Low-frequency whale and seismic airgun sounds recorded in the mid-Atlantic Ocean

Beginning in February 1999, an array of six autonomous hydrophones was moored near the Mid-Atlantic Ridge (35 degrees N-15 degrees N, 50 degrees W-33 degrees W). Two years of data were reviewed for whale vocalizations by visually examining spectrograms. Four distinct sounds were detected that are believed to be of biological origin: (1) a two-part low-frequency moan at roughly 18 Hz lasting 25 s which has previously been attributed to blue whales (Balaenoptera musculus); (2) series of short pulses approximately 18 s apart centered at 22 Hz, which are likely produced by fin whales (B. physalus); (3) series of short, pulsive sounds at 30 Hz and above and approximately 1 s apart that resemble sounds attributed to minke whales (B. acutorostrata); and (4) downswept, pulsive sounds above 30 Hz that are likely from baleen whales. Vocalizations were detected most often in the winter, and blue- and fin whale sounds were detected most often on the northern hydrophones. Sounds from seismic airguns were recorded frequently, particularly during summer, from locations over 3000 km from this array. Whales were detected by these hydrophones despite its location in a very remote part of the Atlantic Ocean that has traditionally been difficult to survey.     

Comparison of beaked whale detection algorithms

Due to recent advances in passive acoustic monitoring techniques, beaked whales are now more effectively detected acoustically than visually during vessel-based (e.g. line-transect) surveys. Beaked whales signals can be discriminated from those of other cetaceans by the unique characteristics of their echolocation clicks (e.g. duration >175 μs, center frequencies between 30 and 40 kHz, inter-click intervals between 0.2 and 0.4 s and frequency upsweeps). Furthermore, these same characteristics make these signals ideal candidates for testing automated detection and classification algorithms. There are several different beaked whale automated detectors currently available for use. However, no comparative analysis of detectors exists. Therefore, comparison between studies and datasets is difficult. The purpose of this study was to test, validate, and compare algorithms for detection of beaked whales in acoustic line-transect survey data. Six different detection algorithms (XBAT, Ishmael, PAMGUARD, ERMA, GMM and FMCD) were evaluated and compared. Detection trials were run on three sample days of towed-hydrophone array recordings collected by NOAA Southwest Fisheries Science Center (SWFSC) during which were confirmed visual sightings of beaked whales (Ziphius cavirostris and Mesoplodon peruvianus). Detections also were compared to human verified acoustic detections for a subset of these data. In order to measure the probabilities of false detection, each detector was also run on three sample recordings containing clicks from another species: Risso’s dolphin (Grampus griseus). Qualitative and quantitative comparisons and the detection performance of the different algorithms are discussed.     

Low-frequency whale sounds recorded on hydrophones moored in the eastern tropical Pacific

An array of autonomous hydrophones moored in the eastern tropical Pacific was monitored for one year to examine the occurrence of whale calls in this region. Six hydrophones which recorded from 0-40 Hz were placed at 8 degrees N, 0 degree, and 8 degrees S along longitudes 95 degrees W and 110 degrees W. Seven types of sounds believed to be produced by large whales were detected. These sound types were categorized as either moan-type (4) or pulse-type (3) calls. Three of the moan-type calls, and probably the fourth, may be attributed to blue whales. The source(s) of the remaining calls is unknown. All of the call types studied showed seasonal and geographical variation. There appeared to be segregation between northern and southern hemispheres, such that call types were recorded primarily on the northern hydrophones in the northern winter and others recorded primarily on the southern hemisphere hydrophones in the southern winter. More calls and more call types were recorded on the eastern hydrophones than on the western hydrophones.     

Classification of broadband echoes from prey of a foraging Blainville's beaked whale

Blainville's beaked whales (Mesoplodon densirostris) use broadband, ultrasonic echolocation signals with a -10 dB bandwidth from 26 to 51 kHz to search for, localize, and approach prey that generally consist of mid-water and deep-water fishes and squid. Although it is well known that the spectral characteristics of broadband echoes from marine organisms vary as a function of size, shape, orientation, and anatomical group, there is little evidence as to whether or not free-ranging toothed whales use spectral cues in discriminating between prey and nonprey. In order to study the prey-classification process, a stereo acoustic tag was deployed on a Blainville's beaked whale so that emitted clicks and the corresponding echoes from targets in the water could be recorded. A comparison of echoes from targets apparently selected by the whale and those from a sample of scatterers that were not selected suggests that spectral features of the echoes, target strengths, or both may have been used by the whale to discriminate between echoes. Specifically, the whale appears to favor targets with one or more nulls in the echo spectra and to seek prey with higher target strengths at deeper depths.     

Blue whale (Balaenoptera musculus) sounds from the North Atlantic

Sounds of blue whales were recorded from U.S. Navy hydrophone arrays in the North Atlantic. The most common signals were long, patterned sequences of very-low-frequency sounds in the 15-20 Hz band. Sounds within a sequence were hierarchically organized into phrases consisting of one or two different sound types. Sequences were typically composed of two-part phrases repeated every 73 s: a constant-frequency tonal "A" part lasting approximately 8 s, followed 5 s later by a frequency-modulated "B" part lasting approximately 11 s. A common sequence variant consisted only of repetitions of part A. Sequences were separated by silent periods averaging just over four minutes. Two other sound types are described: a 2-5 s tone at 9 Hz, and a 5-7 s inflected tone that swept up in frequency to ca. 70 Hz and then rapidly down to 25 Hz. The general characteristics of repeated sequences of simple combinations of long-duration, very-low-frequency sound units repeated every 1-2 min are typical of blue whale sounds recorded in other parts of the world. However, the specific frequency, duration, and repetition interval features of these North Atlantic sounds are different than those reported from other regions, lending further support to the notion that geographically separate blue whale populations have distinctive acoustic displays.     

Recognizing transient low-frequency whale sounds by spectrogram correlation

A method is described for the automatic recognition of transient animal sounds. Automatic recognition can be used in wild animal research, including studies of behavior, population, and impact of anthropogenic noise. The method described here, spectrogram correlation, is well-suited to recognition of animal sounds consisting of tones and frequency sweeps. For a sound type of interest, a two-dimensional synthetic kernel is constructed and cross-correlated with a spectrogram of a recording, producing a recognition function--the likelihood at each point in time that the sound type was present. A threshold is applied to this function to obtain discrete detection events, instants at which the sound type of interest was likely to be present. An extension of this method handles the temporal variation commonly present in animal sounds. Spectrogram correlation was compared to three other methods that have been used for automatic call recognition: matched filters, neural networks, and hidden Markov models. The test data set consisted of bowhead whale (Balaena mysticetus) end notes from songs recorded in Alaska in 1986 and 1988. The method had a success rate of about 97.5% on this problem, and the comparison indicated that it could be especially useful for detecting a call type when relatively few (5-200) instances of the call type are known.     

Compression of Cuvier’s beaked whale (Ziphius cavirostris) vocalizations using fast orthogonal search

Remote sensing of beaked whale vocalizations poses significant problems due to limited communications bandwidths. Many beaked whales vocalize (click) at frequencies up to 50 kHz Hence high bandwidth sampling (typically 100+ kHz) and processing is required in order to detect the clicks, but transmitting the data from a remote sensor using a low-bandwidth (4800 baud) satellite link results in a real-time bottleneck. Even if auto-detection algorithms were used on the remote sensor, some data would need to be relayed to a human operator to verify the classification. Hence, the ability to compress the data in a manner that does not impede the ability to detect and classify the transient signal is required. Typical audio compression techniques have a maximum sampling rate of 48 kHz which is too low to collect beaked whale clicks and still obey the Nyquist criterion. In addition, audio compression algorithms also have a psycho-acoustic model that aids in the compression of the signal but distorts the audio signal.This paper presents a compression algorithm that uses a non-linear modelling technique called fast orthogonal search (FOS) to create a functional expansion of the acoustic data. The candidate functions used in the functional expansion are transient signals that model Cuvier’s beaked whale (Ziphius cavirostris) clicks as well as sinusoidal functions for modelling whale songs. A compression ratio of 93 is achieved by transmitting candidate identification numbers and weights for only the candidate functions that are chosen by the FOS algorithm. The acoustic signal is recreated using the weights and candidate numbers transmitted. The reconstructed time series is used as an input to a band-limited energy detector for whale vocalizations. The raw data and the reconstructed data have comparable probability of detection and missed detections, with fewer false alarms for the reconstructed signal.     

Classification and identification of recorded and synthesized impact sounds by practiced listeners, musicians, and nonmusicians

Three experiments were conducted to test the viability of a low-parameter modal model for synthesizing impact sounds to be used in commercial and psychoacoustic research. The model was constrained to have four physically based parameters dictating the amplitude, frequency, and decay of modes. The values of these parameters were selected by ear to roughly match the recordings of ten different resonant objects suspended by hand and struck with different mallets. In experiment 1, neither 35 professional musicians nor 187 college undergraduates could identify which of the two matched sounds was the real recording with better than chance accuracy, though significantly better than chance performance was obtained when modal parameters were selected without the previously imposed physical constraints. In experiment 2, the undergraduates identified the source corresponding to the recorded and synthesized sounds with the same level of accuracy and largely the same pattern of errors. Finally, experiment 3 showed highly practiced listeners to be largely insensitive to changes in the acoustic waveform resulting from an increase in the number of free parameters used in the modal model beyond 3. The results suggest that low-parameter, modal models might be exploited meaningfully in many commercial and research applications involving human perception of impact sounds.     

Automated detection and localization of bowhead whale sounds in the presence of seismic airgun surveys

An automated procedure has been developed for detecting and localizing frequency-modulated bowhead whale sounds in the presence of seismic airgun surveys. The procedure was applied to four years of data, collected from over 30 directional autonomous recording packages deployed over a 280 km span of continental shelf in the Alaskan Beaufort Sea. The procedure has six sequential stages that begin by extracting 25-element feature vectors from spectrograms of potential call candidates. Two cascaded neural networks then classify some feature vectors as bowhead calls, and the procedure then matches calls between recorders to triangulate locations. To train the networks, manual analysts flagged 219 471 bowhead call examples from 2008 and 2009. Manual analyses were also used to identify 1.17 million transient signals that were not whale calls. The network output thresholds were adjusted to reject 20% of whale calls in the training data. Validation runs using 2007 and 2010 data found that the procedure missed 30%-40% of manually detected calls. Furthermore, 20%-40% of the sounds flagged as calls are not present in the manual analyses; however, these extra detections incorporate legitimate whale calls overlooked by human analysts. Both manual and automated methods produce similar spatial and temporal call distributions.     

Vocalizations produced by humpback whale (Megaptera novaeangliae) calves recorded in Hawaii

Although humpback whale (Megaptera novaeangliae) calves are reported to vocalize, this has not been measurably verified. During March 2006, an underwater video camera and two-element hydrophone array were used to record nonsong vocalizations from a mother-calf escort off Hawaii. Acoustic data were analyzed; measured time delays between hydrophones provided bearings to 21 distinct vocalizations produced by the male calf. Signals were pulsed (71%), frequency modulated (19%), or amplitude modulated (10%). They were of simple structure, low frequency (mean=220 Hz), brief duration (mean=170 ms), and relatively narrow bandwidth (mean=2 kHz). The calf produced three series of "grunts" when approaching the diver. During winters of the years 2001-2005 in Hawaii, nonsong vocalizations were recorded in 109 (65%) of 169 groups with a calf using an underwater video and single (omnidirectional) hydrophone. Nonsong vocalizations were most common (34 of 39) in lone mother-calf pairs. A subsample from this dataset of 60 signals assessed to be vocalizations provided strong evidence that 10 male and 18 female calves vocalized based on statistical similarity to the 21 verified calf signals, proximity to an isolated calf (27 of 28 calves), strong signal-to-noise ratio, and/or bubble emissions coincident to sound.     

Time patterns of sperm whale codas recorded in the Mediterranean Sea 1985-1996

A distinctive vocalization of the sperm whale, Physeter macrocephalus (=P. catodon), is the coda: a short click sequence with a distinctive stereotyped time pattern [Watkins and Schevill, J. Acoust. Soc. Am. 62, 1485-1490 (1977)]. Coda repertoires have been found to vary both geographically and with group affiliation [Weilgart and Whitehead, Behav. Ecol. Sociobiol. 40, 277-285 (1997)]. In this work, the click timings and repetition patterns of sperm whale codas recorded in the Mediterranean Sea are characterized statistically, and the context in which the codas occurred are also taken into consideration. A total of 138 codas were recorded in the central Mediterranean in the years 1985-1996 by several research groups using a number of different detection instruments, including stationary and towed hydrophones, sonobuoys and passive sonars. Nearly all (134) of the recorded codas share the same "3+1" (/// /) click pattern. Coda durations ranged from 456 to 1280 ms, with an average duration of 908 ms and a standard deviation of 176 ms. Most of the codas (a total of 117) belonged to 20 coda series. Each series was produced by an individual, in most cases by a mature male in a small group, and consisted of between 2 and 16 codas, emitted in one or more "bursts" of 1 to 13 codas spaced fairly regularly in time. The mean number of codas in a burst was 3.46, and the standard deviation was 2.65. The time interval ratios within a coda are parameterized by the coda duration and by the first two interclick intervals normalized by coda duration. These three parameters remained highly stable within each coda series, with coefficients of variation within the series averaging less than 5%. The interval ratios varied somewhat across the data sets, but were highly stable over 8 of the 11 data sets, which span 11 years and widely dispersed geographic locations. Somewhat different interval ratios were observed in the other three data sets; in one of these data sets, the variant codas were produced by a young whale. Two sets of presumed sperm whale codas recorded in 1996 had 5- and 6-click patterns; the observation of these new patterns suggests that sperm whale codas in the Mediterranean may have more variations than previously believed.     

Simulating the effect of high-intensity sound on cetaceans: modeling approach and a case study for Cuvier's beaked whale (Ziphius cavirostris)

A finite element model is formulated to study the steady-state vibration response of the anatomy of a whale (Cetacea) submerged in seawater. The anatomy was reconstructed from a combination of two-dimensional (2D) computed tomography (CT) scan images, identification of Hounsfield units with tissue types, and mapping of mechanical properties. A partial differential equation model describes the motion of the tissues within a Lagrangean framework. The computational model was applied to the study of the response of the tissues within the head of a neonate Cuvier's beaked whale Ziphius cavirostris. The characteristics of the sound stimulus was a continuous wave excitation at 3500 Hz and 180 dB re: 1 microPa received level, incident as a plane wave. We model the beaked whale tissues embedded within a volume of seawater. To account for the finite dimensions of the computational volume, we increased the damping for viscous shear stresses within the water volume, in an attempt to reduce the contribution of waves reflected from the boundaries of the computational box. The mechanical response of the tissues was simulated including: strain amplitude; dissipated power; and pressure. The tissues are not likely to suffer direct mechanical or thermal damage, within the range of parameters tested.     

Characteristics of biosonar signals from the northern bottlenose whale, Hyperoodon ampullatus

The biosonar pulses from free-ranging northern bottlenose whales (Hyperoodon ampullatus) were recorded with a linear hydrophone array. Signals fulfilling criteria for being recorded close to the acoustic axis of the animal (a total of 10 clicks) had a frequency upsweep from 20 to 55 kHz and durations of 207 to 377 μs (measured as the time interval containing 95% of the signal energy). The source level of these signals, denoted pulses, was 175-202 dB re 1 μPa rms at 1 m. The pulses had a directionality index of at least 18 dB. Interpulse intervals ranged from 73 to 949 ms (N?=?856). Signals of higher repetition rates had interclick intervals of 5.8-13.1 ms (two sequences, made up of 59 and 410 clicks, respectively). These signals, denoted clicks, had a shorter duration (43-200 μs) and did not have the frequency upsweep characterizing the pulses of low repetition rates. The data show that the northern bottlenose whale emits signals similar to three other species of beaked whale. These signals are distinct from the three other types of biosonar signals of toothed whales. It remains unclear why the signals show this grouping, and what consequences it has on echolocation performance.     

Detection of whale calls in noise: performance comparison between a beluga whale, human listeners, and a neural network

This article examines the masking by anthropogenic noise of beluga whale calls. Results from human masking experiments and a software backpropagation neural network are compared to the performance of a trained beluga whale. The goal was to find an accurate, reliable, and fast model to replace lengthy and expensive animal experiments. A beluga call was masked by three types of noise, an icebreaker's bubbler system and propeller noise, and ambient arctic ice-cracking noise. Both the human experiment and the neural network successfully modeled the beluga data in the sense that they classified the noises in the same order from strongest to weakest masking as the whale and with similar call-detection thresholds. The neural network slightly outperformed the humans. Both models were then used to predict the masking of a fourth type of noise, Gaussian white noise. Their prediction ability was judged by returning to the aquarium to measure masked-hearing thresholds of a beluga in white noise. Both models and the whale identified bubbler noise as the strongest masker, followed by ramming, then white noise. Natural ice-cracking noise masked the least. However, the humans and the neural network slightly overpredicted the amount of masking for white noise. This is neglecting individual variation in belugas, because only one animal could be trained. Comparing the human model to the neural network model, the latter has the advantage of objectivity, reproducibility of results, and efficiency, particularly if the interference of a large number of signals and noise is to be examined.