The acoustic ecology of sei whales (Balaenoptera borealis) is poorly understood due to limited direct observation of the behavioral context of sound production and individual behavior. Suction cup–attached acoustic recording tags were deployed on sei whales to unambiguously assign call types and explore the acoustic behavior of this endangered species. Twelve tag deployments resulted in ∼173 h of acoustic data and 1030 calls. Sound types included downsweeps and three previously undescribed call types. Knocks were short duration (<1 s), with an average peak frequency of 330 Hz. Pulse type 1 and pulse type 2 calls, typically produced in sequences, were short in duration (0.08 and 0.28 s) and low in average peak frequency (50 and 26 Hz), with relatively high received levels. Average call rates for all call types combined were three calls per hour, but increased during twilight. Sex differences in call type usage included a higher use of pulses by females and knocks by males. Calls were almost exclusively produced at depths <10 m, although whales rarely dove deeper in this study. These data provide a more comprehensive picture of the acoustic and behavioral ecology of sei whales than previously possible, which can inform future conservation efforts for this endangered species.

Describing the full acoustic repertoire of a species can be challenging, especially for those that are elusive or live in remote habitats. These challenges apply to many cetacean species (i.e., whales, dolphins, and porpoises), where individuals spend the majority of their lives under water. Short-term, suction cup–attached digital acoustic recording tags (Johnson and Tyack, 2003; Johnson , 2009) are an ideal way to describe and identify the acoustic repertoire of cetaceans without the need to maintain close proximity to the tagged animal. These tag types include a sensor suite that records the fine-scale movement behavior of the animal (e.g., orientation and depth) at a high resolution along with concurrent audio data. Attachment of these tags typically requires close approaches to the animals with a small vessel. This is particularly difficult for many of the rorqual baleen whales (Balaenopteridae), including sei whales (Balaenoptera borealis), which are inconspicuous, fast, and located predominately in deeper water (Horwood, 2009; Prieto , 2012).

The sei whale inhabits all of the world's oceans (Horwood, 2009; Prieto , 2012); however, their population ecology is poorly understood. In the North Atlantic, the International Whaling Commission (IWC) recognizes three stocks: Nova Scotian, Iceland–Denmark Strait, and Eastern North Atlantic stocks (Mitchell and Chapman, 1977; Prieto , 2012; Huijser , 2018; Hayes , 2022). Although not recognized by the IWC, a possible separate Labrador Sea stock has been shown to migrate to the Azores and possibly onward to northwestern Africa (Prieto , 2014), with acoustic detections off Greenland corresponding to the timing of this migration (Davis , 2020). The Nova Scotian stock is thought to occur along the east coast of the United States to at least Florida (Davis , 2020; Hayes , 2022), and the Eastern North Atlantic stock is proposed to migrate between northwestern Africa and the waters around Iceland and the Norwegian Sea (Ingebrigtsen, 1929). However, there is little evidence for genetic divergence among the potential stocks in the North Atlantic (Huijser , 2018). Few data are available on the movements and distribution of this species in the North Atlantic and no data are available on locations of breeding grounds and potential mixing on feeding grounds—further confounding attempts to define sei whale stocks in this ocean basin (Prieto , 2012). Even fewer data are available on other populations of sei whales. Although six management areas are defined by the IWC in the southern hemisphere, the population structure is poorly defined and there can be dynamic movement among these areas (Donovan, 1991). In the Pacific, separate stocks are defined in Hawai'i and the eastern North Pacific by the National Marine Fisheries Service (Caretta , 2022).

The ecology and acoustic behavior of sei whales is also poorly understood. Previous descriptions of the sei whale acoustic repertoire relied on associating recorded sounds with visual observations of whales in the area (Thompson , 1979; Knowlton , 1991; McDonald , 2005; Rankin and Barlow, 2007; Baumgartner , 2008; Romagosa , 2015). These include tonal as well as broadband sounds. One call in particular, the downsweep, has been attributed to sei whales based on its consistent detection concurrent with sei whale visual sightings and based on its distinctive acoustic parameters that differentiate it from low-frequency sounds from other species (Rankin and Barlow, 2007; Baumgartner , 2008; Español-Jiménez , 2019). Downsweeps have since been used to detect sei whales in passive acoustic monitoring (PAM) data without associated visual observations. This method provides a platform to remotely study the distribution of sei whales as well as to describe downsweeps in different regions (Newhall , 2012; Tremblay , 2019; Davis , 2020; Nieukirk , 2020; Romagosa , 2020). PAM has also been used to ascribe additional call types to sei whales based on their detection during bouts of downsweeps (Tremblay , 2019; Cerchio and Weir, 2022).

In the Southern Ocean, sei whale downsweeps, broadband sounds, and up–downsweeps have been detected in summer (McDonald , 2005). Downsweeps have also been recorded in both spring (Español-Jiménez , 2019) and autumn (Calderan , 2014; Buchan , 2022), with upsweeps and up–downsweeps occurring in autumn as well (Calderan , 2014). In the South Atlantic, a variety of tonal calls have been detected in both summer and autumn including downsweeps and upsweeps, with evidence of song occurring in autumn (Cerchio and Weir, 2022). No data are available in the Southern Hemisphere during winter.

In the Northern Hemisphere, only downsweeps have been described from the North Pacific, specifically from Hawaiian stocks of sei whales in autumn (Rankin and Barlow, 2007). The remaining data are from the North Atlantic, where much of the previous acoustic research on sei whales has occurred. Downsweeps have been detected year-round in the Labrador Sea and Nova Scotian populations (Baumgartner , 2008; Newhall , 2012; Romagosa , 2015; Tremblay , 2019; Davis , 2020; Romagosa , 2020), as well as in spring, summer, and autumn in the Iceland–Denmark Strait stock (Nieukirk , 2020). Two additional call types have been recorded from the Nova Scotian stock, including frequency-modulated sweeps (Knowlton , 1991) and pulses (Thompson , 1979) (see supplementary material for an overview of the current data available on the acoustic behavior of sei whales by geographic region and proposed or established stocks).1

Despite the previous research effort in the Western North Atlantic, a gap remains in our knowledge of the acoustic and behavioral ecology of sei whales. The reliance on passive acoustic data thus far has limited our ability to understand both the context of sound production and individual calling behavior, such as call rates and call depths. In the present study, acoustic tag data provided us with the opportunity to unambiguously assign call types to sei whales and capture previously unreported call types that were likely missed during passive recordings. Due to the sampling rate of the accelerometer data, we were able to assign calls as likely originating from the focal animal (Goldbogen , 2014) and correlate vocal activity with depth. In addition, by including data on the sex of the focal animal, we were also able to explore differences in vocal production between males and females.

Data were collected from sei whales in the southern Gulf of Maine, Massachusetts in the spring of 2022 (April 25, May 1, and May 5) using short-term, suction cup–attached multi-sensor digital recording tags. Two tag types were used: digital acoustic recording tags (DTAG) (Johnson and Tyack, 2003) and customized animal tracking solutions (CATS) tags (Goldbogen , 2017). Both tags included a sensor suite that recorded the fine-scale movement behavior of the animal (e.g., orientation and depth) at a high resolution, along with concurrent audio data. DTAG (version 3) and CATS tags were both equipped with tri-axial accelerometers, magnetometers, pressure and temperature sensors, and either a single or stereo hydrophone. CATS tags were also equipped with video cameras. Accelerometers were sampled at 250 Hz for DTAGs and 400 Hz for CATS tags; audio was sampled at 120 kHz for DTAGs, and either 24 or 48 kHz for CATS tags; and pressure was sampled at 50 Hz for both tag types. Tags were deployed either via pole (see Friedlaender , 2013) or drone (Wiley , 2023). No dedicated behavioral sequencing was conducted; however, observational data on the number of individuals and species present as well as the general behavioral state (e.g., travelling, skim feeding, or lunge feeding) were collected before, during, and immediately after tagging.

Acoustic recordings were decimated to 24 kHz for analysis. Spectrograms were generated using a Hann window with a fast Fourier transform (FFT) size of 2048 samples and a 90% overlap. Audio files were browsed visually and aurally in Raven Pro 1.6 (K. Lisa Yang Center for Conservation Bioacoustics, 2022a) by an experienced analyst. Calls were assigned to a priori call types based on previous literature or assigned to a new call type if no prior record existed. Call types were verified by a second analyst.

Only calls deemed to likely be from the focal animal were extracted to measure acoustic features of call types. The high sampling rate of the accelerometer data allowed us to designate calls as likely focal or non-focal (see Goldbogen , 2014). For every call identified in the acoustic record, spectrograms of the corresponding accelerometer data were plotted. Calls were defined as likely focal if they were visible on any of the three accelerometer axes, and non-focal if they were only visible in the audio waveform and/or spectrogram (see supplementary material for Figs. S1–S4).1

The use of accelerometer data to identify the calling animal can sometimes be inconclusive (Saddler , 2017; Stimpert , 2020). Whether a call is present on any of the axes can be affected by a variety of factors, including the sampling rate of the accelerometer, call type and frequency, species, and the presence of background noise from increased movement (Stimpert , 2020). Additionally, the magnitude of corresponding accelerometer signals is directly related to the received sound level of the call (Goldbogen , 2014). One call type (knocks) in the current dataset did not show up in the accelerometer data with the exception of four occurrences, possibly because of low received levels (RL) (see Results). Although it is possible that only these four calls were focal, their appearance amidst other likely focal knocks in a sequence renders this unlikely. Therefore, knocks were determined to likely be from the focal animal based on a calculation of relative signal-to-noise ratio (SNR) threshold.

The SNR was calculated by selecting a section of noise with the same frequency bounds and duration directly prior to or after the call (< 1 s). Inband power [decibels relative to full scale (dBFS)] for both signal and noise were measured in Raven from the spectrogram. Inband power (dBFS) measurements are the integral of the average power spectral density over all of the selected frequency bands (i.e., full range or full scale) with respect to time (i.e., energy per unit frequency per unit time) (K. Lisa Yang Center for Conservation Bioacoustics, 2022a). As zero dBFS represents maximum power, values for inband power are negative, with negative numbers closer to zero indicating more power (i.e., more energy) (Scholkmann, 2019). Per the protocol for SNR estimation provided by the Center for Conservation Bioacoustics (K. Lisa Yang Center for Conservation Bioacoustics, 2022b), dBFS for both signal and noise were converted to linear units. The SNR in units was calculated using the formula (signal-noise/noise) and then converted back to decibels (K. Lisa Yang Center for Conservation Bioacoustics, 2022b).

To determine the SNR threshold for likely focal knocks, a hard clustering technique (partitioning around medoids) was used. This method was selected over traditional k-means clustering because it is more robust to outliers and uses real data points (here, SNR measurements) from the given dataset as the medoids, rather than model-created cluster centers (Rousseeuw and Kaufman, 2009; Maechler , 2022). A dissimilarity matrix was created from the pairwise distances between data points using Gower's coefficient with the cluster package (Maechler , 2022) in R (R Core Team, 2020). The optimal number of clusters for knocks was chosen based on the highest silhouette width, an internal validation metric describing similarity between and within clusters (Rousseeuw and Kaufman, 2009). The optimal number of clusters was three; thus, only the cluster with the highest SNR values was considered to contain likely focal calls and the rest were considered non-focal. Although it is possible that some focal calls were missed using these methods, this more conservative approach helped ensure that calls from nearby conspecifics were not included in analyses on individual behavior.

From likely focal calls, seven acoustic measurements were made in Raven Pro: duration (90%), bandwidth (90%), center frequency (Hz), and 5% (minimum) and 95% (maximum) frequencies (Hz) [see Charif (2010) for details on measurement calculations]. These measurements were chosen because they are considered robust to spectrogram settings and selection bounds (Fristrup and Watkins, 1993; Cortopassi, 2006). Although peak frequency (Hz) is not considered as robust, this parameter was also measured in Raven to allow for comparison with previous studies.

RL of the sound types were calculated in Matlab (MATLAB, 2019). As the sounds were recorded on a tag affixed to the animal, the reported metric here is relative received level of the different sound types rather true source level (SL), and is intended to provide a measure of relative amplitude. Additional studies would be needed to estimate SL of these calls. The root mean square RL (rmsRL) (dB re 1 μPa) and peak-to-peak RL (ppRL) (dB re 1 μPa) of each call was calculated using the normalized energy over the duration of the entire call. Before measuring RL, acoustic data were adjusted for approximate tag sensitivity (–178 dB re 1 V μPa−1) to account for a reduction in hydrophone response at lower frequencies. Tag system sensitivity was sourced from Holt (2017), who derived values from calibration tests of DTAGs (version 3) conducted at tank facilities.

An important feature of sei whale communication may be the temporal patterning of call production, with evidence that calls may sometimes be produced in series (Tremblay , 2019; Cerchio and Weir, 2022). To quantitatively incorporate this feature into the current study, a bout criterion analysis was used (Sibly , 1990). The inter-call interval (ICI) was first calculated as the time between the start of one likely focal call and the end of the preceding likely focal call of the same type (Edds-Walton, 1997). The bout end criterion (BEC) was calculated using the maximum likelihood estimation method (MLM) in the R package DiveMove (Luque and Guinet, 2007; Luque, 2007). The MLM assumes that the distributions of intervals between calls are a mixture of two or more Poisson processes. At the smallest scale, this represents fast processes (i.e., calls within a bout), and at the largest scale, slow processes (i.e., calls in separate bouts) (Luque and Guinet, 2007). Once this threshold was established for each of the call types, individual calls were assigned as belonging to a bout or as singular based on the calculated BEC. For calls within bouts, the ICI was then calculated.

To describe the rate of signal production, diel trends, and depth of calls produced by sei whales, only likely focal calls (i.e., likely from the tagged animal) were used. Data on the sex of the tagged animal was obtained from skin particle samples collected from the suction cups after tag recovery (as in Oleson , 2007). The overall call rate was calculated as the number of likely focal calls on a tag divided by the total duration for that tag, and averaged across all tags. Call rates under various conditions were then calculated: for each call type separately; for both sexes separately with all call types combined; and for both sexes separately for downsweeps only, which may be particularly relevant to PAM.

Last, diel trends in call rates were assessed for day, twilight, and night periods for all animals combined, and then for each sex. The time of sunrise, sunset, and twilight were taken from National Weather Service databases (National Weather Service, 2022). Day was defined as the period between sunrise and sunset. Civil twilight was defined as the time when the geometric center of the sun is between 6° below the horizon and the horizon itself, and includes a short period in the morning before the sun rises and in the evening after the sun sets. Last, night was defined as the period between the end of evening twilight and the beginning of morning twilight. Morning and evening twilight constituted only a small proportion of a 24 h period and so were combined; hereafter, simply referred to as twilight. Call rates were calculated as above but using the duration of each light period during which the tag was deployed. rather than tag duration.

Differences in call type usage during the three light periods and between sexes were assessed using binomial generalized linear models (GLMs) with a logit link function in R with the package stats (Bates , 2015; R Core Team, 2020). Binomial GLMs are weighted regression models that can be used to model proportional data, accounting for unbalanced sample sizes (e.g., unequal time recorded in the three light periods or unequal samples of males and females) by using the sample sizes as weights (Nelder and Wedderburn, 1972). By modelling a GLM with a binomial distribution, the statistical probability of a call occurring in the repertoire in the three light periods or between sexes could be compared to the absence of that call type. The chosen model assumes that the repertoire is finite, and only consists of the calls used in the model. As a result, call types are not considered independent, and the sum of all call types under each condition (e.g., twilight) is equal to 1.0. Results are presented as probabilities, which demonstrate the theoretical composition of the repertoire. Post hoc analyses applied Dunnett-style contrasts with the “mvt” method in the package emmeans (Lenth, 2022). The results of the emmeans analysis are presented as odds ratios (ranging from 0 to +∞), which are calculated as
P ( A ) / ( 1 P A ) P ( B ) / ( 1 P B ) ,
(1)
where P(A) is the probability of a call type occurring under the first condition (e.g., twilight) and P(B) is the probability of that same call type occurring under the second condition (e.g., night) [Eq. (1)].

Last, the pressure sensor on the tags was used to determine the depth at which calls were produced. The average depth of all calls was calculated, as well as the average depth of each call type specifically.

Data were collected from sei whales on three separate days between April 25 and May 5, 2022 (Fig. 1). Although no dedicated behavioral sequencing took place, general behavior was noted while the vessel remained in the area. Observations before and after tag attachment indicated that animals were primarily lunge or skim feeding at the surface. Other baleen whale species were sometimes present within 1 km of the sei whales, including North Atlantic right (Eubalaena glacialis) and humpback whales (Megaptera novaeangliae). On April 25, at least nine individual sei whales were present in a feeding aggregation, including at least one mother/calf pair, as well as several right whales. On May 1, 15–20 individual sei whales were observed in the area, initially travelling before switching to feeding later in the afternoon. Up to seven individual right whales were also skim feeding in the aggregation and at least two humpback whales were in the vicinity. Last, about 12 sei whales were observed lunge and skim feeding on May 5 with no other species observed.

FIG. 1.

(Color online) Location of the tag deployments in the Southern Gulf of Maine off the northeastern coast of the United States. The hatched area indicates Stellwagen Bank National Marine Sanctuary. Dates on the map correspond to the general locations of tag deployments for the three tagging days.

FIG. 1.

(Color online) Location of the tag deployments in the Southern Gulf of Maine off the northeastern coast of the United States. The hatched area indicates Stellwagen Bank National Marine Sanctuary. Dates on the map correspond to the general locations of tag deployments for the three tagging days.

Close modal

Nine DTAGs and three CATS tags were deployed for 173.3 h of high-quality audio recordings (see the supplementary material).1 The longest deployment was 23.9 h, while the shortest was 1.4 h. Two tags had no calls detected, with recording durations of 1.4 and 13.4 h. A total of 1030 calls were detected and extracted for analysis from the remaining tags (Table S2). Of the 12 tags, sex could be determined for 10 tagged animals (7 males, 3 females, Table I). The total recording duration for females was 54.5, 94.3 for males, and 24.5 h for animals of unknown sex. The two whales of unknown sex were excluded from analyses on sex differences.

TABLE I.

Details of tag deployments in this study. n represents the number of likely focal calls (n = 523). Sex is listed as male (M), female (F), or unknown (Unk).

Deployment date Tag type Sex Recording duration (hh:mm) n calls n downsweeps n knocks n pulse type 1 n pulse type 2 n tonal n pulsive
4/25/2022  DTAG  15:32 
4/25/2022  DTAG  Unk  01:26 
4/25/2022  DTAG  23:58  17 
5/1/2022  DTAG  19:09  47  38 
5/1/2022  DTAG  18:29  73  25  47 
5/1/2022  CATS  13:25 
5/1/2022  DTAG  06:38  85  85 
5/5/2022  CATS  04:48 
5/5/2022  DTAG  16:51  91  29  58 
5/5/2022  DTAG  Unk  23:04  124  82  21  21 
5/5/2022  CATS  16:13  33  20 
5/5/2022  DTAG  13:43  52  15  30 
Deployment date Tag type Sex Recording duration (hh:mm) n calls n downsweeps n knocks n pulse type 1 n pulse type 2 n tonal n pulsive
4/25/2022  DTAG  15:32 
4/25/2022  DTAG  Unk  01:26 
4/25/2022  DTAG  23:58  17 
5/1/2022  DTAG  19:09  47  38 
5/1/2022  DTAG  18:29  73  25  47 
5/1/2022  CATS  13:25 
5/1/2022  DTAG  06:38  85  85 
5/5/2022  CATS  04:48 
5/5/2022  DTAG  16:51  91  29  58 
5/5/2022  DTAG  Unk  23:04  124  82  21  21 
5/5/2022  CATS  16:13  33  20 
5/5/2022  DTAG  13:43  52  15  30 

Both tonal and pulsed calls were detected in the audio record, 523 of which were considered likely focal based on accelerometer data (see supplementary material for Figs. S7–S10)1 or SNR threshold (for knocks). Downsweeps and tonal calls were detected, which have previously been described for this species. In addition, four new signal types that have not been previously described for sei whales: “knocks,”pulse type 1,”pulse type 2,” and “pulsive calls” were detected in the acoustic record. Tonal and pulsive calls were never deemed focal and were therefore not included when measuring acoustic parameters (Table II).

TABLE II.

Call parameters for each of the described call types. Values represent mean ± standard deviation and range (minimum–maximum). Measurements were taken from calls deemed likely focal (n = 523). Depth represents depth at the time the call was recorded on the hydrophone. As the two pulse types were sometimes detected in the same sequence, the ICI was calculated for both call types combined.

Downsweep (n = 63) Knock (n = 205) Pulse type 1 (n = 161) Pulse type 2 (n = 94)
90% call duration (s)  0.9 ± 0.2  < 0.1 ± 0.0  < 0.1 ± 0.1  0.28 ± 0.1 
(0.5–1.6)  (< 0.1–0.1)  (< 0.1–0.5)  (< 0.1–0.6) 
90% bandwidth (Hz)  58 ± 36  467 ± 248  81 ± 46  58 ± 35 
(23–281)  (94–1465)  (0–199)  (6–164) 
Center frequency (Hz)  49 ± 9  353 ± 144  51 ± 37  27 ± 15 
(12–70)  (82–844)  (12–164)  (6–59) 
Peak frequency (Hz)  55 ± 11  330 ± 163  50 ± 41  26 ± 17 
(12–70)  (47–891)  (12–176)  (0–94) 
5% frequency (Hz)  24 ± 11  184 ± 112  20 ± 11  11 ± 3 
(12–47)  (35–563)  (0–59)  (0–23) 
95% frequency (Hz)  82 ± 35  651 ± 264  100 ± 53  69 ± 36 
(47–316)  (258 ± 1688)  (12–223)  (12–176) 
rmsRL (dB re 1 μ Pa)  156 ± 2  140 ± 2  159 ± 3  159 ± 3 
(148–158)  (127–156)  (145–169)  (152–170) 
ppRL (dB re 1 μ Pa)  163 ± 4  148 ± 5  166 ± 9  169 ± 7 
(153–169)  (134–162)  (146–183)  (149–182) 
Inter-call interval (s)  1.92 ± 0.32  4.87 ± 3.40    2.43 ± 1.04 
(1.03–2.47)  (0.04–15.67)    (0.14–10.1) 
Depth (m)  3.5 ± 3.2  2.1 ± 1.4  2.0 ± 0.9  2.4 ± 2.3 
(0.7–16.4)  (0.2–8.1)  (0.5–4.6)  (0.2–18.7) 
Downsweep (n = 63) Knock (n = 205) Pulse type 1 (n = 161) Pulse type 2 (n = 94)
90% call duration (s)  0.9 ± 0.2  < 0.1 ± 0.0  < 0.1 ± 0.1  0.28 ± 0.1 
(0.5–1.6)  (< 0.1–0.1)  (< 0.1–0.5)  (< 0.1–0.6) 
90% bandwidth (Hz)  58 ± 36  467 ± 248  81 ± 46  58 ± 35 
(23–281)  (94–1465)  (0–199)  (6–164) 
Center frequency (Hz)  49 ± 9  353 ± 144  51 ± 37  27 ± 15 
(12–70)  (82–844)  (12–164)  (6–59) 
Peak frequency (Hz)  55 ± 11  330 ± 163  50 ± 41  26 ± 17 
(12–70)  (47–891)  (12–176)  (0–94) 
5% frequency (Hz)  24 ± 11  184 ± 112  20 ± 11  11 ± 3 
(12–47)  (35–563)  (0–59)  (0–23) 
95% frequency (Hz)  82 ± 35  651 ± 264  100 ± 53  69 ± 36 
(47–316)  (258 ± 1688)  (12–223)  (12–176) 
rmsRL (dB re 1 μ Pa)  156 ± 2  140 ± 2  159 ± 3  159 ± 3 
(148–158)  (127–156)  (145–169)  (152–170) 
ppRL (dB re 1 μ Pa)  163 ± 4  148 ± 5  166 ± 9  169 ± 7 
(153–169)  (134–162)  (146–183)  (149–182) 
Inter-call interval (s)  1.92 ± 0.32  4.87 ± 3.40    2.43 ± 1.04 
(1.03–2.47)  (0.04–15.67)    (0.14–10.1) 
Depth (m)  3.5 ± 3.2  2.1 ± 1.4  2.0 ± 0.9  2.4 ± 2.3 
(0.7–16.4)  (0.2–8.1)  (0.5–4.6)  (0.2–18.7) 

Downsweeps [Fig. 2(a), Mm. 1] were detected on nine of the 12 tags (n = 63 calls, 12% of all likely focal calls). Downsweeps typically swept from 82 ± 35 to 24 ± 11 Hz (95% and 5% frequency) with an average center frequency of 49 ± 9 Hz and peak frequency of 55 ± 11 Hz (Table II). The 90% duration of this call type averaged 0.93 ± 0.24 s. Downsweeps were primarily produced singly (n = 38, 60%). However, they were also detected in bouts as doublets (n = 11) and once as a triplet. The BEC for downsweeps was 7.8 s, and when produced in bouts, the average ICI within a bout was 1.92 ± 0.32 s (Table II).

FIG. 2.

(Color online) Spectrograms and waveforms for each of the measured call types (Hamming window, FFT size 2048, and 90% overlap). (a) downsweep, (b) knocks, (c) pulse type 1, (d) pulse type 2.

FIG. 2.

(Color online) Spectrograms and waveforms for each of the measured call types (Hamming window, FFT size 2048, and 90% overlap). (a) downsweep, (b) knocks, (c) pulse type 1, (d) pulse type 2.

Close modal
Mm. 1.

Downsweep.

Mm. 1.

Downsweep.

Close modal

The most commonly detected sounds (n = 20, 539% of likely focal calls) were relatively low-amplitude sounds (140 ± 2 dB re 1 μPa rmsRL; 148 ± 5 dB re 1 μPa ppRL) that were similar aurally, visually, and in measured parameters to the knocks produced by humpback whales (Cusano , 2020; Gascón, 2021) and pulses of North Atlantic right whales (Parks , 2019). They were also labelled as knocks here to reduce confusion with other sei whale call types [Fig. 2(b), Mm. 2]. Knocks were very short (< 1 s, on average) and mid-frequency (353 ± 144 Hz center frequency), although the average 95% frequency was 651 ± 264 Hz, substantially higher than the other sound types described here. This sound had the widest bandwidth of all call types measured (467 ± 248 Hz 90% bandwidth) (Table II). Knocks were produced singly (34%) or in bouts of between 2 and 17 calls. The BEC for knocks was 16.9 s, and when produced in bouts, there was relatively high variation in within bout ICI (4.87 ± 3.40 s) (Table II).

Mm. 2.

Knock series.

Mm. 2.

Knock series.

Close modal

Pulse type 1 (PT1) calls (n = 16, 131% of likely focal calls) were low frequency, impulsive, narrowband calls with no visible harmonics [Fig. 2(c), Mm. 3]. They had a mean center frequency of 51 ± 37 Hz, and 5% and 95% frequencies of 20 ± 11 and 100 ± 53 Hz on average (Table II). Another pulse-type call that was lower in frequency and longer in duration than PT1 calls was also detected, and sometimes occurred in the same sequence. These were labelled here as Pulse type 2 (PT2) calls (n = 94, 18% of likely focal calls). PT2 calls were typically longer in 90% duration (0.28 ± 0.12 s) compared to PT1 calls (0.08 ± 0.05 s) [Fig. 2(d), Mm. 4]. They covered the lowest frequency range of all calls recorded, with a center frequency of 27 ± 15 Hz and average 5% and 95% frequencies of 11 ± 3 and 69 ± 36 Hz, respectively. Although variable, PT1 calls and PT2 calls had the highest energy, with maximum ppRLs of 183 and 182 dB re 1 μPa, respectively (Table II).

Mm. 3.

PT1 call sequence.

Mm. 3.

PT1 call sequence.

Close modal
Mm. 4.

PT2 call sequence.

Mm. 4.

PT2 call sequence.

Close modal

A distinguishing feature of both PT1 calls and PT2 calls was their production as sequences. As these calls were sometimes produced in the same sequence, they were combined for the bout analysis. The BEC for pulse sequences was 10.4 s, with an average within bout ICI of 2.43 ± 1.04 s. These calls were infrequently detected individually (0.04% of pulses) and were typically found in bouts of 2–13 calls (Table II).

Focal call rates detected on the tags were low, with an average of 3.06 calls/h. The highest call rates were for knocks (1.56 calls/h). PT1 calls (0.77 calls/h), and PT2 calls (0.42 calls/h) were sometimes found in the same sequence. If grouped together, the average call rate for all PT calls was 1.19 calls/h. Downsweeps, the call type used for PAM of sei whales, had the lowest average call rate (0.32 calls/h). Rates of likely focal calls for females (3.93 calls/h) were higher than for males (2.79 calls/h), as were call rates for downsweeps (0.73 calls/h and 0.23 calls/h). However, caution should be used in interpreting call rate differences between the sexes because of the unbalanced sample sizes (n = 3 females, n = 7 males) and relatively short tag attachment durations.

Diel trends in call rates were also apparent. There were 81 h of data during night periods, 95 h during day periods, and only 8.5 h during twilight periods. Despite the low sample size of twilight data (as a result of the shorter proportion of twilight hours in a 24 h period), the highest average likely focal call rate was during this period at 8.7 calls/h. The average likely focal call rate at night was 1.47 calls/h and 3.01 calls/h during the day.

Diel differences arose in the call rates between males and females in the three light periods, although again, caution should be used in interpreting these results due to small sample sizes and unbalanced samples of both the tags and the total recorded time in the light periods between the sexes. During twilight, females had an average likely focal call rate of 21.9 calls/h as opposed to only 5.5 calls/h for males, despite the fact that males were recorded a total of 4.4 h at night and females 3.1 h. The average call rate for females was also higher at night compared to males (2.5 calls/h and 0.4 calls/h, respectively), again despite discrepancies in the number of hours of recording time (26.5 h for females, 45.5 h for males). Call rates between the sexes during the day were similar, with females calling on average 3.4 calls/h and males 3.1 calls/h.

In addition to differences in call rate, there were differences in the acoustic repertoire during the three light periods (Table III, Fig. 3). The estimated GLM probability of detecting likely focal knocks was greatest at night (0.55 ± 0.04) and lowest at twilight (0.05 ± 0.02; odds ratio between night and twilight 21.9 ± 10.7; z ratio 6.3; p < 0.0001). The probability of detecting likely focal downsweeps was also highest at night (0.24 ± 0.03) but lowest during the day (0.05 ± 0.01; odds ratio between night and day 0.18 ± 0.06; z ratio 5.2; p < 0.0001). In contrast, the probability of detecting both likely focal PT1 and PT2 calls were significantly higher during twilight (0.54 ± 0.05 and 0.29 ± 0.05, respectively) and lowest at night (0.12 ± 0.03 and 0.09 ± 0.02, respectively) (Table III, Fig. 3).

TABLE III.

Results of the generalized linear model indicating the modelled probability of detecting each call type (likely focal calls only) in the three light periods. The estimated mean probability is shown in the first three columns and the pairwise contrasts are shown in the last three columns. See Methods for an explanation of odds ratios. Values represent mean ± standard error. A negative z ratio indicates there is a lower odds ratio of observing that call type in the first light period listed, compared to the second.

Day Night Twilight Day/Night Day/Twilight Night/Twilight
Downsweeps  0.05 ± 0.01  0.24 ± 0.03  0.12 ± 0.03  Odds ratio = 0.18 ± 0.06  Odds ratio = 0.44 ± 0.18  Odds ratio = 2.39 ± 0.89 
z ratio = −5.22  z ratio = −2.0  z ratio = 2.34 
p < 0.0001a  p = 0.1114  p = 0.0503b 
Knocks  0.42 ± 0.03  0.55 ± 0.04  0.05 ± 0.02  Odds ratio = 0.58 ± 0.12  Odds ratio = 12.8 ± 6.08  Odds ratio = 21.9 ± 10.7 
z ratio = −2.65  z ratio = 5.36  z ratio = 6.34 
p = 0.0196b  p < 0.0001a  p < 0.0001a 
Pulse type 1  0.34 ± 0.03  0.12 ± 0.03  0.54 ± 0.05  Odds ratio = 3.8 ± 1.07  Odds ratio = 0.42 ± 0.10  Odds ratio = 0.11 ± 0.04 
z ratio = 4.74  z ratio = −3.53  z ratio = −6.74 
p < 0.0001a  p = 0.0012b  p < 0.0001a 
Pulse type 2  0.19 ± 0.02  0.09 ± 0.02  0.29 ± 0.05  Odds ratio = 2.39 ± 0.76  Odds ratio = 0.59 ± 0.16  Odds ratio = 0.25 ± 0.09 
z ratio = 2.73  z ratio = −1.91  z ratio = −3.86 
p = 0.0171b  p = 0.1344  p = 0.0003a 
Day Night Twilight Day/Night Day/Twilight Night/Twilight
Downsweeps  0.05 ± 0.01  0.24 ± 0.03  0.12 ± 0.03  Odds ratio = 0.18 ± 0.06  Odds ratio = 0.44 ± 0.18  Odds ratio = 2.39 ± 0.89 
z ratio = −5.22  z ratio = −2.0  z ratio = 2.34 
p < 0.0001a  p = 0.1114  p = 0.0503b 
Knocks  0.42 ± 0.03  0.55 ± 0.04  0.05 ± 0.02  Odds ratio = 0.58 ± 0.12  Odds ratio = 12.8 ± 6.08  Odds ratio = 21.9 ± 10.7 
z ratio = −2.65  z ratio = 5.36  z ratio = 6.34 
p = 0.0196b  p < 0.0001a  p < 0.0001a 
Pulse type 1  0.34 ± 0.03  0.12 ± 0.03  0.54 ± 0.05  Odds ratio = 3.8 ± 1.07  Odds ratio = 0.42 ± 0.10  Odds ratio = 0.11 ± 0.04 
z ratio = 4.74  z ratio = −3.53  z ratio = −6.74 
p < 0.0001a  p = 0.0012b  p < 0.0001a 
Pulse type 2  0.19 ± 0.02  0.09 ± 0.02  0.29 ± 0.05  Odds ratio = 2.39 ± 0.76  Odds ratio = 0.59 ± 0.16  Odds ratio = 0.25 ± 0.09 
z ratio = 2.73  z ratio = −1.91  z ratio = −3.86 
p = 0.0171b  p = 0.1344  p = 0.0003a 
a

Statistical significance at the p < 0.001 level.

b

Statistical significance at the p < 0.05 level.

FIG. 3.

(Color online) Diel differences in the GLM probability of detecting each call type during the three light periods.

FIG. 3.

(Color online) Diel differences in the GLM probability of detecting each call type during the three light periods.

Close modal

There were also sex differences in the probability of detecting call types (Table IV, Fig. 4). The GLM estimated probability of detecting PT1 calls was significantly higher in females (0.46) compared to males (0.23; odds ratio 2.87 ± 0.64; z ratio 4.75; p < 0.0001). This trend was also evident in PT2 calls, with the probability of detection higher in females (0.22) compared to males (0.14; odds ratio 1.79 ± 0.48; z ratio 2.16; p = 0.0307). The probability of detecting knocks, however, was significantly higher in males (0.50) compared to females (0.14; odds ratio 0.16 ± 0.04; z ratio –7.42; p < 0.0001). There was no significant difference in the probability of downsweeps between males and females (Table IV, Fig. 4).

TABLE IV.

Results of the generalized linear model indicating the probability of detecting each call type in males and females. The estimated mean probability is shown in the first two columns and the pairwise contrasts are shown in the last two columns. See methods for an explanation of odds ratios. Values represent mean ± standard error. A negative z ratio indicates there is a lower odds ratio of observing that call type in females compared to males.

Female Male Female/Male
Downsweeps  0.18 ± 0.03  0.13 ± 0.02  Odds ratio = 1.43 ± 0.40 
z ratio= 1.284 
p = 0.1991 
Knocks  0.14 ± 0.02  0.50 ± 0.04  Odds ratio = 0.16 ± 0.04 
z ratio = −7.42 
p < 0.0001a 
Pulse type 1  0.46 ± 0.03  0.23 ± 0.03  Odds ratio = 2.87 ± 0.64 
z ratio= 4.75 
p < 0.0001a 
Pulse type 2  0.22 ± 0.03  0.14 ± 0.03  Odds ratio = 1.79 ± 0.48 
z ratio= 2.16 
p = 0.0307b 
Female Male Female/Male
Downsweeps  0.18 ± 0.03  0.13 ± 0.02  Odds ratio = 1.43 ± 0.40 
z ratio= 1.284 
p = 0.1991 
Knocks  0.14 ± 0.02  0.50 ± 0.04  Odds ratio = 0.16 ± 0.04 
z ratio = −7.42 
p < 0.0001a 
Pulse type 1  0.46 ± 0.03  0.23 ± 0.03  Odds ratio = 2.87 ± 0.64 
z ratio= 4.75 
p < 0.0001a 
Pulse type 2  0.22 ± 0.03  0.14 ± 0.03  Odds ratio = 1.79 ± 0.48 
z ratio= 2.16 
p = 0.0307b 
a

Indicates statistical significance at the p < 0.001 level.

b

Indicates statistical significance at the p < 0.05 level.

FIG. 4.

(Color online) Sex differences in the GLM probability of detecting each call type.

FIG. 4.

(Color online) Sex differences in the GLM probability of detecting each call type.

Close modal

The average depth of call production for the tagged whales across all call types was 3.5 m (Table II). This result is unsurprising considering all dives were shallower than 10 m with very few exceptions (13 dives to ∼20 m and 3 dives to ∼25 m across all tag records), likely due to the near surface feeding. Likely focal Downsweeps and PT2 calls were both detected below 16 m at times, while likely focal knocks were never detected deeper than ∼4.5 m and likely focal PT1 calls never deeper than ∼8 m (Table I). See Figs. 5 and 6 for an example dive profile illustrating the depth of the knocks and PT calls.

FIG. 5.

(Color online) Dive profile of a female sei whale tagged on May 1, 2022, with shading to indicate twilight periods (light gray) and night (dark gray), and asterisks to show the time of calls; knocks (blue); pulse type 1 (yellow); pulse type 2 (green).

FIG. 5.

(Color online) Dive profile of a female sei whale tagged on May 1, 2022, with shading to indicate twilight periods (light gray) and night (dark gray), and asterisks to show the time of calls; knocks (blue); pulse type 1 (yellow); pulse type 2 (green).

Close modal
FIG. 6.

(Color online) Dive profile from Fig. 5, zoomed in to highlight the dive behavior while producing likely focal pulse type calls and knocks. Shading indicates twilight, and asterisks show the time of calls; knocks (blue); pulse type 1 (yellow); pulse type 2 (green).

FIG. 6.

(Color online) Dive profile from Fig. 5, zoomed in to highlight the dive behavior while producing likely focal pulse type calls and knocks. Shading indicates twilight, and asterisks show the time of calls; knocks (blue); pulse type 1 (yellow); pulse type 2 (green).

Close modal

This study represents the first acoustic tag study of the calling behavior of sei whales and the first detailed description of their acoustic behavior on their springtime foraging grounds in the Western North Atlantic. Using digital acoustic recording tags, we were able to unambiguously assign call types to sei whales and assess individual calling behavior including call types, call rates, and depths. This study revealed that the acoustic repertoire of sei whales is more diverse than previously described. Diel trends in calling behavior within the aggregation, with call rates highest during twilight periods, suggest that behavioral changes during crepuscular periods (potentially related to prey distributions) may impact the calling rate of the species.

Females produced a higher proportion of pulse type calls and males produced a higher proportion of knocks, suggesting that there may be sex differences in repertoire usage in these call types. No sex differences were detected in downsweep production; however, call rates were low, particularly for the downsweep call. As this call type is predominantly used for PAM of this species, it is important to take this cue rate into account when considering the probability of detection of sei whales, especially for passive acoustic density estimation. Downsweep detection may still be sufficient for presence/absence detection in PAM data, especially given its relatively high amplitude, but the individual call rate reported here may have implications for different PAM objectives.

Six call types were recorded from the acoustic tags, including previously described downsweeps (see the supplementary material for Table S1).1 The downsweeps recorded here were lower in frequency (83 to 25 Hz) than those reported from the Southern Hemisphere (Calderan , 2014; Español-Jiménez , 2019; Cerchio and Weir, 2022), and North Atlantic Labrador Sea population (Romagosa , 2015). For sei whales in the North Pacific, two types of downsweeps have been recorded off Hawai'i (Rankin and Barlow, 2007), and the frequency range of downsweeps in the present study fall between the two North Pacific downsweeps. Unsurprisingly, the downsweeps here are comparable to those described for sei whales in the North Atlantic Nova Scotian stock (Baumgartner , 2008), presumably the same stock as the sei whales in the present study. The results of our research and previous studies indicate that downsweeps represent a global call type, albeit with some differences in acoustic parameters. The observed variation may be related to differences between ocean basins and stocks, as seen with other baleen whales [e.g., blue (Stafford , 2001), humpback (Helweg , 1998; Epp , 2021), right (Parks , 2007), and minke whales (Rankin and Barlow, 2005)], and may present an opportunity to differentiate stocks using acoustics. Although tag technology is subject to flow noise due to the movement of the whale, sometimes obscuring lower frequencies (Stimpert , 2011), tags are able to capture sounds in close proximity to the hydrophone. This can provide details of call features, such as harmonics and higher frequencies, which quickly attenuate with distance from a source (Richardson , 1995). Using digital recording tags in other ocean basins in the future would provide a better understanding of any geographical differences, which could potentially be helpful in differentiating stocks, as with blue (Samaran , 2013) and fin whales (Delarue , 2009; Castellote , 2012).

With the use of tag technology and the identification of sex from skin samples, we were able to unambiguously assign calls to individuals of known sex in this species for the first time. As a result, we were able to increase our understanding of the potential function of sei whale downsweeps. One distinguishing feature of the downsweeps recorded here is the lack of a significant difference in their probability of detection between sexes. There was also evidence of counter-calling (i.e., a likely focal call followed closely by a non-focal call) on at least four tags, two of which were deployed on males and two on females. Along with the fact that downsweeps appear to be a common sei whale call type globally, and are detected in all habitats and during migration, our results corroborate previous findings that downsweeps serve a social function and may be used as a contact call (Baumgartner , 2008), similar to humpback whale wops or whups (Dunlop , 2008), fin whale 20 Hz pulses (Payne and Webb, 1971), North Atlantic right whales (Parks and Clark, 2007), and southern right whale upcalls (Clark, 1983).

We recorded four sounds that have not been reported for sei whales to date: pulsive calls, knocks, PT1 calls, and PT2 calls. No pulsive calls (n = 7 calls) were deemed to be produced by the tagged whale, and so were not quantitatively assessed nor assigned to a particular sex. Of the remaining call types that have yet to be described in the literature, knocks were detected most frequently. These calls are relatively low-amplitude, mid-frequency, and short in duration (Table II), likely contributing to their lack of detection on PAM systems. The knocks detected were similar aurally and visually to call types produced by humpback and North Atlantic right whales (Parks , 2019; Cusano , 2020), which have thus far been entirely limited to tag recordings. However, we are confident that these sounds originated from the tagged animal rather than a nearby animal of a different species because of their ubiquity across tags (n = 8 tags) and throughout deployments (present at the beginning, middle, and end of some tags, e.g., bb121b, 121d, and 125c). It is possible that these sounds are simply a by-product produced during certain behaviors, a possibility noted by Parks (2019) for North Atlantic right whale pulses. However, we detected sei whale knocks on a feeding ground, whereas those described from other baleen whale species were from breeding grounds and along migration corridors, areas where little or no feeding occurs. Additional effort should focus on listening for this sound on tagged sei whales in other habitats and behavioral contexts in order to prove intentionality.

PT1 calls and PT2 calls were less common than knocks, and were detected on seven and four tags, respectively. In contrast to previously described pulses from sei whales on the Nova Scotian shelf, PT1 calls and PT2 calls were substantially lower in frequency (Thompson , 1979; Knowlton , 1991), and unlike any call type ascribed previously to sei whales. PT1 and PT2 calls were sometimes produced in the same sequence, indicating the possibility that they are variations of the same call type; however, the same phenomenon occurs with sei whale Downsweep 1 and Downsweep 2 calls off the Falkland Islands, which are distinct call types (Cerchio and Weir, 2022). Further, the distinct acoustic metrics of PT1 and PT2 calls (Table II), recorded here on tags with relatively little attenuation, currently warrants their separation. As with knocks, although both PT1 and PT2 calls have yet to be described, we are confident that they were produced by the animal and not something bumping the tag because of the presence of knocks on the majority of tags (7 of 12 tags), their production throughout some tag deployments, and the consistent inter-pulse interval across sequences (Table II). More fine-scale behavioral data will be necessary to determine the precise context of the newly recorded sei whale call types and interpret any potential functions for these sounds.

Our tag-based analysis of diel patterns in calling activity showed the lowest call rates occurred during night hours. Baumgartner and Fratantoni (2008) also detected relatively low call rates at night. However, the authors linked this result to a possible increase in social behavior during the day while their prey (Calanus finmarchicus) was at depth, and an increase in feeding at night when prey rise to the surface. Although no dedicated behavioral sequencing occurred during our study, direct visual observations of surface feeding occurred every day of data collection, confirming sei whales do sometimes feed at shallow depths during daylight hours in this habitat in Massachusetts Bay. This observation is in line with Parks (2012) who used net, tow, and active acoustic prey sampling to show C. finmarchicus were present in the top 5 m of the water column during the day. The low call rates of sei whales detected during the day, despite a difference in the time of day that feeding occurs, indicates that there may be alternative drivers for the rate of call production. Additional data are needed to correlate call rates to any specific behavioral state.

The present study provides the first insight into the acoustic behavior of sei whales on the feeding ground using animal-borne biologging tags. Although the study was conducted during a small temporal window and in one geographic area, these results are relevant for future PAM and conservation efforts. Overall call rates were low, particularly for downsweeps, which are the call type used for PAM. Downsweeps were not detected on all tags; 25% of tags had no downsweeps, and 58% of tags had no focal downsweeps. Additionally, most downsweeps were produced as single calls and would not be included as a positive detection in some of the more conservative PAM assessments (e.g., Kowarski , 2022). Combined with the low likely focal call rates (3.06 calls/h), these results indicate that some sei whales will likely go through certain areas undetected. Although including the newly described call types could increase the detectability of sei whales, due to their higher call rates, they have thus far only been detected using suction cup tags, directly applied to the animal. It is possible that these sounds were not previously detected in PAM studies because they were unknown to scientists; however, it is likely that some of these calls would be undetectable anyway due to their characteristics (e.g., low amplitude and short durations for knocks). Current acoustic monitoring techniques should consider these call characteristics for detecting the presence/absence of sei whales or for enforcing vessel restrictions when sei whales are foraging.

The results of this study shed light on the acoustic repertoire and behavior of this endangered species on the feeding ground. Our study has provided details on the acoustic behavior of sei whales, which was previously undocumented, highlighting the benefit of tag technology in providing invaluable data. Our results can provide additional information to inform PAM efforts for this species. The low likely focal call rates for downsweeps, the call type used in PAM of sei whale presence, suggest that a better understanding of cue rates for this call type may be needed for density estimation from PAM. The downsweep was detected on most tags when including non-focal signals, suggesting that for detection of sei whales in this habitat, it remains a suitable signal for detection of sei whales through acoustic presence. We recommend that future research be focused on using tag technology in additional habitats and during other behavioral contexts to better understand the vocal repertoire of sei whales and the behavioral function of the different call types. In addition, future studies should assess the feasibility of including additional call types in PAM to aid in the detection of the acoustic presence of sei whales and ultimately aid in conservation efforts for this endangered species.

This work was supported by funding from the Bureau of Ocean Energy Management and Blue World Research Institute. Funding for Ocean Alliance was provided by the Pamela K. Omidyar Trust and the Sarah K. de Coizart Tenth Perpetual Trust. Data were collected under National Marine Fisheries Service ESA/MMPA Permit No. 18059. Research protocols were approved by the Institutional Animal Care and Use Committee at Syracuse University (IACUC 20-002). We thank J. Dombroski, H. He, K. Knapp, J. Linsky, T. Silva, M. Thompson, V. Perez, L. Waters, R. Finley, and the crew of the R/V Auk. The authors have no conflicts to disclose. The data that support the findings of this study are available from the corresponding author upon reasonable request.

1

See supplementary material at https://doi.org/10.1121/10.0022570 for supplementary tables and figures.

1.
Bates
,
D.
,
Mächler
,
M.
,
Bolker
,
B.
, and
Walker
,
S.
(
2015
). “
Fitting linear mixed-effects models using lme4
,”
J. Stat. Softw.
67
,
1
48
.
2.
Baumgartner
,
M. F.
, and
Fratantoni
,
D. M.
(
2008
). “
Diel periodicity in both sei whale vocalization rates and the vertical migration of their copepod prey observed from ocean gliders
,”
Limnol. Oceanographr.
53
,
2197
2209
.
3.
Baumgartner
,
M. F.
,
Van Parijs
,
S. M.
,
Wenzel
,
F. W.
,
Tremblay
,
C. J.
,
Esch
,
H. C.
, and
Warde
,
A. M.
(
2008
). “
Low frequency vocalizations attributed to sei whales (Balaenoptera borealis)
,”
J. Acoust. Soc. Am.
124
,
1339
1349
.
4.
Buchan
,
S. J.
,
Gutiérrez Cabello
,
L.
,
Baumgartner
,
M.
,
Stafford
,
K. M.
,
Ramirez
,
N.
,
Pizarro
,
O.
, and
Cifuentes
,
J.
(
2022
). “
Distribution of blue and sei whale vocalizations, and temperature—Salinity characteristics from glider surveys in the Northern Chilean Patagonia mega-estuarine system
,”
Front. Mar. Sci.
9
,
903964
.
5.
Calderan
,
S.
,
Miller
,
B.
,
Collins
,
K.
,
Ensor
,
P.
,
Double
,
M.
,
Leaper
,
R.
, and
Barlow
,
J. P.
(
2014
). “
Low-frequency vocalizations of sei whales (Balaenoptera borealis) in the Southern Ocean
,”
J. Acoust. Soc. Am.
136
,
EL418
EL423
.
6.
Caretta
,
J. V.
,
Oleson
,
E. M.
,
Forney
,
K. A.
,
Muto
,
M. M.
,
Weller
,
D. W.
,
Lang
,
A. R.
,
Baker
,
J.
,
Hanson
,
B.
,
Orr
,
A. J.
,
Barlow
,
J.
,
Moore
,
J. E.
, and
Brownell
,
R. L. J.
(
2022
). “
U.S. Pacific marine mammal stock assessments: 2021
,” NOAA-TM-NMFS-SWFSC-663 (
U.S. Department of Commerce
).
7.
Castellote
,
M.
,
Clark
,
C. W.
, and
Lammers
,
M. O.
(
2012
). “
Fin whale (Balaenoptera physalus) population identity in the western Mediterranean Sea
,”
Mar. Mamm. Sci.
28
,
325
344
.
8.
Cerchio
,
S.
, and
Weir
,
C. R.
(
2022
). “
Mid-frequency song and low-frequency calls of sei whales in the Falkland Islands
,”
R. Soc. Open Sci.
9
,
220738
.
9.
Charif
,
R.
,
Waack
,
A.
, and
Strickman
,
L.
(
2010
). Raven Pro 1.4 user's manual (
Cornell Lab of Ornithology, Ithaca
,
NY
).
10.
Clark
,
C. W.
(
1983
). “
Acoustic communication and behavior of the Southern right whale
,” in
Communication and Behavior of Whales
, edited by
R. S.
Payne
(
Westview Press
,
Boulder, CO
), pp.
163
198
.
11.
Cortopassi
,
K. A.
(
2006
). “
Automated and robust measurement of signal features [webpage]
,” Cornell Laboratory Ornithology, http://www.birds.cornell.edu/brp/research/algorithm/automated-and-robust-measurement-of-signal-features (Last viewed November 11, 2022).
12.
Cusano
,
D. A.
,
Indeck
,
K. L.
,
Noad
,
M. J.
, and
Dunlop
,
R. A.
(
2020
). “
Humpback whale (Megaptera novaeangliae) social call production reflects both motivational state and arousal
,”
Bioacoustics
31
,
17
40
.
13.
Davis
,
G. E.
,
Baumgartner
,
M. F.
,
Corkeron
,
P. J.
,
Bell
,
J.
,
Berchok
,
C.
,
Bonnell
,
J. M.
,
Thornton
,
J. B..
,
Brault
,
S.
,
Buchanan
,
G. A.
,
Cholewiak
,
D. M.
,
Clark
,
C. W.
,
Delarue
,
J.
,
Hatch
,
L. T.
,
Klinck
,
H.
,
Kraus
,
S. D.
,
Martin
,
B.
,
Mellinger
,
D. K.
,
Moors-Murphy
,
H. B.
,
Nieukirk
,
S.
,
Nowacek
,
D. P.
,
Parks
,
S. E.
,
Parry
,
D.
,
Pegg
,
N.
,
Read
,
A. J.
,
Rice
,
A. N.
,
Risch
,
D.
,
Scott
,
A.
,
Soldevilla
,
M. S.
,
Stafford
,
K. M.
,
Stanistreet
,
J. E.
,
Summers
,
E.
,
Todd
,
S.
, and
Van Parijs
,
S. M.
(
2020
). “
Exploring movement patterns and changing distributions of baleen whales in the western North Atlantic using a decade of passive acoustic data
,”
Glob. Change Biol.
26
,
4812
4840
.
14.
Delarue
,
J. J.-Y.
,
Todd
,
S. K.
,
Van Parijs
,
S. M.
, and
Di Iorio
,
L.
(
2009
). “
Geographic variation in Northwest Atlantic fin whale (Balaenoptera physalus) song: Implications for stock structure assessment
,”
J. Acoust. Soc. Am.
125
,
1774
1782
.
15.
Donovan
,
G. P.
(
1991
). “
A review of IWC stock boundaries
,”
Rep. Int. Whaling Commission
13
,
39
68
.
16.
Dunlop
,
R. A.
,
Cato
,
D. H.
, and
Noad
,
M. J.
(
2008
). “
Non-song acoustic communication in migrating humpback whales (Megaptera novaeangliae)
,”
Mar. Mamm. Sci.
24
,
613
629
.
17.
Edds-Walton
,
P. L.
(
1997
). “
Acoustic communication signals of mysticete whales
,”
Bioacoustics
8
,
47
60
.
18.
Epp
,
M. V.
,
Fournet
,
M. E.
,
Silber
,
G. K.
, and
Davoren
,
G. K.
(
2021
). “
Allopatric humpback whales of differing generations share call types between foraging and wintering grounds
,”
Sci. Rep.
11
,
16297
.
19.
Español-Jiménez
,
S.
,
Bahamonde
,
P. A.
,
Chiang
,
G.
, and
Häussermann
,
V.
(
2019
). “
Discovering sounds in Patagonia: Characterizing sei whale (Balaenoptera borealis) downsweeps in the south-eastern Pacific Ocean
,”
Ocean Sci.
15
,
75
82
.
20.
Friedlaender
,
A. S.
,
Tyson
,
R. B.
,
Stimpert
,
A. K.
,
Read
,
A. J.
, and
Nowacek
,
D. P.
(
2013
). “
Extreme diel variation in the feeding behavior of humpback whales along the western Antarctic Peninsula during autumn
,”
Mar. Ecol. Prog. Ser.
494
,
281
289
.
21.
Fristrup
,
K. M.
, and
Watkins
,
W. A.
(
1993
). “
Marine animal sound classification
,” WHOI-94-13 (
Woods Hole Oceanographic Institution
).
22.
Gascón
,
M. P.
(
2021
). “
Humpback whale (Megaptera novaeangliae) social calls in the southeast pacific population: Context, diversity, and call bouts analysis
,” Ph.D. dissertation,
Rheinische Friedrich-Wilhelms-Universität Bonn
,
Bonn, Germany
.
23.
Goldbogen
,
J. A.
,
Cade
,
D.
,
Boersma
,
A.
,
Calambokidis
,
J.
,
Kahane‐Rapport
,
S.
,
Segre
,
P.
,
Stimpert
,
A.
, and
Friedlaender
,
A.
(
2017
). “
Using digital tags with integrated video and inertial sensors to study moving morphology and associated function in large aquatic vertebrates
,”
Anat. Rec.
300
,
1935
1941
.
24.
Goldbogen
,
J. A.
,
Stimpert
,
A. K.
,
DeRuiter
,
S. L.
,
Calambokidis
,
J.
,
Friedlaender
,
A. S.
,
Schorr
,
G. S.
,
Moretti
,
D. J.
,
Tyack
,
P. L.
, and
Southall
,
B. L.
(
2014
). “
Using accelerometers to determine the calling behavior of tagged baleen whales
,”
J. Exp. Biol.
217
,
2449
2455
.
25.
Hayes
,
S. A.
,
Josephson
,
E.
,
Maze-Foley
,
K.
,
Rosel
,
P. E.
, and
Turek
,
J.
(
2022
). “
U.S. Atlantic and Gulf of Mexico Marine Mammal Stock Assessments 2021
,” NOAA Technical Memorandum NMFS-NE-288 (
U.S. Department of Commerce
,
Woods Hole, MA
).
26.
Helweg
,
D. A.
,
Cat
,
D. H.
,
Jenkins
,
P. F.
,
Garrigue
,
C.
, and
McCauley
,
R. D.
(
1998
). “
Geographic variation in south Pacific humpback whale songs
,”
Behaviour
135
,
1
27
.
27.
Holt
,
M. M.
,
Hanson
,
M. B.
,
Giles
,
D. A.
,
Emmons
,
C. K.
, and
Hogan
,
J. T.
(
2017
). “
Noise levels received by endangered killer whales Orcinus orca before and after implementation of vessel regulations
,”
Endang. Species Res.
34
,
15
26
.
28.
Horwood
,
J.
(
2009
). “
Sei whale: Balaenoptera borealis
,” in
Encyclopedia of Marine Mammals
, edited by
W.
Perrin
,
B.
Würsig
, and
J.
Thewissen
(
Academic Press
,
Amsterdam, Netherlands
), pp.
1001
1003
.
29.
Huijser
,
L. A. E.
,
Bérubé
,
M.
,
Cabrera
,
A. A.
,
Prieto
,
R.
,
Silva
,
M. A.
,
Robbins
,
J.
,
Kanda
,
N.
,
Pastene
,
L. A.
,
Goto
,
M.
,
Yoshida
,
H.
,
Víkingsson
,
G. A.
, and
Palsbøll
,
P. J.
(
2018
). “
Population structure of North Atlantic and North Pacific sei whales (Balaenoptera borealis) inferred from mitochondrial control region DNA sequences and microsatellite genotypes
,”
Conserv. Genet.
19
,
1007
1024
.
30.
Ingebrigtsen
,
A.
(
1929
). “
Whales caught in the North Atlantic and other seas
,”
Conseil Permanent Int. pour l'Explor. de la Mer. Rapports et Proces-Verbaux des Reunions
56
,
123
135
.
31.
Johnson
,
M.
,
Aguilar de Soto
,
N.
, and
Madsen
,
P. T.
(
2009
). “
Studying the behaviour and sensory ecology of marine mammals using acoustic recording tags: A review
,”
Mar. Ecol. Prog. Ser.
395
,
55
73
.
32.
Johnson
,
M. P.
, and
Tyack
,
P. L.
(
2003
). “
A digital acoustic recording tag for measuring the response of wild marine mammals to sound
,”
IEEE J. Ocean. Eng.
28
,
3
12
.
33.
K. Lisa Yang Center for Conservation Bioacoustics
(
2022a
). “
Raven Pro 1.6: Interactive Sound Analysis Software (version 1.6.4) [computer program]
,”
Cornell Lab of Ornithology
, https://www.ravensoundsoftware.com (Last viewed November 11, 2022).
34.
K. Lisa Yang Center for Conservation Bioacoustics
(
2022b
). “
Raven sound analysis signal-to-noise ratio (SNR) - User protocol [webpage]
,”
Cornell Lab of Ornithology
, https://ravensoundsoftware.com/knowledge-base/signal-to-noise-ratio-snr (Last viewed November 11, 2022).
35.
Knowlton
,
A.
,
Clark
,
C. W.
, and
Kraus
,
S.
(
1991
). “
Sounds recorded in the presence of sei whales, Balaenoptera borealis
,”
J. Acoust. Soc. Am.
89
,
1968
.
36.
Kowarski
,
K. A.
,
Martin
,
S. B.
,
Maxner
,
E. E.
,
Lawrence
,
C. B.
,
Delarue
,
J.-Y.
, and
Miksis‐Olds
,
J. L.
(
2022
). “
Cetacean acoustic occurrence on the US Atlantic Outer Continental Shelf from 2017 to 2020
,”
Mar. Mamm. Sci.
39
,
175
199
.
37.
Lenth
,
R. V.
(
2022
). “
emmeans: Estimated marginal means, aka least-squares means
,” R package version 1.8.1-1.
38.
Luque
,
S.
, and
Guinet
,
C.
(
2007
). “
A maximum likelihood approach for identifying dive bouts improves accuracy, precision and objectivity
,”
Behav.
144
,
1315
1332
.
39.
Luque
, S. P.
(
2007
). “
An Introduction to the diveMove Package
,”
R-News
7
,
8
14
.
40.
Maechler
,
M.
,
Rousseeuw
,
P. J.
,
Struyf
,
A.
,
Hubert
,
M.
, and
Hornik
,
K.
(
2022
). “
cluster: Cluster analysis basics and extensions
,” R package version 2.1.3.
41.
MATLAB
. (
2019
). “
MATLAB R2019b (Version 9.7) [Computer Program]
,” The MathWorks Inc. (Last viewed November 2022).
42.
McDonald
,
M. A.
,
Hildebrand
,
J. A.
,
Wiggins
,
S. M.
,
Thiele
,
D.
,
Glasgow
,
D.
, and
Moore
,
S. E.
(
2005
). “
Sei whale sounds recorded in the Antarctic
,”
J. Acoust. Soc. Am.
118
,
3941
3945
.
43.
Mitchell
,
E.
, and
Chapman
,
D. G.
(
1977
). “
Preliminary assessment of stocks of northwest Atlantic sei whales (Balaenoptera borealis)
,”
Rep. to Int. Whaling Commission
1
,
117
120
.
44.
National Weather Service
(
2022
). “
Boston/Norton, MA Sunrise & Sunset/Moonrise & Moonset [webpage]
,”
National Oceanic and Atmospheric Administration
, https://www.weather.gov/box/sunmoon (Last viewed November 17, 2022).
45.
Nelder
,
J. A.
, and
Wedderburn
,
R. W.
(
1972
). “
Generalized linear models
,”
J. R. Stat. Soc. Ser. A: Stat. Soc.
135
,
370
384
.
46.
Newhall
,
A. E.
,
Lin
,
Y.-T.
,
Lynch
,
J. F.
,
Baumgartner
,
M. F.
, and
Gawarkiewicz
,
G. G.
(
2012
). “
Long distance passive localization of vocalizing sei whales using an acoustic normal mode approach
,”
J. Acoust. Soc. Am.
131
,
1814
1825
.
47.
Nieukirk
,
S. L.
,
Mellinger
,
D. K.
,
Dziak
,
R. P.
,
Matsumoto
,
H.
, and
Klinck
,
H.
(
2020
). “
Multi-year occurrence of sei whale calls in North Atlantic polar waters
,”
J. Acoust. Soc. Am.
147
,
1842
1850
.
48.
Oleson
,
E. M.
,
Wiggins
,
S. M.
, and
Hildebrand
,
J. A.
(
2007
). “
Temporal separation of blue whale call types on a southern California feeding ground
,”
Anim. Behav.
74
,
881
894
.
49.
Parks
,
S. E.
, and
Clark
,
C. W.
(
2007
). “
Acoustic communication: Social sounds and the potential impacts of noise
,” in
The Urban Whale: North Atlantic Right Whales at the Crossroads
, edited by
S. D.
Kraus
and
R. M.
Rolland
(
Harvard University Press
,
Cambridge, MA
), pp.
310
332
.
50.
Parks
,
S. E.
,
Clark
,
C. W.
, and
Tyack
,
P. L.
(
2007
). “
Short- and long-term changes in right whale calling behavior: The potential effects of noise on acoustic communication
,”
J. Acoust. Soc. Am.
122
,
3725
3731
.
51.
Parks
,
S. E.
,
Cusano
,
D. A.
,
Van Parijs
,
S. M.
, and
Nowacek
,
D. P.
(
2019
). “
North Atlantic right whale (Eubalaena glacialis) acoustic behavior on the calving grounds
,”
J. Acoust. Soc. Am.
146
,
EL15
EL21
.
52.
Parks
,
S. E.
,
Warren
,
J. D.
,
Stamieszkin
,
K.
,
Mayo
,
C. A.
, and
Wiley
,
D.
(
2012
). “
Dangerous dining: Surface foraging of North Atlantic right whales increases risk of vessel collisions
,”
Biol. Lett.
8
,
57
60
.
53.
Payne
,
R.
, and
Webb
,
D.
(
1971
). “
Orientation by means of long range acoustic signaling in baleen whales
,”
Ann. N.Y. Acad. Sci.
188
,
110
141
.
54.
Prieto
,
R.
,
Janiger
,
D.
,
Silva
,
M. A.
,
Waring
,
G. T.
, and
Goncalves
,
J. M.
(
2012
). “
The forgotten whale: A bibliometric analysis and literature review of the North Atlantic sei whale Balaenoptera borealis
,”
Mamm. Rev.
42
,
235
272
.
55.
Prieto
,
R.
,
Silva
,
M. A.
,
Waring
,
G. T.
, and
Gonçalves
,
J. M. A.
(
2014
). “
Sei whale movements and behaviour in the North Atlantic inferred from satellite telemetry
,”
Endang. Species Res.
26
,
103
113
.
56.
Rankin
,
S.
, and
Barlow
,
J. P.
(
2005
). “
Source of the North Pacific ‘boing’ sound attributed to minke whales
,”
J. Acoust. Soc. Am.
118
,
3346
3351
.
57.
Rankin
,
S.
, and
Barlow
,
J. P.
(
2007
). “
Vocalizations of the sei whale Balaenoptera borealis off the Hawaiian Islands
,”
Bioacoustics
16
,
137
145
.
58.
R Core Team
(
2020
). “
R: A Language and Environment for Statistical Computing [computer program]
,” R Foundation for Statistical Computing, http://www.R-project.org/.
59.
Richardson
,
W. J.
,
Greene
,
C. R.
, Jr.
,
Malme
,
C. I.
, and
Thomson
,
D. H.
(
1995
).
Marine Mammals and Noise
(
Academic Press
,
San Diego, CA
).
60.
Romagosa
,
M.
,
Baumgartner
,
M. F.
,
Cascão
,
I.
,
Lammers
,
M. O.
,
Marques
,
T. A.
,
Santos
,
R. S.
, and
Silva
,
M. A.
(
2020
). “
Baleen whale acoustic presence and behaviour at a Mid-Atlantic migratory habitat, the Azores Archipelago
,”
Sci. Rep.
10
,
4766
.
61.
Romagosa
,
M.
,
Boisseau
,
O.
,
Cucknell
,
A.-C.
,
Moscrop
,
A.
, and
McLanaghan
,
R.
(
2015
). “
Source level estimates for sei whale (Balaenoptera borealis) vocalizations off the Azores
,”
J. Acoust. Soc. Am.
138
,
2367
2372
.
62.
Rousseeuw
,
P. J.
, and
Kaufman
,
L.
(
2009
).
Finding Groups in Data: An Introduction to Cluster Analysis
(
John Wiley & Sons
,
Hoboken, NJ
).
63.
Saddler
,
M. R.
,
Bocconcelli
,
A.
,
Hickmott
,
L. S.
,
Chiang
,
G.
,
Landea-Briones
,
R.
,
Bahamonde
,
P. A.
,
Howes
,
G.
,
Segre
,
P. S.
, and
Sayigh
,
L. S.
(
2017
). “
Characterizing Chilean blue whale vocalizations with DTAGs: A test of using tag accelerometers for caller identification
,”
J. Exp. Biol.
220
,
4119
4129
.
64.
Samaran
,
F.
,
Stafford
,
K. M.
,
Branch
,
T. A.
,
Gedamke
,
J.
,
Royer
,
J.-Y.
,
Dziak
,
R. P.
, and
Guinet
,
C.
(
2013
). “
Seasonal and geographic variation of southern blue whale subspecies in the Indian Ocean
,”
PLoS One
8
,
e71561
.
65.
Scholkmann
,
F.
(
2019
). “
Exposure to high-frequency sound and ultrasound in public places: Examples from Zurich, Switzerland
,”
Acoustics
1
,
816
824
.
66.
Sibly
,
R.
,
Nott
,
H.
, and
Fletcher
,
D.
(
1990
). “
Splitting behaviour into bouts
,”
Anim. Behav.
39
,
63
69
.
67.
Stafford
,
K. M.
,
Nieukirk
,
S. L.
, and
Fox
,
C. G.
(
2001
). “
Geographic and seasonal variation of blue whale calls in the North Pacific
,”
J. Cetacean Res. Manag.
3
,
65
76
.
68.
Stimpert
,
A. K.
,
Au
,
W. W. L.
,
Parks
,
S. E.
,
Hurst
,
T.
, and
Wiley
,
D. N.
(
2011
). “
Common humpback whale (Megaptera novaeangliae) sound types for passive acoustic monitoring
,”
J. Acoust. Soc. Am.
129
,
476
482
.
69.
Stimpert
,
A. K.
,
Lammers
,
M. O.
,
Pack
,
A. A.
, and
Au
,
W. W. L.
(
2020
). “
Variations in received levels on a sound and movement tag on a singing humpback whale: Implications for caller identification
,”
J. Acoust. Soc. Am.
147
,
3684
3690
.
70.
Thompson
,
T. J.
,
Winn
,
H. E.
, and
Perkins
,
P. J.
(
1979
). “
Mysticete sounds
,” in
Behavior of Marine Animals
, edited by
H. E.
Winn
and
B. L.
Olla
(
Springer
,
Boston, MA
), pp.
403
431
.
71.
Tremblay
,
C. J.
,
Van Parijs
,
S. M.
, and
Cholewiak
,
D. M.
(
2019
). “
50 to 30-Hz triplet and singlet down sweep vocalizations produced by sei whales (Balaenoptera borealis) in the western North Atlantic Ocean
,”
J. Acoust. Soc. Am.
145
,
3351
3358
.
72.
Wiley
,
D. N.
,
Zadra
,
C. J.
,
Friedlaender
,
A. S.
,
Parks
,
S. E.
,
Pensarosa
,
A.
,
Rogan
,
A.
,
Shorter
,
K. A.
,
Urbán
,
J.
, and
Kerr
,
I.
(
2023
). “
Deployment of biologging tags on free swimming large whales using uncrewed aerial systems
,”
R. Soc. Open Sci.
10
,
221376
.

Supplementary Material