To flee or not to flee: anti-predator response of humpback whales on their breeding ground

Summary

Anti-predator behaviors are usually well conserved within species, although they may vary across individuals depending on several factors such as socio-ecological contexts. Using humpback whales (Megaptera novaeangliae), we investigated whether the flight anti-predator strategy observed in high-latitude feeding grounds is conserved in low-latitude breeding areas. Playbacks of predatory killer whale (KW) sounds were conducted in mother-calf pairs, some accompanied by a male escort. We quantified whales’ behavioral responses using multi-sensor tags and visual observations. We revealed clear directional movements of whales away from the sound source, with a significant increase in speeds and a switch to near-surface movement patterns in response to KW playbacks (n = 5). Male escorts switched to synchronous surfacing with mothers. In contrast, no or slight responses were observed in response to control playbacks (n = 4). Our results indicate a cohesive flight anti-predator strategy, conserved on feeding and breeding grounds but complemented by context-dependent social components.

Graphical abstract

Highlights

• •Mother-calf pairs accelerate and move away from killer whale sound playbacks
• •Near-surface movements are observed in response to predator sounds
• •Surface synchrony between escort and mothers reveals cohesive group escape
• •Anti-predator flight response is conserved in both breeding and feeding grounds

Marine organism; Wildlife behavior; Zoology

Introduction

Prey species have developed sophisticated mechanisms to detect predators, assess the level of predation risk, and perform effective protection strategies. According to the threat-sensitive hypothesis, prey modulate their anti-predator responses based on the perceived severity of the threat. Individuals benefit from recognizing the type and level of threats and effectively balance avoidance or defensive reactions with other essential activities.^1,2 For example, impala, zebra, and wildebeest adjust the type and intensity of anti-predator response according to predator hunting strategy (ambush versus coursing) and relative risk (e.g., predation pressure).³ Similarly, waterfowl and deer species flee from humans at greater distances during hunting seasons.^4,5 Predators and prey can detect each other through various sensory modalities including acoustic, visual, and chemical cues, depending on the characteristics of their environment and their intrinsic capabilities.⁶ In many bird and terrestrial mammal species, acoustics serve as an important sensory modality for mediating predator-prey interactions.⁷ A systematic literature review from Blumstein et al.⁶ found that 25 of 30 bird and mammal species responded to vocalizations from their predators. Predators typically remain silent during hunts when prey are capable of detecting their sounds. However, prey species may still opportunistically detect predator presence by recognizing sounds emitted in other social contexts and use this information to anticipate a potential attack. For marine mammal species, transmission of reliable information through visual and chemical cues is limited to short distances, whereas acoustic cues appear particularly relevant for detecting predators at long range.

Like most marine mammals, baleen whales exhibit two primary anti-predation strategies: flight and fight.^8,9 Flight strategies include rapid swimming away from threats, seeking refuges and shallow waters (less than 10 m deep) where predatory killer whales’ (KWs) attack tactics are constrained (e.g., less able to attack from below), “snorkel behavior” with quiet surfacing and minimally visible blows, and diving to deep waters or altering migration routes.⁸ For example, gray whales (Eschrichtius robustus) have been observed fleeing away when exposed to KW vocalizations, seeking refuge into nearby thick kelp when available.¹⁰ Similarly, common minke whales (Balaenoptera acutorostrata) attacked by KWs, have been observed to enter shallow waters, sometimes leading to stranding events.⁸ Fight strategies on the other hand involve formation of groups of conspecifics, including potential exhibition of defensive behaviors. For example, southern right whales (Eubalaena australis) respond with tail- and flipper-slapping when attacked and have been reported to strike KWs with their flukes, while also exhibiting group defense with nearby whales coming to help defend calves.^8,11 These anti-predation behaviors have been mostly documented through opportunistic observations of attacks.

Numerous observations on feeding and breeding grounds reported that humpback whales (Megaptera novaeangliae) engage in defense strategies while being directly attacked by or in proximity to mammal-eating KWs.^8,12 Some observations also documented humpback whales approaching and protecting other species nearby being attacked by KWs.¹³ However, humpback whales in Norwegian feeding grounds exposed to playbacks of mammal-eating KWs’ sounds acoustically simulating the presence of potential predators, did not engage any expected fight behavior. Instead, whales exhibited clear horizontal avoidance that was interpreted as a response to a perceived risk of potential upcoming attack.¹⁴ These avoidance responses were less pronounced inshore at the end of the feeding period than offshore at the beginning, suggesting that environmental and/or body conditions influence anti-predator responses.¹⁵ All together these observations indicate that humpback whales can modulate their anti-predator strategies, depending on several factors such as the perceived predation risk and ecological context.

KWs (Orcinus orca) are apex predators of marine ecosystems, with diverse diets and complex social structures. They are found in all ocean basins and have been categorized into different populations, based on their prey preferences, acoustic features, morphology, genetics, distribution and behaviors.¹⁶ Some populations prefer to consume fish, while others prefer marine mammals, i.e., preying upon pinnipeds and/or cetaceans, with young individual prey being particularly vulnerable to predation by KWs.⁸ Hunting strategies of KWs that eat marine mammals usually include stealthy approaches, coordinated group attacks, and communication methods adapted to isolate and capture prey.^17,18,19 Most marine mammals can hear KW sounds. Mammal-eating KWs do not use echolocation as frequently as foraging fish-eating KWs, and turn mostly silent while approaching their prey until the attack, likely to avoid being acoustically detected.²⁰ Indeed, studies conducted on a range of marine mammal species, including pinniped, toothed whale and baleen whale species, revealed that they change behavior in response to the vocalizations of potential predatory KWs (e.g., gray whales Eschrichtius robustus¹⁰; belugas Delphinapterus leucas²¹; harbor seals Phoca vitulina¹⁹; Blainville’s beaked whales Mesoplodon densirostris²²). Moreover, some species such as humpback whales can discriminate between non-threatening familiar fish-eating KW sounds and potential predatory unfamiliar KW sounds, allowing for appropriate threat assessment and adaptive responses.^15,19,23,24

Anti-predation behavioral responses fundamentally reflect trade-offs between survival benefits and associated costs, including energetic expenditure and lost opportunities for other activities such as foraging or reproduction.² These responses are usually consistent within species, although they are shaped by risk perception and a range of individual and contextual factors.^14,15 Individual-related factors include body condition, prior experience with predators, behavioral state, age, and gender. Social factors such as group composition and environmental features including habitat characteristics and availability of refuge may also shape the type of strategy adopted. Breeding and feeding grounds represent two contrasting socio-ecological contexts for humpback whales. While humpback whales in feeding areas are predominantly solitary or form small temporary groups, breeding grounds are characterized by a range of various social structures including competitive groups, pairs of adults, and mother-calf pairs.²⁵ Females with calves actively avoid male competitive groups and may carry calves out of the water on their back or increase travel speed when caught in aggressive situations.²⁶ Interestingly, mother-calf pairs can be found accompanied by one adult male called an “escort.”^25,27 However, little is known about the role of escorts that remain associated with a mother-calf pair. Two hypotheses have been suggested: Escorts are likely either prospecting for access to females experiencing postpartum ovulation or potentially mate-guarding after copulation.^25,27,28 In both cases, they may also play important roles in protecting the mother-calf pair, particularly in predation threat context,²⁹ as predation risk is much higher for neonates.^25,29 Indeed, Pitman et al.¹² observed escorts vigorously assisting mothers in shielding their calves from KW attacks.

Given the distinct social and ecological dynamics characterizing feeding and breeding grounds, we hypothesized that these contexts influence humpback whale responses to predator threats. More specifically, on breeding grounds, whales form more persistent and structured social groups, including numerous mother-calf pairs in which calves can be very young with limited escape capacity. On feeding grounds, whales do not form stable social bonds and are primarily observed alone or in temporary feeding-related associations, with calves, if present, being generally older and possessing more advanced motor abilities. We therefore expected that, on breeding grounds, detection of predator sounds would initiate cryptic (reduced vocal activity and surface avoidance) and defensive response components rather than the typical “flight” responses previously described on feeding grounds. To test this, we conducted playback experiments of KW sounds to humpback whales on their breeding grounds in the presence of calves and characterized their anti-predator responses. To date, playback experiments on humpback whales’ breeding grounds have focused solely on conspecific vocalizations (social sounds and songs), whereas predator sound playbacks have not been conducted yet and are experimentally explored here for the first time.^30,31,32

Results

We tagged the mother of five mother-calf pairs, three of which were accompanied by an escort. Total duration of tag deployments lasted 5.5 ± 2.9 h (mean ± SD) (Table S2). We conducted playback experiments on the five tagged mother humpback whales. One whale, mn240812, was only exposed to KW stimulus because it became too dark (nighttime) to pursue experiments, and the four other whales were exposed to the two stimuli, resulting in a total of four CTRL playbacks and five KW playbacks (Table S2). Tag data were collected for all three PRE/DUR/POST periods for all playbacks except one: the KW playback conducted in mn240826. In this case, the tag detached prematurely after 10 min of playbacks, likely due to the whale’s rapid acceleration, which prevented data collection for the remaining 5 min of the KW playback as well as for the POST period. Moreover, no visual observations were available for the DUR and POST periods of this playback, as the whale was lost from sight immediately after the playback began. Drone flight totaling 3 h 52 min of video footage covered the experimental phases of playbacks for four out of the five focal groups. There was no drone data on the 26th August 2024 because of connection issues between the drone and the control tablet that day. For the four other days of experiments, available drone data are summarized in Table S2. Drone footage from 2023 consisted of multiple short video segments corresponding to time portions for which the tagged whale or individuals of its group were visible, i.e., at the surface or near-surface. We maintained a near-continuous recording throughout the 2024 experiments with only a couple of minutes of interruptions when we had to return the drone back to the boat for battery changes (see Table S2). We recorded a total of 51.62 min of drone footage for the CTRL playbacks periods (N = 6 over 12 covered periods) including 2 PRE_CTRL (23.2 min), 2 DUR_CTRL (23.1 min) and 2 POST_CTRL (5.3 min), and 94.83 min of drone footage for the KW playback periods (N = 10 over 14 covered periods) including 4 PRE_KW (35.8 min), 3 DUR_KW (26.3 min), 3 POST_KW (32.7 min).

Effect of playback on quantitative response variables

For the response variables “horizontal movement reaction score,” “total surfacing duration” and “mean ODBA,” the ANOVA applied on the full GEE model retained the factor Playback (KW vs. CTRL) as significant. For the response variables “mean horizontal movement speed” and “max horizontal movement speed,” the ANOVA applied on the full GEE model retained both factors Playback and Period (PRE vs. DUR or PRE vs. POST) as significant. For the response variable “total duration of near-surface movements,” the ANOVA applied on the full GEE model retained both factors Playback and Period and their interaction term (Playback:Period) as significant. The factor Order was never retained as significant in the best model fit, indicating that this covariate did not explain the variance in the data for any of these response variables. For the other response variables (see Table 1 in Methods), the ANOVA did not retain any covariates as significant, indicating they are not influenced by any of the covariates Order, Period and Playback.

Table 1.

List of measured response variables and their corresponding analytical approaches

Response variables	Original source of data	Analytical approach
Horizontal movement reaction score (arbitrary unit)	tracking data	statistic (GEE)
Horizontal movement speed (mean, max, in m s-1)	tracking data	statistic (GEE)
Duration of near-surface movements (i.e., within 1–10 m, excluding resting mode) (total, in s)	tag data	statistic (GEE)
Surfacing duration (total, in s)	tag data	statistic (GEE)
ODBA (mean, in m s-2, relative index of motor effort)	tag data	statistic (GEE)
Vocalization (detections)	tag data	descriptive
Simultaneous surfacing (yes/no)	surface behavioral data (visual from boat and drone)	descriptive
Logging (yes/no)	surface behavioral data (visual from boat and drone)	descriptive
Milling (yes/no)	surface behavioral data (visual from boat and drone)	descriptive
Spacing mother-calf (predominant category)	surface behavioral data (visual from boat and drone)	descriptive
Spacing mother-escort (predominant category)	surface behavioral data (visual from boat and drone)	descriptive
Line-up mother-escort (predominant category)	surface behavioral data (visual from boat and drone)	descriptive
Calf between mother and escort (yes/no)	surface behavioral data (drone)	descriptive
Horizontal turn (number of occurrences)	surface behavioral data (drone)	descriptive
Number of whales within 200 m of the focal group	surface behavioral data (visual from boat)	descriptive
Distance to nearest group (predominant category)	surface behavioral data (visual from boat)	descriptive
Surface events (in number of occurrences)	surface behavioral data (visual from boat and drone)	descriptive

Horizontal movement

Movement reaction scores in response to KW playbacks were all negative (i.e., representative of an overall avoidance response) and significantly differed from the close-to-zero reaction scores (i.e., slight responses) in response to the CTRL playbacks (p < 0.001 Table 2) (Figure 1). These scores indicated that whales went further than predicted at the end of the DUR_KW period, whereas their location at the end of the DUR_CTRL period was overall concordant with prediction based on their PRE behavior. Because the movement reaction score depends on the change in both horizontal direction of movement and horizontal speed, its negative value in response to KW playback is due to either one or both components.

Table 2.

Results of the best GEE model fitted to each of the behavioral response variables

Response variable	Covariate	Coefficient estimates	P values
Movement reaction score	playback	KW = −1.77684 ± 0.53895	0. 000978 ∗∗∗
Mean horizontal speed	playback	KW = 0.61228 ± 0.09703	2.8e–10 ∗∗∗
	period	DUR = 0.60383 ± 0.11029	4.4e–08 ∗∗∗
		POST = 0.71158 ± 0.19996	0.00037 ∗∗∗
Max horizontal speed	playback	KW = 0.9085 ± 0.2942	0.00202 ∗∗
	period	DUR = 1.1366 ± 0.2673	2.12e–05 ∗∗∗
		POST = 1.2187 ± 0.1926	2.51e–10 ∗∗∗
Total duration of near-surface movements	playback	KW = −20.92 ± 52.81	0.69201
	period	DUR = −55.62 ± 26.59	0.03644 ∗
		POST = 2.00 ± 25.71	0.93798
	playback: period	KW:DUR = 304.93 ± 96.67	0.00161 ∗∗
		KW:POST = 277.77 ± 96.53	0.00401 ∗∗
Total duration of surfacing	playback	KW = −162.66 ± 76.07	0.0325 ∗

Each GEE model was selected based on the results of the ANOVA (see Table S4). In all models, the reference level was CTRL for the covariate Playback, and PRE for the covariate Period. Note that the full GEE model applied to the movement reaction score did not have any Period covariate because this response variable has just one value representing a change from PRE to DUR. Coefficient estimates and p values are provided for each covariate. The reference level was CTRL for the covariate Playback and PRE for the covariate Period. ∗p < 0.05, ∗∗p < 0.01, and ∗∗∗p < 0.001.

Figure 1.

Movement reaction scores of all tagged whales for the two playbacks (KW or CTRL)

Movement reaction scores quantify changes in horizontal movement between the PRE and the DUR periods of each playback. These scores are based on the difference between the actual distance whale-source at the end of the playback and the predicted distance calculated from the PRE period movement. Score values for each individual whale are plotted with black symbols and connected by black lines. Min, max, and median movement reaction scores for each playback are indicated. Pairwise GEE results show a significant difference between the two playback stimuli (KW versus CTRL: p < 0.001∗∗∗). Note that mn240812 (pointing down triangle symbol) was only exposed to KW playback. A negative movement reaction score indicates an avoidance of the sound source.

Regarding horizontal direction of movement, visual tracks showed that all tagged whales moved away from the playback source during KW playbacks (Figures 2B, 3A, 4A, 5A, and 6B). Four out of five tagged whales changed their initial course to avoid the playback source (Figures 2A–2E) the other whale was resting before the playback and started traveling as soon as the playback started, maintaining a directional course while moving away from the source (mn240806, Figures 4A and 4B). Moreover, one group (mn230819) that was relatively far from the coast (>7 km) before the beginning of KW playback moved toward the coast when the playback started. Whales already near the coast (<2.5 km for mn240806 and < 1 km for mn240826) followed the coastline during their avoidance (Figures 4A and 6B). The two other groups were farther from the coast (>8 km for mn230815 and >15 km for mn240812) and did not approach the coast. By contrast, during CTRL playbacks, only two whales (mn240826 and mn230819) changed their trajectory, but did not avoid the sound source. Mn240826 turned toward the sound source while switching from resting to traveling. Mn230819 changed travel direction during CTRL playback, leaving its near-coast position to head farther out and passing close to the sound source: This shift followed an earlier coastward avoidance movement response during the KW playback (Figure 3A).

Figure 2.

Behavioral response of mn230815 to CTRL and KW playbacks

(A) horizontal track during CTRL playback periods (B) horizontal track during KW playback periods, (C) diving profile, and D) overall dynamic body acceleration.

(A) and (B) tracks correspond respectively to playback #1 and playback #2. Magenta circles and triangles on tracks represent the position of the sound source at the start and end of playback, respectively. Pre- and post-playback exposition periods are colored in light and dark blue, while KW- and CTRL-exposition are colored in red and black, respectively. The presence of the two whales symbol above a 11:00 indicates simultaneous surfacing between the mother and the escort during this period (majority of sightings associated with simultaneous surfacing). Zooms on playback periods are available in Figures S2–S6.

Figure 3.

Behavioral response of mn230819 to KW and CTRL playbacks

(A) horizontal track during KW playback periods (B) horizontal track during CTRL playback periods, (C) diving profile and (D) overall dynamic body acceleration.

(A) and (B) tracks correspond respectively to playback #1 and playback #2. Magenta circles and triangles on tracks represent the position of the sound source at the start and end of playback, respectively. Pre- and post-playback exposition periods are colored in light and dark blue, while KW- and CTRL-exposition are colored in red and black, respectively. The presence of the two whales symbol above 08:30–09:00 indicates simultaneous surfacing between the mother and the escort during this period (majority of sightings associated with simultaneous surfacing).

Figure 4.

Behavioral response of mn240806 to KW and CTRL playbacks

(A) horizontal track during KW playback periods B) horizontal track during CTRL playback periods, (C) diving profile, and D) overall dynamic body acceleration.

(A) and (B) tracks correspond respectively to playback #1 and playback #2. Magenta circles and triangles on tracks represent the position of the sound source at the start and end of playback, respectively. Pre- and post-playback exposition periods are colored in light and dark blue, while KW- and CTRL-exposition are colored in red and black, respectively.

Figure 5.

Behavioral response of mn240812 to KW playback

(A) horizontal track during KW playback periods, (B) diving profile and (C) overall dynamic body acceleration. Magenta circles and triangles on tracks represent the position of the sound source at start and end of playback, respectively. Pre- and post-playback exposition periods are colored in light and dark blue, and the during-exposition period is colored in red.

The presence of the two whales symbol above 13:45–14:00 indicates simultaneous surfacing between the mother and the escort during this period (majority of sightings associated with simultaneous surfacing).

This tagged whale was not exposed to CTRL stimulus.

Figure 6.

Behavioral response of mn240826 to CTRL and KW playbacks

(A) horizontal track during CTRL playback periods (B) horizontal track during KW playback periods, (C) diving profile and D) overall dynamic body acceleration.

(A) and (B) tracks correspond respectively to playback #1 and playback #2. Magenta circles and triangles on tracks represent the position of the sound source at the start and end of playback, respectively. Pre- and post-playback exposition periods are colored in light and dark blue, while KW- and CTRL-exposition are colored in red and black, respectively.

Note that the tag detached during the DUR period of the KW playback.

Regarding the horizontal speed component, each covariate Period and Playback significantly and independently influenced this response variable. Mean and max speeds significantly increased during and after playback compared to the PRE period, independent of the stimulus type (GEE: PRE vs. DUR, p < 0.001 and PRE vs. POST, p < 0.001, Table 2). Moreover, mean and max speeds were significantly higher for KW playbacks than for CTRL playback, independent of the experimental period (GEE: KW vs. CTRL, p < 0.001, Table 2). All 5 tagged whales increased speed from the PRE_KW to DUR_KW period, and two of them (mn230815 and mn240806) increased speed even more over the POST_KW period, with one whale reaching up to 2.9 m s^-1 (Figure 7A). During CTRL playback, mn230815, mn240806 and mn240826 increased speed while switching from milling or resting to slow travel, and mn230819 did not increase speed while passing by the sound source when leaving the coast to go back farther to sea.

Figure 7.

Mean of selected behavioral variables across all tagged whales for each of the three 15 min periods (PRE, DUR and POST) of the two playbacks (KW and CTRL)

(A) within-individual mean horizontal speeds, (B) total durations of near-surface movements, and C) total surfacing duration of all tagged whales. Values for each individual are plotted with distinct black symbols, and values from the same individual are linked by thin black lines. Min, max, and median values are indicated for each period of each playback. Pairwise GEE results showed a significant increase in mean horizontal speed from PRE to DUR (p < 0.001∗∗∗) and to POST (p < 0.001∗∗∗) periods, and significantly higher values during KW than CTRL playback periods (p < 0.001∗∗∗). GEE statistics showed a significant increase in total duration of near-surface movements from PRE_KW to DUR_KW (p < 0.01∗∗) and to POST_KW (p < 0.01∗∗) periods, but no significant differences between KW and CTRL playback (p > 0.05). GEE results confirmed significantly lower values in total surfacing durations during KW than CTRL playback periods (p < 0.5∗). Note that mn240812 (pointing down triangle symbol) was only exposed to KW playback, and that there was no POST_KW data for mn240826 (diamond symbol) as the tag detached during playback.

Diving behavior and locomotor activity

Inspecting the dive profiles and ODBA, the three whales that were traveling in PRE_KW, and the one that was resting (mn240826, Figures 6C and 6D) and that started a dive (ODBA >0.02 m s^-2) just before the end of PRE_KW, switched to near-surface traveling mode during KW playback. The whale (mn240812, Figures 5C and 5D; Figure S5) that was already near-surface traveling in PRE_KW continued in this mode over the KW playback. By contrast, the two whales (mn240806, Figures 4C and 4D, and mn240826; Figures 6C and 6D) that were resting in PRE_CTRL did not interrupt their behavior when exposed to the CTRL playback, and the two other whales either continued their PRE dive pattern while traveling (mn230819, Figures 3C and 3D) or started resting during the playback (mn230815, Figures 2C and 2D). The change in dive patterns in response to KW playback is shown by the significantly increased time spent in near-surface movements (1–10 m depth) during DUR and POST period of KW playbacks compared to the overall lack of change over the three periods of CTRL playbacks (GEE: Playback:Period, p < 0.001, Table 2), and significantly higher mean ODBA for KW playbacks than for CTRL playback, independent of the experimental period (GEE: Playback, p < 0.01, Table 2). Specifically, the time spent in near-surface movements increased 5-fold over the KW playback: from 72 ± 38 s (mean ± SEM) in PRE_KW to 321 ± 109 s in DUR_KW and 351 ± 91 s in POST_KW (Figure 7B). CTRL playbacks did not show such a response, with total near-surface movement durations remaining much lower, with 92 ± 44 s in PRE_CTRL, 37 ± 14 s in DUR_CTRL, and 110 ± 47 s in POST_CTRL (Figure 7B). Moreover, whales spent less time at the surface (0–1 m depth) during KW playback periods than during CTRL playback periods (GEE: KW vs. CTRL, p < 0.05, Table 2), with averaged total surfacing durations of 242 s and 404 s, respectively (Figure 7C).

Descriptive analysis of other behavioral variables

Surface events and social behavior

Three of the five tagged whales were accompanied by an escort (mn230815, mn230819, and mn240812). The combination of visual observations and drone data provided detailed insights into social behaviors between mothers and escorts (Figures S7, S8, and S10). While escorts typically remained out of sight at depth, they returned to the surface during KW playbacks. A clear simultaneous surfacing between mothers and escorts was observed from both visual observations and drone data during the DUR_KW and/or POST_KW periods (Figures 2C, 3C, and 5C). In one case (mn230815), several changes in the focal group’s direction were observed during CTRL playback periods in drone footage. Horizontal turns were initiated by the escort and subsequently followed by the mother (Figure 2A). In contrast, in the two other cases (mn230819 and mn240812), the escort did not initiate any particular trajectory change, and when horizontal turns occurred, they were initiated by the mother. Regarding escort position relative to the mother, no consistent social configuration emerged across groups. One group (mn240812) switched from “next to each other” during PRE_KW and DUR_KW to “following each other” in POST_KW, with first the mother leading and then the escort leading. The two other groups (mn230815 and mn230819) maintained a “next to each other” configuration throughout all three KW playback periods. When exposed to CTRL playback, focal group mn230815 switched from “no formation” in PRE_CTRL to “next to each other” during both DUR_CTRL and POST_CTRL. Group mn230819 maintained a “following each other” configuration throughout all three CTRL playback periods. When escorts were visible, they remained close to mothers (<1 BL, body length), except for mn240812’s escort, which maintained 1–3 BL distance. Calves maintained proximity to their mothers throughout observations (<1 BL). Notably, calves mainly positioned themselves adjacent to the mother on the opposite side of the escort (CME) rather than between the mother and escort (MCE). The observation of the MCE arrangement was recorded twice during both mn230815’s PRE_CTRL period and mn240812’s PRE_KW period; all other observations were calf back-carried on the mother’s rostrum or on the opposite side of the escort (CME). Logging events were observed for two groups (mn230815 and mn240812). All logging events occurred either during CTRL periods or in the PRE period of KW playback but never during DUR_KW or POST_KW periods. Milling events were observed in two groups (mn240812 and mn230815). For mn240812, milling was observed during both DUR_KW and POST_KW periods but not during PRE_KW. For mn230815, milling was observed in all periods for CTRL playback, and only in PRE_KW period for KW playback. No particular other surface displays (e.g., breaches, caudal or pectoral slaps, spyhops) were observed during any playback period, except for one spyhop in mn240826 DUR_CTRL.

Number of individuals in the area and distance to the nearest group

Two parameters were observed: number of whales within 200 m of the focal whale and distance to the nearest other whale(s). Regarding KW playbacks, the two individuals (mn230815 and mn240812) that were more than 8 km from the coast remained at a distance from the nearest other group for all three experimental phases, i.e., 500–1000 m in PRE_KW, >1 km in DUR_KW and POST_KW. Conversely, the two individuals (mn230819, mn240806) that were near the coast in the PRE_KW period ended up closer to the nearest group in DUR_KW and/or POST_KW. In particular, for mn230819, no other group was sighted during the PRE period (within a 1 km radius of observation) whereas a whale was seen within 50–100 m of the focal group during the KW playback, resulting in an increase of the number of individuals within a 200 m radius of the tagged whale. Regarding CTRL playbacks, two groups (mn230819 and mn240806) increased nearest group distances from 100 to 200 m in PRE_CTRL to >1 km in POST_CTRL, thus showing less individuals within 200 m from the tagged whale. Conversely, mn230815’s distance to the nearest group decreased from >1 km in PRE_CTRL to 200–500 m in POST_CTRL. No groups were observed within 1 km of mn240826 during any CTRL or KW playback period.

Each individual behavioral response is summarized in Table S5.

Vocal production

Overall, we detected very few social sounds across all playback periods (range: 0–5 calls per 15-min period; n = 26 periods across all playbacks for both CTRL and KW stimuli). No vocalizations were detected for mn230815 during any playback. For mn230819, five vocalizations were detected in PRE_KW and one in DUR_KW, but none during CTRL playback periods. For mn240806, three vocalizations were detected in PRE_CTRL and six in DUR_CTRL, but no vocalizations were detected during any KW playback periods. One vocalization was detected during DUR_KW for mn240812, which was only exposed to the KW stimulus. For mn240826, one vocalization occurred in PRE_CTRL and two in PRE_KW. Notably, no vocalizations were detected during any POST_KW period across all groups.

Discussion

Our playback experiments revealed consistent flight anti-predator behavioral responses in humpback whale mother-calf pairs with or without male escorts when exposed to predatory KW sounds. These responses were expressed through horizontal and rapid near-surface avoidance and social dynamics between mothers and escorts. The whales barely responded to the CTRL stimulus and did not show any avoidance response, suggesting that they likely did not perceive this stimulus as a threat. This stimulus could resemble a noise generated by the engine of a vessel and is possibly perceived as a sound heard frequently by the whales. The response differences between the negative control and KW sounds show that both stimuli were detected, but that the anti-predator responses are specifically induced by KW calls.

The humpback whales exhibited directional movement away from the KW sound source, increasing their average speed from 1.4 up to 2.2 m s^-1 (i.e., 8 km h^-1). Although mean horizontal speeds also increased during CTRL playbacks, they did not reach the high speeds observed in KW playbacks, increasing their average speed from 0.9 up to 1.5 m s^-1 (i.e., 5 km h^-1) in response to CTRL playbacks. Moreover, increases in speed during CTRL playbacks were associated with changes from resting to baseline traveling mode, and thus may not be indicative of disturbance. The speeds observed in response to KW playbacks are consistent with humpback whale flight responses described during similar playbacks in feeding grounds,¹⁴ suggesting that in breeding areas, even in the presence of vulnerable and slower young individuals, humpback whales keep a horizontal high-speed escape strategy. Furthermore, we had no information about the previous predation experience or body condition of focal animals, both of which could have influenced threat perception and response type and/or intensity. Finally, the use of two different amplifiers (Table S1) and the substantial variability in the distance at start of playback resulting in differences of received levels may have introduced some variability in whale responses. Maximum speeds reached 16 km h^-1, but these speeds would likely be insufficient to evade direct KW attacks, which typically occur at speeds exceeding 15–20 km h^-1.⁸ Humpback whales, as well as gray whales, are typically classified as fight strategists⁹ in response to predator attacks, as their morphology is better adapted for defensive behaviors than for sustained high-speed escape responses.⁸ However, the flight responses to predator sound playbacks observed in humpback whales (present study and Curé et al.¹⁴) and gray whales¹⁰ suggest that this binary fight versus flight classification needs refinement. In fact, the fight strategist species appear to exhibit either flight or fight responses, while flight strategist species have always been observed using flight responses. One possible hypothesis to explain this behavioral flexibility in fight strategists is the stage of predation risk: Flight behavior may serve as the initial response upon detecting distant predation risk, allowing whales to increase distance from potential threats and seek advantageous positions. However, once an attack is initiated, their morphology likely makes defensive behaviors the most effective anti-predation strategy.

We quantified whale horizontal movement responses to playback using a reaction score that is based on a comparison between the real response and a predicted response built from the average behavior of the 15-min period. Because heading and speed could vary substantially within the PRE for some cases, our level of confidence in the predicted movement could also vary (Table S3). Therefore, we carefully interpreted movement reaction scores in conjunction with individual horizontal track and speed data, which controlled for this possible bias and strengthened the robustness of our result.

A notable finding was the prolongation and amplification in some cases of the flight behavior during the POST_KW playback period, with horizontal speed continuing to increase even after the KW playback ended. This two-phase response aligns with observations in feeding grounds.^14,24 This phenomenon may represent an adaptive response to the cessation of predator vocalizations, a situation where KWs are no longer acoustically localizable, which might be perceived as an increased risk of threat since KWs typically turn silent during hunting.³³ The interruption of audible predator cues might be interpreted as an escalated threat level, triggering stronger evasive action. This temporal pattern appears to be a consistent element of humpback whale anti-predator strategy regardless of socio-ecological context.

It is important to note the clear distinction between near surface (1–10 m depth) and surface (0–1 m depth). Anti-predator responses included cryptic components with clearly increased motor activity near the surface associated with fast traveling while spending less time at the surface for breathing or resting. All tagged individuals that were not already traveling near the surface interrupted their diving or resting behaviors during KW playback. These whales switched to near-surface movements with no further resting phases or other dives performed during and after the playback, spending less time at the surface. Such a response (interrupting behavior and remaining sub-surface) has also been documented in right whales in response to alerting stimuli.³⁴ In our case here, remaining cryptic while fleeing away would include behavioral strategies to avoid being detected either acoustically (by remaining near the surface) or visually (by remaining under the surface). Therefore, remaining right below the surface appears a good compromise to prevent visual detection (by remaining undetectable by a KW spying from the surface) and acoustic detection (by not vocalizing or vocalizing near the surface where propagation is constrained). By contrast, the two whales that were resting before the start of the CTRL playback did not cease their resting activity. One whale even started resting during the CTRL playback, showing that this sound stimulus was not perceived as a threat. One could interpret resting as a potentially freezing antipredator strategy as observed in some taxa.⁷ However, freezing has never been documented in cetaceans, which typically respond to predation risk with active avoidance or defensive behaviors.⁸ Although humpback whales are known to typically dive within 60–300 m and can occasionally reach depths down to 600 m,³⁵ a vertical escape strategy would have been limited to 60 m, corresponding to the maximum depth of the study area, unless the whales leave the channel to access deeper offshore waters. A preference for shallow rather than deep dives in response to KW sounds was also reported in feeding ground where humpback whales shifted from a typical pattern of shallow breathing dives interspersed with deep foraging dives, to a regular intermediate shallow depth-diving travel mode during KW playbacks.¹⁴ Similarly, sperm whales interrupt foraging or resting dives and return to the surface when exposed to KW sounds.³⁶ Narwhals have also been observed exhibiting shallow breathing when detecting predation risk, and some air breathing fish species have been reported breathing less frequently and spending more time in cover in the presence of predator.²

Focusing on the vocal production of the focal group, only low-frequency calls were detected through all acoustic recordings. Moreover, the absence of detection of social sounds during KW playback could represent a cryptic anti-predation strategy as observed in various animal species, wherein individuals significantly reduce or stop their vocal activity upon detecting a potential threat to avoid being noticed. This behavior serves as a critical adaptation that enhances survival by minimizing the likelihood of detection by predators, especially for flight strategist species.⁹ Numerous studies across different taxa, including birds and mammals, have documented this phenomenon (e.g., Curé et al.³⁶; Laidre et al.³⁷; Mougeot and Bretagnolle³⁸). Fight-strategy species produce relatively loud calls at frequencies easily detectable by KWs, reflecting their general reliance on group defense rather than concealment.⁹ Alternatively, this absence of sound detection on the tag during KW playback could be due to masking by increased flow noise (approx. frequency band: 0–250 Hz) correlated to increased speed while the whale swims faster. Moreover, social calls can be produced at a low amplitude level^39,40, which might have also contributed to misdetection.

Another striking finding in the anti-predator response was the change of travel direction toward shallower coastal waters, which occurred when whales were sufficiently close to shore for this strategy to represent a realistic option. Three of the five focal whales in our study moved toward the coast during KW playbacks. Movement to shallow coastal waters may indicate a potential refuge-seeking behavior.⁸ Such areas constrain KW hunting tactics: Shallow depths limit attack angles, approach speeds, and the ability to position below or repeatedly ram large prey, collectively reducing the effectiveness of predatory attacks on humpback whales.^8,9 One whale actively moved toward the coastline during its avoidance response and subsequently returned offshore after playback ended. The two other whales, already positioned near the coast, followed the coastline during their escape response. Two whales did not seek coastal waters: One was located too far from shore (>15 km), while the other, though closer (8 km), had the playback source positioned between itself and the coast, which likely influenced its behavioral response. Notably, for whales located close to shore (<8 km), a possible refuge-seeking behavior may have also occurred in non-focal individuals present in the experimental area, suggesting that nearby whales exposed to the playback stimuli exhibited a similar strategy of response. A “collective” coastal movement may indicate either a common strategy shared by mother-calf pairs present in the area that were also exposed to the playback, or a potential initiation of a grouping defensive behavior in preparation for a coming attack as social response appears to be primarily employed when whales are positioned near coast.⁹ Humpback whale mothers with young calves naturally tend to stay close to the shore in breeding grounds,^41,42,43 and our findings confirm that when coastal waters are accessible, mother-calf pairs exhibit this movement toward the coastline. Refuge-seeking behavior represents a widespread anti-predator tactic documented across diverse taxa, from invertebrates to large vertebrates, where prey seek environmental refuges in response to predator presence.^2,8,44 In marine environments, some cetacean species exploit shallow waters that may limit the maneuvering advantage of KWs, thereby increasing the effectiveness of defensive behavior against predation. Large baleen whales such as gray and right whales, as well as smaller cetaceans such as dusky dolphins (Lagenorhynchus obscurus) and narwhals (Monodon monoceros), move into shallower waters, and bowhead whales use sea ice as protected cover from KWs.^{11,13,37,45,46}

The presence of escorts in three of the five tagged mother-calf pairs aligns with established literature documenting the prevalence of these social formations on breeding grounds.²⁶ Escorts were almost never visible at the surface during PRE phases but appeared during KW playbacks. While Henderson et al.⁴⁷ and Ransome et al.²⁶ primarily frame escort behavior within the context of mating competition, Pitman et al.¹² and Seeary et al.²⁹ proposed that escorts may also serve anti-predator functions, exhibiting defensive behaviors such as fluke slapping, charging toward KWs, and trumpeting, though they documented one escort abandoning a pair during an attack. Our study revealed no evidence of difference in non-social behavioral responses between escorted and unescorted mother-calf pairs. Escort presence did not seem to influence horizontal avoidance or diving behavior, as all groups actively avoided the sound source and increased speed while switching to near-surface movements. However, the small sample size (5 tagged groups) may have limited our ability to detect potential effects of escort presence. Additionally, the presence of calves may have constrained response strategies, whereas solitary individuals or competitive male groups might exhibit different anti-predator behaviors. During avoidance responses to predator sounds, escorts remained with their respective pairs throughout all playback experiments and maintained synchronous surfacing patterns and close proximity to mothers, demonstrating coordinated responses to perceived threats.²⁹ This behavioral synchrony may allow group members to maintain spatial coordination and reduce separation risk during flight responses.^48,49 Notably, calves consistently remained close to mothers but rarely positioned themselves between mothers and escorts—a spatial arrangement that may reduce the risk of inadvertent calf injury during escort-mother interactions.²⁶ However, this positioning contrasts with what might be expected during flight responses, where calves between the MCE would gain greater protection from external threats.

In spite of a low sample size (n = 5) and some limitations of our study that may have introduced some variability, our results confirmed a clear and consistent flight anti-predator strategy in humpback whales in early detection of predator presence on breeding grounds, similar to what was demonstrated in a feeding context.¹⁴ Our results show that the categorization of species between fight or flight anti-predator strategies needs to be refined, considering that fight species can also exhibit flight depending on whether the predator has detected the prey or not. Predation risk perception, social composition, and individual-level factors may significantly modulate these behavioral responses. Future research should examine how anti-predator responses vary across these factors. Specifically in humpback whales, studies could compare flight responses among lone males, competitive male groups, and mother-calf pairs to determine whether social context influences anti-predator strategies. Understanding these context-dependent strategies is critical for understanding how species assess and respond to predation threats across different socio-ecological and environmental contexts. These insights into context-dependent anti-predation strategies provide valuable frameworks for predicting species vulnerability to anthropogenic disturbances in various contexts, as animals perceive and respond to anthropogenic stimuli using similar risk-assessment mechanisms.^50,51

Limitations of the study

The small sample size (n = 5 tagged groups) limited the ability to perform statistical analyses and to detect potential effects of escort presence. Additionally, the lack of information on individual animals’ prior predation experience and body condition could have influenced threat perception and response intensity.

Resource availability

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Loanne Pichot (loanne.pichot@univ-eiffel.fr).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• •Tag and drone data reported in this paper will be shared by the lead contact upon request.
• •This paper does not report original code.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

We warmly thank the Cétamada team members and interns who contributed to the data collection. The authors are also grateful to Volker Deecke for providing the D-tag acoustic recordings of killer whale sounds. This study was supported and funded by the Centre National de la Recherche Scientifique (CNRS) and Greenov company. L.P. received a Ph.D. scholarship from the École Normale Supérieure de Lyon. The data collections were conducted under the national research permits #35/23-MPEB/SG/DGPA and #24/MPEB/SG/DGPA and complies with the European Union Directive on the Protection of Animals Used for Scientific Purposes (EU Directive 2010/63/EU).

Author contributions

Conceptualization, O.A., I.C., and C.C.; methodology, L.P., C.H., O.A., I.C., and C.C.; investigation, L.P., M.N.R., O.A., I.C., and C.C.; data curation, L.P. and S.M.M.; formal analysis, L.P., M.N.R., and S.M.M.; writing – original draft, L.P.; writing – review and editing, M.N.R., S.M.M., C.H., O.A., I.C., and C.C.; funding acquisition, I.C. and C.C.; resources, M.N.R. and C.H.; supervision, O.A., I.C., and C.C.; validation, O.A., I.C., and C.C.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Software and algorithms
MATLAB	The MathWorks, Inc.	https://mathworks.com/
R v.3.0.2	R Development Core Team 2013	https://www.r-project.org/
CATS-Methods-Materials	Cade et al.52	https://doi.org/10.1186/s40317-021-00256-w
geepack package for R	Højsgaard et al.53	https://doi.org/10.18637/jss.v015.i0
Boris	Friard and Gamba54	https://doi.org/10.1111/2041-210X.12584
Logger 2010	IFAW	https://www.marineconservationresearch.co.uk/downloads/logger-2000-rainbowclick-software-downloads/
Raven	Cornell Lab of Ornithology	https://www.birds.cornell.edu/ccb/raven-pro/

Method details

All procedures involving live animals followed the American Society of Mammalogists guidelines⁵⁵ and were approved by the Ministry of Fisheries Resources of Madagascar under the national research permits #35/23-MPEB/SG/DGPA and #24/MPEB/SG/DGPA.

Study site

Our study was conducted in the Sainte Marie Channel, located between Sainte Marie Island (16°50′ S, 49°55′ E) and the east coast of mainland Madagascar. We collected data during two consecutive humpback whale breeding seasons in August 2023 and 2024, coinciding with the annual peak of whale encounters in the Sainte Marie Channel between July and August.⁴³

General protocol

Our protocol included three main phases: 1) tagging a mother whale with non-invasive tags followed by a 1-h post-tagging recovery period; 2) conducting sound exposure experiments while simultaneously collecting observational data tracking the position and surface behavior of the tagged whale group; and 3) retrieving detached tags and downloading stored data for a posteriori analysis. To maximize tagging success, two teams operating from two separate 7-meter motorboats were deployed to search for whales and conduct tagging operations when feasible. Following successful tagging, one boat served as the ‘tracking boat’ for visual monitoring while the other became the acoustic ‘source boat’ for playback experiments. Humpback whales were observed in various social configurations. In the present study, we focused on mother–calf pairs either accompanied by a single escort or without escort to minimize variance associated with group composition. Only mothers were tagged for this study.

Tagging procedure and tag sensors

We used Customized Animal Tracking Solutions (CATS) cam tags,⁵⁶ attached to females via four silicone suction cups. These tags are equipped with a range of sensors including a 4 k video camera, a hydrophone (HTI-96-MIN, frequency sampling: 48 kHz), a pressure sensor, 3D magnetometers, and accelerometers (sampled at 10 Hz). These components record over time, respectively, video footage of the tagged whale surrounding, sounds produced by the focal group (i.e., the mother-calf pair and the escort if any) and other sounds received on the tag, depth, and 3D movements⁵⁷ (pitch, roll, and heading). Moreover, a VHF transmitter (ATS F1835B, Advanced Telemetry System, Isanti, Minnesota) was mounted on the tag, enabling the localization of the tagged whale at each surfacing using a VHF receiver from the tracking boat (ATS R410). This VHF receiver was also used to recover the tag after detachment from the whale.

The attachment of the CATS on the whale was conducted from one of the two rigid motorboats and consisted of slowly approaching the targeted mother-calf pair and deploying the tag using a 3-m hand-held carbon-fiber pole (tagging procedure detailed in Ratsimbazafindranahaka et al.⁵⁸). This tagging protocol was followed by a recovery post-tagging period of 58 ± 8 min (range: 49–67 min, n = 5) before starting playback experiments. This duration is longer than the time needed for whales to recover from the known short-term effects, such as a brief dive or change in swimming speed or direction.⁵⁹ All mothers’ caudal and/or dorsal fins were photo-identified to avoid double-sampling within and across breeding seasons.

Playback experiments

Acoustic stimuli

The playback protocol was designed to expose each tagged whale to a set of two sound stimuli broadcasted in a randomized order: a control stimulus (CTRL) and a killer whale sound stimulus (KW). KW stimuli were natural sound sequences of unfamiliar marine-mammal-eating killer whales (North Pacific transient population)⁶⁰ and the associated CTRL stimuli consisting of broadband ambient noise extracted from the inter-call intervals of the KW acoustic recordings and amplified to achieve similar root-mean-square (RMS) acoustic intensity. The main energy of the CTRL signals was concentrated within 1 kHz and 2 kHz, which match the dominant frequency components of the KW stimuli (see Curé et al.³⁶ for details). For each pair of stimulus types (KW and CTRL), two different versions were made, collected from different KW acoustic recordings, to reduce risk of pseudo-replication.⁶¹ Each stimulus lasted 15 min, and a recovery period of 45min was applied between the end of the first stimulus and the start of the second stimulus (mean ± SD: 50 ± 14 min, n = 4). The sound pressure level of the KW stimuli ranged from 151 to 173 dB rms re 1 μPa @1 m (mean ± SD: 161 ± 10 dB rms re 1 μPa @1 m, n = 5 stimuli) which is consistent with natural source level of killer whale vocalizations⁶² and clearly exceeded the background noise level (∼100 dB re 1 μPa). The sound pressure level of CTRL stimuli ranged from 158 to 174 dB re 1 μPa @1 m (mean ± SD: 166 ± 8 dB rms re 1 μPa @1 m, n = 4 stimuli). Estimated received levels ranged from 110 to 133 dB re 1 μPa @1 m (mean ± SD: 120 ± 11 dB rms re 1 μPa @1 m, n = 5 stimuli) at KW playback onset, and from 119 to 135 dB re 1 μPa @1 m (mean ± SD: 125 ± 7 dB rms re 1 μPa @1 m, n = 4 stimuli) at CTRL playback onset. Table S1 summarizes source levels, received levels and distances between tagged whales and playback sources at the beginning of each playback.

Playback chain and acoustic monitoring

Acoustic stimuli were transmitted using a Tascam DR40 player and amplified by a Sony XM-N503 amplifier in 2023, and an AudioSud amplifier associated with a LLAC202 impedance matching interface in 2024. The amplifier was connected to an omnidirectional Lubell LL9642T underwater loudspeaker (frequency range: 0.2–20 kHz) deployed from the source boat at 8 m deep, engine off. This loudspeaker is well-suited for broadcasting the playback stimuli whose dominant frequencies fall within the speaker’s optimal frequency range (1–10 kHz). To ensure afterward that sounds were faithfully played back without distortion and at the target sound level, stimuli were recorded during playbacks using a Bruel & Kjaer 8104 hydrophone, fixed on the rope of the loudspeaker 1 m above, plugged to a pre-amp (Nexus Bruel & Kjaer 2635) and connected subsequently to a recorder (Tascam DR680 MKII) and a sound level meter (Bruel & Kjaer 2250), wired in parallel, allowing simultaneous real-time monitoring and digital recording of the playback signal. The playback levels were also measured in real time during sound exposures using a sound level meter (Bruel & Kjaer 2250) that was previously calibrated with the hydrophone. Differences in source levels measured for the broadcasted stimuli (Table S1) are attributable to the use of two amplifier models but remain within an acceptable range for comparison.

Geometry of the playback experiments

Close communication was maintained between the tracking boat and the source boat throughout the entire experiment. Before each stimulus playback started, the source boat was guided by the instructions provided by the tracking boat through radio communication to take position ahead and to the side of the tagged whale’s trajectory, approximately 500 m away from the tagged whale (same geometry as in Curé et al.¹⁴). Once the position of the boat source was validated by the tracking boat, the source boat engine was turned off and the playback chain was deployed to start the broadcast.

Experimental phases

We defined three experimental phases: (1) a 15-min pre-exposure period (PRE) corresponding to the collected behavior preceding the start of playback, to establish the focal group’s baseline behavior and estimate its direction of travel, (2) a 15-min exposure period (DUR) to quantify potential behavioral changes exhibited during the playback, and (3) a 15-min post-exposure period (POST) following the end of broadcast to assess whether behavioral changes started at the cessation of playback rather than during it, or whether a change that may have occurred during playback persisted beyond its end or returned to the PRE behavior.

Collection of tracking and surface behavioral data

The tracking boat observed the focal whale group throughout the three experimental phases PRE, DUR and POST of each playback experiment, maintaining a parallel trajectory to the whales’ horizontal path, at a minimum of 200 m distance not to interfere with their behavior. Tag data were complemented with observations data collected from the tracking boat. These data were collected visually by a dedicated observer and, if possible, using a video camera mounted on an aerial drone operated by a pilot. Indeed, when weather (calm wind, no rain) and technical conditions (>5 km from the airport, battery remaining sufficient) were favorable, a DJI Phantom 4 unmanned aerial vehicle was deployed at 30 m height above the focal group. The observer collected a panel of parameters every 2 min if the focal whales were resighted at the 2 min point, or more than 2 min as soon as the tagged whale was resighted at the sea surface. Simultaneously, a data recorder equipped with a Panasonic Toughbook CF-19 laptop logged all data transmitted by the visual and drone observers using the Logger 2010 software developed by the International Fund for Animal Welfare (IFAW) to promote benign and non-invasive research. First, these visual data included distance to the tagged whale, visually estimated or measured with a laser range finder (Bushnell 7 × 26 mm Elite 1 mile Arc) and bearing of the tagged whale position relative to the axis of the tracking boat measured with an angle board. The tracking boat was continuously moving at a minimum speed of 2 kn to get the true heading of the boat. Mother and calf were considered as a single unit.⁶³ All together, these data were used to estimate the whale positions when sighted at the surface. Second, visual observations from the boat and the drone included simultaneous surfacing (tagged mother and escort surfacing at same time, categories: yes or no); the spacing between the tagged mother and the escort and between the tagged mother and the calf, in terms of adult body length (BL), with the following categories: very tight (<1 BL), tight (1–3 BL), loose (3–15 BL) and very loose (>15 BL); the line-up configuration of the tagged whale and the escort (next to each other rostrums aligned, next to each other rostrums not aligned or ‘echelon’, following each other, no line-up); occurrence of logging (i.e., whales horizontally resting at the surface; yes/no) and milling (i.e., wandering group behavior without consistent and coordinated directional movement across members; yes/no); occurrence of surface behavioral event (namely: breach, tail or caudal slap, spy-hopping) displayed by one of the individual of the focal group within the 2min following the last stored resighting data point; number of other whales within 200 m of the focal whale; and distance to the nearest other whale or group of whales estimated within a 1 km observation radius which correspond to the maximum reliable visual detection range (categories: <50 m, 50–100 m, 100–200 m, 200–500 m, 500–1000 m, or >1000 m if none in sight).⁶³ Drone footages were a posteriori visualized and annotated using Boris event-logging software.⁵⁴ The same previously described behavioral variables were collected on drone footage data as a backup to verify the dataset collected visually, and to supplement the observation dataset with two additional annotating events: (i) occurrence of visible and clear horizontal turns of the mother-calf pair and the escort, and if so, who initiated this change in heading; and (ii) relative position of the calf, i.e., either between the escort and the mother (MCE), or on the external mother’s side opposite to escort (CME).

Quantification and statistical analysis

Data from the tag sensors were downloaded as CSV files and imported using dedicated scripts.⁵² Tag video data were not used in the present study. Depth, heading and sound recordings from the tag as well as observation data were processed as further explained to assess behavioral responses.

Assessment of behavioral responses

Horizontal movement reaction score and speed

A movement reaction score was defined to assess potential horizontal attraction (positive score) or avoidance (negative score) responses of the tagged animals occurring during playback (DUR period). The method to calculate this reaction score was based on the measure of the distance between the tagged whale position and the sound source at the beginning and at the end of playbacks as detailed in Curé et al.⁶⁴ In short, the score corresponded to the difference between the actual distance between the tagged whale and the playback source observed at the end of the playback and the calculated theoretical distance from the projected position of the whale, i.e., if the focal whale had kept its PRE direction and speed of horizontal movement during the playback (projected course over the playback based on the sightings of the 15 min period preceding the start of playback). Horizontal speed was calculated as the ratio between the distance and time between two successive observation data points of the whale position.¹⁴ Therefore, each position data point was assigned a speed value based on the position of the previous sighting. We then calculated for each 15min playback period, the within-individual mean and max horizontal speeds.

To quantify the consistency of whale movement during the PRE period, we calculated two coefficients of variation: one for swimming speed and one for heading direction. Regarding the speed variability, we used the calculated swimming speed between consecutive GPS positions and then computed the coefficient of variation as the standard deviation divided by the mean speed, expressed as a percentage. This metric indicates how much the whale’s speed varied relative to its average speed. Regarding the heading variability, we calculated the direction of travel (heading angle) for each segment between GPS positions. Because heading is a circular variable (0° = 360°), we used circular statistics to compute the coefficient of variation. This metric reflects the directional consistency of the whale’s trajectory, with lower values indicating straighter paths and higher values indicating more variable directions. Both coefficients of variation are expressed as percentages, allowing comparison across individuals regardless of their absolute speeds or locations, and are presented in Table S3.

Relative index of motor activity

Overall dynamic body acceleration (ODBA) is a proxy for the animal’s motor effort.^65,66 It relies on cyclical angular displacements associated with swimming and is obtained by summing the magnitude of high-pass-filtered acceleration data and the amplitude of the movement. ODBA values were divided by each whale’s median value and then multiplied by the overall median value across whales to normalize deployment effects on ODBA as in Isojunno et al.⁶⁷ Despite normalization, ODBA may be overestimated during surfacing events, notably due to tag movements induced by waves.

Diving behavior

Pressure sampled on the tags was converted to diving depth using calibrated values and decimated to a 10 Hz sampling rate. Given the bathymetry of the study site, dives were limited to 60 m depth. We considered a whale to be diving when going deeper than 10 m,^15,68,69,70 to be at the surface when above 1 m depth and to be near surface when between 1 and 10 m depth. Inspecting the dive profiles, whales performed only 33 dives across the totality of the experimental periods. Therefore, we focus our analysis on the most common surface and near-surface behaviors. Specifically, for all tagged whales, we calculated the total duration spent at the surface and in near-surface movements, over each of the three 15-min PRE, DUR and POST periods. Near-surface movements are defined as active movements (ODBA >0.02 m s^-2) occurring between 1 and 10 m depth. These movements were differentiated from near-surface resting based on both dive profile and ODBA. Indeed, resting events are characterized by ODBA close to zero and either logging mode or a specific depth signature readily identifiable as shown in Ratsimbazafindranahaka et al.⁵⁷ Thus, ascent and descent phases of the few dives (excursions >10 m depth) as well as phases in which the whales exhibited resting behavior (ODBA <0.02 m s^-2) were excluded from near-surface movements.

Vocal production

Humpback whale songs and social sounds were identified in acoustic recordings from tags using Raven software through visual and audio inspection of spectrograms (Hann window; 50% overlap; FFT length: 4096). Songs were nearly ubiquitous in the acoustic environment. Their distinctive frequency modulation and frequency bands, which emerged partly above the flow noise spectrum, facilitated reliable identification. Social sounds, ranging from less than 50 Hz to more than 1 kHz, were classified into three distinct categories: high- (>700 Hz), mid- (170–700 Hz), and low-frequency (<170 Hz) calls.³⁹ Audio recordings from all PRE, DUR, and POST periods were manually annotated for each playback, with a focus on social sound production. However, prominent flow noise in the recordings overlapped with the frequency range of most social sounds, limiting detection. Consequently, only social sounds with high signal-to-noise ratios that were both aurally detectable and clearly visible on spectrograms could be confidently identified and attributed to the tagged whale group (Figure S1). Given social sounds are low-amplitude vocalizations³⁹ and no other whale groups were observed within 100 m to the tagged group when a call was detected (see Figures S7–S11), we are confident that all detected social calls were produced by the focal group (mother, calf, or escort when present). Due to the variable nature of flow noise, which fluctuates with swimming speed, we could not reliably quantify social call rates. Instead, we provide a qualitative assessment of call presence/absence and classification of opportunistically detected vocalizations.

Surface behavior

We scored the presence (yes/no) of milling or logging over each of the three PRE, DUR, or POST periods when at least one milling or logging event, respectively, was recorded within the considered period (Table 1). Simultaneous surfacing of the mother and escort was assessed within each period as a binary response (yes, no), and considered present when a majority of visual resightings over the given period was associated with occurence of simultaneous surfacing. Regarding the following categorical response variables: individual spacing between mother and calf (<1 BL, 1–3 BL, 3–15 BL and >15 BL), individual spacing between mother and escort (same categories), and line-up configuration of the mother and escort (next to each other rostrums aligned, next to each other rostrums not aligned or ‘echelon’, following each other, no line-up), we assigned the predominant category present over each period. Regarding the calf position relative to the mother and the escort, the typical and most common position is the calf at the opposite side of the escort with the mother in between (CME); thus we investigated whether there was at least one occurrence of the calf positioned between the mother and the escort (MCE) during each of the three periods (yes/no). To assess any potential initiation of social defense behavior with possible grouping with other individuals present in the area, we investigated whether the number of other whales within 200 m from the focal whale increased and whether the distance between the focal whale group and the nearest other group of whales decreased (see Table 1). Note that if the nearest group is >200 m away, there are no other whales within 200 m, and that a distance of >1 km to the nearest group means no whales in sight as visual observations can’t be achieved at such distances. Other surface events (breach, caudal or pectoral slap, spy-hopping) and horizontal turns were also scored in terms of occurrences within each period (Table 1).

Analytical approach

Quantitative analysis and statistical approach

To investigate whether the calculated variables for tagged whales’ diving depth, diving duration, surfacing duration and total duration of movements near-surface movements (Table 1) changed in response to playbacks, we compared behaviors between PRE period and DUR or POST periods within and between each playback. As the tagged whales were each exposed up to 2 sound stimuli (CTRL and KW), we used Generalized Estimating Equation (GEE) models, which allowed us to specify a blocking unit (whale ID) within which observations can be correlated.⁷¹

For each response variable, we applied the full GEE model with the following explanatory variables: ‘Playback’ (2 factor levels: CTRL or KW), ‘Period’ (3 factor levels: PRE, DUR, POST), ‘Order’ (2 factor levels: #1 or #2) and the interaction term ‘Playback:Period’ on the selected variable (see Table 1). As the movement reaction score quantifies the change from PRE to DUR, there was no covariate Period for this response variable. Statistical analyses were performed using geepack package⁵³ in R v.3.0.2 (R Development Core Team 2013). To determine the covariates that accounted for significant variance in the response variables and select the best GEE model fit, we performed ANOVA on the full GEE model followed by backward stepwise selection, i.e., after fitting each model, we conducted an ANOVA and removed the covariate with the highest p-value. We then repeated this until all retained covariates were significant at the 5% level.

Descriptive analysis

Due to limited representation in the sample, the other response variables could not be subjected to quantitative analysis and were thus described over the three different periods (PRE, DUR, POST) of the two playbacks (CTRL and KW) (see Table 1).