Beaked and Killer Whales Show How Collective Prey Behaviour Foils Acoustic Predators

Animals aggregate to obtain a range of fitness benefits, but a common cost of aggregation is increased detection by predators. Here we show that, in contrast to visual and chemical signallers, aggregated acoustic signallers need not face higher predator encounter rate. This is the case for prey groups that synchronize vocal behaviour but have negligible signal time-overlap in their vocalizations. Beaked whales tagged with sound and movement loggers exemplify this scenario: they precisely synchronize group vocal and diving activity but produce non-overlapping short acoustic cues. They combine this with acoustic hiding when within reach of eavesdropping predators to effectively annul the cost of aggregation for predation risk from their main predator, the killer whale. We generalize this finding in a mathematical model that predicts the key parameters that social vocal prey, which are widespread across taxa and ecosystems, can use to mitigate detection by eavesdropping predators.


INTRODUCTION
Vital functions such as courtship and foraging are mediated by acoustic signals in taxa as diverse as humans and insects 1 . However, sound-signallers must trade off the benefits of detection by intended receivers against the costs of detection by eavesdropping predators. Strategies for reconciling these conflicting selection pressures remain largely unexplored for sound signals in stark contrast to the intensive study of visual ecology 2 . A common strategy of many prey is to aggregate to reduce risk of predation via dilution or confusion effects 3,4 . These benefits are partially offset by the cost of larger aggregations being more detectable to predators from a distance 3,5 , but the maximum detection distance typically rises sub-linearly with group size. In chemicallyor visually-mediated systems the relation between group size and maximum detection range scales with a power between 0.5 and 1 6,7 , but a general relationship for the scaling factor for acoustic cues has not been established. This is surprising given that collective acoustic signalling is widespread in nature and chorusing has been observed in many invertebrates, fish, amphibians, birds and mammals, in both terrestrial and aquatic environments 1,8 .
The intuitive expectation that a larger number of vocal prey will unavoidably enlarge the acoustic detection range of a group may not always be true. In the case of chemical cues, increasing group size enlarges detection distance because the higher concentration of chemicals means that detection thresholds will be met at larger convective distances 6 . Similarly, enlarged visual cues arising from prey aggregation increase maximum detection ranges 8,9 . In contrast, the acoustic source level of aggregated vocal animals only increases if their sound cues overlap in time, similarly to intermittent and short duty cycle (proportion of time that the signal is on) visual cues, such as the flashes of non-synchronized fireflies 10 . Aggregated vocal individuals that are vulnerable to predation should adopt strategies that maximise their cumulative effect on legitimate receivers 11 but minimise reception by eavesdropping predators. Defining these strategies and how they depend on the characteristics of the habitat and the functions of vocal signals is essential to understand sound-mediated prey predator interactions that are ubiquitous in nature.
Toothed whales provide an ideal case-study to investigate acoustic predator-prey interactions given their reliance on active acoustic detection (echolocation) and passive listening to hunt and sample their environment 12 . Predation pressure from acousticguided killer whales (Orcinus orca) 13 has been proposed as an evolutionary driver for the vocal behaviour of the multiple small toothed whale species that produce cryptic high frequency calls, out of the main spectral band of sensitivity of killer whales: Phocoenidae, Kogiidae, and species of genus Cephalorhynchus 14 . In contrast, larger species forming tight social groups such as female-young sperm whales (Physeter microcephalus) 15 and pilot whales (Globicephala spp) 16 seem to rely on social defences to abate killer whale predation risk 17,18 . This strategy is not practical for medium-sized beaked whales (Ziphiidae) 13 which form small social groups and suffer killer whale predation in a wide latitudinal range 13,19 . This source of mortality can be critical for slow-reproducing beaked whales and thus may constitute a strong evolutive force on the behaviour of these deep-diving species.
As in myriad other social animals, aggregation dilutes individual predation risk to beaked whales. Killer whales, the main predator of beaked whales, seem to require the combined efforts of several individuals to subdue a single whale prey 13,19 , providing opportunities for other beaked whales in the group to escape. But the net benefit of aggregation would reduce if aggregated beaked whales are more detectable by killer whales. Here we use novel biologging data from beaked whales to study how their social behaviour affects encounter probability with killer whales. Beaked whales feed using echolocation signals 20 that can be heard by killer whales. They forage alone or in groups and only vocalise when deeper than 200-500m in deep dives 21 . At these depths they are safe from predation because the short dives of killer whales are insufficient to subdue a beaked whale at depth. However, beaked whales are vulnerable to attack when they surface to breathe if killer whales can locate and track them through a dive.
Here we show that a finely-tuned combination of collective behaviours and acoustic hiding by beaked whales reduces by >90% their encounter probability with killer whales, regardless of beaked whale group size. In comparison, continuous and uncoordinated group vocalization would lead to near-certain post-detection interception of beaked whales by killer whales. We generalise these results to model the general principles of abatement of acoustically mediated predation risk by any vocal prey (Box 1), showing that vocal animals can benefit from aggregation while avoiding the penalty of increased acoustic detectability in larger groups.

The killer whale-beaked whale acoustic predator-prey system
In predator-prey systems, the temporal and spatial availability of prey cues are key factors influencing detection rate of prey by predators. Here, vocal and diving behaviour data from 27 Cuvier´s and Blainville´s beaked whales obtained with suctioncup attached sound and movement recording tags (DTAGs 22 ) (SI) are used to investigate how group size influences beaked whale cue rate and spatial footprint and thus detection probability by killer whales.
Beaked whales are coined extreme divers because they perform stereotyped diving cycles day and night comprising a deep and long foraging dive with maximum duration and length of 2 hrs and 3 km (Cuvier´s beaked whale), followed by a series of shorter and shallower recovery dives separated by brief (mode~2.5min) surface intervals to breath 23-25 . Individual beaked whales are vocal on average 18%-20% of their time, for echolocation and occasional social signalling during deep foraging dives 21,26 . Beaked whales are typically found at the surface in tight groups although these groups lack long-term stability. We tagged pairs of whales in the same social group in three instances finding remarkable activity synchronization within these three whale pairs ( Figure 1 and SI Table 1). While animals were within a group, the most coordinated deep dives (defined as the two deep dives with closest start time performed by the two whales in each whale pair) overlapped on average for 99% of dive duration (SD 0.3%).
The vocalisation phase of such dives overlapped in time by 98% (SD 4%). The most coordinated shallow dives overlapped by a mean of 97% (SD 2.4%). A randomization test showed that in 100% of 4000 iterations the observed dive-profiles rendered a higher overlap of dives than simulated data obtained by random permutation of the dive cycles of one of the whales of the pair (SI). Real overlap exceeded random overlap by an average of 44% (SD 24%) of the time in both deep and shallow dives, and by 63% (SD 31%) of the vocal phase time (SI). Similar group vocal coordination was observed in an additional dataset of 54 deep vocal dives from 12 whales tagged separately in different groups. The mean duration of the vocal phase in these dives was 25 minutes. The time-delay of start/end of clicking between the tagged whale and any conspecific whale within acoustic range of the tag differed by just 1.8 min (SD 1.5, start of clicking) and 0.9 min (SD 1, end of clicking) (Supp. Table 2). These results for single tagged whales in groups from 2 to 6 whales are consistent with the observed 98% overlap in the vocal phase of dives performed by paired tagged whales (SI Table 1).
Adding the mean observed offset in clicking timing of group members to the mean duration of the vocal phase of tagged whales results in a mean of 27.7 min of group vocal activity per dive. Thus, considering the mean dive cycle duration of 120-140 min 23,26 , groups of whales are acoustically available for detection some 20-22% of their time. This is only slightly longer than the 18-20% of time that individual whales within a group are available for acoustic detection 21,26 , meaning that the proportion of time that beaked whales are available for passive acoustic detection by killer whales is almost independent of group size. In comparison, a randomization test simulating a signalling channel with activity slots that can be accessed by one or more whales at random predicts an approximately Gaussian distribution for the time that 6 asynchronous beaked whales would be available for acoustic detection. The mean of this distribution is 69%, i.e. more than three times longer than the observed 22% of the time that a group of six beaked whales is vocally active, showing how vocal coordination can reduce the time that animals are available for predator detection. groups synchronize the vocal phase of their dives. This is because apart from rare short whistles 21 , beaked whales only produce short (~200 µs) echolocation clicks with a mean duty cycle of 0.0007 21 . Moreover, the volume of water ensonified by the highly directional clicks of beaked whales 27-28 increases negligibly in groups. This is because beaked whales diving in tight coordination show a similar circular distribution of the pointing angle of their clicks within a dive, (i.e. they ensonify a similar restricted sector of the circle) (SI).
Inter-animal separation also influences cue spatial availability. Groups cannot be considered an acoustic point source when they disperse. We calculated the separation between pairs of beaked whales tagged simultaneously in the same group using an acoustic travel-time method (SI). Whales were as close as 11 m when they began echolocating at a mean depth of 450 m. They then separated by up to 1500 m while hunting but re-joined at the end of the vocal phase to as close as 28 m before initiating the silent ascent from a mean depth of 760 m ( Figure 1). Taken together, the whale pairs spent 95% of the vocal phase less than 500 m apart. Considering an individual onaxis maximum detection range of 6.5km 29,30 , and the typical 90º coverage of clicks within a dive, the separation of 0.5 km between beaked whales in a group means an increase in the detection area for surface-dwelling killer whales of 16% of a group compared to a single beaked whale.
In sum, the collective diving and vocal behaviour of beaked whales reduces cue time availability by 40% and increases detection footprint by just 16% while still allowing animals to disperse to hunt. This increase in spatial detectability given by group dispersal occurs when beaked whales are at depths that provide them a refuge from shallow diving killer whales. However, diving beaked whales are susceptible to acoustic stalking in which killer whales track them acoustically and then attack when 188 they leave their deep-water refuge during obligate surfacing for air. Here, the collective behaviour of beaked whales is key to foil stalking predators. By coordinating their dives, groups of diving beaked whales are released from a "surface anchor" that would be maintained by the need to re-join with non-diving group members and thus frees groups to choose where to surface from dives. Most deep-diving whales ascend steeply to minimize transport time and hence maximize foraging time at depth 31,32 , however, this behaviour leads to a high encounter probability with killer whales stalking acoustically from the surface. In contrast, both Cuvier´s and Blainville´s beaked whales manoeuvre in a way that confounds surface predators when they ascend to breathe.
These whales silence at an average depth of 760 m and ascend towards the surface with an unpredictable heading and a shallow average pitch angle of 35º with respect to the horizontal 23,33 . This unusual behaviour for an air-breathing mammal creates an uncertainty cone for the position of beaked whales while they ascend in silence. The resulting potential surfacing area is a circle of 3.7 km 2 (~1.1 km radius) centred on the position of the last click emitted by diving beaked whales (Fig 2 and SI). : Post-detection encounter probability is <10% for killer whales acoustically stalking beaked whales due to the uncertainty in their surfacing location following long silent ascents. The coloured lines in the dive profiles of two beaked whales diving in coordination represent the vocal phase of these dives. The histogram is the depth distribution of the clicks of beaked whales (truncated to 900 m), showing that they are silent at the depths to which killer whales usually dive (marked as a dotted line at 200 m depth).
A pod of killer whales that has tracked acoustically deep diving beaked whales could potentially dive to hunt the beaked whales at depth. However, this does not seem feasible given the protracted and intense pack hunting effort required for killer whales to subdue cetaceans at the surface 13,19 , and the restricted 10 min duration of killer whale dives 34 . Thus, killer whales need to wait for beaked whales to be at or near the surface to hunt them. Killer whales are unlikely to use echolocation to track beaked whales to avoid alerting them and elicit avoidance responses 35,36 . This means that killer whales must search visually the uncertainty surfacing area of beaked whales in the short time that beaked whales spend at the surface after a vocal dive, before they dive again. Both Encounter probability increases with killer whale pack size: usual pack size of mammal eating killer whales is 3-4 whales, but up to 12 whales have been observed 19 . Killer whales in large packs and perfectly coordinated to not overlap in search area could cover some 7% of the potential surfacing area of beaked whales.
Thus, the coordinated movement and acoustic hiding behaviour of Cuvier´s and Blainville´s beaked whales results in a maximum probability of interception by stalking predators of 7% irrespective of group size, i.e., a reduction of >90% when compared to the high interception probability for animals that ascend vertically and/or vocalise during the ascent. The unpredictable ascent of beaked whales is only possible due to their coordinated diving behaviour.

DISCUSSION
Beaked whales exemplify a widespread strategy of vocal animals: to broadcast when predators are not detected or when in a safe place with limited predator access (e.g. in the case of beaked whales, deep waters are safe from killer whale attacks), and silence (i.e. hide acoustically) when compelled to leave the refuge or when predators are detected. These behaviours are observed in avian nestlings, as well as in chorusing insects and frogs, that silence in response to alarm calls or predator approaches 39,40 .
Another important commonality among beaked whales and other vocal species is that The results of this paper show that the detectability of beaked whales for their main natural predator, the killer whale, is very similar for individuals and groups. Tagged beaked whales emitted on average 41% (~1500 clicks) of the clicks produced in a dive while the whales were oriented towards the sea surface, at an average rate of 68 (SD 22) upward clicks per min of the vocal phase. This means that killer whales crossing the acoustic footprint of beaked whales at slow speeds of less than 2 m/s 38 have a high probability of detecting a single vocalising beaked whale when passing by the ensonified area, and thus additional clicks from several vocal whales with collective vocal behaviour may be redundant for group location. In contrast, vocal group size will likely influence beaked whale detection probability from non-natural receivers passing at faster speeds, such as ships with hydrophone systems. Natural predators such as killer have limited capacity to swim faster for protracted times to increase their search area, but they would improve their encounter rate of beaked whales by increasing group size and spreading out while performing area restricted search of detected beaked whales. In fact, killer whale groups attacking beaked whales are larger than groups attacking other marine mammals 19 , indicating that cooperative searching is one way that killer whales can combat the abatement tactics of beaked whales.
In addition to predator defence, coordinated diving may provide additional benefits to beaked whales. An advantage could be sharing information 41 via eavesdropping on the foraging activity of group members as has been observed in echolocating bats 42

Acknowledgements.
Thanks to all researchers and students collaborating in data collection. This work was Groups of beaked whales were defined as clusters of whales observed together at the surface. No inferences were made about short or long-term group stability. Whales in these clusters were most often observed to surface together for the duration of the visual follow. In three occasions (one per field site) we tagged two whales in the same social group. Tag deployments on the two members of each of these three whale-pairs overlapped in time during 3, 9 and 12 hours, respectively; the 6 whales forming these whale-pairs performed in total 22 deep and 64 shallow dives (SI Table 1).
Dive coordination of the whales in whale-pairs was assessed by comparing timing and depth of the most coordinated dives performed by the two members of each whale-pair.
These coordinated dive-pairs were defined as the dives with closest start time performed by the two whales of each whale-pair. The analysis was performed separately for deep vocal dives (deeper than 500 m maximum depth) and shallower non-echolocating dives 17 . For the resulting dive-pairs we calculated the time overlap of the dives, as well as the overlap in the vocal phase of vocal (deep) dives. Differences in duration and maximum depth between the dives in each dive-pair were recorded also.
Results were pooled for each whale-pair (SI Table 1) and then for the three whale-pairs given the close similarity in results between study areas and species and the small sample size of Blainville´s beaked whales (all but one dive-pairs were recorded from Cuvier´s beaked whales).
The group of Cuvier´s beaked whales tagged in the Azores was followed by the research boat and observed at a distance during surfacing intervals to monitor group composition via individual photo-identification. Analysis of photographic data showed that the four animals forming the group at the time of tagging continued to surface in close vicinity until some 9.5 hrs after tag deployment. After this, two of the four whales, including one of the tagged whales, were no longer observed in the group. The   Blainville´s beaked whales adopt a fairly constant heading during ascents, covering consistently more than 50% of the maximum horizontal distance assuming a constant heading, and more than 80% of the maximum distance in 55% of the dives 17 . It is possible that beaked whales modulate the horizontal distance covered during ascents according to the distribution of foraging resources and to the presence of predators or other potential disturbing stimuli, such as ships 38 or delphinids, which have been observed to harass beaked whales (Ana Cañadas, pers.com).

General acoustic model formulae derivation
In all acoustic detectors, a requisite for detection is that the signal to noise ratio, i.e. the The SL of a group of n s vocally overlapping individuals relates to individual SL as: 10 10*log ( )