Acoustic occurrence of baleen whales, particularly blue, fin, and humpback whales, off eastern Canada, 2015−2017

: Six baleen whale species occur off eastern Canada, but little is known of their year-round occurrence across this large region. This complicates identifying areas that are important to them and may require critical habitat designation, especially for those species considered at risk. This is particularly true between fall and spring because of a lack of visual survey effort. The main objective of this paper is to provide a year-round and pluriannual description of the minimum acoustic occurrence of baleen whales, particularly blue Balaenoptera musculus , fin B. physalus , and humpback whales Megaptera novaeangliae . We deployed 25 acoustic recorders off eastern Canada between May 2015 and November 2017, and the data were analyzed using a combination of automated detectors and manual validation to identify vocalizations produced by these species. Blue, fin, and humpback whales occurred off eastern Canada year-round, a finding which con-trasts the traditional seasonal latitude migration narrative for these species. The Scotian Shelf region and Flemish Pass−Orphan Basin areas seem particularly important for these animals and should be the focus of future research. Sei B. borealis , minke B. acutorostrata , and North Atlantic right whale Eubalaena glacialis vocalizations also occurred in the data but were not adequately captured by the adopted methodology. Coarse patterns of occurrence are presented for these species as a foundation for more detailed analyses. This study is the first to cover eastern Canadian waters for an extended continuous period and provides a baseline against which future changes can be assessed.


INTRODUCTION
Eastern Canadian waters are home to a rich marine mammal community, with 34 species having reasonable probability of occurrence. Canada's Species at Risk Act (SARA) lists 6 of these species (SARA 2002), including endangered Atlantic blue Balaenoptera musculus and North Atlan tic right whales Eubalaena glacialis and special concern Atlantic fin whales B. physalus. Atlantic sei whales B. borea lis were as-sessed as endangered by the Committee on the Status of Endangered Wildlife in Canada (COSEWIC) in 2019 (COSEWIC 2019) and are currently under consideration for listing under the SARA. Commercial whaling has severely reduced these populations in the past, and the extent to which they have recovered varies but is generally poorly understood. Anthropogenic threats, particularly entanglements in fishing gear, vessel strikes, and underwater noise, affect all baleen whales and are linked directly to the ongoing decline of the North Atlantic right whale (Vanderlaan et al. 2008, van der Hoop et al. 2017, Kenney 2018, and possibly that of Atlantic fin whales in the Gulf of St. Lawrence (GSL) (Schleimer et al. 2019a, Ramp et al. 2021.
The SARA mandates that the Canadian government prepare a recovery strategy defining threats, recovery potential, and management objectives that will sustain species recovery and survival for threatened and endangered species/populations. Critical habitat is a key element of recovery strategies. Important habitats for North Atlantic right whales in Canadian waters were first identified in 2009 (Brown et al. 2009) and legally protected as critical habitat under the SARA in 2017 (DFO 2017). While important habitats of Atlantic blue whales have been assessed (DFO 2018, Lesage et al. 2018, they have yet to be formally identified under the SARA.
One factor affecting the identification of important habitats for baleen whales off eastern Canada is the large range occupied by each of these highly mobile species. Logistical constraints caused by unfavorable weather and sea state conditions that are prevalent in non-summer months make it difficult to provide complete year-round coverage of these ranges via vesselor aircraft-based surveys. While long-term studies on the occurrence, life history, and habitat preferences of some baleen whales have been conducted in certain areas off eastern Canada (e.g. in the Bay of Fundy or GSL; Johnston et al. 2005, Ramp et al. 2006, 2014, Doniol-Valcroze et al. 2007, Ramp & Sears 2013, Gavrilchuk et al. 2014, Davies et al. 2019, Schleimer et al. 2019b, and some large-scale surveys have been conducted (Lawson & Gosselin 2011, relatively limited information exists about whale occurrence and habitat use throughout most of this region, especially in winter. This leads to uncertainty in identifying potential high-use areas or aggregation sites worthy of protection.
Most current knowledge on Atlantic blue whales in eastern Canada comes from the GSL, an area used only by a proportion of the population (Ramp & Sears 2013). Here they are acoustically present year-round in ice-free areas (Simard et al. 2016). Outside the GSL, important habitat areas have been identified based on habitat and prey modeling, visual sightings, and acoustic detections, which include areas along the Scotian Shelf edge and the southern edge of the Grand Banks off Newfoundland (Gomez et al. 2017, 2020, Lesage et al. 2018. A low number of sightings during large-scale, systematic, aerial surveys off eastern Canada precluded abundance estimates (Lawson & Gosselin 2011 but are consistent with the presumed small population size (< 250 mature individuals) (COSEWIC 2002). Limited recruitment (COSEWIC 2002) and threats such as entanglements (Ramp et al. 2021) and underwater noise (McKenna 2011, Melcón et al. 2012 highlight the urgent need to designate and protect critical habitat for this species, both inside and outside the GSL.
Atlantic fin whales are considered common off eastern Canada. They range broadly from the Bay of Fundy to Labrador, including throughout the Scotian Shelf, GSL, and Grand Banks. They were the most sighted and harvested species on the Scotian Shelf in the 1960s to 1970s, while the vast majority of fin whales caught off eastern Canada occurred off Labra dor and northeastern Newfoundland (Sergeant 1966, Mitchell 1974, Breiwick 1994, Lawson 2006. Visual sightings from 1945 to 2015 off eastern Canada, although uncorrected for effort, suggest a broad distribution of fin whales with potential highly suitable habitats identified along the Newfoundland coast, on the southern Labrador Shelf, on the outer Grand Banks, and in Orphan Basin (Lawson 2006, Gomez et al. 2020. Despite evidence of one localized decline in abundance (Schleimer et al. 2019a), there are currently no data to suggest that this trend applies to the Atlantic fin whale population as a whole, with preliminary summer abundance estimates ranging between 3000 and 4000 individuals (Lawson & Gosselin 2011. Although visual sighting records are strongly biased towards summer, some fin whales remain in eastern Canadian waters in winter (Lawson 2006, Roy et al. 2018. The summer presence of humpback whales off eastern Canada is well documented. In some areas, their distribution parallels that of fin whales, possibly due to trophic niche overlap (Whitehead & Carscadden 1985, Gavrilchuk et al. 2014. They are distributed across several feeding aggregations, which can be summarized as the Gulf of Maine−Scotian Shelf (including the Bay of Fundy), GSL, and Newfoundland− Labrador (Katona & Beard 1990). They occur broadly off Newfoundland and Labrador (Gomez et al. 2020), although predictable aggregations occur in coastal areas in relation with spawning aggregations of capelin Mallotus villosus (Whitehead et al. 1982, Johnson & Davoren 2021. Preliminary summer abundance estimates derived from large-scale systematic surveys indicate that about 6000 humpback whales occur off eastern Canada (Lawson & Gosselin 2011. Humpback whales generally migrate from northern summer feeding grounds to lower-latitude winter breeding grounds. However, passive acoustic moni-toring (PAM) studies have revealed that a portion of the population remains in high-latitude waters in winter, or migrates late in the season, such that the species is present in Canadian waters for most of the year , 2021a, Davis et al. 2020.
PAM has become a commonly used method for studying marine mammal occurrence over space and time (Zimmer 2011, Browning et al. 2017. PAM is a cost-effective means for year-round monitoring in areas otherwise hard to access and costly to survey using visual methods. Technological and dataprocessing advances have made the collection and timely treatment of large data sets possible (Kowarski & Moors-Murphy 2021). PAM has been widely applied to monitor baleen whales worldwide. Indeed, the widespread production of stereotypical songs by males (mostly seasonal) with generally high source levels and long propagation ranges (Širović et al. 2007, Stafford et al. 2007, Garcia et al. 2019) makes many baleen whale species ideal subjects for PAM. Non-song vocalizations produced by either sex can also be used to monitor species presence when songs are absent, rare, or are not the dominant vocalization type, although these types of vocalizations tend to be less consistent and regularly produced (see e.g. Simard et al. 2016).
Here, we aimed to provide a year-round pluriannual description of the minimum occurrence of blue, fin, and humpback whales in subarctic eastern Canadian waters using PAM data collected at 25 recording stations from 2015 to 2017. The analysis methodology applied in this study proved inadequate for a thorough description of North Atlantic right, sei, and minke whale occurrence. The acoustic occurrence of these species will be presented briefly here, with more detailed analyses either ongoing (North Atlantic right and sei whales) or left for future work (minke whales). The findings of this study are intended to serve as a baseline for the long-term assessment of changes in the occurrence of blue, fin, and humpback whales off eastern Canada, to guide future research and monitoring efforts, and to inform the identification of important habitats for management and conservation purposes.

Data collection
Acoustic recorders (n = 25) were deployed be tween 2015 and 2017 (Fig. 1). A 2 yr study funded by the Environmental Studies Research Fund (ESRF) allowed the deployment of 20 recorders (Stns 1−20) throughout the Scotian Shelf, Newfoundland Shelf, and the southern Labrador Shelf. The study focused on areas where anthropogenic activities were concentrated or expected in the near future and where marine mammal occurrence has been poorly de scribed. For that reason, areas where baleen whale occurrence is well documented (e.g. GSL, Bay of Fundy, inshore areas of Newfoundland) were not in cluded. Stns 21−25 were deployed as part of an ongoing PAM program conducted by Fisheries and Oceans Canada (DFO) to help address knowledge gaps for species at risk (including blue and North Atlantic right whales) and for marine protected area monitoring purposes around Nova Scotia. Spacing between stations ranged from ca. 40 to 260 km.
All acoustic data were acquired using autonomous multichannel acoustic recorders (AMARs; JASCO Applied Sciences) configured to sample at 8 kHz with a 24-bit resolution and a duty cycle of 11 min 18 s every 20 min. The AMARs were equipped with M36-V35-100, M8Q-51 (GeoSpectrum Technology, GTI), or HTI-99-HF hydrophones (High Tech, HTI). All hydrophones had a flat frequency response between 10 and 100 000 Hz. GTI hydrophones have a sensitivity of −165 dB re 1 V μPa -1 and noise floor of 165 dB re 1 μPa. HTI hydrophones have a sensitivity of −164 dB re 1 V μPa -1 and noise floor of 165 dB re 1 μPa. Recording systems were calibrated using a reference tone at 250 Hz produced by a pistonphone type 42AC precision sound source (GRAS Sound & Vibration) before and after each deployment. The location of Stn 19 was shifted during the second year of data collection, and recorders deployed at Stns 3 and 7 were lost during the first and second deployments, respectively. Data collection was uninterrupted between both years in the ESRF study (except Stns 9 and 10) but not in the DFO study and started earlier and ended later in the DFO study (Table 1).

Detection range modeling
An important parameter for interpreting data on marine mammal acoustic occurrence is the detection range, which in this study represents the maximum range from a recorder at which a signal of a given source level can be identified by an automated detector in given background sound conditions (see Section 2.4 for description of automated detectors). Modeling detection ranges also allows assessing the potential overlap in the detection area of adjacent stations. Detection range modeling was applied to the song notes of blue, fin, and humpback whales, which are the vocalizations on which this study is focused (see Section 2.3). Detection ranges were modeled for the fall (September to November), when these species were most acoustically abundant throughout the area.
To evaluate the detection ranges, the received level, RL(r) (in dB re 1μPa ), measured at the distance r from the source, was modeled as: where SL is the source level (in dB re 1μPa ), modeled as a Gaussian distribution with a mean and standard deviation. Vocalization SL, bandwidth, and calling depth were obtained from the literature (see Table S1 in the Supplement at www.int-res.com/articles/suppl/ n047p265_supp.pdf entire water column over the modeled areas, from 2 m to a maximum of 3000 m, with step sizes that increased with depth. Water depths throughout the modeled area were extracted from the SRTM15+ grid (Smith & Sandwell 1997, Becker et al. 2009). The propagation loss value was calculated using a sound speed profile characteristic of the selected season (fall, i.e. September to December). The sound speed profiles for the modeled site were derived from temperature and salinity profiles of the US Naval Oceanographic Office's Generalized Digital Environmental Model V 3.0 (GDEM; Teague et al. 1990, Carnes 2009). The GDEM temperature−salinity profiles were converted to sound speed profiles according to Coppens (1981).
Where possible, geoacoustic profiles were obtained via inversion modeling performed at 15 of the monitoring sites (Deveau et al. 2018). At the sites where locally derived profiles were unavailable, the profile of the nearest site or a default profile used in previous modeling exercises by JASCO for similar areas was used (see Table S2).
The automated detection of marine mammal vocalization is assumed to occur if RL is greater than the local noise level in the frequency band of vocalizations (NL) by a constant threshold c: The threshold c must be chosen such that there is very little chance of a false positive (FP) due to ambient noise, and such that the probability that the automated algorithms will detect a signal when present is at least 50%. Depending on the species of interest, JASCO's automated detectors use constants of 3−6 dB, which satisfies these constraints.
The distribution of NL was determined empirically from the measurements performed on the data recorded at each station. We used sound pressure level percentiles (see Delarue et al. 2018 for Stns 1−20; JASCO unpubl. data for Stns 21−25) for the deployment that was least influenced by flow, mooring, and seismic noise, which can be dominant at the low frequencies of interest. Thus, NL has discrete values and the final probability of detection, P D , as a function of range is: (3) where CDF SL (NL i + c + PL(r)) is the cumulative probability of the source level exceeding NL i + c + PL(r).
To further constrain the modeling so that the predicted ranges do not become unreasonably long, the maximum SL considered is the 90 th percentile of the SL distribution, and the minimum noise level is the 10 th percentile of the noise distribution. We then discretized the signal model and noise measurements into 0.5 dB bins (SL j and NL i ) and computed PL(r) for all combinations of NL i and SL j . We extracted the 10 th and 50 th percentiles of this distribution for generating plots and tables of maximum and median detection ranges, respectively.

Targeted marine mammal signals
Baleen whales generally produce stereotyped species-specific signals, at least seasonally. In blue, fin, and humpback whales, these signals form the basis of songs, which are defined as a series of notes arranged in a stereotyped pattern (Payne & McVay 1971) and are produced by males (Croll et al. 2002, Darling et al. 2006, Oleson et al. 2007. Although minke and sei whales produce patterned vocalizations, these have not been formally characterized as songs to date, although it has been suggested for sei whales (Tremblay et al. 2019). The song-forming vocalizations targeted in this study are the fin whale 20 Hz pulses (Watkins et al. 1987), blue whale A-B vocalizations (Mellinger & Clark 2003, Berchok et al. 2006, and humpback whale tonal vocalizations, which form the basis of song units in various forms (Au et al. 2006, Girola et al. 2019) but also occur as non-song vocalizations (Dunlop et al. 2007). The other stereotyped vocalizations identified during ana lysis were sei whale downsweeps (Baumgartner et al. 2008), North Atlantic right whale upcalls (Parks et al. 2011, Davis et al. 2017, and minke whale pulse trains (Risch et al. 2014b) (Fig. 2).
The songs of blue, fin, and humpback whales generally occur between August and April coincident with the breeding season in the northern hemisphere, including the study area (Watkins et al. 2000, Stafford et al. 2007. They have been recorded year-round in some areas, though less frequently than during the breeding season (Morano et al. 2012, Vu et al. 2012, Simard et al. 2016, Roy et al. 2018. Minke whales follow the same seasonal pattern of sound production, with pulse trains essentially absent from late spring to late summer (Risch et al. 2013(Risch et al. , 2014a. Our understanding of sei whale seasonal acoustic be havior is poor, but there is evidence for year-round downsweep production (Davis et al. 2020). Right whales also produce upcalls year-round (Davis et al. 2017).
Outside of the breeding season, balaenopterid whale vocalization rates and stereotypy generally decrease. In addition, non-song vocalizations of blue and fin whales, which are the main signals produced from late spring to mid-summer, and sei whale downsweeps show higher similarity in contours (Ou et al. 2015), limiting the efficacy of automated detectors and increasing the difficulty of reliable species identification. These 'summer' non-song vocalizations for blue whales (Berchok et al. 2006) and fin whales (Watkins 1981) were identified manually while validating automated detections (see Section 2.4).

Automated detection and manual validation
Baleen whale acoustic signals were automatically de tected using a JAVA-based contour-following algo rithm that matched contour parameters to a library of templates representing the vocalizations of the target species. The only exceptions were minke whale pulse trains, for which an automated detector was not implemented when the data were analyzed. For each automated detector, specific spectrogram parameters were applied (Table S3) before the algorithm searched the spectrogram for contours. Contours were assigned to a species if they were within the range of values in the template for the corresponding signal (Table S4). The automated detection algorithms applied to the data were described in detail by Kowarski et al. (2020).
A subset of acoustic files was selected from each station and deployment for manual validation by experienced analysts to evaluate automated detector performance, using the methodology described by Kowarski et al. (2021b). An automatic data selection for validation algorithm selected files for validation such that the file subset matched the corresponding (2) the number of automated detectors triggered per file; and (3) the temporal spread of files containing detections for each automated detector across the recording period. This multi-species approach allowed the performance of multiple automated detectors in diverse background noise conditions to be assessed simultaneously. The size of the validation subset chosen for this study was 0.5% of the sound files in each data set (11.5 min files, n = ~160 per station per year, or ~7680 files total representing 1472 h of data) due to analysis time constraints and the overall size of the data set. Although the automated detectors classify individual signals, we validated the presence/ absence of species at the file level, not the detection level.
The following restrictions were further applied to the automated detector results. If a species was automatically detected at a location but no vocalizations were found by analysts during the manual validation process (either during a specific period or the whole recording period), all automated detections during the relevant period were deemed FPs and excluded from further processing. For example, the automated humpback whale detector was falsely triggered by ice noise in northern stations through winter when the species was never manually confirmed to be acoustically present. Therefore, all automated humpback whale detections from these locations and times were deemed FPs and were excluded from the final occurrence results. Periods during which detections were excluded are re ferred to as exclusion periods (see Tables S6−S8  and see Figs . 4, 6, & 8).
The performance of each automated detector and any necessary thresholds (minimum number of automated detections per file to consider species present) was determined using a maximum likelihood estimation algorithm. This algorithm compared the automated and validated results and maximized the probability of detection while minimizing the number of false alarms using the Matthews correlation coefficient (MCC): (4) where TP (true positive) is the number of files with valid automated detections, FP is the number of files with false detections, and FN (false negative) is the number of files with missed detections. The perform-ance of the automated detectors was then described using precision (P) and recall (R). P represents the proportion of files with detections that are TPs. R represents the proportion of files containing the signal of interest that were identified by the detector.
Where the number of validated detections was too low, or the overlap between manual and automated detections was too limited for the calculation of P, R, and MCC, automated detections were ignored, and only validated results were used to describe the acoustic occurrence of a species.
The automated detector performance algorithm determines a threshold based on the number of detections per 11.5 min file for each species, station, and deployment combination that maximizes the MCC score. A species was considered absent if the automated detection count was below that threshold. Here, we only used results from automated detectors whose post-threshold P ≥ 0.75. When P < 0.75, the manual detections were instead used to de scribe the acoustic occurrence of a species. This was the case for North Atlantic right and sei whale vocalizations at all stations. Because the manual de tections of these, along with minke whale vocalizations, were not obtained in a systematic manner, their acoustic occurrence was summarized as overall presence/absence per station, where a single detection during the study was needed to indicate presence.
The minimum acoustic occurrence of blue, fin, and humpback whales is presented following the templates used by Davis et al. (2020). Where automated detections are shown, the exclusion periods and detection count threshold, if any, have been ap plied. Time series showing the number of days per week with automated and/or manual detections for each station over the 2 study years are pro vided. In addition, the proportion of days with automated detections during the entire recording period was plotted at each station by season. Spring is defined as 1 March to 31 May, summer as 1 June to 31 August, fall as 1 September to 31 November, and winter as 1 December to 28/29 February. The total number of detection (manual and/or automated) days was divided by the total number of recording days to account for recording effort. Sea ice extent (© 2021 EUMETSAT) for 15 March 2016 and 2017 (approximate seasonal maximum) was displayed on the spring plot, when the minimum acoustic occurrence was observed. A detailed analysis of sea ice as a driver of seasonal occurrence was beyond the scope of this paper, but this is intended to illustrate the loss of physical habitat to marine mammals occurring in winter off eastern Canada.

Detection range modeling
The detection ranges presented here should be considered with the following points in mind. The results are based on ambient background sound levels as recorded near the ocean bottom by the re corders during fall, when most automated detections occurred. Expected detection ranges could differ under different deployment scenarios (e.g. re corder in the sound channel or at the surface) or different seasons (due to different sound speed profiles or presence of sea ice, which would reduce detection ranges due to increased refraction caused by the irregular underice surface). Perhaps more importantly, although the modeling is based on a range of source levels (SL ± 3−5 dB) which represent a consensus across several relevant publications (see Section 2.2 and Table S1), the actual source level range is likely broader, which could result in substantial variations in detection ranges compared to those presented here. However, these results are valuable to understand the typical acoustic coverage of re corders for each species/ vocalization type in eastern Canada.
The median and maximum detection ranges for fin, blue, and humpback whale vocalizations are provided in Table S5 and Fig. 1. Table S5 specifically lists the minimum and maximum detection ranges for the 10 th and 50 th percentile of noise levels across all modeled radials, further highlighting the wide spectrum of possible ranges. Humpback whale song notes can be detected over much smaller areas than those of blue and fin whales. Overlapping detection ranges at adjacent stations suggest that fin and blue whale vocalizations could be heard at multiple recorders simultaneously. This was confirmed opportunistically for both species at Stns 1 and 8, on either side of the Cabot Strait (73 km apart). The longest detection ranges (using the highest source levels and lowest noise levels in Eq. 3) were limited to the maximum modeling extent (150 km) at 11 and 8 of 25 stations for fin and blue whales, respectively, and 10 of 25 stations for humpback whales (50−100 km extent), indicating that longer maximum ranges are possible.
Most stations along the continental slope showed large variation in detection ranges depending on source direction, reflecting the effect of bathymetry in relation to the source-receiver orientation (e.g. see blue whales at Stn 13, Fig. 1). Northern stations (Stn 19A and all stations north of Stn 19A) had generally larger detection areas for fin and blue whales than those further south. Stn 20 had noticeably longer detection ranges than any other station for all species. The shorter ranges at Stn 19B reflect the ef fects of high noise levels between 13 and 30 Hz due to flowinduced mooring noise (see Delarue et al. 2018).

Automated detector performance
The evaluation of automated detector performance was based on the manual review of 1472 h of data. The automated detector performance metrics are presented in Tables S6−S8 and summarized in Fig. 3. Automated detector performance varied between species, stations, and year. The fin whale detector performed best overall, while the humpback whale detector had the lowest performance. Fin, blue, and humpback whales had 3, 8, and 14 data sets, respectively, for which P was below 0.75 or none of the automated detections were manually validated. R was generally lower than P, the detectors identifying on average 40−70% of sound files containing vocalizations. However, daily R can be expected to be higher than per-file R (Kowarski et al. 2020). Indeed, a single file (out of 72 files per day) with detection is required for daily presence to be confirmed, and it is 273 Fig. 3. Precision (dark grey) and recall (light grey) scores for blue, fin, and humpback whale automated detectors across the stations where they were detected (sample size shown above each bar group) for both recording years. Boxes show the 25th, 50th and 75th percentiles, crosses show the mean, whiskers show the 5th and 95th percentiles, and dots show outliers therefore unlikely that relatively low per-file R would have a significant impact on the detections presented here. Nevertheless, the results should be considered estimates of the minimum presence of each species.

Blue whale minimum presence
Blue whale A-B vocalization detections (both automated and manual) occurred year-round in eastern Canadian waters (Figs. 4 & 5). Detections started as early as late July, increased in August, and continued throughout the fall, albeit more regularly in the southern half of the study area. They stopped as early as November off Labrador and ceased increasingly later from north to south. The acoustic occurrence of blue whales was at its lowest in late spring and early summer.
In fall and winter, blue whales were most consistently detected at stations associated with the Laurentian Channel, the southern edge of the Grand Banks, and the Scotian Shelf. They were almost never heard in the Strait of Belle Isle or on the southern Labrador shelf. Off Newfoundland, blue whale vocalizations were more common at stations along the slope of the Grand Banks than on the shelf, particularly at the northern end of the Flemish Pass (Stn 19B) despite much shorter detection ranges than at nearby sites (Table S5). In waters surrounding Nova Scotia, blue whale vocalizations in fall and winter were more common in the Laurentian Channel and on the Scotian Shelf than at deep stations along the slope of the Scotian Shelf.

Fin whale minimum presence
Fin whale 20 Hz pulses generally occurred following a well-defined seasonal pattern with an increase in the fall/winter and decrease in the spring/summer, with the contraction more prominent at northern stations (Figs. 6 & 7). Some of the southern stations (Stns 24 and 25) had detections year-round. The detection periods were notably shorter at northern stations, specifically on and off the Labrador shelf (Stns 11−13), on the northern Banks (Stns 14 and 20), and in the Strait of Belle Isle. At many stations, the end of the detection period in the spring represents the start of the exclusion period.
The sustained occurrence of fin whale vocalizations at Stn 19B well into the spring is worth noting due to the short signal detection ranges modeled there (see Fig. 1). The abrupt detection onset in Octo-ber at that station is an artefact of the exclusion period applied to exclude FPs caused by seismic airgun pulses. During spring, fin whale vocalizations occurred more frequently at the Flemish Pass and adjacent areas (Stns 17,19A,and 19B), as well as the western Scotian Shelf (Stn 25) than throughout the rest of the study area. Summer acoustic occurrence was generally low, except on the western Scotian Shelf and in the Gully Canyon (Stns 25 and 24).

Humpback whale minimum presence
Humpback whale vocalizations were detected (manually and/or automatically) at all stations except Stn 9 (Figs. 8 & 9). Detections were notably high in the Strait of Belle Isle (Stn 10), eastern Scotian Shelf (Stns 2 and 22), and southern Grand Banks (Stn 7). In the Strait of Belle Isle, humpback whales were detected nearly daily in both years from June to December. Stn 7 was only monitored during the first year but saw sustained detections throughout the monitoring period with a peak in summer and fall. A high number of weekly detection days at Stns 10 and 7 in summer indicate that positive detections are not reliant on the presence of songs, which were not observed in this season during manual validation.
In the Cabot Strait (Stns 1 and 8), detections were consistently higher on the south side. Similarly, along the southern edge of the Laurentian Channel, detections were consistently higher on the shelf (Stn 22) than along the slope of the channel (Stn 21). At deep stations along the continental slope, detections were generally lower than at stations on the shelf and close to shore. Notable exceptions were in the Gully Canyon (Stn 24) and north of the Flemish Pass (Stn 19B). Humpback whales were rarely detected at the northern stations (Stns 11−14), except during fall 2015 at Stn 14. Where detections were rare, they were generally concentrated during the winter months, sometimes with a secondary peak in spring (e.g. Stns 4 and 5). During winter, detections oc cur red at almost all stations with a predominance at southern stations. In contrast, spring and summer had the fewest stations with detections.

Other baleen whale species
North Atlantic right whale upcalls were manually detected sporadically at Stns 1, 2, 22, and 25 from June to December. Isolated detections occurred in September 2016 at Stn 10 and November 2016 at Fig. 4. Weekly presence summary, showing the number of days per calendar week with confirmed blue whale acoustic presence for each recording site (except Stn 9) between May 2015 and November 2017. Stations (numbers on the right) are arranged north to south. Grey blocks indicate weeks with no available data. Pink blocks indicate exclusion periods (when automated detections were ignored). Yellow areas represent deployments for which the output of the automated detector was not used when there was at least one manual detection. The first half of data for Stn 19 represents data collected at Stn 19A, while the second half represents data collected at Stn 19B  (Fig. 10). Several tentative detections were made at other stations, suggesting that a dedicated manual analysis of the data could reveal a broader acoustic presence of this species in eastern Canadian waters.
Sei whales were manually detected at all stations except those associated with the GSL (Stns 1, 8, 9, and 10) and Stn 3. The prime detection area was off the southern Labrador Shelf (Stn 13) and in the Orphan Basin (Stns 15,19A,and 19B) where detections occurred almost exclusively from May to November. The Scotian Slope (Stns 4, 5, and 24), and to a lesser extent the Scotian Shelf (Stn 25), also had relatively high detections with vocalizations recorded primarily from late winter to early summer and again in fall (Fig. 10). Detections occurred more frequently at the deep stations off the continental shelf than on the shelf. Although an automated detector was designed to find sei whale downsweeps, the presence of seismic airgun pulses off northeastern Newfoundland affected its performance. Off Nova Scotia, where blue whales are more common, the similarity between sei whale downsweeps and blue whale D-calls also impacted automated detector performance.

DISCUSSION AND CONCLUSIONS
The acoustic detections presented here represent the minimum occurrence of baleen whales off eastern Canada within the detection range of each re - corder. The detection areas shown in Fig. 1 are based on a subset of source level values that does not encompass the full possible range of source levels for each species. In addition, detection ranges are the result of the interaction of variable source levels and time-varying background noise and sound propagation conditions, making it difficult to know the area being actually monitored at any given time. A lack of detections (particularly when relying on the manual review results only) does not indicate the absence of non-vocalizing animals. Animals may be present but not vocalizing, may be producing vocalizations not targeted by the automated detectors, or their vocalizations may be masked by ambient or anthropogenic noise. Although the manual validation effort was distributed across the duration of each data set, the relatively low number of validated files may have resulted in analysts missing isolated vocalization events when exclusion periods were implemented. However, the results are unlikely to mischaracterize the overall species occurrence.
Call stereotypy and species specificity are key factors that affect the ability of automated detectors to accurately represent species presence. The vocalizations of fin, blue, minke, and sei whales fulfill these requirements. Humpback whale tonal vocalizations are less stereotyped. However, they generally occur in predictable frequency bands. Generic automated detectors developed for humpback whale vocalizations within these bands can have limited performance in noisy environments but perform well in areas where humpback whales vocalize actively ). Because of their potential similarity to humpback whale vocalizations, confidence in the output of the automated right whale upcall detector can be compromised in areas where humpback whales co-occur and are abundant.
Variable performance of the same automated detectors across stations or even between years at the same stations highlights the need to evaluate automated detector results on a case-by-case basis (Kowarski & Moors-Murphy 2021). Inter-annual variations in performance reflect varying background noise conditions and species presence. Factors influencing performance included a low number of TPs and the presence of confounding signals (e.g. vesselassociated tonal sounds in the case of the blue whale detector; airgun pulses in the case of fin and sei whale detectors; humpback whale vocalizations in the case of the North Atlantic right whale detector). For fin and blue whale vocalizations, lower calling rates, greater call variability, and overlap in call characteristics outside of the song production months (August to April) remain a challenge for automated detectors and, to some extent, also for experienced analysts.

Blue whales
The occurrence of blue whales in eastern Canada was reviewed by Lesage et al. (2018) and Moors-Murphy et al. (2019), and several areas of importance to blue whales for migration and foraging were proposed. These include several foraging areas in the GSL and a migration corridor through the Cabot Strait in and out of the GSL. Our consistent acoustic detections in the Cabot Strait, extending from late July to April, suggest a steady flow of individuals in and out of the GSL throughout most of the year but could also indicate that some individuals use this area for purposes beyond migration (e.g. mating and foraging). Indeed, satellite tagging studies have demonstrated individuals foraging in the Cabot Strait (Lesage et al. 2017). The entirety of the Lauren tian Channel appears to be important to blue whales with detections near its end (Stns 23 and 6) and beyond along the southern Grand Banks (Stn 16), extending well into February. Although most satellite-tagged blue whales headed towards the Scotian Shelf after passing through the Cabot Strait (Lesage et al. 2017), the intensity of acoustic detections in fall along the southern edge of the Grand Banks suggests that at least some individuals head east after exiting the Laurentian Channel. The only station in the GSL was Stn 10, located at the eastern edge of the Mecatina Trough area, which was identified by Lesage et al. (2018) as important blue whale habitat based on high predicted krill aggregations and high blue whale catches during whaling operations between 1925 and 1958, despite a scarcity of more recent sightings in the area. The lack of blue whale acoustic detections at Stn 10, which is consistent with the lack of detections during an earlier 1 yr acoustic study in the same area (Simard et al. 2016), suggests that this area is not regularly used by blue whales, although this may reflect short detection ranges and the recorder being located at the edge of the Mecatina Trough.
Two other areas of higher blue whale acoustic occurrence outside the GSL were the Scotian Shelf and the Flemish Pass. Several tagged blue whales transited near Stn 2 on the eastern Scotian Shelf, sometimes repeatedly, on their way out of the GSL (Lesage et al. 2017). Blue whale vocalizations were recorded at this station regularly between late July and the end of April, suggesting that this area may not be just a migratory corridor. In fall and winter, the recorder deployed in Emerald Basin on the Scotian Shelf (Stn 25) also consistently recorded a high number of detections. P and R values for these stations were high in both years, indicating that the acoustic occurrence of blue whales was accurately characterized. This area partially overlaps a cluster of blue whale sightings recorded during a short period of whaling off Nova Scotia (1967Scotia ( −1972, which extended towards the shelf break (Sutcliffe & Brodie 1977). Interestingly, the closest stations along the Scotian Slope (Stns 4, 5, and 24) did not record the same level of detections. These recorders were located inside the shelf edge important foraging habitat identified by Lesage et al. (2018), which consists of a narrow (~50 km) strip encompassing the shelf break of the Scotian Shelf, and the southern and southeastern Grand Banks up to the southern edge of the Flemish Pass. The detection range modeling results indicate that in the most likely scenarios, the blue whales producing vocalizations recorded at Stn 25 were located on the Scotian Shelf within 40-90 km of that recorder. Stations along the Sco tian Slope had restricted detection ranges towards the Scotian Shelf and a larger listening area seaward. There appears to be some potential for individuals present on the outer continental shelf to have gone undetected, even in the best-case scenarios. Nevertheless, our results suggest that the Scotian Shelf region may be equally, if not more, important to blue whales than the Scotian Slope. This is further supported by blue whale habitat suitability models, which indicate potentially highly suitable habitat on the central Scotian Shelf as well as the Scotian Slope .
The detections on either end of the Flemish Pass (Stns 19B and 17) are noteworthy, particularly in light of the relatively small detection ranges modeled for these areas and low automated detector R. Detections ex tended into January to the north and February to the south of the pass. Stns 19A/B and 15 had similar detection periods in both years, even though occurrence at Stn 15 is based on manual detections only. This suggests that Orphan Basin could have a similar importance as the Flemish Pass. Blue whales were comparatively rare at outer shelf locations on the Grand Banks and southern Labrador Shelf (Stns 18,14,20,11,and 12), despite rather extensive detection areas. This points to a preference for slope habitat in the Newfoundland region, in contrast to what is suggested by detections on and off the Scotian Shelf.

Fin whales
Uniformity in the acoustic occurrence of fin whales observed throughout the study area suggests a broad distribution across eastern Canadian waters from late summer to winter and makes it difficult to identify specific areas that may be of greater importance to this species, if any. It may be that their varied diet, made of zooplankton and small schooling fish (Gavril chuk et al. 2014), provides them more flexibility in habitat choice depending on prey availability and distribution, resulting in a more uniform apparent distribution. In contrast, blue whales may be more restricted in terms of potential foraging be cause of their euphausiid-restricted diet (Yochem & Leatherwood 1985). However, some areas are worth highlighting due to a greater presence of fin whale vocalizations in spring and early summer. Emerald Basin on the Scotian Shelf (Stn 25), the Gully Canyon (Stn 24), and stations associated with the Flemish Pass and southern Orphan Basin (Stns 15 and 19A/B) maintained relatively high call presence during this period. Late spring and early summer detections are noteworthy because the quasi-absence of songs and decrease in calling rates and in the proportion of 20 Hz calls in the vocal repertoire of fin whales (Watkins 1981, Stafford et al. 2007, Roy et al. 2018 generally makes the species more acoustically cryptic during this season. At most stations, the scarcity or absence of automated 20 Hz call detections and manual de tections of any type of calls in spring and summer indicate, at least, a limited acoustic presence of fin whales, but we cannot rule out that a greater manual review effort to compensate for reduced calling rates would reveal more consistent occurrence. In the Flemish Pass area, the exclusion periods in late spring and summer generally coincided with the presence of seismic airgun pulses. The lack of TP detections (no manual detections) may reflect reactions of fin whales to airgun pulses, as shown in other areas (Castellote et al. 2012).
The uninterrupted occurrence of 20 Hz calls in Emerald Basin and the Gully in summer compared to other areas could reflect a higher density of fin whales, better call detectability (e.g. closer proximity of vocalizing whales, lower background noise), or other factors causing higher acoustic activity. The Gully Canyon has significantly higher cetacean sighting rates compared to adjacent slope and shelf areas, including baleen whales (Moors-Murphy 2014), which could result in a greater concentration of fin whales and higher calling rates.
There is uncertainty regarding the winter migratory patterns of fin whales that feed in eastern Canadian waters in summer. Our results suggest that nonnegligible numbers of fin whales remain in the study area in winter. Although the total number of fin whales off eastern Canada is considered healthy, fin whales in some areas appear to be declining (Schleimer et al. 2019a). Atlantic fin whales in Canada are managed as a single population, even though evidence indicates some level of population structure (Mitchell 1974, Delarue et al. 2009). Potential spatial and demographic isolation should be considered when assessing areas that may be important for some but not all individuals, especially if they belong to populations with different abundance trajectories.

Humpback whales
The Strait of Belle Isle (Stn 10) was the only known humpback whale aggregation site (Stevick et al. 2006, Comtois et al. 2010) monitored during this study and was found to be consistently utilized from June, when the earliest humpback whale arrivals are typically recorded in the northern GSL (Ramp et al. 2015), until December, when ice starts forming in this area. Detections persisted nearly year-round on the southern Grand Banks (Stn 7), suggesting that some individuals remain in Canadian waters throughout winter, as previously observed in other North Atlan tic areas such as the Gulf of Maine (Davis et al. 2020) and Gully Canyon . None of the deep stations along the continental slope saw sustained periods of humpback whale acoustic occurrence.
Among the 3 baleen whales discussed in this paper, the humpback whale adheres most closely to the typical model of baleen whale migration patterns. Most stations saw detections constrained to well-defined periods which are presumably associated with the movement of migrating animals. A late fall-winter detection wave was apparent at all stations. A spring wave was generally only visible at stations in the southern half of the study area. These fall and spring waves of acoustic occurrence were investigated in more detail by Kowarski et al. (2019) and are consistent with seasonal patterns of detections observed in feeding grounds in the Gulf of Maine (Vu et al. 2012). Migratory movements of hump back whales from and to Canadian waters are poorly documented. Of 22 whales tagged on Caribbean breeding grounds, only 2 returned to Canadian waters. They migrated in deep offshore waters until reaching the Gully Canyon and the junction of the Grand Banks and the Laurentian Channel, near Stn 6. Both individuals started foraging upon reaching these areas (Kennedy et al. 2014). Whitehead et al. (1982) described the migration of humpback whales in Newfoundland. Whales first arrived on the south coast and progressively moved to the northeast coast near Bonavista Bay before crossing to the southern Labrador coast.
Our detections suggest that humpback whales exit the GSL predominantly along the southern edge of the strait. The discrepancy in the number of detections between the eastern Scotian Shelf (Stns 2 and 22) and stations farther south suggest that the whales disperse after leaving the former instead of following an established migration corridor. However, the intensity of detections downstream from the eastern Scotian Shelf region was highest in the Gully Canyon area, indicating that some whales may transit via this area before heading south in deep oceanic waters . It is unclear whether the waves of detections noted at stations along the slope of the Grand Banks represent animals summering in Canadian waters, for instance off Labrador, or whales migrating from feeding areas farther north, particularly off West Greenland. The differences in acoustic detections at Stns 19A and 19B are noteworthy, suggesting that the northern Flemish Pass (and possibly the Flemish Cap) may act as a migration transit point. Interestingly, Stn 17 at the southern end of the Flemish Pass had very few detections while they peaked at the northern end (Stn 19B).

Other species
The main summer feeding grounds of North Atlantic right whales in eastern Canada were historically located in the Bay of Fundy and in Roseway Basin (Davies et al. 2015(Davies et al. , 2019. Although individuals were also occasionally sighted in the GSL, since 2015 most right whale summer observations come from the southern GSL. Despite the limitations of our analysis protocol for establishing the acoustic occurrence of this species, our detections suggest, at the very least, that some individuals wander beyond the main aggregations site in the southern GSL, both inside and outside the GSL. Minke whale detections were concentrated in the southern half of the study area. However, minke whales are also abundant off Newfoundland (Lawson & Gosselin 2011), both inshore and offshore. A lack of pulse trains at high-latitude feeding grounds in the western North Atlantic was also noted by Risch et al. (2014a). Under the assumption that pulse trains are produced by males, as is the case for songs produced by blue, fin, and humpback whales (Croll et al. 2002, Darling et al. 2006, Oleson et al. 2007, the lack of pulse trains in the northern half of the study area could be explained by sexual segregation and a female-biased sex ratio, as observed in minke whales at other high-latitude feeding grounds such as the GSL (Gavrilchuk et al. 2014) and off West Greenland (Laidre et al. 2009). Although PAM for minke whales has been successful in some areas of the North Atlantic (Risch et al. 2014a,b), it may not be as broadly applicable as in other species, depending on the study area. Reassessing this data set using an automated pulse train detector could provide more robust answers regarding the occurrence of these signals in eastern Canadian waters.
The limited manual review of data identified potentially important areas for sei whales. Although detected at nearly all stations (except those in or near the GSL), deep-water stations in the Orphan Basin and off the southern Labrador Shelf concentrated the bulk of the manual detections. These areas coincide with the migratory destinations of sei whales tagged in the Azores in the spring (Prieto et al. 2014) and are consistent with their preference for deep oceanic waters. Given the recent listing of the species as endangered in Canada (COSEWIC 2019), conservation and management of this species would benefit from a dedicated analysis of this data set.

Final remarks
This study highlighted the broad spatio-temporal distribution of blue, fin, and humpback whales and identified areas that may deserve further attention, particularly the Scotian Shelf region and Flemish Pass−Orphan Basin areas. The latter is noteworthy due to ongoing oil and gas exploration and the potential for prolonged noise exposures and behavioral disruption. Further, some areas with active exploration programs (e.g. Flemish Pass) are in international waters. Although seafloor resource extraction is regulated by Canada under the United Nations Convention on the Law of the Sea, the question of the responsibility for the management of free-ranging species affected by the relevant activities has yet to be addressed.
As noted by Davis et al. (2020), baleen whale distribution can change over relatively short time frames. Climate change is likely to accelerate marine mam-mal distributional changes (Moore & Huntington 2008, Silber et al. 2017. In this study, the apparent range contraction observed in spring in blue, fin, and humpback whales coincided with the maximum sea ice extent in eastern Canadian waters (Figs. 5,7,& 9). Declining sea ice cover will increase habitat available in winter and spring, and changes in prey type or distribution may induce changes in foraging areas (Nøttestad et al. 2015). The patterns of occurrence presented here are the first to cover this extensive area for a continuous period and provide a baseline against which future changes can be assessed. This data set (of which data recorded at Stns 1−20 are publicly available) could also be used to investigate which environmental variables influence habitat use by these species.
Going beyond presence−absence, further evaluation of the relative significance of different areas is required. This is particularly relevant for blue whales because some individuals sighted in eastern Canadian waters do not frequent the GSL (Ramp & Sears 2013), which remains the only substantial aggregation site known in the northwest Atlantic. In fin whales, quantifying sound pressure levels within the band of their vocalizations (18−25 Hz) by week or month could allow an assessment of the relative importance of different areas over time (Nieukirk et al. 2012), especially if this metric can be corrected for the size of the listening area. A closer look at acoustic behavior could provide information about the use of habitats by these species. For instance, blue whale A-B calls emitted as song units are produced by lone traveling males, presumably as part of the species' reproductive behavior, while singular A-B calls are produced by males generally associated with other individuals in various behavioral states, including foraging (Oleson et al. 2007). Davis et al. (2020) investigated the acoustic occurrence of baleen whales between 2004 and 2014 be tween Florida and the Scotian Shelf. They noted a de cline in the occurrence of blue, fin, and humpback whales on the Scotian Shelf after 2010. Whitehead (2013) noted a decline in sighting rates of fin and humpback whales in the Gully Marine Protected Area and an increase in blue whale observations between 1993 and 2011. Although not directly comparable to Davis et al. (2020), our results indicate that the Scotian Shelf region remains an important habitat for all 3 species and that ongoing changes in distribution may be occurring. These observations strengthen the call for continued monitoring of the ranges of these species to track distributional shifts and allow for adaptive management.