ref5+_Mode coupling by internal waves for multimegameter acoustic propagation in the ocean (Colosi, John A. Flatté, Stanley M)

Mode coupling by internal waves for multimegameter acoustic
propagation in the ocean

John A. Colosia)
Scripps Institution of Oceanography, University of California, La Jolla, California 92093-0225
Stanley M. Flatté
Physics Department and Institute of Marine Sciences, University of California, Santa Cruz,
California 95064
~Received 20 November 1995; accepted for publication 14 February 1996!

We have performed broadband parabolic-equation simulations with and without sound-speed
fluctuations induced by internal waves. The simulations have a center frequency of 75 Hz, a
bandwidth of 30 Hz, and propagation range R of 1000, 2000, and 3000 km. In these cases it is found
that long-range acoustic propagation through internal waves is strongly nonadiabatic. In terms of
modal travel times, low modes have a negative bias ~they have a higher effective group speed than
without internal waves! because they couple into higher, faster modes, while the higher modes show
a positive bias, indicating preferential coupling into lower, slower modes. The lowest modes show
the least travel-time spread and bias, and these quantities increase rapidly with increasing mode
number. Empirically and approximately it is found that bias grows like R 2 and spread grows like
R 3/2. This average slowing down and spreading of the higher modes causes a depth broadening of
the reception finale. The modeled modal power distributions over frequency at multimegameter
ranges are markedly different from the initial distributions at the source. Power is distributed
roughly equally across the 30-Hz frequency band for each mode, with 5.6-dB scintillations
consistent with an exponential probability distribution function for intensity. In spite of the dramatic
spread and bias in the higher modes, it is found that a synthesis of these modes results in coherent
wavefronts, whose characteristic timing fluctuations at the 3000-km range are two orders of
magnitude less than those of the corresponding modes. The modal spreads found in these
simulations imply limits to the precision with which modal travel times can be measured using
standard techniques. A nonstandard approach, using the geometric mean of independent realizations
after transformation to the frequency domain, effectively averages the phase fluctuations and
eliminates much of the spread. The strong modal coupling nonetheless suggests that nonadiabatic
modal inversions will be necessary to do modal tomography at multimegameter ranges. © 1996
Acoustical Society of America.
PACS numbers: 43.30.Pc, 43.30.Bp @JHM#
INTRODUCTION sions using modal travel-time data.2–9 Of particular impor-

tance to the field of acoustic tomography are the following
In any wave propagation problem both geometrical op- questions: Can the propagation of a particular mode be mod-
tics and mode representations of the wave field can be ren- eled using the adiabatic approximation in which mode cou-
dered. In broadband ocean acoustic wave propagation we pling is negligible? How much fluctuation do we expect in
generally have artificially partitioned the acoustic arrival pat- modal travel times from internal-wave effects, and how do
tern into two regions: one where wavefronts are apparent and random internal-wave sound-speed perturbations affect mode
where geometrical optics may be applied, and another where identity as modes propagate from the source to the receiver?
wavefronts are not apparent, suggesting that a modal repre- If modal energy is coupling randomly between a population
sentation may be more useful. Of fundamental interest is the of modes rather than traveling adiabatically, then the energy
interrelationship of these two representations.1 measured at a receiver array as being in a particular mode
Researchers in acoustic tomography have traditionally has propagated as other modes in its journey from the source.
used the information in wavefronts and have made little use In this case modal identity is complicated and the manner in
of other parts of the arrival pattern, save perhaps the final which a mode has sampled the ocean cannot be easily deter-
cutoff. This is partly due to the fact that most tomography mined.
experiments have been at higher frequencies ~250 Hz!, and The use of matched-field processing for the purpose of
single receivers or sparse vertical arrays have been used. identifying sources or improving signal-to-noise has tended
With the recent deployment of denser receiving arrays, ef- to involve representations of the acoustic field in terms of
forts have been made to enhance acoustic tomography inver- modes, since adiabatic mode theory provides a quick solu-
tion to the wave equation when multiple source and/or re-
a!
Now at the Applied Ocean Physics and Engineering Department, Woods ceiver locations are involved.10 Important issues in matched-
Hole Oceanographic Institution, MS #11, Woods Hole, MA 02543. field processing are therefore similar to those in tomography;
3607 J. Acoust. Soc. Am. 100 (6), December 1996 0001-4966/96/100(6)/3607/14/$10.00 © 1996 Acoustical Society of America 3607
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 129.105.215.146 On: Sat, 20 Dec 2014 16:45:16
how good is the adiabatic approximation, and to what degree TABLE I. Canonical sound-speed parameters.
do internal waves couple modes and render matched filtering
Sound-speed
ineffective. parameters Value
Clearly the efforts in both acoustic tomography and in
matched-field processing require an investigation of the ef- e 0.00238
c axis 1478.33 ~m/s!
fects of internal waves on acoustic mode propagation to bet- B ss 485 ~m!
ter understand what limitations may exist. z axis 700 ~m!
Early work in the field of mode propagation was done
by Munk and Wunsch,1 and Desaubies analyzed the effects
of mesoscale features on mode propagation.11,12 Dozier and
spectrum. We choose the canonical sound-speed parameters
Tappert13,14 analyzed the modal energy redistribution caused
based upon a fit to an experimental profile from the SLICE89
by internal waves, and recent papers by Lynch15 and
experiment.18 In this study we are calculating statistical mo-
Traykovski16 have described and compared internal-wave-
ments of the modal acoustic field and therefore we generate
induced modal and ray travel-time fluctuations for a short-
independent realizations of the stochastic internal-wave dis-
range, shallow-water experiment. However, most mode
placement z~r!, and we do not model the internal-wave varia-
propagation theories for internal waves either rely on the
tions in time.
dubious adiabatic approximation, or make other untested ap-
proximations.
This paper is intended to treat internal-wave influences
on mode propagation using broadband parabolic-equation A. Internal-wave displacements
~PE! numerical simulations, and incorporating sound-speed
The method of generating GM internal-wave perturba-
perturbations obeying the Garrett–Munk ~GM! spectral
tions is discussed in detail in a previous paper,19 but we offer
model at the reference GM energy level. We decompose the
here a brief description. We model the internal-wave dis-
broadband PE data into individual modal arrivals to study the
placement z~r! as a superposition of linear modes, W j (k,z),
statistics of modal travel-time fluctuations and modal energy
with random phases and amplitudes. If a two-dimensional
distributions at 1000, 2000, and 3000 km propagation ranges.
slice of the three-dimensional displacement field is consid-
We examine the case of a canonical sound-speed profile, a
ered, then
75-Hz center frequency, and a 30-Hz bandwidth. The sensi-
tivity of these results to changes in frequency, or mean
sound-speed profile or internal-wave spectra are not ad- z ~ x,z ! 5 z 0 •N G
dressed; one should interpret these results in that light.
The presentation of this paper is as follows. In Sec. I we
discuss numerical modeling techniques. In Sec. II we show
•Re F( ( (
jm km Nu
j51 k5k 0 i51

G
G̃ j ~ k, u i ! W j ~ k,z ! e ikx cos~ u i ! ,
single-frequency results which display the evolution of ~4!

modal energy and phase as a function of range. Section III
has broadband results for low mode numbers which describe where u is the angle between the internal-wave plane wave
the transmission finale and Sec. IV discusses the modeled and the ocean slice; the functions G̃ j (k, u ) are complex
modal travel-time and intensity statistics. Section V ad- Gaussian random variables obeying the GM internal-wave
dresses assorted signal processing issues. Section VI presents spectrum; and N G is a normalization factor. The internal-
a summary of the present work. wave modes, W j (k,z), are determined using an exponential
buoyancy profile ~see Fig. 1 and Table II for parameters!.
I. NUMERICAL SIMULATION METHODS The exponential buoyancy profile is also a fit to SLICE89
data.18
We represent the ocean sound-speed field as the sum of Realizations of the internal-wave field were generated
a deterministic mean profile plus a stochastic perturbation using 16 horizontal wave numbers equally distributed in the
induced by internal waves: range d k/2<k<12 d k/2 ~cpkm!, with d k 5 161 ~cpkm!, and
the set of vertical mode numbers is j51210,11,13,...,49.
c ~ r! 5c̄ ~ z ! 1 d c ~ r! ~1!
Ten horizontal azimuths u ~N u510!, were used every 36°
with from 0° to 324°. Other parameters of the simulation
c̄ ~ z ! 5c axis~ 11 e ~ e 2 h 1 h 21 !! , h 52 ~ z2z axis! /B ss are j 53, n 0 B54219 m cph, vi 51 cpd. To match the
*
~2! GM spectral level the constant z0 is chosen so that
^z~z ref!&rms5z057.3 m where N~z ref!53 cph. These param-
and eters are summarized in Table II. To form an extended
d c ~ r! 5 S D
] c̄
]z p
z ~ r! 5
2c axise 2 h
B ss
e z ~ r! , ~3!
internal-wave field 35 separate 16-km realizations of z are
created and patched together so as to keep ^z2& independent
of range x. The patching can be done a number of ways. We
where the mean profile c̄(z) is a Munk canonical profile17 choose to place the centers of the 16-km realizations 8 km
~see Fig. 1, and Table I for parameters!, and the random apart ~x 058 km!. At each range 0,x,x 0 , the internal-wave
perturbations dc~r! are described by the GM internal-wave displacement is given by
3608 J. Acoust. Soc. Am., Vol. 100, No. 6, December 1996 J. A. Colosi and S. M. Flatté: Mode coupling 3608
TABLE II. Internal-wave model parameters.
Internal-wave
parameters Value
z0 7.3 ~m!
Nr 3 ~cyc/h!
j* 3
f 1 ~cyc/day!
N0 4.57 ~cyc/h!
B 923.18 ~m!
x0 8 ~km!
Nk 16
dk 1/16 ~cyc/km!
k0 d k/2 ~cyc/km!
Nu 10
du 36°
$ j% 1–10,11,13,...,49
F
z ~ x,z ! 5 S D
1
Na
z 1 ~ x,z ! cos2
px
2x 0
1 z 2 ~ x 0 ! sin2
px
2x 0 S DG ,
A S D S D
~5!
px px
N a5 cos 4
1sin4 ,
2x 0 2x 0
where z1 and z2 are two different 16-km realizations of the
internal-wave field. This procedure is carried out to multi-
megameter range by randomly selecting a sequence of the
16-km realizations, and randomly multiplying them by 61
and/or randomly reflecting them about their central range. FIG. 1. The mean and fluctuation sound-speed environments used in this
Figure 1 shows the mean sound-speed and buoyancy study. The canonical ~solid! sound-speed ~upper left! and exponential buoy-
profiles, internal-wave spectra and rms internal-wave sound- ancy profiles ~upper right! used in this study with the SLICE89 experimental
profiles ~dash! from which the analytical profiles were derived. Internal-
speed perturbation as a function of depth. Note that our
wave vertical wave number spectrum ~middle left! as modeled in this study
internal-wave simulation procedure has generated horizontal ~solid!, and the theoretical Garrett–Munk ~GM! spectrum ~dash!. Internal-
wavelengths considerably larger than the maximum wave- wave, horizontal wave number spectrum ~middle right! as modeled in this
length of the internal-wave modes ~32 km!.19 study ~solid!, and the GM spectrum ~dash!. The rms sound-speed perturba-
tion from internal waves as a function of depth ~bottom!.
B. Acoustic propagation methods

The standard parabolic equation ~SPE! is used to simu- The boundary conditions on the acoustic propagation are
late the acoustic propagation: a reflecting ocean surface and an absorbing ocean bottom.
F 22ik 0
] ]2
G
1 2 1k 2 ~ r! 2k 20 F ~ r, v ! 50,
]x ]z
~6!
The reflecting ocean surface is modeled numerically using an
image ocean, so that the actual number of vertical points is
twice that listed in Table III. In order to model an absorbing
where k 0 5 v /c 0 , c 0 is a representative sound speed, and bottom a gradual loss of amplitude is imposed on F~r,v! as
k~r!5v/„c̄(z)1 d c~r!…. While this equation is a small-angle z approaches the ocean bottom. The functional form of the
approximation to the Helmholtz equation ~which cannot be imposed loss at each step is
solved accurately at multimegameter ranges! we assume that
the fluctuation quantities we are calculating will be insensi-
tive to this approximation. The initial condition is a narrow
Gaussian in depth of width one wavelength and the acoustic TABLE III. PE simulation parameters.
field at each frequency is marched forward in range using
PE
Eq. ~6!. The total wave field is obtained by Fourier synthesis, Description parameters
C ~ r,t ! 5 E P ~ v ! F ~ r, v ! e i ~ k 0 x2 v t ! d v ~7!
Number of depth points
Number of frequencies
N52048
M 5512
Number of realizations L511
where Depth spacing Dz52 ~m!
S D
Range step Dx550 ~m!
p~ v2vc! f c 575 ~Hz!
P ~ v ! 5sin ~8! Center frequency
~ 2D v ! Bandwidth D f 530 ~Hz!
Source depth z source5800 ~m!
is the source transfer function. Table III lists the parameters Reference sound speed c 051478.33 ~m/s!
for the broadband PE simulations.
H
L ~ z ! 5exp 2 b dx3exp 2
z2z b 2
azb F S D GJ
~9!
with b 50.04 and a 50.05. 20 This f orm e f f ec-
ti v ely stops any acoustic energy f rom penetrat-
ing about a 3z b .200 m abo v e the ocean bottom
zb .
C. SPE modal arrivals

One advantage of using the SPE is that it can be easily
separated using
F ~ r, v ! 5 (n A n~ x, v ! B n~ z, v ! , ~10!
where the deterministic vertical mode functions, B n (z, v ),

are obtained from the mean sound-speed profile using the
familiar equation
F ]2
]z2 G
1k̄ 2 ~ z ! 2k 2n ~ v ! B n 50, ~11!
and the B n (z, v ) obey the orthonormality relation
E 0
zb
B n ~ z, v ! B m ~ z, v ! dz5 d mn . ~12!
The stochastic horizontal functions, A n (x, v ), obey the equa-

tions
F 2ik 0
]
]x G
1k 2n ~ v ! 2k 20 A n ~ x, v ! 5 (m A m~ x, v ! C mn FIG. 2. The mode functions, B n (z,w) at 75 Hz for odd modes 1 through 9
~upper!. The group slowness curves for modes 1 through 15 ~middle!. Mode
~13! upper turning depth for frequencies 60, 75, and 90 Hz, and mode numbers 1
through 40 ~lower!.
with the stochastic mode coupling matrix
C mn 52 E 0
zb
„k 2 ~ r! 2k̄ 2 ~ z ! …B n B m dz. ~14! f̄ n ~ x, v ! 5
k 2n ~ v ! 2k 20
2k 0
x. ~19!
Rather than attempt the solution of Eq. ~13!, we solve the It is of interest to write down the adiabatic solution to
SPE Eq. ~6! for F~r,v!, calculate the B n (z, v ) mode func- these equations so that we may compare this simple analytic
tions, and obtain the A n (x, v ) functions from approach to the full solution. The adiabatic solution for the
E zb A n (x, v ) is obtained by setting the off-diagonal elements of

A n ~ x, v ! 5 F ~ r, v ! B n ~ z, v ! dz. ~15! C mn equal to zero:
0
Ã n ~ x, v ! . u A n ~ x50,v ! u e i f̃ n ~ x, v ! , ~20!
The arrival pattern for a single mode is expressed as
E
where
E
C n ~ r,t ! 5 P ~ v ! u A n ~ x, v ! u B n ~ z, v ! x 8 5x Dk 2n ~ x 8 , v !
f̃ n ~ x, v ! 5 f̄ n ~ x, v ! 2 dx 8 ~21!
x 8 50 2k 0
3e i„k 0 x2 v t1 f n ~ x, v ! … d v , ~16!
and
where we see that the modal travel time ~group-time-delay!
is given by
] f n ~ x, v !
Dk 2n ~ x, v ! .22 v 2 E 0
zb d c ~ z,x !
c̄ 3
B 2n dz. ~22!
t n ~ x, v ! 5 . ~17!
]v Figure 2 shows a few of the low-mode functions,
B n ~z,v575 Hz!, some group slowness curves, and the mode
The quantities t n (x, v ) are the important observables used in upper-turning-point depth for several modes calculated with-
modal acoustic tomography inversions.1–9 out internal-wave effects. We see from this figure that mode
For d c50, we have 1 has the largest slowness, or smallest group velocity, and
that mode 1 is localized around the sound-channel axis.
Ā n ~ x, v ! 5 u A n ~ x50,v ! u e i f̄ n ~ x, v ! , ~18!
Higher modes have smaller group slowness, and extend fur-
where ther above and below the sound-channel axis. Acoustic
FIG. 3. Single-frequency modal energy and phase from a simulation of propagation through internal waves starting from a point source. The initial modal
energy levels are indicated by dashed lines. The PE results are plotted with solid lines and adiabatic phases are plotted with dashed lines.
modes are generally most sensitive to perturbations near ^ d t 2n ~ x, v ! & 5 ^ ~ t n ~ x, v ! 2 ^ t n ~ x, v ! & ! 2 & , ~24!
their turning depths,2 and therefore we can see from the
turning-point plot and Fig. 1 that the higher modes sample where ^ t n (x, v ) & is the mean travel time of modes with in-
larger internal-wave perturbations, and that higher modes at ternal waves. We define the mode travel-time spread as the
60- and 90-Hz sample depth regions that are different by rms value of mode travel time, that is, the square root of the
more than the internal-wave vertical coherence distance. variance.
Finally a note about notation. We use a superscript l to II. SINGLE-FREQUENCY RESULTS
signify the lth realization of a stochastic variable like
In this section we present single-frequency results at 75
A ln (x, v ), fln (x, v ), or Fln (x, v ).
Hz to show the detailed evolution of the modal power and
phase as a function of range.
D. Mode travel-time fluctuations A. Modal power variations
We define the mode travel-time bias as the mean differ- Figure 3 shows one realization of the modal power evo-
ence between the travel time of a mode with and without lution u A l51
n (x) u for the first nine modes out to 3000 km.
2
internal waves, Our results qualitatively confirm the strong modal energy
exchange predicted by the theoretical work of Dozier and
^ d t n ~ x, v ! & 5 ^ t n ~ x, v ! 2 t n ~ x, v ! & , ~23!
Tappert.13,14 The evolution of modal power shows episodes
where t n (x, v ) is the travel time of the mode without inter- of rapid energy loss or gain and other regions where the
nal waves. The mode travel-time variance is energy has slow variations.
B. Modal phase variations culation shows that internal-wave-induced spreading main-
tains a significant time overlap. By 3000 km the higher
Figure 3 also shows the evolution of modal phase per-
modes show a tremendous internal-wave-induced spread, as
turbations with range, where we define the modal phase per-
well as a bias toward a later arrival time.
turbation as
We can understand this result in terms of mode cou-
D f ln ~ x, v ! 5 f ln ~ x, v ! 2 f̄ n ~ x, v ! . ~25! pling. As an idealized example consider the case where we
produce a single propagating mode, n, at the source, and this
Also shown on this plot are the modal phase perturbations
mode experiences a coupling event at some range, R 1 , which
estimated using the adiabatic approximation from Eq. ~21!.
places some energy from mode n into modes n11, and n21.
In regions where the modal power is diminished we are un-
At a second range R 2 another coupling event takes place
able to accurately track the phase and we see rapid varia-
which places the energy from modes n11 and n21 back
tions; however, it is still apparent that the adiabatic phases
into mode n. At range R we receive mode n which has three
are not fully accounting for the total phase variation. Clearly
peaks in its arrival pattern. Since in deep water low modes
the mode coupling term in Eq. ~14! has a significant phase
travel more slowly than high modes, we observe an early
and amplitude effect.
peak with time t n 5S gn R 1 1S gn11 (R 2 2R 1 )1S gn (R2R 2 ), a
late peak with time t n 5S gn R 1 1S gn21 (R 2 2R 1 )1S gn (R2R 2 ),
III. BROADBAND RESULTS: LOW-MODE NUMBERS
and a middle peak with time t n 5S gn R, where S gm is the group
We focus attention on mode numbers 1 to 10 so as to slowness of mode m. The arrival for mode n is spread in
examine the finale of the pulse transmission. Recent experi- time. Time spreading is the result of the fact that mode n did
ments at the 1000-km and 3250-km range have revealed not maintain its identity in its propagation from the source to
strong internal-wave effects at the finale.18,19,21 the receiver, but had significant correlations to modes n11
and n21. Now if we consider random coupling events
A. Mode resolvability
caused by internal waves we can see how time spreading can
Since ocean acoustic experiments must deploy limited create the complex arrival patterns seen in Fig. 5. We can
vertical arrays, issues of mode resolvability will exist due to also understand the bias in the arrival pattern since the cou-
the fact that the mode arrivals cannot be unambiguously pling of mode n into other modes need not be symmetric for
identified by their depth structure alone. Therefore, one must mode numbers greater than and less than n, and because S gn
rely on anticipated modal time separations, and well-known is not in general a linear function in n.
equations based on the adiabatic approximation exist for de-
termining when a mode can be resolved in time in the ab-
sence of internal-wave effects.1,22 C. Broadband modal arrivals: Frequency domain
We can qualitatively examine mode resolvability by
looking at the arrival patterns, uC~r,t!u2, calculated from Eq. We next examine the modal arrival pattern in the fre-
~7!. Figure 4 shows the last 2 s of the arrival pattern calcu- quency domain by plotting the modal intensity, u A n (x, v ) u 2,
lated at ranges of 1000, 2000, and 3000 km for a case with- and the modal phase f n (x, v ), for the simulations with and
out internal waves and one realization with internal waves. without internal waves at ranges 1000, 2000, and 3000 km.
At 1000 km with no internal waves, the low modes are ar- These quantities are shown in Fig. 6.
riving at similar times and one cannot see groups of modes The modal intensity distribution across frequency for the
clearly. By 3000 km the low modes have spread out in time, simulation with internal waves is markedly different from
and clear groupings are evident up to mode number 9. Nev- the initial distribution at the source ~i.e., the calculation with
ertheless, when internal-wave perturbations are present the no internal waves! and the distribution becomes increasingly
coherent mode arrival pattern is destroyed, and there is a complicated with increasing range. We display the un-
significant time overlap of the modes evidenced by the wrapped modal phase in Fig. 6, which reveals rapid phase
broadening and blunting of the transmission finale. We ex- variations. Recall that the important quantity tn depends on
pand on the modal time spreading concept in the next sec- the slope of the phase, not its absolute value. It is these rapid
tion. phase variations that give rise to time spread and it is the
mean change in slope which gives rise to a time bias as seen
B. Broadband modal arrivals: Time domain in Fig. 5.
We decompose the simulation fields, F~r,v!, using Eq.

~15! and then synthesize the mode amplitudes, A n (x, v ), us- IV. BROADBAND RESULTS: MODE STATISTICS
ing Eq. ~16!, to obtain the broadband arrival-time pattern,
C n ~r,t!, for modes 1 through 10. In Fig. 5 we plot C n ~r,t! We form statistics of modal travel time and modal in-
for the simulations with and without internal waves at ranges tensity, which summarize the simulation results and provide
of 1000, 2000, and 3000 km. At the 1000-km range we see predictions which may be compared to future experimental
that modes 1 through 10 arrive very closely in time, and that results.
internal-wave sound-speed perturbations have the effect of
A. Modal travel-time statistics
distorting and spreading the arrival-time pattern. At 2000 km
the simulation without internal waves reveals that the modes We examine the statistics of modal travel times in order
are beginning to separate in time, yet the internal-wave cal- to quantify the observed bias and spread of the broadband
where
d t ln ~ x, v ! 5 t ln ~ x, v m ! 2 t̄ n ~ x, v ! ~28!
and
] f̄ n ~ x, v !
t̄ n ~ x, v m ! 5 ~29!
]v
with f̄ n (x, v ) given by Eq. ~19!. We restrict these sums to
values of d t ln (x, v m ) which are less than 3 s.d. from the
mean, thereby eliminating spurious values associated with
the ‘‘dropouts’’ evident in Fig. 6. This selection procedure
results in a significant depletion of points for longer ranges
and higher-mode numbers. We only plot values of ^dtn & and
^dt2n & for which all of the M l are greater than 256, and for
which the d t ln (x, v m ) have a Gaussian distribution. In Fig. 7
we plot ^dtn & scaled by the square of the range, R 2, and
~^dt2n &!1/2 scaled by R 3/2, the ranges 1000, 2000, and 3000
km. The R 2 and R 3/2 scalings work well for mode numbers
less than about 25. The modal bias is negative for low
modes, indicating preferential coupling into higher, faster
modes, but the bias turns positive and large for higher modes
indicating preferential coupling into lower, slower modes.
The rms modal spread has the same magnitude as the bias
and grows with increasing mode number.
We reiterate that at present the only theory known to us
that predicts the statistics of modal travel times induced by
internal waves is the adiabatic theory, which we have seen in
Sec. II does not work well. Furthermore, it can be shown that
the adiabatic theory predicts zero modal bias, again high-
lighting the shortcomings of the adiabatic theory for long
ranges. @The proof this statement uses Eqs. ~17!, ~22!, and
~21!, and the fact that ^ d c & 50.# Finally, we note that the
adiabatic theory predicts16 that the modal rms spread scales
like R, where we have found a scaling of R 3/2 in the simu-
lations.
B. Modal scintillations
We calculate the mean log-intensity averaged across fre-
quency for the case with internal waves using
L M
1
^ i n~ x ! & 5
ML ( (
l51 m51
i ln ~ x, v m ! ~30!
FIG. 4. The last 2 s of the acoustic arrival patterns calculated from the PE and we calculate the mean log-intensity for the case without
simulations with and without internal waves for ranges of 1000, 2000, and internal waves using
3000 km. The gray scale is 0 to 235 dB, referenced to the maximum
M
intensity for each case individually. 1
i n5
M (
m51
i n~ v m ! , ~31!
modal arrivals. We estimate the modal bias given by Eq. ~23! where i n (x, v )510 log10~u A n (x, v ) u 2!. We also calculate the
by averaging over both realizations and frequency to gain scintillations in log-intensity averaged over frequency
statistics using
L M
1
1
L
1
M
^ d i 2n ~ x ! & 5 ( ( „i ln ~ x, v m ! 2 ^ i n ~ x ! & …2 . ~32!
^ d t n& 5
L (
l51 Ml (
m51
d t ln ~ x, v m ! ~26! ML l51 m51
We plot these quantities in Fig. 8. The mean log-

and we estimate the modal spread given by Eq. ~24! using intensities ^in (x)& show that an equilibrium is established
L M before the 1000-km range, that is, ^in (x)& is not a function of
1 1
^ d t 2n & 5
L (
l51 Ml (
m51
~ d t ln ~ x, v m !! 2 2 ^ d t n & 2 , ~27! x at all. This is in qualitative agreement with the theory of
Dozier and Tappert which predicts a relaxation range of 200
FIG. 5. The modal arrival patterns, C n (r,t!, at a receiver depth of 800 m for one realization with internal waves ~solid! and one without ~dash!. Mode numbers
1 through 10 are shown, for ranges of 1000, 2000, and 3000 km.
km for 75-Hz frequency.13,14 There is some evidence from model fits the data very well at both the high- and low-
Fig. 8 that modes 1 and 2 reach equilibrium with the other intensity regions and the log-normal curve clearly does not
modes at a range larger than 1000 km. The distribution of the fit. This exponential PDF is predicted by the Dozier and
mean log-intensities at long range seems to have settled on a Tappert theory.13,14 Other observations of the PDF in short
smoothed version of the distribution at the source. In the range, high-frequency ocean acoustic transmissions have
absence of dissipation, Dozier and Tappert predict an equi- looked at raylike arrivals rather than modes, and have seen
partition of model energy: that is, equal energies in all distributions more like the log-normal PDF.23 Observations
modes. This difference between the present simulations and of the PDF in a long-range, single-frequency ocean acoustic
the Dozier and Tappert prediction may be related to the bot- experiment did show an exponential PDF, which was inter-
tom loss imposed on the numerical simulation, which is not preted as due to the addition of deterministic multipath.24
treated in the theory. Our nonequipartition result also sug- Observations of optical propagation through turbulence have
gests that near-neighbor coupling is dominant.
shown PDFs that fit a log-normal convolved with an
It is uncertain if the mean intensity curve calculated here
exponential.25 Physically, the pure exponential result is ex-
represents a universal curve, whose shape is determined by
pected in cases for which the energy whose PDF is being
the bottom dissipation and mode coupling in analogy to the
investigated is the absolute-value-squared of a complex am-
case of Kolmogorov turbulence where dissipation at small
plitude that is made up of a large number of independent
scales and wave number independent energy flux give the
k 5/3 power law. Other simulations with different initial con- phasors. We can imagine one of our modes at the receiver
ditions show that an equilibrium is not reached by 3000 km. range as being made up of many independent contributions,
This is an interesting question which is beyond the scope of perhaps rays, each of which would have a different history of
the present analysis. scattering in traveling from the source to the receiver. The
The mode scintillations ~^di2n (x)&!1/2 plotted in Fig. 8 log-normal convoled with the exponential PDF comes about
also show an equilibrium with roughly 5.6-dB fluctuations, when the number of independent phasors is slowly varying,
independent of mode number. usually due to large-scale effects. In the ocean that would
Figure 9 shows probability density functions ~PDFs! of mean that the large-scale structure of the sound channel is
I n (x) and ln „I n (x)…, (I n 5 u A n u 2 ), for mode 20 at ranges of modulating the independent contributions created by the
1000, 2000, and 3000 km and model exponential and log- small-scale structure, most likely internal waves. Since our
normal PDFs. Results for other mode numbers are similar, simulations do not have any variability in the large-scale
except for mode number 1 at 1000 km and 2000 km where structure of the sound channel, we would not expect any
this mode is still approaching saturation. The exponential log-normal features in our PDFs.
FIG. 7. The modal travel-time bias scaled by R 2 ~lower panel!. The rms
modal travel-time spread scaled by R 3/2 ~upper panel!. The different ranges
are: 1000, km ~d!, 2000 km ~s!, and 3000 km ~L!. The right-hand panels
are expansions of the left-hand panels to show the low-mode numbers.
transmission finale, indicating that ray and mode variability

near the axis may be comparable.28 However, the increase in
variability for high modes is in clear contrast to the decrease
in variability seen for steep rays.
The bias is another matter. Fermat’s principle tells us
that a ray is a stationary-phase path from the source to the
receiver. In the absence of a sound channel the stationary-
phase path is a minimum-time path, and it has been shown
that this implies a negative bias.29 In the presence of the
ocean sound channel, most rays are not paths with absolute
FIG. 6. The same modal arrival patterns shown in Fig. 5 except displayed in
the frequency domain using A n (x,w). Ranges of 1000, 2000, and 3000 km
minimum travel time, but saddle points in the sense that
are shown. Phases are in radians. some nearby paths have shorter or longer travel times. In
these cases the bias may be either positive or negative.29
These principles are a result of the competition between
C. Ray and mode equivalence
travel-time corrections due to variations in geometrical path
The modal travel-time fluctuations shown in Fig. 7 can length and sound speed. No such principles exist for modes,
be compared with well-known effects of internal waves on because there is no counterpart to variations of geometrical
wavefronts or raylike arrivals.13,19,26,27 Generally steep rays path length. In our simulations we see a positive bias for
which have turning points well separated from the sound- modes with high-mode number, and a negative bias for the
channel axis experience the least internal-wave effect, and as modes with very low-mode number.
the rays become more axial the effect becomes very pro- Figure 10 shows the arrival-time patterns for several
nounced. From the mode viewpoint we see weaker effects on modes up to mode 50 for the simulations with and without
the low modes which describe the axially propagating en- internal waves. The increasing modal spread and bias as a
ergy, and larger effects at high-mode numbers which make function of mode number and range are evident. The modal
up the steeply propagating energy. Calculations of internal- arrival patterns are very complex, yet a synthesis of these
wave effects on rays reveal the same depth broadening of the complex patterns over mode number ~see Fig. 11!, i.e., the
FIG. 9. Probability distribution functions ~PDFs! of intensity and log inten-
sity for mode 20 at ranges of 1000, 2000, and 3000 km ~solid!, with model
FIG. 8. The mean log-intensity, ^ i n & ~lower panel! and the rms variation in ~PDFs! for an exponential distribution ~small dash! and a log-normal distri-
log intensity, ^ d l 2n & ~upper panel! as a function of mode number and range. bution ~large dash!.
Ranges are 1000 km ~d!, 2000 km ~s!, and 3000 km ~L!. The log-intensity
distribution at the source, l̄ n , is indicated by the ~n! in the lower panel.
L
1
total wave field, reveals a much more coherent pattern, C̄n ~ r,t ! incoh5 ^ u C n ~ r,t ! u 2 & 5
L ( u C ln~ r,t ! u 2 ,
l51
~33!
where raylike wavefronts are evident. In a companion paper
we will present the analysis of wavefront fluctuations calcu-
lated from these same simulations; briefly, at 3000 km we and the coherent average,
find wavefront timing fluctuations on the order of 20-ms rms.
This is about a factor of 50 less than the modal travel-time
spread. Traykovski and Lynch15,16 compare ray and mode
fluctuations and find about a factor of 3 reduction for a
C̄n ~ r,t ! coh5 u ^ C n ~ r,t ! & u 2 5
1
L
U( L
l51
U
C ln ~ r,t ! ,
2
~34!
300-Hz case, and they predict the reduction factor to de-

crease with decreasing frequency. However, they use ap- are often applied to the data. We show the results of taking
proximate adiabatic scattering equations which are not valid these averages in Fig. 12. The coherent average is similar to
at long ranges. Physically we attribute the observed reduc- the incoherent average except the intensity is somewhat re-
tion in timing variance for rays and modes as a result of the duced, showing that the L realizations have some coherence.
averaging over mode number and frequency required to form The incoherent average reveals the large average spreading
a ray.15,16 of the arrivals. In neither of these cases do we see much
In addition we find a wavefront bias of about 210 ms, improvement of the received signal.
which has opposite sign and is about 100 times smaller than Therefore, we consider a multiplicative noise model:
the bias at the mode number corresponding to that wave- Considering A n to be of the form:
front.
l
A ln ~ x, v ! 5 u A ln ~ x, v ! u e i f n ~ x, v ! , ~35!
V. SIGNAL PROCESSING
In this section we examine various ways of averaging where
the simulation data, in search of a way to mitigate the effects
of internal waves. First, we look at techniques that assume
additive noise. Two averages, the incoherent average, f ln ~ x, v ! 5 f̄ n ~ x, v ! 1 d f ln ~ x, v ! , ~36!
^ A n ~ x, v ! & g 5 S) L
l
A ln ~ x, v ! D 1/L
5 ^ u A n ~ x, v ! u & g e i ^ f n ~ x, v ! & , ~37!

then we average out the phase variations which dominate the
spreading of the arrival. In Fig. 12 we plot the following
quantity:
C̄n ~ r,t ! g 5 UE P ~ v ! ^ A n ~ x, v ! & g
3B n ~ z, v ! e i ~ k 0 x2 v t ! d v . U 2
~38!
While the resulting mode arrival pattern is not perfect, we

have certainly eliminated much of the spread and sharpened
up the peak considerably.
Figure 13 plots u ^ A n (x, v ) & d u 2 and ^ f n (x, v ) & from Eq.
~37!. The effect of this averaging is to smooth out the phase
variations and to give roughly equal energy across the modal
bandwidth.
It is clear from the application of this method that
d f ln (x, v ) has a nonzero mean, which implies a nonzero
travel-time bias.
VI. SUMMARY
We have presented parabolic-equation numerical simu-
lations that show that long-range acoustic propagation in the
ocean is strongly nonadiabatic in the presence of internal-
wave sound-speed fluctuations.
Single-frequency calculations at 75 Hz show that mode
coupling induced by internal waves produces significant
modal energy exchange, which proceeds at varying rates as a
function of range. These same single-frequency calculations
reveal that modal phase variations cannot be explained by a
simple adiabatic model, and therefore that mode coupling
gives rise to significant phase variations as well as energy
variations.
Our broadband calculations at 75-Hz center frequency
and 30-Hz bandwidth further exemplify strong intermodal
energy and phase variations for long-range propagation to
distances of 1000, 2000, and 3000 km, and allow us to quan-
tify modal travel-time statistics. At 3000-km range for the
first few modes we find a modal travel-time bias that is nega-
tive ~faster travel times! and that has a magnitude on the
order of 100 ms. For the higher modes we see a positive bias
~slower travel times!, with a magnitude on the order of 1 s.
At 3000-km range the modal rms travel-time spread starts at
about 200 ms and increases to values in excess of 1 s. Bias
scales like R 2 and rms spread scales roughly like R 3/2, where
R is the source–receiver range.
Modal energy distributions across frequency at ranges of
1000, 2000, and 3000 km are markedly different from the
FIG. 10. The modal arrival patterns, Cn ~r,t!, for one realization with inter-
nal waves ~solid! and one without ~dash!. Mode numbers 1, 10, 15, 20, 25,
initial excitation energy distributions at the source. We ob-
30, 35, 40, 45, and 50 are shown, for ranges of 1000, 2000, and 3000 km. serve a saturation in the distribution of modal energies aver-
aged across frequency at 1000-km range, but we do not ob-
and where d f ln (x, v ) is some phase noise, we realize that if serve the equipartition distribution predicted by the Dozier
we form the geometric average of the A ln (x, v ) over realiza- and Tappert mode-coupling theory. This discrepancy may be
tions due to the bottom attenuation treated in the simulation but
FIG. 11. A synthesis of the modal arrival patterns for one realization with internal waves ~solid! and one without ~dash! for modes 1 through 50 at a depth
of 1500 m. Ranges of 1000, 2000, and 3000 km are shown. The inset magnifies the earlier arrivals.
FIG. 12. The incoherent average, C̄ n ~r,t!incoh ~dark!, the coherent average, C̄n ~r,t!coh ~small dash!, and a nonstandard average, C̄n ~r,t) g ~large dash! of the
simulation data with internal waves plotted in the time domain. The arrivals for the simulation without internal waves are shown by the medium dashed lines.
Ranges of 1000, 2000, and 3000 km are shown.
not treated in the theory. Yet, the observed equilibrium range

and the exponential probability distribution function for in-
tensity are consistent with the Dozier and Tappert theory.
It is demonstrated that the large modal bias, spread, and
energy redistribution for the higher modes does not result in
the loss of coherent wavefronts, implying that while the high
modes may be strongly affected by internal waves, they are
coherent enough that when they are synthesized together lo-
calized wavefronts result.
We find that modal travel-time fluctuations are on the
order of 100 times as large as the fluctuations observed on
the corresponding raylike wavefront fluctuations. Other stud-
ies that have used adiabatic mode theory have found much
smaller enhancement factors ~on the order of 3!.15,16 This
difference is probably due to the use of the adiabatic approxi-
mation which does not hold at long ranges.
We find that modal travel-time biases are much larger
than corresponding ray biases and that they have opposite
signs. We defer a detailed discussion of these phenomena
and the reduction of timing fluctuations to a paper in prepa-
ration, which analyses in detail the wavefront fluctuations
from this same set of simulations.
We have presented a novel signal processing method for
eliminating much of the internal-wave-induced spread of the
mode arrivals, which utilizes the geometric mean of indepen-
dent realizations of A n (x, v ). While this helps to narrow the
modal travel-time patterns the strong modal coupling raises
serious issues regarding the interpretation of modal travel
FIG. 13. The quantities ^ A n (x,w) & g ~solid! and Ā n (x,w) ~dash! are plotted
for odd mode numbers 1 through 9 and ranges of 1000, 2000, and 3000 km. times in tomographic inversions at multimegameter ranges.
Phases are in radians. We suggest that the lowest few modes may be useful in
12
modal acoustic tomography at multimegameter ranges, but Y. Desaubies, C.-S. Chiu, and J. H. Miller, ‘‘Acoustic mode propagation
that the use of higher modes will require more theoretical in a range dependent ocean,’’ J. Acoust. Soc. Am. 80, 1148–1160 ~1986!.
13
L. B. Dozier and F. D. Tappert, ‘‘Statistics of normal mode amplitudes in
work to establish an understanding of how the energy reach- a random ocean. I. Theory,’’ J. Acoust. Soc. Am. 63, 353–365 ~1978!.
ing a receiver array as a particular mode has sampled the 14
L. B. Dozier and F. D. Tappert, ‘‘Statistics of normal mode amplitudes in
ocean on its way from the source. a random ocean. II. Computations,’’ J. Acoust. Soc. Am. 64, 533–547
~1978!.
15
J. F. Lynch, G. Jin, R. Pawlowicz, D. Ray, C. S. Chiu, J. H. Miller, R. H.
ACKNOWLEDGMENTS
Bourke, A. R. Parsons, A. J. Plueddemann, and A. R. Parsons, ‘‘Acoustic
The authors are grateful to Bruce Cornuelle, Matt Dz- travel time perturbations due to shallow water internal waves and internal
tides in the Barents Sea Polar Front: Theory and Experiment,’’ J. Acoust.
ieciuch, Bill Kuperman, Kevin Heaney, Walter Munk, and Soc. Am. 99, 803–821 ~1996!.
Peter Worcester for many useful discussions. This work was 16
P. Traykovski, ‘‘Travel time perturbations due to internal waves: Equiva-
supported by the Strategic Environmental Research and De- lence of modal and ray solutions,’’ J. Acoust. Soc. Am. 99, 822–830
velopment Program through Advanced Research Projects ~1996!.
17
W. Munk, ‘‘Sound channel in an exponentially stratified ocean, with ap-
Agency ~ARPA! Grant No. MDA 972-93-1-0003. We are plication to SOFAR,’’ J. Acoust. Soc. Am. 55, 220–226 ~1974!.
also grateful for a grant from the W. M. Keck Foundation. 18
T. F. Duda, S. M. Flatté, J. A. Colosi, B. D. Cornuelle, J. A. Hildebrand,
This work was partially supported by the Office of Naval W. S. Hodgekiss, P. F. Worcester, B. M. Howe, J. A. Mercer, and R. C.
Research through the Ocean Acoustics Program. We have Spindel, ‘‘Measured wave-front fluctuations in 1000-km pulse propaga-
tion in the Pacific Ocean,’’ J. Acoust. Soc. Am. 92~2!, 939–955 ~1992!.
extensively used a Maspar MP2 parallel processing computer 19
J. A. Colosi, S. M. Flatté, and C. Bracher, ‘‘Internal-wave effects on
for our simulations. 1000-km oceanic acoustic pulse propagation: Simulation and comparison
to experiment,’’ J. Acoust. Soc. Am. 96, 452–468 ~1994!.
20
1
W. Munk and C. Wunsch, ‘‘Ocean acoustic tomography: Rays and S. M. Flatté and F. D. Tappert, ‘‘Calculation of the effects of internal
modes,’’ Rev. Geophys. Space Phys. 21, 777–793 ~1983!. waves on oceanic sound transmission,’’ J. Acoust. Soc. Am. 58, 1151–
2
W. Munk, P. Worcester, and C. Wunsch, Ocean Acoustic Tomography 1159 ~1975!.
21
~Cambridge U. P., Cambridge, MA, 1995!. J. Colosi et al., ‘‘A test of acoustic thermometry at ocean basin scales,’’
3
J. J. Romm, Ph.D. Thesis, Massachusetts Institute of Technology, 1987. EOS Suppl., Trans. IGU 76~3!, 100 ~16 Jan. 1996!.
22
4
P. J. Sutton, W. M. L. Morawitz, B. D. Cornuelle, G. Masters, and P. F. M. G. Brown, and J. Viechnicki, ‘‘On the feasibility of modal group
Worcester, ‘‘The incorporation of acoustic normal mode data into tomo- delay-based tomography in the deep ocean,’’ J. Acoust. Soc. Am. 97,
graphic inversions in the Greenland Sea,’’ J. Geophys. Res. 99, 12 487– 3235 ~1995!.
23
12 502 ~1994!. S. Flatté, S. Reynolds, R. Dashen, B. Buehler, and P. Maciejewski,
5
P. J. Sutton, P. F. Worcester, G. Masters, B. D. Cornuelle, and J. F. Lynch, ‘‘AFAR measurements of intensity and intensity moments,’’ J. Acoust.
‘‘Ocean mixed layers and acoustic pulse propagation in the Greenland Soc. Am. 82, 973–980 ~1987!.
24
Sea,’’ J. Acoust. Soc. Am. 94, 1517–1526 ~1993!. S. Flatté, R. Dashen, W. Munk, K. Watson, and F. Zachariasen, Sound
6
P. J. Sutton, W. M. L. Morawitz, P. F. Worcester, B. D. Cornuelle, and G. Transmission Through a Fluctuating Ocean ~Cambridge U. P., Cam-
Masters, ‘‘Temperature evolution of the upper ocean in the Greenland bridge, MA, 1979!.
25
Sea, January to March 1989,’’ J. Geophys. Res., accepted for publication S. M. Flatté, C. Bracher, and G. Wang, ‘‘Probability-density functions of
~1996!. irradiance for waves in atmospheric turbulence calculated by numerical
7
E. Shang, ‘‘Ocean acoustic tomography based on adiabatic mode theory,’’ simulation,’’ J. Opt. Soc. Am. A 11, 2080–2092 ~1994!.
26
J. Acoust. Soc. Am. 85, 1531–1537 ~1989!. S. M. Flatté and R. B. Stoughton, ‘‘Predictions of internal-wave effects on
8 ocean acoustic coherence, travel-time variance, and intensity moments for
R. M. Jones, E. Shang, and T. M. Georges, ‘‘Nonperturbative modal to-
mography inversion. Part I. Theory,’’ J. Acoust. Soc. Am. 94, 2296–2302 very long-range propagation,’’ J. Acoust. Soc. Am. 84, 1414–1424
~1993!. ~1988!.
9 27
S. E. C., Y. Wang, R. H. Jones, and T. Georges, ‘‘Nonperturbative modal R. B. Stoughton, S. M. Flatté, and B. Howe, ‘‘Acoustic measurements of
tomography inversion. Part II. Numerical simulation,’’ J. Acoust. Soc. internal-wave rms displacement and rms horizontal current off Bermuda
Am. 98, 560–569 ~1995!. in late 1983,’’ J. Geophys. Res. 91, 7721–7732 ~1986!.
10 28
A. B. Baggeroer, W. A. Kuperman, and P. N. Mikhalevski, ‘‘An overview J. Simmen, G. Wang, and S. Flatté, ‘‘Wavefront folding, chaos, and dif-
of matched field methods in ocean acoustics,’’ IEEE J. Ocean Eng. 18, fraction for sound propagation through ocean internal waves,’’ J. Acoust.
401–424 ~1993!. Soc. Am. ~submitted!.
11 29
Y. Desaubies, ‘‘A uniformly valid solution for acoustic normal mode J. L. Codona, D. B. Creamer, S. M. Flatté, R. Frehlich, and F. Henyey,
propagation in a range varying ocean,’’ J. Acoust. Soc. Am. 76, 624–626 ‘‘Average arrival time of wave pulses through continuous random me-
~1984!. dia,’’ Phys. Rev. Lett. 55, 9–12 ~1985!.

ref5+_Mode coupling by internal waves for multimegameter acoustic propagation in the ocean (Colosi, John A. Flatté, Stanley M)

Uploaded by

Document Informationclick to expand document information

Document Informationclick to expand document information

Copyright:

Available Formats

ref5+_Mode coupling by internal waves for multimegameter acoustic propagation in the ocean (Colosi, John A. Flatté, Stanley M)

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

ref5+_Mode coupling by internal waves for multimegameter acoustic propagation in the ocean (Colosi, John A. Flatté, Stanley M)

Uploaded by

Copyright:

Available Formats

Mode coupling by internal waves for multimegameter acoustic

propagation in the ocean

~Received 20 November 1995; accepted for publication 14 February 1996!

INTRODUCTION sions using modal travel-time data.2–9 Of particular impor-

j51 k5k 0 i51

single-frequency results which display the evolution of ~4!

B. Acoustic propagation methods

C. SPE modal arrivals

where the deterministic vertical mode functions, B n (z, v ),

and the B n (z, v ) obey the orthonormality relation

The stochastic horizontal functions, A n (x, v ), obey the equa-

E zb A n (x, v ) is obtained by setting the off-diagonal elements of

We decompose the simulation fields, F~r,v!, using Eq.

We plot these quantities in Fig. 8. The mean log-

transmission finale, indicating that ray and mode variability

300-Hz case, and they predict the reduction factor to de-

5 ^ u A n ~ x, v ! u & g e i ^ f n ~ x, v ! & , ~37!

C̄n ~ r,t ! g 5 UE P ~ v ! ^ A n ~ x, v ! & g

While the resulting mode arrival pattern is not perfect, we

not treated in the theory. Yet, the observed equilibrium range

You might also like