Exploring movement patterns and changing distributions of baleen whales in the western North Atlantic using a decade of passive acoustic data

Abstract Six baleen whale species are found in the temperate western North Atlantic Ocean, with limited information existing on the distribution and movement patterns for most. There is mounting evidence of distributional shifts in many species, including marine mammals, likely because of climate‐driven changes in ocean temperature and circulation. Previous acoustic studies examined the occurrence of minke (Balaenoptera acutorostrata) and North Atlantic right whales (NARW; Eubalaena glacialis). This study assesses the acoustic presence of humpback (Megaptera novaeangliae), sei (B. borealis), fin (B. physalus), and blue whales (B. musculus) over a decade, based on daily detections of their vocalizations. Data collected from 2004 to 2014 on 281 bottom‐mounted recorders, totaling 35,033 days, were processed using automated detection software and screened for each species' presence. A published study on NARW acoustics revealed significant changes in occurrence patterns between the periods of 2004–2010 and 2011–2014; therefore, these same time periods were examined here. All four species were present from the Southeast United States to Greenland; humpback whales were also present in the Caribbean. All species occurred throughout all regions in the winter, suggesting that baleen whales are widely distributed during these months. Each of the species showed significant changes in acoustic occurrence after 2010. Similar to NARWs, sei whales had higher acoustic occurrence in mid‐Atlantic regions after 2010. Fin, blue, and sei whales were more frequently detected in the northern latitudes of the study area after 2010. Despite this general northward shift, all four species were detected less on the Scotian Shelf area after 2010, matching documented shifts in prey availability in this region. A decade of acoustic observations have shown important distributional changes over the range of baleen whales, mirroring known climatic shifts and identifying new habitats that will require further protection from anthropogenic threats like fixed fishing gear, shipping, and noise pollution.


| INTRODUC TI ON
Seasonal migratory patterns are the foundation of long-distance movements and dramatic changes in animal distribution for many taxa in the animal kingdom (Dingle, 2014). Many cetaceans undergo long migrations with the purpose of moving from high-latitude feeding grounds in warmer months, to low-latitude breeding grounds in colder months (Kellogg, 1929). Baleen whales are among the longest traveled mammals, some covering up to 10,000 km annually (Stevick et al., 2011). Movements are thought to be driven by foraging or social behaviors (e.g., Clapham et al., 1993;Tyack & Whitehead, 1982;Visser, Hartman, Pierce, Valavanis, & Huisman, 2011); however, Corkeron and Connor (1999) also suggested that migration could be influenced by predator avoidance, and highlight that not all whale populations migrate annually (Geijer, Notarbartolo di Sciara, & Panigada, 2016). Non-migratory populations that remain in tropical and subtropical waters year-round (Mikhalev, 1997;Širović, Bassett, Johnson, Wiggins, & Hildebrand, 2014) may be supported by yearround productive foraging grounds (Geijer et al., 2016), as well as reduced energetic expenditure afforded by foregoing long migratory movements (Brown, Corkeron, Hale, Schultz, & Bryden, 1995;Kennedy et al., 2014). Even within migratory populations, some individuals remain on feeding grounds over winter (e.g., Brown et al., 1995;Thomisch et al., 2016;Van Opzeeland, Van Parijs, Kindermann, Burkhardt, & Boebel, 2013). Such intraspecies variation in individual movements are still not well understood, and may be further influenced by differences in gender, age, and reproductive state (Geijer et al., 2016). However, it is clear that baleen whale movement patterns are considerably more complex than previously thought.
Over the last few decades, climate change has led to dramatic increases in ocean temperatures, causing shifts in the distribution of prey species, with foraging animals following suit (Chen, Hill, Ohlemüller, Roy, & Thomas, 2011). The Gulf of Maine, an important feeding ground for many baleen whale species, is one of the fastest warming bodies of water in the world (Pershing et al., 2015), which may influence seasonal shifts in baleen whale presence (Ramp, Delarue, Palsbøll, Sears, & Hammond, 2015) in response to range shifts in prey and fish stocks throughout the western North Atlantic (Nye, Link, Hare, & Overholtz, 2009;Staudinger et al., 2019). North Atlantic right whales (NARWs; Eubalaena glacialis), an intensely studied species, are a striking example of these shifts in distributions over the last decade. From 2010 onward, NARWs spent less time in the Gulf of Maine and Bay of Fundy, and more time in mid-Atlantic waters along the US east coast and the Gulf of St. Lawrence (Davis et al., 2017;Davies et al., 2019). Record et al. (2019) showed that these observed changes in NARW seasonal movements reflect temperature-driven changes in the distribution of their primary food source, Calanus finmarchicus. Additional studies reveal bottom-up effects of temperature changes, such as shifts in kelp distribution (Merzouk & Johnson, 2011) and collapses of fisheries (Pershing et al., 2015), eventually leading to changes in communities within the entire marine ecosystems (Beaugrand et al., 2019). It is unclear whether other North Atlantic baleen whale species have undergone similar shifts in their movement patterns to NARWs in response to ocean warming and food source redistribution. While the seasonal distribution of humpback whales (Megaptera novaeangliae) is relatively well-known, the movements and distributions of other large baleen whale species (sei, Balaenoptera borealis; fin, B. physalus; and blue whales, B. musculus) throughout the North Atlantic Ocean remain poorly described.
Within the North Atlantic, the humpback whale range extends from breeding grounds in the Caribbean and Cape Verde Islands to feeding grounds off the eastern United States and Canadian seaboard, Iceland, Greenland, and Norway (Hayes, Josephson, Maze-Foley, & Rosel, 2019;Kennedy et al., 2014). During the spring, summer, and fall, humpback whales in the western North Atlantic are found feeding in the Gulf of Maine, Gulf of St. Lawrence, and in waters off Nova Scotia, Newfoundland, and western Greenland (Katona & Beard, 1990). In winter months, a portion of the North Atlantic humpback whale population visits breeding grounds in the Caribbean and the Cape Verde Islands, and some individuals have even been identified in both breeding grounds (Heenehan et al., 2019;Stevick et al., 2016;Stevick, ØIen, & Mattila, 1998;Wenzel et al., 2009). Passive acoustic data from the western North Atlantic have revealed that humpback whales are present year-round in the Gulf of Maine (Murray, Rice, & Clark, 2014;Vu et al., 2012), and in winter months off the Scotian Shelf (Kowarski, Evers, Moors-Murphy, Martin, & Denes, 2018). Tagging studies provided insight on migration between these known coastal feeding and breeding grounds (Kennedy et al., 2014); however, long-term humpback whale movements among these areas are not well known.
Sei whales are one of the least studied baleen whales, with most information on their distribution derived from historic whaling records, stranding records, and visual surveys (COSEWIC, 2003;Hayes et al., 2019;Mead, 1977). In the western North Atlantic, their range extends from mid-to low-latitudes to as far north as Labrador (Kapel, 1985;Kellogg, 1929;Olsen et al., 2009;Prieto, Silva, Waring, & Gonçalves, 2014) and the Davis Strait (Mitchell, 1974). The southern limit of their range remains unknown; however, stranding reports document sei whales as far south as Florida (Miller, 1924) and Mexico (Miller, 1928). Migratory movements of sei whales in the western North Atlantic are not yet well understood, but they are believed to move northward in June and July from southern New England to eastern Canada (Mitchell, 1975), and move southward in September and October (CETAP, 1982). During the spring and summer, sei whales are sighted in northern portions of the US Atlantic Exclusive Economic Zone (EEZ), including Georges Bank, the Gulf of Maine, and south of New England (Halpin et al., 2009). Often found in the deeper waters off the continental shelf edge, including the Scotian Shelf edge during the spring feeding season (Hain, Hyman, Kenney, & Winn, 1985), sei whales are also seen in shallower waters of the continental shelf in the Great South Channel and Massachusetts Bay (Halpin et al., 2009;Payne et al., 1990). Recently, satellite tag studies revealed westward movements of tagged individuals from the Azores to the Labrador Sea in the summer (Olsen et al., 2009;Prieto et al., 2014). Few studies have documented North Atlantic sei whale vocalizations, until recent work recorded and described sei whale vocalizations off New England and the Azores Romagosa, Boisseau, Cucknell, Moscrop, & McLanaghan, 2015;Tremblay, Van Parijs, & Cholewiak, 2019).
Fin whales are frequently observed in the western North Atlantic, from Cape Hatteras, North Carolina to Greenland (Edwards, Hall, Moore, Sheredy, & Redfern, 2015). A global review of fin whale sightings and acoustic data showed year-round presence throughout most of the US EEZ, commonly occurring in the Gulf of Maine and in Canadian waters off Nova Scotia (Edwards et al., 2015;Hain, Ratnaswamy, Kenney, & Winn, 1992). Acoustic records revealed the year-round presence of fin whales in Massachusetts Bay and the New York Bight (Morano et al., 2012;Muirhead et al., 2018), as well as occurrence from September through June in offshore waters surrounding Bermuda and the Mid-Atlantic Ridge (Clark & Gagnon, 2004;Nieukirk et al., 2012;Nieukirk, Stafford, Mellinger, Dziak, & Fox, 2004;Watkins, Tyack, Moore, & Bird, 1987). While New England waters provide important feeding grounds, mating and calving grounds remain unknown. Hain et al. (1992) suggest US mid-Atlantic latitudes for calving grounds based on neonatal stranding analyses, but this has not been confirmed by at-sea surveys.
While fin whales do undergo seasonal movements (Silva et al., 2011), their broad-scale distribution year-round suggests the possibility that they do not undergo the same large-scale migrations in the North Atlantic as other baleen whales, similar to fin whales in the North Pacific (Oleson, Sirovic, Bayless, & Hildebrand, 2014).
In the western North Atlantic, blue whales are mainly sighted off eastern Canada, with occasional sightings in the Gulf of Maine (Wenzel, Mattila, & Clapham, 1988) and other waters within the US EEZ (CETAP, 1982). The northern part of their range includes waters off Nova Scotia, Newfoundland, and Labrador, and extends as far north as the Davis Strait (Jonsgård, 1955;Moors-Murphy et al., 2019).
From spring through summer, blue whales occur predominantly in the Gulf of St. Lawrence, where the population is well-studied (Sears et al., 1990). In winter months, blue whales are found from southern Newfoundland to the Davis Strait (Mansfield, 1985), while acoustic detections also indicate their presence as far south as the New York Bight and near the Mid-Atlantic Ridge (Muirhead et al., 2018;Nieukirk et al., 2004). They are seen and heard year-round outside the Gulf of St. Lawrence in waters off Nova Scotia (Moors-Murphy et al., 2019).
While their southern range limit is unknown, acoustic detections of blue whales have occurred in deep water north of the West Indies and east of the US EEZ (Clark, 1995;Nieukirk et al., 2004). There have been a few historical strandings in the Caribbean (Harmer, 1923) and the Gulf of Mexico (Baughman, 1946), supporting suggestions that their range extends at least that far south (Yochem & Leatherwood, 1985).
Their tendency to use deeper, rather than coastal waters makes their seasonal movements difficult to study. However, satellite tag studies show movements of blue whales from the Gulf of St. Lawrence to North Carolina, including both on-and off-shelf waters, extending to deeper waters around the New England Seamounts (Lesage, Gavrilchuk, Andrews, & Sears, 2017).
Passive acoustic monitoring (PAM) provides robust data to explore multiple species' simultaneous occurrence across seasons.
Decadal studies using PAM have monitored seasonal distributions of fin whales ; tracked migratory movements of humpback whales (Abileah, Martin, Lewis, & Gisiner, 1996) and blue whales (Stafford, Nieukirk, & Fox, 1999); and provided new information on movements for minke whales (Balaenoptera acutorostrata; Risch et al., 2014) and NARWs (Davis et al., 2017). Within the North Atlantic, well-known songs or call types are unequivocally attributed to each of the species discussed in this paper and are widely used to assess their presence. Here we use patterned song notes and other sounds produced by humpback whales (Payne & McVay, 1971;Stimpert, Au, Parks, Hurst, & Wiley, 2011), downsweeps produced by sei whales , 20 Hz pulses produced by fin whales (Watkins et al., 1987), and song notes produced by blue whales (Mellinger & Clark, 2003) to examine large-scale species distribution. Most of these signals are sex-specific (humpback whale song: Winn & Winn, 1978; fin whale 20 Hz pulses: Croll et al., 2002), and often seasonal (blue whale song: Moore, Stafford, Mellinger, & Hildebrand, 2006;Stafford, Mellinger, Moore, & Fox, 2007). Although we will miss species' presence when they use other call types or are silent, we can still capture large-scale distribution patterns throughout the periods that they use these known vocalizations.
Previously, we conducted a broad-scale PAM study across the western North Atlantic to analyze NARW seasonal distribution (Davis et al., 2017). Based on identified changes in occurrence patterns starting in 2010, we found NARW acoustic detections significantly decreased in the Gulf of Maine region, and increased in mid-Atlantic regions of the US eastern seaboard. Here, we use similar acoustic datasets and protocols to understand the seasonal distribution of humpback, sei, fin, and blue whales within the western North Atlantic Ocean, and to determine whether any of these species exhibited similar shifts in distribution patterns across time.

| Data collection
All available passive acoustic recordings from over 100 research projects throughout the western North Atlantic Ocean were combined to create a decade-long dataset. A total of 35,033 recording days of data were collected from 2004 to 2014 from 281 passive acoustic recorders deployed between Saba in the Caribbean and the Davis Strait off western Greenland (Figure 1). Most recording sites were located on the continental shelf or along the shelf edge with only six sites in offshelf (off eastern Greenland [region 2] and a New England Seamount [Bear Seamount, region 6]) waters; therefore, this analysis was largely restricted to the continental shelf and shelf break region. The dataset was broken up into 11 geographic regions, based on acoustic data availability and biologically relevant areas ( Figure 1; Davis et al., 2017). The Gulf of St. Lawrence was designated as a separate subregion (region 3A) to reflect its biological importance (Meyer-Gutbrod, Greene, & Davies, 2018); however, only 2 months of recordings were made available for our study in this region, so the Gulf of St. Lawrence (region 3A) was combined with the Scotian Shelf (region 3) and incorporated in the results as one region (region 3) for all analyses.
Recordings were collected using five different types of bottom-mounted passive acoustic recorders (Table 1) Fox, Matsumoto, & Lau, 2001), and the Guardbuoy (Akoostix Inc/ Geospectrum Technologies; http://geosp ectrum.ca). Data were collected from 281 recorders, ranging from a minimum of 25 days to a maximum of 2 years (Table 1). Of these recorders, 56 used a duty cycled recording schedule, recording 12%-95% of the time, and 225 recorded continuously. The majority of recordings (206 out of 281) were sampled at 2 kHz, with some ranging up to 250 kHz. All recordings were low-pass filtered and decimated to 2 kHz to ensure comparability and analytical consistency across datasets. Recordings were further resampled to 120 Hz for adequate analyses of lower-frequency signals, in this case, vocalizations of fin and blue whales.
Acoustic detection ranges can vary significantly depending on the recording equipment, location, whale or recorder depth, bathymetry and environmental conditions, as well as by signal type and behavioral context (Cholewiak et al., 2018;Širović, Hildebrand, & Wiggins, 2007;Stafford et al., 2007). Previous acoustic studies examined detection ranges over which the species-specific vocalizations used in this study can be heard in varying oceanographic conditions Cholewiak et al., 2018;Kowarski et al., 2018;Širović et al., 2007;Stafford et al., 2007), in some cases F I G U R E 1 Locations of available passive acoustic recorders used for this study extending from the Caribbean (bottom right map inset) to the northernmost locations in the Davis Strait (top left map inset). Yellow points indicate the locations of recorders available from 2004 to 2010; black points indicate the locations of recorders available from 2011 to 2014; and blue points indicate locations of recorders available for any amount of time across both time periods. Red boundaries outline the designated regions, which were defined following the methods in Davis et al. (2017). Region numbers correspond to the following geographic areas:

Data gaps
Gulf of Maine Recording unit from this deployment used for missed detection rate analysis.

TA B L E 1 (Continued)
for the same datasets used in this study. We used results from these studies here as guidelines for the distance over which each species may be detected within this study's geographic range (see Table 2).
Taking this information into account and to be conservative, only a single recorder was selected for analysis when recorders were congregated in groups or arrays with units spaced at 20 km or less; this approach minimized duplicate detections across receivers as best as possible. Acoustic analyses were focused on data collected between and Roseway Basin, Canada, as these were the only long-term recordings available for these areas.

| Detection and classification of calls
All acoustic data were processed using the Low Frequency are multivariate normal), 75% of pitch-tracks for the call type will have an MD of 3.0 or less (Baumgartner et al., 2013). Setting an MD threshold is necessary to minimize the false detection rates, but in doing so causes some true detections to be missed in the analysis.
The MD threshold of 3.0 was chosen for all call types detected and classified in the humpback, sei, and fin whale call library. However, for blue whales, false detection rates were lower than any of the other species, thus an MD of 5.0 was chosen to decrease the probability of missing true detections.
All vocalizations were classified based on a user-developed call library (expanded from Davis et al., 2017; Table S1); our library for the 2 kHz sampled data included two of our target specieshumpback and sei whales. Given the low frequency characteristics of fin and blue whale vocalizations, an additional call library was created for these two species that matched the decimated 120 Hz sampled data.
All LFDCS detections were manually reviewed by a number of trained acoustic analysts to determine daily presence of each of the four baleen whale species. A true detection was defined as a pitch track that correctly classified a call or song unit to the species that produced it (Bonnell et al., 2016). Given the variability of each species' vocalizations, the specific methodology to determine daily acoustic presence was different for each species. That process is described in more detail below.

| Baleen whale call types used for detection and classification
Humpback whale males produce complex song that changes annually (Payne & McVay, 1971;Payne & Payne, 1985;Winn & Winn, 1978), and has been recorded throughout their entire range and across  (Figure 2a; Clark & Clapham, 2004;Kowarski et al., 2018;Mattila, Guinee, & Mayo, 1987;Vu et al., 2012). Non-song vocalizations, or social sounds, vary with some calls being similar to those found in song while others are completely different. These non-song vocalizations are produced by both sexes and across ages (Dunlop, Cato, & Noad, 2008;Fournet, Jacobsen, Gabriele, Mellinger, & Klinck, 2018;Stimpert, 2010;Stimpert et al., 2011). Given that humpback whale song can be highly variable between years, the call library described in Baumgartner and Mussoline (2011)  Sei whales in the North Atlantic produce low-frequency downsweeps (Figure 2b), from 82 to 34 Hz, as single, doublet, triplet, or more series of vocalizations . These downsweeps can also be found associated with other newly reported call types thought to represent song (Tremblay et al., 2019). It is currently unknown whether these vocalizations are sex-biased, or how they may vary regionally and seasonally. The LFDCS call library described in Baumgartner and Mussoline (2011) contains the 82-34 Hz sei whale downsweep, which was the call type we focused on in this study. All sei whale downsweep detections with an MD of 3.0 or less were manually screened for the daily presence of a doublet or triplet (following the same methods as described by ; doublets and triplets were defined as two or three repeated downsweeps, respectively, with roughly 3.5 s elapsed between the start of successive calls). Sei whales were considered present if a true detection (at least one downsweep detected within a doublet or triplet) was found for that day. As single sei whale downsweeps can resemble some vocalizations produced by fin and blue whales (Berchok, Bradley, & Gabrielson, 2006;Širović, Hildebrand, & Thiele, 2006), only the occurrence of the downsweeps as doublets or triplets were selected to ensure confidence in species identification.
Fin whales produce 20 Hz pulses, occurring in 7-19 s intervals, with bouts lasting up to 32.5 hr (Figure 2c; Julien Delarue, personal communication; Morano et al., 2012;Watkins et al., 1987). These vocalizations are thought to be produced solely by males as a breeding display (Croll et al., 2002). They have been documented throughout the year in their western North Atlantic range, and thus are excellent indicators of male fin whale presence. A call library for fin whale 20 Hz pulses was built for the data sampled at 120 Hz. To validate this call library, a full year of data from nine sites (18 recorders total) were selected across the dataset range (marked with "a" in Table 1). These data were examined every third hour of each day on the first, 11th, and 21st day of the month to look for fin whale presence. These hours were manually verified for true detections with an MD of 3.0 or less. Using the methods described in Baumgartner and Mussoline (2011) Table 1 for a summary of these decisions).
The most common vocalizations documented from blue whales in the North Atlantic are their low frequency song, which is made up of repeated phrases, comprised of song notes, with 1-2 min intervals (Mellinger & Clark, 2003), thought to be produced by males (Oleson et al., 2007). A call library for blue whales was built for the data sampled at 120 Hz, and created for A, B, and AB phrases (as described by Mellinger & Clark, 2003; Figure 2d, Table S1). All detections with an MD of 5.0 or less were manually screened. Daily presence for blue whales was confirmed if there were three song phrases visible, including at least one true detection. The low frequency band in which blue whale song occurs is often overlapped with boat or background noise and in noisy situations it can be difficult to identify song units with confidence. Only accepting detections when three or more phrases occurred ensured our confidence in the presence of the blue whale song.

| Validation of LFDCS performance
The manual verification of each detection ensured a 0% false detection rate in daily presence. To evaluate the missed detection rate of the LFDCS for each of the four species, three regions (Southern New England, Cape Hatteras, and Southeast United States; regions 7, 9, and 10; see Figure 1) were chosen for manual analysis of the recorded audio. Owing to the large size of the dataset, all regions and recorders could not be included. These regions were selected to incorporate variability across the datasets' geographic, water depth, and temporal range, using one recorder type (MARU) for a comparable assessment.
When available, a full year of data from one recording site was taken from the two time periods compared in this analysis (before and after 2010) for regions 7 and 10, and data from the only available time period (after 2010) in region 9 were taken (marked with "b" in Table 1).
Every fifth day was manually screened by a trained acoustic analyst for the daily presence of each call type described above for each of the species. Long-term spectral averages (LTSAs) were viewed using the MATLAB-(Mathworks) based custom sound analysis software program Triton . When further inspection was needed, the sound analysis software Raven Pro 1.5 (Bioacoustics Research Program, 2014) was used to examine the spectrogram in more detail; thus allowing a more accurate assessment of the presence or absence of certain vocalization types. When the vocalizations of a given species were observed, that day was marked as positive for presence of that species. The number of days of each species' presence found by the manual screening of acoustic data was compared to the days marked as present using confirmed detections from LFDCS. Missed detection rates were calculated using the confusion matrix method as described in Baumgartner et al. (2019).

| Review and analysis of call detections
Daily presence of all call types for each of the four species was summa- with detected species-specific vocalizations as the dependent variable, and defined time periods (2004-2010; 2011-2014) and regions as independent variables, with their interaction effects included in the model. A GLM with a Poisson distribution with log-link was run given that the detection data were counts, accounting for zero-inflated, discrete data. Within each year and region, the number of recording days was multiplied by the duty-cycle to correct for non-continuous data.
As recording effort (the number of days during which recorders were present) varied across time and region, we included the log of the number of days during which recorders were present plus 1 (because for some time*region cells, there were no recorders present) as an offset in the model. This procedure resulted in the following model structure: Lastly, results from these analyses were compared to the NARW's daily presence data from Davis et al. (2017) to compare the seasonal presence of five baleen whale species.

| RE SULTS
A total of 840,792 hr of recordings were processed across all available data. Acoustic detection results are presented as weekly presence for each of the corresponding vocalizations for all four species (Figure 3a-d). Each species' acoustic presence was then summarized into seasons, following the seasonality defined in Roberts et al. (2016) as: Winter (November-February); Spring (March-April); Summer (May-July); and Fall (August-October; Figures 4-7). Lastly, data from Davis et al. (2017) on right whale seasonal presence was plotted together with the four species in this study to allow direct comparisons to be made between the presence of all five species ( Figure S1).

| Humpback whales
Humpback whale songs and calls were detected on at least 1 day in all recording regions (Figures 3a and 4). They were detected yearround in the Gulf of Maine (regions 4 and 5), southern Scotian Shelf (region 3), and off eastern Greenland (region 2). They were detected sporadically, but throughout the year, in mid-Atlantic waters off Virginia (region 8), with the majority of humpback whale presence occurring between January and May. Humpback whales were present for a minimum of 5 days in the Davis Strait (region 1) during July and November to January. They were likely present for longer here; however, distinguishing humpback whale song from bowhead whale song in this region remains challenging.
During winter and spring months, they were detected throughout the entire sampled range, from their known Caribbean breeding grounds (region 11) through eastern Greenland (region 2; Figure 4a,b).
They were detected consistently in these seasons in the northern Caribbean (January-May; region 11), but were present only for a few days in inshore waters in the Southeast United States (region 10), with only one additional day of presence in the summer, suggesting that they rarely come onto the continental shelf in this area. Winter and spring had high detection rates of humpback whales in southern New England waters (within the New York Bight to Nantucket Shoals, region 7) and in the mid-Atlantic off Virginia (region 8).
Humpback whales were detected off Cape Hatteras (region 9) primarily between October and January, during their southward migration, with only a few detection days in spring and fall. In most summer and fall months, humpback whale detections decreased noticeably in southern New England waters (region 7), as well as eastern Greenland (region 2), reflecting concentrated humpback whale presence on feeding grounds from the Gulf of Maine to southern Scotian Shelf (regions 3-5; Figure 4c,d). The offshore recorder on the New England Seamounts (near Georges Bank; region 6) had only a few days of song and call occurrence in winter and summer (with no recording effort available in the spring). In this region (6), vocalizations were found more often on recorders along the shelf break around Georges Bank from March through July, suggesting humpback whales likely remain on the shelf, or close to it, in the northern regions.

F I G U R E 3
Weekly presence summary: Boxplots representing the average number of days per calendar week per recording site with confirmed acoustic presence for (a) humpback whales; (b) sei whales; (c) fin whales, and (d) blue whales, across all recorders in each region described in Figure 1 and for all years of the study (2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014). Horizontal lines within the boxes indicate the median, box boundaries indicate the 25th (lower boundary) and 75th (upper boundary) percentiles, vertical lines indicate the largest (upper whisker) and smallest (lower whisker) values no further than 1.5 times the interquartile range, and black dots represent outliers. Grey blocks indicate weeks where no data were available for that region

| Sei whales
Sei whales were detected from south of Cape Hatteras to the Davis Strait (regions 1-10) and exhibited a distinct seasonal pattern in acoustic presence across the different geographic regions (Figures 3b and 5). In the Southeast United States, sei whales were detected only on recorders deployed on the western edge of Blake's Plateau (region 10), with no occurrence found on recorders closer to shore on the shelf. Sei whale calls were absent from recording areas in the Caribbean (region 11).
Sei whale calls occurred nearly year-round in waters south of New England (region 7), with higher detection rates occurring from March through July. In the winter, sei whale acoustic detections occurred along the entire coastline, from Florida (Southeast United States; region 10) to eastern Greenland (region 2), but were sparse on recorders closest to shore, and only detected off eastern Greenland (region 2) in the beginning of November (Figure 5a).

| Fin whales
Fin whale calls were present across the entire dataset from just south of Cape Hatteras to the Davis Strait (regions 1-10; Figures 3c   and 6). Fin whale calls were present on a few Southeast US recorders (region 10); however, all detection days here were on recorders located off the continental shelf, suggesting that fin whales occurred further offshore in the south. There were no fin whale detections on any of the Caribbean (region 11) recorders, or inshore Southeast US (region 10) recorders.
Throughout the entire year, fin whales were detected near-continuously from Virginia (mid-Atlantic; region 8) through eastern Greenland (region 2). Of these regions (2-8), the highest number of days with detections occurred from August through April, with a noticeable decrease in days with detections from May through July.
Fin whales were detected on Georges Bank (region 6) from March to December, with sporadic presence from May to August, and F I G U R E 8 Adjusted means of acoustic occurrence for each time period (2004-2010 in red, 2011-2014 in blue), for each region indicated on the x-axis, for each species. Vertical bars represent 95% confidence intervals. The y-axis represents the expected number of days with acoustic presence, given the average number of recording days for that region and time period. The y-axis is on a logarithmic scale (base 10) and is different for each species. Data for North Atlantic right whales are taken from Davis et al. (2017) no detections in January or February. Acoustic activity decreased slightly in spring months, however, fin whale detections remained present within the range (Figure 6b). From March through April, fin whales were primarily detected from the Scotian Shelf through the mid-Atlantic (regions 3-8), with some detections in the Davis Strait (region 1), eastern Greenland (region 2), and Cape Hatteras (region 9). In summer months, acoustic activity decreased noticeably, where they were absent in the Davis Strait (region 1) from April through June and in waters south of the mid-Atlantic (regions 9 and 10) from April (region 10) or May (region 9) through August (Figure 6c).

| Blue whales
Blue whales had the lowest number of days with detections throughout the dataset (Figure 3d). Overall, they were detected from North Carolina to the Davis Strait (region 1) in the summer (Figure 7c). There were a handful of days where blue whales were detected off Cape Hatteras (region 9) and the northern edge of Blake's Plateau (Southeast United States; region 10) in the summer, but these occurrences were infrequent.

TA B L E 3
Results of the Poisson generalized linear model (GLM) testing whether the annual occurrence of each species across regions differed over two time periods (A: 2004-2010; B: 2011-2014). The number of days per year in which whale calls were detected is the dependent variable, and the time periods and regions are independent variables, with their interaction effects included in the model. Eastern Greenland, Georges Bank,Cape Hatteras,and the Caribbean (regions 2,6,9,and 11) are excluded from the model due to insufficient data in some time*region cells, and the Caribbean (region 11) is included for humpback whales only. For all other regions, both factors and their interactions were significant. Tables show results from the Poisson GLM testing between the two time periods (A-B) for each region separately, using the False Discovery Rate to correct for alpha-value inflation for (a) humpback whales, (b) sei whales, (c) fin whales, and (d) blue whales. Pairwise comparisons of time periods across individual regions were run using the R Package phia

| Comparison of acoustic detections before and after 2010
The annual acoustic presence before and after 2010 was evaluated for all four species, with NARWs included for comparative purposes ( Figure 8). In addition, changes in weekly presence before and after 2010 are illustrated in Figure S2a

| Detector evaluation of missed detection rates for all call types
The LFDCS, with an MD threshold of 3.0, missed an estimated 5% of days for humpback whales, 14% of days for sei whales, and 10% of days for fin whales. With an MD of 5.0, the LFDCS missed an estimated 10% of days for blue whales (Table 4). The number of days analyzed to evaluate the missed detection rate for each species varied, ranging from 247 to 250 days for blue and sei whales, respectively, and from 678 to 1,215 days for humpback and fin whales, respectively. The ability to manually screen for the repetitive calls of humpback and fin whales was greatly facilitated by using LTSAs compared to the less frequent vocalizations of sei and blue whales; therefore, more days were efficiently incorporated in the analysis for humpback and fin whales to provide a more robust validation. It is likely that the missed detection rate for sei whales would decrease if single downsweeps were allowed to indicate sei whales' daily presence. Additionally, the strict protocols used to define the daily presence for each species further reduced acoustic presence rates, but were necessary to increase our confidence in true presence. For all four species, these rates support the LFDCS as a good detector for determining acoustic presence. This is especially evident when the scale of this study is taken into consideration, as the missed detection rate was composed over data from different regions, depths, noise environments, and throughout full years when available. Therefore, the call detections in this study represent the minimum number of vocalizations present across the region but are likely to be a good representation of true seasonal patterns in each recording region.

| D ISCUSS I ON
All four focal baleen whale species were present throughout, from the Southeast United States (region 10) to the Davis Strait and eastern Greenland (regions 1 and 2); humpback whales ranged further south into the Caribbean (region 11). During winter, all four species were acoustically present from the Southeast United States (region 10) up to the Davis Strait and eastern Greenland (regions 1 and 2), suggesting that they occur widely throughout the western North Atlantic Ocean during this season. In interpreting our observations, it is important to recognize the limitations within the dataset, including regional gaps in acoustic coverage, varying detection ranges across species and habitats, as well as acoustic behavior limiting portions of the populations being detected. Recording locations provided widespread, but varying, temporal and spatial coverage, with some regions that had (a) extensive temporal and spatial coverage (e.g.,

Massachusetts
Bay and southern New England; regions 5 and7); (b) partial temporal and spatial coverage (e.g., Gulf of St. Lawrence, Scotian Shelf, Gulf of Maine, and Georges Bank; regions 3A, 3, 4, TA B L E 4 Summary from missed detection rate analysis, showing number of days with true positives (whales were found present both by detector validation and manual screening), false negatives (whales were found present in manual screening but not from detector validation), true negatives (whales were not found present by either detector validation or manual screening), and resulting missed detection rates for each species and 6); or (c) little to no coverage (shelf-edge and off-shelf waters).
Furthermore, despite differences in vocal behavior across species, where some vocalizations change seasonally (e.g., fin whale 20 Hz pulses, song) or are thought to be produced by males only (e.g., song), or where acoustic behavior remains unknown (sei whales), we broadly detected vocalizations used by each species across the entire dataset. These data give a comprehensive overview of the minimum spatial and temporal distribution of each species, adding broad-scale, long-term information to our current understanding of these species, filling in significant gaps, and highlighting potential changes in acoustic occurrence over time.

| Humpback whales
We detected humpback whale song and social sounds throughout all regions in the dataset during winter. Our observations confirm both that humpback whales vocalize throughout their entire range (Mattila et al., 1987;Vu et al., 2012), and that not all humpback whales migrate to southern breeding grounds in the winter (Brown et al., 1995), with at least some individuals remaining on northern feeding grounds such as the Gulf of Maine (regions 4 and 5) and the Scotian Shelf (region 3) throughout this period (Kowarski et al., 2018). This winter distribution is not surprising, as various studies have observed humpback whales in northern latitudes throughout the year (Clapham et al., 1993;Murray et al., 2014). However, the length of time over which they were present across all areas during winter months in this study was extensive.
In addition to the expected detections in the Caribbean (Heenehan et al., 2019;Whitehead & Moore, 1982), humpback whales were present from Cape Hatteras (region 9) to eastern Greenland (region 2) throughout the winter. Detections also remained high across these regions through spring and summer. Additionally, detections showed that the regions south of New England (region 7) and east of Greenland (region 2) were also important areas for humpback whales, similar to NARWs (Davis et al., 2017;Mellinger et al., 2011;Muirhead et al., 2018).
The noticeable decrease in acoustic activity across all available recorders in the fall (Figure 4), as well as recorders on the shelf in the Southeast United States (region 10) in winter through summer months, supports previous studies that suggested migration to and from Caribbean breeding grounds occurs further offshore, beyond the detection range of the recorders used in our study (Clapham & Mattila, 1990;Kennedy et al., 2014;Reeves, Smith, Josephson, Clapham, & Woolmer, 2004). The few detections (33 days) in the Southeast United States (region 10) in late winter and spring suggest that some individuals may travel through or linger in coastal waters, but this is likely an exception rather than the norm. However, decreases in humpback whale detections could also be attributed to changes in vocal behavior, where more sporadic calling could lead to missed or insufficient calls within our defined presence analysis window.

| Sei whales
Sei whales exhibited distinct seasonal movements, with peak occurrence in northern latitudes (regions 1 and 2) during late summer and fall months.
Like the other species, sei whales were detected along almost the entire coast in winter months, from Florida (Southeast United States.; region 10) to eastern Greenland (region 2). In the Southeast United States (region 10), sei whales were not detected on recorders closer to shore than the western edge of Blake's Plateau, indicating a more offshore distribution in this southern area. Sei whales moved into more northern regions, the Davis Strait and eastern Greenland (regions 1 and 2), in summer months, while still occurring south to the New York Bight (region 7). Very little information existed on sei whale distribution prior to this study, with most knowledge coming from whaling records off northern Labrador and the eastern North Atlantic (Jonsgård, 1966;Mead, 1977;Prieto, Janiger, Silva, Waring, & Gonçalves, 2012). This could be due to their use of offshore, pelagic habitats (Hain et al., 1985), or the fact that sei whales can be difficult to distinguish from Bryde's or fin whales in visual surveys. Their summer occurrence near Greenland (regions 1 and 2) matches the movements of satellite tagged sei whales traveling towards the Labrador Sea in May and June (Olsen et al., 2009;Prieto et al., 2014). Acoustic occurrence of sei whales corresponded with the timing reported in previous acoustic studies, with sei whales present in the Great South Channel (Georges Bank; region 6) throughout May (Baumgartner & Fratantoni, 2008;Baumgartner, Lysiak, Schuman, Urban-Rich, & Wenzel, 2011;, and in Massachusetts Bay (region 5) from September to November (Tremblay et al., 2019). Their occurrence along the shelf edge, particularly in Canadian and Northeast US waters (Scotian Shelf and Georges Bank; regions 3 and 6) corresponds with previous reports, however, detections occurring on the shelf in the Gulf of Maine and southern New England (regions 4, 5, and 7) highlight greater use of on-shelf areas here than previously described (COSEWIC, 2003).
This study provides the first comprehensive analysis of sei whale distribution throughout the western North Atlantic Ocean, highlighting movements and important habitat for this species. Their movement northward in summer months suggests that their summer feeding grounds range from the Gulf Maine through the Scotian Shelf New England and the New York Bight (region 7) are highlighted as an important area for sei whales, as this is the one region where they were detected persistently year-round. This area is an important region for baleen whale species in general, and in particular for NARWs who target the same prey as sei whales, specifically C. finmarchicus (Baumgartner & Fratantoni, 2008;.

| Fin whales
Fin whales were present nearly year-round from Virginia (mid-Atlantic; region 8) to eastern Greenland (region 2). These findings correspond with regional studies where fin whales were detected on 99%-100% of recording days in Massachusetts Bay (region 5) and the New York Bight (region 7; Morano et al., 2012;Muirhead et al., 2018). Moreover, these data reflect previous findings of  (Edwards et al., 2015). Overall, these data confirm much of the evidence that fin whales occupy a large portion of the shelf for most of the year (Hain et al., 1992).

| Blue whales
For a typically oceanic and rare species, blue whales were detected in the dataset on the continental shelf far more than expected.
Although blue whale's song travels large distances (see Table 2), it is unlikely that all detections in our data were from individuals far offshore, as sound attenuates rapidly for the recorders in shallow shelf areas, and the presence of blue whales has been corroborated with visual sightings in many of the areas where they were acoustically detected (National Marine Fisheries Service, 2018; Wenzel et al., 1988 (Berchok et al., 2006;Marotte, 2014;Mellinger & Clark, 2003). Like other baleen whales, blue whale song is thought to be produced by males, as a reproductive display (Oleson et al., 2007

| Baleen whale occurrence before and after 2010
All baleen whale species showed significant changes in their acoustic occurrence between the two time periods considered in this study: before and after 2010. In particular, sei whales showed an increased presence in southern New England and mid-Atlantic regions (regions 7 and 8), similar to that reported for NARWs (Davis et al., 2017). As both species are copepod feeders, sei and NARWs can often be found feeding together in some habitats . Shifts in prey distribution in this part of the North Atlantic are already being reported, and are projected to increase with warming sea temperatures Except on the Scotian Shelf and in the Southeast United States (regions 3 and 10), sei whale call occurrence increased after 2010 in most areas (Figure 8; Figure S2b). This contrasts with the dramatic changes in the presence of NARWs observed through visual and acoustic surveys, particularly in the Gulf of Maine (region 4). While some of the reduction in the presence of NARWs is due to the species' decline in recent years (Pace, Corkeron, & Kraus, 2017), their large-scale distributional changes cannot be explained by the decline alone . These differences between sei and NARWs could be due to the differences in feeding strategies.
NARWs are ram feeders, targeting extremely dense patches of C.
finmarchicus with open mouths (Kenney, Hyman, Owen, Scott, & Winn, 1986;Kenney, Mayo, & Winn, 2001), and sei whales can gulp feed, targeting additional prey such as euphausiids or fish by skimming as they swim (Baumgartner & Fratantoni, 2008;Flinn, Trites, Gregr, & Perry, 2002;Laidre et al., 2010). Additionally, warming and melting of arctic sea ice have shown increases in shelf-associated copepods in the North Atlantic, including early copepodid stages of C. finmarchicus, but decreases in abundance of later stages of this species, primarily targeted by NARWs (Greene, Pershing, Cronin, & Ceci, 2008;Grieve, Hare, & Saba, 2017). Thus, as NARWs' distribution shifts to follow the distribution of their primary prey, sei whales may remain in these areas and alter their focal prey. Alternate feeding strategies witnessed in other species, such as humpback whales' ability to prey switch (Fleming, Clark, Calambokidis, & Barlow, 2016), offers plausible explanation as to why results varied between species and across regions.
We observed that fin, blue, and sei whales increased the time Lastly, on a more regional scale, a significant shift in habitat use after 2010 can be seen in the decreased acoustic occurrence of humpback, fin, blue, and NARWs on the Scotian Shelf (region 3).
What might be driving this shift remains unclear; it is possible the shift reflects changes in prey availability similar to that observed in the Gulf of Maine (Sorochan et al., 2019), but there is little data to elucidate this process. Nevertheless, the data suggest that the Scotian Shelf (region 3) has become a less preferred habitat for most baleen whales since 2010.

| CON CLUS IONS
This is the first study to show spatial and temporal occurrence of humpback, fin, blue, and sei whales across the western North Atlantic Ocean over long time spans and large spatial scales and to demonstrate how these species' distributions have changed over time. These species are all protected under the US Marine Mammal Protection Act, with fin, blue, and sei whales also listed as endangered under the US Endangered Species Act. In Canada, blue and sei whales are listed as endangered under the Canadian Species at Risk Act and by the Committee on the Status of Endangered Wildlife in Canada, respectively. Anthropogenic activity, including ship strike, entanglement, and ocean noise, are the leading threats to these species (e.g., Avila, Kaschner, & Dormann, 2018;Thomas, Reeves, & Brownell, 2016). Knowing when and how each of these species frequent areas that overlap with anthropogenic activity is crucial for their conservation, which is even more challenging given their widespread winter distributions. With increasing industrial use of the western North Atlantic seaboard (Gilman et al., 2016;Government of Canada, 2017), and increased concerns around climate change (Pecl et al., 2017), there is a need for cost-effective monitoring of whale distributions and any changes therein. Many years of traditional visual surveys from vessels and aircraft have been conducted in US and Canadian waters (Lawson & Gosselin, 2009;Palka et al., 2017).
Although these surveys can derive estimates of abundance, these estimates are not precise enough to detect the changes in distribution identified in this paper. PAM is effective for monitoring large areas over years, especially in seasons when visual surveys are limited, and is particularly valuable for detecting temporal trends and changes.
Current technology also includes PAM in real-time (Baumgartner et al., 2013(Baumgartner et al., , 2019, which can improve our management response times and inform mitigation efforts. This study highlights the wealth of information available from retroactively analyzing datasets from a wide range of study designs and goals. Continuing these types of cross-institutional collaborations and designing surveys with clear goals in mind can allow for a better understanding of species occurrence, and can be used to recognize large-scale changes as they transpire.

ACK N OWLED G EM ENTS
We