ICES Journal of
Marine Science
ICES Journal of Marine Science (2014), 71(4), 882– 894. doi:10.1093/icesjms/fst177
Original Article
Seabed classification using surface backscattering strength versus
acoustic frequency and incidence angle measured with vertical,
split-beam echosounders
George R. Cutter Jr* and David A. Demer
Fisheries Resources Division, Southwest Fisheries Science Center, National Marine Fisheries Service, National Oceanic and Atmospheric Administration,
8901 La Jolla Shores Drive, La Jolla, CA 92037, USA
*Corresponding Author: tel: +1 858 546 5691; fax: +1 858 334 2809; e-mail: [email protected]
Cutter, G. R. Jr, and Demer, D. A. 2014. Seabed classification using surface backscattering strength versus acoustic frequency and incidence angle
measured with vertical, split-beam echosounders. – ICES Journal of Marine Science, 71: 882 – 894.
Received 16 May 2013; accepted 21 September 2013; advance access publication 25 December 2013.
The multifrequency biplanar interferometric imaging technique (MBI) is applied to data from vertical, split-beam echosounders to produce
sub-beam estimates of seabed surface-backscattering strength (Ss), incidence angle (u), and roughness (R). A simple model is used to quantify the variation of Ss versus u ¼ {2 – 208} and acoustic frequency, f ¼ {18, 38, 70, 120 and 200 kHz}. The coefficients of the angle- and
frequency-dependent terms of the model indicate seabed material properties, principally small- and large-scale roughness and hardness.
These indices are combined with the estimates of u and R to classify the seabed using unsupervised cluster analysis. This technique is
applied to data from the Forty-Three-Fathom Bank, a seamount in the Southern California Bight. The resulting seabed classifications
are consistent with the surficial lithology and the spatial distribution of known rockfish (Sebastes spp.) habitat. The method should be
generally applicable to seabed classification.
Keywords: multifrequency biplanar interferometric imaging, normal incidence, rockfish, seabed backscatter, Sebastes, spectral, specular, surficial
geology.
Introduction
Seabed classification using single-frequency
normal-incidence seabed echoes
Acoustic remote sensing has long been used to efficiently and accurately map seabed bathymetry, e.g. for navigation. Increasingly, it is
also used classify the seabed for studies ranging from habitat characterization to oil exploration (see Anderson et al., 2008). A common
approach to acoustic classification of the seabed involves the probability density function (pdf) or other statistics of echo amplitude
from the first (Stanton, 1984; Sternlicht and DeMoustier, 2003)
or subsequent (Hines and Heald, 2001; Penrose et al., 2005;
Anderson et al., 2008, and references therein) echoes from the
seabed measured with a vertical echosounder. A common assumption is that the acoustic incidence angle (u ¼ 908 – grazing angle) is
zero, but this is rarely correct due to transducer motion and seabed
slope. The combination of sloped or rough seabed and ship and thus
transducer motion may produce a wide range of u (Demer et al.,
2009; Cutter and Demer, 2010). Therefore, considering the first
seabed reflection, the pdf of echo amplitudes from a range of u
intrinsically includes information about the combined effects of
seabed slope, roughness, and composition on the acoustic pulse
during its reflection from an area of the seabed principally defined
by the transducer-beam width and orientation and the seabed
depth and shape. The pdfs of echo amplitudes from subsequent
seabed reflections involve seabed scattering from a larger range of
u, and one or more reflections from the ocean surface whose dynamics modulate echo directions and add uncertainty to estimates of u.
Therefore, the results from these methods are generally interpretable
only relative to their collection instrument and conditions.
Seabed classification using backscatter versus acoustic
incidence angle and frequency
Measures of seabed surface-backscattering coefficient (ss), commonly reported in decibels as surface backscattering strength
Published by Oxford University Press on behalf of the International Council for the Exploration of the Sea 2013. This work is written by (a)
US Government employee(s) and is in the public domain in the US.
883
Seabed classification using split-beam echosounder data
[Ss ¼ 10log10(ss)] e.g. Greenlaw et al., (2004), are typically made
using a pulsed, monostatic, multibeam echosounder operating at
a single frequency between 10 and 500 kHz. These measures of ss,
frequently versus u and infrequently versus both u and f, are used
to invert geoacoustic models and estimate seabed parameters.
Several models have been developed that include numerous parameters, e.g. for large- and small-scale roughness relative to the
acoustic wavelength (Caruthers and Novarini, 1993; Novarini and
Caruthers, 1998; Jackson et al., 1986, 2010, 2011; Jia and
Courtney, 2001), and characteristics, heterogeneity, and layering
of the sediment volume (Jackson et al., 1986; Novarini and
Caruthers, 1998; Jia and Courtney, 2001; Jackson et al., 2010). The
relative contributions of surface- and subsurface –volume backscatter depend on the combined physical attributes of the seabed, or
seabed type, and the acoustic frequency. In this paper, henceforth,
“seabed surface-backscattering” refers to measurements of highly coherent reflections from the seawater–seabed interface which may
include relatively minor contributions from relatively incoherent
volume scattering. These geoacoustic model inversions depend on
the input data, inversion algorithm, number of parameters, initial
conditions, parameter bounds, and convergence criteria (Jia and
Courtney, 2001), and they may not converge reliably, if ever
(Lamarche et al., 2011). Therefore, convergence is often attained by
constraining the model parameters to values assumed to represent
a set of expected seabed types (Fonseca and Mayer, 2007; Fonseca
et al., 2009; Kloser et al., 2010). Consequently, inversion results
may not represent all actual seabed types and their properties.
These issues have inspired the development of more generic, phenomenological models based on measures of ss(u, f) from known
seabed types. These models do not necessarily have parameters that
represent seabed characteristics, butby comparing field measurements
to these empirical models, many seabed types may be inferred or at
least differentiated by statistical analysis of data or generic model
parameter values (e.g. Chakraborty et al., 2003, 2004; Hamilton,
2011; Lamarche et al., 2011), and some seabed properties, e.g.
modal grain size, may be acoustically estimated (Hellequin et al.,
2003; Fonseca et al., 2009; Kloser et al., 2010; Lamarche et al., 2011;
Lucieer and Lamarche, 2011).
Model inversions often begin by compensating Ss for the effects
of u using Lambert’s law (Urick, 1967; Lurton, 2002; Greenlaw et al.,
2004; Lamarche et al., 2011):
ss0 =
ss
,
cos2 (u)
(1)
where ss0 is the angle-independent seabed surface-backscattering
coefficient which, for large incidence angles is equivalent to m in
Greenlaw et al. (2004). However, Lambert’s law only approximately
describes Ss(u) for some (e.g. rough) seabed types (Urick, 1967) and,
while appropriate for large incidence angles, is not an accurate
model for reflection and scattering near normal incidence
(Novarini and Caruthers, 1998; Lamarche et al., 2011). Hence, a
value of ss0 based solely on a Lambert’s-law model neglects specular
reflection and near-specular, henceforth “specular regime”, scattering, yet is useful for data that represent only large incidence angles
e.g. Greenlaw et al. (2004). Specular regime backscattering is often
modelled using an exponential term (Novarini and Caruthers,
1998; Jia and Courtney, 2001; Lamarche et al., 2011) and may be
modelled similarly by:
ss0 =
ss
cosCa (u)
,
(2)
which describes the variation of ss versus u where ss0 is the expected
value for ss at u ¼ 0, and the power of the cosine (Ca) is the angular
response parameter which is not constrained to a constant value as it
is in Eq. (1). Ca varies with seabed composition and roughness, and
acoustic frequency, and can be used to discriminate seabed types (Jia
and Courtney, 2001; Lurton, 2002; Fonseca and Mayer, 2007;
Fonseca et al., 2009; Jackson et al., 2010; Lamarche et al., 2011).
The ss0 is proportional to f Cf:
ss0 = sbs0 f Cf ,
(3)
where the proportionality coefficient, sbs0, is the normal incidence
(u h 08) backscattering cross-sectional area, and f is in Hertz.
Within a frequency band of 10–900 kHz and for small (,158)
grazing angle (908– u) values, the frequency response parameter, Cf,
ranges circa 0–1.5 and is indicative of seabed type (Jia and
Courtney, 2001; Greenlaw et al., 2004; Cable et al., 2006).
Substituting Eq. (3) into Eq. (2) and solving for ss(u, f) results in
ss u, f = sbs0 f Cf cosCa (u),
(4)
which is suitable for backscatter near normal incidence. An additional
term is necessary to model backscatter from normal to grazing angles
(Novarini and Caruthers, 1998; Jia and Courtney, 2001; Lamarche
et al., 2011). Eq. (4) can be expressed as a seabed-surface backscattering strength (Ss; dB):
Ss u, f = 10log10 (sbs0 cosCa (u)f Cf ),
(5)
or equivalently expanded to:
Ss u, f = 10log10 sbs0 + Ca 10log10 (cos(u)) + Cf 10log10 f .
(6)
This log-linear form is useful for estimating the model coefficients by
the method of least squares and avoiding non-linear fitting operations
that may not converge.
Multifrequency biplanar interferometry
and statistical– spectral identification
Large sets of sub-beam seabed detections can be acquired with multifrequency split-beam echosounders having collocated beams
(Demer et al., 1999) using the multifrequency biplanar interferometric technique (MBI; Cutter and Demer, 2010) and statistical –
spectral identification (SSID; Demer et al., 2009). MBI solutions
estimate the spatial location of each echo sample based on the orthogonal split-beam angles and the sample range. Seabed echoes
with high cross-frequency interferometric-phase-angle coherence
(Demer et al., 1999, 2009) are mapped into a local (Cartesian) reference frame (Conti et al., 2005) and then to a global (geographic)
reference frame (Cutter and Demer, 2010). This combination of
methods provides high-resolution measures of within-beam (i.e.
sub-beam) seabed ranges, ss(u, f), and u at multiple f, for each transmission. If motion data are available to estimate the transducer
orientation, these measures of u may be related to seabed slope,
without assumptions about the beam direction or seabed shape
(Cutter and Demer, 2010).
Study objectives
Seabed backscatter is dependent on many confounding factors
such as the seabed composition, roughness, slope, and spatial
884
heterogeneity. The principal objective of our study is to use
sub-beam seabed detections, acquired with multifrequency splitbeam echosounders and the MBI processing technique, to concurrently estimate the angular and frequency response parameters and a
number of other metrics (see Methods) that can be used to classify
seabed. Ultimately, the classifications and model parameters may
be used to identify seabed habitats for demersal fishes and to solve
geoacoustic models of the seabed.
Methods
Study area
The study area was Forty-Three-Fathom Bank (FTFB), 75 km west
of Point Loma, San Diego, California (Figure 1). The area provides
seabed habitat (Demer et al., 2009) to many rockfishes (Sebastes
spp.) and is included in the Cowcod Conservation Areas (Figure 1)
that were created to protect and revitalize the cowcod (Sebastes levis)
stock (Butler et al., 2003). The FTFB has been well mapped using multibeam and single-beam echosounders (e.g. Goldfinger et al., 2007;
Demer et al., 2009; Cutter and Demer, 2010). The top of the bank is
a nearly circular plateau at 100 m depth. Near the centre of the
plateau is a rough, rocky region with depths of 80 m. In the north
and east, the flanks of the bank drop off steeply to 150 m and then
flatten. In the west and southwest, the flanks continue to descend
steeply to the maximum mapped depth, 300 m.
Data acquisition
A survey of FTFB was conducted during five days from 25–29 October
2010, at a nominal speed of 8 knots, along parallel east–west and
northwest–southeast transect lines (Figure 1). Data were acquired
from multifrequency echosounders (Simrad EK60s) configured with
split-beam transducers (Table 1) mounted in a retractable keel on
the NOAA Fisheries Survey Vessel (FSV) “Bell M. Shimada.” The echosounder systems were calibrated prior to the survey, on 23 September
2010, while the ship was anchored in San Diego Bay, using a
38.1-mm-diameter tungsten-carbide sphere (Foote et al., 1987).
Throughout the survey, the echosounders synchronously transmitted 512-ms pulses at each f (Table 1). The transmit pulse rate was
1.1 s21 during the first day and 1.6 s-1 thereafter to allow for synchronization with other echosounders. Echo power and alongships
and athwartships interferometric-phase angles were recorded
(Simrad .raw format) to a maximum range of 500 m.
Processing of multifrequency echosounder data
Within the insonified areas defined by the intersections of the acoustic beams and the seabed, measures of u at each f were estimated
(Demer et al., 2009) using the multifrequency interferometricphase data (Demer et al., 1999). Assuming that the mean pitch
and roll of the ship were approximately zero (i.e. vertical echosounder transmissions), the mean seabed slope (u) was estimated as the
running mean of the beam-wise local u over n ¼ 30 multifrequency
transmissions. During the survey, including turns, the pitch were
measured (Applanix PosMV 350) to be ,3.18 and ,5.28 for 95%
and 99.9% of the time, respectively, and roll were ,2.68 and
,5.38 for 95% and 99.9% of the time, respectively. The pitch was
biased by 1–28. These u values agree with estimates derived from
the differences between adjacent multibeam-estimated depths (see
Figure 2 in Cutter and Demer, 2010).
Within the same areas, roughness (R) was estimated for each f
as the standard deviation of the MBI sub-beam bathymetry solutions (Cutter and Demer, 2010) from the fitted plane (Demer
G. R. Cutter and D. A. Demer
et al., 2009), i.e. the variation in detrended residuals of sample elevations, hence R includes macroscale or facet roughness. To estimate ss
versus a large range of u (potentially from 0 to 908) and versus f
spanning more than three octaves, the data were pooled within
each f, from multiple sequential transmissions. These data were
used to invert Eq. (6) for parameters related to seabed hardness
(density contrast with water) and roughness (elevation variation).
The echosounder measures of received power (pr; W) were converted to estimates of beam-compensated seabed backscattering
cross-sectional area (sbs; m2):
pr 16p2 r 4 10ar/5 0.60206(a2 +b2 −0.18(a2 b2 ))
sbs u, f =
10
,
pt g02 l2
(7)
where r is range (m), a is the absorption coefficient (dB m21), pt
is the transmit power (W), g0 is the system gain (dimensionless),
and l is the acoustic wavelength (m) (Demer et al., 1999). The
2
2
2 2
last term (100.60206(a +b – 0.18(a b ))), compensates for the
theoretical two-way beam directivity of the echosounder transducer
(Hammerstad, 2000), where a ¼ 2a/u23dB,alongships, b ¼ 2b/u23dB,
athwartships, a and b are elevation angles to the scatterer relative to
the maximum response axis of the transducer beam, and u23dB
angles are the transducer half-power (23 dB relative to the beam
axis) beamwidths (8) in two orthogonal planes, alongships and
athwartships (see Demer et al., 1999; Conti et al., 2005). The area of
the seabed insonified by each echosounder, A (m2), was estimated
following Lurton (2002) and Hellequin et al. (2003) by adapting
their expression for instantaneous area to the entire period (dt)
during which the pulse encounters the seabed:
⎧
⎨ 2Hr1 u−3dB tan u, u ≤ uLIM ,
A=
r0 dr u−3dB
⎩
, u . uLIM
sin(2u)
(8)
u−3dB = (u−3dB, alongships + u−3dB,athwartships )/2,
where
uLIM = cos−1 (H/(H + ct/2)), H is the water depth (m), r0 is the
nearest range to the seabed (m), dr ¼ r1 – r0 where ri ¼ c(ti – t )/2,
ti(s) is the echo travel time for sample i, hence r0 and r1 are the
ranges to the initial and final samples encountered within a beam
over dt, c is the speed of sound in water (ms21), and t is the pulse duration (s). This estimation of A defines a planar area and does not
account for sub-beam seabed variation (Hellequin et al., 2003).
Values of ss were estimated for each transmission and f by normalizing
the 0.90-quantile values (q90) of measured sbs [see Eq. (7)] by the estimated insonified area:
ksbs u, f l
ss u, f =
,
A
(9)
where ,. indicates the q90 statistic. The median sbs was explored,
but high variability near normal incidence resulted in some negatively
biased values for small angles.
Model of seabed-surface backscattering strength
Using Ss(u, f) from n transmissions and 28 intervals of u, the coefficients Ca(angular response) and Cf (frequency response) in Eq. (6)
were estimated numerically using a least-squares linear regression.
The Ss(u, f) were estimated for u ¼ {08: 908} and f ¼ {18, 38, 70,
120, and 200 kHz} (Figure 2). However, most values of u were
,508 and therefore the regression analysis was restricted to
Seabed classification using split-beam echosounder data
885
Figure 1. (a) The Forty-Three Fathom Bank (FTFB) study area (bounds indicated by the small bold rectangle) is located within the Cowcod
Conservation Areas (thin lines) in the Southern California Bight off the southwest of the USA (rectangle in inset). (b) Survey transects at the FTFB
sampled with multiple-frequency echosounders (Simrad EK60s), October 2010, aboard the NOAA FSV “Bell M. Shimada.” Axes labels on the inset
map are: (left, bottom) geographic coordinates—latitude, longitude; (top, right) projected coordinates—easting, northing (m), Universal
Transverse Mercator (UTM) zone 11 north. (c) The bathymetry, with 10-m-interval isobaths, results from an interpolation of the seabed detections.
886
G. R. Cutter and D. A. Demer
Table 1. Specifications for the multifrequency split-beam echosounders (Simrad EK60s).
Frequency (kHz)
Transducer model (Simrad)
-3 dB Beamwidths (88 )
Transmit power (W)
Receiver bandwidths (Hz)
18
ES18-11
10.3
2 000
1 749
38
ES38-B
6.8
2 000
3 275
70
ES70-7C
7.0
750
4 687
120
ES120-7C
7.2
250
5 557
200
ES200-7C
7.4
110
5 972
Transmitted power was sufficiently low to avoid non-linear effects.
Figure 2. (a) Frequency and (b) angular response curves for Ss data
from multifrequency echosounder data (18 –200 kHz) collected over
several hundred transmissions, demonstrating the large range of
incidence angles encountered and similarity to previous results.
u ¼ {28: 208} to avoid high variability at normal incidence (08), the
transition from specular to Bragg or diffuse scattering regimes, and
sparse data at large incidence angles. Within this angular span, Ss(u)
typically decreases rapidly with increasing u. In contrast, at larger
angles (208 , u ,808), the typically slow decay of Ss(u) often
approximates Lambert’s Law (a squared cosine term in Eqs. 4 –6).
In any case, the curve typically decreases monotonically with a steepness that is modulated by the relative contributions of large- and
small-scale roughness elements.
Spatial interpolation
The resulting model and MBI-derived parameters u, R, Ca , Cf
and Ss0(u, f), indexed by geographic position, were interpolated
spatially using an inverse-distance-weighted algorithm (Davis,
1986) to produce raster grid maps with 25 × 25 m cells (e.g.
Figures 3 – 4).
implemented in SAGA-GIS (System for Automated Geoscientific
Analyses), to form seven cluster groups. The number of cluster
groups was selected by an iterative method where the classification
was rejected and a different number of classes (3 –14) was used,
and the classification repeated if a single class contained the rocky
peak region and a substantial proportion of the sedimented
plateau or deep plains, or if the Mahalanobis distance [Euclidean
distance standardized by multiplying by the inverse of the covariance matrix (Davis, 1986)] between centroids of normalized data
was , 2.2 (Table 2). Our classification strategy follows that of
Hamilton (2011) such that the resultant mappings have spatial coherency and are consistent with known characteristics, and a larger
number of classes do not improve the spatial correspondence with
published “surficial geological habitat” (SGH) maps (Goldfinger
et al., 2007). The choice of seven groups is the minimum number
of groups that ensures that the rocky peak region (coincident with
known, utilized, rockfish seabed habitat) is distinctly separated
from other morphological regions that are visually evident from
the bathymetry (Figure 5b). A larger number of classes would increase the uncertainty about seabed characteristics because the
SGH maps have only gross spatial divisions of nine primary and secondary lithology combinations resulting from expert interpretation
of multibeam bathymetry and bathymetric derivative maps with reference to video images from manned submersible transects (covering 5.8 km) and 11 grab or core sediment samples (Goldfinger et al.,
2007) (Figure 5c and d). These coarse spatial divisions are much
larger than the spatial variations evident from the acoustic data,
prompting unsupervised classification. The centroids of model parameter values Cf, Ca, and Ss0(u, f)), characteristic of each class, were
input to Eq. (6) to generate curves relating angular and frequency
responses representative of each class (Figure 6).
Results
Seabed surface-backscattering strength
The Ss estimated at each of the five frequencies exhibited similar patterns. For example, the Ss measured at 38 kHz (Ss_38) are lowest
(,235 dB) on the steeply sloping flanks of the bank (Figure 3a).
Ss_38 are low (235 to 230 dB) in a deep area to the west of the
bank and in an area north of the bank. The Ss_38 are higher (235
to 220 dB) and more variable in the rocky region near the centre
of the circular plateau on top of the bank, and are much higher
(225 to 210 dB) in the surrounding flat area and the deep plain
to the east. The Ss_38 are highest (.25 dB) in an area to the southeast of the bank and during some of the survey direction changes in
the north, but these extreme values may be artifacts.
Seabed slope
Classification
The gridded measurements (u, and R) and model results [Cf, Ca,
and Ss0(u, f)] were classified using a cluster analysis based on an
iterative minimum-distance (k-means) algorithm (Forgy, 1965),
The estimates of u are ,58 in the area surrounding the central rocky
region on the plateau and in the central part of deep plain to the
southeast (Figure 3b). The u are ,108 along the eastern part of
the deep plain to the southeast. The u are between 10 and 208 in
Seabed classification using split-beam echosounder data
887
Figure 3. (a) Surface-backscattering strength Ss (dB) for 38 kHz; (b) seabed slope u (8) from within-beam incidence angle averaged over several
transmissions; (c) seabed roughness R (m), estimated from MBI solutions for the seabed. Axes label coordinates are: (left, bottom) latitude,
longitude; (top, right) easting, northing (m), UTM zone 11 north.
888
G. R. Cutter and D. A. Demer
Figure 4. (a) Seabed frequency-response coefficient Cf (dB re 1 Hz); (b) seabed angular-response coefficient Ca (dB re 1 rad). Very low, negative
Ca (,254.5), grey, may be an artifact of regressing Eq. (8) with ss spanning a small range of u; (c) seabed normal-incidence surface backscattering
strength, Ss0 (dB); (d) index representing effective roughness-to-hardness, as the normalized frequency response coefficient Cf/Ss0. Axes label
coordinates are: (left, bottom) latitude, longitude; (top, right) easting, northing (m), UTM zone 11 north.
889
Seabed classification using split-beam echosounder data
Table 2. Mahalanobis distance (Dmahal) between class centroids.
Class
1
2
3
4
5
6
2
2.86
3
3.41
2.34
4
2.82
3.46
2.28
5
3.43
3.15
3.28
3.12
6
3.36
3.28
3.15
3.26
3.45
7
3.46
2.86
3.41
2.82
3.43
3.36
most of the rocky peak area and on the bank flanks, and are .208 on
the steep southwestern edge of the bank.
Seabed roughness
Very low R (,0.1 m) are in the flat area on the top of the bank
(Figure 3c). Low R (0.1–0.25 m) are in areas to the southeast and
northeast of the bank top. Moderate R (0.26–0.75 m) are in the
peak rocky area and along most of the flanks to the southeast and
northwest. High R (0.5–1.5 m) are in the rocky region on the top
of the bank and in the areas of the eastern and northern parts of
the flanks. High and very high R (.1.5 m) cover most of the
areas with the steepest slopes on the southwestern edge of
the bank and occur in small patches in the peak rocky area, on the
flanks and in the southeastern deep plain.
Seabed frequency response
Low negative to zero values of Cf (21.28 to 0) are in the flat area on
the bank-top plateau and a deeper region to the east (Figure 4a).
Very low to intermediate positive Cf (0–0.75) are in the rocky area
on top of the bank and along transitional edges marking the
beginning of the descent of the western slope. Intermediate
Cf (0.75– 1.0) and patches of high (1.0– 1.5) and very high values
(.1.5) are in the deeper area to the southeast. High to very high
values are in the areas of the steep flanks.
Seabed angular response
Patches of very low, negative Ca (,2100) are in areas with steep
slopes or large rough features and in the plateau and the deep
plain areas to the southeast (Figure 4b). Low negative to low positive
Ca (249 to 250) are in the rocky peak region, on most of the steep
bank edge, and on the flanks bounding the bank plateau.
Intermediate Ca (250– 1000) are on the bank-top plateau and
deeper plains to the southeast and northwest of the bank top.
High (500–1000) and very high Ca (.1000) are in patches throughout the bank-top plateau and the southeastern deep plains.
Normalized seabed-surface backscattering strength
The Ss0 values evaluated for f ¼ 0 and u ¼ 0 are high (.210 dB) on
the bank plateau and the deep plain to the east and part of the deep
area to the north –northwest. Ss0 are typically moderate (230
to 220 dB), but variable (including some values ,260
to 210 dB), in the peak rocky region, low (,240 dB) on the northern slope and central part of the deep plain to the southeast, and very
low (,260 dB) on the steep bank flanks (Figure 4c).
Roughness-to-hardness index
The roughness-to-hardness index (Cf/Ss0) values are very low
(,20.02) on the flat bank top plateau and low (20.02
to 20.013) in most of the deep plains to the southeast of the bank
(Figure 4d). Low to intermediate values (20.013 to 20.011) are
throughout the steep southwestern edge and on the bank flanks.
The Cf/Ss0 are high (20.010 to 0.015) in the rocky region near the
centre of the bank plateau. Patches of very high Cf/Ss0 are in the
rocky peak area and a part of the deep plain to the east and deep
northern slope.
Classification and interpretation
The peak of the bank has rocky habitat characterized by large rough
features evident in the interpolated bathymetric data (Figure 5b).
A 7-cluster classification is based on Cf, Ca, Ss0, R and u (Table 3).
The high-relief rock region (class 4) is distinct from sedimentary
plains on the plateau and to the southeast, and similar to transitional
edges along the shoulder of the western slope and an area to the
north (Figure 5a). Class 4 is characterized by nearly flat but slightly
decreasing Ss(u, f)curves (Figure 6), low Ca values, a small positive
frequency dependence Cf, and moderate Ss0 (Table 3). The steep
sedimented flanks (class 1) also have a small angular response but
a large frequency response. Sand with gravel (class 3) covers the
bank plateau and deep areas to the east and north and has a large
angular response and a slightly negative frequency response. The
region along the steep rock western boundary (class 5) has a small
angular response and moderately large frequency response. Sand
with large cobble or boulders (class 2) occurs mainly on the deep
plain to the southeast and has a moderately large angular response
and small frequency response. Sand with cobble (class 7) also
occurs on the deep southeast plain and has a slightly larger
angular response and larger frequency response than class 2. The
patchy regions throughout the centre of the deep southeastern
plain and on the plateau (class 6) has the largest angular response
of all the classes, suggesting that this class represents finer sediments
than previously indicated by the SGH [“SGH_43_Fathom_Bank”
dataset, available from http://nwioos.coas.oregonstate.edu/
datasets.html, and described in Goldfinger et al. (2007)]
(Figure 5c and d and Table 3). Low R (≤0.25 m) are in plateau
and deep sedimentary plain areas (Figure 3c), which are flat according to the bathymetry (Figure 5b). These regions have been previously identified as primarily sand or gravel regions based on SGH
data. Moderate R values (0.26–0.75 m) are in areas with rough
rocky seabed and the near circular flank slope. High R values
(0.76–2.9 m) are on the steep slope bordering the entire bank to
the southwest.
Discussion
Seabed backscatter is modulated by many confounding factors
such as the seabed composition, slope and roughness, and the
acoustic frequency and angle of incidence. Backscatter intensity
versus incidence angle can be exploited to detect differences in
seabed properties (Urick, 1967; Jackson, et al., 1986; Kloser et al.
2010; Lamarche et al. 2011), and to solve geoacoustic models for
seabed properties (Jackson, et al., 1986, 2010; Novarini and
Caruthers, 1998; Jia and Courtney, 2001; Fonseca and Mayer,
2007; Fonseca et al., 2009). Furthermore, as Greenlaw et al.
(2004) showed, the frequency response may indicate seabed type.
Therefore, the principal aims of our study were to quantify seabedbackscatter versus both acoustic incidence angle and frequency,
and to use these and other model parameters to identify classes
of seabed. Our combined approach (i.e. utilizing MBI measurement of backscatter versus acoustic incidence angle and frequency), greatly increases the information available from
“vertical” split-beam echosounders for improved, high-resolution
discrimination of seabed properties.
890
G. R. Cutter and D. A. Demer
Figure 5. (a) Classification of the study area by cluster analysis of model results, including: Ca, Cf, Ss0 u, f , u, and R; with interpreted seabed
lithology and terrain (S ¼ sand, c ¼ cobble, b ¼ boulder, g ¼ gravel, m ¼ mud, r ¼ rock, x ¼ mixture; where upper- and lower-case letters
indicate primary and secondary contributions, respectively). (b) Plan-view, shaded-relief image of the bathymetric surface derived from the seabed
detections. The high-relief rocky region, comprising rockfish habitat, is conspicuous near the centre of the plateau. (c) Primary and (d) secondary
lithology maps created using the “SGH_43_Fathom_Bank” dataset (http://nwioos.coas.oregonstate.edu/datasets.html; Goldfinger et al. 2007).
The colours used in (a) distinguish acoustic classes and do not apply to color schemes of (c) or (d). Axes label coordinates in (a) (c) and (d) are:
(left, bottom) latitude, longitude; (top, right) easting, northing (m), UTM zone 11 north.
891
Seabed classification using split-beam echosounder data
Figure 6. Representative curves for Ss u, f produced by solving Eq. (5) with the modal parameter values characterizing each seabed class (Table 3).
Titles for each panel indicate the class number and the values used for Ca, Cf and Ss0(u, f).
Table 3. Seabed classes and associated model parameter centroids (where Ca, Cf, and Ss0 are the angular and frequency response, and
normal-incidence backscattering strength parameters, respectively); principal spatially concordant primary/secondary lithology from surficial
geological habitat (SGH) data (Goldfinger et al. 2007); lithology and terrain represented by each class based on interpretation of the model
parameters and bathymetry data with regard to SGH; and coverage of the study area.
Seabed class
1
Ca
32
1.06
Ss0
275.2
SGH
Sand/gravel
Interpreted lithology, terrain
Sand/gravel, on steep slopes
Area (km2)
1.84
% cover
9.2
Sand/large cobble, boulders
Sand/cobble or boulders, or
Boulders/mixtures, on flat
terrain
5.76
28.6
Sand/gravel
Sand/gravel, on flat terrain
4.27
21.2
Cf
2
219
0.30
224.5
3
328
20.25
5.5
4
40
0.48
241.2
High-relief rock, Boulders,
Sand/boulders, and Sand/
boulder + low-relief rock
Rock/boulders with patches of
sediments, in moderate- to
high-relief terrain
2.74
13.6
5
31
0.77
260.3
High-relief rock, High-relief rock/
sand, Cobble/sand, Sand/
boulders, and Sand/
boulder + low-relief rock
Rock/mixtures, on steep slopes
1.36
6.7
6
1762
0.44
220.2
Sand/cobble, and Possibly
misclassified
Fine sand or sandy mud, on flat
terrain
0.72
3.6
7
283
0.87
253.9
Sand/cobble
Sand/cobble, on flat terrain
3.46
17.1
Seabed backscatter model
Our seabed backscatter model, Eq. (6), is similar to that in Greenlaw
et al. (2004), except that we do not assume Ca ¼ 2 (Lambert’s Law).
This is because, for most seabed types and ranges of incidence angle,
Ca is neither 2 nor another constant, but varies with seabed composition and roughness, and acoustic frequency. Our model does not
892
include an interaction term that would describe the variation in
terms of combined angle and frequency effect because, within the
relatively small angular range of the constrained data, the effect
may be assumed negligible. Furthermore, to isolate and compensate
for a general effect of slope on seabed backscatter at any frequency
would require the unrealistic assumption of a known and uniform
seabed type. Therefore, Ca is variable in our model, resulting in a
heuristic, generic, Lambert-like model (Lurton, 2002) applicable
to the specular regime. Our model is also similar to the GSAB
model described by Lamarche et al. (2011), providing a simple
generic description of variation of ss and not inverting for the physical seabed attributes based on a theoretical model that may not
apply to our conditions, but instead producing associations
between parameter values and known properties of the seabed.
These may be developed into more formal classifiers for sites with
more ground-truth data. Our model differs from the basic GSAB
model by also including a frequency response term (Cf ) that
describes the variation of ss with f. This follows the approach of
Greenlaw et al. (2004) who estimated a frequency response parameter (that is suggested to be characteristic of seabed type) but only
after separately characterizing the angular response model. Our
model simply and simultaneously quantifies ss with respect to incidence angle and frequency for the near-specular regime.
In this study, u varied from 0 to .508, but were mostly ,208.
Therefore, our model was solved for this near-normal incidence
regime where Lambert’s Law is generally inapplicable as Ss(u)
decays approximately exponentially. Thus, our Ca term effectively
represents the exponential term in Lamarche et al. (2011) or the
mf term of Novarini and Caruthers (1998), and is analogous to
the near-angle-range slope term of Fonseca and Mayer (2007) and
Fonseca et al. (2009). Also, the Ss0 parameter of this model is
similar to the specular maximum amplitude parameter of the
Lamarche et al. (2011) model.
Model parameters and other seabed metrics
Plots of the multifrequency, single-beam echosounder data
(Figure 2b) strongly resemble the classical shapes of multibeam
backscatter versus incidence angle curves. A large positive Ca indicates decreasing Ss with increasing u, indicative of a region with
low roughness, e.g. a sedimented seabed. A near-zero or small positive Ca indicates constant or increasing Ss with increasing u, indicative of a rough rocky or steeply sloped region (Figures 3b and 4b).
However, Ss relates not only to u and seabed composition, but
also to local seabed roughness. For seabeds with low R, Ss decreases
greatly from 0 to – 40 dB as u increases from 08 to 158. For
seabeds with high R or steep slopes, Ss is constant or increases slightly for u between 0 and 208. The frequency response is also dependent on incidence angle, potentially changing from negative to
positive Cf values over the range of u, but typically not within the
analyzed range of theta.
Classification and interpretation
The acoustic model parameters were used to classify and segment
the seabed. For example, Cf values of 0.41 –0.69 indicate either
rough rocky regions or sedimented planes (Figure 4a), however
these two seabed types are distinguished by normalizing Cf by Ss0
(Figure 4d), as is evident from the bathymetry (Figure 5b). This
roughness-to-hardness index, Cf/Ss0, is independent of seabed
depth and slope. Negative Cf values indicate mainly sandy seabed
(Table 3) based on the primary lithology data from the independent
SGH (Goldfinger et al., 2007) and sample data (Reid et al., 2006).
G. R. Cutter and D. A. Demer
This interpretation is also consistent with Urick (1954) for
seabeds comprised of “sand and rock” and “silt and shell.”
However, Greenlaw et al. (2004) observed a linear increase of Ss
with increasing frequency (i.e. positive Cf values) for a seabed
with sandy sediment. This discrepancy may be explained by their assumption that Ca ¼ 2 and their estimation of Cf ≈1.4 for a narrow
range of large incidence angles (u ≈758). In contrast, our Cf are variable (Figure 4a) and our ub are typically ,308. Perhaps, as suggested
by the secondary lithology of the SGH, the negative Cf values in our
study are indicative of heterogeneous seabed, e.g. sand mixed with
shell or rock, or surficial sand over hard consolidated material.
Interpretation or validation based on primary and secondary lithology has limitations. For example, when they differ, the primary
and secondary components can represent from 51 –99%, and 1 –
49% of the material, respectively. Each lithology class can represent
a fairly large range of sediment grain sizes and associated seabed
properties. Nonetheless, when considering the seabed as habitat
for rockfishes, these classifications may be sufficient. Also, gross
delineations based on sparse data in some areas may misclassify conditions, e.g. south of the rocky central peak where the SGH map
indicates sand (Figure 5c and d), but bathymetry (Figure 5b) indicates large boulders, rocks or both.
The map of the cluster analysis results (Figure 5a) is indicative of
broad-area patterns of lithological and morphological properties,
and enables geographic segmentation of the study site (by selecting
all map grid cells for any class). The classes uniquely distinguish
rocky versus steep, sedimented slopes (class 5 versus 1) and
fine-to-coarse sediment from various terrains. Good spatial correspondence exists between class 4 and the high-relief rock region near
the centre of the plateau (Figure 5b). However, the parameters are
variable and the classes are patchy in this region because it also contains small regions of sand, gravel and boulders among the rocks
(Demer et al., 2009).
Representative curves for Ss u, f (Figure 6) were produced by
solving Eq. (5) with the modal values of Ca, Cf, and
Ss0(u, f) that
characterize each seabed class (Table 3). The Ss u, f curves for
each seabed class are distinguishable by their rates of change
versus angle and frequency. These characteristic shapes agree with
the theoretical angular response curves for various sediment grain
size and roughness, and the independent SGH data. Specifically,
seabeds with larger sediment grain sizes or roughness elements
have lower angular response. For instance, the
rocky region
rough
depicted by class 5 (Figure 6) has nearly flat Ss u, f . The region indicated by the SGH data
to
have sand with gravel relates to class 2 with a
relatively steep Ss u, f curve. The curves for class 3 have similar
shapes, but are less steep and correspond to an SGH region containing sand and cobble. The curves for class 2, gravelly sand, are uniquely characterized by negative Cf. Class 6, indicative of finer-grained
sediments, occurs throughout the centre of the deep southeast
plain and has the highest Ca. The weak angular response and moderately strong frequency response of class 7 imply sediments similar
to class 3, but perhaps coarser.
Observation scale
The same seabed, when considered on different spatial scales, may be
considered homogeneous or heterogeneous (see e.g. Reid, 2007;
Anderson et al. 2008). The MBI samples from a single-beam echosounder transmission may span only a small area and range of u, particularly for a smooth flat seabed, and consequently may be too
sparse or variable to accurately identify the seabed type. On the
other hand, a collection of MBI measurements from a large
893
Seabed classification using split-beam echosounder data
number of transmissions spanning large distances may convolve
the effects of multiple seabed types. Therefore, it is important to
consider that Ca and Cf may represent the combined effects of multiple seabed types. In other words, the model parameter values for
the seven classes (Table 3) represent region-scale phenomena at
least at the size of the interpolated grid cell size (25 × 25 m), and
do not indicate finer-scale seabed heterogeneity. This limitation
also applies to data from single transmissions from a multibeam
echosounder. For example, a 1208-swath of a single multibeam
transmission in the shallowest, 90 m deep, portion of the
FTFB, would span 312 m, potentially spanning multiple seabed
types.
Acknowledgements
We thank the other members of the Advanced Survey Technology
Group including Kyle Byers, Josiah Renfree, Steve Sessions and
Juan Zwolinski, and the captain and crew of the FSV “Bell
M. Shimada,” for their parts in calibrating and collecting the multifrequency echosounder data used in this analysis. We thank L. N.
Andersen and Simrad for technical support of EK60 echosounders.
We thank Christian Reiss and Dae Jae Lee for providing comments
and suggestions on a draft of the paper, and Russ Vetter and Bill
Perrin for their roles in the SWFSC internal review process.
Funding
This project was base funded by the authors’ institution.
Future work
The methods that we have developed and demonstrated here
provide high-resolution (sub-beam) metrics for the seabed that approach or exceed the fine spatial scales of optical and physical
samples of the seabed. The methods are applicable to data from multifrequency echosounders that are used to concomitantly survey fish
and their seabed habitat. Therefore, the methods should be broadly
applicable to improve both seabed classifications and habitat characterizations. To better account for seabed heterogeneity, the acoustic classes should be derived from smaller analysis cells, e.g. 10- to
50-m distance. Then, to verify the finer scale lithological characteristics of each class, seabed photographs and physical samples should
be used. Representative seabed photographs or samples are needed
for all of the major class regions. These efforts are underway using
camera images collected with a remotely operated vehicle (ROV).
After sufficient verification data are obtained, the classes may be
refined and used to predict potential seabed habitat for rockfishes
elsewhere, on any similar bank, potentially for any of the banks in
the Southern California Bight. Refinement of classes may involve
splitting some classes or defining new classes with adjusted model
parameter values resulting from classification done using a different
number of groups, a subset of model parameters, or another grouping algorithm. Alternatively, classes may be redefined. These
samples provide a basis for supervised classification where model
parameter values are collected into feature vectors representing specific seabed attributes with known spatial extent. The resulting
classes would represent more precise seabed types.
In the context of acoustic– optical surveys of rockfishes (e.g.
Demer et al., 2009), an accurate model of potential seabed habitat
for rockfishes could serve to optimize the expansion of the ROV
sampling and the reduction of acoustic sampling. For example, previous work to verify the species and sizes of rockfishes contributing
acoustic backscatter over the FTFB (Demer et al., 2009) has involved
ROV transects concentrated on the high-relief rock region near the
centre of the bank. If all class 4 regions comprise potential seabed
habitat for rockfishes, e.g., then additional parts of class 4 areas
should be sampled with ROV transects, particularly along the
western slope edge, in the north, and in small areas in the deeper
regions to the east and southeast of the bank. Also, if rockfishes
were found only to occur in the class 4 regions, then future acoustic
sampling for rockfishes could be reduced on this bank by 10 –50%,
depending on whether the small patches of class 4 in the southeast
may be ignored (Figure 5a). The class 4 region on the northern
flank that seems to contradict the SGH map should be investigated.
The actual increase in survey efficiency will depend on the spatial
dispersion of the important classes on each bank.
References
Anderson, J. T., Holliday, D. V., Kloser, R., Reid, D. G., and Simard, Y.
2008. Acoustic seabed classification: current practice and future
directions. ICES Journal of Marine Science, 65: 1004– 1011.
Butler, J., Jacobson, L., Barnes, J., and Moser, H. 2003. Biology and
population dynamics of cowcod (Sebastes levis) in the Southern
California Bight. Fishery Bulletin US, 101: 260 – 280.
Cable, P. G., Yudichak, T. W., Dorfman, Y., Knobles, D. P., Zhang, R.,
Peng, Z., Li, F., et al. 2006. On shallow-water bottom reverberation
frequency dependence. IEEE Journal of Oceanic Engineering, 31:
145– 155.
Caruthers, J. W., and Novarini, J. C. 1993. Modeling bistatic bottom
scattering strength including a forward scattering lobe. IEEE
Journal of Oceanic Engineering, 18: 100 – 108.
Chakraborty, B., Kodagali, V., and Baracho, J. 2003. 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: 121– 128.
Chakraborty, B., Mahale, V., de Sousa, C., and Das, P. 2004. Seafloor
classification using echo-waveforms: a method employing hybrid
neural network architecture. IEEE Geoscience and Remote Sensing
Letters, 1: 196– 200.
Conti, S. G., Demer, D. A., Soule, M. A., and Conti, J. H. E. 2005. An
improved multiple-frequency method for measuring in situ target
strengths. ICES Journal of Marine Science, 62: 1636– 1646.
Cutter, G. R., and Demer, D. A. 2010. Multifrequency biplanar interferometric imaging. IEEE Geoscience and Remote Sensing Letters
7: 171 – 175.
Davis, J. C. 1986. Statistics and Data Analysis in Geology. John Wiley &
Sons Inc., New York.
Demer, D. A., Cutter, G. R., Renfree, J. S., and Butler, J. L. 2009. A statistical– spectral method for echo classification. ICES Journal of
Marine Science, 66: 1081– 1090.
Demer, D. A., Soule, M. A., and Hewitt, R. P. 1999. A multiple-frequency
method for potentially improving the accuracy and precision of in
situ target strength measurements. Journal of the Acoustical
Society of America, 105: 2359– 2376.
Fonseca, L., Brown, C. J., Calder, B., Mayer, L., and Rzhanov, Y. 2009.
Angular range analysis of the seafloor themes from Stanton Banks
Ireland: a link between visual interpretation and acoustic signatures.
Applied Acoustics 70: 1298– 1304.
Fonseca, L., and Mayer, L. 2007. Remote estimation of surficial seafloor
properties through the application Angular Range Analysis to multibeam sonar data. Marine Geophysical Researches, 28: 119 – 126
Foote, K. G., Knudsen, H. P., Vestnes, G., MacLennan, D. N., and
Simmonds, E. J. 1987. Calibration of acoustic instruments for fish
density estimation: a practical guide. ICES, Cooperative Research
Report, 144, 69 pp.
Forgy, E. W. 1965, Cluster analysis of multivariate data: efficiency vs
interpretability of classifications. Biometrics, 21: 768 – 769.
894
Goldfinger, C., Romsos, C., Chaytor, J., Yoklavich, M., Amend, M.,
Watters, D., Wakefield, W., et al. 2007. Multibeam Sonar Surveys
and Geological Habitat Mapping of the Seafloor within the
Cowcod Conservation Areas (CCA), Southern California
Continental Borderland. Cooperative Research Report, Oregon
State University and National Oceanic and Atmospheric
Administration, ATSML Report 07-01. 40 pp.
Greenlaw, C., Holliday, D. V., and McGehee, D. E. 2004. High-frequency
scattering from saturated sand sediments. Journal of the Acoustical
Society of America, 115: 2818– 2813.
Hamilton, L. J., 2011. Acoustic seabed segmentation for echosounders
through direct statistical clustering of seabed echoes. Continental
Shelf Research, 31: 138 – 148.
Hammerstad, E. 2000. Backscattering and Seabed Image Reflectivity.
EM Technical Note. Kongsberg, Norway.
Hellequin, L., Boucher, J., and Lurton, X. 2003. Processing of highfrequency multibeam echo sounder data for seafloor characterization. IEEE Journal of Oceanic Engineering, 28: 78 – 89.
Hines, P. C., and Heald, G. J. 2001. Seabed classification using normal
incidence backscatter measurements in the 1 – 10 kHz frequency
band. Proceedings of the Institute of Acoustics, 23: 42– 50.In
Acoustical Oceanography. Ed. by T. G. Leighton, G. J. Heald, H.
Griffiths, and G. Griffiths.
Jackson, D. R., Odom, R. I., Boyd, M. L., and Ivakin, A. N. 2010. A geoacoustic bottom interaction model (GABIM). IEEE Journal of
Oceanic Engineering, 35: 603 – 617.
Jackson, D. R., Odom, R. I., Boyd, M. L., and Ivakin, A. N. 2011.
Corrections to “A Geoacoustic Bottom Interaction Model
(GABIM)”. IEEE Journal of Oceanic Engineering, 36: 373.
Jackson, D. R., Winebrenner, D. P., and Ishmaru, A. 1986. Application of
the composite roughness model to high-frequency bottom backscattering. Journal of the Acoustical Society of America, 79: 1410– 1422.
Jia, Y., and Courtney, R. C. 2001. Broad-band frequency and
incident-angle dependence of bottom backscattering on Browns
Bank. IEEE Journal of Oceanic Engineering, 26: 373– 382.
Kloser, R. J., Penrose, J. D., and Butler, A. J. 2010. Multi-beam backscatter measurements used to infer seabed habitats. Continental Shelf
Research, 30: 1772– 1782.
G. R. Cutter and D. A. Demer
Lamarche, G., Lurton, X., Verdier, A-L., and Augustin, J-M. 2011.
Quantitative characterization of seafloor substrate and bedforms
using advanced processing of multibeam backscatter. Application
to the Cook Strait, New Zealand. Continental Shelf Research, 31:
93 – 109.
Lucieer, V., and Lamarche, G. 2011. Unsupervised fuzzy classification
and object-based image analysis of multibeam data to map deep
water substrates, Cook Strait, New Zealand. Continental Shelf
Research, 31: 1236– 1247.
Lurton, X. 2002. An Introduction to Underwater Acoustics.
Springer-Verlag, Berlin.
Novarini, J. C., and Caruthers, J. W. 1998. A simplified approach to
backscattering from a rough seafloor with sediment inhomogeneities. IEEE Journal of Oceanic Engineering, 23: 157– 166.
Penrose, J. D., Siwabessy, P. J. W., Gavrilov, A., Parnum, I., Hamilton, L.
J., Bickers, A., Brooke, B., et al. 2005. Acoustic Techniques for Seabed
Classification. Cooperative Research Centre for Coastal Zone
Estuary and Waterway Management, Technical Report 32.
Reid, D. 2007. Accounting for spatial and temporal scales and interpolation in acoustic seabed classification surveys. In Acoustic Seabed
Classification of Marine Physical and Biological Landscapes, pp.
73 – 93. Ed. by J. T. Anderson, D. V. Holliday, R. Kloser, D. Reid,
and Y. Simard. ICES Cooperative Research Report No. 286.
Reid, J. A., Reid, J. M., Jenkins, C. J., Zimmermann, M., Williams, S. J.,
and Field, M. E. 2006. usSEABED: Pacific Coast (California, Oregon,
Washington) offshore surficial-sediment data release: US Geological
Survey Data Series 182, version 1.0. http://pubs.usgs.gov/ds/2006/
182/ (last accessed 17 October 2013).
Stanton, T. K. 1984. Sonar estimates of seafloor microroughness.
Journal of the Acoustical Society of America, 75: 809– 818.
Sternlicht, D. D., and DeMoustier, C. P. 2003. Remote sensing of sediment characteristics by optimized echo-envelope matching.
Journal of the Acoustical Society of America, 114: 2727– 2743.
Urick, R. J. 1954. The backscattering of sound from a harbor bottom.
Journal of the Acoustical Society of America, 26: 231– 235.
Urick, R. J. 1967. Principles of Underwater Sound for Engineers.
McGraw – Hill, New York.
Handling editor: Howard Browman