Passive acoustic monitoring has been an effective tool to study cetaceans in remote regions of the Arctic. Here, we advance methods to acoustically identify the only two Arctic toothed whales, the beluga (Delphinapterus leucas) and narwhal (Monodon monoceros), using echolocation clicks. Long-term acoustic recordings collected from moorings in Northwest Greenland were analyzed. Beluga and narwhal echolocation signals were distinguishable using spectrograms where beluga clicks had most energy >30 kHz and narwhal clicks had a sharp lower frequency limit near 20 kHz. Changes in one-third octave levels (TOL) between two pairs of one-third octave bands were compared from over one million click spectra. Narwhal clicks had a steep increase between the 16 and 25 kHz TOL bands that was absent in beluga click spectra. Conversely, beluga clicks had a steep increase between the 25 and 40 kHz TOL bands that was absent in narwhal click spectra. Random Forest classification models built using the 16 to 25 kHz and 25 to 40 kHz TOL ratios accurately predicted the species identity of 100% of acoustic events. Our findings support the use of echolocation TOL ratios in future automated click classifiers for acoustic monitoring of Arctic toothed whales and potentially for other odontocete species.

In the Arctic, passive acoustic monitoring enables observations of biotic and abiotic processes over spatiotemporal scales that would otherwise be unfeasible (e.g., Halliday , 2021a; Mattmüller , 2022; Stafford , 2022). Notably, observations of cetacean occurrence acquired from passive acoustic data provide key information about changes in habitat-use and timing of life history events, or phenology, of Arctic species due to climate change (Ahonen , 2021; Moore , 2022; Stafford , 2021). The Arctic has warmed at rates more than three times the global mean, and sea ice extent at its September minimum has declined by more than 12% per decade (relative to 1981–2010 mean; IPCC, 2022; Onarheim , 2018; Rantanen , 2022). Greater accessibility in Arctic waters has spurred increased human activities, such as commercial shipping and oil and gas exploration, that have heightened underwater noise levels and the risk of acoustic disturbance (e.g., behavioral changes, masking, or hearing damage) to marine mammals (Erbe , 2016; Finneran , 2002; Halliday , 2020; Southall , 2021). Studies using passive acoustics effectively track changes to natural and anthropogenic sounds, but they require knowledge of unique properties of specific sounds to accurately identify them in recordings. Given the efficacy and need for long-term passive acoustic monitoring in the Arctic, a firm understanding of acoustic parameters must be established for key species to ensure reliable acoustic classification.

Narwhals (Monodon monoceros) and belugas (Delphinapterus leucas) are ice-associated Arctic toothed whales (Monodontidae family). Narwhals inhabit waters in the Atlantic sector of the Arctic around Greenland, Canada, Svalbard, and Russia, while the beluga range is circumpolar and extends into subarctic territories (Hobbs , 2019; Innes , 2002; Reeves , 2014). Migratory belugas and narwhals follow the annual sea ice cycle where their autumn southward and spring northward movements are linked to sea ice advance and retreat, respectively. They return to the same summer and winter locations with high site fidelity, particularly the narwhal (Dietz , 2001; Heide-Jørgensen , 2003a; Innes , 2002). Phenological and distribution shifts have been documented for both species where animals are departing later during years with delayed sea ice freeze-up and expanding their range (Hauser , 2017; Heide-Jørgensen , 2010; Louis , 2020; Nielsen, 2009; Shuert , 2022). Further, narwhals have been shown to be extremely sensitive to ship noise and seismic airgun pulses (Heide-Jørgensen , 2021; Radtke , 2023; Tervo , 2023; Williams , 2022) and are considered one of the most sensitive Arctic marine mammals to climate change due to their preference for cold waters and specialization in habitat and prey (Heide-Jørgensen , 2020; Laidre , 2008). Belugas are also strongly affected by anthropogenic noise disturbance (Erbe and Farmer, 2000; Halliday , 2021b; Lesage , 1999) and some stocks have had significant declines from cumulative climate and anthropogenic stressors (Hobbs , 2019; Lesage, 2021).

Narwhals and belugas have been studied using short- and long-term passive acoustic methods (e.g., Ahonen , 2019; Au , 1985; Lammers , 2013; Marcoux , 2011; Møhl , 1990; Schevill and Lawrence, 1949). Both species are highly social with complex community structures, and accordingly, individuals produce a variety of vocalizations to interact with conspecifics and sense their environment. Communicative call types produced by these whales are broadly categorized into whistles, pulsed calls (or “burst pulses”), and combined calls that contain both pulsed and tonal sounds (Blackwell , 2018). Vocalizations produced for sensory tasks (e.g., navigation and prey localization) include echolocation clicks and terminal buzzes that are found at the end of click sequences (Blackwell , 2018; Castellote , 2021; Chambault , 2023; Roy , 2010). Pulsed calls and buzzes are defined by their short duration and high-repetition rate signals. Although there is a considerable body of literature studying their sounds, beluga and narwhal vocalizations remain partially described due to recording limitations and the diversity in their social calls within and between subpopulations (Garland , 2015; Marcoux , 2012; Sjare and Smith, 1986). For regions where belugas and narwhals overlap in West Greenland and the Canadian High Arctic, distinguishing between their vocalizations is essential for passive acoustic monitoring of these two species.

Scientific studies have historically estimated key acoustic parameters for each species individually (Au , 1985, 1987; Koblitz , 2016; Miller , 1995; Møhl , 1990; Rasmussen , 2015; Roy , 2010; Stafford , 2012; Zahn , 2021a). The first direct species comparison was done by Frouin-Mouy (2017), where they examined beluga and narwhal recordings and found narwhal clicks had a substantial increase in energy between the 16 and 20 kHz one-third octave bands which was lacking in beluga clicks. Jones (2022) presented differences in two key parameters (peak frequency and inter-click interval). It is now well established that beluga and narwhals produce high-frequency, broadband clicks from ∼20 kHz to well over 100 kHz (Koblitz , 2016; Rasmussen , 2015; Zahn , 2021a), and beluga clicks contain more energy at higher frequencies (>40 kHz) than narwhal clicks (Frouin-Mouy , 2017; Jones , 2022; Zahn , 2021b). Outside of descriptive statistics, Zahn (2021b) reported the first attempt to automatically classify beluga and narwhal clicks from parameter estimates using machine learning Random Forest (RF) models.

Here, we build on previous work and advance methods to classify beluga and narwhal echolocation clicks using manual and automated approaches on recordings from passive acoustic instruments mounted near glacier fronts in Northwest Greenland. Our primary objective was to investigate the predictive capacity of click parameters derived from one-third octave levels (TOL) for use in acoustic classification models. Our findings provide an important step in automating beluga and narwhal detection in long-term acoustic recordings using echolocation signals, and thus improve methods to monitor these species.

This study was conducted in Northwest Greenland and was part of a larger project to study the ecological importance of narwhal summering grounds in Melville Bay (Fig. 1). Melville Bay (Greenlandic: Qimusseriarsuaq) is found on the northwest Greenland continental shelf (74°N – 76.5°N) with a large trough that opens southwest into Baffin Bay. Both belugas and narwhals utilize regions along the eastern portion of Baffin Bay throughout the year. The Eastern High Arctic-Baffin Bay beluga stock occupies estuaries, bays, and inlets in the Canadian Arctic Archipelago during summer months (Hobbs , 2019). Over winter, a portion of the stock resides in the North Water polynya while the majority travel south during fall along the West Greenland coast remaining in mostly ice-free waters (Doidge and Finley, 1993; Heide-Jørgensen , 2003b; Richard , 1998b; Richard , 1998a). The Melville Bay narwhal stock spends summer months near glacier fronts in Melville Bay and migrates south to overwinter in the dense pack-ice of Baffin Bay and Davis Strait (Dietz , 2008; Dietz and Heide-Jørgensen, 1995; Laidre , 2016; Laidre and Heide-Jørgensen, 2011). Narwhals are temporary summer residents of Melville Bay. Belugas transit through the area during their fall and spring migrations.

FIG. 1.

(Color online) Map of the study region in Melville Bay, Northwest Greenland (a). The dashed outline in (a) demarcates the region (b), and the black box in (b) shows the region displayed in (c). Ocean mooring locations are marked with yellow diamonds (b), (c). Mooring positions near Kong Oscar glacier and in the Fisher Islands are shown on a Landsat 8 image converted to natural color from 18 August 2019 (c) courtesy of the U.S. Geological Survey. Ocean bathymetry (a), (b) is from the International Bathymetric Chart of the Arctic Ocean (Jakobsson , 2012).

FIG. 1.

(Color online) Map of the study region in Melville Bay, Northwest Greenland (a). The dashed outline in (a) demarcates the region (b), and the black box in (b) shows the region displayed in (c). Ocean mooring locations are marked with yellow diamonds (b), (c). Mooring positions near Kong Oscar glacier and in the Fisher Islands are shown on a Landsat 8 image converted to natural color from 18 August 2019 (c) courtesy of the U.S. Geological Survey. Ocean bathymetry (a), (b) is from the International Bathymetric Chart of the Arctic Ocean (Jakobsson , 2012).

Close modal

Between 2019 and 2020, three seafloor-mounted ocean moorings were deployed from the R/V Sanna (Greenland Institute of Natural Resources, Nuuk, Greenland) near glacier fronts in Melville Bay. Each mooring was equipped with a sound recorder and an array of oceanographic sensors to study the habitat of narwhals during summer. Instruments and floats were attached to a 6 mm Dyneema line, and an 800 kg anchor connected to an acoustic release (PORT LF, Edgetech Instruments, Hudson, MA) mounted the mooring to the seafloor. After deployment in August 2019, one mooring was destroyed by icebergs and two were successfully retrieved in August 2020. These two moorings were positioned in the Fisher Islands and in front of Kong Oscar glacier, also known as Nuussuup Sermia (Fig. 1). We refer to these mooring locations as the Fisher Islands and Kong Oscar sites.

Moorings included SoundTrap ST500 STD (Ocean Instruments, New Zealand) acoustic recorders (Table I). The SoundTrap instruments sampled at 144 kHz (16-bit) with a Nyquist frequency of 72 kHz to record beluga and narwhal echolocation clicks that contain most of their energy above 20 kHz (Frouin-Mouy , 2017; Jones , 2022; Koblitz , 2016; Zahn , 2021a). Specifications of the SoundTrap ST500 STD recorders included a system peak clipping level of 173 dB re μPa, bandwidth from 20 Hz to 60 kHz (±3 dB), and self-noise <36 dB re 1 μPa above 2 kHz and better than sea-state zero from 100 Hz to 2 kHz. Acoustic data were logged 40 min per hour. This duty cycle was selected to ensure the recorders maximized recording coverage each hour while retaining sufficient battery to last into winter, well after the area was covered in fast ice and whales had departed.

TABLE I.

Summary of acoustic data collected from the moored SoundTrap ST500 STD recorders in Melville Bay.

Site Recording period Latitude (°N) Longitude (°W) Seafloor depth (m) Sampling rate (kHz) Duty cycle Instrument depth (m)
Fisher Islands  5 Aug 2019–20 Dec 2019  76.1038  61.7270  370  144  40 min/h  194 
Kong Oscar  5 Aug 2019– 16 Jan 2020  75.8418  59.8431  250  144  40 min/h  158 
Site Recording period Latitude (°N) Longitude (°W) Seafloor depth (m) Sampling rate (kHz) Duty cycle Instrument depth (m)
Fisher Islands  5 Aug 2019–20 Dec 2019  76.1038  61.7270  370  144  40 min/h  194 
Kong Oscar  5 Aug 2019– 16 Jan 2020  75.8418  59.8431  250  144  40 min/h  158 

Raw acoustic data were processed using the open-source passive acoustic monitoring software PAMGuard (version 2.02.09; Gillespie , 2009). Echolocation clicks were detected using PAMGuard's Click Detector when the amplitude of a transient signal exceeded 14 dB above the measured background noise. Subsets of the SoundTrap data were processed to adjust filter settings and determine an effective approach to reduce false detections (i.e., noise) while maximizing the detection of echolocation clicks. A 4-pole Butterworth 1 kHz high pass filter (“digital pre filter” in PAMGuard) was applied to remove a low frequency noise peak. Then, a 4-pole Butterworth 20 kHz high pass filter (“digital trigger filter”) was applied for click detection. Signal waveforms from click detections were labeled and extracted from the 1 kHz high pass filtered data for subsequent analysis.

Following the click detection stage, sound clips were labeled according to their peak frequency from power spectra within PAMGuard. Each classifier, or detector, in PAMGuard categorized clicks with a unique numeric code (i.e., detector number) when their peak frequencies fell within a specified frequency range: 1 (4–20 kHz), 2 (20–50 kHz), and 3 (50–70 kHz). Initial processing of the data revealed that detector 1 included only false detections (i.e., non-echolocation sounds), so only detectors 2 and 3 were enabled for downstream analyses. Further inspection of false detections showed that power spectra from transient noise signals decreased in energy with increasing frequency. Therefore, to further reduce abiotic transient sounds (e.g., ice or glacier calving), an additional constraint was added to detector 2, whereby clicks were only labeled as echolocation clicks when there was a >5 dB difference in energy between 10–20 kHz and 30–40 kHz. This additional constraint was not applied to detector 3, because there were no significant abiotic transients detected between 50 and 70 kHz. Unclassified detections (detector 0) were discarded.

Detections identified and labeled as clicks within PAMGuard were then processed into hourly acoustic events using the PAMpal package (version 1.0.4; Sakai, 2023) in R (version 4.3.1; R Core Team, 2023). Detections were binned into 1-h long acoustic events with unique identification numbers. Each event contained all the detections for one or the other species that occurred within a given hour from the start to the end date of the recording period for each site (Table I). Events that had fewer than 30 detections were found to generally contain only false detections and were removed.

Concatenated click spectrograms and mean power spectra for all acoustic events were manually examined to identify each event's source and classify them as beluga or narwhal. Events that contained only noise from crackling sea ice, glacier ice, and calving were removed from subsequent classification analyses. Beluga and narwhal events were identified based on differences in the energy distribution of each species' clicks in the frequency domain. Belugas produce higher frequency clicks compared to those generated by narwhals as demonstrated by estimated peak frequencies being consistently >20 kHz higher in beluga clicks than narwhal clicks (Jones , 2022; Koblitz , 2016; Zahn , 2021a; Zahn , 2021b). This is partly because narwhal clicks have considerably more energy between 20 and 30 kHz than beluga clicks (Frouin-Mouy , 2017; Jones , 2022). Beluga clicks also do not contain a low frequency component that is sometimes observed in narwhal echolocation between 3 and 14 kHz (Stafford , 2012). For events where species designation was in doubt from concatenated click spectrograms or mean spectra, raw sound file spectrograms were manually audited in Raven Pro (version 1.6.5; K. Lisa Yang Center for Conservation Bioacoustics at the Cornell Lab of Ornithology, 2023). Additionally, the presence of whistles aided manual species designation. Marcoux (2012) describe beluga whistles as having more diversified contours than those produced by narwhals (e.g., Belikov and Bel'Kovich, 2007; Sjare and Smith, 1986). The presence of highly variable whistles below 20 kHz along with high frequency echolocation (click energy mainly >30 kHz) supported beluga species assignment. Finally, manually classified events were compared with an independently labeled dataset at a 6-h resolution produced by a different human analyst using the original spectrogram data to ensure no whale detections were missed or incorrectly classified.

Upon completion of the manual species classification for all acoustic events, we built upon methods developed by Frouin-Mouy (2017) and calculated the TOL difference between two sets of one-third octave bands in R (version 4.3.1; R Core Team, 2023). One-third octave band edge frequencies were calculated using the American National Standards Institute (ANSI) standard base-ten center frequencies also known as decidecade bands (ANSI, 2004). Three one-third octave bands were selected that best represented differences in spectral energy within the recording frequency bandwidth for the two species: 16 kHz (14 125–17 783 Hz), 25 kHz (22 387–28 184 Hz), and 40 kHz (35 481–44 668 Hz). We refer to these one-third octave bands by their nominal center frequencies throughout in lieu of the actual frequency range for simplicity. In this study, we considered whether the TOL in narwhal clicks consistently increased between 16 to 25 kHz with little difference between 25 to 40 kHz. Conversely, beluga clicks were expected to have little difference between 16 to 25 kHz and a large TOL difference between 25 to 40 kHz.

For TOL calculations, 1 kHz high pass filtered waveforms extracted from PAMGuard were analyzed, and mean power spectra were produced for each acoustic event (512 pt fast Fourier transform, FFT). The TOL was calculated by summing the squared pressures within the upper and lower one-third octave limits from mean spectra. Then the difference was computed between the (1) 16 and 25 kHz and (2) 25 and 40 kHz one-third octave bands. The result was two TOL ratios with potential use in beluga and narwhal click classification. Hereafter we refer to these two metrics as the 16 to 25 kHz TOL ratio and 25 to 40 kHz TOL ratio.

Manual species identification of SoundTrap recordings from the Fisher Islands and Kong Oscar sites yielded two labeled datasets of hourly acoustic events. We evaluated the predictive strength of the TOL ratio metrics for species classification by building RF models using the 16 to 25 kHz and 25 to 40 kHz TOL ratios as predictors. Three RF models were built using different training datasets that included TOL ratio estimates from: (1) Fisher Islands and Kong Oscar (pooled), (2) Fisher Islands, and (3) Kong Oscar. Additionally, the Fisher Islands and Kong Oscar datasets functioned as complimentary training and testing datasets. Thus, site-specific RF models were tested with a dataset different from the one used to train the model. The Fisher Islands RF model was used to predict Kong Oscar acoustic events and the Kong Oscar RF model was used to predict Fisher Islands acoustic events.

RF models have three primary parameters that can be adjusted: mtry defines the number of randomly selected variables used to split observations at each node, sampsize indicates the number of observations (i.e., acoustic events) that are randomly subset to build each tree, and ntree specifies how many individual trees are built in the forest. Since we used two variables for classification (16 to 25 kHz and 25 to 40 kHz TOL ratios), only one variable could be randomly selected so the only possible value for mtry was one. RF models were configured such that equal subsamples (i.e., sampsize) were drawn from each species group without replacement to account for unbalanced datasets and prevent overfitting to the majority class. To evaluate model sensitivity to sampsize, each model was fit over all possible values of sampsize, and model accuracies were determined. Possible values of sampsize ranged from 2 to (N − 1) where N is the total sample size of the smallest species class. For the RF models developed, the accuracies varied by less than 0.01% and therefore were not sensitive to permutations in sampsize. As in Zahn (2021b), we report model results when sampsize was equal to half the total sample size for the smallest (i.e., minority) species class. Models were constructed with ten thousand trees (ntree).

RF models use bootstrap sampling to build thousands of decision trees to produce a final ensemble tree. Random subsets drawn and used as training data are referred to as “in-bag” samples and the remaining observations are called “out-of-bag” (OOB) samples. OOB samples are used as testing data to estimate model prediction errors and accuracy when RF models are built. Therefore, model performance of trained RF models was evaluated using OOB correct classification scores. Additionally, model stability was confirmed through plotting the trace of cumulative OOB error rates in the forest. Consistent performance was observed across ten thousand trees. The accuracy of trained RF models that were tested with new data was determined from the percentage of acoustic events that were correctly classified by the model.

In RF classification models, observations (i.e., acoustic events) are assigned to a class (i.e., species) based on the percentage of trees in the forest voting for each species. For all models, the uncertainty of model predictions was evaluated by examining the distribution of votes for each acoustic event within a species class. Acoustic events were predicted with high model certainty when a majority of the trees voted for the correct species class. All acoustic classification model development and evaluation of RF models were conducted using the randomForest (version 4.7-1.1; Liaw and Wiener, 2002) and rfPermute (version 2.5.1; Archer, 2021) packages in R.

SoundTrap instruments recorded for 138 days (∼4.5 months; 5 August–20 December 2019) at the Fisher Islands site and 165 days (∼5.5 months; 5 August 2019–16 January 2020) at the Kong Oscar site. With a duty cycle of 40 min per hour, recordings totaled approximately 4848 h (∼6.7 months) across sites. The PAMGuard Click Detector identified echolocation clicks including buzzes as well as non-echolocation high repetition rate calls, referred to in the literature as burst pulses, or pulsed calls (Fig. 2). The low frequency component of narwhal clicks was visible between 3 and 5 kHz in some concatenated click spectrograms, possibly when whales were close to the moorings and the signal-to-noise ratio was high. During manual inspection of echolocation events using Raven Pro, lower frequency (<20 kHz) beluga whistles were observed in some events (Fig. 2). However, echolocation clicks remained the dominant call type produced by both species over the recording period.

FIG. 2.

(Color online) Selected spectrograms showing beluga (a), (b) and narwhal (c), (d) echolocation clicks with lighter and darker colors indicating higher and lower sound levels, respectively. All samples were recorded at the Fisher Islands site. Panels (a) and (b) also include beluga whistles in the 5–10 kHz range and (b) shows a few beluga pulsed calls. Panels (c) and (d) show narwhal burst pulses and echolocation click trains. Spectrograms were produced using a Hanning window, 50% overlap, and 512 point FFT for (a), (c) and 1024 point FFT for (b), (d) in Raven Pro 1.6.5. Note: the duration of the spectrogram (in seconds) differs between subplots with one short (3 s) and one long (20 s) duration example for each species.

FIG. 2.

(Color online) Selected spectrograms showing beluga (a), (b) and narwhal (c), (d) echolocation clicks with lighter and darker colors indicating higher and lower sound levels, respectively. All samples were recorded at the Fisher Islands site. Panels (a) and (b) also include beluga whistles in the 5–10 kHz range and (b) shows a few beluga pulsed calls. Panels (c) and (d) show narwhal burst pulses and echolocation click trains. Spectrograms were produced using a Hanning window, 50% overlap, and 512 point FFT for (a), (c) and 1024 point FFT for (b), (d) in Raven Pro 1.6.5. Note: the duration of the spectrogram (in seconds) differs between subplots with one short (3 s) and one long (20 s) duration example for each species.

Close modal

Manual species identification of hourly acoustic event data revealed a clear visible difference between beluga and narwhal echolocation. Beluga clicks contained energy above 30 kHz with peak intensities typically occurring between 40 and 60 kHz (Fig. 3). We note that true spectral peaks occurring in the 60–72 kHz range could not be identified due to the SoundTrap anti-aliasing filter having a −3 dB cut-off frequency at 64.8 kHz. Narwhal clicks consistently presented a relatively sharp low-frequency cut-off around 20 kHz (Fig. 3) including signals produced during burst pulses and buzzes (Fig. 2). Although the exact frequency of this lower limit varied between approximately 18 and 23 kHz, the presence of a steep increase in energy around 20 kHz provided an explicit method to distinguish narwhal clicks from those of beluga. A full catalogue of labeled beluga and narwhal acoustic event concatenated click spectrograms and mean spectra showed clear differences between species (Zahn , 2023).

FIG. 3.

(Color online) Concatenated click spectrograms (a), (c) and mean power spectra (b), (d) for example beluga (a), (b) and narwhal (c), (d) acoustic events. The number of clicks included in each concatenated click spectrogram is provided on the x-axis (a), (c). Spectrograms and spectra were produced using a 512 point FFT and Hanning window from 1 kHz high pass waveforms.

FIG. 3.

(Color online) Concatenated click spectrograms (a), (c) and mean power spectra (b), (d) for example beluga (a), (b) and narwhal (c), (d) acoustic events. The number of clicks included in each concatenated click spectrogram is provided on the x-axis (a), (c). Spectrograms and spectra were produced using a 512 point FFT and Hanning window from 1 kHz high pass waveforms.

Close modal

Belugas and narwhals were detected at both the Fisher Islands and Kong Oscar sites, however, the majority of detections were, by far, narwhal (92%). No mixed species events were discovered. From the more than six months of acoustic data, which was divided into 1-h time bins, a total of 201 h contained narwhal and beluga clicks. From PAMGuard's click detectors 2 and 3 (i.e., clicks with peak frequencies between 20–50 and 50–70 kHz, respectively), there were more detections in total, and for each species, at the Fisher Islands site (117 events total; 776 679 detections) than the Kong Oscar site (84 events total; 241 920 detections). At the Fisher Islands, 12 events (141 098 detections) were beluga and 105 were narwhal (635 581 detections). At the Kong Oscar site, four events were beluga (1653 detections) and 80 events were narwhal (240 267 detections). Narwhals were present in Melville Bay from the start of the recording period in early August until mid-November when fall sea ice formation began (Fig. 4). Belugas were detected only during the month of October (Fig. 4).

FIG. 4.

(Color online) Time series of narwhal (a), (c) and beluga (b), (d) presence and percent sea ice cover (blue lines) at the Fisher Islands (a), (b) and Kong Oscar (c), (d) mooring sites from May 2019 to March 2020. Black bars provide the number of hours per week that narwhals and belugas were detected and thus indicate when whales were present. Gray shaded regions denote when no acoustic data were available. Hourly sea ice concentration data on the secondary (right side) y-axis were sourced from the ERA5 reanalysis product (Hersbach , 2023) and were averaged to a daily resolution.

FIG. 4.

(Color online) Time series of narwhal (a), (c) and beluga (b), (d) presence and percent sea ice cover (blue lines) at the Fisher Islands (a), (b) and Kong Oscar (c), (d) mooring sites from May 2019 to March 2020. Black bars provide the number of hours per week that narwhals and belugas were detected and thus indicate when whales were present. Gray shaded regions denote when no acoustic data were available. Hourly sea ice concentration data on the secondary (right side) y-axis were sourced from the ERA5 reanalysis product (Hersbach , 2023) and were averaged to a daily resolution.

Close modal

Consistent trends between selected TOL were observed between beluga and narwhal acoustic events. Beluga echolocation power spectra had little change in TOL between the 16 and 25 kHz bands in contrast to narwhal power spectra that showed a large increase in magnitude between these bands (Fig. 5). Conversely, a large difference was seen between the 25 and 40 kHz TOL in beluga spectra that was absent in narwhal spectra (Fig. 5). The mean 16 to 25 kHz TOL ratio was 2.69 dB (median: 2.91 dB) in beluga spectra and −7.07 dB (median: −7.03 dB) in narwhal spectra. The mean 25 to 40 kHz TOL ratio was −8.74 dB (median: −10.3 dB) in beluga spectra and 1.35 dB (median: 0.62 dB) in narwhal spectra. All beluga events had a decrease in TOL between the 16 and 25 kHz bands, corresponding to the decreasing noise level with increasing frequency. The difference between TOL ratios (i.e., 16 to 25 kHz TOL ratio minus the 25 to 40 kHz TOL ratio) was positive for beluga spectra and negative for narwhal spectra (Fig. 5).

FIG. 5.

(Color online) Mean power spectra for beluga (a) and narwhal (b) acoustic events. Mean spectra for individual acoustic events are shown in gray, and the overall mean spectra across events are shown in black with the noise floor provided as a dashed line. The average noise spectrum was computed using samples extracted from waveform clips that immediately preceded each click. Primary y axes in (a) and (b) provide normalized power (dB) with respect to the maximum and the secondary y axes show corresponding power spectral density magnitudes (dB re 1 μPa2/Hz). Blue shaded areas in (a) and (b) demarcate the frequency bandwidths of the 16, 25, and 40 kHz one-third octave bands. Violin plots with inset box plots show the distribution of the TOL ratios (dB) between the 16 and 25 kHz and 25 and 40 kHz one-third octave bands and the difference between these two ratios (i.e., 16:25 kHz TOL ratio subtracted from 25:40 kHz TOL ratio) for beluga (c) and narwhal (d) acoustic events.

FIG. 5.

(Color online) Mean power spectra for beluga (a) and narwhal (b) acoustic events. Mean spectra for individual acoustic events are shown in gray, and the overall mean spectra across events are shown in black with the noise floor provided as a dashed line. The average noise spectrum was computed using samples extracted from waveform clips that immediately preceded each click. Primary y axes in (a) and (b) provide normalized power (dB) with respect to the maximum and the secondary y axes show corresponding power spectral density magnitudes (dB re 1 μPa2/Hz). Blue shaded areas in (a) and (b) demarcate the frequency bandwidths of the 16, 25, and 40 kHz one-third octave bands. Violin plots with inset box plots show the distribution of the TOL ratios (dB) between the 16 and 25 kHz and 25 and 40 kHz one-third octave bands and the difference between these two ratios (i.e., 16:25 kHz TOL ratio subtracted from 25:40 kHz TOL ratio) for beluga (c) and narwhal (d) acoustic events.

Close modal

The three RF models trained with different datasets (1, both sites; 2, the Fisher Islands site; and 3, the Kong Oscar site) all had high OOB correct classification rates (>99%; Table II). All acoustic events were correctly assigned in all models except for one narwhal event misclassification from the Fisher Islands. For all RF models developed, variable importance scores indicated the 16 to 25 kHz TOL ratio was the more important variable for correct species classification. However, exploratory runs of RF models with only the 16 to 25 kHz TOL ratio variable resulted in decreased accuracy, and thus the 25 to 40 kHz TOL ratio variable was also important for correct predictions. Visualizing the distribution of RF model votes confirmed high confidence in OOB predictions for all RF models developed (see supplementary material, Fig. S1).

TABLE II.

RF confusion matrices for model development using the TOL ratios between the 16 to 25 kHz and 25 to 40 kHz one-third octave bands. Three RF models were trained using acoustic events from separate datasets: (1) both sites, (2) Fisher Islands site, and (3) Kong Oscar site. Rows indicate the original acoustic event species assignment and columns show predictions by the classifier. OOB percent correct classification rates and percent error (95% confidence interval) indicate model accuracy.

Beluga Narwhal OOB accuracy [95% confidence interval (CI)] Error (95% CI)
Both sites   
Beluga  16  100% (79.4%–100%)  0% (0%–60.2%) 
Narwhal  184  99.5% (97.0%–100%)  0.5% (0%–3.0%) 
Overall      99.5% (97.3%–100%)  0.5% (0%–2.7%) 
Fisher Islands 
Beluga  12  100% (73.5%–100%)  0% (0%–60.2%) 
Narwhal  104  99.0% (94.8%–100%)  1.0% (5.2%–4.5%) 
Overall      99.1% (95.3%–100%)  0.9% (4.7%–4.3%) 
Kong Oscar 
Beluga  100% (39.8%–100%)  0% (0%–60.2%) 
Narwhal  80  100% (95.5%–100%)  0% (0%–4.5%) 
Overall      100% (95.7%–100%)  0% (0%–4.3%) 
Beluga Narwhal OOB accuracy [95% confidence interval (CI)] Error (95% CI)
Both sites   
Beluga  16  100% (79.4%–100%)  0% (0%–60.2%) 
Narwhal  184  99.5% (97.0%–100%)  0.5% (0%–3.0%) 
Overall      99.5% (97.3%–100%)  0.5% (0%–2.7%) 
Fisher Islands 
Beluga  12  100% (73.5%–100%)  0% (0%–60.2%) 
Narwhal  104  99.0% (94.8%–100%)  1.0% (5.2%–4.5%) 
Overall      99.1% (95.3%–100%)  0.9% (4.7%–4.3%) 
Kong Oscar 
Beluga  100% (39.8%–100%)  0% (0%–60.2%) 
Narwhal  80  100% (95.5%–100%)  0% (0%–4.5%) 
Overall      100% (95.7%–100%)  0% (0%–4.3%) 

Site-specific RF models built with TOL ratio variables predicted the species identity of acoustic events with extremely high accuracy (100%; Table III). The Fisher Islands RF model correctly predicted all of the acoustic events from the Kong Oscar dataset. Similarly, the Kong Oscar RF model correctly predicted all of the acoustic events from the Fisher Islands dataset. Visualizing the distribution of RF model votes for each model confirmed high confidence in predictions (see supplementary material, Fig. S2).

TABLE III.

RF confusion matrices for model predictions using the TOL ratios between the 16 to 25 kHz and 25 to 40 kHz bands. Individual RF models built for each site were tested with data from the other site to evaluate model performance. Rows show the original acoustic event species assignment and columns indicate classifier predictions. Model accuracy or percentage of total acoustic events correctly classified (95% confidence interval) and percent error are provided.

Beluga Narwhal Accuracy (95% CI) Error (95% CI)
Fisher Islands predicts Kong Oscar 
Beluga  100% (39.8%–100%)  0% (0%–60.2%) 
Narwhal  80  100% (95.5%–100%)  0% (0%–4.5%) 
Overall      100% (95.7%–100%)  0% (0%–4.3%) 
Kong Oscar predicts Fisher Islands 
Beluga  12  100% (73.5%–100%)  0% (0%–26.5%) 
Narwhal  105  100% (96.5%–100%)  0% (0%–3.5%) 
Overall      100% (96.9%–100%)  0% (0%–3.1%) 
Beluga Narwhal Accuracy (95% CI) Error (95% CI)
Fisher Islands predicts Kong Oscar 
Beluga  100% (39.8%–100%)  0% (0%–60.2%) 
Narwhal  80  100% (95.5%–100%)  0% (0%–4.5%) 
Overall      100% (95.7%–100%)  0% (0%–4.3%) 
Kong Oscar predicts Fisher Islands 
Beluga  12  100% (73.5%–100%)  0% (0%–26.5%) 
Narwhal  105  100% (96.5%–100%)  0% (0%–3.5%) 
Overall      100% (96.9%–100%)  0% (0%–3.1%) 

Our results provide compelling evidence for identifying beluga and narwhal clicks with high certainty. We demonstrated the predictive strength of two new acoustic classification parameters for echolocation signals derived from one-third octave frequency bands. The differences in TOL between the 16 to 25 and 25 to 40 kHz one-third octave bands proved to be robust metrics for automated and manual species identification. Our TOL classification results suggest modest sampling rates (≥96 kHz) are needed to identify beluga and narwhal clicks which may enable longer duty cycling or extended deployments in remote areas of the Arctic. This study builds on existing literature (e.g., Frouin-Mouy , 2017; Jones , 2022) and together indicate that among the diverse vocalizations produced by Arctic toothed whales, echolocation clicks are dependable signals for beluga and narwhal acoustic classification.

We show beluga and narwhal clicks can reliably be identified through visual inspection of spectrograms. Across all acoustic events examined here, narwhal clicks contained a distinct spectral peak just above 20 kHz where spectral energy sharply decreased below 20 kHz and remained relatively flat between 20 and 60 kHz. The frequency peak near 20 kHz is consistent with existing literature documenting narwhal spectra (Frouin-Mouy , 2017; Jones , 2022; Koblitz , 2016; Møhl , 1990; Stafford , 2012; Zahn , 2021b). Pulsed calls and buzzes produced by narwhals and detected by the PAMGuard Click Detector also contained increased energy near 20 kHz. Only in lower ambient noise conditions were the low frequency components of narwhal clicks between 3 and 5 kHz observed. In contrast to narwhal clicks, beluga click spectra mostly had energy above 30 kHz. Unlike the sharp 20 kHz lower limit seen in narwhal spectra, spectral energy in beluga clicks increased more gradually from 30 kHz and peaked between 40 and 50 kHz, which is potentially a consequence of a 72 kHz Nyquist frequency and an anti-aliasing filter attenuating energy above ∼60 kHz. Beluga pulsed calls that were detected in our analyses contained energy from 20 kHz and above, but they did not present a clear 20 kHz edge like those produced by narwhals.

These results corroborate findings by Jones (2022) where beluga and narwhal clicks from moored High-frequency Acoustic Recording Packages (HARPs) with a 200 kHz sampling rate were analyzed. Jones (2022) found beluga clicks had higher peak frequencies (55–60 kHz) than narwhal clicks (23 kHz) at lower received levels (<130 dB peak-peak). At high received levels (>150 dB peak-peak), both species had peak frequencies between 50 and 60 kHz. Broadly, Jones (2022) summarize that narwhal clicks had more energy than beluga clicks below 40 kHz. Yet, spectral peaks vary depending on the orientation of the whale relative to the receiver and the specifications of the recorder. For example, peak frequencies of on-axis clicks were higher for both species, estimated to be 71 ± 15 kHz for narwhals (Koblitz , 2016) and 97 ± 7 kHz for belugas (Zahn , 2021a). Nonetheless, existing literature supports the assertion that belugas produce clicks with greater energy at higher frequencies compared to narwhals.

RF models had strong classification scores for models trained and tested with TOL statistics. Our results confirm that narwhal clicks had a larger difference between the 16 and 25 kHz one-third octave bands compared to belugas, whereas beluga clicks had a larger difference between the 25 and 40 kHz bands compared to narwhals. OOB correct classification rates were high for all RF models developed (>99%; Table II). Similarly, RF models trained with data from one site (i.e., Kong Oscar or Fisher Islands) correctly predicted 100% of the acoustic events from the other site (Table III). Since the beluga acoustic event sample size was much smaller than the narwhal species class, beluga event predictions had greater uncertainty. We expect that correct prediction rates would remain high and confidence intervals would narrow if there were a larger beluga event sample size since the model would have more data for species differentiation.

In summary, TOL ratios were strong predictors for acoustic classification of Arctic toothed whale echolocation signals. While echolocation parameters, such as peak frequency, will vary depending on the sampling rate and frequency response of the recording system, the 16 to 25 and 25 to 40 kHz TOL ratios will be unaffected as long as the sampling rate is at least 96 kHz with a flat frequency response between 14 and 45 kHz. However, ambient noise levels will influence 16 and 25 kHz TOL estimates where higher noise levels will decrease the relative magnitude within these bands. Still, the TOL ratios between the 16 to 25 and 25 to 40 kHz bands were sufficiently large that beluga and narwhal clicks remained differentiable across two locations. Importantly, using TOL ratios for species classification may not just be useful for belugas and narwhals. Employing TOL ratios for acoustic classification can be applied to other odontocete species using selected one-third octave bands that best capture click spectra variability.

The present study marks an important step in automating beluga and narwhal acoustic detection for future passive acoustic monitoring programs. Based on the results presented herein, we highlight that the lowermost part of echolocation click spectra (<50 kHz) for narwhals and belugas contain enough critical information for species classification. Therefore, we recommend a minimum sampling rate of 96 kHz for the classification of these two species, although recorders with higher sampling rates (∼400 kHz or more) are needed for general signal characterization involving parameters such as peak frequency and centroid frequency. Given the sensory function of echolocation, acoustic properties of clicks appear to be less variable than social calls, making them consistent metrics for species classification across subpopulations. To fully implement an auto-detector and classifier for recordings near glaciated coastlines, future work must describe and classify transient ice sounds to isolate biotic and abiotic sounds. While this was outside of the scope of the present study, we show highly accurate automated species predictions once cetacean clicks are separated.

Arctic ecosystems are being altered by climate change and trans-Arctic shipping routes will be used with greater frequency in the next decade (Lannuzel , 2020; Meier , 2014; Melia , 2016). Efforts to monitor changes to ambient noise levels and endemic Arctic odontocete distributions are becoming increasingly important, especially for the effective management of beluga and narwhal stocks for harvest by Arctic communities in Canada and Greenland (Hobbs , 2019). We verified that beluga and narwhal clicks are differentiable, and moreover, discrete parameters exist to automatically classify them at high success rates. With the increasing prevalence of autonomous recorders being used to monitor cetaceans globally, our results are directly applicable to future passive acoustics research. Going forward, we encourage sustained observations using long-term passive acoustics from fixed platforms (e.g., moored HARP or SoundTrap) to track species occurrence of resident and non-resident species and monitor increased human activity in the Arctic.

See the supplementary material for additional information. A catalogue of concatenated click spectrograms and mean power spectra for all acoustic event data from this study is available on Zenodo at https://doi.org/10.5281/zenodo.10076260 (Zahn et al., 2023).

This work was funded by the U.S. Office of Naval Research (Award No. N00014-17-1-2774) and supported by the NASA Oceans Melting Greenland EVS-2 mission. M.J.Z. was partially supported by the Cooperative Institute for Climate, Ocean, & Ecosystem Studies (CICOES) under NOAA Cooperative Agreement NA20OAR4320271, Contribution No. 2023-1312. M.S. and M.L. were supported by the Danish Cooperation for Environment in the Arctic (DANCEA, MST-2020-64692). We thank everyone who contributed to the data collection. We are also grateful to Jennifer L. K. McCullough for her assistance with PAMGuard, Ben Cohen for Landsat 8 imagery support, and Alex Douglass for his statistical advice. We thank Shannon Rankin for her helpful input on previous versions of this manuscript.

The authors have no conflicts of interest to disclose.

Acoustic data from moorings is available upon request from authors. Sea ice concentration ERA5 hourly data are supplied by the Copernicus Climate Change Service Climate Data Store at https://doi.org/10.24381/cds.adbb2d47. R and Python code used to produce all analyses including classification models and figures for this manuscript are publicly available on Zenodo: https://doi.org/10.5281/zenodo.10668629.

1.
Ahonen
,
H.
,
Stafford
,
K. M.
,
Lydersen
,
C.
,
Berchok
,
C. L.
,
Moore
,
S. E.
, and
Kovacs
,
K. M.
(
2021
). “
Inter-annual variability in acoustic detection of blue and fin whale calls in the Northeast Atlantic High Arctic between 2008 and 2018
,”
Endang. Species Res.
45
,
209
224
.
2.
Ahonen
,
H.
,
Stafford
,
K. M.
,
Lydersen
,
C.
,
Steur
,
L. D.
, and
Kovacs
,
K. M.
(
2019
). “
A multi-year study of narwhal occurrence in the western Fram Strait–detected via passive acoustic monitoring
,”
Polar Res.
38
,
3468
.
3.
American National Standards Institute
(
2004
).
ANSI S1.11-2004, Specification for Octave-Band and Fractional-Octave-Band Analog and Digital Filters
(
ANSI
,
New York
).
4.
Archer
,
E.
(
2021
). “
rfPermute: Estimate permutation p-values for Random Forest importance metrics (v2.5)
.”
5.
Au
,
W. W. L.
,
Penner
,
R. H.
,
Carder
,
D. A.
, and
Scronce
,
B.
(
1985
). “
Demonstration of adaptation in beluga whale echolocation signals
,”
J. Acoust. Soc. Am.
77
,
726
730
.
6.
Au
,
W. W. L.
,
Penner
,
R. H.
, and
Turl
,
C. W.
(
1987
). “
Propagation of beluga echolocation signals
,”
J. Acoust. Soc. Am.
82
,
807
813
.
7.
Belikov
,
R. A.
, and
Bel'Kovich
,
V. M.
(
2007
). “
Whistles of beluga whales in the reproductive gathering off Solovetskii Island in the White Sea
,”
Acoust. Phys.
53
,
528
534
.
8.
Blackwell
,
S. B.
,
Tervo
,
O. M.
,
Conrad
,
A. S.
,
Sinding
,
M. H. S.
,
Hansen
,
R. G.
,
Ditlevsen
,
S.
, and
Heide-Jørgensen
,
M. P.
(
2018
). “
Spatial and temporal patterns of sound production in East Greenland narwhals
,”
PLoS One
13
,
e0198295
.
9.
Castellote
,
M.
,
Mooney
,
A.
,
Andrews
,
R.
,
Deruiter
,
S.
,
Lee
,
W. J.
,
Ferguson
,
M.
, and
Wade
,
P.
(
2021
). “
Beluga whale (Delphinapterus leucas) acoustic foraging behavior and applications for long term monitoring
,”
PLoS One
16
,
e0260485
.
10.
Chambault
,
P.
,
Blackwell
,
S. B.
, and
Heide-Jørgensen
,
M. P.
(
2023
). “
Extremely low seasonal prey capture efficiency in a deep-diving whale, the narwhal
,”
Biol. Lett.
19
,
20220423
.
11.
Dietz
,
R.
, and
Heide-Jørgensen
,
M. P.
(
1995
). “
Movements and swimming speed of narwhals, Monodon monoceros, equipped with satellite transmitters in Melville Bay, northwest Greenland
,”
Can. J. Zool.
73
,
2106
2119
.
12.
Dietz
,
R.
,
Heide-Jørgensen
,
M. P.
,
Richard
,
P. R.
, and
Acquarone
,
M.
(
2001
). “
Summer and fall movements of narwhals (Monodon monoceros) from northeastern Baffin Island towards northern Davis Strait
,”
Arctic
54
,
244
261
.
13.
Dietz
,
R.
,
Heide-Jørgensen
,
M. P.
,
Richard
,
P.
,
Orr
,
J.
,
Laidre
,
K.
, and
Schmidt
,
H. C.
(
2008
). “
Movements of narwhals (Monodon monoceros) from Admiralty Inlet monitored by satellite telemetry
,”
Polar Biol.
31
,
1295
1306
.
14.
Doidge
,
D. W.
, and
Finley
,
K. J.
(
1993
). “
Status of the Baffin Bay population of beluga, Delphinapterus leucas
,”
Can. Field-Nat.
107
,
533
546
.
15.
Erbe
,
C.
, and
Farmer
,
D. M.
(
2000
). “
Zones of impact around icebreakers affecting beluga whales in the Beaufort Sea
,”
J. Acoust. Soc. Am.
108
,
1332
1340
.
16.
Erbe
,
C.
,
Reichmuth
,
C.
,
Cunningham
,
K.
,
Lucke
,
K.
, and
Dooling
,
R.
(
2016
). “
Communication masking in marine mammals: A review and research strategy
,”
Mar. Pollut. Bull.
103
,
15
38
.
17.
Finneran
,
J. J.
,
Schlundt
,
C. E.
,
Dear
,
R.
,
Carder
,
D. A.
, and
Ridgway
,
S. H.
(
2002
). “
Temporary shift in masked hearing thresholds in odontocetes after exposure to single underwater impulses from a seismic watergun
,”
J. Acoust. Soc. Am.
111
,
2929
2940
.
18.
Frouin-Mouy
,
H.
,
Kowarski
,
K.
,
Martin
,
B.
, and
Bröker
,
K.
(
2017
). “
Seasonal trends in acoustic detection of marine mammals in Baffin Bay and Melville Bay, Northwest Greenland
,”
Arctic
70
,
59
76
.
19.
Garland
,
E. C.
,
Castellote
,
M.
, and
Berchok
,
C. L.
(
2015
). “
Beluga whale (Delphinapterus leucas) vocalizations and call classification from the eastern Beaufort Sea population
,”
J. Acoust. Soc. Am.
137
,
3054
3067
.
20.
Gillespie
,
D.
,
Mellinger
,
D. K.
,
Gordon
,
J.
,
McLaren
,
D.
,
Redmond
,
P.
,
McHugh
,
R.
,
Trinder
,
P.
,
Deng
,
X. Y.
, and
Thode
,
A.
(
2009
). “
PAMGUARD: Semiautomated, open source software for real-time acoustic detection and localization of cetaceans
,”
J. Acoust. Soc. Am.
125
,
2547
.
21.
Halliday
,
W. D.
,
Barclay
,
D.
,
Barkley
,
A. N.
,
Cook
,
E.
,
Hilliard
,
R. C.
,
Hussey
,
N. E.
,
Jones
,
J. M.
,
Juanes
,
F.
,
Marcoux
,
M.
,
Niemi
,
A.
,
Nudds
,
S.
,
Pine
,
M.
,
Richards
,
C.
,
Schraffenberg
,
K.
,
Westdal
,
K.
, and
Insley
,
S.
(
2021a
). “
Underwater sound levels in the Canadian Arctic, 2014–2019
,”
Mar. Pollut. Bull.
168
,
112437
.
22.
Halliday
,
W. D.
,
Pine
,
M. K.
,
Citta
,
J. J.
,
Harwood
,
L.
,
Donna
,
D.
,
Hauser
,
W.
,
Hilliard
,
R. C.
,
Lea
,
E. V.
,
Loseto
,
L.
,
Quakenbush
,
L.
, and
Insley
,
S.
(
2021b
). “
Potential exposure of beluga and bowhead whales to underwater noise from ship traffic in the Beaufort and Chukchi Seas
,”
Ocean Coast. Manage.
204
,
105473
.
23.
Halliday
,
W. D.
,
Pine
,
M. K.
, and
Insley
,
S. J.
(
2020
). “
Underwater noise and Arctic marine mammals: Review and policy recommendations
,”
Environ. Rev.
28
,
438
448
.
24.
Hauser
,
D. D. W.
,
Laidre
,
K. L.
,
Stafford
,
K. M.
,
Stern
,
H. L.
,
Suydam
,
R. S.
, and
Richard
,
P. R.
(
2017
). “
Decadal shifts in autumn migration timing by Pacific Arctic beluga whales are related to delayed annual sea ice formation
,”
Glob. Chang. Biol.
23
,
2206
2217
.
25.
Heide-Jørgensen
,
M. P.
,
Blackwell
,
S. B.
,
Tervo
,
O. M.
,
Samson
,
A. L.
,
Garde
,
E.
,
Hansen
,
R. G.
,
Ngô
,
M. C.
,
Conrad
,
A.
,
Trinhammer
,
P.
,
Schmidt
,
H.
,
Sinding
,
M.-H. S.
,
Williams
,
T.
, and
Ditlevsen
,
S.
(
2021
). “
Behavioral response study on seismic airgun and vessel exposures in narwhals
,”
Front. Mar. Sci.
8
,
658173
.
26.
Heide-Jørgensen
,
M. P.
,
Blackwell
,
S. B.
,
Williams
,
T. M.
,
Sinding
,
M. H. S.
,
Skovrind
,
M.
,
Tervo
,
O. M.
,
Garde
,
E.
,
Hansen
,
R.
,
Nielsen
,
N. H.
,
Ngô
,
M. C.
, and
Ditlevsen
,
S.
(
2020
). “
Some like it cold: Temperature-dependent habitat selection by narwhals
,”
Ecol. Evol.
10
,
8073
8090
.
27.
Heide-Jørgensen
,
M. P.
,
Dietz
,
R.
,
Laidre
,
K. L.
,
Richard
,
P.
,
Orr
,
J.
, and
Schmidt
,
H. C.
(
2003a
). “
The migratory behaviour of narwhals (Monodon monoceros)
,”
Can. J. Zool.
81
,
1298
1305
.
28.
Heide-Jørgensen
,
M. P.
,
Laidre
,
K. L.
,
Borchers
,
D.
,
Marques
,
T. A.
,
Stern
,
H.
, and
Simon
,
M.
(
2010
). “
The effect of sea-ice loss on beluga whales (Delphinapterus leucas) in West Greenland
,”
Polar Res.
29
,
198
208
.
29.
Heide-Jørgensen
,
M. P.
,
Richard
,
P.
,
Dietz
,
R.
,
Laidre
,
K. L.
,
Orr
,
J.
, and
Schmidt
,
H. C.
(
2003b
). “
An estimate of the fraction of belugas (Delphinapterus leucas) in the Canadian high Arctic that winter in West Greenland
,”
Polar Biol.
26
,
318
326
.
30.
Hersbach
,
H.
,
Bell
,
B.
,
Berrisford
,
P.
,
Biavati
,
G.
,
Horányi
,
A.
,
Muñoz Sabater
,
J.
,
Nicolas
,
J.
,
Peubey
,
C.
,
Radu
,
R.
,
Rozum
,
I.
,
Schepers
,
D.
,
Simmons
,
A.
,
Soci
,
C.
,
Dee
,
D.
, and
Thépaut
,
J-N.
(
2023
). “
ERA5 hourly data on single levels from 1979 to present
,”
Copernicus Climate Change Service
(C3S Climate Data Store (CDS).
31.
Hobbs
,
R.
,
Reeves
,
R.
,
Prewitt
,
J. S.
,
Desportes
,
G.
,
Breton-Honeyman
,
K.
,
Christensen
,
T.
,
Citta
,
J. J.
,
Ferguson
,
S. H.
,
Frost
,
K. J.
,
Garde
,
E.
,
Gavrilo
,
M.
,
Ghazal
,
M.
,
Glazov
,
D. M.
,
Gosselin
,
J.-F.
,
Hammill
,
M.
,
Hansen
,
R. G.
,
Harwood
,
L.
,
Heide-Jorgensen
,
M. P.
,
Inglangasuk
,
G.
, and
Watt
,
C. A.
(
2019
). “
Global review of the conservation status of Monodontid stocks
,”
Mar. Fish. Rev.
81
,
1
53
.
32.
Innes
,
S.
,
Heide-Jørgensen
,
M. P.
,
Laake
,
J. L.
,
Laidre
,
K. L.
,
Cleator
,
H. J.
,
Richard
,
P.
, and
Stewart
,
R. E. A.
(
2002
). “
Surveys of belugas and narwhals in the Canadian High Arctic in 1996
,”
NAMMCO Sci. Publ.
4
,
169
190
.
33.
IPCC
(
2022
). “
Polar regions
,” in
The Ocean and Cryosphere in a Changing Climate: Special Report of the Intergovernmental Panel on Climate Change
(
Cambridge University Press
,
Cambridge, UK
), pp.
203
320
.
34.
Jakobsson
,
M.
,
Mayer
,
L.
,
Coakley
,
B.
,
Dowdeswell
,
J. A.
,
Forbes
,
S.
,
Fridman
,
B.
, and
Hodnesdal
,
H.
(
2012
). “
The international bathymetric chart of the Arctic Ocean (IBCAO) Version 3.0
,”
Geophys. Res. Lett.
39
,
L12609
, https://doi.org/10.1029/2012GL052219.
35.
Jones
,
J. M.
,
Frasier
,
K. E.
,
Westdal
,
K. H.
,
Ootoowak
,
A. J.
,
Wiggins
,
S. M.
, and
Hildebrand
,
J. A.
(
2022
). “
Beluga (Delphinapterus leucas) and narwhal (Monodon monoceros) echolocation click detection and differentiation from long-term Arctic acoustic recordings
,”
Polar Biol.
45
,
449
463
.
36.
K. Lisa Yang Center for Conservation Bioacoustics at the Cornell Lab of Ornithology
(
2023
). “
Raven Pro: Interactive Sound Analysis Software (Version 1.6.5) [computer software]
,” https://ravensoundsoftware.com/ (Last viewed October 13, 2023).
37.
Koblitz
,
J. C.
,
Stilz
,
P.
,
Rasmussen
,
M. H.
, and
Laidre
,
K. L.
(
2016
). “
Highly directional sonar beam of narwhals (Monodon monoceros) measured with a vertical 16 hydrophone array
,”
PLoS One
11
,
e0162069
.
38.
Laidre
,
K. L.
, and
Heide-Jørgensen
,
M. P.
(
2011
). “
Life in the lead: Extreme densities of narwhals Monodon monoceros in the offshore pack ice
,”
Mar. Ecol. Prog. Ser.
423
,
269
278
.
39.
Laidre
,
K. L.
,
Moon
,
T.
,
Hauser
,
D. D. W.
,
McGovern
,
R.
,
Heide-Jørgensen
,
M. P.
,
Dietz
,
R.
, and
Hudson
,
B.
(
2016
). “
Use of glacial fronts by narwhals (Monodon monoceros) in West Greenland
,”
Biol. Lett.
12
,
20160457
.
40.
Laidre
,
K. L.
,
Stirling
,
I.
,
Lowry
,
L. F.
,
Wiig
,
Ø.
,
Heide-Jørgensen
,
M. P.
, and
Ferguson
,
S. H.
(
2008
). “
Quantifying the sensitivity of Arctic marine mammals to climate-induced habitat change
,”
Ecol. Appl.
18
,
S97
S125
.
41.
Lammers
,
M. O.
,
Castellote
,
M.
,
Small
,
R. J.
,
Atkinson
,
S.
,
Jenniges
,
J.
,
Rosinski
,
A.
,
Oswald
,
J. N.
, and
Garner
,
C.
(
2013
). “
Passive acoustic monitoring of Cook Inlet beluga whales (Delphinapterus leucas
),”
J. Acoust. Soc. Am.
134
,
2497
2504
.
42.
Lannuzel
,
D.
,
Tedesco
,
L.
,
van Leeuwe
,
M.
,
Campbell
,
K.
,
Flores
,
H.
,
Delille
,
B.
,
Miller
,
L.
,
Stefels
,
J.
, and
the BEPSII community
(
2020
). “
The future of Arctic sea-ice biogeochemistry and ice-associated ecosystems
,”
Nat. Clim. Chang.
10
,
983
992
.
43.
Lesage
,
V.
(
2021
). “
The challenges of a small population exposed to multiple anthropogenic stressors and a changing climate: The St. Lawrence Estuary beluga
,”
Polar Res.
40
,
5523
.
44.
Lesage
,
V.
,
Barrette
,
C.
,
Kingsley
,
M. C. S.
, and
Sjare
,
B.
(
1999
). “
The effect of vessel noise on the vocal behavior of belugas in the St. Lawrence River estuary, Canada
,”
Mar. Mamm. Sci.
15
,
65
84
.
45.
Liaw
,
A.
, and
Wiener
,
M.
(
2002
). “
Classification and regression by randomForest
,”
R News
2
,
18
22
.
46.
Louis
,
M.
,
Skovrind
,
M.
,
Castruita
,
J. A. S.
,
Garilao
,
C.
,
Kaschner
,
K.
,
Gopalakrishnan
,
S.
,
Haile
,
J. S.
,
Lydersen
,
C.
,
Kovacs
,
K. M.
,
Garde
,
E.
,
Heide-Jørgensen
,
M. P.
,
Postma
,
L.
,
Ferguson
,
S. H.
,
Willerslev
,
E.
, and
Lorenzen
,
E.
(
2020
). “
Influence of past climate change on phylogeography and demographic history of narwhals, Monodon monoceros
,”
Proc. R. Soc. B
287
,
20192964
.
47.
Marcoux
,
M.
,
Auger-Methe
,
M.
,
Chmelnitsky
,
E.
,
Ferguson
,
S. H.
, and
Humphries
,
M. M.
(
2011
). “
Local passive acoustic monitoring of narwhal presence in the Canadian Arctic: A pilot project
,”
Arctic
64
,
307
316
.
48.
Marcoux
,
M.
,
Auger-Méthé
,
M.
, and
Humphries
,
M. M.
(
2012
). “
Variability and context specificity of narwhal (Monodon monoceros) whistles and pulsed calls
,”
Mar. Mamm. Sci.
28
,
649
665
.
49.
Mattmüller
,
R. M.
,
Thomisch
,
K.
,
Van Opzeeland
,
I.
,
Laidre
,
K. L.
, and
Simon
,
M.
(
2022
). “
Passive acoustic monitoring reveals year-round marine mammal community composition off Tasiilaq, Southeast Greenland
,”
J. Acoust. Soc. Am.
151
,
1380
1392
.
50.
Meier
,
W. N.
,
Hovelsrud
,
G. K.
,
van Oort
,
B. E. H.
,
Key
,
J. R.
,
Kovacs
,
K. M.
,
Michel
,
C.
,
Haas
,
C.
,
Granskog
,
M.
,
Gerland
,
S.
,
Perovich
,
D. K.
,
Makshtas
,
A.
, and
Resit
,
J. D.
(
2014
). “
Arctic sea ice in transformation: A review of recent observed changes and impacts on biology and human activity
,”
Rev. Geophys.
52
,
185
217
, https://doi.org/10.1002/2013RG000431.
51.
Melia
,
N.
,
Haines
,
K.
, and
Hawkins
,
E.
(
2016
). “
Sea ice decline and 21st century trans-Arctic shipping routes
,”
Geophys. Res. Lett.
43
,
9720
9728
, https://doi.org/10.1002/2016GL069315.
52.
Miller
,
L. A.
,
Pristed
,
J.
,
Mohl
,
B.
, and
Surlykke
,
A.
(
1995
). “
The click-sounds of narwhals (Monodon monoceros) in Inglefield Bay, Northwest Greenland
,”
Mar. Mamm. Sci.
11
,
491
502
.
53.
Møhl
,
B.
,
Surlykke
,
A.
, and
Miller
,
L. A.
(
1990
). “
High intensity narwhal clicks
,” in
Sensory Abilities of Cetaceans: Laboratory and Field Evidence
, edited by
J. A.
Thomas
and
R. A.
Kastelein
(
Springer US
,
Boston, MA
), pp.
295
303
.
54.
Moore
,
S. E.
,
Clarke
,
J. T.
,
Okkonen
,
S. R.
,
Grebmeier
,
J. M.
,
Berchok
,
C. L.
, and
Stafford
,
K. M.
(
2022
). “
Changes in gray whale phenology and distribution related to prey variability and ocean biophysics in the northern Bering and eastern Chukchi seas
,”
PLoS One
17
,
e0265934
.
55.
Nielsen
,
M. R.
(
2009
). “
Is climate change causing the increasing narwhal (Monodon monoceros) catches in Smith Sound, Greenland?
,”
Polar Res.
28
,
238
245
.
56.
Onarheim
,
I. H.
,
Eldevik
,
T.
,
Smedsrud
,
L. H.
, and
Stroeve
,
J. C.
(
2018
). “
Seasonal and regional manifestation of Arctic sea ice loss
,”
J. Clim.
31
,
4917
4932
.
57.
Radtke
,
C. L.
,
Terhune
,
J. M.
,
Frouin-Mouy
,
H.
, and
Rouget
,
P. A.
(
2023
). “
Vocal count responses of narwhals to bulk carrier noise in Milne Inlet, Nunavut, Canada
,”
Mar. Mamm. Sci.
39
,
1057
1075
.
58.
Rantanen
,
M.
,
Karpechko
,
A. Y.
,
Lipponen
,
A.
,
Nordling
,
K.
,
Hyvärinen
,
O.
,
Ruosteenoja
,
K.
,
Vihma
,
T.
, and
Laaksonen
,
A.
(
2022
). “
The Arctic has warmed nearly four times faster than the globe since 1979
,”
Commun. Earth Environ.
3
,
168
.
59.
Rasmussen
,
M. H.
,
Koblitz
,
J. C.
, and
Laidre
,
K. L.
(
2015
). “
Buzzes and high-frequency clicks recorded from narwhals (Monodon monoceros) at their wintering ground
,”
Aquat. Mamm.
41
,
256
264
.
60.
R Core Team
(
2023
).
R: A Language and Environment for Statistical Computing
,
R Foundation for Statistical Computing
, http://www.r-project.org/ (Last viewed February 14, 2024).
61.
Reeves
,
R. R.
,
Ewins
,
P. J.
,
Agbayani
,
S.
,
Heide-Jørgensen
,
M. P.
,
Kovacs
,
K. M.
,
Lydersen
,
C.
,
Suydam
,
R.
,
Elliott
,
Q.
,
Polet
,
G.
,
van Dijk
,
Y.
, and
Blijleven
,
R.
(
2014
). “
Distribution of endemic cetaceans in relation to hydrocarbon development and commercial shipping in a warming Arctic
,”
Mar. Policy
44
,
375
389
.
62.
Richard
,
P. R.
,
Heide-Jørgensen
,
M. P.
, and
St. Aubin
,
D.
(
1998a
). “
Fall movements of belugas (Delphinapterus leucas) with satellite-linked transmitters in Lancaster Sound, Jones Sound, and Northern Baffin Bay
,”
Arctic
51
,
5
16
.
63.
Richard
,
P. R.
,
Orr
,
J. R.
,
Dietz
,
R.
, and
Dueck
,
L.
(
1998b
). “
Sightings of belugas and other marine mammals in the North Water, late March 1993
,”
Arctic
51
,
1
4
.
64.
Roy
,
N.
,
Simard
,
Y.
,
Gervaise
,
C.
, and
Dtn
,
E.
(
2010
). “
3D tracking of foraging belugas from their clicks: Experiment from a coastal hydrophone array
,”
Appl. Acoust.
71
,
1050
1056
.
65.
Sakai
,
T.
(
2023
). “
PAMpal: Load and process passive acoustic data
.”
66.
Schevill
,
W. E.
, and
Lawrence
,
B.
(
1949
). “
Underwater listening to the white porpoise (Delphinapterus leucas)
,”
Science
109
,
143
144
.
67.
Shuert
,
C. R.
,
Marcoux
,
M.
,
Hussey
,
N. E.
,
Heide-Jørgensen
,
M. P.
,
Dietz
,
R.
, and
Auger-Méthé
,
M.
(
2022
). “
Decadal migration phenology of a long-lived Arctic icon keeps pace with climate change
,”
Proc. Natl. Acad. Sci. U.S.A.
119
,
e2121092119
.
68.
Sjare
,
B. L.
, and
Smith
,
T. G.
(
1986
). “
The vocal repertoire of white whales, Delphinapterus leucas, summering in Cunningham Inlet, Northwest Territories
,”
Can. J. Zool.
64
,
407
415
.
69.
Southall
,
B. L.
,
Nowacek
,
D. P.
,
Bowles
,
A. E.
,
Senigaglia
,
V.
,
Bejder
,
L.
, and
Tyack
,
P. L.
(
2021
). “
Marine mammal noise exposure criteria: Assessing the severity of marine mammal behavioral responses to human noise
,”
Aquat. Mamm.
47
,
421
464
.
70.
Stafford
,
K. M.
,
Citta
,
J. J.
,
Okkonen
,
S. R.
, and
Zhang
,
J.
(
2021
). “
Bowhead and beluga whale acoustic detections in the western Beaufort Sea 2008–2018
,”
PLoS One
16
,
e0253929
.
71.
Stafford
,
K. M.
,
Farley
,
E. V.
,
Ferguson
,
M.
,
Kuletz
,
K. J.
, and
Levine
,
R.
(
2022
). “
Northward range expansion of subarctic upper trophic level animals into the Pacific Arctic region
,”
Oceanography
35
,
158
166
.
72.
Stafford
,
K. M.
,
Laidre
,
K. L.
, and
Heide-Jørgensen
,
M. P.
(
2012
). “
First acoustic recordings of narwhals (Monodon monoceros) in winter
,”
Mar. Mamm. Sci.
28
,
197
207
.
73.
Tervo
,
O. M.
,
Blackwell
,
S. B.
,
Ditlevsen
,
S.
,
Garde
,
E.
,
Hansen
,
R. G.
,
Samson
,
A. L.
,
Conrad
,
A. S.
, and
Heide-Jørgensen
,
M. P.
(
2023
). “
Stuck in a corner: Anthropogenic noise threatens narwhals in their once pristine Arctic habitat
,”
Sci. Adv.
9
,
eade0440
.
74.
Williams
,
T. M.
,
Blackwell
,
S. B.
,
Tervo
,
O.
,
Garde
,
E.
,
Sinding
,
H. S.
,
Richter
,
B.
, and
Peter
,
M.
(
2022
). “
Physiological responses of narwhals to anthropogenic noise: A case study with seismic airguns and vessel traffic in the Arctic
,”
Funct. Ecol.
36
,
2251
2266
.
75.
Zahn
,
M. J.
,
Ladegaard
,
M.
,
Simon
,
M.
,
Stafford
,
K. M.
,
Sakai
,
T.
, and
Laidre
,
K. L.
(
2023
). “Catalogue of Arctic toothed whale click spectra and spectrograms from soundtrap recordings (1.0.0) [Data set],”
Zenodo
https://doi.org/10.5281/zenodo.10076260.
76.
Zahn
,
M. J.
,
Laidre
,
K. L.
,
Stilz
,
P.
,
Rasmussen
,
M. H.
, and
Koblitz
,
J. C.
(
2021a
). “
Vertical sonar beam width and scanning behavior of wild belugas (Delphinapterus leucas) in West Greenland
,”
PLoS One
16
,
e0257054
.
77.
Zahn
,
M. J.
,
Rankin
,
S.
,
Mccullough
,
J. L. K.
,
Koblitz
,
J. C.
,
Archer
,
F.
,
Rasmussen
,
M. H.
, and
Laidre
,
K. L.
(
2021b
). “
Acoustic differentiation and classification of wild belugas and narwhals using echolocation clicks
,”
Sci. Rep.
11
,
22141
.

Supplementary Material