram of (a) 1^st station, (b) 2^nd station, and (c) 3^rd station. The first and second seabed echoes in each station is evidently appeared. The expanded seabed features in red rectangle in the right indicating the different strength of acoustic signal in relation to the sediment at seabed. Time and water depth are seen.

Fig. 3.Sediment composition based on acoustic characterization.

Fig. 4.The substrate collected by the Van veen grab sampler, (a) sand in 1^st station, (b) mud in 2^nd station, and (c) shell in 3^rd station.

Fig. 5.Seabed classification map as the result of acoustic seabed classification in study area (a). The expanded map show the detail acoustic classes and precise location of grab sample in 1^st station (b), 2^nd station (c), and 3^rd station (d).

Table

Table 1.Specification of transducer

Table 2.Backscattering strength value of each substrate type in three stations

Reference

Amiri-SimkooeiAR , SnellenM and SimonsDG . 2011. Principal component analysis of single-beam echo-sounder signal features for seafloor classification . IEEE J Oceanic Eng36, 259-272. (DOI:)
AndersonJT . 2007. Acoustic seabed classification of marine physical and biological landscapes . ICES Coop Res Rep286, 1-183.
BertholdT , LeichterA , RosenhahnB , BerkhahnV and ValeriusJ . 2017. Seabed sediment classification of side-scan sonar data using convolutional neural networks. IEEE Sym Ser on Comp Int, 1-8.
BriggsKB , WilliamsKL , JacksonDR , JonesCD , IvakinAN , and OrsiTH . 2002. Fine-scale sedimentary structure: implications for acoustic remote sensing . Mar Geol182, 141-159. (DOI:)
ChangJK , KimHJ , JouHT , SukBC , ParkGT , YooHS and YangSJ . 1998. Seabed classification using the K-L (Karhunen-Love) transform of chirp acoustic profiling data: an effective approach to geoacoustic modeling . J Kor Soc Oce3(3), 158-164.
CollierJS and BrownCJ . 2005 Correlation of side-scan backscatter with grain size distribution of surficial seabed substrates . Mar Geol214, 431-449. (DOI:)
Echoview. 2018. Help file 9.0.18 for Echoview. Retrieved from http://support.echoview.com/WebHelp/Echoview.htm, Accessed 11 Nov 2018.
Fauziyah, Priatna A , Prakoso WF, Hidayat T, Surbakti H and Nurjuliasti E. 2018. Measurement and analysis of acoustic backscattering strength for characteristics of seafloor substrate in Indian Ocean WPP 572-573. IOP Conf Ser Earth Environ Sci 162, 12-24. (DOI:)
FreitasR , SilvaS , QuintinoV , RodriguesAM , RhynasK and CollinsWT . 2003. Acoustic seabed classification of marine habitats: studies in the western coastal-shelf area of Portugal . ICES J Mar Sci60, 599-608. (DOI:)
GoodJA , KraftBJ , MayerLA , SchockSG , SommerfieldCK , OlsonHC , GulickSPS and NordfjordS . 2009. Seabed characterization on the New Jersey middle and outer shelf; correlatability and spatial variability of seafloor substrate properties . Mar Geol209, 147-172. (DOI: )
HamiltonLJ . 2011. Acoustic seabed segmentation for echosounders through direct statistical clustering ofseabed echoes . Cont Shelf Res31, 2000-2011. (DOI:)
HealdGJ and PaceNG . 1996. An analysis of the 1st and 2nd backscatter for seabed classification. Proc. 3rd Eur Conf on Und Acoust 2, 649-654.
KatsnelsonB , KatsmanR , LunkovA and OstrovskyI . 2015. Lowfrequency sound propagation and substrate characterization in Lake Kinnerat . Proc Ins Acoust37(1), 102-109.
KimGY , KimDC , KimYE , LeeKH , ParkSC , ParkJW and SeoYK . 2002. Remote seabed classification based on characteristics of the acoustic response of echosounder: prelim result of Suyoung Bay . J Kor Fish Soc35(3), 273-281.
LeeGH , KimHJ , KimDC , YiBY , NamSM , KhimBK and LimMS . 2009. The acoustic diversity of the seabed based on the similarity index computed from chirp seismic data . ICES J Mar Sci66, 227-236. (DOI:)
LegendreP , EllingsenKE , BjørnbomE and CasgrainP . 2002. Acoustic seabed classification: improved statistical method . Can J Fish Aquat Sci59, 1085-1089. (DOI:)
LiedTT , WaldayM , OlsgardF , EllingsenKE and HolmS . 2004. SEABEC - a single beam echosounder, seabed classification system. MTS/IEEE Kobe/Techno-Ocean 4, 2024-2028.
MacDonaldIR , BenderLC , VardaroM , BernardB , and BrooksJM . 2005. Thermal and visual time-series at a seafloor gas hydrate deposit on the Gulf of Mexico slope . Earth Planet Sci Lett233, 45-59. (DOI: )
PenroseJD , SiwabessyPJW , GavrilovA , ParnumI , Hamilton LJ , BickersA , BrookeB , RyanDA and KennedyP . 2005. Acoustic techniques for seabed classification. TechRep 32. Coop Res Centre for Coastal Zone Est and Water Manag, 1-118.
Rodríguez-PérezD , Sánchez-CarnN and FreireJ . 2014. A pulse-length correction to improve energy-based seabedclassification in coastal areas . Cont Shelf Res.77, 1-13. (DOI: )
SimonsDG , SnellenM and AinslieMA . 2007. A multivariate correlation analysis of high frequency bottom backscattering strength measurements with geotechnical parameters . IEEE J Oceanic Eng32(3), 640-650. (DOI: )
SimonsDG and SnellenM . 2009. A Bayesian approach to seafloor classification using multibeam echosounder backscatter data . Appl Acoust70(10), 1258-1268. (DOI:)
SiwabessyJ , PenroseJ , KloserR and FoxD . 1999. Seabed habitat classification. Shallow Survey '99. Int Conf on High Res Surveys in Shallow Water, 1-81.
TegowskiJ. 2005. Acoustical classification of bottom substrates in southern Baltic Sea . Quatern Int130, 153-161. (DOI: )
WienbergC and BartholomäA . 2005. Acoustic seabed classification in a coastal environment (outer Weser Estuary, German Bight) - a new approach to monitordredging and dredge spoil disposal . Cont Shelf Res25, 1143-1156. (DOI:)