Marine Mammal Vocalizations Research Articles

A source separation approach to enhancing marine mammal vocalizations

A common problem in passive acoustic based marine mammal monitoring is the contamination of vocalizations by a noise source, such as a surface vessel. The conventional approach in improving the vocalization signal to noise ratio (SNR) is to suppress the unwanted noise sources by beamforming the measurements made using an array. In this paper, an alternative approach to multi-channel underwater signal enhancement is proposed. Specifically, a blind source separation algorithm that extracts the vocalization signal from two-channel noisy measurements is derived and implemented. The proposed algorithm uses a robust decorrelation criterion to separate the vocalization from background noise, and hence is suitable for low SNR measurements. To overcome the convergence limitations resulting from temporally correlated recordings, the supervised affine projection filter update rule is adapted to the unsupervised source separation framework. The proposed method is evaluated using real West Indian manatee (Trichechus manatus latirostris) vocalizations and watercraft emitted noise measurements made within a typical manatee habitat in Florida. The results suggest that the proposed algorithm can improve the detection range of a passive acoustic detector five times on average (for input SNR between -10 and 5 dB) using only two receivers.

Detection of killer whale presence using low duty cycle recorders.

Passive aquatic listeners (PALs) have been developed to make long-term (up to a year) autonomous measurements of the underwater ambient noise in the marine environment. Because of data storage limitations and low-power requirements, PALs have a very low duty cycle, of the order of 1 percent. In fact, they are optimized for acoustic rainfall detection and have been successfully used to measure wind speed and rainfall rate at sea using spectral components of the sound field. These geophysical sounds have relatively long time scales (minutes) compared to vocalizations of marine mammals (seconds) or mooring noise (chain clanking). Data collection was designed to detect and reject short temporal signals. However, these signals include marine mammal vocalizations. Consequently, the sampling strategy was modified to evaluate the temporal content of the data sample and save the temporal data sample only if it contains a signal consistent with killer whale vocalizations. These sound bites have demonstrated detection of pod-specific vocalizations of resident-type killer whales, transient-type killer whales, bone crunching from likely killer whale predation (eating) on sea lions, and humpback and N. Pacific right whales.

Understanding the relationship between marine mammals and their environment in the Bering Sea.

Passive aquatic listening (PAL) instruments were deployed on sub‐surface moorings along the 70‐m isobath of the Bering Sea shelf. PALs operate according to an adaptive sampling protocol as opposed to recording continuously, thus enabling data collection for a full year at a sampling rate of 100 kHz. Year long time series indicate that marine mammal vocalizations dominate the regional soundscape in the late winter/early spring time period. Species acoustically identified include North Pacific right, killer, beluga, humpback, fin, and bowhead whales, plus bearded seal and walrus. Synoptic measurements of prey abundance with three‐frequency active acoustic sensors, ice cover, and physical oceanographic processes provide vital information for understanding the relationship between marine mammals and their continually changing environment. These data represent important baseline measurements of environmental variability, marine mammal species richness, and temporal distribution of animals in an area of the world that is rapidly transforming due to global climate change. [Research supported by ONR.]

Automatic detection of marine mammals using information entropy

This article describes an automatic detector for marine mammal vocalizations. Even though there has been previous research on optimizing automatic detectors for specific calls or specific species, the detection of any type of call by a diversity of marine mammal species still poses quite a challenge--and one that is faced more frequently as the scope of passive acoustic monitoring studies and the amount of data collected increase. Information (Shannon) entropy measures the amount of information in a signal. A detector based on spectral entropy surpassed two commonly used detectors based on peak-energy detection. Receiver operating characteristic curves were computed for performance comparison. The entropy detector performed considerably faster than real time. It can be used as a first step in an automatic signal analysis yielding potential signals. It should be followed by automatic classification, recognition, and identification algorithms to group and identify signals. Examples are shown from underwater recordings in the Western Canadian Arctic. Calls of a variety of cetacean and pinniped species were detected.

Spectrogram denoising and automated extraction of the fundamental frequency variation of dolphin whistles

Marine mammal vocalizations are often analyzed using time-frequency representations (TFRs) which highlight their nonstationarities. One commonly used TFR is the spectrogram. The characteristic spectrogram time-frequency (TF) contours of marine mammal vocalizations play a significant role in whistle classification and individual or group identification. A major hurdle in the robust automated extraction of TF contours from spectrograms is underwater noise. An image-based algorithm has been developed for denoising and extraction of TF contours from noisy underwater recordings. An objective procedure for measuring the accuracy of extracted spectrogram contours is also proposed. This method is shown to perform well when dealing with the challenging problem of denoising broadband transients commonly encountered in warm shallow waters inhabited by snapping shrimp. Furthermore, it would also be useful with other types of broadband transient noise.

Historical ambient noise levels in a topographically isolated basin

During the late 1970s and early 1980s, careful measurements of ambient noise were made in the Tongue of the Ocean (TOTO) in the Bahamas. The measurements were made using a vertical array of five omni‐directional hydrophones moored in 2200 m water depth with an average hydrophone depth of 122 m. Data were collected within a 20 Hz ‐ 20 kHz band at ten specific 1/3 octave band frequencies, and then reported in spectrum level as a function of surface wind speed. Records contaminated by anthropogenic sources (e.g., local vessel traffic), biological sources (e.g., marine mammal vocalizations) and other sources (e.g., rain) were eliminated to allow analysis of background ocean ambient noise only. TOTO is a deep basin, topographically isolated from the Atlantic Ocean, and therefore largely acoustically decoupled from the Atlantic Ocean deep sound channel. The basin's geographical isolation and the controlled data collection methodology employed over a four‐year period make this three‐decade‐old data set a potential proxy for preindustrial era ocean noise levels. The environmental setting, data collection methodology, analysis methodology and results are presented and discussed within the context of historical ocean ambient noise levels.

Data Handling of a Perennial Acoustic Observatory in the Antarctic Ocean

In December 2005, 4 broadband hydrophones were deployed through boreholes under the 100m thick floating ice shelf near Atka Bay, Antarctica. Since then, more than 10,000 hours of underwater sounds have been recorded, containing marine mammal vocalizations and the unique background sounds generated by ice and the occasional ship. These records from this pristine region are analyzed in terms of the local noise budget, the behaviour of marine mammals and anthropogenic impacts. The observatory is energetically self-sufficient and connected by a 15km WLAN link to the year-round manned German Neumayer Base, where a leased satellite line allows live streaming of highly compressed audio to http://icecast.awi.de. While the recording device is capable of four channels / 192kHz / 24Bit, this cannot be transferred continuously through the WLAN and satellite link, so a multi-stage automatic buffering and selection scheme is implemented to allow for both, continuous coverage and high quality recordings of selected events. About two terabytes per year are automatically streamed into the petabyte data storage of the World Data Centre for Marine Environmental Data, hosted by AWI, where they are analysed and prepared for publishing under an open access license.

Detection and classification of call types in the vocalizations of north‐east pacific blue whales

Characterization of marine mammal vocalizations is an essential part of any program of marine mammal monitoring as well as being of great help for furthering understanding of subjects such as underwater communication, sonar, etc. The vocalizations of the North‐East Pacific (NEPAC) blue whales are known to be made of at least three different call types: the A call, the B call and the C call. This study aims at the development of a wholly automatic process of detection and classification for the two most common call types of the NEPAC population which are the A call and the B call. We created one template for the A call and one for the B call in order to extract features with matched filtering operations. We show that a simple Gaussian Mixture Model classifier can be used to accurately track and identify the call types in 24‐hour long records. The proposed methodology is applied to real data sets recorded by seismic sensors.

An opportunistic passive acoustics study of the spatial and temporal distribution and vocal behavior of Blainville's beaked whale (“Mesoplodon densirostris”) in the presence of mid‐frequency active sonar

The effect of mid‐frequency active sonar, has increasingly become an issue with navies worldwide. The U.S. navy ranges have been used to develop passive acoustic algorithms and tools to detect, classify, and localize marine mammal vocalizations which have been applied to an opportunistic passive acoustic study of Blainville's beaked whales. Based on Woods Hole Oceanographic Digital Tags (Dtags), these animals are known to produce echo‐location clicks only during deep foraging dives. Using passive acoustics detection of vocalizations, foraging groups of animals were isolated and the duration of vocalizations was used as a measure of foraging behavior. The animals' vocal behavior and spatial and temporal distribution were characterized during periods with no active sonar on range. These results are compared to those derived from opportunistic data obtained during multi‐ship active mid‐frequency sonar operations.

Visual and Passive Acoustic Marine Mammal Observations and High-Frequency Seismic Source Characteristics Recorded During a Seismic Survey

In this paper, we present marine mammal observation statistics, high-frequency seismic source characteristics, and example denoising of marine mammal acoustical recordings using data collected during the mitigation and monitoring program for a 3-D seismic survey by EnCana Corporation, Calgary, AB, Canada, in the Northwest Atlantic during 2003. Marine mammals were observed both visually and acoustically. No marine mammal incidents or adverse reactions were observed during the survey. Acoustical observations were made by the Sea map Passive Acoustic Cetacean Monitoring System (SPACMS), consisting of two hydrophones placed 50 m apart, towed ahead of and to one side of the seismic source. Visual and acoustical detections were uncorrelated, indicating the complementary nature of the two observational techniques. Visual detections were more common per hour of effort than acoustical detections. Acoustical detection rates showed no significant day-night difference. Marine mammals appeared to have avoided very close ranges (100 m) from the seismic array during seismic acquisition, but the overall number of marine mammals in the observable radius (1-2 km) did not change significantly when the seismic source was ldquoonrdquo compared to ldquooff.rdquo Marine mammals were observed in larger groups and appeared to have become less vocal when the seismic source was active. It should be noted however, that the results from this data gathering effort may be affected by potential sources of bias (such as the combination of data from toothed and baleen whales). Signal processing of seismic source signatures indicated some high-frequency energy content consistent with expectations from earlier work. This analysis confirmed that most of the seismic energy was concentrated at lower frequencies (500 Hz). No low-frequency comparisons with near-field data could be made due to the geometry of the SPACMS recording hydrophones and seismic source, which resulted in the Lloyd's mirror effect obliterating low-frequency components in the SPACMS records. A wavelet-based denoising method was applied to improve the visibility of marine mammal vocalizations on a spectrogram display.

FLIP (Floating Instrument Platform): A major Fred Fisher contribution to ocean science

Frederick H. Fisher, a loyal and zealous member of the Acoustical Society of America, was an imaginative and effective developer of new techniques for research in both laboratory and seagoing acoustics. Most notable among his contributions was his work in bringing into being and enhancing the usefulness of the spar buoy laboratory, FLIP, from its inception in 1960. Not only did Fred use FLIP in his own research, its existence and many of its ancillary capabilities constituted a base for the seagoing research of others. The authors of this paper have benefited from FLIP’s unique capabilities, starting with long-range sound propagation studies in the 1960’s and 1970’s. FLIP’s stability and deep draft structure provided the platform for development of acoustic Doppler techniques for the measurement of ocean currents. Most recently, FLIP has been involved in studies of marine mammal vocalizations and use of multielement arrays to investigate details of shallow-water propagation. Fred’s initial studies of sonar bearing accuracy, for which FLIP’s construction was funded, and his dedication to advancing FLIP’s ability to contribute to ocean science, constitute a legacy that is being utilized today, more than 40 years after FLIP’s launching.

Call detection and extraction using Bayesian inference

Marine mammal vocalizations have always presented an intriguing topic for researchers not only because they provide an insight on their interaction, but also because they are a way for scientists to extract information on their location, number and various other parameters needed for their monitoring and tracking. In the past years field researchers have used submersible microphones to record underwater sounds in the hopes of being able to understand and label marine life. One of the emerging problems for both on site and off site researchers is the ability to detect and extract marine mammal vocalizations automatically and in real time given the copious amounts of existing recordings. In this paper, we focus on signal types that have a well-defined single frequency maxima and offer a method based on Sine wave modeling and Bayesian inference that will automatically detect and extract such possible vocalizations belonging to marine mammals while minimizing human interference. The procedure presented in this paper is based on global characteristics of these calls thus rendering it a species independent call detector/extractor.

Passive acoustic detection and localization of sperm whales ( Physeter macrocephalus) in the tongue of the ocean

A set of algorithms for real-time detection and localization of vocalizing marine mammals has been developed as part of the Marine Mammal Monitoring on Navy Ranges (M3R) program. These algorithms work on a broad variety of vocalizations including sperm whale clicks. The detection algorithm is a two stage process utilizing a binary thresholded FFT as the first stage. The second stage examines the FFT output to determine whether a click is present in a given FFT window. Detected clicks are split out of the data stream and sent to a data association algorithm called a scanning sieve. Time differences of arrival (TDOAs) are calculated which are then fed into 2D and 3D hyperbolic localization algorithms. Software written to implement the algorithms was used to process a data set consisting of sperm whale vocalizations provided as part of the 2nd International Workshop on Detection and Localization of Marine Mammals using Passive Acoustics. Real-time detection and localization results from the data set are provided, along with a detailed description of the algorithms.

Acoustic analysis for the Haro Strait on 5 May 2003

On 5 May 2003, USS SHOUP conducted sonar training for 4 h while transiting through the Strait of Juan de Fuca and the Haro Strait. During the time period when the sonar was configured for long-range operations, the transmissions were alleged by some observers to have influenced the behavior of the resident J-pod of orcas. Additionally, the transmissions were received and recorded on various hydrophones deployed for monitoring marine mammal vocalizations. The U.S.-Navy-commissioned detailed reconstruction of the event that relates the location of SHOUP, the J-pod, and the monitoring hydrophones during the time period of SHOUP’s long-range sonar operations is presented. By combining high-fidelity predictions of the acoustic field with the in situ acoustic recordings, the reconstruction provides a detailed, moment-to-moment picture of the worst case estimates of the sonar sound levels experienced by the J-pod. [Work supported by the U.S. Navy Pacific Fleet.]

Detection and classification of right whales in the Bay of Fundy using independent component analysis

A novel method of the detection and classification for marine mammals is presented which uses techniques from independent component analysis to solve the blind source separation problem for right whales in the Bay of Fundy. Using the fundamentally non-Gaussian nature of marine mammal vocalizations and data collected on multiple hydrophones, we are able to separate right whale source spectra, up to an unknown scale, from ambient noise. This technique assumes that the array data is a linear combination of the source signals but does not require specific knowledge of the array geometry. A detector/classifier algorithm is demonstrated which compares the estimated source spectra against known right whale vocalizations.

An autonomous acoustic recorder using a directional sensor for locating calling bowhead whales

A new directional acoustic recorder has been developed to assess the movement and distribution of vocalizing marine mammals. Called DASAR (Directional Autonomous Seafloor Acoustic Recorder), the instrument incorporates an omnidirectional and two orthogonal horizontal sensors from a DIFAR (DIrectional Frequency and Recording) sonobuoy, as well as a magnetic compass. With appropriate processing, the horizontal directions can be computed for received sounds. The acoustic bandwidth is 1000 Hz, appropriate for bowhead calls, and recording is continuous for 45 days. DASARs were deployed in a hexagonal array seaward of an artificial gravel island recently constructed for oil and gas production northwest of Prudhoe Bay, Alaska. Operation was during the fall migration of bowhead whales past the island. The objective was to relate calling-whale locations to the character and levels of sounds emanating from the island and from nearby tugs and barges. [Work supported by BP Exploration (Alaska).]

Techniques for passive detection and localization of marine mammals using wide aperture arrays

SACLANTCEN is sponsoring a program entitled Sound Oceanography and Living Marine Resources Program, whose objective is to demonstrate dual use technology for passive detection and localization of marine mammals, which will assist in the implementation of acoustic risk mitigation policies when using high-power acoustic sources. A sea trial was conducted in August 1999 to evaluate marine mammal detection and localization techniques using a number of advanced passive arrays. Passive, bearing-only data from marine mammal vocalizations received on a towed horizontal line array and target motion analysis are used to localize whale position. The multipath arrival structure of received signals and a ray trace model are used to determine whale range and depth over time. Data from several stations, executed using three deployed vertical line arrays in the vicinity of diving whales, and triangulation of arrivals are used to track whale location in the horizontal plane, while beamforming of the vertical array data is used to track whale depth. Evaluation of these data and techniques will be used to formulate passive procedures for determining the presence of marine mammals prior to activation of high-power acoustic sources.

Stochastic dynamical modeling of marine mammal vocalization waveforms

Stochastic models have been developed for approximating acoustic waveforms of marine animal sounds. The models provide a general method for replicating the time-frequency features and variations of vocalization waveforms based on limited databases of recorded samples. For each sample waveform, a linear-system model is constructed, and the system parameters are determined so that the impulse response of the system optimally fits the sampled waveform. Then, the statistical variability of the system parameters within the ensemble of recorded samples is characterized, and procedures for generating new waveforms based on those statistical distributions are defined. The linear-system models generate waveforms that are similar to the recorded samples and interpolate among the feature variations of those samples. In addition, the models suggest possible physical mechanisms of sound generation by marine animals. The approach is then extended to provide for representation and simulation of extended time series consisting of patterned repetitions of marine animal sounds, e.g., click bursts and codas of dolphins and whales. Applications to modeling of underwater noise generated by marine animals are presented. [Work supported by the Office of Naval Research.]

Matched-field processing of baleen whale vocalizations using calibrated environmental models

During the 1997 Marine Mammal Vocalization (MMV) experiment conducted by the Scripps Marine Physical Laboratory last September, J15 source tows were conducted in a highly range-dependent environment northwest of San Nicolas Island in the Southern California Bight region. The tows consisted of eight radial tracks ranging from 1 to 6 nautical miles from the research platform FLIP, along all cardinal directions. Among the signals broadcast were low-frequency combs (18–110 Hz), middle-frequency combs (50–200 Hz), FM downsweeps (900 to 50 Hz over 9 s), and an Eastern Pacific type A blue whale recorded during the 1996 MMV experiment. Multiple CTD casts were also taken, and a bathymetric database was obtained from the National Oceanographic Service. All these data are used to calibrate matched-field processing environmental models, which are subsequently applied to find the location in 3D space of a series of blue whale vocalizations. [Work sponsored by ONR.]

Removal of multipath effects from marine mammal vocalizations.

Long‐duration (≳1 s) marine mammal vocalizations (i.e., blue whale calls) generate echos and other multipath effects [McDonald et al., J. Acoust. Soc. Am. 98, 712–721 (1995)], obscuring the original structure of the call, which may contain valuable anatomical information about the animal. Several deconvolution techniques are evaluated for their effectiveness in removing multipath effects from both a hypothetical and actual marine mammal call. The evaluation is first performed on a model of a whale call in an isovelocity waveguide with noise, where the source signal is known and can be compared with the output of the deconvolution schemes. The methods are then applied to two blue whale calls recorded during an experiment (SWellEx‐3) in July, 1994 in 200‐m water. The depth and range of the animal at those two times had previously been determined by matched field processing. As the position of the animal is known, the resulting multipath propagation can be computed and the accuracy of the techniques in identifying those multipaths determined.