Multidimensional comparison of underwater soundscapes using the soundscape code^a)

Aquatic species use sound cues in local environments for a variety of activities: foraging, navigation, habitat selection, predator detection, migration, and breeding (Wartzok and Ketten, 1999; Richardson , 1995; Tyack, 1998; Bass and McKibben, 2003). Consequently, by studying the ambient sound field, it is possible to learn about a specific region from the sounds recorded in the environment. The study of ambient sound and acoustic ecology has led to the development of the underwater soundscape concept, where the soundscape is an acoustic environment that provides insight into the function of a given location. The soundscape is made up of all the acoustic signals that arrive at an animal receiver or acoustic recorder (Pijanowski , 2011). The underwater soundscape was formally defined by ISO 18405 as the “characterization of ambient sound in terms of its spatial, temporal, frequency attributes, and the types of sources contributing to the sound field” (ISO, 2017).

The combination of sounds present in an environment produce unique and distinct soundscapes, which can be linked to ecosystem function and variable environmental conditions of proximate and distant habitats (Radford , 2010; Radford , 2014; Staaterman , 2013; Lillis , 2014; Bertucci , 2015). Healthy coral reef habitats, whether they are thriving or have recovered from damage, produce soundscapes that are significantly different than degraded reef habitats due to increased biodiversity and abundance of acoustically active marine life (Piercy , 2014; Kaplan , 2015; Bertucci , 2016; Lamont , 2022). Presence of marine mammal species and biodiversity of deep ocean environments are also assessed by studying soundscapes (Parks , 2014; Erbe , 2015; Ahonen , 2017; Kowarski , 2023). Underwater soundscapes may also contain significant contributions from anthropogenic sources, predominantly commercial shipping and signals from seismic surveys, which can be seasonal and pervasive (Nieukirk , 2004; McDonald , 2006; Hildebrand, 2009; Martin , 2017; Ahonen , 2017). Recreational boating contributes significantly to shallow, coastal soundscapes (Hermannsen , 2019). When habitat quality is assessed with regard to anthropogenic impact, soundscapes can be used to deduce how human-generated noise may be affecting an environment.

In this study, soundscapes were assessed to determine whether deep, cold-water coral reef soundscapes differ from deep-sea soundscapes of sandy substrate habitats in the same region and whether warm and cold-water coral reefs exhibit similar soundscape characteristics. The soundscape of a coral reef is connected to important ecological information pertaining to overall reef health and diversity of reef organisms due to a variety of soniferous fish and invertebrate inhabitants that contribute to a reef soundscape (Piercy , 2014). Biologically louder and more diverse soundscapes relating to healthier reef habitats are more attractive to marine life and act as acoustic cues in larval settlement (Montgomery , 2006; Simpson , 2008). Evidence that shallow water environments, including coral reef habitats, have unique soundscapes (Bertucci , 2015; Lillis , 2014; Radford , 2010; Radford ., 2014) suggests cold, deep-sea reef environments may also have unique soundscape signatures important to marine life; thus, soundscape information may provide insight into these poorly understood, deep-sea environments.

Deep-sea coral habitats are common in the Outer Continental Shelf (OCS) region off the southeastern U.S. within the exclusive economic zone, but few have been mapped or characterized in terms of benthic biology (Reed , 2006). A variety of coral species dominate this Southeastern U.S. region, including Oculina varicose, Lophelia pertusa, Enallopsamia profunda, Madrepora oculata, and Solenosmilia variabilis (Reed, 2002a, 2002b). These deepwater reefs provide hard bottom substrate and habitat for a variety of sessile macrofauna, which, in turn, provide habitat for biologically rich and diverse communities of mobile macrofauna, including fish, crustaceans, and mollusks (Reed , 2006). Direct observations of deep-sea, benthic communities enabled by submersibles identified 142 taxa of benthic invertebrates and 58 taxa of fish at 6 deepwater reefs in the Southeastern U.S. region of the OCS (Reed , 2006). Qualitative differences in species richness among these six OCS sites suggest that unique habitats and community structures exist in the cold, deepwater coral environments; considerable differences in benthic communities were even observed between adjacent sites. There is mounting evidence that deepwater corals provide important fish habitat (Costello , 2005; Ross, 2006) analogous to warm-water reefs and can be used to glean information regarding historic environmental change and habitat connectivity (Adkins , 1998; Williams , 2006). In this study, two different reef comparisons were made. A general warm- vs cold-water coral reef soundscape comparison was examined between two disparate locations [the Great Barrier Reef (GBR; warm) and corals reefs of the Atlantic OCS (cold)] while a second comparison captured similarities/differences among cold-water reef soundscapes at relatively close locations within the Atlantic OCS. The value of including the single GBR exemplar for comparison to the two OCS reefs was twofold as it provided (1) a benchmark of performance for the soundscape code (SSC) in identifying quantitative differences in what was expected to be dissimilar geographical soundscapes due to differences in sound propagation, vocalizing species, and level of anthropogenic contributions (Wilford , 2021); and (2) the first soundscape comparison between a warm- and cold-water reef. It was recognized during the study design that the northern and southern hemisphere locations did not align in season or marine life, which made the reef comparisons attractive in assessing the multidimensional performance of the newly developed SSC method (Wilford , 2021) in characterizing two soundscapes as distinctly different. During aural review of the deep and shallow water reefs by human analysts, it was clear that these are different acoustic environments. If the SSC analysis does not quantitatively discriminate what the human auditory system could, then confidence in the value and application of the SSC approach would be low.

Deep-sea soundscapes have the potential to be an important sensory cue for a variety of marine animals, especially with regard to deep-sea coral habitats. Fishing and mining interests are beginning to expand into deeper regions of the ocean (Roberts, 2002; Lin , 2019), which makes understanding these environments and establishing baseline habitat information paramount for monitoring changes. The goals of this study were to (1) evaluate how habitats varying in depth and bottom substrate differ in soundscape properties and (2) establish baseline soundscape information for poorly understood deep-sea regions.

Sound recordings at four deep-sea sites (depths greater than 450 m) along the OCS region of the southeastern U.S. consisted of deep-sea coral (Savannah Deep, SAV; Richardson Hills, RH) and non-coral, sandy bottom (Wilmington, WIL; Blake Escarpment, BLE; Fig. 1). SAV, WIL, and BLE were sites in the Atlantic Deepwater Ecosystem Observatory Network (ADEON) network (Miksis-Olds , 2021; Kowarski , 2023), and recordings at RH were acquired as part of the DEEPSEACH project (McPherson, 2020). The ADEON and RH locations are located adjacent to or within the Gulf Stream, which is the dominant oceanographic feature in the OCS area. Vessel traffic (Miksis-Olds , 2022), multiple species of marine mammals (Kowarski , 2023), and transient storms/hurricanes (Tripathy , 2021) were the predominant sources recorded across the OCS sites. All OCS recordings were made in 2019. A fifth recording from Wheeler Reef in 2013, a mid-shelf, shallow (6 –18 m) coral reef environment part of the GBR), was used for the shallow coral soundscape comparison. In a 2014 study of several GBR habitats, Wheeler Reef was found to have higher coral cover, structural complexity, and coral genera richness than nearby reefs that had been disturbed by crown-of-thorns starfish outbreaks and coral bleaching events (Graham , 2014). Wheeler Reef is also protected from fishing and dominant southeastern wave action by three large adjacent reefs (Marshall, 1986); water flow and wave action contribute natural abiotic sounds to the soundscape (Bertucci , 2015), hence, the sheltering reefs adjacent to Wheeler Reef have the effect of reducing noise from these sound sources.

A mid-April to mid-May time period was analyzed at the five sites. Time series were analyzed using the 1-min metrics of the SSC (Wilford , 2021) to quantify soundscape information and compare the soundscapes of deep-sea coral and sandy bottom environments with the soundscape of a shallow coral environment. This analysis period was chosen such that all recordings were related to a similar time period as recordings from RH and GBR were the limiting recordings with only April/May data available. This presented a seasonal disparity between the Northern hemisphere sites (SAV, WIL, BLE, and RH) and Southern hemisphere site (GBR; Table I). As a result of the data limitations at RH and GBR, bias in terms of biological activity and seasonal vocal behaviors between the OCS and GBR sites could have influenced the comparisons; however, due to the large difference in ecological habitat in terms of depth and temperature between the cold (OCS reefs) and warm (GRB) water reefs, analogous marine life and seasonal biological activity conditions were not possible or expected. Data from SAV, WIL, and BLE were recorded on an AMAR G3 recorder (JASCO Applied Sciences, Dartmouth, NS, Canada) and consisted of 2057 min of passive acoustic data per site sampled at 375 kHz. Data from the RH were recorded on an icListen Smart Hydrophone (Ocean Sonics, Truro Heights, NS, Canada) and consisted of approximately 1440 min of passive acoustic data sampled at 128 kHz. Data from the GBR were recorded on an AMAR G3 recorder (JASCO Applied Sciences, Dartmouth, NS, Canada) and consisted of 2899 min of passive acoustic data sampled at 64 kHz (Table I). All hydrophones were calibrated before and after deployment, therefore, absolute pressure level comparisons could be made. Data from the OCS sites were low-pass filtered such that the upper analyzed frequency ranges matched those of GBR (32 kHz).

These sites were expected to offer valuable comparisons in depth and bottom type. Specifically, coral soundscapes were hypothesized to be different than sandy, soft sediment habitat soundscapes (Fig. 1). The WIL site is unique in that it is within a general region classified as live hard bottom coral (NOAA, 2023), but ROV (Remotely Operated Vehicle) Jason video footage showed the exact location as a mix of sand and small coral rubble (dead coral pieces). This site offered the opportunity to determine if soundscape characteristics can inform a degraded or non-functioning reef compared to adjacent healthy reefs with live coral cover.

The SSC metrics compared the five soundscapes in terms of amplitude, impulsiveness, acoustic dissimilarity (or uniformity), and periodicity. Sound pressure level root-mean-square (SPL) and peak sound pressure level (PK), well-established measures of soundscape amplitude (Merchant , 2015), were used to report average and maximum sound pressure levels of the soundscapes. Sound pressure kurtosis captured the impulsiveness of the soundscapes, which can be used to assess a variety of soundscape details, including the potential for auditory injury among any present marine animals as well as the content of transient signals (Lei , 1994; Hamernik , 2007; Qiu , 2013; Qiu , 2020; Martin , 2020). Measures of dissimilarity (D-index), proposed as a proxy for biodiversity (Sueur , 2008), were used to capture and report minute-to-minute spectral and temporal changes in the soundscapes. Small D-index values suggest soundscapes are more “acoustically uniform.” The periodicity metric, first proposed in Wilford (2021), reported the content of repetitive signals by quantifying the number of peaks in the autocorrelation of averaged 1-min pressure time series, a process inspired by Martin (2019). Echolocation clicks and seismic survey sounds are examples of periodic signals that are captured by the periodicity metric. Details of metric calculation can be found in Wilford (2021) and Table II.

To compare the SSC metrics, median and central 95th percentage (C95) 1-min metric values were reported in five frequency bands as described in Wilford (2021; broadband, 10 Hz–32 kHz; low, 10–100 Hz; mid, 100 Hz–1 kHz; high, 1–10 kHz; ultra-high, 10–32 kHz). The metric medians and C95s were reported in the tabular structure of the SSC to make soundscape comparisons quick and easy. To aid in visual interpretation, a color-coding scheme was adopted whereby the metric range across all sites of individual metric medians and C95s were divided into quarters, and each cell of the SSC table was colored based on quartile (Fig. 2). The current color-coding range values are specific to the soundscapes analyzed in this study and may not be optimal in all cases. Visual review of spectrograms was used to verify signals at the extreme range of metric variability.

SSC metric time series were compared between sites with the Dunn method for joint ranking for each frequency band (Dunn, 1964, Appendix). Statistical operations were performed using JMP Pro^TM 14.0.0 (Cary, NC). Connections between SSC metric groupings and habitat type of the different soundscapes were explored further using k-means clustering, which was also carried out in JMP Pro^TM 14.0.0. The 1-min broadband metric values were grouped in daily bins and the numbers of clusters analyzed (k) were related to the number of sites (n) analyzed (k = n ±1). For clustering with all the sites, 4–6 clusters were analyzed; for clustering excluding GBR, 3–5 clusters were analyzed. GBR was removed for a portion of the clustering operation to see if logical groups were formed from the metrics corresponding only to the deepwater sites. The sample size totals in each of the clustering analysis table columns in Sec. III are the number of principle component values at each site over the 30 day analysis period. The column totals are not the same for each of the five sites because the different recorder duty cycles (Table I) impacted the overall site sample size.

Temporal patterns are also important factors to consider in soundscape assessments but are not captured in the SSC analysis. To explore daily temporal patterns among the five sites, broadband SSC metrics were grouped in hourly bins for the duration of the recordings and presented in polar plots. Broadband metrics were considered as a cursory investigation into temporal trends. Low, mid, high, and ultra-high bands could be considered for this analysis in the future, but it was not within the scope of the project to extend this particular analysis beyond broadband. Periodicity metrics showed no visible patterns and were not included in the polar plots.

The SSC outputs (Fig. 2) provided a wealth of information about the soundscapes and produced quantitative results that highlight salient similarities and differences, which informed subsequent analysis. Significance between each site pairing is included in the  Appendix.

GBR amplitude metrics were consistently larger than OCS site metrics in frequencies over 100 Hz, which suggests that the soundscape of the shallow, tropical environment was driven by acoustic activity in the mid, high, and ultra-high bands. All OCS site broadband amplitude metric medians were within 3 dB (across site), but some nuanced differences in the amplitude metrics in other frequencies suggested important soundscape differences among the deep OCS sites. At SAV and RH, the SPL medians were similar in the low, mid, and high bands; the ultra-high band had smaller median SPL levels. The sandy bottom BLE and WIL locations also reported a lower SPL in the ultra-high band compared to the cold-water coral sites of SAV and RH. The SPL medians at BLE and WIL were only similar in the mid and high bands. The metric medians were highest in the low band at BLE and WIL, suggesting more low frequency acoustic signals at these sites compared to the other OCS sites at SAV and RH. The soundscape at WIL produced the largest high and ultra-high band PK medians of the OCS sites, which otherwise had comparable PK medians in frequencies over 1 kHz. The amplitude metric variability was lowest for GBR across all frequency bands, whereas the RH and SAV metrics were generally the most variable. BLE and WIL SSCs had substantial variability in all frequencies but at a lower level relative to the RH and SAV sites.

The largest kurtosis medians and C95s revealed the GBR soundscape as the most impulsive. The OCS sites all reported identical kurtosis medians of three, therefore, the soundscape impulsiveness was further assessed by examining the C95s. RH was the second most impulsive site after GBR and had the largest C95 values among the OCS sites. The impulsive signals at RH were detected in all frequencies and peaked in the ultra-high band. Two of the non-coral OCS sites reported large kurtosis C95 values in specific frequency bands: the BLE soundscape generated a large kurtosis C95 (= 1042) in the ultra-high band, and the WIL soundscape generated substantial kurtosis C95s in the mid (C95 = 134) and ultra-high bands (C95 = 122). This suggested that the non-coral sites BLE and WIL were being influenced by impulsive acoustic signals. SAV was the least impulsive soundscape, suggested by kurtosis C95 values greater than 3 but less than 40 in the mid, high, and ultra-high bands. Much of the impulsive signals suggested by kurtosis values appeared to be biological in origin, and new research confirmed the presence of many acoustically active echolocating cetaceans at these sites at different times during a given year (Kowarski , 2023). Analysis of spectrograms from recordings that generated kurtosis values above k = 3 revealed an abundance of biological signals at BLE, WIL, SAV, and RH. At BLE, echolocation of foraging cetaceans and delphinid whistles were observed in frequencies over 10 kHz; similar signals were observed at WIL but less frequently; at SAV, these signals were mostly faint and infrequent; at RH, an abundance and diversity of acoustic biological activity in all frequencies was observed.

Similar to kurtosis, all the sites reported identical median periodicity metric values of zero for all frequency bands, thus, the C95s were used to assess periodicity of these sites. The acoustic activities at BLE, SAV, and GBR were not periodic; the periodicity C95 values for these sites were the lowest of all the sites. Considerable variability (C95 > 3) in the periodicity metric at WIL and RH in the lower frequency bands suggested the potential for these soundscapes to contain intermittent periodic signals. Visual review of spectrograms showed some periodic signals of biological origin. The SSCs also suggested that among the OCS sites, there was a disparity in the content of periodic signals. RH had the most periodic signals of the four OCS sites, followed by WIL, SAV, and BLE.

The median D-index values of uniformity were similar for all sites in the broadband category with the largest median value found at WIL in the low band (D = 0.031). In higher frequencies (>100 Hz), D-index medians of the OCS sites remained low, whereas they dominated at GBR. This suggested that the minute-to-minute changes in frequency content of the acoustic activity was greatest at GBR in frequencies over 100 Hz while in the low band, these changes were greatest at WIL. Frequent but unidentified low frequency signals were observed in spectrograms of WIL acoustic data, which are most likely responsible for the increased low band D-index median at that site. The relatively small C95 of the D-index at GBR in the high and ultra-high bands suggested that this site was acoustically consistent in these frequencies. RH experienced the largest ranges in the D-index, which suggested that at RH, there were more transient events that shifted the D-index to values higher than those at the other sites, but not enough to result in larger medians. The D-index at RH was driven mostly by bouts of what appeared to be echolocation of foraging cetaceans. Also, influential to RH, D-index values were low frequency (<100 Hz) flow noise. Biological signals were consistently observed at the OCS sites (Kowarski , 2023), and subtle differences in the D-index medians and ranges among the OCS sites suggested nuanced biological and anthropogenic acoustic activity in respective soundscapes.

A clustering analysis was performed to further examine overall soundscape similarities in support of the objective of comparing and contrasting soundscapes from different depths (shallow GBR vs deep OCS sites) and habitats (sandy vs coral). Amplitude clustering grouped GBR apart from the OCS sites for all numbers of clusters (Fig. 3). The GBR datapoints in cluster 4 for SPL [Fig. 3(A)] showed two relatively separate subclusters. On additional review of the GBR data points, it was determined that the two GBR subclusters relate to a separate subcluster for day and night hours. Two GBR subclusters were also present in the GBR impulsiveness [Fig. 4(A)]. Because the GBR day and night subclusters were more similar to each other than to any of the deepwater OCS sites, additional analysis was not pursued because day vs night comparisons were not in alignment with the study objectives.

Combined with the SSC output matrix results, the GBR clustering results suggested that the shallow coral reef had a soundscape that was significantly louder than the OCS sites [Fig. 3(A) for SPL]. Distinction among the OCS sites in terms of soundscape amplitude was also observed, where the coral habitat of RH grouped apart from the other three OCS sites for SPL [Fig. 3(A)]. Separation of RH and SAV PK metrics into multiple, isolated clusters was observed in clustering, including and excluding GBR [Fig. 3(B)], which suggested the amplitudes of RH and SAV cold-water coral habitat soundscapes were different from the other sandy bottom OCS soundscapes.

Like amplitude, impulsiveness drew a sharp contrast between GBR and the OCS sites (Fig. 4). Unlike amplitude metric clusters, no clear differences among the OCS sites in terms of impulsiveness were observed when GBR was included [Fig. 4(A)] or excluded [Fig. 4(B)] from the clustering. OCS site impulsiveness metrics clustered into one group for all numbers of clusters analyzed, and separate groups were formed by small numbers of outliers only. This result is similar to the SSC metric analysis, where impulsiveness metric medians for all sites and frequencies (besides GBR) were three, and further distinction needed to be made using variability of the metric.

Periodicity results grouped a sandy bottom OCS site (WIL) and coral bottom OCS site (RH) together and showed trends similar to SSC outputs by revealing a tendency for RH and WIL to separate, including [Fig. 5(A) cluster 3] and excluding GBR [Fig. 5(B) cluster 1]. The remaining OCS sites grouped predominantly in one main cluster [cluster 2 in Figs. 5(A) and 5(B)].

Similar to amplitude metrics, the uniformity metrics of the cold-water reef soundscapes RH and SAV showed a weak tendency to separate from the other sites [Fig. 6(A) clusters 1 and 2], which mostly formed one main cluster including GBR [Fig. 6(A) cluster 4].

Uniformity metrics showed some distinction among the OCS sites, but did not distinguish between the OCS sites and GBR. The only noticeable difference in uniformity clustering results for GBR were that for all numbers of clusters, the GBR populated only one cluster, whereas the OCS sites tended to be spread across multiple clusters.

Temporal trends occurring over daily time scales are relatively well-documented in established soundscape literature for shallow, warm-water coral reefs (Radford , 2008; Staaterman , 2014; Bertucci , 2015) but not captured in the SSC analyses. Polar plots of broadband soundscape metrics suggested that the tropical coral reef habitat of GBR featured classic dusk peaks and nighttime increases in sound amplitude compared to daytime hours (Radford , 2008; Bertucci , 2015; Fig. 7). Soundscape impulsiveness also appeared to increase in the nighttime hours at GBR, which links impulsiveness and amplitude in this soundscape.

The OCS sites, in contrast, did not show any clear and/or consistent temporal patterns across sites like the shallow GBR amplitude metrics. However, the hourly BLE metrics did show crepuscular peaks in PK amplitude, kurtosis, and uniformity, and RH had an amplitude peak during nighttime hours. These OCS site sunset and nighttime peaks in amplitude at RH and BLE could be related to a temporal soundscape trend related to diel migration, tides, or nocturnal sound production, which would link it to the shallow, tropical GBR soundscape and should be further explored. Other finer diel cycle features (e.g., afternoon kurtosis peak at SAV and late morning peak in WIL D-index) are evident within sites, which could also provide valuable information at specific sites or within specific OCS habitats.

The goal of this work was to apply the SSC metrics and explore differences in SSC output matrices relative to known differences in depth and presence of coral across the study sites. Differences in soundscapes have been connected to differences in biodiversity, habitat type, ecosystem health, season, lunar phase, and time of day (Radford , 2008; Radford , 2010; Radford , 2014; Kaplan , 2015; Bertucci , 2015). The SSC output and clustering of SSC metrics showed salient differences between the shallow coral environment of GBR and the deeper OCS sites. Differences among the OCS SSC outputs suggest differences in soundscapes among the deep ocean habitats. The results from the comparison of the five sites using the SSC structure directed the subsequent clustering and temporal analyses. SSC results for the OCS sites also provided a quantitative soundscape assessment of the deep ocean sites, which is the first of its kind and establishes a baseline for these deep-sea acoustic environments.

The tropical, shallow GBR soundscape generated a SSC that was remarkably different from the OCS sites in terms of all SSC properties. The large median amplitude, impulsiveness, and uniformity metric values in the mid, high and ultra-high bands suggest that GBR is dominated by acoustic activity in the higher frequencies (>100 Hz). The dominant acoustic energy in higher frequencies at GBR is likely tied, in part, to healthy function of its biodiverse environment. The reduction in all metrics at lower frequencies was likely due to the shallow water attenuation for frequencies less than ∼50 Hz. GBR amplitudes were much higher in frequency bands associated with signals from snapping shrimp, sea urchins, and vocalizing reef fish (Radford , 2014; McWilliam , 2018). In comparison, SSC outputs corresponding to SAV, WIL, and BLE suggested that dominant acoustic activity was in the lower frequencies. Proximity of the OCS sites to busy shipping lanes and a wealth of biological activity at GBR are other factors that most likely influence the differences noted in amplitude properties. RH was intermediate in its characteristics between GBR and the other OCS sites. The dominant acoustic energy in the upper frequency bands at GBR is also likely caused by attenuation of low frequency sound in shallow water. The disparity in propagation conditions would explain why the deeper OCS sites experienced larger amplitude medians in the low band relative to the shallow GBR site (Urick, 1986; Forrest , 1993; Hermannsen , 2019; Heaney , 2021).

Large impulsiveness ranges in all frequency bands and large amplitude ranges in frequencies over 1 kHz suggest that RH is similar to GBR in terms of amplitude and impulsiveness, but large uniformity and periodicity metric ranges and large amplitude metric medians in frequencies under 1 kHz suggested that RH is also similar to the OCS soundscapes. Nuanced differences in amplitude metrics among the deep ocean sites could represent a connection between the deep and shallow coral environments. SAV and RH have the largest broadband amplitude metric ranges, which appear mostly influenced by acoustic activity in the low, mid, and high bands. Healthier reefs in shallow water were found to be significantly louder than degraded environments (Piercy , 2014), and if the increased amplitudes of SAV and RH compared to the sandy bottom habitats of BLE and WIL are connected to the content of coral at these sites, it represents a significant finding that can be explored in future work.

Soundscape impulsiveness drew stark contrasts between the deep and shallow ocean soundscapes, which were present in the SSC results as well as in the clustering. GBR had high medians and variability, whereas the deep ocean sites all had comparatively lower medians and considerable variability among sites and frequency bands. The large amount of variability in impulsiveness metric values at RH sets it apart from the other OCS sites and does not necessarily suggest it is similar to GBR but rather that it is intermediate in its characteristics. Due to the connection between increased acoustic diversity and coral reef health in shallow water (Piercy , 2014), the sound sources responsible for the kurtosis variability at the OCS sites should be explored for potential proxies of health or biodiversity in cold, deep-sea coral habitats. Impulsiveness observed in the OCS sites was associated with the sounds of foraging and echolocation by cetaceans, whereas, at GBR, the impulsiveness was attributed mostly to snapping shrimp (Cato and Bell, 1992; Readhead, 1997). The OCS sites exhibit unique impulsive characteristics and exploring the nature of the signals as they relate to respective environments could illuminate important soundscape and habitat relationships in deep ocean environments. Soundscapes influenced heavily by biological signals may indicate the existence of unique ecosystems that should be considered in the management of offshore energy development or other anthropogenic activity in the region.

The disparity in amplitude and impulsiveness metrics between the deep and shallow soundscapes was most obvious, but influence by periodic signals also suggests a difference in the soundscapes. The periodicity values of the OCS sites provide a valuable comparison between the soundscapes, and two of the OCS sites (RH and WIL) produce SSCs that indicate substantial influence by periodic signals, which could be a distinguishing soundscape feature. To understand the nature of the increased periodicity of the OCS sites, the sound sources responsible for the periodicity values would have to be determined, which was beyond the scope of this study but in alignment with recommendations to inspire future research and validation of source mechanisms presented here.

Slight differences in D-index ranges suggest subtle variances among the OCS soundscapes in terms of acoustic uniformity. The large low band median at WIL suggests that some type of influential acoustic activity exists there but not as much as at the other OCS sites. Without knowing more about the sound source mechanisms, it is difficult to understand what this disparity in uniformity ranges represents; unidentified low band signals and noise appear to influence the acoustic uniformity of WIL. If D-index values are directly related to biological acoustic activity, then increased D-index ranges at RH and SAV might suggest that these sites are more biologically active than WIL and/or BLE. If SAV and RH feature more biological signals, it would represent an interesting and important connection between soundscape, coral cover, structural complexity, and biological activity, a connection already well-documented in shallow water coral reef environments (Risk, 1972; Luckhurst and Luckhurst, 1978; Grigg, 1994; Graham and Nash, 2013; Graham , 2014). The rugged bottom topography of SAV and RH could be environments more attractive to deep-sea marine life just like shallow water environments with increased structural complexity provide more robust and attractive habitats for marine life. Clustering of the D-index drew little distinction between GBR and the OCS sites but suggested that the soundscapes at SAV and RH were different than the other two sandy bottom OCS sites, which aligned with the other SSC results. The strong tendency for RH and SAV to cluster uniformity metrics into separate groups while the sandy bottom OCS sites remained mostly in one cluster strengthens the case that the SSC metrics are drawing distinctions between the OCS sites in agreement with differences in bottom type.

Daily temporal soundscape trends were identified in the shallow GBR soundscape as well as at some of the OCS sites. Given the depth of the OCS sites, it is interesting that there were any crepuscular or nocturnal observations in acoustic activity in the absence or low level of light compared to the shallow GBR reef. Trends in acoustic activity relating to lunar, diel, and seasonal cycles/trends are well-documented in shallow reef environments (Radford , 2008; Staaterman , 2013; Staaterman , 2014; Bertucci , 2015) but to our knowledge, have not been thoroughly explored in deep ocean environments. It is reasonable to attribute diel patterns in the OCS sites to surface and water column biological activity, but longer annual or multiyear soundscape time series of amplitude and impulsiveness could illuminate whether seasonal or lunar cycles influence biological activity in deep ocean habitats.

In summary, the most obvious differences in respective SSC outputs group all the deep ocean soundscapes into one group and the shallow, tropical reef soundscape into another group. The SSC features that most clearly distinguish these two groups are the SSC metric ranges and amplitude and impulsiveness median values. The consistency of the biological signals of GBR produced SSC metric ranges that are small compared to the OCS sites, where kurtosis is the main exception. Dominant acoustic signals at GBR appear to be relating to nightly fish chorus and daily snapping shrimp activity (Readhead, 1997; McWilliam , 2018). Although fish choruses have been detected in deep-sea environments (Cato, 1978; Kelly , 1985; Bolgan and Parmentier, 2020), they remain poorly understood, and it is not clear how/if these types of signals are influencing the OCS soundscapes featured in this research. Many of these influential signals at the OCS sites do appear to be biological in nature, but specific signals observed in this study were echolocation and communication by foraging cetaceans, not chorusing fish.

SSC results provide valuable soundscape information and highlight areas that subsequent analysis should explore to better understand the soundscape dynamics of the five sites. Further investigation into daily and/or seasonal trends in soundscape of the OCS sites would be beneficial as initial results pertaining to temporal trends in deep-sea soundscapes from this study indicate that the daily temporal patterns of deep ocean reefs are not the same as those for shallow reefs. More in-depth assessments of the sound sources driving the SSC metric differences would illuminate potentially unique soundscape features among the OCS sites and help us understand any connections between deep and shallow coral reef environments. The SSC metrics appear to capture salient and obvious differences in soundscapes of deep and shallow marine environments and suggest that similarities between deep and shallow coral reef habitats do exist. Soundscape and sound source analysis combined with traditional methods of ecosystem and biodiversity assessment will aid in developing a better understanding of these important but poorly understood deep-sea habitats.

This research would not have been possible without collaboration from JASCO Applied Sciences for the GBR data (in collaboration with James Cook University). This work was supported by the NOAA Grant No. NA15NOS4000200 provided to the Center for Coastal and Ocean Mapping at the University of New Hampshire; U.S. Department of the Interior, Bureau of Ocean Energy Management, Environmental Studies Program, Washington, DC, Contract No. M17PC00009 to DEEP Sea Exploration to Advance Research on Coral/Canyon/Cold seep Habitats (DEEPSEARCH); and U.S. Department of the Interior, Bureau of Ocean Energy Management, Environmental Studies Program, Washington, DC, Contract No. M16PC00003, in partnership with Office of Naval Research (ONR) and NOAA, for Atlantic Deepwater Ecosystem Observatory Network (ADEON). Funding for ADEON ship time was provided under separate contracts by ONR, Code 32.

The authors have no conflicts of interest to disclose.

The ADEON data that support the findings of this study are openly available from the National Centers for Environmental Information in https://www.ncei.noaa.gov/products/passive-acoustic-data. The GBR data that support the findings of this study are available from the corresponding author upon reasonable request.

SSC metric time series over the 30 day analysis period were compared between sites with the nonparametric Dunn method for joint ranking for each frequency band (Dunn, 1964, Appendix; see Figs. 8–12). All statistical operations were performed using JMP Pro^TM 14.0.0.