NOVEMBER 03 2023

Efficient approach to computing travel time with the

parabolic equation model
Kevin D. Heaney; Richard L. Campbell

JASA Express Lett. 3, 116001 (2023)

https://doi.org/10.1121/10.0022267

CrossMark

 
View Export
Online Citation

08 November 2023 03:20:12

 ARTICLE asa.scitation.org/journal/jel

Efficient approach to computing travel time

with the parabolic equation model
Kevin D. Heaney and Richard L. Campbell
Applied Ocean Sciences, 11006 Clara Barton Drive, Fairfax Station, Virginia 22039, USA
[email protected], [email protected]

Abstract: An efficient method for computing the travel time of an acoustic wave using the parabolic equation model is pre-
sented. The frequency derivative of the acoustic phase is the differential travel time associated with a propagation in range. By
taking this difference across closely spaced frequencies this method computes the acoustic travel time. This method requires
the computation of the field at two frequencies rather than over the full band. The method compares well with other travel
time methods for four different cases, including deep water, upslope and shallow water, and a three-dimensional propagation
environment. V C 2023 Author(s). All article content, except where otherwise noted, is licensed under a Creative Commons Attribution (CC BY)

license (http://creativecommons.org/licenses/by/4.0/).
[Editor: Linus Y. S. Chiu] https://doi.org/10.1121/10.0022267
Received: 31 July 2023 Accepted: 20 October 2023 Published Online: 3 November 2023

1. Introduction
A commonly used acoustic propagation model for low-frequency acoustic propagation is the parabolic equation (PE). The
model handles range-dependence in the ocean, seafloor, and sediment type. It was introduced into underwater acoustics

08 November 2023 03:20:12

by Tappert (1977) and the efficient high-angle split-step Pade-dependent acoustic model (RAM) was introduced by Collins
(1993). The PE model, separates the wave equation into incoming and outgoing energy, discards the incoming energy and
solves for the outgoing (one-way propagation) in range using the paraxial approximation by applying a finite-difference
propagator matrix to the field at each range step. As a frequency domain model with a finite-difference implementation
for the propagator matrix it is well suited to low frequency narrowband propagation modeling. It has been extended to
three-dimensional (3D) modelling applying a propagation step in the horizontal (Lin et al., 2012; Heaney and Campbell,
2016). The computational grid is generally defined by a points-per-wavelength in depth (16) and range (0.2–1), which
leads to computational cost which scales as frequency-squared ( f 2). For 3D propagation this is frequency-cubed ( f 3). For
this reason, the PE is considered a favorable model for lower propagation frequencies. Impulse responses and travel times
can be computed by the PE, via a sequence of narrowband runs across the source bandwidth followed by an inverse fast
Fourier transform (IFFT). The frequency spacing for the IFFT is set by the travel time dispersion of the acoustic propaga-
tion such that df ¼ 1/(tmax  tmin), where tmin is the fastest propagation energy (ray or mode) and tmax is the slowest. For
very long-range propagation, where the dispersion can be on the order of 10s of seconds, this can lead to the requirement
of the computation of many frequency components. As an example, to compute the impulse response from the ATOC
Pioneer Seamount source (near Monterey California) to a vertical line array off the coast of Hawaii, with a range of
3515 km, the time spread was 10 s (source bandwidth was 30 Hz). This leads to the requirement to compute at least 300
frequency bins to generate an unambiguous arrival time (Heaney, 1997). Note that there are many global propagation
examples that are significantly greater than 3515 km (Heaney et al., 1991). There are faster models for computing arrival
times, in particular ray trace codes, but these sometimes do not have the physics that is being investigated (3D diffraction,
or rough surface scattering). For complex propagation, if the PE can handle the physics, this method will be able to be
applied. This algorithm has been implemented for this paper using the split-step Pade PE, but it can be similarly applied
to the split-step Fourier method (Tappert, 1977; Lin et al., 2012).
In this paper we present a method for computing the elapsed travel time by taking the finite-difference derivative
of the pressure phase across two narrowly spaced frequencies. This method, which we will refer to as the differential phase
(DP) method, with a computation of only two frequencies, is shown to reproduce the arrival time consistent with the full
broadband computation. In Sec. 2 the mathematical derivation and numerical implementation of the algorithm are
described. In Sec. 3 four examples with comparisons to other travel time computations are shown covering deep water,
shallow water, and 3D propagation environments. Section 4 is the conclusion.

2. Numerical methodology
To derive the differential phase approach, we begin with the acoustic wave (Helmholtz) equation for a uniform density
medium in the region without a source term in Eq. (1),

JASA Express Lett. 3 (11), 116001 (2023) C Author(s) 2023.

V 3, 116001-1
 ARTICLE asa.scitation.org/journal/jel

1 @2u
r2 u þ ¼ 0: (1)
c2 @t 2
Solutions to this equation can be shown to be a sum of traveling waves with the following form:

u  eixtikr ¼ ei/ ; (2)

where x is the frequency and k is the local wave number. Solutions utilizing this approach include normal mode solutions
and the PE (Jensen et al., 2011). The derivative of the phase [Eq. (2)] with respect to frequency yields the travel time as a
function of frequency
@/ @k r
¼ t ðxÞ  r ¼ t ðxÞ  : (3)
@x @x vg
The computational approach is to run the PE for adjacent frequencies, unwrap the phase (taking out 2p loops)
and compute the incremental travel time. The travel time is then associated with the travel time of the dominant arrival.
This is strictly a post-processing step on the PE output. The differential phase df must be small enough to encompass the
dispersion between the slowest and fastest paths, namely, df < 1/(range/cpmin  range/cpmax). For the examples below, the
frequency pairs computed were 20/20.4 Hz, 100/102 Hz, and 400/408 Hz, corresponding to a frequency differential of 0.02.
The results were found to be independent of the phase differential below a multiplier of 0.1  f0. In Sec. 3, four examples
comparing the differential phase approach with other methods for travel time computation are shown.
The DP method can be used with multiple frequency pairs (or a list). The method computes the difference in
circular (complex) phase between two frequencies. This vector is a superposition of all arriving multipaths and will be
dominated (but not equal to) by the path with the highest intensity. For complex multipath regions, or those with high
dispersion, the true arrival time (centroid) will be different than the dominant path arrival time. Use of multiple pairs of
frequencies for the differential phase computation are easily performed but have not been found to add much stability.

3. Propagation examples
The examples below cover propagation in deep water, an upslope environment and a 3D propagation example around an

08 November 2023 03:20:12

island. For each example the DP approach is compared with the explicit broadband PE (BB method), where the FFT is
taken from the full-band frequency-domain computation. The source function for all examples is a flat source spectrum
across the specified bandwidth with zero phase delay (a band limited delta function).

3.1 Deep-water convergence zone (CZ) propagation

The first example is for a shallow source in the deep-water environment of the North Atlantic. The source location is
25 N, 52.5 W, with a propagation range of 500 km due east. For this, and all other examples, the sound-speed field is
taken from the World Ocean Atlas (Locarnini et al., 2023a,b) database, the bathymetry is from GEBCO 2021 (Group,
2021), and the seafloor is uniformly medium silt, corresponding to a grain-size phi of 5. All runs are run using the data of
July 1st in the WOA database. The source center frequency is 100 Hz, with a bandwidth of 10 Hz. The expected time-
spread of the pulse by 500 km is 4 s so the frequency spacing was taken to be 0.25 Hz. There are, therefore, 40 frequencies
computed in the BB method. The source depth is 100 m. The DP method is used computing the travel time based upon
the phase difference between 100 and 102 Hz, a frequency difference factor of 0.02. The results are shown in Fig. 1. The
upper left panel is the propagation loss (PL) indicating the convergence-zone propagation as the sound refocuses in at a
range of approximately 65 km at the source depth. As the sound travels east, particularly beyond 150 km it becomes bot-
tom interacting and is eventually attenuated. The convergence zones spread due to the dispersive sound speed profile and
the increasing bottom interactions. The steep angle bottom bounce, or fathometer return, is evident at 15 km. The upper
and lower right panels are the travel time offset [relative to a constant moveout of 1500 m/s in range (dt ¼ t  r/1500) for
the differential phase method (upper) and full broadband method (lower)]. The figures are masked to only plot times
where the PL was less than 115 dB. The CZ structure is visible and the paths within the CZ have a phase offset of 0 to 1 s
for both computational approaches. The fathometer return at a range of 15 km and shallow receiver depths is shown to
have a differential time of 3–4 s, which is consistent with the time taken to go to the seafloor (at 4500 m) and back. The
bottom bounce energy, which is evident in shadow regions outside of the main convergence zone refraction, is shown in
both models to arrive later. The lower left panel is the travel time offset as a function of range for a receiver at the source
depth of 100 m. The notable place where there are differences are near the surface at ranges of 80 to 100 km. This is
driven by the high dispersion of sound in the surface duct which is not well represented by the single dominant propaga-
tion arrival estimated by the DP method.

3.2 Deep-water SOFAR channel propagation

A low-frequency, deep-water axial example is presented next using the same propagation path. For this case the source
depth is at 1100 m (the sound speed minimum) and the center frequency is 20 Hz, with a 10 Hz bandwidth. The DP fre-
quencies are 20 and 20.4 Hz. The results are presented below in Fig. 2. In the upper left, the PL shows that the sound is

JASA Express Lett. 3 (11), 116001 (2023) 3, 116001-2

 ARTICLE asa.scitation.org/journal/jel

Fig. 1. Convergence zone propagation example, 400 Hz, 200 m source. (Upper left) PL in dB. (Lower left) Travel time offset at 200 m receiver
depth for full broadband (BB) computation (black) and differential phase (DP) method (blue). (Upper right) Travel time offset vs range and
depth for the DP method. (Lower right) Travel time offset relative to 1500 m/s vs range and depth for the BB method. All offsets are relative
to 1500 m/s. The source position is 25 N, 52.5 W with a propagation direction due east.

well contained within the sound channel and loss is low out to 500 km. The travel-time offsets shown on the right show
that the DP method matches the PL method in most regions. The increase in the time offset with range (the mode group
velocity is likely around 1480 m/s) is evident for axially trapped sound. The higher offsets for bottom bounce energy,

08 November 2023 03:20:12

which appears in the shadow regions is clear at near the surface and the seafloor. The DP method does have spatial noise,
which could be removed via smoothing, or taking a lower resolution output figure. The direct comparison of the time off-
set in the lower panel for the receiver depth of 1100 m shows excellent agreement, with a linear increase in the offset with
range. The DP method does appear to have a small bias to a higher offset (slower propagation speed).

3.3 Upslope and shallow water propagation

The next example is for a shallow source propagating up the continental shelf into shallow water off the east coast of the
United States. The source location is 30 N, 79.4 W, with propagation 200 km due west. As in example 1, the source cen-
ter frequency is 100 Hz, with a bandwidth of 10 Hz. The BB computation frequency step is 0.25 Hz. The source depth is

Fig. 2. Deep water axial propagation example, 20 Hz, 1100 m source. (Upper left) PL in dB. (Lower left) Travel time offset at 1100 m receiver
depth for full broadband (BB) computation (black) and differential phase (DP) method (blue). (Upper right) Travel time offset vs range and
depth for the DP method. (Lower right) Travel time offset relative to 1500 m/s vs range and depth for the BB method. All offsets are relative
to 1500 m/s. The source position is 25 N, 52.5 W with a propagation direction due west.

JASA Express Lett. 3 (11), 116001 (2023) 3, 116001-3

 ARTICLE asa.scitation.org/journal/jel

Fig. 3. Upslope and shallow-water propagation example, 100 Hz, 30 m source. (Upper left) PL in dB. (Lower left) Travel time offset at 30 m
receiver depth for full broadband (BB) computation (black) and differential phase (DP) method (blue). (Upper right) Travel time offset vs
range and depth for the DP method. (Lower right) Travel time offset relative to 1500 m/s vs range and depth for the BB method. All offsets
are relative to 1500 m/s. The source position is 30 N, 79.4 W with a propagation direction due west.

30 m. The results are shown in Fig. 3. The upper left is the PL for this computation. This case begins in a bottom-
interacting environment, with a water depth of 900 m and then propagates up the continental slope from 40 to 85 km and
then along the shelf in 80 m of water. As the sound propagates upslope, the vertical angles of propagation steepen (due to

08 November 2023 03:20:12

specular reflections off the sloping seafloor), until 80 km when the seafloor is flat and then only a few modes propagate
until the signal is attenuated. The travel-time offset plots on the right show three features of this propagation. There is a
small increase in offset (slowdown) in the water column as the field propagates upslope (the higher the propagation angle,
the slower the group velocity), a large increase in offset in the sediment (sediment speeds are slower than 1500 m/s for
medium silt) and then a decrease in offset (speed increase) in the shallow water portion beyond a range of 80 km, where
the group velocity is faster than 1500 m/s. The line plot comparison of the offset in the lower left panel, shows the increase
in the offset upslope, then the decrease once in shallow water. The algorithm fails at ranges beyond 100 km, where the
field levels approach numerical noise.

3.4 3D global propagation

One of the motivations for a PE based travel-time estimation algorithm is generation of 3D travel-time tables for acoustic
localization. The Comprehensive Test Ban Treaty Organization (CTBTO, 2023) utilizes these coverage maps and travel-
time tables to localize events from their deployed hydroacoustic stations. For most other cases, either the broadband PE or
raytrace methods are preferable. Propagation of low frequency (< 100 Hz) sound in the ocean is impacted by refraction
and diffraction from coasts, islands and seamounts (Lin et al., 2013; Heaney et al., 1991; Heaney and Murray, 2009;
Heaney et al., 2013; Heaney and Campbell, 2016; Heaney et al., 2017). The DP method provides the opportunity to get a
travel time from a full 3D PE computation. The example presented is from the Juan Fernandez North (H03N1) CTBTO
receiver site to a position behind (south of) the island, in the acoustic shadow. The “notional source” position is 64.6 km
south of the receiver, on the other side of the island. For this case a center frequency of 20 Hz was used. The DP method
(using df ¼ 0.02f0), run within the 3D PE model, generates travel times as a function of position (longitude, latitude) and
depth, in this case from 100 m to 1000 m. To view the times, an offset speed of 1475 m/s is removed and the travel time
residuals are plotted in Fig. 4. The BB method, with a bandwidth of 5 Hz, is compared with the DP method. The region of
focus is the acoustic shadow south of the island. The geodesic, with a range of 54.6 km, from the source to receiver is
shown in light blue, passing over Juan Fernandez Island. The path around, shown in red, is a total distance of 56.9 km.
The travel time residual for the selected point is 1.6 s, computed by the DP computation. The difference in path length is
2300 m, which corresponds to 1.55 s for a predicted travel time offset, using 1475 m/s as a moveout speed. This validates
the PE travel time algorithm for computing the time in the diffractive shadow behind an island.
Note the ring of travel time delay to the north and west, is at a range of 15 km. The figure plotted is the mean
travel time for depths 100 to 1000 m. At shallow receiver depths, there is an acoustic shadow (see Fig. 1). This region is
mostly in the shadow and hence the energy that arrives there is bouncing off the seafloor and should come in nearly 2 s
late. There are regions, particularly in the acoustic shadow where the DP method has a greater travel time (by up to 0.5 s)
than the BB method.

JASA Express Lett. 3 (11), 116001 (2023) 3, 116001-4

 ARTICLE asa.scitation.org/journal/jel

Fig. 4. 20 Hz PE travel-time residual (for 1475 m/s moveout), for H03N1 around the island of Juan Fernandez. The full broadband (BB)
method is on the left, the differential phase (DP) is on the right. A point in the acoustic shadow is selected with a travel time offset of 1.6 s is
chosen. The geodesic path (white fine dash) is 54.6 km and the path around the island (white course dash) is 56.9 km, corresponding to a dif-
ferential distance of 2300 m, or 1.55 s offset.

4 Conclusion
In this paper we have presented a novel approach for computing the acoustic travel time using the PE. As a solution to
the wave equation, it is expected that phase fronts of waves will move at the local group velocity. By taking the difference
in the phase term across narrowly spaced frequencies we can estimate the travel time of the dominant arrival. This method
is applied to four example cases: deep water convergence zone, deep water axial, upslope and shallow water, and 3D prop-
agation. In each of these cases the differential phase time estimate closely matches that computed by other methods. The
approach struggles when the acoustic field is very small, which may be either a numerical problem, or a physics one (there
is no longer a dominant wave). This method has the capabilities of computing acoustic travel-time tables for global propa-

08 November 2023 03:20:12

gation including of 3D propagation affects.

Acknowledgments
This work was supported by the Preparatory Commission for the Comprehensive Test Ban Treaty Organization (Contract
2021-1251) and the Office of Naval Research (Contract No. N000142212584). The authors would like to acknowledge many
productive discussions, including reviews of the manuscript, with Tiago Oliveira, Mario Zampolli, Mark Prior, Georgios
Haralabus, Dirk Metz (CTBTO), and Emanuel Coelho (AOS). Andrew Heaney (AOS) helped on this project with
computational runs and figure generation.

Author Declarations
Conflict of Interest
The authors have no conflict of interest to report.

Ethics Approval
No animal subjects or human participants were used in this study.

Data Availability
The acoustic model is not available to the public. The databases used for the seafloor, ocean, and sediment can be found
in the references.
References
Collins, M. (1993). “A split-step Pade solution for the parabolic equation method,” J. Acoust. Soc. Am. 93, 1736–1742.
CTBTO (2023). https://ctbto.org/ (Last viewed October 20, 2023).
Group, G. B. C. (2021). “The GEBCO_2021 Grid—A continuous terrain model of the global oceans and land,” NERC EDS British
Oceanographic Data Centre NOC.
Heaney, K. D. (1997). Inverting for Source Location and Internal Wave Strength Using Long Range Ocean Acoustic Signals, Ph.D. thesis,
University of California San Diego, San Diego, CA.
Heaney, K. D., and Campbell, R. L. (2016). “Three-dimensional parabolic equation modeling of mesoscale eddy deflection,” J. Acoust. Soc.
Am. 139, 918–926.
Heaney, K. D., Campbell, R. L., and Snellen, M. (2013). “Long range acoustic measurements of an undersea volcano,” J. Acoust. Soc. Am. 134,
3299–3306.

JASA Express Lett. 3 (11), 116001 (2023) 3, 116001-5

 ARTICLE asa.scitation.org/journal/jel

Heaney, K. D., Kuperman, W. A., and McDonald, B. E. (1991). “Perth-Bermuda sound propagation (1960): Adiabatic mode interpretation,”
J. Acoust. Soc. Am. 90, 2586–2594.
Heaney, K. D., and Murray, J. J. (2009). “Three dimensional propagation measurements in the continental shelf,” J. Acoust. Soc. Am. 125,
1394–1402.
Heaney, K. D., Prior, M., and Campbell, R. L. (2017). “Bathymetric diffraction of basin-scale hydroacoustic signals,” J. Acoust. Soc. Am. 141,
878–885.
Jensen, F. B., Kuperman, W. A., Porter, M. B., and Schmidt, H. (2011). Computational Ocean Acoustics (Springer Science & Business Media,
New York).
Lin, Y.-T., Collis, J. M., and Duda, T. F. (2012). “A three-dimensional parabolic equation model of sound propagation using higher-order
operator splitting and Pade approximants,” J. Acoust. Soc. Am. 132, EL364–EL370.
Lin, Y.-T., Duda, T. F., and Newhall, A. E. (2013). “Three-dimensional sound propagation models using the parabolic-equation
approximation and the split-step Fourier method,” J. Comput. Acoust. 21, 1250018.
Locarnini, R. A., Baranova, O., Mishonov, A. V., Boyer, T. P., Reagan, J. R., Dukhovskkoy, D., Siedov, D., Garcia, H. E., Bouchard, C., Cross,
S., Paver, C. R., and Wang, Z. (2023a). Temperature, World Ocean Atlas 2023 Vol. 1 (National Oceanic and Atmospheric Administration,
Washington, DC).
Locarnini, R. A., Baranova, O., Mishonov, A. V., Boyer, T. P., Reagan, J. R., Dukhovskkoy, D., Siedov, D., Garcia, H. E., Bouchard, C., Cross,
S., Paver, C. R., and Wang, Z. (2023b). Salinity, World Ocean Atlas 2023 Vol. 2 (National Oceanic and Atmospheric Administration,
Washington, DC).
Tappert, F. D. (1977). “The parabolic approximation method,” in Wave Propagation and Underwater Acoustics, edited by J. B. Keller and J. S.
Papadakis (Springer, Berlin), Vol. 70, pp. 224–287.

08 November 2023 03:20:12

JASA Express Lett. 3 (11), 116001 (2023) 3, 116001-6