Exploring offshore particle motion soundscapes

Particle motion refers to the back-and-forth vibratory motion of individual fluid particles in a sound field, whose collective motion of compression (particles closer together) and rarefaction (particles further apart) leads to propagation of sound pressure waves. All species of fishes utilize particle motion for hearing, as do many invertebrate taxa, including mollusks and crustaceans (Popper and Hawkins, 2018; Radford , 2022; Samson , 2016). Sound pressure is utilized by marine mammals and some fishes. Fishes and aquatic invertebrates depend on soundscape cues for key ecological functions, including intraspecific communication (e.g., for courtship), navigation, deterring competitors, and detecting predators and prey (Popper , 2001; Putland , 2019). Most acoustic communication among these taxa is thought to occur in the acoustic nearfield (Higgs and Radford, 2016; Jézéquel , 2021), where energy associated with particle motion exceeds that from pressure close to a sound source. Detectability of the overall soundscape (or “auditory scene”) for navigation, settlement, and other functions by these taxa can extend into the acoustic far field (Raick , 2021; Salas , 2022), where energies associated with acoustic pressure and particle motion are equal. Soundscape cues in the near and far field influence the settlement of fish and invertebrate larvae, including corals and oysters (Lillis , 2013; Lillis , 2018; Parmentier , 2015). Most soundscapes are only reported using sound pressure with limited understanding of (1) particle motion soundscapes, (2) how natural soundscapes are linked to fish and invertebrate ecology, and (3) how anthropogenic noise impacts these taxa (Hawkins , 2015; Mooney , 2020; Popper and Hawkins, 2018).

Soundscape components fall under three broad categories: biotic sounds produced by living organisms, abiotic sounds from natural physical phenomena such as weather and geological events, and anthropogenic sounds from human activity such as transportation and construction. In offshore ocean soundscapes, commonly reported biotic sounds contain vocalizations from fishes and marine mammals, including whales, dolphins, and porpoises; wind is a predominant abiotic sound source (Hildebrand , 2021; Kowarski , 2022). Anthropogenic sounds in offshore environments include those from air guns during seismic surveys, construction of offshore energy platforms, drilling, and shipping. Of these, shipping is especially prevalent both temporally and spatially (Duarte , 2021).

All fishes possess inner ears capable of detecting particle motion. Many fishes that possess a gas-filled swim bladder are also capable of sound pressure detection, and those that possess specialized extensions of, and connections between, the swim bladder and inner ear are more sensitive to sound pressure (Popper and Fay, 2011). Conversely, aquatic invertebrates generally lack low-density gas-filled organs that readily respond to sound pressure, and therefore likely only sense particle motion, via statocysts and other sensory organs (Budelmann, 1992a,b). These taxa tend to be most sensitive to frequencies below 1 kHz, including infrasound (Nedwell, 2004; Roberts and Elliott, 2017; Samson , 2016). This frequency range includes many sounds present in offshore soundscapes, including those from shipping and wind, as well as low- and mid- frequency cetacean calls (Kowarski , 2022; Slabbekoorn , 2010). Most fish vocalizations also have predominant energy <1 kHz (Amorim, 2006).

Particle motion can be reported as particle displacement, velocity, acceleration, or further derivatives. Of these, particle acceleration is considered the most relevant sensory stimulus for hearing organs across fish and aquatic invertebrate taxa (Popper and Hawkins, 2018). Importantly, particle motion is a vector with a given magnitude and direction (axis of motion), as opposed to pressure which is a scalar. Among fishes, particle motion vectors can serve as sound localization cues (Zeddies , 2012), and some fishes use both sound pressure and particle motion cues for directional hearing (Veith , 2024).

Several recent studies have described particle motion levels of soundscapes in underwater habitats ranging from coral reefs (Jones , 2022; Kaplan and Mooney, 2016) to temperate, sandy bays and harbors (Horch and Salmon, 1973; Jesus , 2020; Rogers , 2021) as well as freshwater rivers and streams (Lugli and Fine, 2007; Lumsdon , 2018). Some have also reported particle motion of vessel noise (Magnhagen , 2017; Picciulin , 2010; Wahlberg , 2008). A recent study also measured particle motion of the ambient soundscape and boat noise in a nearshore marine reserve to report listening range reduction (masking) by boat noise for multiple fish and invertebrate species (Wilson , 2023). Many of these studies have compared sound pressure level to the particle motion levels of soundscapes, and soundscapes in terms of both pressure and particle motion to known hearing sensitivities of fishes (and sometimes invertebrates). Most existing studies reporting particle motion of underwater soundscapes have been in bays and coastal waters, whereas particle motion in offshore soundscapes has received less attention. Notably, recent studies in open-ocean environments have compared sound pressure [expressed as potential energy (PE)] and particle motion [expressed as kinetic energy (KE)] of empirical and simulated vessel, seismic airgun, explosive, and sonar sources (Dahl , 2024; Dahl and Dall'Osto, 2022; Flamant and Bonnel, 2023). However, to the authors' knowledge, no studies have taken a whole soundscape perspective, including natural sounds, for comparing pressure and particle motion at offshore sites.

To calculate particle motion from sound pressure measurements, it is common practice to use an approximation (hereafter termed “plane wave approximation”), which assumes an idealized scenario of a plane wave propagating in a free-field and lossless environment (Jansen , 2019; Oppeneer , 2023). Given these assumptions, this approximation states that the magnitude of particle velocity is equal to the sound pressure divided by the characteristic-specific acoustic impedance (the density of the medium times the sound speed), and it can be calculated using pressure data from a single hydrophone. It is not considered a valid approximation of particle motion in the near field (within about one wavelength of a source, where particle motion will be greater relative to pressure), or when frequencies are below the waveguide cut off frequency, which depends on the water depth and sediment properties (Jansen , 2019; Nedelec , 2021; Oppeneer , 2023). Further, where propagating waves (including plane waves) reflect off the sea surface and seabed and constructively and destructively interfere, measured particle motion and pressure can depart from this plane wave approximation within about 1/4 of a wavelength from the sea surface and seabed (Oppeneer , 2023). Notably, in an underwater waveguide scenario of fluctuating PE and KE due to interference of modes, sufficient bandwidths (e.g., 1/3 octave) can average out PE and KE differences throughout most of the water column (Flamant and Bonnel, 2023). In above conditions, however, it is recommended that particle motion be measured directly either with an accelerometer or velocimeter (e.g., geophone) or via hydrophone arrays used to estimate particle acceleration; such methods do not rely on a free-field plane wave assumption (Nedelec , 2021). Further, single hydrophone estimates of particle motion do not give any directional information, which is of relevance to fish and invertebrate hearing and to underwater acousticians for estimating the bearing of sound sources.

The present study utilized 1 year of soundscape data from ocean landers deployed on the seafloor at six deep-water sites on the U.S. Mid and South Atlantic Outer Continental Shelf. The compact, volumetric hydrophone arrays at these sites allowed measurement of both sound pressure and particle motion. This study addressed three major questions about deep-water offshore soundscapes in this region: (1) Do particle motion levels and trends of ambient offshore soundscapes across time differ from those of sound pressure? (2) Are particle motion levels predicted from the plane wave approximation valid (within ±3 dB)? (3) Can offshore particle motion and sound pressure soundscape levels be equivalently predicted by non-acoustic indicators, including wind, currents, temperature, and vessel presence? The results inform discussion on the types of sounds and scientific objectives for which particle motion measurement is essential in offshore soundscape studies, and those for which sound pressure and particle motion data are redundant.

Co-located passive acoustic, temperature, salinity, and dissolved oxygen data were collected from Acoustic Long Term Observatory landers (JASCO Applied Sciences, Halifax, Nova Scotia, Canada) deployed on the seafloor at six sites on the U.S. Mid- and South Atlantic Outer Continental Shelf, part of the Atlantic Deepwater Ecosystem Observatory Network (ADEON) (Fig. 1; Table I). These sites ranged from about 300 to 900 m deep and were at distances of 45–280 km offshore. Sites Wilmington (WIL) and Savannah Deep (SAV) were near deep-sea coral sites (Gasbarro , 2022). Hatteras South (HAT) was closest to major shipping lanes and to shore, whereas Blake Escarpment (BLE) was furthest from shore. Each lander was equipped with a compact four-channel array of M36-V35-100 hydrophones (Geospectrum Technologies, Dartmouth, Nova Scotia, Canada; calibrated frequency range: 2–250 000 Hz; nominal sensitivity: –165 dB re 1 V/μPa; spectral noise floor level for each hydrophone: 34 dB re 1 μPa/Hz) connected to an Autonomous Multichannel Acoustic Recorder (AMAR G4; JASCO Applied Sciences; 24-bit, 8 kHz sample rate; 6 dB gain; Fig. 2). The hydrophones were orthogonally positioned with a nominal spacing of 0.5 m, forming two horizontal axes (x, y) and one vertical axis (z). The lower three hydrophones were placed ∼1.25 m above the seafloor, and 0.25 cm above the flat lander surface. Within 24 h before deployment and after retrieval, hydrophone spacing was measured, and hydrophones were calibrated with a GRAS 42AC pistonphone, which played a calibration tone at 250 Hz. Landers at two sites [HAT, Jacksonville (JAX)] had upward facing echosounders, although echosounder data were not utilized in the present study. For further details on deployed equipment, see the ADEON Hardware Specification (Martin , 2018). Each lander had a Sea-Bird SBE 37-SMP-ODO MicroCAT (Sea-Bird Scientific, Bellevue, WA) recorder that measured salinity (PSU), dissolved oxygen (ml L⁻¹), and temperature (°C) at 30 min intervals. Due to sensor malfunction, salinity, oxygen, and temperature data were not available at HAT from July 10, 2018 to November 11, 2018, and bottom temperature was not logged at WIL in October and November 2018. The MicroCAT emitted noise when recording for <90-s periods (at 30 min intervals) at all sites, and echosounders placed on HAT and JAX landers emitted broadband noise pulses when on for 12 min periods (at 1 h intervals). Noise from both instruments was present within the 10–750 Hz range for soundscape analysis, therefore 60 s time samples with this noise were excluded from analyses.

Additional time series of modeled oceanographic data, including current speed and wind speed, were obtained from HYCOM and Copernicus Marine Service online databases, respectively (Copernicus Marine Service, 2023; HYCOM Consortium, 2023). Current speed data had a 3-h resolution, and wind speed data had a 1-h resolution. Time series of both variables were interpolated to a quarter-hourly resolution, then subset to time points within ±5 min of the half-hourly Sea-Bird data timestamps. Time series were spatially subset to a single location per ADEON site, closest to each site (within 0.04° latitude and longitude for HYCOM, and within 0.062° for surface wind data). The norms (square root of the sum of squares) of east and north current and wind speed vectors were calculated. Due to zero-inflation of HYCOM current data at and close to the seafloor, current data from 100 m above the seafloor at each site were used.

Data from two deployments were analyzed spanning November 2017 to November 2018 for each site. Landers for the second deployment were launched within 7 h after landers from the first deployment were retrieved. For simultaneous recording of the four hydrophone channels sampled at 8 kHz, duty cycle varied by deployment. From November 2017 to June 2018, the duty cycle was 5 min on, 16 min off (on for about 14 min per hour). From June 2018 to November 2018, the duty cycle was 19 min on, 1 min off, 19 min on, 1 min off, 5 min on, 15 min off (on for 43 min per hour). The duty cycles were chosen to serve multiple monitoring priorities beyond the focus of the present study. For instance, to detect high-frequency cetaceans, single-channel recordings sampled at 375 kHz took place during parts duty cycle where the four-channel recording was “off.”

Acoustic-level metrics were calculated for both sound pressure (L_p) and measured particle acceleration (L_a) as 60 s means, including spectral density levels (power spectral density for pressure; acceleration spectral density for acceleration) in 1 Hz bins, decidecade band levels, and broadband mean square levels (Table II). To obtain spectral levels, Welch's spectral density was calculated for 1 s time windows with a Hanning window, 50% overlap, and FFT size of 2048. The spectral density values were summed over decidecade bands with center frequencies ranging from 10 to 630 Hz. Broadband mean square levels were calculated after applying a finite impulse response bandpass filter with 10 and 50 Hz window cutoffs for a “low” frequency band, and 100 and 750 Hz window cutoffs for a “high” frequency band; each filter had an order equal to the sample rate (2000). Flow noise was dominant in the low band, and the high band was chosen to avoid frequencies of strong flow signals and include frequencies where wind sounds had more energy. The 750 Hz upper cut-off frequency omitted higher frequencies with ambiguous directionality due to the spacing of hydrophones, i.e., frequencies with wavelengths shorter than 4 days (Nedelec , 2021). Although this wavelength cutoff has been suggested in a recent best practice guide for underwater particle motion measurement (Nedelec , 2021), note that some sources recommend a more conservative limit to lower frequencies, e.g., wavelengths not exceeding 6d (Fahy, 1977), which would limit analysis with spacing in the present study to 500 Hz. Thus, errors in the present study associated with the finite-difference approximation may be relatively greater at frequencies above 500 Hz than below, although such a trend was not observed in sensitivity analyses (described below).

A sensitivity analysis for L_a used 100 randomly selected 60 s audio samples from each site and each deployment (12 sets of 100 samples). Sound pressure data [input into Eq. (1)] was reanalyzed with an applied –4 mm hydrophone spacing measurement error (shorter spacing for all hydrophone pairs) and a –0.5 dB calibration measurement error (applied to only the hydrophone at the origin). These values represent expected error given practical measurement precision. Particle acceleration amplitude error was defined as the ratio of L_a with no error applied (i.e., using actual measured spacings and calibration values) to that with both the –4 mm spacing and –0.5 calibration error applied to measured values, and was visualized as percentiles in 1 Hz bins (see supplementary material). This resulted in median [interquartile range (IQR)] error of less than 3.7 [2.5, 4.0] dB re 1 (μm s⁻²)²/Hz between 10 and 200 Hz and less than 1.1 [0.9, 1.2] dB re 1 (μm s⁻²)²/Hz between 200 and 750 Hz for each site.

For L_p metrics, the arithmetic mean of the four hydrophones was taken to encompass pressure data from all hydrophones to compare with L_a, which was calculated using pressure data from all four hydrophones. The median [IQR] of the range of L_p among the four hydrophones (for each 60 s mean) was 1.11 [0.55, 4.70] dB re 1 μPa² for the low-frequency band and 0.67 [0.38, 1.16] dB re 1 μPa² for the high-frequency band. Therefore, some variability over time for single hydrophone data may be lost by averaging, yet averaging encompassed variability from multiple hydrophones that could be missed by using pressure data from only one hydrophone.

“Excess particle motion,” here defined as L_a − L_apw, was calculated for each 60 s window to investigate potential causes of measured particle motion being higher than expected for free-field plane waves. Within each frequency analysis band (low and high) and each site, audio samples with L_a within 3 dB of L_apw were compared with those with excess particle motion above the 95th percentile (“outlier”). For manual annotation of sounds in these time windows, 50 unique windows were randomly subsampled from the “plane wave” group and 50 were selected from the “outlier” group. Two sites had sample sizes for this analysis less than n = 50 in the low-frequency band for the “plane wave” group due to few audio samples having excess particle motion below 3 dB: HAT (n = 40) and Charleston Bump (CHB) (n = 33). This resulted in 1173 samples total for manual annotation; these samples were taken from the same dataset used for generalized additive models (GAMs) (see Sec. II C). These samples were manually tagged (by viewing spectrograms) for signal types, including mammal vocalizations, putative fish vocalizations, vessel sounds, signals of unknown origin which could be either part of the soundscape or instrument self-noise, and flow noise. For each site and frequency band, Fisher's Exact Tests compared proportions of each signal type in the plane wave group versus the outlier particle motion group.

Times of vessel sounds were identified with an automated detector, which operated on calibrated audio files, in 60-s segments, by detecting shipping tonals on overlapped short FFTs (FFT size = 64 000 frequency resolution: 0.125 Hz, time resolution 8 s) over a 40–315 Hz band (Martin, 2013). A total of 50% of the detections were manually checked for false positives. The false positive rates for each site were as follows: HAT, 0.7%; WIL, 1%; CHB, 5.3%; SAV, 16.3%; JAX, 0%; BLE, 0%. False positives at CHB and SAV may have been due to minke whales, which were usually present in files of false positive boat detection at these sites (minke whales were present at other times as well, with and without vessel sounds). Manually verified false positives of vessel sounds were corrected by tagging these 60-s means as “ambient,” defined as a time where no vessel sounds were present.

Statistical analyses were performed in RStudio version 2022.07.2.576 (RStudio Team, 2022). GAMs (using the “gam” function) with multiple predictor variables were performed separately for each site, and separately on L_p and L_a. These models aimed to determine what oceanographic, anthropogenic, and temporal variables best explained variation in L_p and L_a, and whether pressure and particle acceleration could be equivalently predicted by the same variables. GAMs were also performed separately for soundscape data in the two frequency bands (“low” from 10 to 50 Hz, and “high” from 100 to 750 Hz). The following predictor variables were investigated: month of year (numerical), vessel presence (binary, as determined from the automated vessel noise detector), surface wind speed, current speed 100 m above the seafloor, and bottom temperature, salinity, and oxygen (from MicroCAT sensors on the lander, all continuous variables). Month of year was the only nonlinear (smoothed) predictor, whereas other predictor variables were linear. Data from November 2017 and November 2018 were not included for models because, at most sites, data were only available for a few days of those months. Because temperature, salinity, and oxygen data were only available until July 10, 2018 at HAT, data points from that date onward were not included for HAT models. Soundscape levels, i.e., 60 s means of L_p and L_a, were subset to those at times within ±5 min of each time point of co-located temperature, salinity, and oxygen data (which wind and current data were also time-matched to). Due to the duty cycle and data gaps, not every 30-min interval had soundscape data within a ±5-min window of the non-acoustic data. Therefore, soundscape data were matched to irregular sample intervals resulting in sample sizes and median [IQR] time sample intervals for soundscape data given in Table III. The Log₁₀ of wind speed above 5 m s⁻¹ was taken for GAMs, as in the present study and prior studies the relationship between the log wind speed and sound pressure level had a change in slope at 5 m s⁻¹ (Hildebrand , 2021).

Significant autocorrelation of GAM residuals was found for each time series (i.e., site) and attempts to reduce this using autocorrelation structures within GAMs were unsuccessful. This may limit confidence in model parameters, and temporal trends outside of the “month” variable may be unaccounted for. However, the main goal of the present study was to determine which environmental factors were most important in explaining soundscape levels and to compare models for sound pressure levels with those for particle acceleration, rather than to optimize predictive models.

Sounds visible in long-term spectral averages (Fig. 3) at all sites included broadband wind sound with greatest energy above 100 Hz, and broadband vessel sounds (often spanning the entire analyzed frequency range) with greatest energy below 100 Hz. There were seasonal vocalizations from several low-frequency cetacean species, including minke, blue, fin, humpback, and sei whales, and a single occurrence of a North Atlantic right whale at HAT in January 2018; see Kowarski (2022) for a detailed report of marine mammal occurrence recorded from same sites and time period of the present study. Putative unidentified fish sounds (individual calls, no chorusing) were observed at BLE, all sharing the same spectrotemporal pattern, with predominant energy between 100 and 2800 Hz and durations of 40–50 ms (Fig. 4). Both the greater amplitude 100–2800 Hz calls and lower amplitude 200–900 Hz calls in Fig. 4 are believed to be the same call type from the same species, as they share the same temporal structure and peak frequencies (between 400 and 500 Hz). Fish sounds were restricted to the high band (100–750 Hz) for subsequent analyses. All signal types were visible in both sound pressure and particle acceleration spectrograms. Low-frequency flow noise was present at all sites, mostly below 50 Hz, but at CHB it sometimes exceeded 100 Hz.

Correlations between measured particle acceleration levels [L_a, Eq. (2)] and pressure levels (L_p) indicate how closely particle acceleration levels scale with pressure [Figs. 5(A) and 5(C); Table IV]. Correlations between L_a and plane wave particle acceleration [L_apw, Eq. (3)] indicate to what extent particle acceleration could be estimated with the plane wave approximation where values within ±3 dB of a 1:1 relationship indicate sounds that are approximated by Eq. (3) [Figs. 5(B) and 5(D); Table V]. In the low-frequency band, residuals of this 1:1 line had large spread, with a 5th percentile as small as 0 dB (at WIL) and 95th percentiles as great as 34 dB (at CHB and JAX). In the high-frequency band, residuals had smaller spread, with a 5th to 95th percentile range from –6 dB (at HAT) to 12 dB (at CHB). Across sites, only 0.3%–13.3% of points were within 3 dB of the 1:1 line (of L_a vs L_apw) for the low band versus 58.5%–77.8% of data points for the high band. After repeating this analysis for each decidecade band (not shown), the transition to a majority of data fitting the free-field plane wave approximation (defined as when >50% of data points were within 3 dB of the 1:1 line) was found to occur at the following decidecade bands for each site: HAT, 63 Hz; WIL, 63 Hz; CHB, 200 Hz; SAV, 200 Hz; JAX, 250 Hz; BLE, 50 Hz. Note that with this low threshold (50%), many L_a values were still >3 dB above L_apw values at higher frequencies. Some data points were below –3 dB from the 1:1 line in the high band as well, ranging from 7.6% (at CHB) to 36.6% (at HAT) compared with <1% at all sites in the low band. Across sites (n = 6) the variance of residuals of the 1:1 line was significantly less for the high band compared to the low band (paired t test, p < 0.01). There was no significant difference in R² values of L_a regressed with L_p (paired t test, df = 5, t = –1.76, p = 0.14; Table IV) or L_a regressed with L_apw (t = –1.38, df = 5, p = 0.23; Table V) when comparing low versus high bands. Thus, differences between low and high bands were best captured by variance (rather than the correlation coefficient, R²), with a greater variance and significantly greater spread of residuals about the 1:1 line for the low band compared to the high band. The slight positive curve of low amplitude (approaching 20 dB re 1 μm s⁻²) broadband L_a away from the 1:1 line could be caused by L_p being closer to the recorder noise floor (see Discussion).

Data points with excess particle motion (L_a − L_apw) above the 95th percentile, referred to as the “outlier” group, were compared against those with low (within ±3 dB) excess particle motion, referred to as the “plane wave” group, to determine possible causes of measured particle motion exceeding free-field plane wave predictions. At all sites in the low band and high band, flow noise was present in a significantly greater proportion of samples in the outlier group compared to the plane wave group (p < 0.05, Fisher's Exact Tests; Fig. 6; Table VI). Flow noise was most pervasive at CHB, in terms of its wide frequency range (often exceeding 100 Hz) and the proportion of audio files affected. Vessel sounds made up a significantly greater proportion of files in the plane wave group compared to the outlier group for all sites in the low band, and at HAT for the high band (p < 0.01), consistent with vessel sounds received in the far field and propagating as plane waves. Conversely, the proportion of vessel sounds in the outlier group was significantly greater than in the plane wave group for the high band at WIL (p < 0.05). Mammal sounds were statistically similar between plane wave and outlier groups, except at HAT in the low-frequency band where a significantly lower proportion of mammal vocalizations was found in the outlier group (potentially due to masking by flow noise). “Unknown” signals were significantly greater in the outlier group than the plane wave group for the low band at SAV (p < 0.05) and high band at WIL (p < 0.01).

At BLE, fish sounds were only in the outlier group, with a proportion significantly greater than in the plane wave group (p < 0.01), suggesting fish were calling in the near field. These fish sounds had excess particle motion in the high band between 4.8 and 12.6 dB (mean: 7.6 dB; n = 8 files out of a subset of 50). Modeling the fish signal as a monopole point source [using Nedelec , 2021, their Eq. (39)] the range r of the fish to the receiver was estimated assuming a frequency f of 100 Hz or 750 Hz (the edges of the high-frequency band) and a sound speed c of 1500 m s⁻¹. For the 4.8 dB and 12.6 dB excess particle motion values, kr was estimated at 0.7 and 0.25, respectively, leading to estimated r of 22–167 cm for the 4.8 dB signal and 8–60 cm for the 12.6 signal. Detection at such distances is plausible given opportunistic observations of fishes at the lander at other sites (see Fig. 9). Putative fish sounds were not found at other sites.

Bivariate histograms of particle motion signals (horizontal x or y axis versus vertical z axis) were visualized for five fish signal time windows and five flow noise time windows (100 ms for each), belonging to the “outlier” group at the BLE site (examples shown in supplementary material). This helped determine which particle motion signals may not be accurately estimated by Eq. (3), due to flow noise or reflections off the seabed which are expected to lead to more circular particle motion, i.e., wider ellipses (Campbell , 2019; Campbell and Nedelec, 2023; Dahl and Bonnel, 2022). Conversely, acoustic signals that do not experience interference from reflected sound waves [and accordingly should be more accurately estimated by Eq. (3)] are expected to have more linear particle motion, i.e., narrower ellipses. The fish signal and flow noise ellipticities were analyzed in a decidecade bands centered at 500 and 40 Hz, respectively, to focus on peak frequencies of each. Qualitative comparisons revealed fish particle motion signals were highly linearly polarized, whereas flow particle motion signals were less linearly polarized and more circular.

Monthly low band L_a had more variability compared to low band L_p (Fig. 7). However, monthly medians and distributions of soundscape levels for each site followed a similar pattern across time when comparing L_a and L_p. Low band L_a and L_p had weak seasonal patterns, with most sites having median monthly pressure levels 3–5 dB greater in winter months than in summer and early fall, although levels at CHB were much more variable month-to-month (possibly due to strong flow noise). High band L_a and L_p had larger seasonal differences for each site, about 10 dB greater in winter than in summer. Seasonal trends in the high band were more similar between L_a and L_p compared to those in the low band.

Surface wind speed at each site had a strong seasonal trend consistent across sites, with greater wind speed in winter than summer [Fig. 8(A); see supplementary material]. WIL had many outlier wind speed values on September 12–13, 2018, coinciding with Hurricane Florence (Tripathy , 2021). Median bottom temperatures were the coolest, down to about 5 °C, at BLE and SAV (the deepest sites) and warmest, up to about 16 °C, at HAT [the shallowest site; Fig. 8(B); see supplementary material]. Within each site, bottom temperatures were warmest between March and June. Current speeds (100 m above the seafloor) had site-specific seasonal variability [Fig. 8(C); see supplementary material]. For example, median current speeds were greatest between May and June at BLE, but greatest between December and February at WIL. Median current speeds had less variability across months at SAV compared to other sites, and BLE had the smallest median current speeds of all sites, below 0.1 m (in December–February and August).

Vessel sound detections were highly variable across sites, and across time at most sites [Fig. 8(D); see supplementary material]. BLE had the most vessel detections in December, and August and September. JAX had few detections in December through May (below 10%) with a peak at 27% in July; SAV and HAT had a similar seasonal trend to JAX but with peak vessel activity in May (14%) and July (25%), respectively. WIL had increasing vessel detections from December 2017 to September 2018 with a peak of 30% in August 2018. Vessel detections were fewest at CHB, not exceeding 5% of files per month (see supplementary material), likely due to masking of vessel sounds by strong flow noise.

All GAMs had multiple statistically significant predictors (p < 0.001); however owing to large sample sizes, regressions with weak correlation (adjusted partial R², hereafter “partial R²,” <0.1) could still be statistically significant (Table VII). Therefore, these results primarily focus on comparing R² values rather than p values. Such an approach is called “effect size statistics,” which focuses on reporting the magnitude of an effect regardless of statistical significance and is useful when working with large sample sizes (Nakagawa and Cuthill, 2007). Across sites, surface wind speed was a strong predictor of L_p and L_a in the high band (100–750 Hz; mean partial R² = 0.42 and 0.37, respectively) and weaker predictor of low band L_p and L_a (10–50 Hz; mean partial R² = 0.06 and 0.08, respectively). Vessel presence was a strong predictor of L_p in the low band and high band (mean partial R² = 0.15 and 0.16, respectively) but weaker predictor of L_a in the low and high bands (mean partial R² = 0.08 and 0.07, respectively) Temperature had weak correlations with low band L_p and L_a (partial R² ≤ 0.20) and weaker correlations with high band L_p and L_a (partial R² ≤ 0.08). Current speed (100 m above the seafloor) had even weaker correlations with L_p and L_a, with partial R² ≤ 0.16 and ≤0.07 for low and high bands respectively. Month was a significant weak predictor for each model (partial R² was not evaluated as this was a nonlinear predictor). Most GAM estimates indicated positive correlations (68 of 96 linear estimates, including all models). Wind speed and vessel presence had negative correlations with L_p or L_a (at some sites) only for the low band, with partial R² < 0.14. Bottom temperature and current speed had negative correlations with L_p or L_a in some models, but correlations of these variables were positive for models with the highest partial R² (>0.14) (Table VII).

Adjusted total R² values (hereafter, “model R²”) were not significantly different for models of L_p versus those of L_a in the high band (two-tail paired t test, t = 1.46, p = 0.20), nor in the low band (t = 0.48, p = 0.65). For high band L_p and L_a, CHB had the poorest fit (model R²: 0.27, 0.2, respectively), whereas WIL, SAV, and BLE had the best fits (model R²: 0.47–0.62). For low band L_p and L_a, model fit was poorest for CHB (model R²: 0.08, 0.09, respectively), better for SAV and JAX (model R²: 0.10–0.22), and best for HAT, WIL, and BLE (model R²: 0.20–0.31).

To further investigate whether excess particle motion levels could be explained by hydrodynamic flow (i.e., current speed), we repeated GAMs (with the same predictor variables) on low band particle acceleration levels separately for the outlier group and the plane wave group (see supplementary material for model summary tables). Adjusted partial R² for current speed was higher for the outlier group than for the plane wave group at BLE, CHB, HAT, and WIL, and not at SAV and JAX. Partial R² for current speed (mean of all six sites) was not significantly different between the outlier and plane wave group (paired two-tailed t test, df = 5, t = 1.29, p = 0.25). However, GAMs indicated current speed was a significant predictor (p < 0.001) of low band acceleration in the outlier group and not a significant predictor in the plane wave group at four sites: BLE, CHB, JAX, and WIL. Coefficients for current speed in the outlier group indicated positive correlation with low band acceleration levels at BLE (4.95), JAX (13.18), and WIL (4.40). Curiously, the coefficient in the outlier group at CHB was negative (–8.05), although the adjusted partial R² for current speed in the outlier group was lowest at this site (R² = 0.06) compared to other sites with positive correlations (R² range of 0.08–0.19). Together, these results indicated that excess particle motion levels were in part related to (and positively correlated with) current flow.

This study is the first to report long-term (1-year) measured particle motion alongside sound pressure of deep-water continental slope soundscapes. Soundscapes in the ADEON study region were dominated by wind and vessel sounds, all sites contained marine mammal vocalizations, and one site, BLE, had putative fish sounds. Across sites, measured particle motion (L_a) correlated poorly with sound pressure (L_p) and free-field plane wave particle acceleration (L_apw) in the low band (10–50 Hz) where L_a exceeded L_apw by over 30 dB. Conversely, L_a correlated more strongly with L_p and L_apw in the high band (100–750 Hz); this result is consistent with a recent study that reported closer KE and PE equivalence with increasing bandwidth, where KE ≈ PE in 1/3 octave bands (Flamant and Bonnel, 2023). Yet, outliers of excess particle acceleration (L_a − L_apw) up to 24 dB were measured in the high band in the present study. At all sites, flow noise led to outliers of excess particle acceleration in both frequency bands (although this was most prevalent in the low band), and at BLE in the high band some of the outlier excess particle acceleration levels were due to fish vocalizations. Seasonal temporal trends of L_a closely followed those of L_p at each site, but with higher variability for L_a in the low band. Variation of both L_a and L_p (over an 11-month dataset) was similarly explained by environmental indicator variables, including surface wind speed and vessel presence (and month) which were the strongest predictors, followed by bottom temperature and mid-water current speed which were weaker predictors.

Potential causes of L_a in the present study deviating from L_apw include the following: (1) acoustic signals received in the near-field, (2) hydrodynamic flow noise, (3) the waveguide cutoff frequency and (4) reflections from the seafloor.

Sounds received in the near field have greater particle motion magnitudes than those expected for a free-field plane wave. Sounds likely to occur in the near field at these sites include those from soniferous fishes and invertebrates at and nearby the lander. The strongly directional (i.e., close to linear) bivariate histograms of fish particle motion signals match expected particle motion ellipticity for a free-field plane wave (see supplementary material). However, particle motion magnitudes measured for fish signals were far higher (up to 12.6 dB) than free-field plane wave predictions [Eq. (3)]. In the high band at BLE, a greater proportion of audio samples in the outlier excess particle motion group had fish signals compared to the plane wave group (Figs. 4 and 6). Together, these results suggest fish sounds were received in the near field. Lack of video data and limited databases for fish vocalizations precluded identifying what species produced these sounds. However, opportunistic video (Fig. 9) from the ROV Jason during recovery of landers at a later deployment at WIL (December 10, 2020) and JAX (December 13, 2020) revealed the presence of potentially soniferous fishes, including white hake (Urophycis tenuis) and blackbelly rosefish (Helicolenus dactylopterus) at both sites, and conger eels (Conger oceanicus) at WIL. To the author's knowledge, neither intentional vocalizations (e.g., for communication) nor incidental sounds (e.g., from feeding) have been documented for these fish species. However, red hake (Urophycis chuss) which shares a genus with white hake, and other eels in the order Anguilliformes such as the American eel (Anguilla rostrata) have documented sound production (Fish and Mowbray, 1970; Kaschner, 2012). No signals were found that the authors could confidently label as invertebrate sounds. Notably, however, spider crabs (superfamily Majoidea) were present at both sites. Sounds from at least one spider crab species, Maja brachydactyla (which does not inhabit the ADEON region) have been recorded, including sounds associated with feeding behaviors, e.g., tearing prey (Coquereau , 2016). Shortfin squid (Illex illecebrosus) were also observed at WIL; although no squid species are known to purposefully vocalize, squid sense and behaviorally respond to acoustic particle motion (Mooney , 2010; Samson , 2016) Although these observations took place 2 years after the present study, they demonstrated that soniferous fish and invertebrate taxa were present around these landers in December at the same locations of the present study. Simultaneous and co-located video data with audio data in future offshore deployments would be useful to connect nearby fish and invertebrate sounds with the species visually present, as has been done in nearshore habitats including coral reefs and kelp forests (Pagniello , 2021; Tricas and Boyle, 2014).

Hydrodynamic flow noise was a major cause of excess particle motion measured in the present study. Flow noise was expected considering all sites were located beneath the Gulf Stream. For all sites, a significantly greater proportion of audio samples with outlier (above the 95th percentile) excess particle motion levels contained flow noise compared to samples with particle motion near (within 3 dB of) levels estimated for free-field plane waves (Fig. 6). This was true for both low- and high-frequency bands, although the differences were greatest in the low band where most of the energy of flow noise was contained. Further, GAMs for low band particle acceleration levels in the outlier group indicated significant positive correlation between these levels and current speed (versus nonsignificant correlation in the plane wave group) at BLE, JAX, and WIL. Bivariate particle motion data (see supplementary material) of flow noise were widely elliptical, further indicating these were not far field acoustic signals propagating as plane waves. Flow noise is usually considered unwanted noise because they mask the detection of truly acoustic sources (Gray , 2016; Nedelec , 2021). However, when present, flow signal picked up by hydrophones could potentially be a useful proxy for measuring local current flow (discussed in Sec. IV B).

Signals from vessels and marine mammals were received in the far field. Accordingly, there were either significantly smaller proportions of these signals in the outlier excess particle motion group than in the plane wave group, or no significant differences in proportions between these groups. Flow noise may have masked some of these signals in the outlier group, but the present results suggest that, for these continental slope sites, the plane wave approximation is suitable for reporting particle motion magnitude of these sources. The same is expected for sounds from the surface due to wind, also received in the far field at these depths, although wind sounds were not quantitatively compared between outlier and plane wave particle motion groups.

Sound below the waveguide cutoff frequency [f₀, Eq. (4)] can also lead to deviation of measured particle motion from that expected for plane waves (Nedelec , 2021). The lowest frequency analyzed in the present study, 10 Hz, was above f₀ for all sites (f₀ estimated to be no higher than 4.2 Hz for the shallowest site, HAT, and lower for deeper sites). Thus, this waveguide cutoff is unlikely to have influenced particle motion measurements in the present study.

In shallow-water waveguides, interference of out-of-phase reflected sound waves (including plane waves) off the seabed and sea surface can result fluctuating measured particle motion to pressure ratios that depart from the relationship of the free-field “plane wave approximation,” i.e., Eq. (3) (Gray , 2016; Parsons and Duncan, 2011). Measured particle acceleration departed from the plane wave approximation with 5th to 95th percentile ranges of 0 to 34 dB in the low-frequency band and –6 to 12 dB in the high-frequency band. This variability extended well beyond that estimated via a sensitivity analysis for particle motion measurement error (absolute error up to 6 dB at 100 Hz). This result along with the fact that not all excess particle acceleration values in the “outlier” group (above the 95th percentile) included flow noise indicates wave interference from seabed reflections may contribute to large positive (and in some cases negative) excess particle motion. In the present study, this could also include waves reflected off the structure of the ALTO landers. Future studies measuring ratios of pressure to particle motion at varying distances closer to the seafloor, and for a variety of substrate types and depths, would be useful to better understand influences of these boundary effects on particle motion measurements and the acoustic fields that benthic fishes and invertebrates experience (Nedelec , 2016).

Notably, some data with low measured broadband particle acceleration levels (e.g., <20 dB re 1 μm s⁻²) deviated more from plane wave estimated particle acceleration, with a slight positive curve of the 1:1 line (Fig. 5). This is potentially caused by L_p being closer to the recorder noise floor for these data. Incorporating data from two hydrophones to calculate particle acceleration in Eq. (1) adds 3 dB to the recorder noise floor for an effective noise floor of 37 dB re 1 μPa²/Hz. Time samples with low broadband L_a had corresponding sound pressure power spectral density as low as 50 dB re 1 μPa²/Hz above 100 Hz and as low as 40 dB re 1 μPa²/Hz below 100 Hz (Fig. 3). Signals even as much as 20 dB above the noise floor can lead to overestimates of particle acceleration (calculated using Eq. (1) by several dB with smaller signal-to-noise ratios leading to greater expected overestimates (Gray , 2016). Given this limitation and observations in the present study, future soundscape studies quantifying particle motion should consider applying minimum signal-to-noise thresholds of at least 20 dB re 1 μPa²/Hz for all frequencies of interest.

Wind speed and vessel presence were strong predictors of sound pressure and particle motion soundscape levels in both frequency bands, and especially in the high-frequency band (>100 Hz), consistent with prior studies of offshore ambient soundscapes (Crouch and Burt, 1972; Haver , 2021). Also, month of the year was a significant predictor for soundscape levels at all sites, corresponding to strong seasonal differences in wind speed at all sites and seasonal difference in vessel sounds at some sites. This seasonal variation is typical of temperate and subtropical North Atlantic offshore soundscapes, reported as sound pressure by previous studies, including at ADEON sites (Haver , 2021; Miksis-Olds , 2022). Less commonly reported in the literature, however, are correlations of soundscape levels with bottom temperature and current speed, which were weak but statistically significant in the present study.

Sound pressure and particle acceleration levels had strong (highest partial R²) and positive correlations with temperature at WIL HAT, and SAV. Negative correlations between sound pressure or particle acceleration levels and temperature were found at CHB and JAX, and these were weaker on average than the positive correlations at other sites. Regarding potential mechanisms linking these variables, bottom temperature could be a proxy for local current flow (leading to flow noise, increasing L_rms; discussed further below), or alternatively a proxy for sound speed profiles. Regarding sound speed profiles, CTD casts of temperature depth profiles taken in June and November at each site (supplementary material) provide some evidence for association between changes in bottom temperature and an overall change in temperature profile and sound-speed profile. Temperature profiles at WIL, HAT, and SAV steadily decreased with depth (after a shallow thermocline and/or isothermal surface layer), with a thin or non-existent deep-water isothermal layer just above the seafloor. This suggests a sound speed minimum likely occurred at (rather than above) the seafloor, which would lead to more sounds from the surface (e.g., from vessels) refracting to reach the seafloor and increasing L_rms (Ainslie, 2010). Conversely, JAX and CHB had weak correlation between soundscape levels and temperature; these sites had a thick deep-water isothermal layer, which could limit downward refraction of sounds and lead to more upward refraction. A recent soundscape model included seasonal variations in sea surface temperature as a proxy for critical surface angle (below which sound rays are trapped in a waterborne waveguide and above which rays are refracted downward to interact with the seabed) as a predictor for spatially averaged sound fields in the Northeast Pacific (Ainslie , 2021). In that study, warmer sea surface temperature corresponded to a lower critical surface angle and lower mid-water soundscape levels. In the present study, warmer bottom temperature levels at WIL, HAT, and SAV could be associated with an overall positive shift in temperature profile and sound speed profile where more low-frequency sounds near the surface are refracted downward at steeper angles (relative to the surface) leading to higher received soundscape levels near the seafloor. This hypothesis could be explored with additional temperature profile measurements to compare bottom temperature, sea surface temperature, and the difference between the two (i.e., temperature stratification) with soundscape data.

Past studies in nearshore environments have used hydrophone arrays to detect tidal currents and measure tidal current speed, although in these studies tidal signals were likely acoustic and received further away (versus non-acoustic flow noise present in the present study), or were measured via cross-correlations of acoustic data from hydrophones spaced much further apart than in the present study, i.e., >5 km (Godin , 2014; Rogers , 2021). A possible link between current speed and soundscape levels is flow noise, which would be expected to result in a positive relationship between the two variables; importantly however, flow noise is unique to a given measurement configuration (i.e., lander and hydrophones in the present study); thus, comparisons across studies or different passive acoustic systems are not straightforward. Among sites where current speed was a significant predictor in GAMs (all but BLE), CHB had negative correlations in the low band with pressure and acceleration and SAV had a negative correlation in the high band with pressure. These negative correlations are counterintuitive and may be a result of large sample size leading to statistical significance for a negligible trend; notably, partial R² values (Table VII) were much smaller for current speed than for other environmental predictors, suggesting current speed is a poor predictor of soundscape levels at these sites. However, currents can be highly variable with small changes in depth in this region (Andres , 2023). Because currents from HYCOM models were taken 100 m above the benthos for the present study, these data may not reflect small spatial scale shifts in current speed at the lander. HYCOM data at the seabed were initially used, but these data were zero-inflated and led to even weaker correlations (lower R²); hence, data further above the seabed were used. The temporal resolution of the HYCOM model data may also be too coarse to reveal short term (less than an hour) variability in currents. To further investigate possible use of local current data to predict soundscape levels, co-located current data (e.g., with an Acoustic Doppler Current Profiler) should be collected simultaneously in situ with acoustic data.

Of the environmental variables considered, wind speed and vessel presence remain superior predictors for soundscapes of these continental slope sites, explaining more of the variability in both pressure and particle acceleration than bottom temperature and current speed.

The largest differences between measured particle acceleration data compared to free-field plane wave particle acceleration (and compared to sound pressure) were in the 10–50 Hz band. This result is similar to a recent study of shallow (7–11 m deep) nearshore coral reefs, in which measured particle acceleration levels at short time scales (L_rms integrations over a few seconds) and at frequencies below 100 Hz were furthest above those predicted from a free-field plane wave assumption (Jones , 2022). The coral reef study attributed low-frequency excess particle motion to nearfield biological signals and boundary interactions in shallow water. In the present study, the greatest excess particle motion levels (at frequencies below 100 Hz) were attributed to hydrodynamic flow. Interference of waves reflecting off the seabed may have also contributed to excess particle motion levels below 300 Hz, as below that frequency the hydrophones on the lander (positioned as close as 1.25 m above the seabed) are within ¼ wavelength from the seabed (Oppeneer , 2023). Based on the present results, when measuring sounds from vessels and marine mammals (and likely wind) in offshore habitats sound pressure from single hydrophones may approximate 25th to 75th percentile magnitudes of particle motion within ±3 dB, at least above 100 Hz. However, even above 100 Hz received particle motion of individual signals (60 s integrations in the present study) could be underestimated by the plane wave approximation by at least 12 dB (at the 95th percentile). Fish sounds at BLE often had large excess particle motion, and potentially were received in the near field. Notably, fish communication often occurs in the near field (Higgs and Radford, 2016), so measurement of nearfield signals is of great biological relevance to fishes and assuming free-field plane wave conditions would underpredict their particle motion levels. Further, analysis of excess particle motion, as done in the present study, can be a useful tool to quickly identify potential nearfield transients of biological origin (after accounting for flow noise). Importantly, the present study primarily focused on magnitudes of particle motion rather than directional information, which cannot be obtained from single hydrophone measurements of sound pressure. The directionality of particle motion (including main axes of motions for individual sounds and particle ellipticity) is likely relevant to many fishes and invertebrates for sound detection and localization and should be investigated further in future soundscape studies (Popper and Hawkins, 2018).

Measuring sound pressure of offshore soundscapes alone (without particle motion measurement) appears sufficient for capturing longer-term temporal patterns (11 months in the present study) and for predicting soundscape levels from environmental variables, at least those variables assessed at the sites in the present study. Although more variable in the 10–50 Hz band than at higher frequencies (likely due to flow noise), particle acceleration followed similar seasonal patterns to sound pressure at each site. Sound pressure and particle acceleration levels were also similarly predicted by month, vessel presence, wind speed, and bottom temperature and currents in GAMs. Notably, GAMs for the high band explained significantly more variation in the sound pressure level data than in the particle acceleration data. Thus, sound pressure may be more suitable than particle acceleration to use in developing models explaining and predicting soundscape levels at these sites, based on the environmental indicators presently investigated. Importantly, the same cannot be assumed for shallower and nearshore sites, especially at low frequencies. Shallower water will have a higher low-frequency cutoff below which sounds will not propagate as free-field plane waves, and stronger boundary interactions with the surface and seafloor that can lead to pressure and particle motion further deviating from the “plane wave approximation,” i.e., Eq. (3) (Nedelec , 2021).

Soundscapes throughout the U.S. Mid- and South Atlantic Outer Continental Shelf were quantified with sound pressure and acoustic particle motion using compact hydrophone arrays on bottom landers at six sites. Trends of measured particle motion levels were compared with those of sound pressure and to particle motion levels predicted for a free-field plane wave (using single hydrophone measurements and characteristic-specific acoustic impedance), a commonly used approximation for particle motion in offshore environments. This study found that measured particle motion levels were more variable than sound pressure levels and were in excess of free-field plane wave-predicted levels by up to 34 dB at the 95th percentile at lower frequencies (<50 Hz). At higher frequencies (>100 Hz), measured particle motion levels correlated more closely with sound pressure and plane wave-predicted levels, although still with a wider range outside of estimated measurement error (beyond ±6 dB). Excess measured particle motion in the 10–50 Hz band was primarily attributed to hydrodynamic flow noise. Unidentified fish sounds were found at one site, Blake Escarpment, which were likely received in the near field and significantly contributed to higher measured particle motion levels above those that would be predicted for free-field plane waves. This study revealed that, although many sounds in offshore habitats, including those from marine mammals and vessels, can be approximated as free-field plane waves to within ±3 dB if only median values (or those within an interquartile range) are of interest, but individual signals can have excess measured acceleration further above plane wave predictions (up to 12 dB at the 95th percentile, for a 100–750 Hz band). As well, near field biologically relevant sounds can also be recorded that necessitate measurement with vector sensors to accurately report the particle motion of these individual sounds. For comparing temporal patterns of offshore soundscapes, e.g., assessing seasonal differences, particle motion, and sound pressure provided similar results.

Variability in sound pressure was similarly, if not better predicted by month and environmental indicators including wind speed, vessel presence, bottom temperature, and current speed, compared to particle motion. Wind speed and vessel presence were the strongest predictors, consistent with prior studies of offshore soundscapes. Bottom temperature and current speed weakly and significantly correlated with soundscape levels, and hypothesized mechanisms for these relationships require further investigation.

Overall, this study established a useful baseline of acoustic particle motion levels in Northwest Atlantic continental slope environments and their predictability from sound pressure measurements and key environmental indicators. These results can be compared with future studies of particle motion soundscapes in other offshore and nearshore environments to enhance understanding of relationships between pressure and particle motion in different habitats, which is of high relevance to bioacoustics studies involving fishes and aquatic invertebrates. Although the present study focused on magnitudes, directionality of particle motion is also biologically relevant to these taxa and should be given further focus in future studies. As well, potential ties between bottom temperature, currents, and soundscape data should be further investigated with simultaneous direct measurements of temperature and current profiles with soundscapes.

See the supplementary material for measured particle acceleration, sensitivity analysis results, monthly trends of environmental variables at additional sites, bivariate histograms of measured particle acceleration, detailed Generalized Additive model outputs, and example temperature and salinity depth profiles from study sites.

The ADEON project was supported by the U.S. Department of the Interior, Bureau of Ocean Energy Management, Environmental Studies Program, Washington, DC, Contract No. M16PC00003. Funding for ADEON ship time and additional funding for I.T.J. were provided by the Office of Naval Research, Code 32. We thank the crews of the RV Endeavor and RV Niel Armstrong and the many students, scientists, and collaborators responsible for data collection during the ADEON project cruises; Theresa Ridgeway and Research Computing Center staff at University of New Hampshire for assistance with data management; Thomas Butkiewicz and Ilya Atkin, University of New Hampshire Center for Coastal & Ocean Mapping, for developing online spectrogram visualizations utilized in this study; Kevin Heaney for providing sediment grain size estimates at ADEON sites; and two anonymous reviewers whose comments helped improve our manuscript.

The authors have no conflicts of interests to disclose.

Raw hydrophone data (.wav files) and temperature data (MicroCAT CTD) utilized for this paper are openly available at https://adeon.unh.edu/data_portal (Miksis-Olds, 2023) and on NCEI databases at https://www.ncei.noaa.gov/maps/passive-acoustic-data. Derived data are available from the corresponding author upon request.