The use of acoustic devices to warn marine mammals of tidal-stream energy devices

Requirements on an acoustic warning system

It is clear from this basic consideration of underwater sound in tidal-sites and existing acoustic devices that using existing off-the-shelf equipment to warn marine mammals of turbines would be an unsatisfactory and simplistic approach. However, given the discrete point source of threat (i.e. individual turbines and the rotors in particular) coupled with the investment and infrastructure associated with tidal-stream developments, then more sophisticated acoustic warning devices would not be out of place to help mitigate a collision issue, should one prove to exist. If formulating warning systems, the following paragraphs outline seven attributes that should be considered. These are not intended to be a recipe for designing an acoustic warning system but rather outline the operating requirements that such a system would need to satisfy.

Attribute 1: The signal must elicit an appropriate response: To keep animals away from a discrete point of extreme danger, knowing the precise mode and extent of exclusion resulting from an acoustic warning is not essential. On the other hand, for more complex scenarios the way animals respond and the spatial extent of those behaviours are more critical. This is particularly the case for tidal turbine developments. Subsequently to test-devices, commercial-scale turbines are unlikely to be deployed singly but instead be placed in arrays of tens or ultimately hundreds in spaced-grid or more complex configurations (Bai et al., 2009). Thus, animals' reactions around a single turbine are highly relevant particularly if those responses take them towards the path of neighbouring turbines. Acoustic warnings that elicit startle responses and rapid flight, for example, may suit a single turbine but become inappropriate for multiple turbines. Furthermore, marine mammals, particularly odontocete cetaceans are social and their movements are often coordinated among individuals (Gibson 2006). Accordingly a response such as fright and flight by one animal on the periphery of a school can be propagated to more distant individuals (a phenomenon evident and well documented in fish, Domenici and Batty 1997; Gerlotto 2006). For species forming large or fast moving schools (e.g. common dolphins), their communication and coordination has the potential to be over ranges relevant to multiple turbines simultaneously. An extreme and inappropriate response to the warning sound associated with one turbine could therefore steer more distant individuals towards another.

One of the more obvious options for a warning stimulus would be to use the sounds of a natural predator. Killer whales prey to a greater or lesser extent on all coastal marine mammal species and can be highly vocal. Given that most marine mammals appear to have an innate fear of killer whales and their vocalisations, playbacks of their calls (or a proxy) would seem like a reasonable approach. However, mimicking a natural sound may have unintended consequences. For example prey species, such as seals, may eventually learn to associate playback killer whale sounds with turbines rather than the original predators and so show inappropriate responses on encountering real whales. Also the killer whales themselves may respond to these sounds by approaching as if they were interacting with real conspecifics. Furthermore, the responses of prey to predators are often sophisticated and vary depending on context and may range from flight or extreme avoidance to concealment or no outward response at all (Deecke et al., 2002). Mimicking the sounds of a predator may therefore turn out to be inappropriate when used in the context of fixed turbine(s).

Acoustic Harassment Devices, such as seal scarers, use amplitude as the primary feature, where the sound is sufficiently loud and unpleasant that animals elect not to approach the source and therefore the resource being protected. However, with water being an excellent sound conductor, these emissions will also propagate beyond the area of concern and can introduce acoustic energy (pollution) well outside of the footprint of a tidal stream development. Depending on the circumstances this may sometimes be appropriate (see other Attributes) or simply ensonify otherwise suitable habitat or movement corridors.

Other acoustic devices ( ADDs in particular, Table 2) emit lower amplitude sounds that are entirely artificial in nature and usually not intended to emulate the sound of anything in particular. Despite this, they are known to elicit responses either through directional avoidance or more simply, by providing novel sound encouraging animals to switch to a more spatially alert status. Our understanding of precisely which components make these sounds effective is limited, but there is good evidence that they do work (e.g. Kraus et al., 1997; Carlström et al. 2002; Culik et al., 2001; Cox et al., 2003). Therefore, the use of abstract artificial sounds (i.e. pure tones, frequency sweeps etc.) may be appropriate in a renewables context so long as the other Attributes listed below are met. In addition, refinement of the signal to maximise their aversive properties through features such as 'roughness' (i.e. bandwidth and frequency modulation, Götz and Janik 2010) may strengthen the responses.

An alternative to creating artificial sound would be to tune the self-noise generated by the turbine itself, for example by influencing the vibration of the rotor tips or the gearing. This would have the added advantage of the sound scaling with the motion of the turbine. Though an attractive prospect, adjusting variables such as blade design for their sonic properties would be logistically challenging and costly especially once turbines are deployed. This would be particularly so as our thinking on the potential mammal-turbine collision issue is likely to develop as more information comes to light and refinements become available.

Finally, the complexity of finding an appropriate warning stimulus should not be underestimated. A terrestrial equivalent is the development of warning sounds of hybrid and all-electric cars for humans. Running on electricity these can be near-silent during operation and pose a collision risk for pedestrians (Simpson, 2008). Despite all we know about human hearing and acoustic perceptions (Tandy and Lawrence 1998), urban soundscapes and the ease of directly questioning people, there still remains uncertainty over what added sound(s) would be most effective in this context (as exemplified by the on-going ELVIN study, University of Warwick).

Attribute 2: Emission rates must suit approach velocities: The timing that sounds are issued is also important whether near-continuous, intermittent-regular or intermittent-random. Aquaculture or fishery related AHD and ADD emissions tend to be pulsed with either regular or sporadic duty cycles (Table 2). However, unlike nets or cages, the rate at which animals approach tidal turbines are likely to be more rapid and the duration of interactions much shorter. Thus while random signals may discourage habituation in a fisheries context they have less relevance in a renewables one where animals have the potential to rapidly approach a discrete point of danger by swimming within a mobile water mass. Similarly for a regular signal, the cues must come frequently enough to give an approaching animal sufficient spatial warning. Calculating the closing animal-turbine speeds along with the distance thresholds and the number of pulses required to elicit an appropriate response should help define appropriate inter-pulse intervals of a warning emission.

One way to optimise pulse rates and minimise unnecessary site ensonification would be to link an acoustic alarm with an active-detection sonar. Thus an acoustic warning signal would only be triggered upon the detection of a mammal-like target on a strike trajectory. However, while such devices are being developed for detecting upstream targets (e.g. MCT 2010) they cannot yet pick up all approaching animals due to the technical challenges of covering the entire water column. False detections triggering unnecessary warning sounds however are likely to have fewer implications than the alternative of the sonar being used to trigger a turbine shut-down.

Attribute 3: Emission frequencies must be audible for target species: Though it may appear obvious, any warning signal needs to be audible to the animals of concern (McKinley et al., 1988). This, however, is not as straightforward as it may seem because the hearing capabilities of marine mammals span an extremely wide and differing range of frequencies and for many species, particularly baleen whales, their precise sensitivities remain unknown (Richardson et al. 1995). Furthermore, our understanding of how hearing abilities vary among individuals in wild populations is limited even for the best known species. Coupled with hearing acuity, sound frequencies themselves propagate to different extents in water and to be effective in this context must exceed often frequency specific ambient noise.

Most previous attempts at acoustic warning have targeted a broad (but not comprehensive) range of species whether deliberately or not and have not been tuned to local conditions. However some recent developments (e.g. Götz and Janik 2010) have considered the choice of precise frequencies with a particular recipient species in mind. To be effective in a renewable energy context, the choice of the frequency(ies) chosen for a warning stimulus require consideration of the species targeted, any collateral species (including hearing-generalist or hearing-specialist fish) and the ambient conditions through which the sounds are intended to propagate.

Attribute 4: Amplitude must be appropriate for detection range and site: The warning signal must be sufficiently loud to be audible to the intended recipient species at a long enough range that they can take either avoiding or evasive manoeuvres. In addition to hearing sensitivity, development of the correct intensity also requires knowledge of the background noise in the site as well as propagation characteristics.

A pragmatic (and common) approach to sound intensity is to simply err on the generous side and produce overly loud stimuli. This however ensonifies more of the environment than is necessary and elevates the risk of animals (motivated to stay in the area for foraging or breeding) habituating to the stimulus. Two potential refinements that could improve on this simple approach would be to: firstly, make the signal itself directional so that inappropriate areas (particularly off to the sides) are not needlessly ensonified. And secondly, link the signal to a monitoring hydrophone so that the signal strength can be varied to keep it at an appropriate level above ambient noise. Should background noise be particularly high for some reason (storm conditions, passing ship, maintenance vessel operating on site etc.) then the signal can be emphasised and conversely in quiet circumstances it can be reduced.

Attribute 5: Signal must be directionally resolvable: Whether or not an animal can determine the direction of a stimulus impacts how it responds (Blaxter and Hoss 1981). This may or may not be important depending on the type of warning sound and desired outcome used. However, this feature should be considered particularly because of the critical need for animals to make directionally relevant responses to turbines or their arrays.

Figure 13. Fictitious diagram to illustrate how the acoustic signature of a turbine will originate from multiple sources and locations around a device. For example, the machinery's gearing and power conversion equipment (red), the rotors (yellow) and general flow noise around the device and substructure (green). Thus at close-range there will not be a single acoustic stimulus from a device and furthermore, the sound emanating may not clearly indicate the parts posing the greatest risk for an approaching animal.

Attribute 6: Warning should be co-ordinated with threat: Introducing artificial sound to the marine environment comes at the risk of introducing additional noise pollution. If the mammal-turbine collision issue turns out to be real, then collision risk is likely to scale to tidal flow because 1) water flow increases the closing speed for any animals approaching from upstream 2) manoeuvring options are constrained by the directional flow and 3) because the flow drives rotor motion at velocities greater than the water speed itself. Should warning sounds be used, then scaling them to the flow rate (as well as ambient noise, see above) may be appropriate particularly so that unnecessary or habituating sounds are not produced at inappropriate times such as slack water.

Attribute 7: Location of sound source must be suitable: The sphere of acoustic warning, by whatever means, can operate over a wide range of scales. At the closest range, a warning may simply prompt attention and provide more precise spatial information to an animal so that it could evade (i.e. dodge) a particular part of a single device as it passes. The flux of water and the various operations of a turbine will produce an assortment of acoustic outputs (Figure 13) with high spatial resolution. Acoustically sensitive animals with directional awareness have the potential to respond to this, though it is currently unclear whether at close-range they will perceive the entire structure of the turbine or simply parts of it. Furthermore it is unclear whether they will appropriately prioritise these stimuli to avoid the parts that pose the greatest risk (e.g. rotor tips more than the nacelle). Given the size of currently commercial scale turbines (12-20 m diameter) relative to marine mammals (0.3-4.4 m diameter) it may be possible to promote appropriate evasion by highlighting close range cues and accentuate the more dangerous parts, whether through local acoustic warning, visual stimuli (colour or lights) or other means (e.g. echolocation reflectors).

Figure 14. Spatial scenarios for turbine array acoustic warning (plan view). Top left panel: Acoustic footprint only extends to immediate vicinity of each turbine. Acoustically sensitive species navigate in response to individual turbines. Array entry and interaction with multiple turbines is likely. Top right: Acoustic footprints abut or overlap so approaching animals can perceive multiple turbines at once and have the opportunity to skirt an array without entering it. Fewer turbines are likely encountered. Bottom left: Warning sound created from inside the array and independent of individual turbines. Response threshold extends beyond turbines. Animals can skirt array without nearing turbines but also excluded from additional habitat. Bottom Right: Warning sounds created independent of turbines at array perimeter. Animals can skirt array without approaching the turbines. However, if that perimeter is breached (pale blue) then the acoustic warning may encourage animals to stay within array.

At a larger scale - avoiding entire turbines - there is a wide-range of acoustic warning possibilities (Figure 14). These could range from a simple scale-up of the scenario discussed above to encourage avoidance of an entire turbine rather than specific parts of it. If this acoustic footprint only extends to immediate vicinity of each machine (Figure 14, top left) then acoustically sensitive species are likely to navigate in response to approach of individual turbines themselves.

With this scale of information, opportunities for animals to perceive an array are limited - akin to a person coming across a tree trunk in thick fog, there is no opportunity to know that that tree is part of a forest. For a marine mammal, therefore, array entry and interaction with multiple turbines is likely. However, if the acoustic footprints around turbines are larger and extend to either directly abut or overlap one another (Figure 14, top right), then there is potential for multiple devices to be perceived simultaneously and thus animals can both avoid individual devices and skirt around an array. Depending on the array shape, extent and the animal's approach angle, this will ultimately lead to fewer animal-device interactions.

An alternative to equipping each turbine would be to locate a single, longer range warning sound centrally within a development site, independent of the turbines so that its reach covers the area of concern (Figure 14, bottom left). Again, if responding to a threshold value of noise intensity, animals have the opportunity to skirt an array and directly interact with fewer (or no) actual turbines. One disadvantage of this over the turbine-based noise sources is that animals are less likely to ultimately associate the warning noise with turbines and also there is greater danger of areas of suitable habitat (not associated with an array) being ensonified. This is particularly important for sites in or spanning potential movement corridors such as straits or fjord/bay mouths.

Finally, given the likely literalised nature of tidal-stream sites and the parallels with fish movements in rivers, there is the possibility to construct something akin to an acoustic barrier outside of an array (Figure 14, bottom right). Again this would have the advantages of reducing the number of turbines responding animals would need to interact with to potentially zero but ensonify much less water than the whole-site scenario (bottom left). Externalising the noise source from an array, does however run the risk of responding animals being trapped within the site. However (and depending on array shape and animal behaviour), only the side facing upstream could be turned on to reduce this effect.