Methods for tracking fine scale underwater movements of marine mammals around marine tidal devices

3 Fine scale underwater movement

The objective is to detect and describe fine scale movements of marine mammals in the vicinity of an operating underwater turbine. We thus have to consider what is meant by 'fine scale'. The questions underlying the main objective are to investigate if and when an animal becomes aware of proximity to a turbine, what its reaction is, and whether the encounter results in a collision. There is no direct way of determining awareness. We can however detect change in movement behaviour. But there is no simple mapping between awareness and behaviour. Behaviour change may be the result of other stimuli. Similarly, awareness of the proximity of a turbine may not invoke behavioural change if the individual does not perceive risk.

Whilst acknowledging the imperfect nature of inferences that can made from movement behaviour, we must propose a target precision and range for the required data. We therefore propose that:

• the range of measurement should be within 100m of a turbine,
• temporal resolution 1s
• spatial precision 1m

This target is a yardstick to assess technological candidates. It may well be that none achieve this target, and thus the iterative process of matching technology to biological questions may lower the target.

It is likely that even detailed tracks may fail to distinguish between an impact and a near miss. Impact detection methods are thus discussed in Section 5.

3.1 Candidate Technologies

There is a diverse set of telemetry (McConnell et al. 2010) and detection (Hastie 2012) techniques that are available to study marine mammal movements and behaviour, which the authors have considered extensively. However the data requirements of our objective to record fine scale interaction places a heavy demand on existing technology. However, we identify three generic methodologies that show potential:

1. Animal-borne telemetry devices,

2. Passive sonar arrays to track animals that vocalise or that carry acoustic pingers, and

3. Active sonar systems.

Each is discussed in turn below. In Table 1 we provide an overview of their attributes for comparison. An explanation of these attributes follows:

• Identify to species. Can the technique discriminate to species level?
• Identify to individual. Can the technique reliably identify the individual involved?
• Precision. What is the expected spatial and temporal precision that can be expected? Note that a system ( e.g. dead reckoning) can have good precision relative to previous locations, but poor absolute precision in our ability to place the track in real space ( e.g. latitude and, longitude coordinates).
• Plausible sample size. This describes the likely power of the observation technique - in other words the number of 'close interactions' we might expect to document. This is primarily a function of local animal density and duration of study. For techniques that involve the capture and tagging of individuals it is, more precisely, a function of the local density of tagged individuals. In turn, this is a function of both the numbers and movements of locally tagged animals.
• Data Latency. This is the delay in getting the data processed to a level where detections can be determined. Zero data latency would be real time data relay and processing.
• Range. This is the distance from the device within which an animal can be detected and tracked.
• Taxa suitability. The applicability of each method to different taxa of marine mammals.

The values in Table 1 are for illustrative purposes. The real values depend upon the location of the study and the density of marine mammals found there. For the most, part they can only be refined with pilot/experimental studies. As we mention in the Introduction, the aim is to focus discussion on the parameters that are operationally significant, rather than making definitive predictions.

3.2 Animal-borne telemetry

In this section we investigate the use of active electronic tags that can be applied to captured marine mammals ¹ . This approach permits the detailed longitudinal study (up to about six months) of the tagged individuals. However there is a risk when there is a single geographical focus interest such as a turbine array. Marine mammals are wide ranging and thus those tagged close to such a focal point may subsequently emigrate. Thus we need to know a priori the target species' movement patterns, and thus the required sample size to achieve the statistical power to capture interactions at a focal point.

Note that the catching and tagging of seals in the UK, whilst requiring a high level of expertise and permitting ² , are now reasonably routine procedures. This expertise does not currently exist in the UK for small cetacea, although by-caught porpoises are regularly tagged off Denmark (Geertsen et al. 2004). Thus we exclude cetacea from this technology.

It is logistically difficult to recapture specific, instrumented grey and harbour seals. Whilst an automatically-timed or remotely triggered tag release mechanism is a possibility, it is usually logistically difficult and costly to recover detached tags from animals that can roam far. This fact thus excludes tag types where detailed information is stored in memory for subsequent physical retrieval ( e.g. time-depth recorders and the Animal Diary dead-reckoning tags (Wilson et al. 2007b)). We thus limit our discussion to those tag technologies that relay data ashore. Existing seal tag designs can relay data ashore by either Argos satellites, or the GSM (mobile phone system). Whilst there are other radio systems potentially available, none combines the required small size, efficiency and rapid start up (to accommodate short surfacing periods) ³ .

3.2.1 Argos satellite tags

Historically (and currently for many marine mammal species) the Argos satellite system (Argos 2008) has provided both a means to relay data and to estimate approximate locations. Its primary advantage is its global coverage and that data (including stored behavioural information) can be sent immediately on an animal surfacing. For these reasons, it has been used to track seals (McConnell et al. 1999), large whales (Mate, Mesecar & Lagerquist 2007), and small odontocetes (Sveegaard et al. 2011). However, the location information it provides are sparse (perhaps one to six locations per day) and of low precision (errors of more than 1 km are common - see Vincent et al. (2002)). Argos-based telemetry data are therefore of insufficient quantity or quality for investigating fine scale movements around marine renewable devices.

3.2.2 GPS/ GSM tags

3.2.2.1 GPS

The use of the Global Positioning System ( GPS) is an option for increasing location accuracy. However, the brevity of surfacing intervals (effectively shortened even further by periods of wave wash over the GPS antenna) is generally shorter than the time required to calculate a fix from a cold start (time to first fix from cold). This issue was resolved by the Fastloc innovation. Fastloc (Costa et al. 2010) which obtains a snapshot (< 0.2 s) of GPS satellite transmission when the animal surfaces. This is then processed and condensed into about 32 bytes of pseudo-range data which are time-stamped and stored for subsequent transmission. Once these data are received ashore, the pseudo-range data are post-processed to provide a series of accurate GPS fixes. Fastloc data can be relayed within Argos uplinks (transmissions that are successfully received and relayed by the satellite segment), but this imposes severe restrictions on the amount of GPS fixes that can be relayed. Lonergan et al. (2009) emphasised that accurate track recreation depends not just on fix precision, but also on the number of fixes per day.

3.2.2.2 GSM Mobile phone technology

GSM mobile (cell) phone technology is one solution to the Argos data bottleneck. Since 2004 tags developed by the Sea Mammal Research Unit ( SMRU) ⁴ deployed on seals have used the mobile (cell) phone network to relay data ashore. In these GPS/ GSM tags, data (including stored Fastloc and depth data) are collected routinely over periods of six months or more. Every time a seal swims within suitable GSM network coverage the stored data are sent ashore using GSM 2.5G link ⁵ . This allows high data rates to be achieved - and at low energy and financial cost. In their usual configuration the tags store data for up to two days before attempting to relay them ashore due to the energy overhead associated with establishing each GPRS session. However data latency could be potentially reduced to less than one hour if the site of potential interactions was within GSM coverage and software parameters were changed. However there is still a chance that the animal may not spend sufficient time at the surface (20 s uninterrupted) for successful GSM registration until it next hauls out. The short surfacing periods of cetaceans prohibits the use of GSM data relay.

GPS/ GSM tags also record and relay detailed depth profiles within each dive. However these profiles are time based. An attempt to geo-reference them relies upon a linear interpolation between GPS fixes at the start and end of each dive. This introduces uncertainly into the track and thus to the locations at which to dive depths occurred. Since grey and harbour seals have dive durations in the order of 3- 5 minutes, this uncertainly may extend to many tens of meters. Thus we do not recommend standard GPS/ GSM tags to determine fine scale interactions in seals or cetacea.

3.2.3 Dead reckoning

Dead reckoning ( DR) uses data from on-animal movement (acceleration, attitude and speed through water) sensors to estimate the position of a tagged animal. In contrast to the other localization methods described above, DR is not a stand-alone method. It can be used to interpolate localizations and so improve the temporal resolution of infrequent positions, such as from surface GPS fixes, but DR cannot, by itself, give the geographic position of an animal. DR estimates the displacement of an animal from a starting position by integrating the velocity vector of the animal with respect to time. This requires regular measurements of the animal's velocity, i.e., its speed and direction. Both of these parameters can be estimated from sensors in a tag rigidly attached to the animal: the forward speed through water can be measured by a paddlewheel while the direction of movement, which is assumed to coincide with the longitudinal body axis, can be estimated from tri-axial accelerometers and magnetometers.

Although the idea behind DR is straightforward, there are two practical issues that complicate the picture. The first is the large amount of information that must be retrieved from a tag to recreate a track. Speed, direction and depth need to be sampled by a tag at least once per second to recreate a reliable track of an agile animal such as a seal. This means that about 500 bytes per minute (= 7 Mbytes per day) are collected by the tag, which is about 70 times the typical data relay rates of GPS/ GSM tags in their standard configuration. Whilst we estimate that these tags could handle this elevated data relay rate, there would be a concomitant decrease in battery life.

The second issue is more serious and relates to the way that velocity is measured. The velocity over ground is required for geographic tracking but with respect to water the sensors in a candidate tag measure speed and direction with respect to the water. If the water current in the area is known, this can be added to the tag velocity measurement to estimate the velocity over ground. But any uncertainty in the total speed or direction will be integrated when computing the DR track leading to incremental positioning errors that grow with time (Shiomi et al. 2008; Shiomi et al. 2010). Errors can become substantial: a speed error of 1 m/s (2 knots) will result in a track error of up to 150 m in a 5 minute dive, assuming that GPS positions are taken at the start and end of the dive. It is important to note that not every attempt to obtain a GPS fix using Fastloc is successful and so only a proportion of dives will have start and end GPS location pairs. Thus, in areas with high currents such as those favoured for energy generation, a good estimate of the current field in the vicinity ( e.g. 500m radius) of the tagged animal is essential for DR to be accurate. This may be especially challenging if the current is highly dynamic or varies spatially. As error builds with time, frequent positioning during dives, e.g., using a passive acoustic tracking system (see the next section), would reduce errors.

3.2.4 System readiness

GPS/ GSM tags are available 'off the shelf'. To reduce data latency and for use in areas with low GSM coverage, a dedicated link from a GPS tag to a shore- (or turbine-) UHF receiver can be established Sensors to facilitate DR have been developed and implemented in a number of retrievable tag types. Examples include the D-Tag (Johnson & Tyack 2003), the Animal Diary tag (Wilson et al. 2007b) and the Little Leonardo series of data loggers (Mitani et al. 2004). However the combination of the two systems to provide inter-fix dead reckoning would take significant development time due to the need for on-board processing of raw sensor data. However, the major drawback is the requirement to accurately predict water current in time and space ⁶ - without which the combined GPS/ GSM and DR tag is not a feasible option.

3.3 Passive acoustic monitoring ( PAM)

Many marine mammal species use both passive and/or active acoustic detection as a means of sensing their environment; dolphins and porpoises in particular produce echolocation clicks for navigation and finding prey and these potentially provide a means of locating and tracking individual animals in 3D space. It is therefore possible to use passive acoustic monitoring systems mounted on, or in the vicinity of, tidal energy devices to localise the vocalisations of cetaceans swimming around the devices.

3.3.1 Vocalisation

Harbour porpoise ( Phocoena phocoena) are the most likely cetacean to be present in the same areas as tidal turbines in European waters. In addition several species of dolphins may also be present. Porpoises produce trains of characteristic narrow band ultrasonic clicks (peak frequency @140kHz) which are projected forward in a narrow beam (3dB beam width of 16˚) and have an on-beam source level of 178-205 dB re 1µ Pa p-p (Villadsgaard, Wahlberg & Tougaard 2007). The primary function of these clicks is echolocation and click rate varies with behaviour and the echolocation task being undertaken. Click rates typically vary between 5 and 35 clicks per second but can reach rates of over 1000 clicks per second (Clausen et al. 2010). In the wild porpoises vocalise frequently, with 90% of intervals between clicks are less than 20 seconds (Akamatsu et al. 2007). Several species of dolphins are also found in inshore waters and are likely to interact with tidal energy devices. They produce communication whistle vocalisations as well as echolocation clicks. Their clicks are louder and have a broader bandwidth than those of porpoises and their rate of click production may be more variable.

3.3.2 Mode of operation

Passive acoustic systems have been used extensively for detecting vocalising animals ⁷ but their use to localise and track animals is less well developed. The location of a vocalising animal can be determined from the time difference for a sound arriving at two or more hydrophones (Time Of Arrival Differences - TOAD). A single TOAD from a pair of hydrophones allows the location of the sound source to be determined along a hyperbolic surface of infinite area.

Arrays with a larger number of sensors provide a greater number of TOADs. In fact for an array with N sensors there are N*(N-1)/2 time of arrival differences, each of which will provide a hyperbolic surface; however, only (N-1) of these will be independent. The location of the acoustic source is estimated as the (theoretical) point where these surfaces cross. Hence at least two independent time delays are required to calculate a source location in two dimensions and three time delays (requiring a 4 sensor array) to provide a three dimensional location.

Simple two hydrophone towed arrays are now used routinely to carry out passive acoustic line transect surveys for marine mammals. However, because this type of array is towed, and therefore moving, a sequence of detections results in a series of surfaces all of which cross at a common location. This allows a range to the vocalising animal to be estimated, which is key information for line transects using distance based methods. However with a spatial precision of many tens of meters it is unlikely to be a practical method for investigating fine scale responses at these sites. In addition the required boats are both expensive and likely to affect the behaviour of the subjects. We thus focus on static hydrophones in the remainder of this section.

3.3.3 Accuracy

The accuracy of locating an acoustic source depends on a range of factors including the physical environment, the array design, the accuracy with which the hydrophone locations are known and the acoustic behaviour of the marine mammals being studied. Changing sound speed profiles within water masses which results in acoustic refraction leading to curved sound paths and a number of concomitant errors. However, these effects are less likely to be a problem in strong tidal current areas where waters are well mixed. Also reverberation, background noise and the directional nature of cetacean vocalisations can all result in variable signal waveforms at different hydrophones within an array, often introducing timing errors. In addition, for towed arrays and other configurations where the hydrophones are not rigidly fixed, error in the location of the hydrophones is a substantial potential source of error.

Generally, the effect of these errors is determined by their magnitude in proportion to the size of the TOADs themselves. Larger arrays will therefore, tend to provide more reliable locations than smaller arrays. As a rule of thumb, the range that good locations may be estimated is about ten times the maximum dimension of an array. However large array dimensions can bring their own problems. Some stem from the practical difficulty of deploying and maintaining a large rigid array in an extremely energetic marine environment whilst others relate to the nature of the signals themselves. For example, cetacean echolocation clicks are highly directional, so as two hydrophones are moved further apart the waveforms received on each hydrophone will become increasingly dissimilar resulting in increasing timing errors. With large separations and low received levels it is likely that some clicks will be detected on only a subset of hydrophones raising concerns about the practicality of tracking cetacean movements using widely separate hydrophones. This issue was explored empirically as part of a Scottish Government funded project by deploying a large 3D array (dimensions ~20m) on a fish farm (a cost effective means of deploying a large floating array in porpoise habitat) which showed that it is possible to detect coherent clicks from porpoises within 150-200m range. We were able to localise and track animals, although the fact that the hydrophones were not rigidly fixed in this array compromised location accuracy that could be achieved. Most importantly, the exercise has served to demonstrate empirically that sufficiently coherent clicks are detected at multiple hydrophones in an array sufficiently large to allow tracking within a few hundred meters.

Based on simulations and modelling, we estimate that a rigid 3D array with dimensions of tens of metres is required to achieve sufficient accuracy (better than 1m) to examine interactions with an underwater turbine. It would be challenging and expensive to provide a structure to hold sensors in an appropriately configured array in high energy tidal areas. However, many turbine designs include substantial support structures that could provide a cost effective rigid support for an appropriately sized array (suitable examples include, ANDRITS HYDRO Hammerfest HS1000, Atlantis, Tidal Energy Delta Stream device, Open Hydro and Voith Hydro). Field data collected in pilot studies permits the performance of different array configurations to be modelled (for example see Ehrenberg &Steig (2002)).

Hydrophones need to be synchronised to determine TOADs with appropriate accuracy and this will require either all the hydrophone signals to be brought to a single digitising system, or remote digitising systems to very precisely synchronised, almost certainly through a connecting cable or fibre. Where turbines have appropriate large support structures and the incorporation of hydrophone arrays are planned from an early stage, it should be feasible to deploy systems with multiple rigidly fixed hydrophones which are hard-wired to a single digital acquisition device within the structure, with digital data being streamed ashore for detailed analysis. However, this may not always be feasible, and is clearly limited to data collection once the turbine has been installed. One possible solution to this would be to deploy clusters of hydrophones in small arrays of the order of a meter or so with waveforms and/or click detections being recorded autonomously within each cluster. TOAD analysis of these synchronised signals within clusters would provide accurate and unambiguous bearing and azimuth data for vocalising animals, and "crossing" such 3D bearings from multiple clusters deployed around a tidal turbine should provide locations and tracks for vocalising animals. While these locations might be less accurate than those that could be provided from a larger rigid array of synchronised hydrophones, they should still provide data of value for management applications. A system like this should certainly reveal avoidance and larger scale evasion at a scale of meters to tens of meters, but probably not at sub-meter accuracy.

3.3.4 Data processing and latency

Systems incorporating hydrophones that are wired to shore, perhaps including basic signal processing offshore will provide the raw data (clicks or waveforms) in real time. Processing required to detect and characterise signals should also run in real time, especially in the case of porpoise clicks and active acoustic tags which have very characteristic signals. Currently the next step of calculating 3D location for these detections and joining them into tracks requires substantial user input and a time lag of several minutes would be expected, even if an experienced operator was on duty 24-7. The main need for real time data would be as part of a mitigation procedure involving a shut-down, or some other action such as activation of an acoustic alarm, if an animal came within a zone where it was believed to be at risk.

Hydrophone clusters (discussed above) and autonomous data loggers have to be recovered periodically and the data amalgamated. Thus they do not provide real time information.

3.3.5 System readiness

Although the type of systems we describe here have never been deployed in tidal current areas, we have, in large part with Scottish Government support, been making progress in developing some of the essential software that would underpin a system like this and in addressing some of the key uncertainties. Thus, software routines which are highly relevant to 3D tracking have been developed within PAMGUARD, in the first place to analyse vertical array data, and the work deploying large arrays at fish farms mentioned earlier, addressed many concerns about the practicality of tracking porpoises with arrays of the order of 20m or so. Additional programming will be required to marshal and archive the huge amounts of data that such a system could provide and substantial software development might be anticipate to achieve a goal or real time automated localisation and tracking ⁸ .

The hardware required consists of hydrophones, signal conditioning units and digitisers. It is also likely that digital signals will be converted to optical to facilitate bringing the signals ashore. The exact configuration will in part depend on the particular opportunities and constraints of each deployment. However, generic discussions with turbine engineers indicate that there will be sufficient room within the waterproofed chambers of the structure to be able to accommodate off the shelf digitisers, signal processing and computers and network equipment. However this requires appropriate planning and early cooperation from the developer

We are working with one developer, TEL, to deploy a system with these capabilities in spring 2013, using largely off the shelf equipment. Thus, there seems no reason why such systems should not be considered within the time frames of the earliest commercial developments in Scotland.

3.3.6 Detection of acoustic pinger tags

So far, only localisation derived from the animals' own vocalisations have been considered. Grey and harbour seals do not regularly and predictably vocalise underwater, but they could also be locally tracked with a passive acoustic array if fitted with acoustic pingers. These could either be glued to the seal's fur or attached to flipper tags. Care must be taken with the choice of tag frequency in order that they are not detected by the seals (this is expanded in the active sonar section below).

Acoustic pingers are routinely used to track fish (Cooke et al. 2011) and a variety of tags and receivers systems are available. A number of companies offer bespoke solutions with the potential to track suitably tagged seals ⁹ . For example Wright et al. (2007) monitored the locations of tagged harbour seals in an estuary using a fixed array of 15 acoustic receivers. Harcourt et al. (2000) and Simpkins et al. (2001) used similar a similar technique to study the movement of polar seals in relation to breathing holes in ice. However none of these deployments had to contend with the challenges of strong tidal currents.

At any site at which there was the possibility that small cetaceans would be present, it would make most sense to establish an array of hydrophones for cetacean tracking (as discussed above), and incorporate detectors for individually coded acoustic tags into the detection and classification software. As the tag signals can be designed to be optimal for localisation and have a distinctive signature tracking them with the array should be somewhat easier than tracking small cetaceans. Calculations based on a 'typical' fish acoustic tag suggest that a static hydrophone array (of the type discussed above) running PAMGUARD ¹⁰ detection software should be able to identify a tag at a range of up to 200m. For tracking, the useful range should be about 10 times the array dimension. That is, an array with a foot print of 20m across should be able to track out to 200m. This would require tag-specific PAMGUARD detection module to be designed and the cooperation of tag manufacturers to release the details of their tag coding systems. However such an array would have the benefit of being able to track both vocal cetacea and acoustically tagged individual seals.

Acoustic pingers are an order of magnitude cheaper than GPS/ GSM tags (£100-300 compared with £3000). Thus a greater number of seals could be tagged. An extra advantage is that their small size may permit acoustic tags being attached to the flipper webbing, and thus they would not detach at the annual moult.

3.4 Active sonar

Active sonar is like underwater radar, but uses sound rather than radio waves. In recent years, there has been rapid development of active sonar systems for a wide range of uses. These include underwater imaging in low visibility and diver intrusion into secure zones.

The mode of operation is, in essence simple. Pulses of sound ('pings') are produced electronically using a sonar projector and the system then monitors for echoes of these pulses as they reflect off objects using one or more hydrophones. The speed of sound in water divided by half of the echo delay provides distance to target. To measure the bearing, several hydrophones are used to measure the relative arrival time at each, or with a receiver array of hydrophones, by measuring the relative amplitude in beams formed through a process called beam-forming.

Sonar efficiency can be affected by variations in sound speed, particularly in the vertical plane. The speed is determined by temperature, salinity and pressure. Furthermore, scattering from small objects in the sea, from the seabed, and from the surface can be a major source of interference. Together, these effects can make using active sonar to detect and track marine mammals in energetic tidal areas particularly challenging.

3.4.1 Available systems

There are a large number of commercially available active sonar systems. A recent review collated an inventory detailing over 200 systems from 39 sonar manufacturers (Hastie 2012). These are designed for a wide range of uses including mapping ( e.g. with swathe bathymetry), underwater navigation, fisheries research, and seabed profiling. Fundamental transmission frequencies typically range from 12 to 2,250 kHz. Source levels were also provided by manufacturers of 99 of the systems and ranged from approximately 187 to 237 dB re 1µPa at 1m. Twenty four systems incorporated automated target detection and tracking software; however, most of these were designed for vessel or port security rather than for marine wildlife tracking.

To be able to measure the behaviour of marine mammals around tidal energy devices, a sonar system must meet a number of essential specifications including:

• Appropriate spatial coverage (both horizontally and vertically); this effectively determines the volume of water that can be monitored around the turbine.
• Sufficient temporal resolution (ping rate), angular and range resolution to allow marine mammals to be effectively detected, classified, and tracked.
• No interference with the behaviour of target and non-target species.

3.4.2 Target strength

The system must have a reliable detection capability, which depends on the proportion of sound that is reflected by the animal back to the receiver array. This is often termed the "target strength" and is frequency dependent and is usually expressed in decibels (dB). For smaller marine mammals, there are few empirical target strength data available. However, Au (1996) reported that the target strength of a stationary bottlenose dolphin under controlled conditions was relatively low, with mean broadside target strengths ranging from −11 to −24 dB dependent on transmission frequency. Most acoustic energy was reflected from the area between the dorsal and pectoral fins, corresponding to the location of the dolphin's lungs. Similarly, Doksæter et al. (2009) measured target strengths of 22 marine mammals (assumed to be dolphins or small whales) from a seabed mounted Simrad EK60 (38kHz); mean target strengths ranged from -5 to -35 dB, with an overall mean of -20 dB. Target strength measurements such as these provide an indication of the effective range over which a marine mammal species could be detected by sonar and also potentially provide a basis for discriminating marine mammals from other marine targets ( e.g. fish, debris). Air, for example in the lungs, is compressed with increasing pressure as animals dive deeper. As noted above, air sacs may make the largest contribution to target strength and consequently target strength of marine mammals with decrease with depth.

3.4.3 Frequency choice

Most marine mammals rely heavily on sound as a means of navigation, and for detecting prey, and the hearing and vocal ranges of many species overlap with the transmission frequencies of many of the commercially available sonar systems (approximately 12 to 150 kHz). Thus appropriate active sonar frequencies must be chosen to avoid potential negative impacts including from auditory injury (Southall et al. 2007), changes in behaviour (Richardson et al. 1991) or interference with communication (Fristrup, Hatch & Clark 2003). While an animal's hearing may be most vulnerable to damage from sounds within its auditory range, intense sounds outside this range can also cause damage. Similarly, although the fundamental frequency of a sonar signal may be above the auditory range of a marine mammal, the source may also produce, unwanted, lower frequency energy which may be audible. Therefore, when considering the choice of active sonar, it is important to review the auditory capabilities of both the target and non-target species. These capabilities vary significantly between marine mammal species. For example, harbour porpoise hearing threshold at 1 kHz is about 80 dB re 1µPa (Kastelein et al. 2002), while it is about 26 dB re 1µPa for harbour seals. This means that a sound with a pressure level of 80 dB re 1µPa and a frequency of 1 kHz would be relatively loud to the seal (Kastak & Schusterman 1998). However, the same sound be perceived as barely audible to the porpoise.

3.4.4 System tests

Although a number of published studies have used active sonar to measure some aspects of marine mammal behaviour (Nottestad et al. 2002; Benoit-Bird & Au 2003; Benoit-Bird, Wursig & McFadden 2004; Doksaeter et al. 2009; Gonzalez-Socoloske, Olivera-Gomez & Ford 2009; Gonzalez-Socoloske & Olivera-Gomez 2012), none have measured fine scale marine mammal interactions with tidal turbines.

In a recent study of marine mammal interactions with tidal turbines, two 375 kHz manually-scanning sonar systems (Tritech Super SeaKing ¹¹ ) were deployed on the SeaGen 1.2 MW tidal turbine at Strangford Narrows (Northern Ireland) in the vicinity of a harbour seal colony (Hastie 2009). The primary aim of the study was to evaluate:

• The efficiency and reliability of the sonar as a monitoring and mitigation tool for marine mammals on an operational tidal turbine.
• The frequency of close range interactions between marine mammals and tidal turbines, and to compare movement metrics of marine mammals and other mobile targets as a basis for automated classification of marine mammals.

A total of 135 hours of real-time monitoring was carried out using a combination of visual and active sonar techniques. In all 72 marine mammals were sighted close to the turbine; this compares to a total of 87 other mobile targets that were detected using the active sonar. Comparison of the sonar targets to the spatial and temporal information on sightings made by the visual observer information suggested that a number of the sonar targets (22 targets; 16% of all targets) were marine mammals. These included harbour seals, harbour porpoises, and grey seals. The overall target detection rate was 1.18 targets per hour while the rate for confirmed marine mammal targets was 0.16 per hour. When sightings of marine mammals within the area covered by the sonar were compared with sonar targets, the percentage of sightings that could be matched with sonar targets was 46.7%.

The results of this study illustrate that small marine mammals (and other mobile targets) can be detected in a tidally turbulent water column in real time using sonar up to ranges of around 50m. The relatively low detection rate (46.7%) is potentially due to limitations with this sonar system ( e.g. poor temporal and spatial resolution) and to the inherent problems associated with high frequency acoustics in tidal environments and targets close to the water surface. It is known that the highly heterogeneous water characteristics ( e.g. density) near the surface or wind generated clutter (Kozak 2006) can have significant impacts on the imaging capabilities of sonar.

Although the initial results of these trials were encouraging (Hastie 2009), it was clear that a certain amount of development was required to produce an effective sonar system for use around tidal turbines. Hastie (2012) therefore carried out an R&D study in collaboration with sonar manufacturers to develop a system. Through correspondence with 39 sonar manufacturers and subsequent behavioural response tests with captive grey seals and wild harbours seals, and field validation trials, the Tritech Gemini system was chosen to be developed and trialled on a tidal turbine.

The Tritech Gemini ¹² is a 720 kHz forward looking multi-beam sonar that is designed for detecting objects in the water column. It is a 2D system that allows detection and localisation of objects in the X-Y plane but does not provide information on the depth of the target. It has the following features:

• update rate: between 7 and 30Hz
• angular range resolution: 0.5°
• range resolution: 0.8 cm
• horizontal and vertical swathe widths: 120° and 20° respectively (up to 4 heads can be synchronised by pinging in sequence)

3.4.5 Data processing

Post-processing detection and classification software (SeaTec) was developed (by Tritech software engineers with input from SMRU Ltd) that used target similarity to a marine mammal (using flood-fill techniques), tide-weighted target velocity, and target path to estimate the likelihood that a target was a marine mammal. Field tests near a grey seal colony suggested a reliable (probability >0.95) detection range of around 44 m. Classification in real time using the SeaTec software was encouraging with all marine mammals being correctly classified. However, it tended to be relatively conservative with a proportion of other targets (such as floating logs or buoys) also being classified as marine mammals. Further post hoc processing can be carried out to refine these classifications and successfully reduce false positives; however, this currently requires user-intensive manual analysis.

In 2011 this Gemini system was trialled on the 1.2 MW turbine at Strangford Narrows. Over a period of 42 days a total of 109 targets (average of 5.9 per day) were classed as 'highly probable marine mammals' by processing with the SeaTec algorithms in real time and running the post hoc analyses on the results (see Figure 1). It should be noted that although there were no external data ( e.g. visual observations) for validation of these targets, the raw sonar data for a proportion of the targets were reviewed and in most cases appeared to be marine mammals. As described above, there was scope for improvement through automation of the post hoc classification analyses.

3.4.6 System readiness

It is clear that active sonar can be used to detect and track marine mammals in the vicinity of fixed structures such as tidal turbines. However, few off-the-shelf systems have the spatial and temporal resolution, range and 3D detection capabilities required to track marine mammals. Furthermore, it is critical to carefully consider the acoustic characteristics of the system and the hearing ranges of the species of interest in order to avoid difficulties in teasing apart responses due to the turbine with responses due to the sonar. The main task now is to improve and automate the marine mammal classification process to reduce the level of manual post hoc analyses ¹³ . Also, active acoustic systems are directional and multiple units (which will need to be synchronised to avoid interference) may be required to provide good coverage.