Institute of Ocean Sciences, 9860 W Saanich Rd., Sidney BC, V8L 4B2, ErbeC@pac.dfo-mpo.gc.ca

Marine mammal species include whales, dolphins and porpoises (Order Cetacea); seals, sea lions, walrus, otters and polar bears (Order Carnivora); and manatees and dugongs (Order Sirenia). Censuses have been carried out for many decades to determine population trends, classify the risk status of species, decide how many animals can be taken per year without depleting stocks, monitor migration, identify critical habitat, assess the short-term and long-term effects of noise (displacement, mortality, reduced reproductivity) and so forth. Various methods for marine mammal censuses exist. They are described and compared in the following. At the moment, no single technique by itself suffices. Techniques need to be combined in order to design correction factors for missed or multiply-counted animals. In my opinion, the future lies with automatic acoustic surveys, as discussed at the end.

The historical and still standard method for taking a census of marine mammals is visual surveying. This is done either from aircraft or boats or shore-stations, and on rare occasions from kites, balloons or remote-controlled planes. As visual surveying has been the standard technique for decades, the associated literature is vast, for example see Wilson et al. 1996 for census methods for mammals in general, and Garner et al. 1999 for the most recent advances of marine mammal (visual) censuses in particular. Visual survey techniques have been optimized by biometricians and biostatisticians yielding standard ranges, standard speeds and altitudes, standard numbers of observers and rotation schedules, standard training for observers, and standard correction factors for the animals' time spent at the surface, sea state, visibility etc. Visual surveys are normally done along line transects (e.g. Morgan 1986, Palka and Pollard 1999). Subsequent biostatistical modelling estimates the total number of animals in a population from the number of animals seen.

Visual surveys from close ranges potentially allow individual identification, either by experienced observers or by simultaneous photo-ID. Another advantage over other techniques (e.g. the current status of acoustic surveying) is the fact that visual surveys yield absolute minimum animal numbers as well as relative numbers provided by other techniques.

Despite being the standard technique and being so well developed, visual censuses have a number of downsides, which are mostly impossible to overcome:

• Visual surveys are impossible at night, unless an infrared camera is used. IR sensors proved useful in the detection of whale blows at night in the study by Perryman and Laake (1995).
• Detectability declines rapidly with distance (e.g. Cummings and Holliday 1985).
• Animals can only be counted at or near the surface, where some species spend only about 10% of their time.
• Observer bias. Subjectivity.
• Richardson (1998) showed that one "biologically qualified" observer counted fewer animals than two.
• Depending on the transect lines chosen, there is a chance of counting the same individual twice, unless surveys are accompanied by photo-ID or animals are otherwise tagged or marked.
• Visual surveys are often very difficult to compare to one another (particularly to historical surveys), because of bad consistency and lack of detailed information, such as observed ranges, training level of observers, weather etc. Also, methods have changed. For example, in 1977, the Alaska bowhead census estimated about 1000 animals, compared to 8000 in 1993. This population did not suddenly experience a boost in reproductivity, but rather methods of accounting for missed whales improved (Zeh 1999).
• The presence of the survey boat or plane can disturb animals, leading to avoidance and thus lower population counts (Salvado et al. 1992).

Some unusual visual techniques include Kingsley et al.'s (1990) use of infrared sensors to detect seal lairs and polar bear dens under snowdrifts. Lavigne and Oritsland (1974) and Lavigne (1976) described the use of ultraviolet sensors to detect white polar bears and seals on ice. Radford et al. (1994) used airborne synthetic aperture radar to detect whale wakes through clouds and at night. As was pointed out by Yvan Simard and Bill Karp during the meeting, (military) satellites are able to see individual whales. Access to these satellite data would be of immense value for marine mammal monitoring.

Photography is the standard technique used for censuses of pinnipeds on haul-out sites, where hundreds of them can be sitting squeezed next to each other. Photos are taken from aircraft for later animal counting in the laboratory. This method gives absolute minimum animal numbers. Correction factors are used to account for the number of animals at sea.

Photo-identification can occasionally be applied as a type of capture-mark-recapture census. In this type of census, a number of animals M of a population are captured and marked before being released back into the environment. At a later stage, a visual survey is conducted counting n animals out of which m were marked. The total population is estimated as N=n*M/m. Current techniques can be fairly advanced (Pollock et al. 1990). Photo-ID is good for short-term monitoring at a given location, e.g. to see if animals return after human disturbance. Photo-ID is also useful for long-term monitoring of survival.

Most marine mammals are highly vocal. Sound propagates very well over long ranges in the ocean. These two facts support the application of passive acoustics to marine mammal censuses. One or more hydrophones are deployed commonly in a fixed array or towed behind a boat or suspended from freely drifting sonobuoys. Animal calls are received and animal locations calculated. Depending on the algorithm used, this can be done in real-time, or sounds are recorded for later analysis in the laboratory.

An example for a fixed array is the US Navy sound surveillance system (SOSUS) and now integrated undersea surveillance system (IUSS). This is a network of bottom-mounted hydrophones. SOSUS provides good coverage of the North Atlantic and North Pacific, in deep offshore water, where visual surveys are very costly (Nishimura and Conlon 1993). SOSUS works best for baleen whales, or any species calling at low frequencies (a few Hz to a few 100 Hz). SOSUS' feasibility has been demonstrated in a few studies documenting seasonal movement patterns and relative vocal activity of baleen whales (Moore et al. 1998, Stafford et al. 1998, Stafford et al. 1999, Watkins et al. 2000).

Linear hydrophone arrays towed behind a ship have widely been employed (Thomas et al. 1986, Spikes and Clark 1996, Clark and Fristrup 1997, Norris et al. 1999). They yield the direction (bearing) to a calling animal; and if the inter-hydrophone distance is roughly less than the wavelength of the animal call or if the array is very long, the whale's position can be located.

Hydrophones can also be suspended from sonobuoys (which are generally freely drifting, but may be anchored). DIFAR (directional low-frequency analysis and recording) buoys can determine the calling animal's direction. Any array of three or more sonobuoys can locate the animal. Sound recordings (and GPS position of the buoy) can be stored on the buoy for later retrieval (e.g. Hayes et al. 2000). Alternatively, sound can be transmitted to a shore station (or boat or plane) via satellite or VHF radio (e.g. Cummings and Holliday 1985).

Hydrophones can further be deployed by various other means, attached to anything. Janik and Thompson (2000) deployed a 3-hydrophone array to locate bottlenose dolphins and harbor seals. Clark and Ellison (2000) suspended hydrophones from the ice edge in Alaska. Apart from being suspended, autonomous recorders can be bottom mounted, in which case an acoustic release can be used to retrieve the recorder (Moore et al. 1998). Aubauer et al. (2000) showed that under certain circumstances (short range, shallow water), location can be achieved with just one hydrophone using different arrival paths (direct and reflected).

The low-frequency sound of baleen whales travels well into most sea floors and can therefore also be detected with geophones originally deployed for seismic monitoring.

The standard method for determining location with an array of hydrophones is by time of arrival differences (Watkins and Schevill 1974, Cummings and Holliday 1985, Mohl et al. 1990, Freitag and Tyack 1993, Frankel et al. 1995, Janik and Thompson 2000, Clark and Ellison 2000). Spiesberger and Fristrup (1990) studied how bottom topography affected location. The method of time of arrival differences works best at short range, when the animal call arrives along direct paths at the array of hydrophones. In some environments, and if the animal is not too far away (or the water too shallow, or the calls too long), reflected rays can be identified easily and sorted out of the recordings (Freitag and Tyack 1993). In long-range detection, multiple rays from different paths can arrive simultaneously, making the picking of arrivals impossible. In such cases, matched-field processing has recently been successfully applied to marine mammal location. Thode et al. (2000) used a tilted vertical line array to determine range and depth of a calling blue whale over ranges of 60 times the water depth. With one separate hydrophone, the location could be fixed in all three dimensions.

The advantages of acoustic censuses over visual censuses are manyfold:

• Acoustic surveys can be done at night, 24 hours per day, under ice, in high sea state, in fog etc.
• Hydrophones survey whales underwater, where they spend most of their time.
• Longer detection ranges: Cummings and Holliday (1985) deployed a hydrophone array off Alaska locating bowhead whale sounds to determine if whales migrated past census stations beyond visual range and were uncounted. "Based on a sample of 182 whale sounds (over 48 h) from closest point of approach (CPA) distances out to more than 10 km, 68% originated beyond 2 km (CPA), where only 1% of the 242 whales were sighted. No whales were sighted beyond 3 km during this time, but 53% of the located sounds originated that far and beyond."
• The detectable range is well predictable in the local environment and ambient noise. This is not the case with visual surveys.
• Acoustic surveys have the potential of being done fully automatically, not requiring any permanent personnel. Detection, location and tracking can be automated.
• Automatic acoustic surveys are not prone to observer bias, and are therefore objective and repeatable and comparable to other automatic surveys.
• For some species, pods and even subgroups can already be identified automatically using "intelligent" software (Deecke et al. 1999). No trained observers will be needed, also eliminating observer bias.
• Acoustic surveys currently work well to identify the geographic distribution of species and population trends over time.

Passive acoustic surveys, on the other hand, fail to predict absolute animal numbers (whereas visual surveys can at least determine absolute minimum numbers). Correction factors are badly needed to adjust the percentage of animals singing at any one time. This will depend on the social structure of the group (e.g. in humpback whales, only males sing) and on the behavioral state of the animals (silent resting versus socializing versus foraging etc.). For some species, particular calls have been linked to particular behavioral states (Ford 1989, Ford 1991). This is the first step towards successful designing of correction factors for the percentage of animals vocalizing during any particular activity. Photo-identification allows the identification of individual whales. Acoustic individual identification still lies in the future. Passive acoustic surveys therefore have a very large risk of re-counting the same individual. At the moment, the lack of correction factors for acoustic surveys makes the determination of absolute population numbers infeasible. Correction factors need to be developed by combining acoustic surveys with visual surveys (to determine typical group size and minimum numbers) or other techniques, e.g. acoustic tagging (to determine which and how many animals vocalize at the same time). Obviously, if two calls originated from roughly the same location, passive acoustics alone would not be able to tell whether there were two individuals, or whether one individual called twice. Signal detection procedures further need to be able to deal with the case of a number of animals calling at the same time.

Active acoustic surveys, i.e. using active sonar for marine mammal censuses, could be useful to detect silently resting marine mammals floating near the surface in order to avoid collision with ships. Particularly the North Atlantic right whale population, prone to frequent collisions, would be a good candidate. Such a system is being developed by Miller et al. (1999). There have been various measurements of marine mammal target strengths (e.g. Au 1996). Very high frequencies are needed to detect objects floating at the surface. High frequencies generally give higher target strength (because of higher surface impedance) and are thus good in noisy environments. However, due to increased absorption by ocean water and scattering, they don't travel as far as low frequencies. It is generally easier to generate loud sources at high frequencies than at low frequencies. A concern with active sonar, though, is that the source must not be too loud to harm marine life.

Tagging is a most useful tool for determining correction factors for marine mammal surveys. For example, time-depth-recorders readily yield the amount of time the tagged animal spends at the surface. Tagging a few animals within one population can thus give a correction factor for visual surveys of this population or species. If a GPS is included in the tag, or a radio transmitter that allows location of the animal every time it surfaces, the swim speed can be calculated, which is useful to design proper transects during visual surveys. Acoustic tags are needed to determine correction factors for acoustical surveys, such as the amount of time vocalizing and the percentage of animals vocalizing. Such tags include a device that flashes when the animal produces a sound (Tyack 1985) or that records the sound of the tagged animal (Tyack 1986). An acoustic tag fitted with a GPS can be used to identify calling animals in an acoustic survey with a hydrophone array. Recorded data can be stored inside the tag for later retrieval by acoustic releases or, e.g., when pinnipeds molt, tags will fall off. The other option is VHF and satellite telemetry (via the ARGOS satellite system) to send data back to a receiving station on shore, on a boat or airplane. A wide range of potentially useful telemetry sensors is available, including temperature, pressure (depth), salinity, GPS, light intensity, sound, heart rate, body temperature, and swim speed (Kraus 1998). Data can also be sent to a receiver by acoustic modem.

From my experience, tagging does not appear to affect whales behaviourally. I.e., whales did not seem to notice being tagged, nor did the tagged individuals nor their untagged neighbours alter their behaviour. This was also stated by Watkins et al. (1981).

Tags, even if they are only there to identify the animal (e.g. branding pinnipeds) are useful for long-term (until the animal molts) monitoring of survival, distribution and site-fidelity. They also help to correct for multiple counting of the same individual in visual surveys. They can be used in capture-mark-recapture surveys.

This depends on the particular problem posed. For example, in the case of human activities emitting underwater noise, marine mammal protection regulations often require that animals be counted in the vicinity and that the noisy operation be shut down if animals occur within a critical range. In this case, you need a technique that detects a high proportion of marine mammals over the extent of the critical range. In a different case, where animal numbers need to be compared (e.g. between an ensonified versus control area), it may be sufficient to detect a consistent proportion.

At the moment, no single technique by itself can give absolute numbers of animals present with much confidence. Therefore, techniques are and should be combined. For example, the LFA SURTASS Marine Mammal Research Program combines all three techniques discussed above: visual observations from aircraft, ships and shore with acoustic tracking by bottom-mounted hydrophones and towed arrays, and tagging. Combined visual and passive acoustic surveys have been carried out in a number of other studies (Clark and Fristrup 1997, Moore et al. 1998, Ko et al. 1986, Zeh 1999). A combination of methods also allows a cross-validation of methods. This should result in the improvement of all methods combined as well as in correction factors (for the time spent at the surface, the time spent calling etc.).

In my opinion, only acoustics has the potential for long-term automatic and objective censusing without recurring and ongoing costs of personnel and ship or plane time. In particular, the following points need to be addressed with future research.

• Double or triple techniques are required to determine correction factors.
• Correction factors need to be a function of behavioural state and social structure of the group. More research is needed to link particular calls (and percentage animals calling) to behaviour and social structure.
• Automatic signal detection algorithms (and noise cancellation) need to be optimized for marine mammal call detection. Mellinger and Clark (2000) used a matched filter, a hidden Markov model and spectrogram image convolution to detect bowhead whale vocalizations in noise. The methods were compared with respect to their false alarm and miss rates. Spectrogram image convolution performed the best, having the smallest combined error rate. The hidden Markov model followed; matched filtering had the highest error rate. Potter et al. (1994) designed an artificial neural network trained with back-propagation and tested it on the same bowhead data set. The neural net performed even better than spectrogram image convolution. Mellinger and Clark (1997) tested matched filtering and spectrogram image convolution on a few mysticete sounds from blue, fin and minke whales. They found that matched filtering worked well if the signals to be detected were buried in white background noise. Spectrogram image convolution excelled if the interfering noise was structured, e.g. harmonic. In the case that the animal call was a highly repetitive sequence of sounds occurring at regular intervals (a pulse train), a summed auto-correlation method proved useful. Erbe et al. (1999) and Erbe (2000) studied the detectability of beluga whale calls in icebreaker noise, naturally occurring icecracking noise and Gaussian white noise. Combining detectors at a level of 50% correct detections, human listeners generally outperformed a beluga whale, followed by automatic detectors in the order of spectrogram cross-correlation, a back-propagation neural network and finally a matched filter. Clearly, more work needs to be done to optimize automatic detection of marine mammal calls in underwater noise.
• In killer whales, vocal dialects have been identified, and pods and even subpods can be identified by ear or automatically using a neural network (Deecke et al. 1999). Individual identification, which can easily be achieved by photography in some species, is still a challenge acoustically. Gender identification, on the other hand, can be easier acoustically than visually, e.g. in humpback whales where only males sing. More research on marine mammal communication is needed to tell age classes apart.
• At the moment, efforts are being made at the University of Santa Cruz (Hayes et al. 2000) to develop a low-cost long-lasting sound recording and storing sonobuoy. While such a system would immediately be of immense value to marine mammal researchers, I strongly believe the future lies with on-buoy data processing and radioing (or storing) of results.
• Software for automatic detection of marine mammal calls, automatic arrival time picking (e.g. by cross-correlation), arrival-time-difference computation (in the case of this method) and location should be encoded on digital chips and be done on-site, on the sonobuoy or other. Processed information (what species, when, where, how many) can then be transmitted to shore, rather than transmitting the raw acoustic signal. Alternatively, the results can be stored on the buoy for later retrieval. This would cut down data storage requirements considerably.
• In the developmental stages, or in any case for further analysis in the laboratory, it would be profitable if only useful information could be stored on the buoy. At the moment, recorders usually manage to record for a total of a few hours. If an automatic call detection algorithm was implemented on-site, the recorder could repeatedly overwrite short samples until a signal of interest was detected. Only this would have to be stored for later location analysis.
• Battery requirements need to be researched and likely improved for on-site processing. If hydrophones are deployed not too far from shore, cables rather than batteries could be used. Janik and Thompson (2000) had 150m power cables connecting a 3-element array with shore.
• Ideally, all automatic processing would take place in real-time.
• A problem with acoustic surveys is posed by marine mammals that decide to be silent. Acoustic surveys of deer were very efficient, when the animals were disturbed eliciting a vocal response (Reby et al. 1998). I would suggest a project studying the possibility of evoking a vocal response in marine mammals. Play-back of animal sounds and artificial stimuli should be tested.
• The detection of new species using passive acoustics is likely very far in the future, as the repertoire of known species is still an active area of research.

With increasing public concern about marine mammal survival and biodiversity, and persistent conflicts with subsistence hunting and whaling countries, accurate censusing of marine mammals has never been more urgent. Methods need to be objective (repeatable and comparable), as well as cost and time efficient. In my opinion, passive acoustics is the basis for the future.

This page is maintained and was last updated 21-Nov-00 by Pip Sumsion (sumsionp@dfo-mpo.gc.ca).