High-resolution acoustically informed maps of sound speed

The New England shelf break is a dynamic and energetic region, where cold fresh water from the shelf collides with the warmer saltier water of the slope forming a prominent shelf break front with shoaling isopycnals in the offshore direction (Chapman, 2000; Chapman and Lentz, 1994; Fratantoni , 2001; Lozier and Reed, 2005). This economically important region (Gawarkiewicz , 2018) is also the site of major exchange of heat, salinity, and biogeochemical processes and tracers (Zhang and Gawarkiewicz, 2015). Frontal meanders cause temporal changes in the hydrodynamic conditions on scales ranging from hourly to inter-annually (Bisagni , 2009; Chapman, 2000; Fratantoni , 2001; Linder and Gawarkiewicz, 1998; Zhang and Gawarkiewicz, 2015; Zhang , 2011). Movement of the shelf break front can be driven by meanders in the Gulf Stream, the passage of warm core rings, seasonal stratification, and storms (Bisagni , 2009; Gawarkiewicz , 2004; Houghton , 1994; Linder and Gawarkiewicz, 1998; Todd , 2013; Zhang and Gawarkiewicz, 2015; Zhang , 2011).

Understanding this dynamic environment requires high-spatiotemporal-resolution measurements of environmental parameters such as temperature, salinity, density, sound speed, and fluorescence. The spatial distribution of these water properties can be measured, or derived from direct measurements, by a variety of methods including shipboard conductivity, temperature, and density sensor (CTD) casts, tow-yos, moored arrays, passive drifting platforms, and uncrewed vehicles. Over ocean basin scales, water column properties such as sound speed are also inferred from long-range propagation inversions (Mercer , 2009). Many of these methods provide high-resolution data in either depth or position, but typically cannot provide a holistic view of oceanographic properties on two-dimensional cross-sections without considerable interpolation. Shipboard CTD casts, for example, provide high resolution in depth at a single cast location, however, data between casts must be interpolated to give horizontal resolution. CTD casts also suffer from issues of temporal aliasing resulting in interpolation across both temporal and spatial dimensions to approximate horizontal resolution. Tow-yos partially address this issue, however, this method still suffers from temporal aliasing issues due to the slow ship speed when towing a platform. Combining multiple methods, such as CTDs on moored arrays and AUVs, can provide high resolution in depth and horizontal position at certain locations, however, interpolations in regions of sharp spatial gradients, such as frontal transitions, can introduce significant errors into estimates. Active acoustics on surface vessels and other mobile platforms provide the means for making high-resolution measurements of acoustic scattering, in both depth and horizontal position. Acoustic scattering techniques have been used to identify water masses (Ross and Lavery, 2012; Stranne , 2017), and to characterize temperature and salinity gradients associated with microstructure (Lavery , 2010; Lavery , 2013; Lavery , 2003).

Acoustic scattering techniques also hold the potential to map large features such as shelf break fronts more quickly with higher resolution and with less interpolation than traditional CTD-based methods. In this study, scientific echosounders were used to map the water masses at the New England shelf break front. The acoustically-inferred maps of sound speed (AIMSS), informed by a single CTD profile, were compared with sound speed cross-sections inferred from multiple CTD profiles with a traditional interpolation approach. Ground truthing was performed by comparing water column properties from the acoustically determined profile to those measured at the same location by CTD casts that were not used in the acoustically informed maps. Long-range propagation modeling was performed to assess the impact of the different water column property characterization methods. Suggestions for further improvements to the AIMSS method are then discussed.

Acoustic scattering surveys of the New England shelf break front were conducted as part of the 2021 Office of Naval Research funded New England Shelf Break Acoustics Experiments (NESBA). Cross-shelf transects were conducted approaching the NESBA experimental site (DeCourcy , 2022; Lin , 2022) at a survey speed of 7 knots. Acoustic data were collected using shipboard EK80 split-beam scientific echosounders mounted on the R/V Neil Armstrong. The settings used during shelf break transects are outlined in Table 1. Opportunistic CTD casts were made during the transects, with sampling density dictated by the available time for acoustic surveys. CTD deployment locations were informed by real-time visualization of the shelf break front with the EK80 software, observing when scattering interfaces between water masses were present/absent and observing how the interfaces changed in depth. Water-column properties were measured by a CTD rosette equipped with a Seabird SBE 911 Plus CTD. The CTD recorded pressure, temperature, conductivity, oxygen concentration, fluorescence, and turbidity at 0.5 m intervals. Sound speed was computed according to the Chen-Millero equation (Chen and Millero, 1977) and density by TEOS-10 (IOC, 2010).

Multi-CTD interpolated sound speed cross-sections (hereafter referred to as multi-CTD cross-sections), EK80 echograms (Fig. 1), and acoustically informed maps of sound speed (AIMSS) were created by projecting cruise tracks onto cross-shelf transects perpendicular to the isobaths at the NESBA location. For this region, the isobaths run roughly east-west, and therefore cruise tracks were projected into north/south transects. Projecting by averaging pings and interpolating CTD data perpendicular to local isobaths made it easier to visualize the wedge-shaped front with condensed isopycnals rising toward offshore and avoided the impact of “smearing” of the front from transects that were not run perpendicular to the bathymetry. Multi-CTD cross-sections were created by linearly interpolating, in 1 m depth bins and 20 m bins horizontally (projecting onto transects running perpendicular to the local isobaths) between CTD profiles. Acoustic echograms were created following the method described in Loranger (2022), where the volume backscattering coefficient, σ, was averaged in 70 m bins north-south and 1 m depth bins.

The echograms were combined with sound speed measurements made by an individual CTD profile to map the water masses at the shelf break. Of the five echosounders listed in Table 1, the 18 kHz provided the clearest image of the shelf break front in this study. The 18 kHz echogram showed wedge-shaped lines, consistent with scattering from interfaces between water masses at the front (Fig. 1). While the 38 kHz echo sounder was also able to identify the layers, they were less clear than at 18 kHz. Acoustic scatter in the other higher frequency echo sounders was largely dominated by what appears to be planktonic organisms, a result that is consistent with previous research at the New England shelf break (Loranger , 2022). Scattering from turbulence has a higher scattering strength at low frequency compared with other non-resonating targets, such as planktonic organisms or suspended sediment (Lavery , 2007), making the 18 kHz echosounder the preferred instrument for this study. The layers on the echograms were identified as scattering from the condensed frontal isopycnals with relatively strong density gradients, as opposed to scattering from biological targets or sediments, by comparing the calibrated frequency response across all five echosounders with models of scattering from biological targets, sediments, and turbulent microstructure (data not presented here). The broadband backscatter in the interfaces was consistent with scattering from salinity and temperature fluctuations associated with turbulent microstructure. For a comprehensive discussion on the use of broadband acoustic backscatter to identify turbulent microstructures in the presence of other scatterers, please see Bassett (2023). The frontal interfaces were mapped by manually picking points along the frontal lines shown in the echogram (Fig. 1).

Once the frontal interfaces were mapped from the echogram, a two-dimensional AIMSS was created by correlating each interface with a change in sound speed at the location of a single CTD profile. The AIMSS methodology described in this study could have been used to create maps of any of the water-column properties measured by the CTD, or computed from CTD measurements, to generate acoustically-inferred maps of water column properties (AIMWCP). However, special care must be taken when using AIMSS or AIMWCP to determine if a water column property corresponds to a watermass detected acoustically. The CTD profile that intersected all of the interfaces identified in the echogram analysis (see cast 10 in Fig. 1) was used to create the AIMSS. A single CTD cast was used in this analysis so that the other casts could be used as ground-truth to validate this approach. The depth of each interface at the location of the CTD cast was overlaid on the sound speed profile. The sound speed profile from the CTD was broken into sections bounded by the depth of each interface at the CTD location. These sections were then propagated to the north and south following the lines of interfaces identified in the echogram data. When the width of a layer between interfaces changed relative to the width at the CTD profile, the properties were linearly interpolated to either shrink the data or expand it relative to the width of the layer at the CTD profile. For example, the sound speed profile in cast 10 from 0 to 50 m (from the surface to the top of the first scattering interface) was propagated by contracting that range of the CTD profile into a smaller depth range moving to the south as the frontal isopycnals approached the surface, and expanded moving to the north as the frontal isopycnals approached the seafloor. The same process was repeated for the sound speed between all interfaces, as well as the profile from the deepest interface to the seafloor. Because of the relatively shallow depths in this study, we have assumed that pressure is not a significant factor in the sound speed.

Four acoustic propagation simulations were performed using a two-dimensional wide-angle Parabolic Equation (2DWAPE) acoustics model (Lin , 2013). Each simulation contains a permeable acoustic half-space ocean bottom with a sound speed of 1650 m/s, density of $1800 g / cm 3$⁠, and attenuation of $0.5 dB / wavelength$⁠, representing a sandy bottom. Model 1 simulates upslope propagation from a 50-meter-deep 500 Hz acoustic point source located at $39 ° 54 ′ 00 ″$ North, $71 ° 36 ′ 23 ″$ West (P1) to $40 ° 7 ′ 30 ″$ North, $70 ° 36 ′ 23 ″$ West (P2) approximately 25 kilometers away. Model 2 simulates downslope propagation from P2 to P1, retaining the 50-meter source depth. Each model was run using the multi-CTD cross section and the AIMSS separately, and the results of the four simulations are compared to examine the influence of the sound speed profile interpolation error on the modeled sound propagation.

Four scattering interfaces were identified in the echogram of the shelf break front (Fig. 1). Multiple sets of what appear to be internal waves can be seen in each interface, especially at the northernmost extent and between CTD casts 9 and 11. On the shelf end of the echogram, a large aggregation of scatterers can be seen in the following the shallower interfaces, consistent with previous surveys at the shelf break (Loranger , 2022).

The multi-CTD cross section (Fig. 2, upper) highlights the wedge-shaped front, where the warm and more saline waters of the slope meet the colder, fresher waters of the shelf. The linear interpolation between CTD profiles leads to a smoothed cross section showing sloping lines of constant sound speed. The AIMSS (Fig. 2, lower), shows the same overall shape of the front, but with higher resolution, resolving small scale features missed in the multi-CTD cross section. In the AIMSS, the location of the foot of the front, where the front meets the seafloor, is readily observed between casts 9 and 10 for the deeper interfaces, and between casts 7 and 8 in the shallower interfaces. The location of where the front meets the seafloor is less readily apparent in the CTD cross section. The presence of a “bump” at the foot of the front, where the front rises in the water column, is clearly visible in the AIMSS and entirely absent in the multi-CTD cross section. The AIMSS also contains information about the water column's physical properties throughout the profile from the surface to the seafloor, while the multi-CTD cross section is limited to the maximum depth of the shallower profile between two profiles.

The single CTD profile used to generate the AIMSS was cast 10, which intersected all four of the scattering interfaces. Figure 3 shows the sound speed profiles from the CTD measurements (solid lines) and from the AIMSS at the same latitude as the CTD profiles (dashed line). Cast 10 was used for the AIMSS method and, therefore, the CTD profile and the AIMSS are identical at that location. Cast 11 is at the extreme southern end of the study area. The profile from cast 11 compared with the profile at that location from the AIMSS shows good agreement in the upper water column, however, there is evidence of two layers not detected in the acoustic analysis at about 83 and 144 m. It is possible these deeper interfaces would have been detected if the acoustic survey had continued further offshore, as there is evidence of the layers in the echogram (Fig. 1), however, these interfaces were at the outer edge of the survey area and were not included in this analysis. The echogram also shows evidence of another interface at around 30 m depth at the location of cast 9. This interface is coincident with a change in sound speed at that depth in the CTD profile for cast 9 [Fig. 3(c)]. This interface was not initially identified in the echogram due to ringing of the transducer from the surface to about 25 m depth, which masked the interface in the upper region of the water column.

In casts 8 and 9, the deeper warmer slope water layer appears higher in the water column in the AIMSS than in the CTD profile. In cast 9, the warm layer is about 6.0 m shallower in the AIMSS than the CTD profile and in cast 8 the layer is about 9.0 m shallower in the AIMSS. The echogram shows the presence of internal waves at these locations (Fig. 1) with a peak to trough height of about 15 m. During the acoustic surveys, acoustic scattering data were collected continuously, however, as the vessel slowed to begin CTD casts, the ship's fathometer and Doppler velocimeter log were turned on to aid the vessel in the CTD deployment. These instruments interfered with the scientific echosounder data, making interpretation difficult and no data from the time when the additional acoustic navigation instruments were turned on were used in this analysis. The time from when the ship's navigational equipment was turned on to the time of the CTD recovery was about 1 to 2 h. The acoustic data, therefore, represents the position of the internal waves approximately 1 to 2 h prior to the CTD cast. The discrepancy between the depth of the warmer layer in the acoustic profile and the CTD profile could therefore be due to the propagation of the internal wave through the area during the 1 to 2 h difference in data collection. Comparing the shipboard collection to a stationary array or to expendable bathythermograph (XBT) casts where the time between acoustic collection and direct water column property measurements is minimized would reduce this uncertainty.

To highlight the impact of the differences between the two methods, long range acoustic propagation modeling was conducted. Figures 4(a) and 4(b) show the results from long range propagation modeling of a 500 Hz source located at 50 m depth on the slope end of the profile. Figure 4(a) shows the results of modeling using the multi-CTD cross section, and Fig. 4(b) shows the results from the AIMSS. Figures 4(c) and 4(d) show the results from long range propagation modeling of a 500 Hz source located at 50 m depth on the shelf end of the profile. A visual comparison of the upslope CTD based and AIMSS model based runs [Figs. 4(a) and 4(b)] suggest that the multi-CTD cross section provides a more stable ducting effect in the top 75 meters. The propagation model runs approximately between CTD casts 11 and 9, and inspection of the AIMSS performance at cast 9 in Fig. 3(c) highlights the sharp sound speed gradient that is lost in the AIMSS calculation, and which is responsible for the difference in acoustic ducting. As discussed previously, the gradient smoothing is largely due to ringing in the transducer near the surface masking the presence of another interface not captured in this AIMSS analysis. While that interface was not captured in this AIMSS analysis, the AIMSS model does provide more vertical fluctuation of the shelf and slope water transition, which produces more detailed acoustic propagation in shallow water. Likewise, the downslope models [Figs. 4(c) and 4(d)] demonstrate the acoustic significance of shelf front definition. The shallower position of the front in the AIMSS model near casts 9 and 11 refracts acoustic energy into the slower upper shelf waters more effectively than the deeper CTD transition. Given the stark differences in acoustic propagation produced by the choice of sound speed model, further ground truthing to validate the AIMSS model is critical to understanding the benefits and limitations of high resolution mapping of the shelf front.

The benefits to the AIMSS method include limiting the distance over which interpolations need to be performed, decreased ship time spent performing CTD casts, and enhanced resolution of frontal features such as internal waves that can be difficult to capture with shipboard CTD casts. Further time savings using the AIMSS method could be performed when using XBTs or fast CTD profilers, instead of traditional CTDs providing further time savings, as XBTs and fast CTD profilers can be deployed during full-speed ship transit. Entire frontal profiles could be generated in the time it takes the vessel to traverse the region of interest at survey speed.

During the NESBA experiments, the benefits of the AIMSS method became immediately apparent when the R/V Armstrong was forced to leave the experiment site due to inclement weather. As a storm approached, the vessel left the slope area to find safe harbor, and there was only time for a single CTD cast. Water column properties cannot be resolved horizontally with a single cast, but with the AIMSS method, we were able to resolve distinct changes to the front as an eddy traversed the study area (see the supplementary material). This eddy appears to have generated sufficient upwelling to pull intruding slope water, typically trapped in the lower portion of the water column on the shelf, up to near the surface.

While the AIMSS method shows promise for real-time high-resolution water column mapping, further ground truthing is necessary to determine the accuracy and limitations of acoustic based water column property mapping. In this study, only one CTD profile was used to create the AIMSS so that other CTD profiles could be used as a means of ground truthing the method. With other ground truthing methods, it would be possible to develop a more robust model, such as a bootstrapping method that iteratively determines the sound speed profile based on multiple CTD/XBT casts until the AIMSS converges. Concurrent long-range propagation measurements would provide a critical comparison between the long-range modeling based on multi-CTD cross-sections and AIMSS and elucidate the limitations and benefits of these two methods. Limiting the intensive human-in-the-loop frontal interface mapping component of the AIMSS method would also greatly improve this method's applicability. A machine learning based target detection method, trained to automatically detect the frontal interfaces using both image processing and broadband backscatter analysis could rapidly determine the front location and therefore perform AIMSS in near real time. These improvements to the current method, and improved validation through other sources of ground truthing, would greatly improve the applicability of this method as a standard survey technique.

See the supplementary material for a figure showing the use of AIMSS when only one CTD cast was available due to approaching inclement weather during the study period.

The authors would like to thank the captain and crew of the R/V Armstrong. We would also like to thank our fellow scientists and students whom participated in the New England Shelf Break Acoustics Experiments (NESBAEx). Funding was provided by the Office of Naval Research Task Force Ocean.

The authors have no conflict of interest to disclose.

The data that support the findings of this study are available from the corresponding author upon reasonable request.