Save

Individual calling behaviour and movements of North Atlantic minke whales (Balaenoptera acutorostrata)

In: Behaviour
View More View Less
  • 1 aIntegrated Statistics, 172 Shearwater Way, Falmouth, MA 02540, USA
  • | 2 bInstitute for Terrestrial and Aquatic Wildlife Research, University of Veterinary Medicine Hannover Foundation, Werftstrasse 6, 25761 Büsum, Germany
  • | 3 cNOAA, Northeast Fisheries Science Center, 166 Water Street, Woods Hole, MA 02543, USA
Open Access

Information on individual calling behaviour and source levels are important for understanding acoustically mediated social interactions of marine mammals, for which visual observations are difficult to obtain. Our study, conducted in the Stellwagen Bank National Marine Sanctuary (SBNMS), located in the Gulf of Maine, USA, used passive acoustic arrays to track North Atlantic minke whales and assess the sound production behaviour of individuals. A total of 18 minke whales were acoustically tracked in this study. Individual calling rates were variable, with a median intercall interval (ICI) of 60.3 s. Average source levels (SLrms) for minke whales pulse trains ranged between 164 and 168 dB re 1 μPa, resulting in a minimum detection range of 0.4–10.2 km for these calls in this urban, coastal environment. All tracked animals were actively swimming at a speed of 5.0 ± 1.2 km/h, which matches swimming speeds of migrating minke whales from other areas and confirms SBNMS as part of the migration route of this species in the Western North Atlantic. Tracked minke whales produced 7 discrete call types belonging to 3 main categories, yet no individual produced all call types. Instead, minke whales produced 2 multisyllabic call sequences (A and B) by combining 3–4 different call types in a non-random order. While 7 of the tracked individuals produced calling pattern A, 10 whales used calling pattern B, and only 1 animal combined call types differently. Animals producing different call sequences were in acoustic range of one another on several occasions, suggesting they may use these sequences for mediating social interactions. The fact that the same calling patterns were shared by several individuals suggests that these patterns may contain information related to sex, age or behavioural context.

Abstract

Information on individual calling behaviour and source levels are important for understanding acoustically mediated social interactions of marine mammals, for which visual observations are difficult to obtain. Our study, conducted in the Stellwagen Bank National Marine Sanctuary (SBNMS), located in the Gulf of Maine, USA, used passive acoustic arrays to track North Atlantic minke whales and assess the sound production behaviour of individuals. A total of 18 minke whales were acoustically tracked in this study. Individual calling rates were variable, with a median intercall interval (ICI) of 60.3 s. Average source levels (SLrms) for minke whales pulse trains ranged between 164 and 168 dB re 1 μPa, resulting in a minimum detection range of 0.4–10.2 km for these calls in this urban, coastal environment. All tracked animals were actively swimming at a speed of 5.0 ± 1.2 km/h, which matches swimming speeds of migrating minke whales from other areas and confirms SBNMS as part of the migration route of this species in the Western North Atlantic. Tracked minke whales produced 7 discrete call types belonging to 3 main categories, yet no individual produced all call types. Instead, minke whales produced 2 multisyllabic call sequences (A and B) by combining 3–4 different call types in a non-random order. While 7 of the tracked individuals produced calling pattern A, 10 whales used calling pattern B, and only 1 animal combined call types differently. Animals producing different call sequences were in acoustic range of one another on several occasions, suggesting they may use these sequences for mediating social interactions. The fact that the same calling patterns were shared by several individuals suggests that these patterns may contain information related to sex, age or behavioural context.

1. Introduction

Understanding the behavioural context of individual calling behaviour and the temporal patterns of call production is an important aspect of studying animal communication systems. Several taxa arrange different calls or syllables into larger units of sound. Such combinations allow for syntactical rule building, and increase information transfer over monosyllable communication. Songbirds, in particular, recombine simple calls to form a variety of higher-order songs that function primarily in a reproductive context (Nowicki & Searcy, 2004; Catchpole & Slater, 2008). In addition, multi-syllable phrases may convey information about group membership, as well as the size and threat of predators (Templeton et al., 2005; Briefer et al., 2013). In mammals, males of the Brazilian free-tailed bat produce songs that share several structural and functional traits with bird song and vary based on social context (Bohn et al., 2013). Several non-human primate species also produce higher-order call combinations, which may carry meaning related to predator presence and type or food source (Clarke et al., 2006; Ouattara et al., 2009; Clay & Zuberbühler, 2011).

In a marine context, many cetacean species exhibit highly advanced vocal systems, some of which have been studied extensively. In an environment, where light is attenuated quickly, behaviours such as the advertisement of breeding condition, coordination of group movements or the maintenance of social bonds are often mediated through sound (e.g., Tyack & Clark, 2000). However, due to the fact that most marine mammals spend only little time at the surface, and underwater observations are often infeasible, the visual quantification of behaviour and identification of individuals at sea is severely limited. Thus, there is a lack of knowledge on individual calling behaviour and the behavioural context of vocalizations.

Call sequences and their behavioural correlates have mainly been studied in odontocetes. For example, bottlenose dolphin signature whistles, which encode individual identity, are often produced in sequence both by individual animals, as well as groups (e.g., Quick & Janik, 2012; Janik & King, 2013). Similarly, short-finned pilot whales and killer whales produce non-random sequences of stereotyped call types, which most likely function in individual recognition and to maintain group cohesion (e.g., Ford et al., 1989; Sayigh et al., 2013).

Several species of baleen whales combine individual sound units to form songs that, similar to bird song, represent a series of notes arranged in a recognizable temporal pattern (Payne & McVay, 1971). These patterned sequences have been termed songs, based on the definition by Broughton (1963) that song is: ‘…a series of notes, generally of more than one type, uttered in succession and so related as to form a recognizable sequence or pattern in time’. In particular, humpback whales (Megaptera novaeangliae) produce complex, hierarchically structured songs (Payne & McVay, 1971; Cholewiak et al., 2013). Similarly, bowhead whales (Balaena mysticetus) sing elaborate songs (Stafford et al., 2008; Tervo et al., 2011), while blue (Balaenoptera musculus) and fin (Balaenoptera physalus) whales produce high intensity song units at very low frequencies (approx. 15–30 Hz) (McDonald et al., 2001; Croll et al., 2002). In humpback, fin, and blue whales it has been shown that only males produce songs (Glockner, 1983; Croll et al., 2002; Oleson et al., 2007a). In these species, songs function as male advertisement or to mediate interactions between competing males (Tyack, 1981; Tyack & Whitehead, 1983; Oleson et al., 2007a). Recent acoustic recording efforts of marine mammals are beginning to show that song occurs not only during the traditional breeding season but also on feeding grounds and during migration (Stafford et al., 2007; Simon et al., 2010; Vu et al., 2012). And alternative functions of songs, such as navigation and prey detection, have also been suggested (Clark & Ellison, 2004). In addition to song production, in most species of baleen whales, both sexes produce a range of different call types in various contexts. Several species produce feeding-associated vocalizations that may be repeated in monosyllabic sequences (Cerchio & Dahlheim, 2001; Oleson et al., 2007b; Širović et al., 2013). In addition, sequences of frequency modulated call-counter calls occur in fin, blue and right whales (Eubalaena spp.) and serve as contact calls to maintain group cohesion (Clark, 1982; Oleson et al., 2007b; Širović et al., 2013). A variety of variable social calls have been described for most species (Oleson et al., 2007a; Dunlop et al., 2008; Stafford et al., 2008; Parks et al., 2011; Stimpert et al., 2011); the function of these calls is largely unknown but many of the calls are stable over several years suggesting an important role in mediating social interactions (Rekdahl et al., 2013).

Sounds produced by North Atlantic minke whales have only recently been described in more detail. Mellinger et al. (2000) described low-frequency pulse trains with a varying interpulse interval (IPI) structure; and a recent study in the Gulf of Maine described 7 distinct pulse train types, which fall into 3 main categories and occur with varying frequency (Risch et al., 2013). However, the behavioural significance of these vocalizations and whether they are specific to sex, age, recording site or season is unknown.

Given the identification of several stereotypic call types in the minke whale vocal repertoire, the main objective of the current study was to investigate whether individual minke whales use the full vocal repertoire, whether they combine pulse trains in predictable vocal sequences and how individuals use these sounds when engaging in vocal exchanges with conspecifics. Despite the reliance on primarily passive acoustic data and the lack of visually observed context, answers to these questions will allow the development of testable theories with respect to the behavioural function of minke whale pulse trains.

In addition, passive acoustic localization can also elucidate other, non-vocal aspects of behaviour, such as swimming speeds and movement behaviour (e.g., Stanistreet et al., 2013). Thus, a secondary goal of this study was to use acoustic data to quantify fine-scale movements of minke whales in the Gulf of Maine study area, where little is known about the general behaviour of this species.

Lastly, basic data on individual vocal behaviour, source level and detection range, as obtained by acoustic localization, is necessary for passive acoustic monitoring (PAM) applications. North Atlantic minke whales are still exploited commercially. Thus, although the species is currently listed as ‘a species of least concern’ in the IUCN Red List (Reilly et al., 2008), accurate monitoring of population size and structure is essential for its conservation. Most current monitoring and abundance estimates for minke whales are based on visual sightings data (Skaug et al., 2004; de Boer, 2010; Bartha et al., 2011). However, new methods such as PAM, coupled with new analytical approaches (Marques et al., 2013), offer an opportunity to significantly improve abundance estimates for this cryptic species (Oswald et al., 2011; Martin et al., 2013). Since PAM depends on the detection of vocalizing animals, it is critical to understand how individual calling behaviours influence calling rates, as well as any sex, season or site specificity of different call types. Only a few studies have collected such vocalization data for baleen whales using either acoustic recording tags or passive acoustic array configurations (Matthews et al., 2001; Parks et al., 2011, 2012; Stanistreet et al., 2013). For North Atlantic right whales, considerable variability in individual calling rate patterns, related to behaviour, age, sex and season has been documented (Parks & Tyack, 2005; Parks et al., 2005, 2011; Van Parijs et al., 2009). These results reinforce the importance of describing and considering such variability when interpreting passive acoustic data.

Figure 1.
Figure 1.

Map of Massachusetts Bay with the Stellwagen Bank National Marine Sanctuary (SBNMS) outlined and shaded in gray. Inset map in upper right corner shows the position of the study area along the US East coast. Filled dots represent acoustic recording units deployed from 2 October–30 November 2009; triangles represent recorders deployed from 17 August to 11 October 2011.

Citation: Behaviour 151, 9 (2014) ; 10.1163/1568539X-00003187

2. Material and methods

2.1. Acoustic data collection

During 2 October–30 November 2009 and 17 August–11 October 2011, acoustic data were continuously recorded in the Stellwagen Bank National Marine Sanctuary (SBNMS) located in the Southern Gulf of Maine, Northwest Atlantic Ocean (Figure 1). Arrays of 10 and 11 (2009 and 2011, respectively) marine acoustic recording units (MARUs) (Calupca et al., 2000) were deployed at depths ranging from 25 to 114 m, and anchored 1–2 m above the sea floor. Units were spaced approx. 11 and 5 km apart in 2009 and 2011, respectively (Figure 1). Each MARU was equipped with a HTI-94-SSQ hydrophone (sensitivity −168 dB re 1 V/μPa), connected to a pre-amplifier and A/D converter, resulting in an effective system sensitivity of −151.7 dB re 1 V/μPa. All units sampled at 2000 Hz and 12 bit resolution, yielding an effective analysis bandwidth of 10–1000 Hz, with a flat frequency response (±1 dB) between 55 and 585 Hz. Recordings from individual units were time-aligned using calibration signals recorded at the beginning and end of the deployments and compiled into multi-channel data files.

2.2. Individual calling behaviour

All acoustic data were examined manually for the presence of minke whale pulse trains by generating multi-channel spectrograms using the sound analysis software XBAT (Figueroa & Robbins, 2008; FFT size 1024, 85% overlap, Hanning window). Pulse train types were assigned to one of seven categories within three main groups, based on interpulse interval (IPI) structure, as described in (Risch et al., 2013). These main groups were slow-down (sd), constant (c) and speed-up (sp) pulse trains (Figure 2). All pulse trains that were not stereotypic, or of too low quality for categorization, were placed in a variable (v) group. Calling rates for each animal were calculated as the total number of calls/min, and intercall interval (ICI) was calculated as the difference between the start times of two consecutive pulse trains produced by the same individual. All temporal measurements were carried out in XBAT based on manually delineated event boxes.

Figure 2.
Figure 2.

Spectrograms of North Atlantic minke whale pulse trains, as described in Risch et al. (2013). Identified calling patterns are based on transition frequencies and association patterns of individual pulse trains. (a–c) Calling pattern A, consisting of pulse train types: sd1, sd2 and c3. (d–g) Calling pattern B, consisting of pulse train types: sd3, c1, c2 and sp. Note the different time scales for spectrograms. Spectrogram parameters: FFT = 512, overlap = 75%, sample rate = 2000, resulting in a spectrogram resolution of 3.9 Hz and 64 ms.

Citation: Behaviour 151, 9 (2014) ; 10.1163/1568539X-00003187

To determine whether transitions between call types were random, transition frequencies were analysed as a first-order Markov chain, where a succeeding event is only dependent on the immediately preceding event. R package msm (Jackson, 2011) was used to arrange the continuous call data into a 2-way contingency table of preceding and following pulse train types and a transition probability matrix was estimated based on maximum likelihood. Observed and expected transition matrices were compared using a goodness-of-fit test and a 2-tailed Z-test for proportions (Fleiss, 1981) was used to compare each observed transition to its corresponding expected transition.

The transition frequencies for each call type combination were used to calculate an index of association (Ford, 1989). This index, based on Dice’s coefficient of association, normalizes the data to account for call type abundance:

2(Nij+Nji)(Si+Sj),

where i and j are consecutive pulse trains, N is the number of transitions from one pulse train to the next, and S is the total number of transitions of a particular type. Hierarchical clustering of this association matrix was performed by applying the UPGMA method, using function hclust in the R stats package (R Core Team, 2013). The cluster results were then plotted as a dendrogram.

2.3. Localization and movement

Series of pulse trains recorded on three or more channels and occurring for a period of at least 20 min, with gaps not exceeding 10 min, were chosen for localization. Individual pulse trains were localized using a correlation sum estimation algorithm (CSE), applied in XBAT (Cortopassi & Fristrup, 2005). This localization method differs from techniques based on time differences of arrival (TDOA) and hyperbolic fixing as it does not rely on the selection of waveform cross-correlation peaks to estimate locations. Instead, it calculates accumulated cross-correlation sums for all channel pairs across a grid of spatial points and chooses the point that maximizes the correlation sum as the most likely location. Due to this process, the method is considered to be more robust to background noise (Cortopassi & Fristrup, 2005). Each localized signal was verified visually in multi-channel spectrograms to ensure that the same pulse train was picked on all channels and the candidate location, as determined by the CSE algorithm, agreed with the observed TDOAs. Pulse trains for which reliable and repeatable location estimates could not be obtained using CSE, were removed from all further movement analyses. However, if the visually observed TDOA estimates of these calls agreed with the general pattern of the tracked animal, they were still included in temporal calling pattern analyses (see next paragraph). After manual review of each localization, animal tracks were defined as the time-ordered collection of locations from a single source connected by a straight line (Turchin, 1998). Tracks were smoothed with a 5-point moving average (MA) to reduce the influence of localization error. Statistical simulation tests using the CSE algorithm and comparable array geometry show, that localization accuracy depends on the position of the source relative to the array and increases with distance from the centre of the array (Urazghildiiev & Clark, 2013). Thus, movement characteristics were calculated only for tracks within 5 km of the array boundaries to reduce the impact of increasing error outside of the array. R package adehabitat (Calenge, 2006) was used to calculate track statistics, including track duration (h), net displacement (km), total distance (km) and average speed (km/h). A straightness index (SI), defined as the quotient of net displacement and total distance (1 = straight line path, 0 = meandering path), was calculated to assess directness of movement. All analyses using R were performed using version 2.15.3 (R Core Team, 2013). Location error for the arrays was determined by conducting calibration experiments on 22 October 2009 and on 9 October 2011 at two and five sites within the array, respectively. A series of 5 to 10 frequency modulated sweep tones were played at each site. The source location of each playback sweep was estimated using the CSE algorithm. Location error in meters was then quantified by subtracting the estimated position for each locatable sweep from the known speaker location. Differences between known and estimated source locations were averaged over all sweeps and transmission sites.

2.4. Source level estimation

To estimate pulse train source levels (SL), received levels (RL) were measured for a subset of the 2011 data, based on several detection criteria that included (a) a high signal-to-noise ratio (SNR > 10 dB); (b) could be reliably located; and (c) did not overlap with other sounds. The signals were bandpass filtered between 50 and 250 Hz. RL measurements were carried out in Raven Pro version 1.5 (Bioacoustics Research Program, 2013) for every fifth individual pulse and for the entire pulse train. Minimum and maximum frequencies of the measured signals were defined as the −10 dB end points relative to the signal peak in the power spectrum. Measurements included peak-to-peak (RLpp) and root-mean-square (RLrms) sound pressure levels (dB re 1 μPa) for every pulse, and RLrms for the whole pulse train. RLrms was measured over a time window encompassing 90% of the total signal energy in the selection window (Madsen & Wahlberg, 2007). Following these measurements, SL was calculated from RL by compensating for transmission loss (TL). Under the assumption of mainly spherical spreading, TL equals 20log(R), where R is the range of the whale from the receiver (Urick, 1983). In shallow water environments refraction and reflections from the sea bottom or surface will considerably affect TL, making the cylindrical spreading law (TL=10log(R)) more appropriate, while in many environments an intermediate term is most appropriate. Due to empirical measurements showing that 17log(R) is a reasonable approximation for TL in the study area (unpublished data), we chose this term for all SL estimations. SL results were averaged by pulse train and reported in terms of peak-to-peak (SLpp, dB re 1 μPa), root-mean-square (SLrms, dB re 1 μPa) and energy flux density (SLefd, dB re 1 μPa2s).

Figure 3.
Figure 3.

Boxplot of source levels (SLrms) showing results for individual pulses measured from pulse train types: c2 and sd3 for three individuals (animal ba1–ba3). Lower and upper bounds of boxes represent lower and upper quartiles, respectively. Solid lines represent medians and non-filled circles represent means. Whiskers represent farthest data points within 1.5× interquartile range (IQR) of the lower and higher quartile, respectively. Histogram shows frequency of occurrence of different source levels (binwidth = 2 dB) and a smoothed Gaussian kernel density plot for all measured pulses/pulse trains (N=993 (N=57)).

Citation: Behaviour 151, 9 (2014) ; 10.1163/1568539X-00003187

Assuming source and receiver depths of 20 m and using averaged minke whale SLs from our study, signal propagation in the SBNMS was then modelled using the Acoustic Integration Model (Hatch et al., 2012). Hourly ambient noise level (NL) values were calculated for the week of 4–10 October 2009. NLs were averaged over the frequency band containing the most pulse train energy (56.2–355 Hz) and summarized as 5th, 25th, 50th 75th and 95th percentiles. In the absence of hearing and detection threshold data for minke whales, these NL values were compared with the signal propagation curve, and the range at which SNR = 0 was determined.

3. Results

3.1. Individual calling behaviour

Since whales could only be successfully tracked to a certain range (< 8 km) outside the array boundaries, the data analysed were effectively censored; that is, start and end times of tracks did not necessarily mark the beginning or end of a calling bout. In addition, in order to reliably track individual animals and minimize the possibility of switching individuals, we only analysed tracks for which gaps in calling did not exceed 10 min (see Methods). Thus, all calling rate parameters are based on the time period during which an animal could be reliably tracked, given the acoustic range of the hydrophone arrays and limitations set by the analysis approach. Calling rates for individual whales ranged from a minimum of 8.7 to a high of 133.3 pulse trains/h. Figure 4 illustrates the distribution of observed intercall intervals (ICI). Tracked minke whales tended to call at a regular rate, with a median ICI of about 60.3 s (Figure 4, Table 1). The average call rate was 48.6 ± 27.5 calls/h, and the maximum silence between tracked calls extended to an average of 392.6 ± 292.1 s, with a maximum of about 6.5 min (Figure 4).

Figure 4.
Figure 4.

Histogram showing the frequency of occurrence of different Intercall Intervals (ICI) (binwidth = 20 s). The dotted line represents the median (N=1060).

Citation: Behaviour 151, 9 (2014) ; 10.1163/1568539X-00003187

Table 1.

Summary statistics of movements and calling rate parameters for individual minke whales recorded in the Stellwagen Bank National Marine Sanctuary (SBNMS) during autumn 2009 and 2011.

Table 1.

The 18 individual minke whales tracked in this study produced all major pulse train categories defined in a larger scale study from the same area (Risch et al., 2013): slow-down pulse trains (sd1–sd3), constant pulse trains (c1–c3), and speed-up pulse trains (sp) (Figure 2). However, pulse train type transitions were not random. The results of the Markov chain analysis showed that some types were highly likely to occur before or after other pulse train types (Goodness-of-fit test, χ2=733.25, df = 56, p<0.001, Figure 5). This pattern was not evenly distributed between pulse train type transitions. While 15 out of 64 transitions were positively correlated, 17 were negatively correlated (Figure 5). The calculated index of association between pulse train types showed strong positive associations between types sd1–c3 (0.39), sd3–c1 (0.29) and sd3–c2 (0.30) (Figures 5 and 6). Hierarchical cluster analysis of the association matrix grouped pulse train types sd1, sd2 and c3 as calling pattern A, while pulse train types sd3, c1, c2 and sp were grouped as calling pattern B (Figures 2 and 6). When Markov chain analyses were run by individual, the resulting probability matrices showed that pulse train type association patterns reflected differential call type usage by individual minke whales. While 39% (N=7) of the tracked animals preferably used pattern A, 56% (N=10) used pattern B, and only one animal combined call types in a different pattern (pattern C, Figure 7). In our sample, tracked minke whales used calling patterns A and B simultaneously during five occasions, when vocalizing individuals were at an average distance of 4.6 ± 2.5 km, and thus likely within acoustic range of one another.

Figure 5.
Figure 5.

Matrix of transition probabilities between different pulse trains. Preceding pulse trains are shown vertically and following pulse trains are plotted horizontally. (+/−) indicate transitions that are significantly greater or smaller than expected (p<0.05).

Citation: Behaviour 151, 9 (2014) ; 10.1163/1568539X-00003187

Figure 6.
Figure 6.

Dendrogram of association index, based on transition frequencies of pulse trains produced by individual North Atlantic minke whales, showing two distinct groups of associated pulse train types, resulting in calling patterns A and B.

Citation: Behaviour 151, 9 (2014) ; 10.1163/1568539X-00003187

Figure 7.