Improving near real-time acoustic monitoring of two endangered whale species that make similar calls

Abstract

Protecting marine mammals from anthropogenic threats, such as vessel strikes, requires accurate species occurrence data to ensure that management actions are justifiable. In Northwest Atlantic waters, several regions are seasonally monitored for baleen whales using near real-time (NRT) passive acoustic gliders. Two endangered species monitored, sei and blue whales, produce acoustically similar calls, complicating species discrimination in NRT. In this study, we sought to improve NRT blue whale monitoring in a region with irregular sei whale presence by implementing a new blue whale call library for automated detection and classification, and revised NRT analyst validation protocols. Acoustic detections from three deployments in 2025 were compared to six deployments from 2022-2024. NRT detections were reviewed audibly and visually to validate them post-deployment. The mean proportion of deployment days with detections misclassified as sei whale in NRT decreased from 3% to 0 in 2025, while true NRT blue whale detections increased by 89%. The new blue whale call library correctly classified blue whale calls in an average of 35% of blue whale ‘detected’ periods. Blue whale acoustic detections varied by season and location, with audible calls contributing to more detections than infrasonic calls. Overall, NRT misclassifications were reduced, and blue whale call discrimination improved, which is important for monitoring endangered species whose detection data inform conservation measures. These improvements are particularly relevant if either species is incorporated into future dynamic management strategies, such as the current NRT dynamic speed restriction program used to mitigate ship strike risk to North Atlantic right whales in the Gulf of St. Lawrence.

1 Introduction

Passive acoustic monitoring (PAM) is a valuable tool for determining marine mammal occurrence, and many automated software programs have been developed to improve the efficiency of detecting and classifying marine mammal sounds in PAM data (e.g., Baumgartner and Mussoline, 2011; Bermant et al., 2019; Buchan et al., 2020). ‘Detection’ refers to identifying acoustic signals, while ‘classification’ refers to assigning species identity to those signals. Automated detection and classification techniques - including neural networks (e.g., Bermant et al., 2019; Potter et al., 1994; Rasmussen and Širović, 2021; Shiu et al., 2020), Hidden Markov Models (e.g., Buchan et al., 2020; Trawicki, 2021), and discriminant function analyses (e.g., Baumgartner and Mussoline, 2011) - are popular because they substantially reduce processing time and can perform comparably to manual analysis (Baumgartner and Mussoline, 2011; Bermant et al., 2019; Buchan et al., 2020).

Near real-time (NRT) PAM extends these capabilities by enabling reliable, efficient species detection and classification data from hydrophones integrated with autonomous platforms such as moored buoys and underwater gliders, supporting important advancements in marine mammal conservation (e.g., Indeck et al., 2025b). These systems send passive acoustic data to shore via satellite or undersea cabling, where human analysts validate classifications and disseminate detection information for immediate management use (e.g., mitigating the risks of human activities, when and where at-risk species are detected; e.g., Indeck et al., 2025a, b; Sanguineti et al., 2021; Van Parijs et al., 2009). However, NRT PAM presents challenges not present in conventional PAM (Baumgartner et al., 2013; Gibb et al., 2019; Kowarski et al., 2020). Because satellite transmission is expensive, only a small subset of the recorded audio or data products derived from spectrograms are sent ashore; full acoustic spectrograms and audio files are not available to analysts, limiting contextual information often used to more reliably differentiate species (Baumgartner et al., 2013, 2019, 2020, 2021; Gibb et al., 2019; Kowarski et al., 2020). Autonomous platforms are also power-limited, preventing the use of computationally intensive machine learning classifiers. These constraints are typically addressed by only transmitting data products, such as ‘pitch tracks’ (Baumgartner et al., 2013; Baumgartner and Mussoline, 2011), or only the highest-priority candidate detections (e.g., endangered species) as contour tracings (Kowarski et al., 2020).

Species discrimination can be particularly challenging when co-occurring species produce calls with overlapping acoustic features – such as the frequency, duration, amplitude, pattern, slope, inter-note intervals (INI), or spectrogram shape (Baumgartner and Mussoline, 2011; Podos and Webster, 2022; Wilder et al., 2022) – which increases the risk of misclassification. A prominent example of this in the Northwest Atlantic is distinguishing sei whale (Balaenoptera borealis) calls from those of blue whales (Balaenoptera musculus), whose vocalizations overlap in frequency and structure.

Sei and blue whales occupy overlapping ranges in the Northwest Atlantic, and their presence can co-occur in acoustic monitoring records. Both species are listed as endangered in Canada: the Northwest Atlantic sei whale population is estimated at fewer than 1000 individuals, while the blue whale population is estimated to include no more than 250 mature individuals (COSEWIC, 2012, 2019; Fisheries and Oceans Canada, 2020; Rocha et al., 2015; Thomas et al., 2016). Sei whales, typically associated with shelf-break habitats, are regularly detected along the Scotian Shelf, off the coast of Newfoundland into the Labrador Sea, and in the Gulf of Maine (Baumgartner et al., 2008; COSEWIC, 2019; Davis et al., 2020; Delarue et al., 2022; Gomez et al., 2020; Horwood, 1987; Macklin et al., 2024; Prieto et al., 2014). Similarly, blue whales frequent the coast of Newfoundland and the Scotian Shelf, and are commonly observed in the Gulf of St. Lawrence (GoSL), where sei whale detections are rare (Beauchamp et al., 2009; COSEWIC, 2019; Lesage et al., 2017; Moors-Murphy et al., 2019; Ramp and Sears, 2013; Simard et al., 2016). Seasonal migration patterns further contribute to overlap, with both species exhibiting increased presence and acoustic activity during the summer and fall (Davis et al., 2020; Delarue et al., 2022; Moors-Murphy et al., 2019; Wingfield et al., 2022).

Discriminating between sei and blue whales is challenging because both produce downsweeping calls that overlap in frequency (~150–30 Hz) and overall shape (Baumgartner et al., 2008; Calderan et al., 2014; Cerchio and Weir, 2022; Español-Jiménez et al., 2019; Macklin et al., 2024; Ou et al., 2015; Rankin and Barlow, 2007; Tremblay et al., 2019). Sei whale downsweeps are 1–2 seconds in duration and typically span 80–30 Hz (Baumgartner et al., 2008; Davis et al., 2020; Macklin et al., 2024). Their concave-up shape (e.g., Figure 1A) and characteristic production patterns are diagnostic of the species (Baumgartner et al., 2008; Macklin et al., 2024; Wilder et al., 2022). Sei whale calls occur as singlets, doublets, or triplets, with inter-note intervals (INI) of ~3.5 seconds in the Gulf of Maine and 2-2.5 seconds off the Scotian Shelf (Baumgartner et al., 2008; Macklin et al., 2024; Tremblay et al., 2019; Wilder et al., 2022). Although the function of these downsweeps remains uncertain, they are thought to act as contact calls and are produced by both sexes (Baumgartner et al., 2008; Cusano et al., 2023).

Figure 1

Blue whales have a broader call repertoire, comprised of infrasonic calls (i.e., outside the human audible frequency range of ~20Hz-20kHz) typically between 10–20 Hz, produced as repetitive ‘songs’ observed only from males and thought to be a breeding display, and audible calls (i.e., within the human audible range) associated with foraging or contact and produced by both sexes (Berchok et al., 2006; Mellinger and Clark, 2003; Oleson et al., 2007a, b; Wingfield et al., 2022). Audible calls vary widely in their frequency and temporal characteristics, including ‘Short’, ‘Medium’, ‘Steep’, and ‘Long’ downsweeps, as well as ‘Arch’ calls – all of which fall within the same frequency range as sei whale downsweeps, complicating species differentiation (Figures 1, 2) (Department of Fisheries and Oceans (DFO), 2024, unpublished). A key distinguishing feature of blue whale audible calls is their downward concavity, which reliably indicates blue whale acoustic presence (Berchok et al., 2006; Mellinger and Clark, 2003; Wilder et al., 2022).

Figure 2

Accurate discrimination between sei and blue whale detections is crucial for understanding their occurrence in the Northwest Atlantic, including the GoSL, which is bisected by a heavily trafficked shipping route connecting Eastern Canada’s major cities to the rest of the world (Figure 3). Each year, underwater gliders conduct NRT acoustic monitoring in these shipping lanes and are able to detect five baleen whale species, including sei and blue whales. The NRT data sent to shore are pitch-tracked contours derived from spectrograms, so analysts make whale classification decisions in NRT based only on visual review and no audio data (Baumgartner et al., 2013; Baumgartner and Mussoline, 2011). Here, NRT detections of the critically endangered North Atlantic right whale (Eubalaena glacialis, NARW) trigger mandatory 15-day regional vessel slowdowns to reduce the risk of lethal ship strike (Indeck et al., 2025a; Transport Canada, 2026a, b). Although slowdowns are only implemented for NARWs, sei and blue whales share similar ship strike threats, making accurate acoustic validations important baseline data for conservation and management.

Figure 3

As a precursor to this study, NRT sei whale detections from these monitoring missions were manually analyzed using the full recorded audio after glider recovery. Nearly all (97%) sei whale detections were misclassifications, as most (91%) contained definite or probable evidence of blue whale acoustic presence (MacDonald, 2025). These results demonstrated that the similar call features of the two species led to sei and blue whales being over- and underrepresented, respectively, during NRT acoustic monitoring surveys. This was largely because former detection protocols lacked a reliable method for discriminating between the two species.

Therefore, the primary objective of this study was to determine whether NRT acoustic detection of blue whales could be improved in a region where sei whale presence is irregular but their call features are similar. Updates to the NRT detection procedure included refining the detection protocols for both species and incorporating a new blue whale audible call library for automated detection and classification. These updates were implemented for the first time during the 2025 monitoring season (May–November 2025), and NRT detections of both species were compared to detections from the previous three monitoring years (2022-2024) in the same region, to broadly assess changes in detection reliability. Additionally, spatial and temporal variability in blue whale NRT detections had not previously been examined in this region. We therefore evaluated whether seasonal, locational, and annual differences were associated with blue whale acoustic presence and call category (audible vs. infrasonic) use. Finally, we assessed the species-discriminating ability of the new blue whale call library to determine whether it improved differentiation between blue and sei whale detections.

2 Methods

2.1 Data collection

2.1.1 Study system and monitoring platform

A portion of the Laurentian Channel (LC) in the GoSL is divided into dynamic shipping zones (DSZs) to facilitate area-restricted vessel speed limits under Transport Canada’s dynamic management strategy for reducing NARW ship-strike risk (Indeck et al., 2025a; Transport Canada, 2026a) (Figure 3). NRT baleen whale acoustic detections were collected from nine underwater glider deployments across DSZs A, B, C, and E: six deployments from 2022–2024 and three in 2025, each lasting 100–185 days between April and November. Deployment dates and the DSZ(s) monitored by each glider are provided in Supplementary Table 1 of the Supplement.

We used battery-powered Slocum G3 and G3s electric underwater gliders (Teledyne Webb Research). The gliders travelled at ~0.15 m/s, performing sawtooth dives in the water column as they transited their assigned DSZ. All gliders were equipped with a digital acoustic monitoring (DMON) instrument that contained a single hydrophone (flat response: 10–7500 Hz; sensitivity: –203 dB re 1 V μPa–1; total system gain: 33.2 dB). The DMON continuously recorded and archived 16-bit audio at 2 kHz throughout each deployment, a sampling rate sufficient to capture sei and blue whale vocalizations. Archived audio was available upon glider retrieval at the end of each deployment.

All DMONs were integrated with the Low Frequency Detection and Classification System (LFDCS), which automatically detects and classifies narrowband tonal sounds in real-time by comparing the attributes of pitch tracks to those of call exemplars in species-specific libraries (Baumgartner and Mussoline, 2011). LFDCS detections can either be correctly classified, misclassified as another species, or left unclassified; similarly, abiotic sounds can be correctly left unclassified or misclassified as whale calls. An example of a blue whale arch call pitch track misclassified as a sei whale downsweep is shown in Supplementary Figure 1.

2.1.2 NRT detection and validation of baleen whale calls

Adapting the procedure described in Baumgartner et al. (2013), the gliders surfaced every ~4–6 hours and transmitted a subset of pitch track data and LFDCS classifications (up to 8 kb/hr) in 15-minute ‘tally periods’ to onshore servers via an onboard Iridium satellite modem. Data beyond the 8 kb limit were archived, but not available for NRT analysis. Once daily, trained analysts validated all tally periods, assessing their associated pitch tracks and classifications according to published NRT species detection protocols and scoring each tally period as either ‘detected’, ‘possibly detected’, or ‘not detected’ for all monitored species (Wilder et al., 2022). Validation scores incorporate objective evidence (i.e., LFDCS classifications), subjective evidence (e.g., pitch track shape and amplitude), and contextual information (i.e., whether other species’ calls are present) (Wilder et al., 2022). A ‘detected’ score indicates that a tally period contains unambiguous objective and subjective evidence of species presence and meets the minimum requirement for high-quality calls. A ‘possibly detected’ score also requires a minimum threshold of calls, but reflects lower certainty because the pitch tracks are fewer or less reliable. A ‘not detected’ score is assigned when there is insufficient evidence to identify the species producing the signals.

Original NRT detection protocols were developed for sei, fin (Balaenoptera physalus), humpback (Megaptera novaeangliae), and NARWs (Baumgartner et al., 2019; Wilder et al., 2022). Before 2024, the absence of a blue whale call library meant that blue whale acoustic presence in the GoSL was scored opportunistically (i.e., without supporting LFDCS classifications). A protocol to opportunistically but systematically score blue whale acoustic presence, requiring a minimum threshold of high-quality calls per tally period, was developed in 2022 and has been implemented for all subsequent GoSL monitoring deployments; prior to 2022, opportunistic blue whale detections were not standardized.

2.1.3 Updates to the NRT detection protocols

We revised the NRT species detection protocols published by Wilder et al. (2022) for both sei and blue whales in the GoSL prior to the 2025 monitoring season (May–November 2025). These updates were informed by MacDonald (2025), which showed that blue whale acoustic presence was frequently misclassified as sei whale presence in the GoSL (MacDonald, 2025). The original sei whale protocol was based on species occurrence in the Gulf of Maine, where these whales occur regularly and exhibit predictable seasonal patterns (COSEWIC, 2019; Davis et al., 2020; Delarue et al., 2022; Johnson et al., 2021). Applying this protocol in the GoSL, where sei whale visual and acoustic detections are irregular and unpredictable, likely contributed to the high misclassification rate (MacDonald, 2025). Therefore, the sei whale protocol was made more conservative for this region.

Under the revised protocol, only sei whale doublets or triplets with the species-expected INIs (noted above) are sufficient for a ‘detected’ score, as these call patterns are diagnostic of the species. A ‘detected’ score can no longer be assigned based solely on 3 or more classified singlets within a tally period, as archival analyses revealed that periods containing up to six classified sei whale singlets were manually scored as blue whale (MacDonald, 2025). Instead, this criterion is now used for ‘possibly detected’ scores, whereas previously only 1–2 classified calls were required. A comparison of the updated and previous protocols is provided in Table 1.

A new blue whale audible call library for the LFDCS was developed in 2024 and includes five call types (Figure 2). With its addition, criteria were needed for validating tally periods with LFDCS-classified blue whale calls. A single LFDCS blue whale classification was considered sufficient for a ‘detected’ score because a previous analysis applying the new library to manually validated archival data found that all 73 periods containing at least one LFDCS blue whale classification were manually scored as blue whale (MacDonald, 2025). It was also necessary to continue scoring blue whale acoustic presence using unclassified calls, as the same trial showed that manually validated blue whale periods contained classified calls only 31% of the time (MacDonald, 2025). Given that blue whales were frequently misclassified and underrepresented in NRT detections (MacDonald, 2025), the validation criteria for blue whales needed to become less restrictive.

This was achieved by decreasing the number of audible calls (loosely grouped as ‘D’ and ‘arch’ calls) required for both a ‘detected’ score (from 3 to 2) and ‘possibly detected’ score (from 2 to 1), while keeping downward concavity, diagnostic of blue whales, as a requirement in both cases. A comparison of the updated and previous protocols is provided in Table 1. It is important to note that the ‘arch’ calls referred to hereafter are consistent with those previously described for the GoSL (i.e., ~5–12 second calls that arch from approximately 56-69–35 Hz) (Mellinger and Clark, 2003). These calls align more closely with the ‘long downsweep’ call type in the new library (compare the first call in Figure 1B with Figure 2E) than with the ‘Arch’ call type, which is ~5 seconds in duration and spans a similar frequency range (41-65–34 Hz), but has a sharper, ‘inverted-V’ shape (Babin et al., 2025) (Figure 2C).

Although blue whale audible calls were the primary call category considered in this study, infrasonic calls also provide important information on blue whale acoustic presence and can be detected (just not classified) by the LFDCS. Infrasonic calls in the Northwest Atlantic are low-frequency (15–19 Hz), ~8-11-second, single-tone calls (Berchok et al., 2006; Boisseau et al., 2008; Mellinger and Clark, 2003). Until 2025, no standardized protocol existed for systematically scoring blue whale acoustic presence based on infrasonic calls in the GoSL; infrasonic detections were only ever scored opportunistically as ‘possibly detected’. Recent protocol updates outside the present study, designed to maximize blue whale detections in the GoSL, now allow analysts to use infrasonic calls for both ‘detected’ and ‘possibly detected’ scoring decisions. As a result, 2025 is the first year with standardized NRT detections of blue whale infrasonic calls.

2.1.4 Deployment details

During the 2022–2024 deployments, the LFDCS firmware (version 1.3) on each glider contained the original collection of call libraries. Sei whale presence was scored using the original, cross-region NRT species detection protocol (Table 1, Wilder et al., 2022), and blue whale presence was scored opportunistically but systematically using the GoSL-specific protocol developed in early 2022 (Wilder et al., 2022). Before the 2025 deployments, the LFDCS firmware (version 2.0) was modified to include the new blue whale audible call library, and sei and blue whale presence were scored according to the updated NRT protocols for both species (Table 1).

Each year, glider deployment locations (DSZs) and deployment/retrieval dates are determined in collaboration with Transport Canada to optimize NARW monitoring, prioritizing areas and seasons NARWs are expected to be present. As a result, glider coverage varies annually (Supplementary Table 1); for example, DSZ E is only monitored during the spring, when NARWs migrate through the Cabot Strait. When gliders were deployed or retrieved outside of the DSZs, the number of days a DSZ was acoustically monitored, hereafter ‘monitoring days’, differed from the deployment length. However, LFDCS data were still collected and validated in NRT during these periods, so these ‘deployment days’ were included in all analyses that did not consider DSZ.

2.2 Data processing and analyses

All NRT data (e.g., tally period validation scores - either ‘detected’, ‘possibly detected’ or ‘not detected’) were obtained for all nine deployments. Each tally period scored ‘detected’ or ‘possibly detected’ for blue whales was annotated with the call category (i.e., audible or infrasonic) that contributed to the NRT score. For tally periods in which both call categories were present, annotation was assigned to audible calls if the period would have received the same score in the absence of infrasonic calls, following the appropriate protocol; similarly, annotation was assigned to infrasonic calls if the score would have remained unchanged without audible calls. Tally periods were omitted when neither call category alone satisfied the requirements for the assigned NRT score.

2.2.1 Daily validation score analyses

We compared daily NRT validation scores from the years before (2022-2024) and after (2025) the library and protocol updates to determine whether the number of sei and blue whale detections differed significantly. Because no formal protocol existed for systematically scoring blue whale presence using infrasonic calls prior to 2025, tally periods scored ‘possibly detected’ (2022-2025) or ‘detected’ (2025) based solely on infrasonic calls were excluded from this analysis for consistency. For each deployment, we then calculated the number of days containing at least one tally period scored ‘detected’ or ‘possibly detected’ for sei and blue whales separately (i.e., four score categories). These counts were divided by the total number of deployment days to standardize by deployment duration, yielding the proportion of survey days in which each species was ‘detected’ or ‘possibly detected’ at least once. We then calculated the mean proportion of deployment days with at least one detection for each score category across 2022–2024 and 2025 deployments separately, allowing percent change (relative increase or decrease) between pre- and post-update means to be quantified.

Chi-square tests of proportions were run using the R package rstatix (Kassambara, 2023) to test whether the mean proportion of daily validation scores across the three 2025 deployments (observed) significantly deviated from the mean proportion across the six 2022–2024 deployments (expected). One test was run for each score category. All statistical analyses in this study were performed in R (R version 4.5.2) using a significance level of 0.05.

2.2.1.1 Manual analysis of the 2025 archival data

To assess the reliability of sei whale scoring following the protocol updates, we conducted a manual analysis to (1) determine the proportion of accurately scored tally periods and (2) assess whether sei whale presence was misclassified in 2025. For each tally period with a sei whale NRT detection score, we manually analyzed the corresponding archival audio, including the 15 minutes before and after the tally period, to confirm species presence. This was done by aurally and visually inspecting audio spectrograms (e.g., Figure 1) (1024-point fast Fourier transform (FFT), Hann window, 2.81 Hz frequency resolution, 90% frame-to-frame overlap) using Raven Pro (version 1.6.5, K. Lisa Yang Center for Conservation Bioacoustics at the Cornell Lab of Ornithology, 2026). The extra 30-minute buffer provided broader acoustic context to support confident species identification.

Each 45-minute period was assigned one of five scores: ‘definite sei whale’, ‘probable sei whale’, ‘definite blue whale’, ‘probable blue whale’, or ‘unknown’, based on call characteristics and contextual information. A period was scored ‘definite sei whale’ when characteristic concave-up downsweeps descending from ~80–30 Hz were present (e.g., Figure 1A) and no blue whale calls (i.e., concave-down calls) were detected. ‘Probable sei whale’ was assigned when characteristic sei whale downsweeps were present but blue whale presence could not be fully ruled out (e.g., when variable downsweeps were also observed). A ‘definite blue whale’ score was given when diagnostic arch calls (e.g., Figure 1B, call 1) or concave-down D calls (e.g., Figure 1B, calls 2 and 4) were present. ‘Probable blue whale’ was assigned when variable downsweeps suggested blue whale presence, but diagnostic arch or concave-down D calls were absent. Periods were scored ‘unknown’ when species identity could not be determined or too few calls were present. The precision of sei whale NRT detections was calculated using these manual scores. Recall was not calculated, as time limitations prevented analysis of the full archival record, so false negatives could not be quantified.

2.2.2 Variability in blue whale detections

2.2.2.1 Spatio-temporal variability

Generalized linear mixed models (GLMMs) were used to model spatio-temporal variability in blue whale NRT acoustic detections across all deployments at both the 15-minute tally period scale and the daily scale. Modelling at the tally period scale allowed finer-resolution assessment of relationships between predictor variables and blue whale detection, while daily scale models provided a more interpretable temporal resolution relevant for management and often used to evaluate seasonal patterns. GLMMs were selected because they allow for the inclusion of random effects to partially account for temporal non-independence. Temporal independence could not be guaranteed in these data, especially at the tally period scale, because consecutive detections may originate from the same calling individual. To address this, each model included a temporal random intercept created by grouping all deployments into consecutive time bins, annotated with either a unique datetime (tally period models) or date (daily models). Bin sizes sufficiently reduced autocorrelation, confirmed using the ‘acf’ function in the stats package (R Core Team, 2025) applied to the model residuals (Supplementary Figures 2, S3). All models met the assumptions of logistic GLMMs, verified using variance inflation factor tests, autocorrelation plots, and DHARMa residual diagnostics (Hartig, 2024).

Four binomial GLMMs were run to model the probability of acoustically detecting blue whales. Separate models were fit for data collected before (2022-2024) and after (2025) updates to the NRT detection protocols and call library, as these updates altered blue whale detectability. These models (Models 1–4) evaluated how locational, seasonal and annual differences were associated with blue whale acoustic detections across the two time periods and two temporal scales. Blue whale NRT presence (1) or absence (0) was used as the response variable, and predictor variables included Season, DSZ, and Year (for 2022–2024 models only), with datetime or date included as a random effect to account for temporal nonindependence (Table 2).

Models 1 and 3 used tally period data (n= 71,771 and 43,489 tally periods for 2022–2024 and 2025, respectively). Models 2 and 4 used daily data, where a deployment day was considered detected if ≥ 1 tally period was scored ‘detected’ for blue whales in NRT (n = 827 and 533 days for 2022–2024 and 2025, respectively). Models 3 and 4 (2025 data) included an interaction between Season and DSZ to assess whether blue whale detectability varied with glider monitoring location and/or seasonal whale distribution (Table 2). This interaction was not included in Models 1 and 2 (2022–2024 data) because some Season x DSZ combinations lacked observations (e.g., DSZ E was never monitored in the fall), and all combinations must be represented to produce finite estimates. Across all models, only tally periods or days scored ‘detected’ based on audible calls or ‘not detected’ were included for analysis; tally periods or days scored ‘detected’ based on infrasonic calls (which were not standardized across years) or ‘possibly detected’ were excluded. The equations for Models 1–4 were:

All GLMMs were fit using ‘glmmTMB’ from the R package glmmTMB (Brooks et al., 2017). Model selection was conducted using the ‘dredge’ function in the R package MuMln (Bartoń, 2025), allowing candidate models to be compared based on their corrected Akaike Information Criterion (AICc). Models within ΔAIC_C ≤ 2 were considered to have comparable support, and the best-fitting model was retained for interpretation. The ‘emmeans’ and ‘pairs(emmeans)’ functions in the emmeans R package (Lenth and Piaskowski, 2025) were used to calculate the estimated marginal means for each predictor variable, which were plotted, and to perform post-hoc pairwise comparisons of the estimated marginal means, respectively. All calculations were conducted on the ‘response’ scale so that the effect estimates were back-transformed to the units of the response variable (0 = not detected, 1 = detected, for blue whale presence).

2.2.2.2 Blue whale call category analyses

To examine any seasonal patterns in blue whale call category use, we calculated the proportion of 2025 blue whale ‘detections’ informed by either audible or infrasonic calls during spring (April 1–June 30), summer (July 1–September 30) and fall (October 1–December 31). This analysis included only tally periods scored ‘detected’ and annotated for either call category (total n= 2,979), so only 2025 data were used. Because monitoring effort was higher in the summer (276 days) than the spring (116 days) or fall (141 days), pairwise chi-square tests from the R package rstatix (Kassambara, 2023) were used to generate effort-weighted expected proportions by specifying the proportion of days monitored each season. Each test compared observed seasonal detection counts to these expected values to determine whether call detections differed more than would be expected based on effort alone. Separate chi-square analyses were run for audible and infrasonic call detections.

2.2.3 Blue whale call library analyses

Finally, we determined the proportion of 2025 blue whale audible call ‘detections’ that were informed by an LFDCS-classified blue whale call, as an indicator of how effectively the new call library identified true calls. For all tally periods scored ‘detected’ for blue whales that contained at least one LFDCS-classified blue whale call, we recorded the audible call type assigned by the LFDCS, whether the classification represented a true call (i.e., resembled the diagnostic calls of the species), and whether that classification contributed to the tally period’s ‘detected’ score. Using these annotations, we also determined the distribution of LFDCS-classified blue whale call types (i.e., types 30-34) that contributed to tally periods scored ‘detected’.

3 Results

3.1 Glider monitoring effort

The nine glider surveys from 2022–2025 resulted in 1360 monitoring days during the spring, summer and fall across DSZs A, B, C and E (Supplementary Table 2). The DSZ/season combination with the highest effort was summertime monitoring of DSZ C, which represented 24% (331 days) of total effort and was monitored each year. The six 2024/2025 deployments accounted for 72% (988 days) of total monitoring effort. DSZ E is not dynamically managed in the summer or fall, so no monitoring was conducted during those seasons in that zone. DSZ A has only been monitored since 2024.

3.2 Daily validation score analyses

The number of unique deployment days on which sei and blue whales were ‘detected’ or ‘possibly detected’ at least once, excluding tally periods scored using infrasonic blue whale calls, are presented in Table 3. Following the NRT updates in 2025, there were zero sei whale ‘detections’, a decrease from previous years when sei whales were scored ‘detected’ on an average of 3% of deployment days (Figure 4A). The mean proportion of deployment days with a sei whale ‘possible detection’ also decreased significantly, from 53% in 2022–2024 to 2% in 2025 (p < 0.0001, Supplementary Table 3) (Figure 4B).

Figure 4

In contrast, blue whale detections increased markedly in 2025. The mean proportion of deployment days on which blue whales were ‘detected’ increased significantly from 35% in 2022–2024 to 66% in 2025 (p < 0.0001, Supplementary Table 3), an 89% increase (Figure 4C). The mean proportion of deployment days with at least one blue whale ‘possible detection’ in 2025 was 60%, which did not differ significantly from the 57% in 2022-2024 (p = 0.1135, Supplementary Table 3).

3.2.1 Manual analysis of 2025 archival data

Fourteen tally periods were scored ‘possibly detected’ for sei whales in NRT in 2025. Manual analysis resulted in one tally period scored ‘probable sei whale’ (7.1%), six scored ‘definite blue whale’ (42.9%), five scored ‘probable blue whale’ (35.7%), and two scored ‘unknown’ (14.3%) because too few calls were present to confidently determine species identity. The subsequent precision of NRT ‘possible detection’ scores was 0.08. The tally period manually scored ‘probable sei whale’ also contained evidence of blue whale presence and was therefore also scored ‘probable blue whale’, as amplitude differences between the species’ putative calls suggested there may have been two calling animals present. As such, the period was appropriately identified in NRT as possibly containing a sei whale.

3.3 Variability in blue whale detections

3.3.1 2022–2024 GLMM

For Model 1 (Equation 1), only one candidate model fell within a ΔAIC_C ≤ 2: the global model including Season, DSZ, and Year (Equation 1). For Model 2 (Equation 2), four candidate models fell within a ΔAIC_C ≤ 2; unlike Model 1, the top-ranked model included Season and DSZ. Autocorrelation functions confirmed that binning datetime into 1-hour increments (Model 1) and using date (Model 2) most effectively reduced temporal autocorrelation (Supplementary Figure 2). Analyses showed no concerning multicollinearity in either model, and simulated residuals showed that both models fit the data well, with only minor deviation observed in Model 1.

Both Models 1 and 2 showed significant relationships between Season and DSZ with blue whale detections (Table 4). Detection odds were substantially higher in fall than in either spring (272-fold and 564-fold increases in Models 1 and 2, respectively) or summer (12-fold and 143-fold increases in Models 1 and 2, respectively), with the largest contrast occurring between spring and fall (Table 5). Detection odds were also higher in summer compared to spring, highlighting a clear seasonal progression in blue whale detection probability from spring to summer to fall (Table 5).

Model 1 also showed DSZ effects: DSZs A-B and A-C differed significantly (Table 5), with pairwise comparisons indicating that detection odds in DSZ A were ~2.3 times higher than in DSZ B and ~2.8 times higher than in DSZ C (Table 5). No other DSZ contrasts were significant in Model 1, and none were significant in Model 2 (Table 5). Year effects in Model 1 were also non-significant, suggesting that blue whale acoustic detection was relatively consistent across years (Table 5).

Figure 5A shows the probability that any 15-minute tally period (96 possible per day) is scored ‘detected’ for blue whales, based on Model 1. Because detections were aggregated across six deployments over three years, tally periods scored ‘not detected’ greatly outnumbered those scored ‘detected’, resulting in low overall detection probabilities. For comparison, Figure 5B shows the probability that a deployment day contains at least one tally period scored ‘detected’, based on Model 2. Both panels reflect the broader trends described above. For example, detection probability is noticeably higher in fall than in summer or spring and decreases from west (DSZ A) to east (DSZ C) (Figure 5). Although DSZ E has a relatively high detection probability at the tally period scale, its confidence interval is wide, likely reflecting the small number of detections recorded in DSZ E during 2022-2024, all of which occurred in spring. In contrast, DSZ E has the lowest detection probability of all the DSZs in the daily model, and the uncertainty around this estimate is comparatively smaller (Figure 5B). No observable differences in detection probability were found among years (Figure 5A).

Figure 5

3.3.2 2025 GLMM

For Model 3 (Equation 3), only one candidate model fell within a ΔAIC_C ≤ 2: the global model including DSZ, Season, and their interaction (Equation 3). For Model 4 (Equation 4), two candidate models fell within a ΔAIC_C ≤ 2; unlike Model 3, the top-ranked model only included Season. Autocorrelation functions confirmed that binning datetime into 4-hour increments (Model 3) and using date (Model 4) most effectively reduced temporal autocorrelation (Supplementary Figure 3). Analyses showed no concerning multicollinearity in either model, and simulated residuals showed that both models fit the data well.

Both Models 3 and 4 showed that blue whale detections varied significantly with Season (Table 4). Model 3, which included the Season x DSZ interaction, also showed significant relationships between detections, DSZ, and the spatiotemporal interaction (Table 4). Pairwise comparisons indicated that all seasonal contrasts were significant in Model 4; however, without the interaction term, Model 4 (daily scale) provided less detail than Model 3. In contrast, Model 3 (tally period scale) identified significant Season x DSZ interaction effects for all fall and several spring comparisons (Table 6).

Across both models, detection odds were consistently highest in the fall, followed by summer and then spring, and this seasonal pattern was observed across all DSZs in Model 3 (Table 6). In Model 3, the strongest interaction occurred in fall, where all DSZ contrasts (A-B, A-C, and B-C) were significant (Table 6). During fall, detection odds were highest in DSZ A, followed by DSZ B, and lowest in DSZ C. In spring, DSZ A-B and A-C comparisons were significant, with lower detection odds in DSZ A than in DSZs B or C, indicating an east-to-west decrease in detection probability (C > B > A) (Table 6). This pattern reversed in fall, when detection probability increased east-to-west (C < B < A) (Table 6). No significant differences among DSZs were found during summer, and no other interaction contrasts were significant (Table 6).

The probability that one 15-minute tally period is scored ‘detected’ for blue whales in Model 3 was low because detections were irregular relative to the total number of tally periods (Figure 6A). For comparison, Figure 6B shows the probability that a deployment day contains at least one tally period scored ‘detected’, based on Model 4. Both panels in Figure 6 reflect the seasonal trend described above: detection probability is highest in the fall, followed by summer and then spring (Figure 6). Because the daily scale model did not include the Season x DSZ interaction, it did not capture the spatiotemporal patterns evident in the tally period model. For example, in Figure 6A, the east-to-west increase in detection probability is visible during fall. Summer detection probabilities show no observable differences among DSZs, indicating a more even spatial distribution of blue whale detections during that season, while the very small effect size in spring makes differences among DSZs difficult to interpret (Figure 6A).

Figure 6

3.3.3 Blue whale call category analysis

Across all three seasons monitored in 2025, blue whale audible calls contributed to more NRT detections than infrasonic calls, indicating that audible calls are the predominant call category in the region (Figure 7). Infrasonic detections were most frequent in fall, followed by summer and then spring (Figure 7). Observed detections of both call categories differed significantly from expected values for each season (Supplementary Table 4). In fall, both audible and infrasonic calls occurred significantly more often than expected (Figure 7C). In summer, observed proportions of both call categories were closest to expected values, although both were detected less often than expected (Figure 7B). In spring, audible and infrasonic detections were significantly lower than expected, and no infrasonic calls were detected (Figure 7A). Together, these results show that audible and infrasonic call detections exhibit clear seasonality in blue whale acoustic occurrence.

Figure 7

3.4 Blue whale call library analyses

Across the three 2025 deployments (May–November), 236 of 764 (30.9%), 301 of 685 (43.9%) and 263 of 855 (30.8%) blue whale audible call ‘detections’ (BWDs) from Fundy, Peggy and Sable/Scotia, respectively, contained at least one LFDCS-classified blue whale call. Interestingly, Peggy, despite having the fewest BWDs, had the highest number of classified calls, while Fundy had the fewest overall. Classified blue whale calls were an optional criterion for scoring tally periods ‘detected’ for blue whales in 2025 (i.e., opportunistic scoring was possible using other criteria in the absence of LFDCS classifications, Table 1). Among the 2025 BWDs that contained at least one classified call, LFDCS classifications were the deciding factor for ‘detected’ scores in only 74 of 236 (31.4%), 108 of 301 (35.9%), and 75 of 236 (28.5%) of Fundy, Peggy and Sable/Scotia BWDs, respectively.

Overall, the new blue whale call library classified calls in an average of 35% of BWDs across the three deployments, but only an average of 32% (257 total) of those detections required the classification to achieve a ‘detected’ score. Therefore, the majority of BWDs (on average, 65%) were scored entirely based on unclassified blue whale calls, indicating that the LFDCS exhibited poor classification performance at the tally period scale. However, classification performance improved substantially when evaluated on a daily timescale. Of the 148 days on which blue whales were detected in NRT in 2025, 132 days (89%) included at least one detection containing an LFDCS-classified call, reducing the missed-classification rate to 11% at the daily level.

The number of calls classified by the LFDCS using the blue whale call library differed drastically by call type. The distribution of classified call types across the 2025 deployments is shown in Figure 8. Call type 30, the ‘short downsweep’, was classified most frequently, and this pattern was consistent across all deployments. Call type 33, the ‘steep downsweep’, was the next most frequently classified type. ‘Medium downsweeps’ (call type 31), ‘arches’ (call type 32) and ‘long downsweeps’ (call type 34) were classified least often.

Figure 8

4 Discussion

4.1 Sei and blue whale NRT detection

This study evaluated a new blue whale call library for the LFDCS and updated sei and blue whale detection protocols for glider-based NRT monitoring in the LC. Daily validation scores showed that species acoustic discrimination improved following these updates; blue whales were detected more accurately, and sei whale misclassifications were reduced. Since 91% of sei whale detections in previous years were manually attributed to blue whales (MacDonald, 2025), the significant decrease in sei whale detections and corresponding increase in blue whale detections in 2025 indicates that NRT validations have become more reliable. These results highlight the importance of continually assessing and refining NRT detection systems to optimize detector, software, or protocol performance and ensure effective species discrimination (Baumgartner et al., 2014; Kowarski et al., 2020).

Manual analysis confirmed misclassifications were reduced and true sei whale detections were not missed. Although most (11 of 14) tally periods scored ‘possibly detected’ for sei whales in 2025 were misclassified, the more conservative 2025 sei whale protocol resulted in zero sei whale ‘detections’ and far fewer ‘possible detections’, thereby reducing overall misclassification rates. Precision of sei whale ‘possible detections’ increased from 0.01 (2020-2023) to 0.08 in 2025 (section 3.2.1). Although precision remains low, indicating that misclassifications still occur, the substantial reduction in sei whale detections limits the total number of misclassifications. Importantly, the only tally period manually scored as containing sei whale presence (‘probable sei whale’) was also correctly scored ‘possibly detected’ in NRT, suggesting that the updated procedure can capture true sei whale acoustic occurrence. Together, these results indicate that the updated protocols are adequate for future NRT monitoring in the GoSL.

Improving NRT detections of these species is important because they provide foundational species occurrence data. Acoustic occurrence supports numerous ecological applications, including estimating population density (e.g., Dawson and Efford, 2009), characterizing species distribution and habitat use (e.g., Davis et al., 2020; Fenton, 1997), and informing species-specific monitoring for conservation action (Baumgartner et al., 2019; Indeck et al., 2025a; Van Parijs et al., 2009). More reliable blue whale acoustic detections will therefore expand current knowledge of the Northwest Atlantic population in the GoSL. Additionally, the absence of sei whale acoustic detections in 2025 more accurately reflects visual observations from the region than in previous years (COSEWIC, 2019; Gomez et al., 2020; Johnson et al., 2021; Lesage et al., 2017; Moors-Murphy et al., 2019; Ramp and Sears, 2013; Sears and Calambokidis, 2002), providing the most accurate indication to date of this species’ acoustic presence in the GoSL. These improvements are valuable, as they can guide future studies and resource allocation toward identifying and monitoring regions important for each species.

4.2 Blue whale detection in the Laurentian Channel

4.2.1 Factors related to blue whale acoustic detection

Modelling demonstrated that blue whale acoustic presence varied with season and DSZ, but not year (Tables 4-6, Figures 5, 6). Across all models, detection probability increased as the seasons progressed, with the highest probability in the fall. The 2022–2024 models showed that detection probability was highest in DSZ A, while the Season x DSZ interaction in the 2025 tally period model revealed that spatial patterns in blue whale detection differed among seasons. Although the two daily models could not resolve fine-scale interactions between fixed effects, the seasonal and locational patterns they identified were consistent with those observed in the tally period models and had less uncertain probability estimates. Because daily presence is often more interpretable and directly relevant for management, the daily scale estimates may be particularly useful for predicting blue whale detections in the GoSL.

The seasonal and spatial patterns revealed by all models, especially the Season x DSZ interaction in the 2025 tally period model, align well with current knowledge of the Northwest Atlantic blue whale population. In spring, blue whale acoustic presence was extremely low, but detections were statistically more likely in the southeast (DSZ C) than the northwest (DSZ A) of the LC. Both the density and distribution of detections reversed in the fall (Table 6). It remains unknown whether acoustic detection density is related to whale density, but blue whales are known to occur year-round in the GoSL (Lesage et al., 2017; Simard et al., 2016) and forage seasonally in the Lower St. Lawrence Estuary (Maps et al., 2015), which is closest to DSZ A. This suggests that higher fall detections in the western domain may reflect proximity to feeding areas. Visual observations from August to November support this pattern (Lesage et al., 2017; Sears and Calambokidis, 2002; Sears et al., 1990). The DSZs also overlap with critical habitat for blue whales. The Cabot and Honguedo Straits (encompassing DSZ A) are important transit corridors for the species (Fisheries and Oceans Canada, 2020). The fact that both the 2022–2024 and 2025 models showed similar seasonal and spatial trends, and that regions with higher predicted detection probabilities correspond to documented blue whale occurrence, indicates that the models effectively captured variability in blue whale acoustic presence in the LC.

4.2.2 Seasonality in blue whale call category use

Blue whale call category detection in the LC was strongly seasonally influenced. Infrasonic call detections were most frequent in the fall and least common in the spring (Figure 7), which is consistent with infrasonic call production documented in the GoSL from August to October (Berchok et al., 2006), as well as on the Scotian Shelf and off the coasts of Newfoundland and Labrador, where detections peak from November to January (Delarue et al., 2022; Moors-Murphy et al., 2019; Wingfield et al., 2022). Based on the proportions of call detections observed here (Figure 7) and in other glider-based monitoring studies in the region (Indeck et al., 2024), audible calls are the predominant call category produced and detected in the GoSL foraging region. This aligns with research linking blue whale audible calls to feeding behavior (Oleson et al., 2007a, b). As these calls are detected most frequently, continued optimization of audible call discrimination is essential for accurate NRT monitoring. The protocol and library updates implemented for the 2025 season, which focused on improving correct detection of blue whale audible calls, are therefore well justified and represent an important first step toward enhancing blue whale detection in the LC.

4.3 Evaluation of the blue whale call library

In 2025, an average of 35% of periods scored ‘detected’ for blue whales in NRT based on audible calls contained at least one LFDCS-classified call. Most of these were either call type 30 (‘short downsweep’) or call type 33 (‘steep downsweep’) (Figure 8). This result closely matches that observed by MacDonald (2025), who found that 31% of manually validated ‘definite blue whale’ periods contained at least one LFDCS-classified call, with call types 30 and 33 classified most often. Therefore, the new library performed similarly in both field and archival contexts. Because 257 blue whale ‘detections’ in 2025 were attributable solely to LFDCS classifications, the blue whale library remains an important tool for NRT monitoring.

However, the library did not perform as optimally as possible. Many blue whale pitch tracks observed in NRT were not classified by the LFDCS in 2025. On average, 65% of blue whale audible call ‘detections’ were scored opportunistically, meaning that although each contained sufficient evidence to score the period ‘detected’, the LFDCS did not classify calls it theoretically could have. Additionally, three of the five blue whale call types in the library were rarely classified, including arch calls (i.e., ‘long downsweeps’) (Figure 8). Arch calls are a common, stereotyped blue whale call type in the GoSL and were frequently used to opportunistically score periods ‘detected’ in 2025 (anecdotal evidence). Their stereotypy and prevalence should make them good candidates for automated detection and classification (Baumgartner et al., 2020; Gibb et al., 2019), yet the low number of arch calls classified suggests that the arch call exemplars in the library did not statistically match those observed in the GoSL. Poor library performance can occur when exemplars originate from a different geographic region than the one in which the library is applied (e.g., Baumgartner et al., 2014). The new library contained blue whale call exemplars recorded off the coasts of Newfoundland and Nova Scotia, as well as in the Honguedo Strait (Babin et al., 2025). However, the distribution and characteristics of these exemplars may not fully represent the variation present in GoSL call types, which seems to be the case here.

The library’s relatively poor classification performance indicates clear room for improvement. A new LFDCS library containing only GoSL-recorded exemplars has been compiled and will be tested, with the expectation that a regionally specific library will further improve species discrimination. An additional refinement could include adjusting relative prior probabilities within the LFDCS software, to guide LFDCS classifications toward acoustically prevalent species in the region (in this case, blue whales) (see Baumgartner and Mussoline, 2011). These updates may allow the blue whale library’s performance to approach that of other LFDCS libraries, which have shown correct classification rates of 52% and 65% for individual right and sei whale calls, respectively, in archival data (Baumgartner and Mussoline, 2011), and recall rates of 33-73% for sei, humpback, right and fin whale detections in NRT (Baumgartner et al., 2019). Continued development of the blue whale audible call library is essential. Blue whale acoustic detection in the North Atlantic is often based on infrasonic calls (e.g., Davis et al., 2020; Simard et al., 2016), so optimizing audible call classification in NRT ensures that acoustic data are not biased by age or sex and reflect the call types most commonly detected on GoSL foraging grounds. Additionally, LFDCS classifications provide objective evidence for analysts during NRT decision-making, which is crucial for standardizing monitoring practices.

4.4 Primary sources of uncertainty

One limitation of this study is that it could not isolate the individual contributions of the new blue whale call library and the protocol modifications, as all updates were implemented simultaneously. Although this constrains our ability to evaluate the effectiveness of each update independently, it does not impact the main conclusion that the combined updates were beneficial. Moving forward, this means the three updates should continue to be used and optimized together, or further research will be required to assess their individual efficacy. Additionally, differences in glider effort across deployments included in the 2022–2024 model led to complete separation in the blue whale call category data, restricting the analyses we could run. Estimating interactions in GLMMs requires observations for every combination of predictor variable (e.g., monitoring in each season, DSZ, and year), which was not the case for the 2022–2024 dataset. As a result, interaction terms such as Season x DSZ were not included in the 2022–2024 models, even though they could have provided additional insight into spatiotemporal variation in blue whale detections during those years. However, since the trends observed were similar to those of the 2025 model that included the interaction, the overall patterns in blue whale detection probability were likely captured. Finally, since there were no blue whale detections attributable to infrasonic calls in the spring of 2025, complete separation prevented GLMMs from being run for the call category analysis and a chi-square test was used instead. Although this approach reduced inferential power, the qualitative differences still provided clear indications of which seasons were associated with higher observed detections of audible and infrasonic calls.

Another consideration is the efficacy of detecting blue whale calls under variable ambient noise levels. Many non-biological processes contribute to ambient noise, such as weather (e.g., wind, rain), vessel traffic, and platform-related noise (Gehrmann et al., 2023; Hildebrand, 2009; Indeck et al., 2024). Since the LFDCS firmware only pitch-tracks calls with an average amplitude of at least 10 dB above background noise (Baumgartner and Mussoline, 2011), elevated noise levels can restrict the number of calls detected in NRT. In the Honguedo Strait, glider-based monitoring research has shown that noise in the 63 Hz one-third octave band (centered over the primary frequency range of blue whale audible calls), increases by a few decibels inside the shipping lanes compared to outside them. This suggests that the shipping lanes may have elevated noise that restricts call detections; however, this effect was considered minimal for blue whale detection probabilities overall (Gehrmann et al., 2023; Indeck et al., 2024). Monitoring within the shipping lanes does, however, increase the likelihood of nearby vessel transits, which can temporarily reduce call detection probability (Gehrmann et al., 2023; Indeck et al., 2024). Although this study did not assess noise levels directly, it is important to acknowledge the potential bias they may introduce into the acoustic detections assessed here, particularly as vessel noise is the primary contributor to ambient noise at blue whale call frequencies, and the deployments assessed in this study transited the DSZs.

Finally, it is important to consider the results of this study in the context of blue whale detection ranges. Previous work has estimated that blue whale infrasonic calls can be detected at distances of up to 200 km in the Southern Ocean (Širović et al., 2007), 10–190 km in the Gulf of Alaska (Stafford et al., 2007), and that audible calls can propagate 30 km in the Indian Ocean (Torterotot et al., 2023) and more than 100 km for a subset of audible calls in the GoSL (personal communication). However, many factors including bathymetric features, environmental conditions (e.g., sound-speed profiles), background noise, source and receiver depth, and call characteristics (e.g., source level, frequency, duration) can influence sound propagation and detection range. As a result, spatial variability in blue whale detections, such as the patterns modelled among the DSZs, may be misestimated if calls produced in one region are detected in another. Despite this uncertainty, spatial trends observed here are consistent with documented blue whale occurrence. Detection probability was highest in DSZs toward the west of the GoSL, where blue whales are frequently observed in foraging areas (Lesage et al., 2017; Sears and Calambokidis, 2002; Sears et al., 1990). Although the DSZs (~83–87 km long) may be too fine-scale to fully isolate detections of longer-ranging audible calls, especially when gliders operate near DSZ boundaries, detection ranges are highly dynamic. Therefore, although it is possible some detections attributed to one DSZ originated from a whale in another, this uncertainty should not diminish the importance of the detections themselves: any whale present within or near the DSZs is susceptible to ship strikes and therefore important to detect at any spatial scale. The challenge then becomes determining the most appropriate management strategy for the species, given the detection range information available.

4.5 Management implications in the Northwest Atlantic

In the GoSL, blue whales are endangered and susceptible to vessel-related mortality (Fisheries and Oceans Canada, 2020; Indeck et al., 2024). Many areas designated as critical foraging or transiting habitat are in close proximity to the region’s shipping lanes (Figure 3) (Fisheries and Oceans Canada, 2020). Conservation measures that reduce vessel-strike risk in these habitats are therefore crucial for sustaining the population. Blue whales are acoustically prevalent in the region, and this study shows that we are now better equipped to detect them in NRT. Vessel-strike mitigation programs will therefore be important to consider, but the most effective strategy has yet to be determined.

NRT detections of NARWs from the same glider deployments included in this study were used by Transport Canada to implement 15-day, 10-knot vessel speed limits in the DSZ where the detection occurred. If a similar dynamic management approach were implemented for blue whales, the improvements to blue whale NRT detection accuracy demonstrated here would help. However, this study found that in 2025, after updates improved blue whale detection, blue whales were detected on an average of 66% of deployment days in the LC, and the mean detection probability in fall exceeded 0.9 (Figures 4, 6B). Thus, the current NARW management framework would effectively lead to near-continuous implementation of the dynamic zones; i.e., a static management area for blue whales. Future research on detection ranges of different blue whale call types will also be crucial for evaluating the feasibility of using acoustics for management in the DSZs. For example, if a whale could be 150 km away from the detection platform, implementing vessel speed restrictions may not provide meaningful risk reduction for either the animal or industry.

Seasonal or voluntary speed-restricted zones represent static management approaches that have been used to reduce ship-strike risk for NARWs (Laist et al., 2014) and Pacific blue whales (McKenna et al., 2012), respectively, and may offer options for protecting blue whales in the GoSL that are less reliant on acoustic detections. However, the DSZs are vital corridors for ship traffic, connecting major Canadian cities to global trade routes, so seasonal or voluntary restrictions may not be economically or logistically feasible given the high volume of ship traffic. Overall, the trade-offs associated with different management strategies highlight the need for continued, accurate monitoring and further research on blue whales in the LC. Improved understanding of blue whale acoustic occurrence will help inform managers and policymakers and support the development of effective conservation plans.

4.6 Implications for other systems

In the GoSL, blue whales are common while sei whales are rare, and the protocol updates implemented here were designed so that acoustic detections more accurately reflected the known presence of both species. The GoSL-specific protocol updates performed well; by making the sei whale protocol more restrictive and including objective evidence for blue whale detections, correct blue whale detections increased. Although it is possible that true sei whale detections were missed due to the more restrictive criteria, this trade-off is appropriate for the region, as missing the occasional sei whale is preferable to falsely detecting a species that is rarely present while overlooking detections of a commonly occurring, locally important species.

However, in other regions the relative abundance of sei and blue whales may differ. Sei whales may be common and blue whales rare, one species may occur without the other, or both may be common. In these cases, the performance of the GoSL protocols may change. For example, in a system where blue whales are rare or absent and sei whales are common, applying the GoSL protocols could lead to missed sei whale detections and an unreasonably high number of blue whale misclassifications. In regions where both species are common, the GoSL protocols would likely continue to capture blue whale detections effectively, as requiring either objective evidence (classified calls) or concave-down calls (opportunistic scoring) ensures that detections are based on high-quality, diagnostic evidence. Since the GoSL protocols detect sei whale acoustic presence using objective evidence and diagnostic doublets or triplets, their occurrence should also be adequately detected in such regions; however, sei whale calls that do not meet these more conservative criteria may be missed more often than under the original protocol. Importantly, the acceptable balance between misclassifications and missed detections depends on the objectives of a given monitoring program. Detecting rarer species can sometimes be more important. For example, humpback and NARW calls can overlap acoustically (Mussoline et al., 2012; Wilder et al., 2022), and in regions where their distributions overlap and NARWs are dynamically managed, misclassifying a humpback as a NARW can have severe management consequences.

In general, caution should be exercised when using protocols that require only one classified call for a ‘detection’ or ‘possible detection’ because NRT detection scores are intended to be conservative and minimize false detections (Baumgartner et al., 2019). In the GoSL, a one-classified-call threshold was included as an option in the updated blue whale protocol because preliminary testing of the blue whale library demonstrated that this criterion was appropriate (MacDonald, 2025). Additionally, the objective statistical evidence provided by LFDCS blue whale classifications, now directly comparable to the sei whale call library, means that blue whale classifications are selected over sei whale classifications with a high level of confidence. However, using a one-classified-call threshold can increase the risk of misclassification. Prior to the updated protocol, sei whales could be scored ‘possibly detected’ using one or more classified calls, which contributed to misclassifications in the GoSL, although at the time no blue whale library existed for comparison (MacDonald, 2025). Therefore, for regions where sei and blue whale distributions resemble those in the GoSL, or where both species are common, the one-classified-call threshold established in this study could be used for blue whale detection, alongside opportunistic scoring criteria. However, for regions with different species distributions, or where prior knowledge of species occurrence is limited, detections based on LFDCS classifications should require two or more classified calls for both species to maintain low misclassification rates.

4.7 Conclusion

As methods for species acoustic detection and classification continue to advance, it is important that species discrimination remains accurate. This study demonstrated that implementing a new blue whale call library alongside updated NRT species detection protocols effectively improved acoustic discrimination between previously misclassified sei and blue whales in the LC (MacDonald, 2025). Improving discrimination between these species is crucial since blue whales are endangered and inhabit a socioeconomically important region. When using NRT data to inform conservation measures, where biological and economic trade-offs must be balanced, species detection must be reliable so that any mitigation action is justifiable. Overall, this study highlights the importance of continually evaluating and refining acoustic detection methods and provides a strategy to do so effectively. However, monitoring groups facing similar species discrimination challenges should first evaluate regional species distributions and tailor their objectives accordingly, before implementing the updates tested in the GoSL.

References

• 1

  BabinA. B.MacklinG. F.McLeanS. N.WingfieldJ. E.BeslinW. A. M.AdamsM. J.et al. (2025). Characterization of blue whale (Balaenoptera musculus) audible calls recorded in the Northwest Atlantic. J. Acoust. Soc Am.158, 3767–3780. doi: 10.1121/10.0039869

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 2

  BartońK. (2025). Mumin: Multi-Model Inference Version 1.48.11. Available online at: https://CRAN.R-project.org/package=MuMIn (Accessed January 10, 2026).

  • Google Scholar
  • View reference in article

• 3

  BaumgartnerM. F.BallK.PartanJ.PelletierL.-P.BonnellJ.HotchkinC.et al. (2021). Near real-time detection of low-frequency baleen whale calls from an autonomous surface vehicle: Implementation, evaluation, and remaining challenges. J. Acoust. Soc Am.149, 2950–2962. doi: 10.1121/10.0004817

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 4

  BaumgartnerM. F.BonnellJ.CorkeronP. J.Van ParijsS. M.HotchkinC.HodgesB. A.et al. (2020). Slocum gliders provide accurate near real-time estimates of baleen whale presence from human-reviewed passive acoustic detection information. Front. Mar. Sci.7. doi: 10.3389/fmars.2020.00100

  • CrossRef
  • Google Scholar
  • View reference in article

• 5

  BaumgartnerM. F.BonnellJ.Van ParijsS. M.CorkeronP. J.HotchkinC.BallK.et al. (2019). Persistent near real‐time passive acoustic monitoring for baleen whales from a moored buoy: System description and evaluation. Methods Ecol. Evol.10, 1476–1489. doi: 10.1111/2041-210X.13244

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 6

  BaumgartnerM. F.FratantoniD. M.HurstT. P.BrownM. W.ColeT. V. N.Van ParijsS. M.et al. (2013). Real-time reporting of baleen whale passive acoustic detections from ocean gliders. J. Acoust. Soc Am.134, 1814–1823. doi: 10.1121/1.4816406

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 7

  BaumgartnerM. F.MussolineS. E. (2011). A generalized baleen whale call detection and classification system. J. Acoust. Soc Am.129, 2889–2902. doi: 10.1121/1.3562166

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 8

  BaumgartnerM. F.StaffordK. M.WinsorP.StatscewichH.FratantoniD. M. (2014). Glider-based passive acoustic monitoring in the arctic. Mar. Technol. Soc J.48, 40–51. doi: 10.4031/MTSJ.48.5.2

  • CrossRef
  • Google Scholar
  • View reference in article

• 9

  BaumgartnerM. F.Van ParijsS. M.WenzelF. W.TremblayC. J.Carter EschH.WardeA. M. (2008). Low frequency vocalizations attributed to sei whales (Balaenoptera borealis). J. Acoust. Soc Am.124, 1339–1349. doi: 10.1121/1.2945155

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 10

  BeauchampJ.BouchardH.de MargerieP.OtisN.SavariaJ.-Y. (2009). “ Recovery Strategy for the blue whale (Balaenoptera musculus), Northwest Atlantic population, in Canada [FINAL,” in Species at Risk Act Recovery Strategy Series ( Fisheries and Oceans Canada, Ottawa), 1–62.

  • Google Scholar
  • View reference in article

• 11

  BerchokC. L.BradleyD. L.GabrielsonT. B. (2006). St. Lawrence blue whale vocalizations revisited: Characterization of calls detected from 1998 to 2001. J. Acoust. Soc Am.120, 2340–2354. doi: 10.1121/1.2335676

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 12

  BermantP. C.BronsteinM. M.WoodR. J.GeroS.GruberD. F. (2019). Deep machine learning techniques for the detection and classification of sperm whale bioacoustics. Sci. Rep.9, 12588. doi: 10.1038/s41598-019-48909-4

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 13

  BoisseauO.GillespieD.LeaperR.MoscropA. (2008). Blue (Balaenoptera musculus) and fin (B. physalus) whale vocalisations measured from northern latitudes of the Atlantic Ocean. J. Cetacean Res. Manage.10, 23–30. doi: 10.47536/jcrm.v10i1.656

  • CrossRef
  • Google Scholar
  • View reference in article

• 14

  BrooksM. E.KristensenK.van BenthemK. J.MagnussonA.BergC. W.NielsenA.et al. (2017). glmmTMB balances speed and flexibility among packages for zero-inflated generalized linear mixed modeling. R J.9, 378–400. doi: 10.32614/RJ-2017-066

  • CrossRef
  • Google Scholar
  • View reference in article

• 15

  BuchanS. J.MahúR.WuthJ.Balcazar-CabreraN.GutierrezL.NeiraS.et al. (2020). An unsupervised Hidden Markov Model-based system for the detection and classification of blue whale vocalizations off Chile. Bioacoustics29, 140–167. doi: 10.1080/09524622.2018.1563758

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 16

  CalderanS.MillerB.CollinsK.EnsorP.DoubleM.LeaperR.et al. (2014). Low-frequency vocalizations of sei whales (Balaenoptera borealis) in the Southern Ocean. J. Acoust. Soc Am.136, EL418–EL423. doi: 10.1121/1.4902422

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 17

  CerchioS.WeirC. R. (2022). Mid-frequency song and low-frequency calls of sei whales in the Falkland Islands. R. Soc Open Sci.9, 220738. doi: 10.1098/rsos.220738

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 18

  COSEWIC (2012). “ COSEWIC status appraisal summary on the Blue Whale Balaenoptera musculus, Atlantic population, in Canada,” in Committee on the Status of Endangered Wildlife in Canada ( Committee on the Status of Endangered Wildlife in Canada, Ottawa), 1–12. Available online at: https://www.Canada.ca/en/environment-climate-change/services/species-risk-public-registry.html (Accessed April 7, 2026).

  • Google Scholar
  • View reference in article

• 19

  COSEWIC (2019). “ COSEWIC assessment and status report on the Sei Whale Balaenoptera borealis, Atlantic population, in Canada,” in Committee on the Status of Endangered Wildlife in Canada ( Committee on the Status of Endangered Wildlife in Canada, Ottawa), 1–48. Available online at: https://www.Canada.ca/en/environment-climate-change/services/species-risk-public-registry.html (Accessed April 7, 2026).

  • Google Scholar
  • View reference in article

• 20

  CusanoD. A.WileyD.ZehJ. M.KerrI.PensarosaA.ZadraC.et al. (2023). Acoustic recording tags provide insight into the springtime acoustic behavior of sei whales in Massachusetts Bay. J. Acoust. Soc Am.154, 3543–3555. doi: 10.1121/10.0022570

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 21

  DavisG. E.BaumgartnerM. F.CorkeronP. J.BellJ.BerchokC.BonnellJ. M.et al. (2020). Exploring movement patterns and changing distributions of baleen whales in the western North Atlantic using a decade of passive acoustic data. Global Change Biol.26, 4812–4840. doi: 10.1111/gcb.15191

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 22

  DawsonD. K.EffordM. G. (2009). Bird population density estimated from acoustic signals. J. Appl. Ecol.46, 1201–1209. doi: 10.1111/j.1365-2664.2009.01731.x

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 23

  DelarueJ. J.-Y.Moors-MurphyH.KowarskiK. A.DavisG. E.UrazghildiievI. R.MartinS. B. (2022). Acoustic occurrence of baleen whales, particularly blue, fin, and humpback whales, off eastern Canada 2015-2017. Endanger. Species Res.47, 265–289. doi: 10.3354/esr01176

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 24

  Español-JiménezS.BahamondeP. A.ChiangG.HäussermannV. (2019). Discovering sounds in Patagonia: Characterizing sei whale (Balaenoptera borealis) downsweeps in the south-eastern Pacific Ocean. Ocean Sci.15, 75–82. doi: 10.5194/os-15-75-2019

  • CrossRef
  • Google Scholar
  • View reference in article

• 25

  FentonM. B. (1997). Science and the conservation of bats. J. Mammal.78, 1–14. doi: 10.2307/1382633

  • CrossRef
  • Google Scholar
  • View reference in article

• 26

  Fisheries and Oceans Canada (2020). “ Action plan for the Blue Whale (Balaenoptera musculus), Northwest Atlantic population, in Canada,” in Species at Risk Act Action Plan Series ( Fisheries and Oceans Canada, Ottawa), 1–23.

  • Google Scholar
  • View reference in article

• 27

  GehrmannR. A. S.BarclayD. R.JohnsonH.ShajahanN.NoletV.DaviesK. T. A. (2023). Ambient noise levels with depth from an underwater glider survey across shipping lanes in the Gulf of St. Lawrence, Canada. J. Acoust. Soc Am.154, 1735–1745. doi: 10.1121/10.0020908

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 28

  GibbR.BrowningE.Glover-KapferP.JonesK. E. (2019). Emerging opportunities and challenges for passive acoustics in ecological assessment and monitoring. Methods Ecol. Evol.10, 169–185. doi: 10.1111/2041-210X.13101

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 29

  GomezC.Konrad ClarkeC.VanderlaanA.Moors-MurphyH.MarotteE.LawsonJ.et al. (2020). Identifying priority areas to enhance monitoring of cetaceans in the Northwest Atlantic Ocean. Can. Tech. Rep. Fish. Aquat. Sci.3370, 1–103.

  • Google Scholar
  • View reference in article

• 30

  HartigF. (2024). DHARMa: Residual diagnostics for hierarchical (multi-level/mixed) regression models Version R package version 0.4.7. Available online at: https://CRAN.R-project.org/package=DHARMa (Accessed April 12, 2026).

  • Google Scholar
  • View reference in article

• 31

  HildebrandJ. A. (2009). Anthropogenic and natural sources of ambient noise in the ocean. Mar. Ecol. Prog. Ser.395, 5–20. doi: 10.3354/meps08353

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 32

  HorwoodJ. (1987). The Sei Whale: Population Biology, Ecology & Management (Beckenham: Croom Helm).

  • Google Scholar

• 33

  IndeckK. L.BaumgartnerM. F.LecavalierL.WhoriskeyF.DaviesK. T. A. (2025a). Passive acoustic gliders are effective monitoring tools for dynamic management plans aimed at mitigating whale-vessel strikes. Conserv. Lett.18, e13102. doi: 10.1111/conl.13102

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 34

  IndeckK. L.BaumgartnerM. F.LecavalierL.WhoriskeyF.Durette-MorinD.PettigrewN. R.et al. (2025b). Glider surveillance for near-real-time detection and spatial management of North Atlantic right whales. Oceanography38, 13–21. doi: 10.5670/oceanog.2025e111

  • CrossRef
  • Google Scholar

• 35

  IndeckK. L.GehrmannR.RichardsonA. L.BarclayD.BaumgartnerM. F.NoletV.et al. (2024). Variation in glider-detected North Atlantic right, blue, and fin whale calls in proximity to high-traffic shipping lanes. Endanger. Species Res.54, 191–217. doi: 10.3354/esr01327

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 36

  JohnsonH.MorrisonD.TaggartC. (2021). WhaleMap: A tool to collate and display whale survey results in near real-time. J. Open Source Software6, 3094. doi: 10.21105/joss.03094

  • CrossRef
  • Google Scholar

• 37

  KassambaraA. (2023). rstatix: Pipe-Friendly Framework for Basic Statistical Tests. Version R package version 0.7.2. Available online at: https://rpkgs.datanovia.com/rstatix/ (Accessed August 30, 2026).

  • Google Scholar

• 38

  K. Lisa Yang Center for Conservation Bioacoustics at the Cornell Lab of Ornithology (2026). Raven Pro: Interactive Sound Analysis Software Version 1.6.5 (Ithaca, NY: The Cornell Lab of Ornithology). Available online at: https://www.ravensoundsoftware.com/. Computer software. (Accessed April 13, 2026)

  • Google Scholar

• 39

  KowarskiK. A.GaudetB. J.ColeA. J.MaxnerE. E.TurnerS. P.MartinS. B.et al. (2020). Near real-time marine mammal monitoring from gliders: Practical challenges, system development, and management implications. J. Acoust. Soc Am.148, 1215–1230. doi: 10.1121/10.0001811

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 40

  LaistD. W.KnowltonA.PendletonD. (2014). Effectiveness of mandatory vessel speed limits for protecting North Atlantic right whales. Endanger. Species Res.23, 133–147. doi: 10.3354/esr00586

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 41

  Lenth and Piaskowski (2025). emmeans: Estimated marginal means, aka least-squares means Version 2.0.0. Available online at: https://CRAN.R-project.org/package=emmeans (Accessed April 12, 2026).

  • Google Scholar

• 42

  LesageV.GavrilchukK.AndrewsR.SearsR. (2017). Foraging areas, migratory movements and winter destinations of blue whales from the western North Atlantic. Endanger. Species Res.34, 27–43. doi: 10.3354/esr00838

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 43

  MacDonaldK. (2025). Audial, Visual and Statistical Discrimination of Similar Calls Made by Sei and Blue Whales (UNB Scholar). Fredericton, Canada: Honours thesis, University of New Brunswick. Available online at: https://unbscholar.lib.unb.ca/handle/1882/38421 (Accessed October 21, 2025).

  • Google Scholar

• 44

  MacklinG. F.Moors-MurphyH. B.LeonardM. L. (2024). Characteristics and spatiotemporal variation of sei whale (Balaenoptera borealis) downsweeps recorded in Atlantic Canada. J. Acoust. Soc Am.155, 145–155. doi: 10.1121/10.0024213

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 45

  MapsF.PlourdeS.McQuinnI. H.St-Onge-DrouinS.LavoieD.ChasséJ.et al. (2015). Linking acoustics and finite-time Lyapunov exponents reveals areas and mechanisms of krill aggregation within the Gulf of St. Lawrence, eastern Canada. Limnol. Oceanogr.60, 1965–1975. doi: 10.1002/lno.10145

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 46

  McKennaM. F.KatzS. L.ConditC.WalbridgeS. (2012). Response of commercial ships to a voluntary speed reduction measure: Are voluntary strategies adequate for mitigating ship-strike risk? Coast. Manage.40, 634–650. doi: 10.1080/08920753.2012.727749

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 47

  MellingerD. K.ClarkC. W. (2003). Blue whale (Balaenoptera musculus) sounds from the North Atlantic. J. Acoust. Soc Am.114, 1108–1119. doi: 10.1121/1.1593066

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 48

  Moors-MurphyH. B.LawsonJ. W.RubinB.MarotteE.RenaudG.Fuentes-YacoC. (2019). “ Occurrence of blue whales (Balaenoptera musculus) off Nova Scotia, Newfoundland, and Labrador,” in Dfo can. Sci. Advis. Sec. Res. Doc, vol. 2018/007. Ottawa, Canada: DFO Canada Canadian Science Advisory Secretariat, 1–55.

  • Google Scholar

• 49

  MussolineS.RischD.ClarkC.HatchL.WeinrichM.WileyD.et al. (2012). Seasonal and diel variation in North Atlantic right whale up-calls: Implications for management and conservation in the northwestern Atlantic Ocean. Endanger. Species Res.17, 17–26. doi: 10.3354/esr00411

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 50

  OlesonE. M.CalambokidisJ.BurgessW. C.McDonaldM. A.LeDucC. A.HildebrandJ. A. (2007a). Behavioral context of call production by eastern North Pacific blue whales. Mar. Ecol. Prog. Ser.330, 269–284. doi: 10.3354/meps330269

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 51

  OlesonE. M.WigginsS. M.HildebrandJ. A. (2007b). Temporal separation of blue whale call types on a southern California feeding ground. Anim. Behav.74, 881–894. doi: 10.1016/j.anbehav.2007.01.022

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 52

  OuH.AuW. W. L.Van ParijsS.OlesonE. M.RankinS. (2015). Discrimination of frequency-modulated baleen whale downsweep calls with overlapping frequencies. J. Acoust. Soc Am.137, 3024–3032. doi: 10.1121/1.4919304

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 53

  PodosJ.WebsterM. S. (2022). Ecology and evolution of bird sounds. Curr. Biol.32, R1100–R1104. doi: 10.1016/j.cub.2022.07.073

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 54

  PotterJ. R.MellingerD. K.ClarkC. W. (1994). Marine mammal call discrimination using artificial neural networks. J. Acoust. Soc Am.96, 1255–1262. doi: 10.1121/1.410274

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 55

  PrietoR.SilvaM.WaringG.GonçalvesJ. (2014). Sei whale movements and behaviour in the North Atlantic inferred from satellite telemetry. Endanger. Species Res.26, 103–113. doi: 10.3354/esr00630

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 56

  RampC.SearsR. (2013). “ Distribution, densities, and annual occurrence of individual blue whales (Balaenoptera musculus) in the Gulf of St. Lawrence, Canada from 1980-2008,” in Dfo can. Sci. Advis. Sec. Res. Doc, vol. 2012/157. Ottawa, Canada: DFO Canada Canadian Science Advisory Secretariat, 37.

  • Google Scholar

• 57

  RankinS.BarlowJ. (2007). Vocalizations of the sei whale Balaenoptera Borealis off the Hawaiian Islands. Bioacoustics16, 137–145. doi: 10.1080/09524622.2007.9753572

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 58

  RasmussenJ. H.ŠirovićA. (2021). Automatic detection and classification of baleen whale social calls using convolutional neural networks. J. Acoust. Soc Am.149, 3635–3644. doi: 10.1121/10.0005047

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 59

  R Core Team (2025). R: A Language and Environment for Statistical Computing ( R Foundation for Statistical Computing), Vienna, Austria: Computer software. Available online at: https://www.R-project.org (Accessed April 12, 2026).

  • Google Scholar

• 60

  RochaR. C.Jr.ClaphamP. J.IvashchenkoY. (2015). Emptying the oceans: A summary of industrial whaling catches in the 20th century. Mar. Fish. Rev.76, 37–48. doi: 10.7755/MFR.76.4.3

  • CrossRef
  • Google Scholar

• 61

  SanguinetiM.AlessiJ.BrunoldiM.CannarileG.CavalleriO.CerrutiR.et al. (2021). An automated passive acoustic monitoring system for real time sperm whale (Physeter macrocephalus) threat prevention in the Mediterranean Sea. Appl. Acoustics172, 107650. doi: 10.1016/j.apacoust.2020.107650

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 62

  SearsR.CalambokidisJ. (2002). “ Update COSEWIC status report on the Blue Whale Balaenoptera musculus in Canada, in COSEWIC assessment and update status report on the Blue Whale Balaenoptera musculus in Canada,” in Committee on the Status of Endangered Wildlife in Canada (Ottawa). Ottawa: Committee on the Status of Endangered Wildlife in Canada, vi + 1–v32 pp. doi: 10.1016/b978-0-12-804327-1.00070-4

  • CrossRef
  • Google Scholar

• 63

  SearsR.WilliamsonJ. M.WenzelF. W.BérubéM.GendronD.JonesP. (1990). “ Photographic identification of the blue whale (Balaenoptera musculus) in the Gulf of St. Lawrence, Canada,” in Report of the International Whaling Commission, vol. 12. Cambridge, England: Reports of the International Whaling Commission, 335–342.

  • Google Scholar

• 64

  ShiuY.PalmerK. J.RochM. A.FleishmanE.LiuX.NosalE.-M.et al. (2020). Deep neural networks for automated detection of marine mammal species. Sci. Rep.10, 607. doi: 10.1038/s41598-020-57549-y

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 65

  SimardY.RoyN.AulanierF.GiardS. (2016). “ Blue whale continuous frequentations of St. Lawrence habitats from multi-year PAM series,” in Dfo can. Sci. Advis. Sec. Res. Doc. 2016/091. Ottawa, Canada: DFO Canada Canadian Science Advisory Secretariat, v + 14 p.

  • Google Scholar

• 66

  ŠirovićA.HildebrandJ. A.WigginsS. M. (2007). Blue and fin whale call source levels and propagation range in the Southern Ocean. J. Acoust. Soc Am.122, 1208–1215. doi: 10.1121/1.2749452

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 67

  StaffordK. M.MellingerD. K.MooreS. E.FoxC. G. (2007). Seasonal variability and detection range modeling of baleen whale calls in the Gulf of Alaska 1999–2002. J. Acoust. Soc Am.122, 3378–3390. doi: 10.1121/1.2799905

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 68

  ThomasP. O.ReevesR. R.BrownellR. L.Jr. (2016). Status of the world’s baleen whales. Mar. Mammal Sci.32, 682–734. doi: 10.1111/mms.12281

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 69

  TorterototM.SamaranF.RoyerJ.-Y. (2023). Long‐term acoustic monitoring of nonstereotyped blue whale calls in the southern Indian Ocean. Mar. Mammal Sci.39, 594–610. doi: 10.1111/mms.12998

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 70

  Transport Canada (2026a). Transport Canada Actions Taken to Date to Protect North Atlantic Right Whales (Ottawa, Canada: Transport Canada). Available online at: https://tc.Canada.ca/en/marine-transportation/navigation-marine-conditions/protecting-north-atlantic-right-whales-collisions-vessels-gulf-st-lawrence/transport-Canada-actions-taken-date-protect-north-atlantic-right-whales (Accessed April 11, 2026).

  • Google Scholar

• 71

  Transport Canada (2026b). Protecting North Atlantic Right Whales From Collisions With Vessels in the Gulf of St. Lawrence (Ottawa, Canada: Transport Canada). Available online at: https://tc.Canada.ca/en/marine-transportation/navigation-marine-conditions/protecting-north-atlantic-right-whales-collisions-vessels-gulf-st-lawrence (Accessed April 11, 2026).

  • Google Scholar

• 72

  TrawickiM. B. (2021). Multispecies discrimination of whales (cetaceans) using Hidden Markov Models (HMMS). Ecol. Inf.61, 101223. doi: 10.1016/j.ecoinf.2021.101223

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 73

  TremblayC. J.Van ParijsS. M.CholewiakD. (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. doi: 10.1121/1.5110713

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 74

  Van ParijsS. M. V.ClarkC. W.Sousa-LimaR. S.ParksS. E.RankinS.RischD.et al. (2009). Management and research applications of real-time and archival passive acoustic sensors over varying temporal and spatial scales. Mar. Ecol. Prog. Ser.395, 21–36. doi: 10.3354/meps08123

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 75

  WilderJ. M.DavisG. E.DeAngelisA. I.Van ParijsS. M.BaumgartnerM. F.RichardsonA. L.et al. (2022). “ LFDCS reference guide—Version 1.3,” in Noaa Nefsc Passive Acoustics Research GroupWoods Hole: NOAA NEFSC Passive Acoustics Research Group.

  • Google Scholar

• 76

  WingfieldJ.RubinB.XuJ.StanistreetJ.Moors-MurphyH. (2022). Annual, seasonal, and diel patterns in blue whale call occurrence off Eastern Canada. Endanger. Species Res.49, 71–86. doi: 10.3354/esr01204

  • Pubmed Abstract
  • CrossRef
  • Google Scholar