You are on page 1of 8

ANALYSIS OF SINGLE BEAM, MULTIBEAM AND SIDESCAN

SONAR DATA FOR BENTHIC HABITAT CLASSIFICATION IN THE


SOUTHERN BALTIC SEA

Jaroslaw Tegowskia, Natalia Gorskab,a, Aleksandra Krussb, Jaroslaw Nowakc,


Piotr Blenskia
a

Institute of Oceanography, University of Gdansk, al. Marszalka Pilsudskiego 46, 81-378


Gdynia, Poland,
b
Institute of Oceanology Polish Academy of Science, Powstancow Warszawy 55, 81-712
Sopot, Poland,
c
Maritime Institute in Gdansk, Dlugi Targ 41/42, 80-830 Gdansk, Poland

Jaroslaw Tegowski, Institute of Oceanography, University of Gdansk, al. Marszalka


Pilsudskiego 46, 81-378 Gdynia, Poland,
tel: (+48 58) 523 68 82, fax: (+48 58) 523 66 78, e-mail: tegowski@iopan.gda.pl
Abstract: Benthic habitat characterization is important for the study and conservation of the
biodiversity of the Baltic ecosystem. The main objective of this paper is the development of
complementary acoustic techniques for monitoring of Baltic benthic habitats. The study area
was located in the southern Baltic Sea and characterised by a considerable diversity of
geomorphologic forms and benthic assemblages. The simultaneous registration of the
acoustical data was conducted with two single-beam echosounders working at different
frequencies, a multibeam echosounder and a sidescan sonar. The high resolution multibeam
data were used to estimate seabed corrugation, a crucial feature for bottom surface
characterization. To identify morphological forms and benthic habitats, a parametric
approach was applied to the multibeam data. Firstly, spectral, wavelet, and fractal
parameters were computed in windows sliding along the separated bathymetric transects.
The vectors of computed parameters were then used as an input into Principal Component
Analysis and subsequently to fuzzy C-means clustering classification system. Moreover,
angular dependency of the backscattering intensity was analysed. Also the information from
single beam echosounders and sidescan sonar was utilised. The classification algorithms
were validated with video records and biological sampling.
Keywords: benthic habitats, multibeam echosounder, classification algorithms

1. INTRODUCTION
The acoustical maps, containing the information on the shape and geological nature of the
seabed itself and the benthic marine organisms present, represent an essential tool for the
conservation and management of the seafloor of the Polish Exclusive Economic Zone within
the Baltic Sea and allow to predict accurately the impact of anthropogenic activities on the
habitats. Taking it into account, the sets of acoustical backscattering data was collected by
complementarily used different acoustical tools as multibeam echosounder, sidescan sonar,
sub-bottom profiling system and single beam echosounders. Special attention was focused on
the narrow euphotic zone of the depth between 4-20m elongated parallel to the Polish cost
and containing different forms of benthic habitats. The total length of the surveyed area was
about 220 km and of a width slightly above 1 km. For the habitat mapping purpose a special
test polygon was chosen. The surveyed test site featured a diversified seafloor
geomorphologic forms and associated habitats and was located 1.2 km NE of the Rowy
harbour (see Fig.1.).

a)

b)
Fig.1: The study area - measurement polygon located 1.2 km NE of the Rowy harbour
(f=5440'02N, l= 1703'10E).

In the Polish marine areas, bottom covered with boulders and pebbles is rare, and therefore
the area of boulder field located near the Rowy harbor stands out against the practically bare
of benthic fauna and vegetations Polish inshore zone. The high biological values of this area
are undoubtedly a great impact on the varied morphology of abrasion platform with boulders
and pebbles scattered over the surface of the bottom, allowing attachment of organisms [1].
Moreover, the small depth of the area provides favorable light conditions for plants and
indispensable conditions for photosynthesis.
The acoustical measurements were accompanied by biological and geological sampling
and video inspection. The classification methodology were concentrated on the multibeam
echosounder data, which have been developed in two directions. The first classical method
utilised the shape parameters of the angular dependency of the backscattering intensity, when
the second method used the shape parameters of the bathymetric transects computed in a
sliding window. The acoustical data were integrated with the collected biological and
geological data in order to verify the developed classification algorithms.

1.1. MORPHOLOGY OF SEAFLOOR TEST AREA


The polygon is located in the inshore area within the zone of bottom relief having
polygenetic origin with relicts of periglacial forms together with contemporary forms of
marine origin. The polygon depth varies from ~4m up to ~14m. The right map in Fig.1.
shows the investigated area with depicted MBES imagery of measurement polygon.
The bottom surface is rough and varied with clearly formed embankment of ~300m in
width stretching in the NW-SE direction. The highest part of the embankment at depths of
about 4-7m, is a abrasive platform, with many young inselbergs and stony gravely abrasive
pavement on the bottom surface. The embankment slopes are furrowed with numerous,
relatively broad erosion gorges. The south slope is relatively short (up to 50m) and adjoins to
the shore slope, which arose as a result of sand accumulation. The north slope 100m long at a
depth of approximately 15m becomes a nearly flat surface of accumulated marine sands and
fine-grained sands and muddy sands. This is an area of relict moraine embankment
occurrence, made of till covered in stony gravely abrasive pavement and numerous relict and
contemporary erosion gorges with surfaces covered with contemporary accumulative marine
sands.
2. EXPERIMENT METODOLOGY
The measurements were conducted from the r/v IMOR equipped with precise
navigational instrumentation, multibeam echosounder Reson 8125 (455kHz, range: 0,5m 120m, no. of beams: 248, scan width: 120 , beam width: 0.5), chirp dual frequency sidescan
sonar EdgeTech 4200SP (300/600 kHz), single beam echosounders: Simrad EK-500 (120
kHz) and BioSonics DT-X (420 kHz). The USBL underwater positioning system was used to
calibrate measurements, where the acoustical signals were backscatter from biologically
recognized areas. Divers collected biological samples from the eight areas limited by frames
(1x1m) and made video recordings of benthic habitats. Figure 2 shows the metallic frame
surrounding area before pick up of biological samples (left photo) and after sampling (right
picture). Based on information from divers, video and photographic

Fig.2: The metallic frame surrounding biological measurement area before samples
picking (left photo) and after samples collection (right photo). Transducer of the USBL
positioning system is visible attached to the frame.

documentation, laboratory analysis of benthic material and information from the literature
[1], the characteristics of the individual stations were extracted.
It should be noted, that for the purpose of resolution enhancement, the number of
acoustical transects exceeds the number which assures the needs of polygon area coverage.
For that reason the spatial resolution of the bathymetric map obtained from multibeam
echosounder measurements reaches 0.05m.
3. DATA PROCESSING
The MBES data processing delivers segmented maps of different geomorphologic and
associated habitable areas. For this targets realisation were made two classification
algorithms based on different ideas the parameterisation of the shape of angular
dependency of backscattering intensity and the second method the parameterisation of the
high resolution bathymetric transects.
3.1. PARAMETRICAL ANALYSIS OF BOTTOM BACKSCATTER INTENSITY
The idea of MBSE seafloor classification based on angular dependency of the
backscattered intensity features is known in several classification systems [e.g. 2,3,4,5]. In
contrast to these systems, the method presented here utilises only the shape parameters of the
backscattering intensity computed for the separated two sides of returning MBSE signals. For
each backscattering intensity function were computed 26 spectral, fractal, and wavelet
transformation parameters.
The normalized power spectrum of backscattering intensity angular dependency in
logarithmic form is defined as [6]:
A S
1 log10 A 1 ,
C f log10

max

(1)

where A=105 const., S() - power spectral density function and its maximum value
Smax(). The classification parameters were defined as the relationships between parts of
spectral density functions:
f Ny

S f1

df ,

S fm

1
S f1

1
f Ny
m

df ,

(2)

where m=2, 4, 8, 16 and fNy is the Nyquist frequency.


The spectral moments of the r-th order are very sensitive for signal shape variation are
defined as:

mr r S d ,
0

(3)

where S is the Fourier power spectral density with moment order of r=0, 1,..,7. The
spectral widths 2, 2 and spectral skewness are defined as [7]:

m0 m4 m22
,
m0 m 4

~
m

m0 m 2
- 1,
m12

~ 332 .
m

(4)

The other parameter based on power spectral density function is the fractal dimension
computed from the spectrum slope and defined by Mandelbrot [8] as DFFT= (5-)/2.
The next group of very useful parameters in data segmentation process are wavelet
transformation coefficients and related wavelet energies. For the backscattering intensity
signal, the wavelet energy content was computed using the 7-channel dyadic decomposition
(scale a=2j, j=1,..,7) with a 3th-order Coiflet wavelet:

E j ,Coif 3

bmax

C a, b db ,
2

(6)

bmin

where C(a,b) is the wavelet transformation coefficient, bmin and bmax are boundary values of
scale b (time). The energy distribution indicator - entropy hCoif3 is defined as [9]:
7

hCoif 3 E j ,Coif 3 lnE j ,Coif 3 ,

(7)

j 1

The technique of Hurst exponent determination via the averaged wavelet coefficient
method [10, 7] was used in this calculation. The Hurst exponent H and subsequently the
Hausdorff dimension is equal DH,Daub7=2-H. The above defined parameters formed 26element vectors for either side of backscattering intensity. For elimination of the excessive
fluctuation of parameters values the moving average procedure was used. The strong
correlation between some of the parameters required the elimination of redundant
information. A Principal Component Analysis (PCA) has been applied to the data for the
removal of this redundancy. The number of chosen Principal Components is determined by
their summed variations. In successive calculations we used seven firsts PCs, which ensured
above 96% (96.40%) of cumulative variation and resulted in a loss of less than 4% of
information (Fig.3.b).

a)

b)

Fig.3: PCA plot containing four clusters indicating separate features of seafloor sound
reflectivity a), summed variations of first seven PCAs b).

In the next step, the Calinski-Harabasz index [11] was calculated to determine the number
of clusters centers needed for the classification procedures. The Principal Components were
the input to fuzzy c-means (FCM) data clustering algorithm [12]. The example of the result of
this algorithm product is presented below (Fig.4).

Fig.4: Example of result of segmentation using shape parameters of the angular


dependency of backscattering intensity.

3.1. PARAMETRICAL ANALYSIS OF THE SHAPE OF BATHYMETRIC


TRANSECTS
The second classification method utilises the information included in the shape of bottom
surface. From the high resolution bathymetric 3D map of tested polygon (the white rectangle
area in Fig 1.b) 150 vertical and 150 horizontal parallel bathymetry cross- sections were
extracted. An example of one bathymetric vertical cross-section taken in the middle part of
investigated area is presented in Fig.5. Such cross-section was the object of high-pass
filtration procedure necessary for elimination of the depth level dependency on the
parameters values.

Fig.5: Example of one bathymetric vertical cross-section taken in the middle part of
investigated area (white rectangle in Fig.1.a).
For each consecutive cross-section the shape parameters in sliding window were
computed. There were 26 spectral, fractal and wavelet transformation parameters defined in
section 3.1. The spatial resolution of such a parameterised bathymetric map were depended
on sliding window width (256, 512 or 1024 samples) and the distance between consecutive
cross-sections. The segmentation procedure was almost the same as method presented in
section 3.1. The set of parameters were object of the PCA process. After the choice of first 6

Principal Components and the computation of the Calinski-Harabasz index [11], Principal
Components were input to the FCM [12] classification algorithm. Fig. 6 presents a
comparative set of a bathymetric map and segmented bottom imageries.

Fig.6: Comparative set of a bathymetric map (a) and segmented bottom images for 3 clusters
(b) four clusters (c) and five clusters (d).
4.CONCLUSIONS
The both MBES bottom imagery segmentation schemes presented in this study have
many promising features which allow them to be applied for extracting morphological forms
of seabed and habitats. The first method based on angular dependency of the backscattered
intensity delivers information about the reflectivity of the measured area, when the second
method is strongly associated with the morphology of the investigated polygon. Both
techniques precisely indicated areas of relicts of periglacial forms as well as contemporary
forms of marine origin. The results of sidescan sonar bottom imageries, echograms made
using single beam echosounders, sedimentological and biological sampling, and video frames
analyses, confirm precision and effectiveness of both supplementary segmentation systems.
The benthic flora and fauna settled in bottom geomorphologic forms create separated habitats
detectable by both systems. The correctness of the method was verified by the results of
underwater video recordings, single beam echosounder registrations and biological samples
taken in situ.

4. ACKNOWLEDGEMENTS
This work was supported by the Ministry of Science and Higher Education of Poland
(research project no. N306296933).
REFERENCES
[1] Osowiecki A., Kruk-Dowgiao L., Biodiversity in the coastal pebbles field Rowy next to
Slowiski National Park, Edit. Maritime Institute in Gdansk, 127, (in Polish), 2006.
[2] Clarke, J. H., Toward remote seafloor classification using the angular response of
acoustic backscattering: A case study from multiple overlapping GLORIA data. IEEE
Journal of Oceanic Engineering, 19 (1), 112-127, 1994.
[3] Chakraborty, B.; Kodagali, V.; Baracho, J., Sea-floor classification using multibeam
echo-sounding angular backscatter data: a real-time approach employing hybrid neural
network architecture, IEEE Journal of Oceanic Engineering, 28 (1), 121128, 2003.
[4] Hughes-Clarke, J.E., Danforth, B.W., Valentine, P., Areal seabed classification using
backscatter angular response at 95 kHz , Proceedings of SACLANT Conference -High
Frequency Acoustics in Shallow Water, June 1997, Lerici, Italy, SACLANT CP-45, 243250, 1997.
[5] Lurton X., Augustin J-M., Dugelay S., Hellequin L., Voisset M., Shallow-water
seafloor characterization for high-frequency multibeam echosounder: image segmentation
using angular backscatter, in High Frequency Shallow Water Acoustics (Pace ed.),
Saclantec Conference Proceedings CP-45, 313-321, 1997.
[6] Pace, N.G., Gao, H., Swathe Seabed Classification, IEEE Jour. Oc. Eng., 13 (2), 83-90,
1988.
[7] Tegowski J., Gorska N., Klusek Z., Statistical analysis of acoustic echoes from
underwater meadows in the eutrophic Puck Bay (southern Baltic Sea), Aquat. Living
Resour. 16, 215-221, 2003.
[8] Mandelbrot, B.B., The fractal geometry of nature, Freeman, San Francisco,1982..
[9] Faghani D., Tegowski J., Gorska N., Klusek Z., Recognition Of Underwater
Vegetation Species In The Baltic Sea , Proceedings of the seventh European Conference
on Underwater Acoustics, ECUA 2004, Delft, The Netherlands, 373-378, 2004.
[10]
Simonsen, I., Hansen, A., Nes O. M., Determination of the Hurst exponent by use
of wavelet transforms, Phys. Review, 58, (3), 2779-2787, 1998.
[11]
Calinski, T., Harabasz, J., A., Dendrite method for cluster analysis,
Communications in Statistics, 3, 1-27, 1974.
[12]
Bezdeck, J.C, Ehrlich, R., Full, W., FCM: Fuzzy CMeans Algorithm, Computers
and Geoscience, 10 (2-3), 191-203, 1984.

You might also like