Ocean Propagation Models PDF
Ocean Propagation Models PDF
Ocean Propagation Models PDF
C. H. H a r r i s o n
YARD, ScientificHouse, 40--44 Coombe Road, New Malden, Surrey KT3 4QF, UK
A BS TRA C T
Four approaches to underwater sound propagation modelling are reviewed
including rays, normal modes, Green's function integral, and parabolic
equation. Specific programs are discussed. Particular reference is made to the
problems in running models and the applicability in various regimes,
especially deep-water environments.
1 INTRODUCTION
The wave equation and the Helmholtz equation occur in many branches of
physics, and the basic types of solution are always the same. However, the
problems in different fields vary tremendously because of emphasis on
different types of source, receiver, environment, geometry and so on.
Whereas propagation of light and radar in air is relatively straightforward,
underwater sound propagation can be extremely complex and calculations
may require several hours of computation on a CRAY-size computer. In the
underwater environment there are many different types of problem, some far
simpler than others. These include detecting distant ships or submarines,
echo sounding, short baseline location, seismic surveying and acoustic
tomography.
A typical problem of interest in this paper is to calculate intensity (and
phase) from a receiver somewhere in the water column separated by
distances of 1-100km in water of depth 100-5000m. The water is usually
assumed to be stratified (i.e. depth-dependent sound speed with constant
163
Applied Acoustics 0003-682X/89/$03-50 O 1989 ElsevierSciencePublishers Ltd, England.
Printed in Great Britain
C. H. Harrison
164
2 REGIMES A N D E N V I R O N M E N T S
It is difficult to appreciate the complexity of propagation modelling without
understanding the many environments and propagation regimes. These are
distinguished partly according to the various mathematical approaches and
partly according to oceanography and geophysics; many references discuss
the subject.l - 5 On top of this there are variants of these environments and
other applications of the usual models such as propagation from air to
water,6-a propagation in air, 9 seismic modelling, 1 ultrasound, 1 and radio
propagation over undulating terrain (the multiple-knife-edge diffraction
problem which is often tackled fairly crudely l~). Despite the complications
of diffraction and wave treatment the phenomena that need to be modelled
can be understood very simply by considering the corresponding ray paths.
2.1 Shallow water
The term 'shallow water' is often taken as synonymous with the Continental
Shelf where it has special strategic significance. For shipping-line frequencies of less than a few hundred hertz, where the wavelength is a
significant proportion of the water depth, shallow water behaves like a
waveguide in which propagation is dominated by bottom reflection. Ray
paths may be relatively steep, and consequently relatively straight, in the
165
water column so that the influence of the water column velocity profile is
minimal. On the other hand, the propagation may be strongly influenced by
the structure of the sediment and the underlying rock since sudden changes
in velocity and density and the inevitable gradual increase in velocity with
depth provide strong reflections, particularly if there is a critical angle, and
upward refracted paths with potentially low losses.
Absorption in the sediment increases with frequency and typical values 12
are between 0.05 and 0.5dB/m/kHz, i.e. at 100Hz the loss might reach
l0 dB after a 1 km path length. By contrast, the losses in water are extremely
low, although they still rise with frequency. At 100 Hz a 10 dB loss is reached
after l0 000 km. The corresponding ranges for 10 kHz, though, are 10 m for
sediment and 10 km for water.
An additional factor in shallow water is the possibility of a solid rock sea
bed which supports shear waves. Under certain conditions the boundary
between the solid and fluid layer (whether water or sediment) may exhibit
interface waves, the effects of which are significant for sources and receivers
near the bottom. 13 A common phenomenon is the result of the fact that the
shear critical (grazing) angle, if it exists, is always smaller than the
compression wave critical angle, so that there is a tendency for energy to leak
into the bottom by generation of downward-propagating shear waves. This
effect can be screened out or toned down by an intermediate layer of fluid
sediment between the rock and water.
2.2 Deep water
In deep water propagation can be very complex because there are so many
different types of path. The sound speed increases roughly linearly with
depth, temperature and salinity. The salinity in the oceans usually plays a
minor role, although in the polar regions, in landlocked seas such as the Baltic,
and in the vicinity of some ocean fronts and eddies, there are significant
effects. In the North Atlantic in winter there is typically a well mixed warm
surface layer, several hundreds of metres thick which, when superimposed
on the depth effect, causes a minimum in sound speed at about 1000 m with a
maximum at several hundred metres. Above this maximum there is a surface
duct where rays from a shallow source can be trapped by upward refraction
and surface reflection.
Below the maximum is a deep sound channel centred on 1000 m, bounded
entirely by refraction. However, to make use of this regime both source and
receiver need to be at depths below the maximum. Steeper rays from a nearsurface source travel more deeply, and the roughly linear increase of velocity
with depth causes upward refraction. There is often a slight focusing effect so
that the rays that nearly hit the bottom bunch up near the surface at around
166
C. H, Harrison
167
ray tracing;
normal mode;
Green's function types;
C. H, Harrison
168
(4)
(5)
(6)
(7)
The first six types have been reviewed by De Santo 27 and the first four by
Jensen. 2s This review will restrict itself to the first four types although some
references are given for coupled mode, 29-33 finite element 27'34'35 and
empirical 1,36 - 3 8 models.
Examples of computer codes are given below: a more complete list is given
in references 39 - 4 2 :
ray,
mode,
Green's function,
PE,
There are other models such as FACT 57 and R A Y M O D E ss that do use ray
concepts, although transmission loss is calculated without going through the
motion of tracing the rays. These models are designed to be used 'hands-off"
and so mode calculations are inserted automatically as appropriate, source
and receiver may be swapped, etc. Therefore for the research user it is often
difficult to control what calculations are executed.
A brief description of the four approaches is given below, but more detail
can be found in Refs 2-4, 27-28 and 59-61. All four can be derived from the
inhomogeneous Helmholtz equation, i.e. the wave equation with a point
harmonic source,
V % + k2(r)4, = - ~(r)
(1)
where k =
w/c is the
local wavenumber.
(1 dfl=
dy=
~-~[(1
ds[(1
oc
fl2)~y 2 ~c
Uy-y
oc1
Oc - o~flOc]
Ox[
)'fl-~z
Oc fl ~c7
(2)
169
t = S (c sin 0)- 1 dz
(3,4)
using Snelrs law, that csec 0 is constant for each ray. Also, the radius of
curvature of the ray, R, can be written in terms of velocity, its gradient and
the grazing angle at one depth as
R = c see O/(dc/dz)
(5)
('. H. Harrison
170
(7)
where q~, are the normal modes evaluated at the source and receiver depths zs
and z,, and K. are the eigenvalues. The discrete sum represents loss free
modes equivalent to up- and down-going rays travelling at a well defined
angle given by the horizontal wavenumber K, = k(z) cos O(z). The branch cut
integrals represent lossy modes travelling at lower wavenumbers (steeper
angles), and the contributions that are only effective at short range where the
up- and down-going waves do not balance properly yet.
The normal modes are calculated by choosing trial values of K. and using
numerical techniques such as Runge-Kutta iteratively, until the boundary
conditions are matched at depths zero and infinity with the correct number
of zero crossings. 6v These methods are adequate for shallow water in which
there are a small number of modes, and they are the basis of the model
SNAP. In deep water the far larger number of modes is a problem in itself,
but the close packing in wavenumber coupled with the refracting ducts
means that low-order modes are likely to contain upward- and downwarddecaying portions which lead to stability problems with standard shooting
methods such as Runge-Kutta. Porter and Reiss 4v'6a have used a finite
difference approach which is adopted in SUPERSNAP to formulate the
mode calculation as an algebraic eigenvalue problem which is more stable.
Ferla et al. report some high-frequency normal mode calculations in deep
w a t e r . 69
171
(8)
This is the basis of the Fast Field Program (FFP) 51'52 and SAFARI. 48 - s0
In both programs the bulk of the computation time is taken up with
calculation of the Green's function G (as a function of horizontal
wavenumber K). The Hankel transform is approximated by a Fourier
transform which is implemented by F F T algorithm. The original F F P used
the Thomson-Haskell method at each K to evaluate G by matching the
solutions at each layer boundary and the source. S A F A R I starts with the
same equations but uses a global matrix scheme to solve for all K and all
layers simultaneously.
The Green's function, as a function of K for fixed source and receiver
depth, is an intermediate output of the programs, and it is extremely useful
not only for checking performance but for providing insight into the physics.
The discrete and virtual modes are shown in Fig. l(a), and the corresponding
transmission loss is shown in Fig. l(b). The distinction between the Green's
function and the normal modes can be seen by analogy with a violin string
stretched across the water column. The normal modes are the shapes of the
violin string at its many resonances (the violin frequency corresponds to the
horizontal wavenumber). Between resonances the amplitude is zero, and so
the sequence is discrete. The Green's functions are the shapes of the violin
string when driven by a harmonic source at some particular position along
the length (i.e. the source depth). These exist for any frequency of the
harmonic source, although there will be a large amplitude at each resonance.
C. H. Harrison
172
SAFARI
20.0
F
* 15.0
50.OH~
SD =
50.OH
RD:
IO0.OH
10.0
C3
Z
I%
(._b
LLI
F--
5.0
z
'~
0,0
314
4734.2
9154.2
13574.2
17994.2
2241
.2
( 1 O* *-- 5)
(a)
50
SAFARI
-TLRA
613 0
70
u'l
80
O
J
90
1013
1113
13
0.0
15.O
RANGE
313.O
45.0
60.0
tKM;
(b)
Fig. 1. (a) SAFARI discrete and continuous mode spectrum. (b) SAFARI transmission loss.
The effect o f including bottom losses or any other losses is to broaden and
shorten the resonance peaks in a calculable way. When the broadening is
severe the modes are referred to as 'virtual' modes 75-77 and the modes
decay exponentially with range (as is apparent from considering the Fourier
transform o f an exponential). Similarly, the method can handle mode cutoffs, shear waves in solid layers (sediment, seabed, or ice) and near field as
173
q~ = (r, z)S(r)
(9)
02
Oz---f + 2ik o ~r + k2(n 2 - 1) = 0
(10)
174
C. H. Harrison
~,~(~(r,z))} (11)
1/2
exp(iko(n 2 - 1)Ar/2)
x fo~ ~,(r,z')exp{[iko(z
J--
z')2]/2Ar} dz'
oO
Given the field on the vertical plane at r the second term in the integral is
precisely the Fresnel diffraction term in passing from an arbitrary point (r, z')
on the first screen to the observation point on the second at (r + Ar, z). The
exponential term outside the integral corrects the phase already included in
S(r) (i.e. kor, see eqn (9)) by adding in the phase due to the slowly varying
refractive index but taking out kor. The result, koAr(n-1) is equal to
koAr(n 2 - 1)/2 to first order since n is always very close to unity. The
multiplier turns into the usual ko/2niAr of the Kirchhoff approximation
when allowance is made for the other dimension of the diffraction screen
(horizontal, out of the plane of propagation) which supplies an extra
(ko/2niAr) x/2 term.
The method has been extended to wider angles (40-60 ) and stronger
bottom interaction by using finite difference methods, and one such model is
Implicit Finite Difference (IFD). ~9 There have been developments in which
density is treated more comprehensively.8 A number of operator techniques
for manipulating the parabolic equation have been presented at Yale. al
Recent extensions include calculating the field in a uniform a2 or sheared a3
current, and close approaches to three-dimensional modelling. 25'a4- 86
175
4 PRACTICAL LIMITATIONS A N D T U N I N G
Having briefly reviewed the mode of operation of the models it is useful to
look at some of the more practical limitations. These depend on validity in
the given regime or environment, computation time and very often a lot of
fine tuning and user experience. As well as the oceanographic and geoacoustic inputs, which are themselves often difficult to define, most models
have numerical inputs which need to be chosen by experience or trial and
error. For instance, ray-tracing models require a ray density at the source;
parabolic equation models require a reference sound speed and an
understanding of the angle limits and constraints on medium variability;
S A F A R I requires various compromises to be made in numerical
integration. One can usually devise simple formulae as an aid to choosing a
176
C. H. Harrison
starting point, but there is no substitute for repeating runs with altered
numerical inputs until convergence is reached.
The following discussions are restricted to GRASS (ray), SNAP and
S U P E R S N A P (normal mode), S A F A R I (Green's function) and P A R E Q
and I F D (parabolic equation). The comments are aimed at deep-water
environments with source and receiver within a few hundred metres of the
surface.
4.1 GRASS
It is obvious that the chosen ray density at the source must provide adequate
ray coverage at distant points of interest. However, it is not always easy to
predict what that density should be without a trial run. Ray traces with
coarse and fine ray densities for an identical shallow surface duct are shown
in Figs 2(a) and (b). The apparent size of the wedge-shaped shadow zone is
quite different in the two cases. To a certain extent the worries are illusory
because these rays (surface-reflected) do not contain much energy, but the
incoherently added intensity (Fig. 2(c)) shows steps corresponding to the
wide ray spacing in range, even for the high-density example. This kind of
problem arises when the ray location at a distance is very sensitive to initial
ray angle, i.e. lar/aOol is large, but this is exactly the condition that the
intensity contribution is low. Whether or not the result is important depends
on whether or not there are stronger contributions from elsewhere.
The obvious solution is to carry on increasing the ray density, but one
penalty is increased numerical error, and another more practical limitation
is computation time, which is roughly proportional to total range covered
and number of rays.
Although GRASS is essentially a high-frequency approximation, the
'coherent' option sums rays with regard to their phase and is capable of
showing some frequency dependence other than simple absorption effects.
In principle, a ray treatment can handle bottom reflections and bottomrefracted paths, but G R A S S cannot handle both correctly. One choice is to
use a compromise reflection loss curve which attempts to cover both cases,
but this cannot possibly treat the horizontal offset associated with the
refracted path correctly. Another is only to use the model with the true
reflection loss curve in the case where refracted paths are non-existent or
extremely weak. The converse case where refracted paths dominate over
reflected paths cannot be handled, despite the fact that refracted rays can be
traced. This is because the sediment layer would have to be treated as part of
the water column, and G R A S S does not have the facility to include
absorbing layers in the water.
In a duct ray treatments are valid as long as the duct supports a
(a)
177
ZS = 3 8 m
O,
20-
. ~.60-
801
100
,
1
,
3
Ray trace
(b)
,
4
. . . .
5
6
Range (km)
10
10
ZS:38m
.C:
1
0
Range (kin)
(c)
50"
607o--
8o j
110
\
,
210
'
4.'0
610
8"0
10-0
Fig. 2. (a) Coarse ray trace: 1 ray/degree; 16 spread. (b) Fine ray trace: 100 ray/degree;
1 spread. (c) Transmission loss showing steps due to finite but small ray spacing.
C. H. Harrison
178
substantial number of modes (i.e. definitely greater than one, but not
necessarily an extremely large number). Thus, treating the duct crudely as a
rectangular well of height h, the mode number is related to ray angle by
2H sin 0 = n2. For weak ducts with velocity contrast, Ac,
sin 0 = (1
(12)
f 2 = c3/8h2Ac
(13)
For instance, ray treatment in a surface duct with Ac = 4 m/s and h = 500 m
requires frequencies to be well above 20 Hz. In fact, since the duct crosssection is closer to linear than rectangular, the values of h and kc should be
reduced to compensate, so that the cut-off frequency is two or three times
higher. Better approximations to the formula may be made using the WKB
phase integral 1 for n. The cut-off frequency for the complete water column
is usually far lower, and for typical frequencies ray treatments are valid in
deep water (provided that some allowance is made for Lloyd's mirror, the
bottom dipole effect, etc. as appropriate). The cut-off frequency is now
roughly
f2 = [4H2(co 2 _ c12)] -1
(14)
where H is the total depth and the larger velocity spread is defined by the
limits c o and c 1. For Co -- 1500 m/s, c 1 = 2000 m/s and H = 4000 m the cut-off
frequency is 0.3 Hz.
Above the mode cut-off the dependence of intensity on mode number is
weak, because although the number of modes increases in proportion to
frequency the wavenumber in the denominator of the mode sum formula
(originating from the square of the Hankel function Ho(Knr ) in eqn (7)) also
increases in proportion to frequency. So, apart from some saw-tooth effects
with very low mode number the response is flat. A more important effect for
sources and receivers removed from the centre of the duct is the frequencydependent reduction in mode amplitude near the boundaries. Crudely, the
intensity for one mode is proportional to sin 2 ~.zs sin 2 ~.z,, and with both
source and receiver near the surface this reduces to ~,4.z~z,Z, where (in terms of
the horizontal wavenumber, K,, and the wavenumber in the medium, k)
V. = (k2 - Kz.) 1/2 = (2rtf/c)sin 0.
(15)
Thus intensity will rise with the square of frequency until saturation when
sin 2 ?,z approaches 1 for most modes. This can only be modelled with
G R A S S by a very large number of coherent rays.
A clear indication of the importance of modal effects at frequencies as
high as 5 kHz (in typical shallow surface ducts) is shown by the comparison
of Fig. 3(a) (an I F D intensity contour run on a C R A Y II) with Fig. 3(b) (ray
o r
24
5000.OOHZ,
I~.~o
SO.
5.00M
~~~--~'--~-
~
.~
179
CONDR
..,
"
.,
.,
.o,I
,.,.
,
.
.'.
~
.
OI
;:~*':
o
.:
.:
o
(B)
.;
48
, v 0
~ 0 0
72
I--
o~
Ld
CD
~~.,,~I~I~,
II~'~
.~,~
~o".~.
~,"'~,''
~,.
..
o.~
.~
96
~ .',~
,
,~ ',o
.-~
<3
:o
120
0.0
2.0
.i .0
RANGE
6.0
8.0
I0
(KM)
(a)
RAY
t'~/
TRACE
PANCE
ZS-SM
(KM)
(b)
Fig.
3.
(a) I F D at 5 kHz in a surface duct, showing leakage in the form of downward beams
emerging at every cycle. (b) Corresponding ray trace.
180
C. H. Harrison
plot) for a surface duct. There are strong similarities in the shadow zone
boundaries, but the sharp-edged shadow envisaged by ray exponents is
clearly a myth. This area is partially covered by duct leakage which clearly
includes downward-propagating beams repeating at the cycle distance and
emerging from the duct just before the ray turning points at ranges of 1"3, 4.6
and 7.6 km.
4.2 S N A P and S U P E R S N A P
Normal-mode models are relatively robust, and they have the potential
advantage of being able to calculate intensity at any range or depth without
making computations at all intermediate ranges. Computation time consists
of the 'overhead' of calculating the modes, which increases in proportion to
the number of depth points, the number of modes and the uncertain but
usually small number of iterations, and then the mode summation for each
chosen range and depth point.
For most purposes the discrete mode sum is adequate. However, there are
several weaknesses in deep water apart from the large number of modes and
the convergence problems already mentioned. As a further extension of the
analogy with the violin string which was introduced in section 3.3, a normal
mode is a vertical standing wave caused by upward- and downward-going
waves interfering. The amplitudes of the two waves need to be comparable
to form a true standing wave, and this cannot be so until there have been at
least a few reflections or refractions to turn the down-going rays around.
Therefore the normal mode solution is incomplete for ranges shorter than
the cycle distance 9a for each mode (or ray family).
There are a number of other important effects that cannot be handled by
discrete normal mode models, essentially because the effects require lossy
modes to be included. Lossy modes form a continuous, rather than discrete
spectrum, and this requires evaluation of a branch line integral as well as the
sum. An important return at low frequencies in deep water at ranges before
the first bottom-refracted arrival is bottom reflection. Reflections from a
density discontinuity where velocity is more or less continuous are
necessarily lossy because the Rayleigh reflection coefficient is a constant
independent of angle, and there is no critical angle. The reflections therefore
do not feature in the usual mode sum. Nevertheless, these reflections may be
quite significant, as seen in the comparison of S U P E R S N A P with S A F A R I
(Fig. 4). This example has a water depth of about 1000 m. At ranges beyond
about 10km there are many deep bottom-refracted paths, and the good
agreement implies that a discrete sum is perfectly adequate. At shorter
ranges S A F A R I shows many rapidly interfering b o t t o m reflections which
are not seen in S U P E R S N A P .
50.0
181
SUPERSNAP
I'LRAN,COH
SD30.OOM
RD40.OOM
F
50.OOHZ
SS024
&
n
70.0 i
'JI
0
._I
90.0
110.0
0.0
15 .0
RANGE
30.0
45.0
GO.O
(KM)
(a)
SAFARI-TLRA
SD30.OOM
RD40.00M
F
50.00HZ
50 0
60
70 0
Q
U')
O
.-1
80
90
I00
110
0
0.0
15.0
30.0
RANGE {KM)
(b)
45.0
60.0
Fig. 4. (a) SUPERSNAP and (b) S A F A R I showing the neglected short-range reflections
due to density differences at the sea bed.
182
C. H. Harrison
Shear effects have been included in normal mode models, 67 but in many
environments of interest lossy modes are again important, and these can be
handled more faithfully by SAFARI. For instance, the shear critical angle is
always smaller than the compression critical angle (since shear speed is less
than compression speed) so that the spectrum includes a set of decaying
modes, which is neglected by the discrete sum, but which may nonetheless be
significant.
The adiabatic approximation can be used for 'slowly varying' environments, but it is sometimes difficult to see what this means in any other way
than the statement that there is no mode coupling. In discussing applications
of ray invariants, the author of Ref. 21 looked at the equivalence between
adiabatic modes and ray angles and their joint conditions for validity in a
reflecting duct. Rays become steeper in shallow water in a reversible way,
provided that the reflecting surfaces are smooth and their shapes change
slowly. This is equivalent to the adiabatic approximation. It is clear that in
crossing a ridge, say, the process is not reversible if the critical angle is passed
at any stage; it is also clear that undulations in between bottom bounces will
produce a virtually random additional angle to the ray. Therefore the
condition is roughly that ray angles (at the sea bed) should always be
considerably greater than the slope of the sea bed, so that there are many
bounces within a small change of environment.
4.3 SAFARI
2nN=AKrF
N =fc-~ (1 c _ ~ )
(16, 17)
183
because the Green's function has been matched at all the layer boundaries
and its form is known in each layer so that it is easily evaluated anywhere
without starting from scratch. Without increasing N the only way of
improving the resolution in deep water where there is a large number of lossfree modes (each corresponding to a spike in wavenumber, i.e. a pole in the
complex wavenumber plane) is to offset the line of integration from the real
wavenumber axis by adding a small imaginary part to K. This has the effect
of moving the integration line further from the poles (which are on or very
close to the real axis) so that the integral appears smoother; consequently
resolution with the same Nis improved. A criterion has been given 4s for the
offset which can be written in terms of dB per wavelength, J, as
J = 60cl/frr
(18)
C. H. Harrison
184
(a)
SAFARI
20.Q
SO=
40.0H7
50,OH
RD =
IOO.OM
15.0
C3
IO,O
z
c~
-5
Ld
5.O
O.O
4188
6938 8
9688.8
(b)
12438.8
15188.8
I ?938 8
(10"*-5)
SAFAR]-TLRA
50.0
60.0
70.0
80.0
90.0
I00.0
110.0
0.0
15.0
RANGE
Fig. 5.
30.0
45.0
GO.O
IKMI
The effect of altering the integration line offset from the real K-axis: (a), (b) small
offset; (c), (d) optimum offset; (e), (f) large offset.
(i.e. fixed N) the user may trade range for velocity contrast or frequency. At a
fixed frequency this usually means that he or she has to open up the phase
velocity limits Cl and c2 to accommodate the real environment. This puts a
restriction on the maximum range. When there is an abrupt density change
at the sea bed the wavenumber plot (integrand) often has significant
amplitudes for phase velocities well above the maximum velocity in the
sediment (i.e. wavenumbers much lower than 2rcf/cB), and a suitable phase
velocity must either be found by trial and error or set to infinity.
An interesting check for SAFARI in a non-trivial environment is given by
an analytical image calculation. Imagining source images in the multiple
surface and bottom image planes at depths -I-2nil +_ z, one can group the
185
SAFARI
40.0
(c)
0
* 30.0
0
20.0
E3
Z
13~ 1 0 . 0
t----
0.0
4188.8
6938.8
9688.8
12438.8
15188.8
?938.8
(10"*-5)
(d)
SAFARI
50.0
SDRDF -
60.0
-TLRA
50.OOM
IO0.OOM
40.00HZ
70.0
o
ul
80.0
o
_3
90.0
100.0
110.0
.0
51.0
RANGE
30.0
45.0
60.0
(KM}
Fig. 5.--contd
sources according to the number of bottom interactions (in isovelocity
water). The true source and one image (in the surface) have no bottom
interactions and produce a Lloyd's mirror effect. The next four images (or
four rays) have one bottom interaction (but different numbers of surface
interactions). The next four have two bottom interactions, and so on. It is
assumed that the angle differences between each order of reflection will be so
great that very rapid oscillations will result, but that they can be neglected
because usually one order of reflection dominates for most ranges. Thus the
main point o f interest is the remaining relatively slow beats in the spatial
pattern. The transmission loss for each reflection order is easily shown to be
TL = - 1 0 l o g P
P = 16(R"/r') 2
(20)
186
C. H. Harrison
SAFARI
40.0
(e)
0
30.0
F =
40.OHZ
SD=
50.OM
RD=
~00. i
20.0
L--%
Z
El". I0.0
0
W
Z.
0.0
4~88.a
6938
9688.8
(0
12438.8
15188.8
7938.8
(I0"*-5)
SAFAR]
50.0
SDRDF
60.0
--rLRA
50.OOM
lO0.OOM
aO.OOHZ
70.0
,4
ul
c3
80.0
96.G
lO0.G
110.0
.C
15.0
30.0
RANGE
45.0
60 .0
IKMI
Fig. 5.--contd.
for n > 1, where
r '2
= (2nil) 2 + r 2
a, = 2knHz,/r'
(21)
a s = 2knHz,/r'
as, = k z : , / r '
(a)
187
o
i
N.0
SD-
50,,
N-t
RD-
100-,,
[]
N-2
-4 0 0 0 ~
- lO.OdB
u3
If}
0
J
50H~
0
O3
,
vo
40
20
RANGE
(b)
SAFARI
50.0
G O .O
m
1:3
70.O
80.0
O
J
60
(KH)
SORDF
-TLRA
50.OOM
IO0.OOM
50.OOHZ
9oo
lO0.O
I|O.O
O.O
5.0
RANGE
Fig. 6.
30.0
45.0
60.0
(KMI
An image calculation (a) used as a diagnostic tool to compare with SAFARI (b).
can at least partly be accounted for by the neglected beating between paths
with different numbers of bottom reflections. Other examples have shown
that the image calculation is extremely useful for filling in the short range
reflections which are omitted by other models.
4.4 PAREQ and IFD
There is a considerable literature on the parabolic equation and its
shortcomings, s1.99-12 The PE has a fundamental elevation angle
188
C. H. Harrison
c~
_c)
oO.t~0
12.00
6.00
lB,O0
RANGE
~4.00
30.00
36.00
42,00
(KM)
(a)
PAREQ-CONDR
0.0
50.OOHZ,
SO-
(B)
15.00M
....."~.'.! "';'"
'
i'"" - "
'.~
I000.0
2000.0
3000.0
W
0
4000.0
5000.0
0.0
8 .0
12 .O
18 .0
RANGE
t
24 .0
J
30.0
J
36 .0
l
42 .0
(KMI
(b)
Fig. 7. Angle or phase errors in the parabolic equation, illustrated by the discrepancies in
range to the first bottom refracted arrival at the sea surface. (a) GRASS ray trace; (b) PAREQ.
50.OOHZ, SO.
189
IS.OOM
CONOR
"
.0
~4o00
"
":.,~
'
-~L\ ~ .,, .
L- 1600
(B)
"
i,
~.~
,''! ~-
0.0
9.0
RANGE
"
18.0
(KM)
27.0
36.0
45.0
(c)
0.O
50.OOHZ, SO-
IS.OOM
'~ ~ ~
.....
CONDR
:.
:('
(B)
1600.0
"l- 2400.0
p-
8_
W
0
4000~0
0.0
S.0
RANGE
18.0
27.0
36.
45.0
IKMI
(d)
Fig. 7.--comd
(c) I F D (N); (d) IFD (W). A strongly refracting sediment layer between 4000
and 5000 m, is assumed.
190
C. H. Harrison
191
192
C. H. Harrison
(23)
lid"
Hw
"(lib + Hw)
where the subscripts refer to the surface duct (d), water column (w) and
sediment (b), and H is the appropriate layer thickness.
The 'effective angle' is defined, rather loosely, as the steepest angle that
succeeds in making a significant contribution for the given source-receiver
combination at the given range. It is especially useful in defining the acoustic
effects of the environment because it can be compared directly with angle
limits of models or translated into wavenumber limits as appropriate. There
are, of course, many complications, particularly at low frequency, such as
near-boundary effects, which need to be borne in mind as well.
At a single frequency there is a curve in the r-O plane which defines the
angles 0elf that must be catered for at each range. Examples are shown in
Figs 8(a) and (b) for low and high frequency respectively. The region that
must be covered is above and to the left of the curve, i.e. for a given range r,
all 0 < [01 < Ocff(r).
With a dense sediment there will be reflections at angles up to 90 , but at
extremely short range these will be overpowered by the direct path. The
effective angle will therefore be smaller than 90 depending on the detailed
geometry, but steep angles will become significant at ranges comparable
with the water depth. F r o m then on the effective angle will fall to the critical
angle and level out after a range of about the cycle distance 9s (2Hcot 0c). If
the energy is spread more or less uniformly in angle each contribution will
drop out at a range where its total loss, RL, exceeds a certain level. The
relation between R L and range r is easily calculated by considering the
absorption in transitting the sediment, and for small grazing angles, 0,
R L = rfaO2/Hc '
(24)
where f, a, H and c' are frequency, absorption in dB per wavelength, water
193
Range
~'J . . . . . . . . . . . . . . . . . . . . . . . .
o l a f
....
n/2[..."""'~"=2H cotO~
T-d
fu
(a)
o
Range
Oduct
0~
rJ2
(b)
Fig. 8. The regime defined by the environment in terms of a single parameter, the "effective
angle', versus range for (a) low and (b) high frequencies.
depth and velocity gradient in the sediment respectively. If the drop-out level
is taken as a constant for all contributions then the range is
r w. H c ' / f ~ O 2
(25)
and has already been calculated for mode stripping. 17 If there is a surface
duct (above cut-off) the effective angle reduces finally to a value defined by
the velocity contrast and depth of the duct.
Naturally, energy partitioning between surface duct, water column and
sediment-refracted arrival must also b e considered. At low frequencies
where the surface duct cannot support any modes its energy contribution is
zero so that the surface duct can be ignored. At higher frequencies, despite
the low spread of angles, there may be relatively high energy density in the
duct because of its limited depth. In extreme circumstances this may make it
possible to ignore the bottom returns despite their high angles. It is difficult
to invent a general rule, though, because the relative strengths of the various
returns change in an arbitrary manner. The best way to find out is to perform
a pilot run to see which returns dominate and where, and then to use a ray
trace to estimate the appropriate angles.
C. H. Harrison
194
Range
S3
S~
Range
S 3 Si
. ........... Parabolic
Equation
Sa
I
I
_~
S~
S2
Normal
Mode
..-"
Si
Equation I
Normal
Mode
GRASS
7
Fig. 9.
vr ~ { 1 -2c o s 0 ) ~l
(a)
~2
fl
(b)
Thc areas of coverage given by G R A S S {ray),normal-mode parabolicequation and
SAFARI models for (a) low and (b) high frequencies.
(26)
195
(27)
(28)
This is shown by the dotted line in Fig. 9. The user can choose any point on
this dotted line to define 0ef f and rr, and so he has a choice of the rectangular
boxes of validity defined by S 1, S 2, $3 etc. However, he does not have a free
hand because the answers will be incorrect or suspect if the effective angle in
Fig. 9 does not cover the required angle shown in Fig. 8, so there is always a
limit on range.
A possible, although tedious, way ofextending the range of SAFARI is to
calculate loss for short extensions in range from non-zero starting points in a
piecewise fashion. Another way is to calculate short-range TL from a high
velocity contrast run, and then calculate long-range TL from a separate low
velocity contrast run. The overlap is usually well behaved.
5.5 Angles covered by PAREQ and IFD
The parabolic equation is valid for large ranges and angles below some limit
for the implementation regardless of frequency. The effect is a flat but hazy
cut-off, as shown by the solid line in Fig. 9. At short ranges there may be
phase errors, and at very short ranges there are guaranteed angle violations
unless the receiver is at the same depth as the source. The practical limitation
of computation time (and ultimately numerical accuracy) is shown by the
dotted line on the right.
5.6 General comments on coverage
At short range the only wave treatment to give reliable coverage is SAFARI.
This is complemented by and overlaps with SUPERSNAP at longer ranges.
At frequencies where SAFARI becomes too expensive to run there is no other
model that can handle short-range returns, and simple image calculations
may be able to fill the gap. In a strongly range-dependent environment where
neither SAFARI nor S U P E R S N A P is applicable there is a strong possibility
of a hole in coverage at high frequencies, which becomes a certainty at low
frequencies. One hopes that, in practice, the environment only requires small
angles that either GRASS or PE can handle.
196
c. H. Harrison
6 CONCLUSIONS
A review of four types of propagation model has been given including
G R A S S (rays), S N A P and S U P E R S N A P (normal mode), S A F A R I and
F F P (Green's function), and P A R E Q and I F D (parabolic equation).
References have been m a d e to various model comparisons. Brief
descriptions of the models and the environments to which they apply have
led to a discussion of their limitations in terms of validity, computation time
and tuning. Finally, an attempt has been m a d e to map out regions of
applicability and to present a way of choosing models in practice.
ACKNOWLEDGEMENT
A substantial part of the work on which this review is based was carried out
under contract to the Admiralty Research Establishment (UK).
REFERENCES
1. Urick, R. J., Principles of Underwater Sound. McGraw-Hill, New York, 1967.
2. Tolstoy, I. & Clay, C. S., Ocean Acoustics: Theory and Experiment in
Underwater Sound. McGraw-Hill, New York, 1966.
3. Clay, C. S. & Medwin, H., Acoustical Oceanography. John Wiley, New York,
1977.
4. Officer, C. B., Introduction to the Theory of Sound Transmission. McGraw-Hill,
New York, 1958.
5. Kinsler, L. E,, Frey, A. R., Coppens, A. B. & Sanders, J. V., Fundamentals of
Acoustics. John Wiley, New York, 1982.
6. Weinstein, M. S. & Henney, A. G., Wave solution for air-to-water sound
transmission. J. Acoust. Soc. Am., 37 (1965)899-901.
7. Urick, R. J., Noise signature of aircraft in level flight over a hydrophone in the
sea. J. Acoust. Soc. Am., 52 (1972) 993-9.
8. Chapman, D. M. F., The effects of seabed interaction on air-to-water sound
transmissions. J. Acoust. Soc. Am. 82 Suppl. 1 (1987) DDD2.
9. Harrison, C. H., Propagation in air using SAFARI (unpublished).
10. Schmidt, H., Modelling of pulse propagation in layered media using a new fast
field program. In Hybrid Formulation of Wave Propagation and Scattering, ed.
L. B. Feisen. Martinus Nijhoff, Dordrecht, 1984, pp. 337-56.
11. Meeks, M. L., Radar Propagation at Low Altitudes. Artech House, MA, 1982.
12. Hamilton, E. L., Geoacoustic modeling of the sea floor. J. Acoust. Soc. Am., 68
(1980) 1313--40.
13. Rauch, D., Experimental and theoretical studies of seismic interface waves in
coastal waters. In Bottom Interacting Ocean Acoustics, ed. W. A. Kuperman &
F. B. Jensen. Plenum PresS, New York, 1980, pp. 307-27.
14. Crowther, P. A., Acoustical scattering from near surface bubble layers. In
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
197
198
C. H. Harrison
36. Marsh, H. W. & Schulkin, M., Shallow water transmission. J. Acoust. Soc. Am.,
34 (1962) 863.
37. Lindop, P. M., The deduction of approximate values of sediment depth from
propagation loss measurements. In Ocean Seismo-Acoustics, ed. T. Akal
& J. M. Berkson. Plenum Press, New York, 1986, pp. 243-52.
38. Harrison, C. H. & Cousins, P. L., A study of propagation loss dependence on
sediment layer thickness using the fast field program. In Ocean SeismoAcoustics, ed. T. Akal & J. M. Berkson. Plenum Press, New York, 1986,
pp. 139-48.
39. Urick, R. J. Sound propagation in the sea. Peninsula Publishing, Los Altos, 1982.
40. Etter, P. C., Underwater acoustic modelling techniques. Shock and Vibration
Digest, 13 (1981) 11-20.
41. Etter, P. C., Underwater acoustic modelling techniques. Shock and Vibration
Digest, 16 (1984) 17-23.
42. Etter, P. C., Numerical modelling techniques in underwater acoustics, aT.
Acoust. Soc. Am., 82, Suppl. 1 (1987) UU1.
43. Cornyn, J. J., GRASS: a digital-computer ray-tracing and transmission-lossprediction system. Volume I--overall description. Nav. Res. Lab., Report No.
7621, 1973.
44. Cornyn, J, J., GRASS: a digital-computer ray-tracing and transmission-lossprediction system. Volume II--user's manual. Nav. Res. Lab., Report No.
7642, 1973.
45. Bartburger, C. L., P L R A Y - - a ray propagation loss program. Nav. Air. Dev.
Center, Report No. NADC-77296-30, 1978.
46. Jensen, F. B. & Ferla, M. C., SNAP: The SACLANTCEN normal-mode
acoustic propagation model. SACLANT Memo SM~121, 1979.
47. Porter, M. & Reiss, E. L., A numerical method for ocean-acoustic normal
modes. J. Acoust. Soc. Am., 76 (1984) 244-52.
48. Schmidt, H., S A F A R I User's Guide. SACLANT ASW Research Centre, La
Spezia, Italy, 1987.
49. Schmidt, H. & Jensen, F. B., A full wave solution for propagation in
multilayered viscoelastic media with application to Gaussian beam reflection
and fluid-solid interfaces. J. Acoust. Soc. Am., 77 (1985) 813-25.
50. Schmidt, H. & Glattetre, J., A fast field model for three-dimensional wave
propagation in stratified environments based on the global matrix method. J.
Acoust. Soc. Am., 78 (1985) 2105-14.
51. Kutschale, H. W., Rapid computation by wave theory of propagation loss in
the Arctic Ocean. Report No. CU-8-73. Columbia University, Palisades, New
York, 1973.
52. Di Napoli, F. R., Fast field program for multilayered media. NUSC Tech.
Report 4103, Naval Underwater Systems Center, 1971.
53. Tappert, F. D., The parabolic approximation method. In Wave Propagation
and Underwater Acoustics, ed. J. B. Keller & J. S. Papadakis, Springer-Verlag,
Berlin, 1977, pp. 224-87.
54. Jensen, F. B. & Krol, H. R., The use of the parabolic equation method in sound
propagation modelling. Report No. SM-72, SACLANT ASW Centre, La
Spezia, Italy, 1975.
55. Lee, D. & Botseas, G., IFD: an implicit finite-difference computer model for
solving the parabolic equation. NUSC Tech. Report No. 6659, New London,
CT, 1982.
199
56. Botseas, G. & Lee, D., IFD: wide angle capability. NUSC Tech. Report No.
6905, New London, CT, 1983.
57. Spofford, C. W., The FACT model, Vol. 1. AESD, Office of Naval Research,
Report No. 109, 1974.
58. Yarger, D. F., The user's guide for the RAYMODE propagation loss program.
NUSC Tech. Memo. No. 222-10-76, 1976.
59. Keller, J. B. & Papadakis, J. S., Wave Propagation and Underwater Acoustics.
Springer-Verlag, Berlin, 1977.
60. Brekhovskikh, L. M. & Lysanov, Yu., Fundamentals of Ocean Acoustics.
Springer-Verlag, Berlin, 1982.
61. Brekhovskikh, L. M., Waves in Layered Media. Academic Press, New York,
1980.
62. Weston, D. E., Acoustic flux methods for oceanic guided waves. J. Acoust. Soc.
Am., 68 (1980) 287-96.
63. Weston, D. E., Acoustic flux formulas for range-dependent ocean ducts. J.
Acoust. Soc. Am., 68 (1980) 269-81.
64. Harrison, C. H. & Ainslie, M. A., The suitability of the WKB method to fast
numerical acoustic pressure calculations. In Acoustics and Ocean Bottom, ed.
A. Lara, C. Ranz & R. Carbo, Federation of the Acoustic Societies of Europe,
Imprenta Moderna, 1987.
65. Biot, M. A. & Tolstoy, I., Formulation of wave propagation in infinite media by
normal coordinates with an application to diffraction. J. Acoust. Soc. Am., 29
(1957) 381-91.
66. Porter, M. B. & Bucker, H. P., Gaussian beam tracing for computing ocean
acoustic fields. J. Acoust. Soc. Am., 82 (1987) 1349-59.
67. Miller, J. F. & Ingenito, F., Normal mode FORTRAN programs for
calculating sound propagation in the ocean. NRL Memo. Report No. 3071,
1975.
68. Porter, M. B. & Reiss, E. L., A numerical method for bottom interacting ocean
acoustic normal modes. J. Acoust. Soc. Am., 77 (1985) 1760--7.
69. Ferla, M. C., Jensen, F. B. & Kuperman, W. A., High frequency normal-mode
calculations in deep water. J. Acoust. Soc. Am., 72 (1982) 505-9.
70. lngenito, F. & Wolf, S., Acoustic propagation in shallow water overlying a
consolidated bottom. J. Acoust. Soc. Am., 60 (1976) 611-7.
71. Kuperman, W. A. & lngenito, F., Attenuation of the coherent component of
sound propagating in shallow water with rough boundaries. J. Acoust. Soc.
Am., 61 (1977) 1178-87.
72. Labianca, F. M., Normal modes, virtual modes, and alternative representations in the theory of surface-duct sound propagation. J. Acoust. Soc. Am., 53
(1973) 1137-47.
73. Stickler, D. C., Normal-mode program with both discrete and branch line
contributions. J. Acoust. Soc. Am., 57 (1974) 856-61.
74. Bartberger, C., Comparison of two normal-mode solutions based on different
branch-cuts. J. Acoust. Soc. Am., 61 (1977) 1643.
75. Williams, A. O. Pseudoresonances and virtual modes in underwater sound
propagation. J. Acoust. Soc. Am., 64 (1978) 1487-91.
76. Tindle, C. T., Virtual modes and mode amplitudes near cut off. J. Acoust. Soc.
Am., 65 (1979) 1423-8.
77. Tindle, C. T., Stamp, A. & Guthrie, K. M., Virtual modes and the surface
boundary condition in underwater acoustics. J. Sound Vibr., 49 (1976) 231--40.
200
C. H. Harrison
201
95. Davis, J. A., White, D. & Cavanagh, R. C., NORDA Parabolic Equation
Workshop. Report TN-143, Naval Ocean Research and Development
Activity, NSTL Station, MS, 1982.
96. Quality assessment of numerical codes. Special session in: J. Acoust. Soc. Am.,
80, Suppl. 1 (1986) I, R.
97. Numerical solutions of two benchmark problems. Special session in: J. Acoust.
Soc. Am., 81, Suppl 1 (1987) Q.
98. Tindle, C. T. & Weston, D. E., Connection of acoustic beam displacement,
cycle distances and attenuations for rays and normal modes. J. Acoust. Soc.
Am., 67 (1980) 1614-22.
99. Fitzgerald, R. M., Helmholtz equation as an initial value problem with
application to acoustic propagation. J. Acoust. Soc. Am., 57 (1975) 839-42.
100. McDaniel, S. T., Propagation of normal modes in the parabolic approximation. J. Acoust. Soc. Am., 57 (1975) 307-11.
101. McDaniel, S. T., Parabolic approximations for underwater sound propagation. J. Acoust. Soc. Am., 58 (1975) 1178-85.
102. Palmer, D. R., Eikonal approximation and the parabolic equation. J. Acoust.
Soc. Am., 60 (1976) 343-54.
103. Claerbout, J. F., Fundamentals of Geophysical Data Processing. McGraw-Hill,
New York (1976), pp. 206-7.
104. Thompson, D. J. & Wood, D. H., A post processing method for removing
phase errors in the parabolic equation. J. Acoust. Soc. Am., 82 (1987) 224-32.
105. Pierce, A. D., The natural reference wavenumber for parabolic approximations in ocean acoustics. Comput. Maths. Applics, 11 (1985) 831-41.
106. Harrison, C. H., A formula for reference sound speed (unpublished).
107. Weston, D. E., Intensity-range relations in oceanographic acoustics. J. Sound
Vibr., 18 (1971) 271-87.