The data for this study were collected off west Maui, Hawai’i in the Hawaiian Islands Humpback Whale National Marine Sanctuary (Supplementary Appendix S1) during three winter breeding seasons (Jan-Mar) from 2019 to 2021. Acoustic recordings were made using suction cup Acousonde^TM tags temporarily deployed on 11 adult humpback whales in three different group types: competitive groups, singly escorted mother-calf pairs, and dyads.

The two Acousonde^TM tags used for this study (B010 and B046) were equipped with hydrophones recording 16-bit audio with the sampling rate set to 12,226 Hz, an 8th order elliptic anti-alias filter at 4646 Hz, a 4-stage cascaded high pass filter at 22 Hz, and a total path gain of +2.4 dB. The acoustic sensitivity of the B010 and B046 tags were -187.2 dB re 1 V/µPa and -187.9 dB re 1 V/µPa, respectively. Each Acousonde^TM tag was equipped with suction cups and deployed on a whale’s dorsal surface just forward of the dorsal fin from a 9 m long hand-held carbon fiber pole by a tagger situated at the bow of an 11.5 m outboard vessel (see Stimpert et al., 2012 for best practices). Prior to tagging, the observers identified the group composition, recorded its behavioral state, and obtained identification images opportunistically of the ventral surface of each whale’s tail flukes (Katona et al., 1979). After a successful tagging event, the boat continued to follow the group at a distance of 200–300 m and the observers continued to opportunistically document whale behavior, group composition, and individual identities. These contextual notes continued while the research vessel remained with the animals, although this typically did not account for the entire tag deployment, which sometimes extended into evening and nighttime hours. To account for periods when context could not be recorded, recordings were only analyzed throughout the period of observation (starting from 10 minutes after tag deployment) and for the 2 hours following the end of observation. This 2-h grace period was chosen under the assumption that the group composition did not change immediately at the end of observation, but we acknowledge that over time the likelihood for group composition to change increases.

Spectrograms were generated using Raven Pro 1.6 (The Cornell Laboratory of Ornithology, 2019) with a 4096-point DFT and 80% overlap. All recordings were visually and aurally inspected in their entirety, and all selections were made by one observer (JC). To account for any disturbances in vocal behavior that may have been caused by the tagging event (Williamson et al., 2016), calls from the initial 10 min of recording were eliminated from analyses. All calls were selected individually in Raven Pro, with boundaries set as tight as possible to the produced signal to ensure the highest accuracy for call measurements. For calls to be included in the selection process, it was necessary that there be a distinguishably clear start and endpoint, and little to no overlap with calls from conspecifics. Overlap did not occur often, and in total only 11 calls with overlap were included in analysis (all with overlap of <110 milliseconds). If any calls were produced in a repetitive pattern that resembled that of whale song from the current season (as per observations of background song recorded on the tags), they were not included in analyses. To be included in quantitative analysis, calls needed to have a signal-to-noise ratio (SNR) of at least 10 dB or higher. A SNR of greater than 10 dB is an accepted threshold established in previous studies (Dunlop et al., 2007; Dunlop et al., 2008; Stimpert et al., 2011; Rekdahl et al., 2013; Fournet et al., 2015) at which calls can be attributed to either the signaler or accompanying whales with relative confidence. However, without concurrent scan sampling to quantify the positions of nearby groups, it is difficult to determine with certainty if calls recorded with acoustic tags are produced by the tagged individual and group, or if they have been produced by other nearby conspecifics. It is important to note that, for this study, no call classification was performed. This was intentional as the naming schema of call types is still currently debated.

Due to the nature of acoustic tag recording, most files contained some flow noise in the 0–200 Hz range. With high rates of tourism in the study area, vessel noise was also occasionally disruptive in the recordings. To account for these disturbances, calls from all groups were measured using custom-written noise subtraction algorithms in MATLAB (Indeck et al., 2020). For every call selection, a corresponding noise selection (1–3 s long) was made in the time either immediately preceding or following the call selection to capture the ambient noise at the time of signaling. Occasionally, during a period of repetitive calls, one noise file was used for more than one signal file to ensure that noise selections corresponded as closely as possible to the respective signal. The spectrum from noise selections was subtracted from the corresponding signal file to remove most of the energy from flow noise, vessel traffic, and song from nearby conspecifics.

Once noise energy was removed from the call files, acoustic characteristics were measured from the observer-selected bounds of the call files. The temporal boundaries of call selections were made as close as possible to the visible start and endpoint of the call, and frequency boundaries were selected as close to the lowest and highest frequencies as possible. The measured acoustic characteristics of the calls included duration and several frequency measurements (i.e., peak frequency, center frequency, fifth percentile frequency, etc.) as used previously (Indeck et al., 2020; Supplementary Appendix S2).

To investigate the temporal differences in calling behavior, we looked at the time periods between calls. For this analysis, all calls were used, including those eliminated from other analyses due to poor SNR. This was done so as not to skew the period between calls, although it is inherently possible that some calls were not detected, due to ambient and flow noise. Every call selection made in Raven measured the beginning and end time of the call based on the time boundaries of the selection. These time measurements were then used to calculate the time between each social call by subtracting the end time of a call from the beginning time of the following call. This yielded a measurement of the inter-call interval (ICI). Call rates (calls/whale/hour) were also compared across the different group compositions. These were calculated for each tag deployment by dividing the total number of calls by the average number of whales in the group, by the number of recording hours.

Select frequency measurements (peak, center, first quartile, third quartile, fifth percentile, and ninety-fifth percentile frequencies) from high SNR (>10 dB) calls were compared across all group compositions. To analyze these measurements, generalized linear mixed models (GLMMs) were run using Tag ID as a random effect and a gamma distribution with a log link (suitable for non-normal data), using the glmmTMB package in R (Brooks et al., 2017). The emmeans package (Lenth, 2022) was then used post hoc to provide least squares means, which are adjusted to predict the effect of the factor variable on the response assuming equal sample sizes and, therefore, are more robust for unbalanced data than are observed averages (Harvey, 1960). Pairwise comparisons of call parameters across group composition were calculated using the “multivariate t" adjustment method, as it takes into consideration the correlation structure of the model (Lenth, 2022). Significance was set to p < 0.05. These statistical tests were also used to compare the temporal differences (duration and ICI) in calling across all groups. All means are presented as the least squares mean +/- standard deviation.

Table 2 shows the model-adjusted mean frequency values for all spectral parameters, and results of statistical comparisons across group compositions. All mean frequency values were highest in competitive groups. The fifth percentile frequency (a proxy for minimum frequency) had a mean of 384 Hz in dyads, 375 Hz in escorted mother-calf pairs, and 445 Hz in competitive groups (Figure 2). The mean peak frequency was 536 Hz, 544 Hz, and 657 Hz, while the mean ninety-fifth percentile frequency (a proxy for maximum frequency) was 1136 Hz, 1228, and 1276 Hz in dyads, escorted mother-calf pairs, and competitive groups, respectively.

Additionally, the duration of calls was compared across all groups. Duration was longest in competitive groups, with calls averaging 0.501 s in length, followed by dyads (0.475 s) and escorted mother-calf pairs (0.461 s).

There are many possible factors that can affect the temporal patterns of humpback whale social calls. Two such factors we considered strongly when interpreting these results were the effect of arousal and motivational context of the signaler. These factors have previously been shown to affect calling behavior (August and Anderson, 1987; Morton, 1977; Cusano et al., 2020), with call rates increasing in high arousal contexts, such as those associated with mating behavior (e.g., red deer, Cervus elaphus, Clutton-Brock and Albon, 1979). The calling patterns we observed in competitive groups reflected the nature of these social interactions. Competitions can last for many hours (Tyack and Whitehead, 1983; Clapham et al., 1992; Herman et al., 2007) and can be energetically costly for the whales involved. Intense periods of male displays are often observed (e.g., head lunging, jaw claps, linear bubble trails, blocking), as well as more aggressive actions such as chases and body strikes (Baker and Herman, 1984; Herman et al., 2007). On the Hawaiian breeding grounds, these competitions have been observed to also include short periods of rest between aggressive displays (Herman et al., 2007). The inter-call intervals in competitive groups were shortest out of all compositions, with only a few periods of extended silence between calls. It is possible that short inter-call intervals occurred during periods of aggression, while periods of silence coincided with rest, although further observation would be necessary to confirm this. The pattern of frequent calling in competitive groups observed in this study may indicate that calling plays an important function in humpback whale competitions.

In contrast to competitive groups, escorted mother-calf pairs produced calls less frequently with longer inter-call intervals. The longer ICIs observed in escorted mother-calf pairs were somewhat expected and are likely explained by known mother-calf calling behavior (Indeck et al., 2022) and the group interactions observed during the tag deployments. Although escorts show a preference for females without a calf, as they typically have a greater reproductive potential than maternal females (Craig et al., 2003), they will also prospect among mothers for those that may be in post-partum estrus (Chittleborough, 1958; Pallin et al., 2018). The addition of one or more escorts to a mother-calf pair increases their energy expenditure (Craig et al., 2014) and could result in injuries to the calf or its separation from its mother (Smultea, 1994). As such, these pairs must balance communication with each other with the potential of being joined by escorts, which is believed to be reflected by lower call rates and levels in these groups (Videsen et al., 2017; Indeck et al., 2022). Nevertheless, these pairs have been found to increase calling rates during periods of increased separation as a means of maintaining acoustic contact while continuing to limit detectability by nearby whales (Indeck et al., 2022). Similar calling behavior is likely also useful during non-agonistic interactions with a single escort, as a means of minimizing the risk of mother-calf separation, as well as reducing the possibility of attracting additional escorts. This could explain the lower calling rates observed in escorted mother-calf pairs here. With the exception of one tag deployment during which increased calling was associated with aggressive behaviors between the mother and escort, MCE groups in this study predominantly displayed low intensity interactions accompanied by low call rates and longer inter-call intervals.

Of all the groups, dyads were expected to produce the fewest social calls, consistent with observations made in previous studies (Silber, 1986; Cusano et al., 2020). While these groups did produce the fewest calls out of the three groups, we did not find call rates (calls/hour/whale) to vary significantly between groups. The temporal pattern of call production in dyads was similar to escorted mother-calf pairs with longer inter-call intervals observed in these groups. On the Hawaiian breeding grounds, dyads are most often comprised of a male and female (Pack et al., 2012) that often spend extended periods resting below the surface with no signs of agonism (Jones, 2010). The results observed here may be partly due to a lack of agonism associated with these groups (Jones, 2010) or may reflect a reduced need for these whales to communicate acoustically when in close proximity (Cusano et al., 2020). Furthermore, the simple fact that there were fewer whales in these groups may have contributed to the longer inter-call intervals observed.

Competitive groups, which are the most aggressive in nature, exhibited the highest frequency calls on average. This may be due in part to an increased level of intensity expected from the agonistic context of these groups, which has previously been linked with higher frequency calls (Lemasson et al., 2012). Additionally, the contribution of aversive calls in competitive groups may contribute to the higher frequencies observed, considering that aversive calls are typically associated with higher frequencies (Morton, 1977, August and Anderson, 1987; Dunlop, 2017).

Escorted mother-calf pairs produced calls with frequencies between that of dyads and competitive groups for all parameters, except fifth percentile. The contribution of calf calls could potentially contribute to increasing the frequency levels of calls in these groups. To an extent, the minimum frequency of calls in mysticete whales is limited by animal size (May-Collado et al., 2007; Martin et al., 2017), with smaller animals possibly being incapable of producing the lowest frequency calls. It is possible that a balance between calf calls, which are generally produced at higher frequencies than those of adults (Zoidis et al., 2008; Indeck et al., 2020), and calls from the adults in these groups, contributed to the frequency values observed

Frequency measurements were relatively consistent with expected patterns, which was also true for observed call durations. High arousal during high intensity contexts (i.e., competitive groups) is frequently associated with longer duration calls (Lemasson et al., 2012). This pattern was observed here, with competitive groups producing the longest duration calls. However, dyads, which were not observed interacting aggressively, were expected to have the shortest call durations, which were instead exhibited by escorted mother-calf pairs. It is possible, however, that the duration of calls could have been affected by the physical attributes of the signaler (i.e., body size, age class, vocal ontogeny/calling experience; Martin et al., 2017), rather than the dynamics of the group.