Three-dimensional observations of tidal plume fronts in estuaries using a synthetic aperture sonar array

From 2017 to 2021 the Undersea Remote Sensing Directed Research Initiative (USRS DRI), sponsored by the Office of Naval Research (ONR), funded the deployment of an autonomous underwater vehicle (AUV) mounted synthetic aperture sonar (sas) system in three different estuaries [Connecticut River (CTR) James River, and Mobile Bay]. The purpose of the deployments was to ascertain how the tidal features and fresh and salt-water interactions in these environments affect sonar imaging—and to determine if sas systems could sense hydrodynamic features in a way that may provide oceanographic insight. Thus, sas derived inferences associated with the spatial scales of plume features and propagation speeds are compared to estimates determined from other in situ and remote sensing techniques. In this context, favorable comparisons to widely adopted measurement techniques reveal the potential of sas systems as an additional tool in physical oceanographic studies.

The usage of a synthetic aperture system for making observations of mid-to-upper water column hydrodynamic features diverges significantly from most previous sas applications. Synthetic aperture sonar imaging exploits the motion of a transmitter and hydrophone array along a trajectory to create, via coherent post-processing, an artificially long “synthetic” aperture.^1,2 This synthetic aperture can be used to create imagery with resolution that is, allowing for certain caveats, both range and frequency independent. This functionality, however, is closely linked to stationarity and medium homogeneity assumptions that are violated by the phenomena linked to estuarine hydrodynamics that this study wishes to observe. Furthermore, the signal-to-noise ratio (SNR) limiting effects of multipath in shallow water environments have had a major impact on the design of sas arrays,^3–5 and most systems, including the one used in the present study, are specifically designed to attenuate the type of acoustic backscattering the present study wishes to observe, namely, backscattering originating from features located in the mid-to-upper water column regions.

If these challenges can be overcome, however, the sas system deployed during the USRS DRI has much to offer: it has a planar array that can be used to estimate azimuthal and vertical angle-of-arrival (AOA) with sufficient accuracy to create meaningful three-dimensional (3D) “point clouds” (xyz scatter plots) from each ping. The broad field of view allows multiple observations to be made of the same locations, potentially allowing the evolution of hydrodynamic features to be observed. In subsequent sections of this paper, the method used for generating point clouds from the deployed sas system will be described, along with a method by which the velocities of point clouds corresponding to hydrodynamic features were determined. The results of applying the developed methods to observe the features and velocities of two different tidal plume fronts encountered during a survey operation at the mouth of the CTR will be discussed and compared to concurrent measurements made via a land-based radar.

Many synthetic aperture systems use a pair of horizontal real-aperture arrays with a vertical offset to make vertical (elevation) angle-of-arrival (AOA) estimates for incoming signals. Examples include the HiSAS 1032^6,7 and Kraken MinSAS.⁸ The vertical baseline between arrays creates a time-difference of arrival for backscattering echoes that can be used to create an interferogram containing bathymetric information. One of the most common difficulties faced by interferometric processing is the resolving of angle ambiguities caused by phase wraps. Much literature has been devoted to the topic of interferogram phase unwrapping, with most papers being published in the related field of synthetic aperture radar,^9–11 though literature specifically directed toward unwrapping interferograms created by synthetic aperture sonar systems also exists.^12–16 Sonar systems often have a wider fractional bandwidth than radar systems, and several resources discussing various methods for exploiting bandwidth to resolve phase wrap ambiguities are available.^17–20 

Unlike the more common two-stave interferometric systems, the sas system used in the present study has a planar array with more than two elements in the vertical dimension. Nevertheless, the spacing of the elements in the vertical dimension is still multiple wavelengths, and phase ambiguities must be dealt with for AOA estimation. Furthermore, the fractional bandwidth of the system (i.e., the signal bandwidth divided by the center frequency) is just at the edge of being sufficiently broad to leverage bandwidth for ambiguity resolution using the techniques described in Saebo¹⁸ or Synnes et al.:²⁰ ambiguity resolution can be quite good when the SNR is very high but tends to wain when the SNR drops. SNR can be increased at the cost of range-resolution using the cross correlation strategy described in Saebo¹⁷ or Cook.¹⁹ However, this strategy is ordinarily reserved for stave pairs, whereas the current system employs a vertical array and data is available from a larger number of staves. To leverage the SNR-enhancing properties of correlation but retain compatibility with the vertical array of the sas system, a cross correlation beamforming framework²¹ was adopted for AOA estimation.

Consider a hypothetical sonar system analogous to the deployed sonar, with a center frequency f₀ of 150 kHz, a bandwidth f_bw of 45 kHz, and a planar array with five elements in the vertical direction. The vertical width w_rcv of the elements is 3λ₀, where λ₀ is the wavelength (0.01 m) at the center frequency f₀ given a sound-speed of c = 1500 m/s. The vertical spacing of the elements is 1.5w_rcv. Furthermore, assume the system has a rectangular transmitter w_trn = 4.5 λ₀ tall, and that the transmitter and receiver array are both oriented with a depression angle $θ dep$ = –5 degrees relative to horizontal. To illustrate the correlation beamforming process, a data collection from this system was simulated for a sonar six meters above a distribution of random scatterers located along an interface, as shown in Fig. 1.

In Fig. 3(a), a clear signal trace is visible both at close ranges and steep grazing angles, and out to the maximum range despite the low SNR. Determining the peak AOA value and comparing the resulting bathymetry estimates with those of the narrow-band technique, it is clear that the noise is reduced and the phase-wrap ambiguities have been resolved.

Note that the division of data into bins results in a reduction of range resolution. With a bin length of 16 samples (24 cm) and overlap of 50%, there is an eightfold reduction in the number of data-points in the correlation beamformer results shown in Fig. 4. The benefits of the beamformer, e.g., phase unwrapping and improvements in accuracy, outweigh the consequences of resolution reduction in the current study, in which hydrodynamic features having scales on the order of meters to tens of meters are being observed.

Several additional notes need to be made regarding the usage of this technique with the deployed sas system: prior to performing the previously described AOA estimation procedure, the real-aperture array data from each horizontal row of elements were first beamformed into polar coordinates (range and azimuth). Following azimuthal beamforming, correlation beamforming is performed and peak values are extracted to create a point cloud in spherical coordinates: range, azimuth, and elevation. Finally, to geo-rectify the point cloud, the spherical coordinate representation is transformed into Cartesian coordinates and vehicle-to-earth coordinate transformations supplied by the gyroscope of the AUV’s inertial navigation system (INS) are applied to the point cloud data, alongside translations based on INS estimates for the AUV location. The rapid oscillations frequently experienced by the AUV, particularly roll, were found to cause serious point cloud misalignments if the earth-coordinate transformations were applied uniformly to each point within a ping. This was found to be due to the failure to account for the rapidly changing orientation of the array during the reception of ping echoes. This problem was resolved by sub-sampling the INS orientation state estimates to achieve unique corrections for the point cloud data corresponding to each range bin of the point cloud data. Figure 5 is an example of a point cloud generated using the deployed system, showing sediment backscattering (magenta, near −10 m), a fish cloud (yellow), and bubbles near the surface (green).

Note that in Fig. 5, backscattering intensity information has been retained in the point cloud using luminosity. In this figure and subsequent point clouds shown in this paper, luminosity is determined by the magnitude of the correlation beamforming result after undergoing dynamic range compression via the square-root operator. This method of visualization was found to improve the visibility of features with high backscattering intensity in point cloud visualizations.

This section summarizes inferred plume properties based on sas-derived measurements and provides quantitative comparisons to similar measurements derived from other in situ and remote sensing techniques to demonstrate the effectiveness of sas systems in this context. To do this we provide relevant information about the CTR plume and supporting sampling methods including optical drone, X-band radar, echosounder, and current velocity and water property profiles. Combined, these are used to make direct comparisons of spatial-scales corresponding to plume features and estimates of propagation speeds both for the expanding plume front and hydrodynamic instabilities propagating along the front.

In 2017, the research teams associated with the USRS DRI performed a set of surveys and measurements of the ebb plume fronts that occur at the mouth of the CTR. The tidally pulsed freshwater outflow from the CTR initially discharges into a crossflow corresponding to the ebbing waters of Long Island Sound, meeting at a strongly convergent plume front a few meters thick.^26–28 The location and water properties in the vicinity of the front vary temporally and with river discharge conditions, but the ebb plume front consistently provides conditions that result in elevated acoustic scattering from physical oceanographic features. First, the convergence of the river discharge with coastal water causes waves to steepen and break near the front, as described by Baschek et al.²⁹ Measurements from the CTR and elsewhere show that this bubble entrainment occurs in the immediate vicinity of the front.^29–31 Directly corresponding to the horizontal velocity convergence at the CTR ebb plume front are strong downwelling currents that have been measured to reach 0.20 to 0.25 m/s.^28,32 This downward vertical flow can draw bubbles deeper into the water column, whereby advection driven by ambient currents can create diffuse bubble clouds extending tens of meters or more behind the front. Frontal features and processes that can entrain bubbles are associated with strong salinity gradients that can also be notable sources of backscattering at near-normal incidence.³³ Bubbles, however, dominate the scattering in the near-field of the frontal features both in the CTR and in other estuarine environments.^29,31

During the experiment, radar coverage of the CTR mouth was constantly maintained. The radar was stationed atop a telescoping tower located 1 km inside the river mouth. The CTR plume front was visible in radar images as it formed at the mouth and as it expanded westward. Figure 6 shows a map of the estuary as visible in Google Earth, and a corresponding example of a radar image. A tidal front is visible within the radar image. The 360-degree radar footprint updated at 0.8 Hz (48 RPM), and 80-s averages (i.e., 64-snapshot averages) were used to create de-speckled image frames. By measuring the temporal changes in the manually extracted location of the tidal front, the frontal velocities could be ascertained from the radar data.³⁴ Uncertainty in these velocities varies with manual ridge-finding accuracy and with the smearing induced by time averaging. A conservative estimate of front position uncertainty is 13.6 m, half the 3 m antenna beam width at the measurement location. Smearing uncertainty must be calculated a posteriori, since it varies with front speed. Combined, the uncertainty can be estimated as $2 σ x front/Δt$⁠, where $σ x front=13.6+80 V plume ( RADAR )/2$ and $Δt=900 s$ between front sample times.

Representative examples of the physical oceanographic conditions at the CTR ebb plume front, in addition to upward-looking echosounder measurements from a REMUS 100 vehicle³⁵ are shown in Fig. 7. This example, which is characteristic of the CTR front, shows that a large plume of bubbles is injected into the water column near the plume front and that these bubbles remain in the upper water column for 10 to 100+ meters behind the plume front.

Relevant to AOA estimation, the sound velocity profile shown in Fig. 7 is not as extreme as might be expected given the strong salinity gradient. This is due to the thermal profile offsetting some of the effects of salinity: the freshwater layer at the time of measurement was warmer than the salt water in the Long Island Sound. The sound velocity that is present is upward refracting within the plume, which will bias the height estimation of scatterers within the plume to positions below their true location. This is because the current method used for AOA estimation assumes a homogenous medium and linear propagation. In an upward refracting medium the vertical angle of the ray between the sonar and scatterer will be at a minimum at the face of the sonar: the linear ray extending from the face of the sonar at the observed AOA lies beneath the actual ray, which is upward refracted during propagation inside the plume. The dense bubble cloud present at the interface between the plume and saltwater was generally observed to occlude scatterers deep within the plume, however, and the effect of this gradient on the geometry of the plume does not appear to strongly affect the reconstruction of observed plume features, which are mostly scattered from bubbles entrained near the plume boundaries.

Figure 8 shows a 3D point cloud generated from a single ping in which the trajectory of the sas system was oblique to the front and on course to intersect it. As in Fig. 5, the color scheme used to visualize the point cloud data encodes height information via hue and magnitude information using luminance. The color-bar on the right side of Fig. 8 shows the hue and luminosity values that may be used to describe height and magnitude. A ground-plane projected photograph of the front captured by an aerial drone during the same experiment but from a different tidal cycle is included to show the surface manifestation of the lobe-and-cleft features observed by the sonar.

Several characteristics of the under-water features of the tidal plume front are visible in the point cloud of Fig. 8. For example, the plume exhibits a prominent lobed structure: two lobes, both ∼11 m long, are shown in the figure, while other similarly sized or smaller lobes are also present. Optical imagery captured from a drone that was deployed during the study [e.g., Fig. 8(B)] also shows the presence of lobes in wave breaking and bubble patterns on the water surface at the front. The presence of these structures was primarily attributed to instabilities arising from cross-front shear in the along-front velocity.³⁶ The along-front length scales revealed in the sas derived point cloud are on the low end of those reported by Simpson et al.³⁶ (roughly 20 m), but not inconsistent with the spatial and temporal variability observed using the drone.

The lobes appear to have bubble scattering signatures trailing behind them and these trailing signatures are not oriented perpendicular to the front, but rather trail at an angle in a south-easterly direction. This is consistent with the velocity vectors of the Long Island Sound current, which drives the south-easterly advection of the bubbles in the mixing layer at the base of the plume. The depth of the bubble plume as a function of range along the direction of the central beam is visible in Fig. 8(C). The distribution qualitatively appears to be linear near the leading edge indicating that a spatial descent rate may be calculated using a fit with a linear slope model. A plume front slope estimation was made by projecting bubbles along a portion of the front onto a vertical plane oriented perpendicular to the front, and culling points greater than 1 meter above the estimated water surface (note that points above z = 0 exist due to system biases, which will be discussed later), and less than 4 meters below the interface. The first 5 meters of the front were used for slope estimation. Slope estimation was performed using total-least squares via a singular value decomposition. The projected points used for the fit and the resulting best-fit line are shown in Fig. 9. The estimated slope is –0.62 meters in altitude for every meter behind the front, approximately −32 degrees, which is a value consistent with plume front structure observed using echo sounder measurements such as the one shown in Fig. 7(A), as well as others captured during the experiment.

Figures 8 and 9 were generated using data from a single ping; data from multiple pings were also combined to achieve a larger scale picture of frontal features. Because the front is travelling, successful multi-ping mosaicking requires estimation of the relative shifts between point clouds prior to combination. Fortuitously, estimating the relative translations between pings can also be used to estimate the speed of travel of the features along the tidal front. The alignment process begins with combining point clouds of pings containing tidal plume front features in non-overlapping sets; in the present case, sets of five pings were combined to form dense, independent point cloud observations of the front. Following grouped point cloud creation, two-dimensional (2D) histograms of the x and y distributions of the points were created to make images of the front. To reject stationary sediment contributions and accentuate the frontal features only, the points having height values within the vertical span –4 m < z < 2 m relative to the air/water interface were used to create the histograms. The positive values in the span, corresponding to heights above the air/water interface, have been included due to the uncertainty in AOA measurements. Ambient acoustic noise, thermal noise in the sonar hardware, ray bending caused by sound-velocity profile inhomogeneities, vehicle depth, or residual roll-correction errors can erroneously position points above the interface. This effect is visible, for example, in Fig. 9, in which the leading edge of the front appears to be about half a meter above the interface. Culling points above the interface would result in the loss of the leading edge of the front, as indicated by the presence of coherent, high intensity scatterers, so the rejection threshold has been extended above the front by two meters. Another source of points that can appear near the air/water interface is multipath,^3,4 particularly the bottom-surface path described in Belettini and Wang³ which, due to the reflection off of the surface and the Lloyd Mirror effect,³⁷ will cause bottom scatterers to appear above the air/water interface. Similarly, the reciprocal surface-bottom path described in Belettini and Wang appears below the sediment interface. As shown in Belettini and Wang, the bottom-surface path defines the lower-bound (in the height dimension) of all multipath contributions for which the air/water interface is the last interface interaction. Likewise, the surface-bottom multipath, which maps below the sediment, is the upper bound of all multipath arrival angles for multipath rays which have the sediment as the last interface reflection. The principle can therefore be inferred that, in an isovelocity medium, multipath signals never map between bounding water column interfaces. This principle is useful for ruling out observed phenomena within the water column as multipath scattering.

In Fig. 10(A), the dominant features in the histogram image are the prominent lobe and cleft structures at the leading edge of the front and the current-driven bubble plume that persists downstream of the front. Notably, as discussed later in greater detail, regions of relatively weak intensity trail behind the cleft features. Several band-like regions corresponding to dropouts are visible in the histogram. These dropout regions correspond to locations in which bright crests of sand-ripples were the dominant scatterers rather than the plume front. As described previously, the values in these dropout bands were “inpainted” using the median of the surrounding pixels to prevent the large discontinuities they create from biasing correlations between image pairs.

Finally, a new fully aligned point cloud is generated from the entire set of pings visualizing the front by interpolating the offset locations $x τ 1 … N , y τ 1 … N$ to the time-stamps corresponding to the individual pings. In the present implementation, linear interpolation is used during this step. The point clouds from all contributing pings, each of which is now compensated for the motion of the plume features, are then combined. A combined set of visualizations in a stationary, geo-rectified frame should undergo defocusing due to “motion blur” induced by the non-stationarity of the front. The motion-compensated version, an image created in the travelling coordinate system of the local plume observations, should have improved focus and enable better resolution of plume features. As shown in Fig. 11, this is in fact what is observed.

A 3D visualization of the set of points along the plume was created by generating a 3D intensity-weighted histogram and rendering using alpha-blending, a technique that allows for occlusion by assigning each voxel a transparency value.³⁸ Voxel transparency was linearly mapped to the intensity-weighted histogram value of the voxel. Figures 13 and 14 show the spatial scales of the motion-compensated plume features. The scalloping of the injected bubble-plume travelling behind and underneath the front is especially visible in Fig. 14, which shows the bubble-plume from underneath, after having removed the sediment contributions to prevent occlusion. A notable feature is that the depths of the bubbles extending from the cleft structures appear to be shallower, as evidenced by the green (i.e., higher altitude) veins trailing behind the clefts. The mechanisms for this observation are unclear but it suggests that areas extending from the clefts may be areas of localized upwelling, or at least weaker than downwelling right at the front. The lobe and cleft structures observed at CTR vary from another type of ebb plume instability referred to as “lobe-cleft instability.” Previous observations made by Horner-Devine and Chickadel using infrared images from the Merrimack River ebb plume showed unique thermal signatures extending from similar-looking cleft like features.³⁹ These features are related to localized upwelling drawing cooler waters from below. The instabilities driving the lobes and clefts at the CTR are hypothesized to be associated with the cross-front shear, which varies from those in Horner-Devine and Chickadel⁴⁰ where there is no significant shear across the front. Simpson et al.³⁶ hypothesize a different mechanism for the generation of the features in CTR. Further insight into the source of these lobe and cleft structures is beyond the scope of this manuscript.

The same plume motion estimation and alignment procedure was applied to an observation of the sonar front on the 28th of June 2017, two days after the previous example. In this scan, the AUV was travelling along to the front at a standoff distance of ∼50 meters. Figure 15 contains volume renderings of data from this scan. The velocity magnitude of the features in Fig. 15 is much higher than in the previous encounter (0.4 ± 0.08 m/s); however, the observed front normal velocity was much smaller: ∼0.06 ± 0.01 m/s, and once again was similar to the radar-estimated front-normal velocity of 0.08 ± (0.04) m/s.

The vertical scalloping effect readily visible behind the clefts of The front in Figs. 13 and 14 appears mostly absent in Fig. 15. Not only are the depths of the bubbles much more homogenous, but the slope of the plume as it descends appears more gradual and is followed by a region in which bubbles ascend back to the surface. This is consistent with bubble kinematics, in that bubbles are brought to depth by downwelling at the front, but the downwelling signal only extends a few meters behind the front. Beyond this range, bubbles are advected by local currents and will gradually rise to the surface driven by their own buoyancy or will dissolve. As shown in Fig. 7, similar features are observed using echosounders. Another feature of importance in Fig. 15(C) is the strong amount of scattering that appears to be originating from significant distances above the surface. Though some of the near-surface contributions to this region may be due to the previously mentioned uncertainties in arrival angle and vehicle roll, these are not significant enough to account for the vertical spread shown in the figure. In fact, these vertical features are more likely due to the higher order multipath contributions described in Belettini and Pinto,⁴ specifically, the multipath rays in which the last interaction is with the surface or, in the present case, with near-surface bubbles.

Though the ray-path bending effects of hydrodynamic features such as internal waves have been observed and studied using sas systems,^41–43  sas systems have not previously been used as a tool to make direct observations of hydrodynamic features in the mid-to-upper water column. Despite being an unorthodox usage of a sas array, the field data collected at the mouth of the CTR during the USRS DRI revealed that sas systems are capable of capturing near-surface estuarine-specific hydrodynamic features such as the tidal-plume fronts shown in this paper. Furthermore, these features were captured without any extra efforts oriented at suppressing sediment features; rather, backscattering from bubbles in the tidal plume front was frequently the dominant backscattering feature for the portions of imagery containing their backscattering contributions.

The challenges associated with coherently processing non-stationary scatterers in a standard synthetic aperture multi-ping coherent post-processing framework were circumvented by operating in an alternative post-processing framework that leveraged point cloud data created using physical array processing applied to real-aperture data collected from individual pings. Dense multi-ping point clouds and 3D histograms were synthesized from combinations of these single-ping point clouds. To leverage the planar array of the deployed sas system, solve angular ambiguities caused by phase wraps, and reveal the 3D structure of the observed environment features, a correlation-beamforming algorithm was exploited to determine the elevation angles corresponding to the dominant scatterer observed at any given range and azimuth. Using this approach, point clouds could be created from every single ping transmitted by the sonar. Matching point cloud features between pings enabled the velocities of features to be determined and improved the focus of multi-ping two- and 3D histogram renderings of the plume front.

A variety of tidal plume front features were observed with the deployed sas system and the post-processing described in this paper: a lobe-and-cleft structure was present in both encounters of the front, and the lobe features had length scales ranging from several meters to just over ten meters. In both observations, the along-front speed of the plume features matched or exceeded the front-normal velocity. Furthermore, the front-normal velocities in both cases nearly matched independent estimates derived from land-based radar observations.

The first Encounter with the front showed an initial descent of –0.62 m meters in height per meter traveled behind the front, and this tapered off when the bubbles reached approximately 3 meters in depth. The measured plume depth is consistent with other sensors and the estimated descent rate is also reasonably consistent with most measurements obtained using other acoustic sensors.^32,35 The lobe-and-cleft structure of the plume front was accompanied by a scalloping in the vertical distribution of the bubble-plume which was visible when viewed from below (Fig. 14). In contrast, the second encounter did not show a steep initial descent, and bubbles seemed to be distributed more homogeneously and more closely to the surface. Furthermore, approximately 15 meters after the leading edge of the front, the bubbles appear to rise in elevation. This is likely driven by buoyant bubbles slowly rising in the water column in the absence of downwelling currents away from the front. Ray bending is an unlikely explanation in this scenario because the observed pattern would require downward refraction, whereas when crossing the front, an upward refracting sound speed profile was observed within the buoyant plume waters.

An argument based on the Lloyd's Mirror effect was put forward that the angles of arrival corresponding to multipath arriving from the surface never map beneath the surface in an iso-velocity environment. Combined with the similar reality that, as shown in Belletini et al.,^3,4 multipath arriving from the sediment never maps above the sediment, the general rule for multipath arrivals in an isovelocity medium is that they never map between the sediment and water interfaces. This principle can be used to reject multipath as a potential source for many of the bright near-surface features visible in the sonar point clouds and 3D histograms shown in this paper, but it also explains the distribution of energy farther above the surface that was particularly noticeable in the second observation of the front and is shown in Fig. 15(C).

The potential benefits to oceanographic studies based on sas are derived from the ability to resolve much larger areas, in 3D, than alternative sonar and in situ sampling techniques. For example, echosounder or acoustic Doppler velocity current profiles, which have been used in estuarine hydrodynamics studies for decades (Famer,⁴⁴ Trump,^30,32 Kilcher,⁴⁵ Lavery³³) can achieve much higher resolutions than physical sampling techniques and can resolve scattering from features bubbles, turbulent microstructure, stratification, and suspended sediment. Because their sampling is limited to the relatively small ensonified beam, temporal and spatial variability must either be assumed or inferred from differences between transects occurring at different times. Additional spatial coverage can be obtained through the use of multibeam or sidescan sonars (Thorpe⁴⁶) but overall coverage is still small when compared to the sas approach described here. The ability to capture 3D features of an environment over a large region in a single snapshot enables feature tracking that is much more difficult with systems that sample more sparsely. Herein, the ability to effectively identify and track features of interest associated with a tidal plume from using a sas system is demonstrated and results generally agree with observations derived from commonly employed sampling techniques. This suggests that new methods for observing features using sas could potentially lead to new insights into the hydrodynamic processes that are observed by the system.

This work was supported as part of the Undersea Remote Sensing Directed Research Initiative, and funded via Office of Naval Research grant numbers: N00014-16-1-2854, N00014-16-1-2948, N00024-17-D-6421, and N00014-19-1-2593. The authors would additionally like to thank Rocky Geyer, David Ralston, Andone Lavery, and Joe Jurisa for providing the supporting physical oceanographic and acoustic scattering measurements presented here.