Limited vocal compensation for elevated ambient noise in bearded seals: implications for an industrializing Arctic Ocean

Abstract

Vocalizing animals have several strategies to compensate for elevated ambient noise. These behaviours evolved under historical conditions, but compensation limits are quickly being reached in the Anthropocene. Acoustic communication is essential to male bearded seals that vocalize for courtship and defending territories. As Arctic sea ice declines, industrial activities and associated anthropogenic noise are likely to increase. Documenting how seals respond to noise and identifying naturally occurring behavioural thresholds would indicate either their resilience or vulnerability to changing soundscapes. We investigated whether male bearded seals modified call amplitudes in response to changing ambient noise levels. Vocalizing seals increased their call amplitudes until ambient noise levels reached an observable threshold, above which call source levels stopped increasing. The presence of a threshold indicates limited noise compensation for seals, which still renders them vulnerable to acoustic masking of vocal signals. This behavioural threshold and response to noise is critical for developing management plans for an industrializing Arctic.

1. Introduction

The Arctic is one of the most rapidly changing ecosystems on Earth [1,2]. Climate-driven decreases in the Arctic icecap result in the loss of summer sea ice, threatening northern species that rely on sea ice habitat. In the new ‘low-ice' Alaskan Arctic, anthropogenic noise is expected to increase with expanded commercial shipping, fishing, energy exploration and extraction, and tourism [3,4]. Anthropogenic noise poses an established threat to marine mammal species and other wildlife that rely on sound for vital life functions, however, most of the research investigating behavioural responses to noise in the marine environment has been conducted in cetaceans [5–7]. Despite the fact that most seal species (family Phocidae) mate under water and rely on sound to mediate courtship [8], how seals respond to elevated noise conditions is largely unknown.

Bearded seals (Erignathus barbatus) mate under water and rely on sea ice for resting, pupping and nursing during spring and early summer. Males produce species-specific underwater vocalizations to attract mates, defend territories, and facilitate social interactions [8,9]. As sea ice declines, it is likely that a greater number of males will compete for smaller amounts of viable breeding habitat [10], resulting in increased acoustic competition. The vocalizations of aquatically breeding pinnipeds, like bearded seals, can contribute significantly to coastal ambient noise levels [11,12], and may act as a source of biological noise that prevents the exchange of acoustic information by reducing communication space (i.e. ‘acoustic masking'). For bearded seals to detect a conspecific vocalization, that signal must reach a critical signal to noise ratio above ambient sound levels at a particular frequency (i.e. a ‘critical ratio') to avoid being masked [13].

Excessive ambient noise can have deleterious impacts on animals that depend on sound, including hindering foraging [14], preventing social interactions [15], reducing reproductive success [16] and increasing individual stress levels [17]. There are several strategies that animals employ to compensate for elevated ambient noise levels: individuals can alter frequency, duration, loudness (the ‘Lombard effect') or call rates in order to maintain acoustic contact (reviewed in [6]). Despite both predicted and observed changes in the Arctic soundscape [4,18], whether bearded seals employ anti-masking strategies to compensate for ambient noise is poorly understood. Because bearded seals rely on acoustic communication to facilitate mating, there are concerns of possible fitness consequences of acoustic masking [16,19], although this has not yet been empirically demonstrated in most taxa. The ability for acoustically communicating species, including bearded seals, to compensate for large-scale changes in noise may ultimately impact their vulnerability to increasing human activities.

There is some evidence of vocal plasticity in phocids, suggesting a potential to compensate for changing acoustic conditions. Some seals have the ability to adjust their acoustic behaviours based on context. Harp seals (Pagophilus groenlandicus) employ a wide range of source levels [20], and harbour seals (Phoca vitulina) can mimic human voices indicating they can adjust acoustic signals based on external stimuli [21]. However, evidence suggests that during the breeding season, advertising males of vocally plastic species already vocalize at upper physiological performance limits, and therefore lack adaptive capability. Weddell seals exhibit vocal learning, but do not alter the duration of breeding advertisement signals to avoid being acoustically masked by calling conspecifics [22–24]. Similarly, male harbour seals failed to adequately increase their call amplitude in response to vessel noise [25], presumably because they were already calling at their maximum physiological limit. Beyond these studies, little has been described on the topic of seal calling behaviours and ambient noise, leaving little room for meaningful management actions pertaining to behavioural responses to changing soundscapes.

While the risks associated with elevated noise are of global concern [26], bearded seals are of particular conservation relevance given that Alaskan Native communities rely heavily on this species as a subsistence resource [27,28]. Bearded seals are culturally, nutritionally, and ecologically important to indigenous communities in the Arctic and subArctic [27,29]. In the United States, bearded seals are considered a single stock that is co-managed by the Alaska Native Ice Seal Committee (ISC) and the NOAA National Marine Fisheries Service. These groups collectively make bearded seal management decisions in the Alaskan Arctic and have identified the impacts of noise and changing habitat conditions as primary conservation concerns [27].

Ice seals are listed as threatened under the US Endangered Species Act (ESA) and are protected by the US Marine Mammal Protection Act (MMPA); the MMPA includes explicit provisions for limiting noise exposure to marine mammals. NOAA Fisheries has published guidelines for assessing marine mammal harassment as it pertains to ambient noise levels. Despite an acknowledgement that elevated ambient noise conditions can have potentially deleterious impacts on marine mammal behaviour and should be integrated into acoustic management plans, recommended limitations on noise levels that elicit behavioural responses or may result in acoustic masking have not been incorporated in management plans. This is in part due to the absence of empirical data. How bearded seals respond to noise is an important component for decision makers and conservation scientists investigating how shifting ecological processes in the Arctic impact bearded seals and subsistence users. This is important not only as it pertains to potential acoustic masking of conspecific signals, but also as it pertains to masking of other ecologically important acoustic signals (e.g. predators, prey, environment)

In order to understand the risks associated with changing Arctic noise conditions, it is essential to identify biologically relevant noise thresholds that have the potential to influence critical behaviours [30,31]. In the context of this study, we define a biologically relevant noise threshold as one that has the demonstrated potential to impair acoustically and contextually mediated biological activities (e.g. reproduction, predator detection). We investigated whether male bearded seals modified their call amplitudes in response to changes in ambient noise levels associated with conspecific callers during the breeding season. We sought to identify a noise threshold above which callers no longer effectively adjusted call amplitude to compensate for conspecific noise. Evidence that bearded seals increase call amplitude in response to elevated noise would indicate that this species exhibits some resilience to predicted changes in Arctic soundscapes- although, vocal compensation does not minimize the risk of acoustic masking of other biologically relevant acoustic signals (i.e. predators or prey). A lack of adaptive capacity beyond a certain noise level, however, may indicate that elevated noise in the Arctic poses a threat to seals and the communities that rely on them.

2. Methods

Acoustic data were collected from arrays of Marine Autonomous Recording Units (MARUs) as part of the census for Bering-Chukchi-Beaufort bowhead whales during spring migration [32,33]. Units were deployed along the ice edge adjacent to Utqiaġvik, Alaska from April–July 2010 (5 elements; figure 1) and May–July 2011 (6 elements; figure 1). In 2010, MARUs were separated by approximately 4 km; in 2011, MARUs were separated by approximately 7 km. MARUs recorded continuously at a 2000 Hz sampling rate, with an effective bandwidth of 10–900 Hz after accounting for high and low-pass filters, with a flat sensitivity of −145.5 dB re 1 V µPa⁻¹ over the 10–900 Hz frequency band. Recordings were time aligned and merged into single multi-channel files for analysis.

Figure 1.

Recording locations for 2010 (triangles) and 2011 (squares) in the Beaufort Sea.

Twenty days were randomly sampled from each year; for each day, the 15:00–16:00 h Alaska Daylight Time (GMT −8) was manually annotated. Bearded seal calling rates peak at the 04:00 h in this region [34]; the 15:00 h was selected to encompass a broad range of ambient noise values, and ensure periods of non-overlapping calls required for localization. Bearded seal calls were logged in time and frequency in Raven Pro 2.0 [35] (DFT = 256 points, Hamming window, frequency resolution = 15.6 Hz, time resolution = 0.128 s). Calls were batch localized using the near-beamforming method integrated into Raven 2.0 [36]. This method generates an estimated range from receiver to caller and a latitude and longitude (with uncertainty in m) for each caller [11,12,37]. Locations were overlaid on a map of the region that included maximum ice extent to assess biological plausibility (i.e. to ensure no callers were erroneously mapped on land). Caller locations were included in analysis if the associated error in latitude, longitude, or range was less than 50 m and the signal-to-noise ratio (SNR) was at least 10 dB at all hydrophones. We calculated the distance between each set of coordinates and the nearest hydrophone using a custom written haversine function in R 3.5.1 [38]. Detection ranges for bearded seal calls in the Alaskan–Canadian Arctic region are up to 25 km [39]; data exploration revealed that for calls produced within 5 km of the nearest hydrophone there was no relationship between distance and source levels; only calls localized to within 5 km of a hydrophone were included in analysis.

Estimated source levels (SL) in dB_RMS re: 1 µPa @ 1 m in the visible bandwidth of the call (minimum to maximum frequency range: 33 Hz–900 Hz; median minimum frequency: 274 Hz) were estimated for each call using the Passive Sonar Equation (equation (2.1)). Here, RL is the calibrated RMS received level extracted in Raven Pro 2.0 for the entirety of each call (i.e. trills and plumes), and TL is the estimated transmission loss calculated as 15 × log(distance in m) [11,12,40]. This propagation loss model is simplistic—it accounts for geometric spreading loss only—but more complex transmission loss estimates made in this region in 2016 confirm that characteristics of multipath propagation made off the coast of Utquiagvik are consistent with geometric spreading loss used here [41]. Because the maximum frequency of bearded seal calls exceeds the upper frequency range of recordings, SL estimates for seals are band-limited (lowest frequency to highest identifiable frequency) and not inclusive of the entire vocal range.

$$\mathrm{SL}=\mathrm{RL}+\mathrm{TL}.$$ 2.1

Inband RMS ambient noise levels (100–900 Hz; dB_RMS) were extracted for the 1 s period directly preceding each call in order to measure signal excess above ambient (SL − ambient noise) [11,12,25,42] and in order to compare SL to ambient noise conditions. In hearing studies bearded seals are capable of interpreting and responding to signals as short as 0.5 s [13]; 1 s was selected to represent an instantaneous snapshot of noise experienced by the seals prior to vocalizing. This frequency range was selected to encompass the range of highest hearing sensitivity within recording limits [13], while holding noise levels constant across the study. Only a very small subset of calls (n = 9, 2.7% of total) had lower frequencies below 100 Hz; exploratory analyses excluding these calls did not alter the results and they were included in the interest of maximizing the number of individuals potentially represented in the study.

Linear models were built to test the relationship between ambient noise and source levels (hereafter ‘full model') and signal excess (excess model). Visual inspection of the SL distribution, ambient noise levels and signal excess indicated that data and model residuals were normally distributed and homoscedastic. A quadratic equation was built into the models and the Akaike information criterion (AIC) was used for model selection.

It has been previously determined that annually there were approximately 22–40 individual males advertising near Utqiaġvik [9]. To account for potential pseudoreplication, data were aggregated into 5 dB bins (e.g. 80–85 dB_RMS) and median SLs were calculated for each bin. Linear models were built to test for the effect of ambient noise (dB_RMS) on median SL (binned model); a quadratic equation was built into the model and AIC was used for model selection.

3. Results

Out of the total of 777 calls, 327 calls fit the inclusion criteria (2010, n = 134; 2011, n = 193). Ambient noise values ranged from 83 dB_RMS to 111 dB_RMS. AIC selection indicated that the full model with a quadratic term was preferred to models without (ΔAIC = 15). Both ambient noise and the quadratic of ambient noise significantly related to SL (F = 48, p_2,324 < 0.000 001). Ambient noise was positively associated with SL (p < 0.00001, ß₁ = 6.4, 95% CI = 3.6–9.3); the quadratic of ambient noise was slightly negatively associated with SL (p < 0.00001, ß₂ = − 0.03, 95% CI = −0.046–0.16), but was functionally flat given the technical resolution of the system. This indicates that beyond a certain ambient noise level, SL stopped increasing and levelled off (figure 2). Inspection of the SL dataset indicates that the threshold occurs between 100 dB–105 dB_RMS. Beyond this, SL stopped increasing and the slope of the line approached zero (figure 2).

Figure 2.

Relationship between band limited bearded seal source levels and ambient noise. Note that source levels are indicative of total measurable energy over the visibly identified bandwidth of the call, while ambient noise is standardized in the 100–900 Hz range. Model results are indicated by solid lines, and 95% confidence intervals indicated by shaded grey. Putative noise threshold of approximately 103 dB indicated by vertical dashed line.

AIC selection indicated both ambient noise and the quadratic of ambient noise should be included in the excess model (ΔAIC = 15). Both ambient noise and the quadratic of ambient noise were significantly related to signal excess (F = 55, p_2,324 < 0.000001). The relationship between ambient noise and signal excess was curvilinear, but generally negative. Ambient noise was positively associated with signal excess (p = 0.0002, ß₁ = 5.4, 95% CI = 2.6–8.3); the quadratic of ambient noise was negatively associated with signal excess (p < 0.0001, ß₂ = − 0.03, 95% CI = −0.05–−0.02), indicating that beyond a certain threshold signal excess began to precipitously decrease with ambient noise (figure 3).

Figure 3.

Relationship between bearded seal signal excess (calculated as the band limited source level minus ambient noise) and ambient noise. Empirical acoustic measurements are indicated by black circles, statistical model output and 95% confidence interval indicated by solid lines and shaded ribbons.

AIC model selection indicated that ambient noise and the quadratic of ambient noise should remain in the binned model. Models indicated that ambient noise and the quadratic of ambient noise were significantly related to median SL (F = 68, p_2,3 = 0.003). Ambient noise was positively associated with median SL (p = 0.03, ß₁ = 4.4, 95% CI = 0.6–8.2); the quadratic of ambient noise was slightly negatively associated with median SL (p = 0.047, ß₂ = − 0.02, 95% CI = −0.04 to −0.0005), indicating that beyond a certain threshold source level stopped increasing with ambient noise and started to level off (figure 4).

Figure 4.

Median band limited source level within each 5 min noise bin plotted against ambient noise in the 100–900 Hz range. Note that source levels are calculated by measuring the energy context of the call, and may not exactly encompass the same bandwidth as ambient noise. Data points indicated by dots, model results and 95% confidence interval indicated by solid line and shaded ribbon.

4. Discussion

This study is the first to identify evidence of a Lombard effect in any pinniped species; however, the increase in SL by bearded seals was not sufficient to prevent a reduction in signal excess and indicates that there is an apparent noise level beyond which males fail to increase call amplitude. That bearded seal signal excess declined as ambient noise levels increased suggests that advertising seals did not sufficiently adjust calling amplitude to maintain communication space beyond a given ambient noise level. It is possible that advertising males alternatively temporally stratify their vocalizations or adjust calling frequency (i.e. pitch) as noise from other seals increases in order to avoid acoustic overlap. Adjusting the frequency or timing of calls in addition to increasing call amplitude may provide a release from acoustic masking that comes at a lower energetic cost than compensating by calling louder [43]; moreover, employing several anti-masking strategies may increase the overall likelihood of being detected by a relevant receiver than a single strategy alone [44]. Masking release hypotheses were tested in Weddell seals, who similarly rely on ornate courtship vocalizations; however, Weddell seals were not found to adjust the timing or duration of their vocalizations to avoid acoustic overlap with conspecifics [24]. Future work investigating consequences of masking in bearded seals and alternative strategies for masking release would be valuable for assessing vocal resilience of this species.

This study suggests that there is a behavioural threshold at approximately 100–105 dB, beyond which bearded seal call amplitude stops increasing. This may indicate that bearded seals lack the physiological ability to increase source levels above a certain point. Male bearded seals use vocalizations to establish acoustic territories and attract mates. They may be advertising at or near a biological maximum amplitude throughout the breeding season and therefore have minimal ability to increase call amplitude, as seen in advertising harbour seals that do not employ a Lombard effect to compensate for noise [25].

As sea ice retreats earlier in the summer, ice seal breeding habitat will probably decrease [45,46]. As a result, advertising males may be forced to compete for smaller areas of viable breeding territories. Given that calling loudly and often is likely a critical reproductive strategy, it is reasonable to assume that underwater ambient sound levels in the call frequency band would vary as a function of caller density and acoustic masking by conspecifics is likely to occur. This study demonstrates that even though conspecifics are a natural noise source, bearded seals may not necessarily have the ability to adjust their calling amplitude to eliminate the risk of acoustic masking from conspecific or other noise sources.

It is important to include the caveat that this study only investigated responses to ambient noise below 900 Hz, and vocal adjustments above that range would not have been observable given the system recording constraints. We measured AL1(T) call types [47], which are long duration, narrowband, downsweeping trills with a terminal plume. The upper frequency of this call type (average max frequency = 4.32 kHz) occurs at the call's onset and only exceeds the recording limits of this system for a proportion of the total call duration (approx. 15–20% of total duration). This call type is not amplitude modulated, and there is no available evidence to suggest that bearded seals would adjust amplitude of frequencies above 900 Hz in a manner contrary to what was observed in this study. However, while the limited Lombard effect observed at lower frequencies was evident, the limitations of this system preclude our ability to speak to vocal responses above 900 Hz.

Species that use the Lombard effect to account for natural sound, often transfer this strategy to anthropogenic noise (e.g. humpback whales [11,12]; túngara frogs Physalaemus pustulosus [48]; oyster toadfish Opsanus tau [49]). The results of this study provide insight into how bearded seals respond to natural changes in the marine soundscape and may act as an important guideline for understanding when noise levels in a species' acoustic habitat are high enough to elicit a behavioural shift. Here, the conclusion is that bearded seal callers may not be able to prevent a reduction in signal excess, and therefore a reduction in communication range, resulting from elevated ambient noise. In order for a bearded seal to detect conspecific calls in the 100–900 Hz range the signal at the receiver needs to be at least 12 dB higher than background noise [13]. In this study, the predicted signal excess when ambient noise was 111 dB (maximum observed) was 17 dB lower than when ambient noise values were 83 dB (minimum observed). Depending on the proximity of a seal to a calling conspecific, this reduction in signal excess is enough to prevent conspecific detection. The risk of masking by conspecifics may be exacerbated by increasing anthropogenic noise in this region, which is louder than calling bearded seals and can persist for weeks to months [18]. Because advertisement behaviour is believed to be directly related to reproduction in bearded seals [8,50], the implication is that acoustic masking may have fitness implications in this species.

Moreover, bearded seals are just one of several Arctic seal species that rely on sound to facilitate reproduction (e.g. ribbon seals Histriophoca fasciata, ringed seals Pusa hispida), and probably to face similar influences by the expanding anthropogenic footprint in the Arctic. Acknowledging that bearded seals were among the first phocine species to diverge at approximately 11 Ma [51], Arctic seals nonetheless share reproductive strategies and similar evolutionary pathways and environmental pressures shaped in part by the presence Arctic sea ice dating back 13–14 Ma [52]. Our findings of vocal compensation thresholds and limited resilience to anthropogenic noise may be more broadly applicable to Arctic phocids contending with rapidly shifting ocean soundscapes.

Data accessibility

The dataset supporting this article has been uploaded as part of the electronic supplementary material.

Author contributions

M.E.H.F. was responsible for project conceptualization, data curation, formal analysis, investigation, methodology, visualization, supervision, writing of the original manuscript, and review and editing of the manuscript. M.S. participated in data curation and project investigation and reviewed and edited the manuscript. C.W.C. participated in project investigation, data collection, methodology, and reviewed and edited the manuscript. H.K. participated in conceptualization, project administration, supervision, and reviewed and edited the manuscript. A.N.R. participated in conceptualization, project administration, supervision, data visualization, resources, and reviewed and edited the manuscript.

Competing interests

We declare we have no competing interests.

Funding

Funding for instrumentation and data collection was provided by the North Slope Borough (Alaska) Department of Wildlife Management.