9

A statistical pattern recognition technique for classification by supervised learning was developed and applied to automated recognition of marine mammal sounds [J. Acoust. Soc. Am., 101(3), March 1997]. Training data identified by human experts are characterised by occupancy statistics associated with a multiple-resolution, binary partition of the unreduced observation space. Classification of a new sample is performed by Bayesian inference applied to these occupancy statistics. The classification algorithm is implemented in a simple, highly efficient computer program. The present work describes efficient encoding of the training data distributions for large numbers of training samples and classes, and efficient evaluation of a posteriori probabilities of class membership for classification of new samples. Data storage requirements and computational efficiency of the classification algorithm are compared with theoretical bounds.

Man made homing systems have evolved more or less independently of any knowledge of biological echolocation systems since the early decades of this century. The prime objectives and general operating environments of homing missiles and torpedoes, however, have much in common with dolphins and other echolocating predators. They need to detect and classify all potential targets in all directions over as great a range as possible, identify and spatially localise the most interesting one, determine the most efficient interception course, and then intercept and attack the target despite evasive manoeuvres and potentially distracting alternative targets. In this paper, experience in modelling and simulating radar and sonar homing systems is combined with the available knowledge of dolphin echolocation to produce estimates of the potential physical capabilities of the dolphin systems and the effects of various environmental factors such as multipaths and reverberation, glint and multiple targets, ambient noise levels, and fluctuations due to turbulence and other random variations in the seawater medium.

The Atlantic bottlenose dolphin Tursiops truncatus is the best studied example of successful adaptation of biosonar to the reverberant and variable VSW/SW environment. With evolved and learned adaptations, it can successfully search for, detect and classify objects in conditions that usually defeat the best artificial sonars. The adaptations of the transmitting, receiving and analysing systems that reduce reverberation and noise are reviewed as background for recent studies of the dynamics of biosonar performance in open-fields; work that suggests still further means of enhancing the S/N ratio. The open-field work uses new experimental methods that control for additional extraneous variables as well as instruments that continuously record the animal's location, head-attitude, head-azimuth, interpulse intervals and emission spectra. The instruments include a high- speed data acquisition package that is carried by the animal and a guide boat that is equipped with a high-accuracy DGPS, an interactive field-display for the driver and a computer that coordinates time, position and data acquisition. Systematic measurements with these methods allow for the analysis of the animal's distribution of biosonar energy in 3D space, time and frequency as a function of practice effects, object distance, grazing angle and environmental variables.

A subtidal-zone (<10 m) system for sound transduction, acquisition, and archival was developed to study and monitor harbour seals. In developing the transduction system - a multi-channel hydrophone array - several engineering challenges were solved: cable armour was applied and smooth low-noise hydrophone cages were designed to protect against animal bites. Helical anchors were used to attach the cable to the sandy bottom; they are adjusted as necessary when sand is moved by currents. The cable was weighted so that it sank into the sand to prevent cable strum from wave action. The acquisition and archival system had several engineering goals: simultaneous multi-channel sound acquisition; changeable sampling rate; high data rate (200 Kbyte/s); large storage volume to allow continuous long- term monitoring; configurable recording times. The system developed was a PC with a multi-channel data acquisition board (SignaLogic), custom software that allows automatic reconfiguration (channels recorded, sampling rate, etc.) at preset times, and a high-capacity (4 Gbyte uncompressed) data DAT drive. The tape drive software allows acquired data to be accessed directly as computer files, obviating a separate re-acquisition step. The acquisition system compares favourably to multi-track tape recorders in price and very favourably in recording capacity.

Real-time and continuous passive acoustic monitoring of the ocean is one goal of the Monterey Bay Aquarium Research lnstitutes Ocean Acoustic Observatory project. Continuous real-time monitoring makes it possible to observe episodic and non-predictable phenomena such as biological and geological events. Animal calls, turbidity flows, underwater landslides and volcanic eruptions are examples of such events. MBARI has developed a state- of-art real-time system for multi-channel acquisition, beamforming, processing and archiving of acoustic data. The system is capable of collecting data from up to 32 hydrophones at the maximum sampling rate of 200 kHz per channel and can form up to 68 simultaneous beams in real time. A four- processor application accelerator performs various signal processing functions on the raw or beamformed data. Spectral analysis such as spectrogram computation, automatic detection, localisation, classification and display of sounds can be performed in real time. An event-based recording scheme can be selected to record only acoustic events of interest. The system is expandable to handle up to 512 channels of data from fixed or towed hydrophone arrays. This paper presents a detailed description of the acoustic observatory system, implemented algorithms and results from fixed or towed arrays from the Monterey Bay ocean region in Northern California.

The study of marine mammals is generally characterised by individuals or small teams working on a very tight budget. The use of advanced electronic technology to assist these studies has traditionally been severely limited by funding. However, the trend in microelectronics, driven by the demands of the personal computer and home entertainment markets is to continually reduce the cost of technology and to make more powerful systems available to the mass market. The marine mammal research area can now benefit from this advanced technology to provide a number of useful tools. At the lowest levels, the advances in basic analogue device performance now allows the building of hydrophone-preamplifier units with good noise performance and high dynamic ranges to achieve small, low-cost analogue channels for underwater acoustic data acquisition. The current generation of analogue-to-digital convertors allows the full bandwidth of these signals to be converted for subsequent digital processing. The demands of the personal computer market are making computer power more easily available to the researcher at a number of levels, ranging from the new RISC-based personal digital assistants which are ideal for data acquisition in the field to high performance desk-top machines based on Pentium or PowerPC processors capable of demanding spectrum analysis tasks. This paper explores the advances made in recent years in both analogue and digital devices to demonstrate the level of technology now available and conceptually design a number of systems that could be assembled to aid marine mammal research.

Shipping has been responsible recently for many collisions with cetaceans in the Canary Islands. A series of experiments, including the playback of artificial sounds of different frequencies, was conducted to test a system designed to keep sperm whales apart from the ferries routes. The results showed that the whales did not react to low frequency playbacks which suggests sperm whales from an area which has heavy vessel traffic have a high tolerance for noise. After the collision of a ferry with two sperm whales, ears were extracted from the two individuals, in order to assess the health of their inner ears. CT scans showed that there were no fractures or other overt evidence of impact, or ship strike related injuries; however, ears from both animals had reduced auditory nerve volumes. 0ne animal also had patches of dense tissue in the inner ear. These findings are consistent with auditory nerve degeneration and fibrous growth in response to inner ear damage. In combination with the results from the playback experiment, these results suggest that low frequency sounds from shipping may be affecting hearing and increasing collision rates. These findings are however, preliminary, and histologic analyses are underway to determine whether the primary cause of the ear changes seen with CT are disease or noise induced.

The U.S. Navy's SURTASS LFA is an active sonar system under development since the late 1980's, for employment from a small number of auxiliary ships assigned to conduct undersea surveillance to detect, classify and track potential threat submarines. SURTASS LFA system development is important because traditional passive systems are facing increasing challenges in their detection capabilities due to improved submarine quieting technologies. SURTASS LFA hardware consists of an array of acoustic transmitting components suspended an average of 100 meters beneath the host ship, which travels at a maximum of 3-4 knots This vertical array of transducers transmits waveforms (100-500 Hz) which are much longer than other active sonar systems. Echoes reflected off objects such as submarines are then detected on passive towed horizontal line arrays (HLA), and processed and evaluated to identify and classify them. The U.S. Navy will soon be at the stage of transitioning the system from a test and evaluation status to the fleet for worldwide employment to enhance undersea warfare capabilities. At present, the U.S. Navy has only one system. If proposals are carried through to completion, a total of four systems will be procured, two each for the Atlantic and Pacific fleets. High sound levels can potentially be harmful, either by affecting hearing or by causing other physiological effects; however, the combination of high transmission losses in the marine environment and the implementation of proven mitigation measures (visual and acoustic monitoring, source ramp- up, sound pressure level monitoring, shut-down criteria), reduce the potential for harmful sound levels from the SIJRTASS LFA system reaching marine animals and divers to a negligible level. During the development of the SURTASS LFA system, the U.S. Navy prepares various environmental analyses prior to sea tests, which are coordinated with the appropriate agencies responsible for wildlife protection. In July, 1996, the U.S. Navy announced a proposal for operational employment of the system and the initiation of a worldwide environmental impact statement (EIS) to evaluate the potential environmental effects of such deployment. The EIS is one element of a synergistic plan composed of three primary thrusts; 1) a comprehensive scientific research program (SRP) under the aegis of some of the world's most prominent marine bioacousticians, using the SURTASS LFA system as a scientific tool to collect much-needed data on the potential effects of low frequency sound on marine mammals; 2) an intensive diver risk assessment involving in-water low frequency sound measurements with human subjects under the direction of the U.S. Navy Submarine Medical Research Laboratory; and 3) the EIS, which will be available to the public in draft form in 1998.

The role of underwater sound as a potential stressor in the marine environment is now widely recognised and the designers of sonars find themselves increasingly constrained by environmental legislation which requires them to consider the possible harmful effects of high power sound transmissions on marine life (e.g. fish and marine mammals) and on human beings. This paper describes a formal process of environmental impact assessment being developed in support of the procurement of future sonars in the UK. The basis of this process in environmental legislation is briefly reviewed but the main purpose of the paper will be to consider the complex scientific and technical issues surrounding environmental impact assessment. In particular, Environmental Assessment (EA) for sonar systems requires a process of cause and affect modelling to be undertaken. Sonars may produce both energy and substance pollution (e.g. explosives may release toxic compounds). The 'precautionary principle' which is enshrined in environmental legislation puts the onus on the polluter to prove that his particular form of pollution does not have a harmful effect on the environment. But how are environmental impact criteria defined? Toxic effects are relatively easy to test for and to quantify. Sound energy on the other hand is rather more difficult. Consideration of the hearing sensitivity of fish, for example, leads to the notion of safe exposure level and probability of avoidance. But how representative are the experiments on which these criteria are based (e.g. the impact of seismic airguns on fish catch rates)? How can we assess the reliability of the scientific evidence given the uncertainties elsewhere, e.g. poor or inadequate knowledge of sound propagation characteristics (including non-linear effects associated with impulsive sound sources), uncertainty in environmental conditions, natural variability and the cumulative effects of repeated exposure to sound energy transmissions? There have been few coincident measurements of sound intensity in the ocean at the ranges at which a particular species exhibits avoidance behaviour and many studies make simplifying assumptions regarding acoustic propagation, e.g. spherical spreading out to unrealistically long ranges, These and other topics will be reviewed with example calculations 'to illustrate particular aspects of the EA process which is being developed by the authors.

In recent years research into the whistle structure of different species of odontocete calls has attempted to identify distinctive individual, pod, or simply species-specific, features, that might assist in acoustic identification and discrimination. So far most of these studies have concentrated on the frequency modulated signal by examining these for characteristic features in the frequency-time domain using FFT analysis. Such analysis has concentrated on the centre frequency, frequency deviation, number of inflection points (frequency reversals), and more significantly the general shape or contour of the call. The Source Level of echolocation signals has also been studied for other reasons but this parameter appears to reflect some body size dependency. This paper discusses the feasibility of using the maximum Source Level of the narrow band social calls (whistles) as an additional cue when attempting to distinguish between species while studying free-ranging animals. The data used for spectral and statistical analysis was recorded from a Dutch fishing research vessel (FRV Tridens) during the 1996 and 1997 CETASEL project sea trials. A single hydrophone was attached to a pelagic trawl net fishing in relatively deep water (100-200 m depth) along the edge of the continental shelf in the Eastern North Atlantic sea areas SW of Eire, through Biscay and towards Finnisterre. Maximum sound pressure levels of social calls were extracted and converted into Source Levels (SL re 1µPa at 1m) using a calibrated 13 kHz pulsed cw pinger (Dukane) to provide a known SL reference. The necessary range information (distance between hydrophone and vocalising animal) was obtained by means of a multi-path echo-ranging method (Kaschner et al 1997). Parallel visual observation in good weather conditions provided the species identification and a time series analysis established the probability of the association between these sighted animals and the acoustic recordings of species whistles. On one occasion a very vocal group of bowriding Common Dolphins Delphinus delphis were recorded simultaneously in air, using a gun microphone directly above them at the bow, and underwater using the hydrophone attached to the trawl at a distance of 518m behind them. The maximum underwater Source Levels of individual signals which were recognisable in both recordings were calculated and compared with the other results in order to test the validity of the method described above. This paper discusses the methodology and limitations of this relatively simple technique but the results achieved to date suggest considerable potential for SL estimation in open sea conditions. Alternative, quieter, platforms than a pelagic trawler are recommended.

Kaschner, K. et al. (1997). Analysis and interpretation of cetaceans sounds obtained by a hydrophone attached to a pelagic trawl. In European Research on Cetaceans - 11. Proc. 11th Ann. Conf. ECS, Stralsund Germany (Eds. P.G.H. Evans) European Cetacean Society, Cambridge