The effect of underwater sounds on shark behaviour

The effect of sound on the behaviour of sharks has not been investigated since the 1970s. Sound is, however, an important sensory stimulus underwater, as it can spread in all directions quickly and propagate further than any other sensory cue. We used a baited underwater camera rig to record the behavioural responses of eight species of sharks (seven reef and coastal shark species and the white shark, Carcharodon carcharias) to the playback of two distinct sound stimuli in the wild: an orca call sequence and an artificially generated sound. When sounds were playing, reef and coastal sharks were less numerous in the area, were responsible for fewer interactions with the baited test rigs, and displayed less ‘inquisitive’ behaviour, compared to during silent control trials. White sharks spent less time around the baited camera rig when the artificial sound was presented, but showed no significant difference in behaviour in response to orca calls. The use of the presented acoustic stimuli alone is not an effective deterrent for C. carcharias. The behavioural response of reef sharks to sound raises concern about the effects of anthropogenic noise on these taxa.

. Schematic diagram of the stereo video camera system used for remotely monitoring shark behaviour and playback of sound. A: waterproof floating container for battery, power amplifier and MP3 player; B: surface buoy; C: anchoring chain to seabed; D: GoPro Hero3 video cameras; E: underwater speaker in protective plastic cage; F: bait bag, G: weights. Not to scale. Shark 3D model designed by KangarooOz 3D and used with permission from CGTrader.
Each species responded differently to the sounds; the amount of time that sharks spent in close proximity of the rig was affected significantly by the factor 'species' (Likelihood ratio test, χ2 = 113.53, df = 6, p < 0.01). N. acutidens and C. plumbeus were observed on screen more frequently than other species in all treatments.
It took on average 42.1 ± 6.6 min (mean ± SD) for the first shark of any species to appear on screen when the artificial sound was playing, which was significantly higher than the 26.3 ± 3.2 min observed when there was no sound playing (control treatment) (Table S2, gaussian GLMM, df = 3, p = 0.01). Again, there was a strong effect of the factor 'species' on the time of arrival (likelihood ratio test, χ 2 = 17.455, df = 6, p < 0.01): N. acutidens and C. brachyurus were generally the first species to be observed. There was, however, no difference in the arrival time of the first shark when comparing the orca and control trials (Table S2, gaussian GLMM, df = 3, p = 0.18).
The different behaviours ('pass' , 'touch rig' , 'bump rig' , 'bump bait' , 'taste bait' , 'bite bait') were translated into scores and summed, where higher scores corresponded to a greater level of physical interaction with the rig (see methods). The scores appeared significantly lower for both acoustic treatments (mean of sums across all species ± SD, 8.1 ± 3.1 for orca and 6.0 ± 1.8 for artificial sound) than for the control, which averaged 20.5 ± 4.5 (Table S2, Fig. 2D). Here too, 'species' was found to be a significant factor when placed as a fixed factor (likelihood ratio test, χ 2 = 12.572, df = 6, p < 0.05). Again, C. obscurus was the highest scoring species (mean score ± SD, 1.74 ± 1.58, although they only interacted 82 times), closely followed by N. acutidens (mean score 1.71 ± 1.47 for a total of 432 interactions).
White sharks, Mossel Bay, South Africa. Only C. carcharias were encountered in South Africa ( Table 2).
However, C. carcharias spent significantly less time in proximity of the rig when the artificial sound was playing, while the orca sound had no effect (Fig. 2C, Table S3, gaussian GLMM, df = 3, orca: p < 0.01, artificial: p = 0.15). The time spent on screen was also significantly affected by the identity factor 'ID' (Likelihood ratio test, www.nature.com/scientificreports www.nature.com/scientificreports/ χ 2 = 142.56, df = 38, p < 0.01). Overall, we identified 38 individual white sharks. The time of arrival of the first individual for each deployment was not significantly different between treatments (Table S3, gaussian GLMM, df = 3, orca: p = 0.5, artificial: p = 0.08). However, again, there was a strong effect of the IDs (likelihood ratio test,   Table S3, negative binomial GLMM, df = 1, orca: p = 0.69, artificial: p = 0.72). Here too, the IDs were found to be an important factor when considered as a fixed factor and some individuals scored significantly higher than others (Likelihood ratio test, χ 2 = 123.65, df = 38, p < 0.01).
We tested whether the total time on screen changed significantly with repeated encounters with the equipment and found that 'experience' was significant as a fixed factor interacting with 'time on screen' (likelihood ratio test, χ 2 = 13.428, df = 3, p < 0.01). Experienced sharks spent comparatively less time in the area when the artificial sound was playing with respect to the control (Table S3, gaussian GLMM, df = 3, control vs artificial: p = 0.05, control vs orca: p = 0.79). There was, however, no effect on the behavioural scores when the factor 'experience' was considered as a fixed factor (likelihood ratio test, χ 2 = 37.5, df = 3, p = 0.54).

Discussion
We investigated whether two sound stimuli altered the behaviour of sharks in the wild, using a baited downward facing midwater stereo-video system rigged with underwater speakers. An artificial sound and a recording of orca (Orcinus orca) calls were played back to seven different species of reef and coastal sharks around Exmouth, Western Australia, and to white sharks (Carcharodon carcharias) in Mossel Bay, South Africa. As a group, the reef and coastal sharks were found to behave differently in response to both sound treatments when compared to the control, with fewer sharks approaching the rig/bait and fewer interactions overall when the sounds were playing. There was also a decrease in behavioural scores ('inquisitiveness'), as the sharks exhibited lower score behaviours (e.g. 'pass' rather than 'bite') when the sounds were playing. In addition, the orca sound decreased the overall time that the reef and coastal sharks spent on screen, and the artificial sound delayed the time of appearance. Likewise, C. carcharias took longer to arrive and spent less time on screen when the artificial sound was played back. However, the orca calls did not elicit a change in behaviour in C. carcharias. These results suggest species-specific differences in sensitivity and/or reactivity to certain types of auditory stimuli under the conditions tested.
Our results support earlier findings that certain underwater sounds can alter the behaviour of some sharks and potentially deter them from entering an area and/or interacting with a potential food source 20 . The natural soundscape of a shark comprises ambient sea noise that consists of both abiotic (wind, waves, etc.) and biotic (sounds made by marine organisms: mammals, fish, invertebrates) components [60][61][62][63] . The sounds perceptible to sharks (below 1.5 kHz) would mostly include continuous and/or rhythmic sounds, such as waves and bubbles, hydrodynamic flow of fish schools and the lower frequency components of some animal calls, like fish calls. An arrhythmic and chaotic sound (such as the artificial sound used in this study), with quick variations of intensities and frequencies, would represent an atypical and unfamiliar acoustic signal. This unnatural cue may trigger either investigative or aversive behaviour in some species of sharks. Given that the particle acceleration level of the orca audio stimulus was smaller than that of the artificial sound ( Fig. 3D), the amplitude may not have been high enough to trigger a response in C. carcharias. Although we cannot eliminate the possibility that C. carcharias are insensitive to these calls, this lack of reaction may also be due to the specific signature of the tested orca calls. Orca are known to have pod-specific calling behaviour and repertoire 64 and even within-pod-specific call types 48 . Discriminatory antipredator behavioural responses to orca calls have been observed in harbour seals (Phoca vitulina), which responded strongly to calls of mammal-eating orcas but not to the calls of a local fish-eating population 53 . The white sharks of Mossel Bay in South Africa may not be reactive to the particular orca playbacks used in this study, whose signatures were recorded from a pod in South Australia, although it is currently unknown whether white sharks are sensitive to regionally-specific orca calls. The white sharks encountered in South Africa were, on average, larger (approximately 4 m in length) than the reef and coastal sharks (approximately 2-2.5 m) in Western Australia, which may be an easier target for orca.
Reef and coastal sharks spent relatively less time in the vicinity of the presented sounds when compared with white sharks and were also less likely to directly interact with the rig. We also found interspecific differences in www.nature.com/scientificreports www.nature.com/scientificreports/ behavioural responses between the reef sharks tested; the 'species' factor was a strong predictor of time spent in the area, time of arrival and the total behavioural scores. For example, we observed N. acutidens biting down on the bait and not releasing for a few seconds, while other species, such as C. plumbeus and S. lewini failed to touch the bait. Overall, C. carcharias appeared to be the most inquisitive of all the species encountered, having the most interactions with the bait and obtaining the highest behavioural scores.
Different shark species occupy different ecological niches and present a large array of life history traits: body size, habitat, mobility, diet and mode of reproduction are examples of factors that are highly variable in sharks 65 . Additionally, sharks have been shown to possess a large repertoire of complex behaviours and social displays [66][67][68] . In this study, the response to the rig and bait, as well as the playback speaker and type of sound, may be partially shaped by the combination of those life history traits. Specifically, the reaction to sound may be related to the characteristics of the soundscape niche occupied by different species of sharks. The habitat and its local ambient acoustics, and each species' mode of locomotion and diet are potential factors that could shape the soundscape niche occupied by different species. As an example, we would expect reef and coastal sharks to live in a soundscape defined by the ambient sounds of the reef, characterised by invertebrates and fish sounds 69,70 , while white sharks (C. carcharias) would experience a diverse arrays of environments, from coastal soundscapes influenced by the dynamics of moving bodies of water, to open water environments with weather-related noise to complex rocky reef soundscapes.
Our findings not only suggest that species respond differently to auditory stimuli, but also that behavioural reactions to acoustic stimuli vary between individuals. Individual white sharks showed significant differences in the time spent around the rig and the scores of behaviours exhibited, independently of their prior experience to the rig. Intraspecific differences could be influenced by a range of factors, including sex, size, life stages, group dynamics and social hierarchy.
We also measured the level of tolerance (i.e. the intensity of disturbance that an individual tolerates without responding in a defined way 71 ) that individual white sharks (C. carcharias) showed towards a stimulus after a few encounters ('experience'), by observing the change in the time they spent in the vicinity of the speaker or the change in behaviour towards the rig and bait (scores). Contrary to our expectations, we did not observe any tolerance to the sensory stimuli over time. In fact, we observed that experienced white sharks spent comparatively less time in the area around the rig with the artificial sound playing than with the control treatment. After an initial inspection, www.nature.com/scientificreports www.nature.com/scientificreports/ some individuals may have refrained from further investigation due to the lack of food reward, or to the potential that the stimulus could be aversive. This process could lead to sensitisation, where the animals learn that a repeated or ongoing stimulus has significant consequences 72 . Each of the tests were run for only one hour, and over the course of 12 days, which is not sufficient to be able to explore the potential longer-term aspects of habituation and/ or sensitisation. Nevertheless, our artificial sound was designed to have a constantly changing frequency and amplitude content that would perhaps be expected to reduce a potential tolerance and habituation effect.
Ultimately, the large variability shown in our results (Fig. 2) agrees with other studies investigating the effects of sounds and noise on marine fauna, where interspecific differences, intrapopulation variation, context of exposure and prior experience may change the responses of the animals to the stimulus 73 . For example, cooperative breeding cichlid fish species Neolamprologus pulcher exhibit sex-dependent behavioural responses to the same playback of boat noise 74 . Similarly, in a study exploring the responses of Orcinus orca to ship noise, the behavioural responses differed with the time of the year and the age of the animals 75 . Such variability might also be expected in the responses of apex predators like sharks, given that received sound or noise is only one of the many factors amongst a multimodal array of sensory cues that they must assimilate as they explore their environment 76 .
From a conservation perspective, it is concerning that some sharks changed their behaviour in response to a relatively low sound level (received levels at a distance of 2 m from source: Z-axis acceleration = 0.0245 m/s 2 , SPLrms = 150 dB re 1 μPa). Most anthropogenic sources (not only high intensity sources such as seismic air guns, pile driving and sonar, but also background noise like shipping) have much higher sound levels 2 . For example, McCauley et al. 77 explored the effect of an airgun on pink snapper (Pagrus auratus), with an airgun which had a source level at 1 m of 203.6 dB re 1 μPa (SPLrms). The fish exposure to such a stimulus caused significant damage to the hair cells of the inner ear. Here, we have shown that a relatively low intensity sound played with a small underwater speaker is able to significantly modify reef shark behaviour. The frequency sensitivity of sharks overlaps with the range of anthropogenic noise, most of the latter lying in the low frequency range (<2 kHz) 2 . Although this study did not allow us to ascertain which of the temporal or the spectral attributes of a sound were most efficient in deterring sharks, previous observations on wild sharks showed it may be abrupt changes in amplitude levels rather than the spectral attributes of a sound that may trigger the behavioural responses. While this agrees with the general understanding that the fish hearing system has adapted as a temporal analyser, where the temporal patterns (rather than spectral) are the physiologically and behaviourally important parts of a sound 78 . This is in contrast to marine mammals, for which the spectral attributes of anthropogenic noise and the context of the animal (behavioural state at the time of exposure and demographic factors) act as a predictor for a withdrawal response, rather than the amplitude alone 75,79 . Considering the lack of knowledge of hearing physiology and acoustic behaviour in sharks 80 , we propose that there is a critical need for more studies on the impact of anthropogenic noise in cartilaginous fishes.
Although acoustic deterrents have proven successful in reducing bycatch of some marine organisms, such as cetaceans 81 , pinnipeds 6,82 and bony fishes 12,13,83 , this study shows that further work is required to assess their efficacy with sharks. Species-specific and individual differences documented in behaviour, ecology, and the peripheral and central nervous systems 66,[84][85][86] suggest that one acoustic stimulus will be unlikely to deter all species of sharks equally. In fact, targeting more than one sensory modality may prove to be a better strategy 87 . Combinations of sounds, lights and bubbles, for example, which target the auditory, visual and lateral line systems of fishes, respectively, have proven successful in various applications 37,[88][89][90][91] . There is also a substantial technical challenge in developing underwater acoustic repellents. The transducer must be large enough to produce low frequency components and provide enough energy and particle motion to spread several metres in order to affect the sharks' auditory system, which, at present, is financially and technically challenging. With these constraints in mind, we are still far from the development of an individual acoustic repellent device that the public would be able to use in their daily aquatic activities or a beach-based mitigation device that can be attached to nets or longlines. However, confined areas like beaches could potentially be enclosed by repellent sounds and thus reduce the incidence of shark-human interactions and/or bycatch. In this context, the effect of such a sound on other marine organisms present in the area should be carefully considered. Nevertheless, in conjunction with underwater acoustic technological advancement, there may be a possibility of developing static, long-term management strategies for shark mitigation, especially if combined with another stimulus as a multimodal system.

Experimental rig.
To record the behaviour of sharks in the wild, we used a downward facing midwater stereo-video camera system attached to a rig (Fig. 1) previously described by Kempster et al. 92 . Rigs were deployed from an anchored boat and suspended in mid-water with a pair of buoys at the surface. Once deployed, the boat left the area and the cameras recorded continuously for a maximum of 90 minutes. The sound device was composed of an underwater speaker (Diluvio from Clark Synthesis) positioned between the cameras pointed towards the bait (Fig. 1). The speaker was powered by a 12 V battery, through an automotive audio amplifier (PBR300X4; Rockford Fosgate), linked to an MP3 player (PhilipsGoGear). The MP3 player, amplifier and battery were enclosed within a waterproof container floating at the surface of the water, attached to the surface buoys, and the audio signal delivered to the submerged speaker via insulated electrical cables.
www.nature.com/scientificreports www.nature.com/scientificreports/ Experimental design and sound treatments. The experiment involved the presentation of two different treatment sounds and one control treatment, comprising a similar but non-functional speaker and floating container. The first sound treatment was an artificially produced sound (referred to as the 'artificial sound'), composed with Adobe Audition CS5.5, the digital audio workstation software Reaper v.42 (Cockos Inc.), and the virtual instrument Granite (New Sonic Arts Inc.) as an audio unit instrument plug-in. The sound consisted of mixed tones of different intensities and frequencies, from 20 Hz to 10 kHz, but with 95% of the energy at frequencies lower than 1 kHz (Fig. 3B) to fall within the presumed shark hearing frequency range. The artificial sound contained most of its power in the lower frequencies (overlapping with the peak sensitivity of the auditory system in sharks), chaotic rhythms (i.e. abrupt changes, no temporal domain pattern) and sudden increases and decreases in intensity (Fig. 3B). With the aid of the granular texture generator (Granite, New Sonic Arts Inc.), we used a technique known as granular synthesis [93][94][95] to build the sound, which allowed us to randomly alter the temporal domain without changing the desired frequency (pitch) to restrict to the low range 96 . A 30 second extract of the artificial sound is provided as Supplemental Information (S5).
The second sound treatment consisted of a combination of orca calls (Orcinus orca) (referred to as 'orca') (Fig. 3A). Orca populations usually specialise in one or several prey species of mammals, penguins or fishes 97,98 and their acoustic behaviours are known to vary with the type of prey hunted 99 . More specifically, groups of killer whales communicate extensively while hunting bony fishes 99,100 . The calls used as playback in this study were recorded by David and Jennene Riggs (Riggs Australia, www.riggsaustralia.com) in South Australia in February 2014 within a mile of a pod of approximately 20-30 individuals. Although the limited information on the distribution, movements, and population status of Australian orca do not allow us to unequivocally class the observed pod as a shark-eating population, we have selected calls that were recorded prior to a mass predation on ocean sunfish (Mola ramsayi). The pattern of the recorded pulsed calls would thus represent calls typically used by the pod when predating on large fishes. Although orca calls can peak to about 25 kHz, we preferentially selected low frequency components (LFC, <1000 Hz) in the recordings (Fig. 3A). For both sounds, we built a recording of 15 minutes duration (of unique sound waves), which we repeated four times to make two final sound files of an hour each.
Calibration of sound device. The sound device was calibrated in a calm location in a river to avoid boat noise (Swan River, Western Australia, salinity 33 ppt, temperature 21 °C, average depth 3.5 m). The speaker was deployed at a depth of 3 m and both sounds were played back and recorded at a depth of 1.6 m with two HTI 90 U hydrophones (High Tech, Inc.), where responses were considered linear from 2 Hz-20 kHz. The recordings were made with a NI USB 6353 Data Acquisition device (National Instruments), and the system gain was estimated with a white noise calibrator. The sounds were measured at three distances (2 m, 4 m and 6 m) from the speaker, to estimate propagation loss. All parameters were calculated with a custom-made code in MATLAB (2017a, The MathWorks, Inc.) (Table S1). Particle accelerations were estimated from pressure gradient measurements using the Euler equation, as described by Mann 101 (Table S1, Fig. 3C) with two hydrophones mounted in an array. The magnitude of the particle acceleration was calculated as the sum of the squared accelerations for each axis (Table S1). Figure 3D shows the loss of the particle acceleration magnitude, with the distance from the underwater speaker. This calibration could not be considered as an exact reference for all our data, as field conditions were not consistent each day and at each location, which would have impacted the acoustic behaviour and propagation loss of the sound device. However, we considered these sound parameters to represent a general reference for stimulus magnitude.
To determine the effect of any electronic signal emanating from the speaker, the electrical signature of the sound device was characterised as described and reported by Ryan et al. 37 . The artificial sound and the orca sound produced a peak-to-peak signal of 0.115 mV and 0.118 mV, respectively. Therefore, we considered that the electrical signature of the speaker would not significantly influence the behaviour of the sharks encountered, unless they came within 30 cm of the speaker, which never occurred in the sound (active) treatments.
Field sites. The fieldwork was conducted in two distinct areas in order to target different shark species. The effects of sound stimuli on reef and coastal shark species were investigated in Exmouth, Western Australia, where three field sites were visited over a period of seven days in April 2014, and five days in April 2015: VLF Bay, Burrows Reef and North West of the Murion Islands (Fig. S1A) The effects of sound stimuli on white sharks were investigated in Mossel Bay, South Africa, where two field sites were visited over a period of 12 days in June 2014: Seal Island and Hartenbos river mouth (Fig. S1B). During deployment of the video-camera rigs, additional bait (chum) was introduced into the water to attract white sharks into these areas. Overall, 101 deployments were carried out in South Africa (44 control, 24 orca, 33 artificial sound; 58 at Hartenbos, 43 at Seal Island), at depths ranging from 10-18 m (mean = 15.0 ± 1.6 m).

Video analysis.
Only the 60 minutes of treatment time on the video footage was analysed. The white sharks in South Africa were each identified by individual markings, scars and dorsal fin profiles 102 . Since no such obvious markings could be identified in the reef and coastal sharks in Western Australia, individuals could not be discriminated. The data acquired from the video analysis consisted of species, individual shark IDs (for white sharks), time of arrival, total time the shark was present in the field of view of the cameras (on screen), number of interactions, and notable behaviours. A shark swimming by, present in the field of view, was accounted as an 'interaction' , even though it did not physically 'interact' with the rig/bait. Observed behaviours for each interaction were classified into one of six categories: (1) 'pass' (shark in the field of view but did not make contact with the rig), (2) 'touch rig' (shark touched any part of the rig), (3) 'bump rig' (shark touched the rig elsewhere than the www.nature.com/scientificreports www.nature.com/scientificreports/ bait bag or canister, with snout), (4) 'bump bait' (shark touched the bait bag or canister with snout), (5) 'tastes bait' (shark touched the bait canister with an open mouth), and (6) 'bite bait' (shark was observed to make full contact with the bait banister in the form of a bite). We scored these behaviours in a progression of interactivity from 1-6, the lowest interaction being a 'pass' (scoring a 1) and the highest a 'bite bait' (scoring a 6). Data analysis. The effect of the sound treatments was determined using mixed model analyses performed in R 103 with the packages 'lme4' 104 and 'glmmADMB' 105 . The treatment was set as a fixed factor and the date, field site, time of day (morning, midday, afternoon), trials (matching treatment and control) were random factors. For the Mossel Bay dataset, shark identity (ID) and the number of previous interactions with the rig ('experience') were added as random factors. To determine if the presence or absence of sharks on the footage was defined by treatments, we performed a binomial generalised linear mixed model (GLMM). To investigate if there were differences in the number of interactions per treatment, we started with a Poisson general linear model (GLM) and obtained non-linear residual patterns and overdispersion. We then fitted a Negative Binomial GLM and verified levels of independence. The same model was used to test the scores. In some cases, transformation of the time data was required (see Tables S2 and S3 for details) to achieve linearity of the residuals. A stepwise procedure was used to consider all possible combinations of predictors and lowest Akaike Information Criteria (AIC) was used to select the final models. To determine the difference between treatments, a multiple comparison for parametric models was performed using the R package 'multcomp' 106 . All statistical plots presented in this study were designed with R package 'ggplot2' 107 .

Data Availability
The datasets generated and analysed during the current study are available from the corresponding author on reasonable request.