Whale Calls Research Articles

PART II: Whale Songs of Wars Not Yet Waged: The Demise of Natural-Born Killers through Human-Machine Teamings Yet to Come

Current human-machine dynamics in security affairs positions the human operator in the loop with artificial intelligence to conduct decisions and actions. As technological advancements in AI cognition, speed, and weapon sophistication increase, human operators are increasingly being shifted to an on the loop where AI takes more responsibility in warfare and defense decisions, whether tactical or even strategic. Human operators are also falling off the loop, trailing enhanced AI systems as the biological and physical limits because humans are not the same for artificial intelligence in narrow applications. Those likely will expand toward general AI in the coming decades, presenting significant strategic, organizational, and even existential concerns. Further, how natural humans respond and engage with increasingly advanced, even superintelligent AI as well as a singularity event will feature disruptive, transformative impacts on security affairs and even at a philosophical level discerning what war is.

Geographic and seasonal variation of blue whale calls in the North Pacific

The call characteristics and distribution of blue whales in the North Pacific were examined by use of acoustic surveys. Two distinct vocalisation types have been previously attributed to blue whales from limited regions in the North Pacific (cf. Thompson and Friedl, 1982; Rivers, 1997). Hydrophone data from sixteen sites in the North Pacific were examined for these blue whale vocalisations. There were distinct geographic and seasonal differences between the occurrence of the two vocalisation types. The hydrophones that were more westerly recorded the ‘northwestern’ Pacific vocalisation, those in the eastern Pacific recorded the ‘northeastern’ Pacific vocalisation and those in the central Pacific recorded both types. Northeastern vocalisations were recorded from July-December in the northeast Pacific and February-May in the eastern tropical Pacific. Northwestern vocalisations were recorded most often from July-December, and were essentially absent from March-May in the northwestern Pacific. These results suggest that the different vocalisation types may represent at least two distinct groups of blue whales in the North Pacific.

Arabian Sea Humpback Whale (Megaptera novaeangliae) Singing Activity off Netrani Island, India

Arabian Sea humpback whales (Megaptera novaeangliae), listed as “Endangered” on the International Union for Conservation of Nature’s Red List, remain resident throughout the year in the waters of the Arabian Sea and constitute a genetically isolated population. In the eastern Arabian Sea, information on humpback whales off the Indian coast has largely been limited to stranding records, local ecological knowledge, and opportunistic visual sighting data. These data, along with information from a long-term study off Oman, suggest that humpback whales migrate across the Arabian Sea into Indian territorial waters from October to March. To study the presence of Arabian Sea humpback whales in Indian waters more comprehensively, passive acoustic monitoring (PAM) was initiated along the west coast of India in 2019. Male humpback whales produce complex songs with a stereotyped structure; these songs are shared within a population, and song patterns are known to evolve progressively over time. In this article, a structural analysis of humpback whale song recorded over four days off the coast of Netrani Island, Karnataka, India, in December 2019 is presented. Time-frequency features of 2,641 individual call units were analysed. Call units had a fundamental frequency bandwidth ranging from 149.98 to 541.65 Hz, with a duration ranging from 1.19 to 5.5 s. The call units were used to identify phrases and themes required to construct the structure of the song, which can potentially help identify the population to which singing individuals belong. This study indicates the need for a long-term PAM program across the Arabian Sea to compare whale songs across the region. Simultaneous recordings over multiple seasons will best assess population connectivity, seasonal occurrence, and movement patterns within and between populations across the Arabian Sea and the Indian Ocean.

An HMM-DNN-Based System for the Detection and Classification of Low-Frequency Acoustic Signals from Baleen Whales, Earthquakes, and Air Guns off Chile

Marine passive acoustic monitoring can be used to study biological, geophysical, and anthropogenic phenomena in the ocean. The wide range of characteristics from geophysical, biological, and anthropogenic sounds sources makes the simultaneous automatic detection and classification of these sounds a significant challenge. Here, we propose a single Hidden Markov Model-based system with a Deep Neural Network (HMM-DNN) for the detection and classification of low-frequency biological (baleen whales), geophysical (earthquakes), and anthropogenic (air guns) sounds. Acoustic data were obtained from the Preparatory Commission for the Comprehensive Nuclear-Test-Ban Treaty Organization station off Juan Fernandez, Chile (station HA03) and annotated by an analyst (498 h of audio data containing 30,873 events from 19 different classes), and then divided into training (60%), testing (20%), and tuning (20%) subsets. Each audio frame was represented as an observation vector obtained through a filterbank-based spectral feature extraction procedure. The HMM-DNN training procedure was carried out discriminatively by setting HMM states as targets. A model with Gaussian Mixtures Models and HMM (HMM-GMM) was trained to obtain an initial set of HMM target states. Feature transformation based on Linear Discriminant Analysis and Maximum Likelihood Linear Transform was also incorporated. The HMM-DNN system displayed good capacity for correctly detecting and classifying events, with high event-level accuracy (84.46%), high weighted average sensitivity (84.46%), and high weighted average precision (89.54%). Event-level accuracy increased with higher event signal-to-noise ratios. Event-level metrics per class also showed that our HMM-DNN system generalized well for most classes but performances were best for classes that either had a high number of training exemplars (e.g., generally above 50) and/or were for classes of signals that had low variability in spectral features, duration, and energy levels. Fin whale and Antarctic blue whale song and air guns performed particularly well.

Calls recorded from North Pacific right whales (Eubalaena japonica) in the eastern Bering Sea

Calls from North Pacific right whales (Eubalaena japonica) were recorded in the eastern Bering Sea during a visual and acoustic survey aboard the US Coast Guard buoy tender Sweetbrier, in July 1999. Calls were commonly detected to 20km, and in one case approximately 30km, via deployment of arrays of directional sonobuoys. Acoustic detections (clusters of right whale calls separated by time and location) numbered 26, but only five right whales were seen. Only one right whale produced calls while under visual observation. The types of calls recorded from North Pacific right whales were similar in duration and frequency to calls recorded from right whales in the South Atlantic. The predominant call type (85%; 436 of 511 calls) was the ‘up’ call, a signal sweeping from about 90Hz to 150Hz in 0.7s. Two call types are described as ‘down’ and ‘constant’ calls, based upon nomenclature established for southern right whales (Eubalaena australis). One call type, the ‘down-up’ was unique to the North Pacific repertoire. Right whales commonly produced calls in series lasting several minutes and then became silent for an hour or more, with some animals not calling for periods of at least four hours. Other cetaceans detected acoustically by ‘random’ sonobuoy deployments during the cruise included fin whales (19 times), killer whales (3 times) and sperm whales (once).

Simultaneous tracking of multiple whales using two fiber-optic cables in the Arctic

Climate change is impacting the Arctic faster than anywhere else in the world. As a response, ecosystems are rapidly changing. As a result, we can expect rapid shifts in whale migration and habitat use concurrent with changes in human patterns. In this context, responsible management and conservation requires improved monitoring of whale presence and movement over large ranges, at fine scales and in near-real-time compared to legacy tools. We demonstrate that this could be enabled by Distributed Acoustic Sensing (DAS). DAS converts an existing fiber optic telecommunication cable into a widespread, densely sampled acoustic sensing array capable of recording low-frequency whale vocalizations. This work proposes and compares two independent methods to estimate whale positions and tracks; a brute-force grid search and a Bayesian filter. The methods are applied to data from two 260 km long, nearly parallel telecommunication cables offshore Svalbard, Norway. First, our two methods are validated using a dedicated active air gun experiment, from which we deduce that the localization errors of both methods are 100 m. Then, using fin whale songs, we demonstrate the methods' capability to estimate the positions and tracks of eight fin whales over a period of five hours along a cable section between 40 and 95 km from the interrogator unit, constrained by increasing noise with range, variability in the coupling of the cable to the sea floor and water depths. The methods produce similar and consistent tracks, where the main difference arises from the Bayesian filter incorporating knowledge of previously estimated locations, inferring information on speed, and heading. This work demonstrates the simultaneous localization of several whales over a 800 km area, with a relatively low infrastructural investment. This approach could promptly inform management and stakeholders of whale presence and movement and be used to mitigate negative human-whale interaction.

Variability in humpback whale songs reveals how individuals can be distinctive when sharing a complex vocal display.

Individually distinctive acoustic signals in animal vocal communication are taxonomically widespread, however, the investigation of these signal types in marine mammals has focused only on a few species. Humpback whale songs are a stereotyped, hierarchically structured vocal display performed by males, and hence thought to be sexually selected. Within a population, whales conform to a common version of the song despite the song constantly evolving. While humpback songs have been studied extensively at the population level, individual level variation has been rarely described, with inconclusive results. Here, we quantified inter- and intra-individual variability at different levels in the song hierarchy using songs from 25 singers across two song types from the eastern Australian population song of 2002 (12 singers), and the revolutionary song introduced in 2003 (13 singers). Inter-individual variability was found heterogeneously across all hierarchical levels of the song structure. In addition, distinct and individually specific patterns of song production were consistently recorded across song levels, with clear structural differences between the two song types. These results suggest that within the constraints of song conformity, males can produce individually distinctive patterns that could function as an advertisement to females to convey individual qualities.

Humpback whale song variations in La Reunion Island from 2016 to 2018 and in Madagascar in 2018

AbstractMale Megaptera novaeangliae produce complex and structured songs which are shared at the population level. Song patterns are culturally transmitted and evolve progressively through time, both over the breeding season and among years. The songs also undergo periods of relatively rapid change, termed “revolutions.” Acoustic monitoring was conducted from 2016 to 2018 in Reunion and throughout 2018 in Madagascar to assess spatiotemporal variation in song structures and population connectivity. A total of 46 high‐quality song samples were selected, representing 2,760 min of recordings in Reunion. In Madagascar, 12 samples representing 240 min of recordings were analyzed. Analysis of songs revealed 11 phrases and their variants. Low levels of temporal variations were observed over the breeding season. Songs recorded in June were very similar to those recorded in September. Greater variation was observed between years, and some phrases identified in 2018 may have evolved from phrases recorded in 2017. More variants were described for each phrase type in 2018 compared to 2016. All themes recorded in Reunion were shared with Madagascar, suggesting a high degree of population connectivity during the breeding season.

La Niña conditions influence interannual call detections of pygmy blue whales in the eastern Indian Ocean

Oceans across the globe are warming rapidly and marine ecosystems are changing as a result. However, there is a lack of information regarding how blue whales are responding to these changing environments, especially in the Southern Hemisphere. This is because long term data are needed to determine whether blue whales respond to variability in environmental conditions. Using over 16 years of passive acoustic data recorded at Cape Leeuwin, we investigated whether oceanic environmental drivers are correlated with the migration patterns of eastern Indian Ocean (EIO) pygmy blue whales off Western Australia. To determine which environmental variables may influence migration patterns, we modelled the number of acoustic call detections of EIO pygmy blue whale calls with broad and fine scale environmental variables. We found a positive correlation between total annual whale call detections and El Niño Southern Oscillation (ENSO) cycles and the Indian Ocean Dipole (IOD), with more whale calls detected during La Niña years. We also found that monthly whale call detections correlated with sea surface height around the hydrophone and chlorophyll-a concentration at a prominent blue whale feeding aggregation area (Bonney Upwelling) where whales feed during the summer before migrating up the west Australian coast. At the interannual scale, ENSO had a stronger relationship with call detections than IOD. During La Niña years, up to ten times more EIO pygmy blue whale calls were detected than in neutral or El Niño years. This is likely linked to changes in productivity in the feeding areas of the Great Australian Bight and Indian Ocean. We propose that in lower productivity years whales either skipped migration or altered their habitat use and moved further offshore from the hydrophones and therefore were not detected. The frequency and intensity of ENSO events are predicted to increase with climate change, which is likely to impact the productivity of the areas used by blue whales. These changes in productivity may affect the physical condition and reproductive success of individual whales. A reduction in reproductive success could have a significant impact on blue whale recovery from historical whaling and their ability to adapt to a changing environment.

Occurrence and patterns of fin whale songs reveal alternative migration strategies in Svalbard Islands, Norway

The Arctic marine environment is highly affected by global warming with notable changes in habitat conditions, which have great consequences on migrating species. For example, the timing of their migration can be altered leading to changes in their occurrence in suitable areas, which are critical for their survival. In this study, seven years of acoustic data were analysed in Svalbard Islands from 2014 to 2020, revealing that the occurrence of fin whales (Balaenoptera physalus) happened all year-round. The sea surface temperature recorded reveals conditions which could be favorable for these species to persist until the Polar Night. The occurrence of songs indicated that certain individuals did not undertake the migratory journey through the southern breeding grounds, possibly using the area for mating purposes. The analyses of the Inter-Note-Interval (INI) demonstrated that over the years songs with different patterns were found. This suggests that either the fin whales are able to switch their INI patterns or that populations with different INIs are visiting during the Winter. Therefore, this study unveils the undertaking of an alternative strategy to migration movements, and the possible potential origin of the fin whales overwintering in Svalbard.

A review of beaked whale behaviour and ecology in relation to assessing and mitigating impacts of anthropogenic noise

Little is known about the ecology and behaviour of species within the family Ziphiidae. In this paper, five aspects of beaked whale ecology and behaviour are reviewed in relation to possible anthropogenic impacts upon them: social structure; life history; foraging/diving ecology; form and function of beaked whale sounds; and habitat characteristics. Differences in social structure within and between species may affect how anthropogenic activities affect local populations. Life history parameters may likewise vary within and between species and may influence the extent of and ability to recovery from population level impacts. Foraging and diving ecology determine where beaked whales spend most of their time and therefore, where in the water column they are most likely to encounter anthropogenic activities. The form and function of beaked whale sounds may be important in determining whether and how beaked whales are affected by anthropogenic noise. Finally, habitat characteristics determine whether beaked whales are likely to occur in a specific area where anthropogenic activities are to be undertaken and may also determine exactly how beaked whales are affected by it within a local area. To help fill the gaps in our knowledge of beaked whale behaviour and ecology, available opportunities for data collection must be maximised. This includes greater levels of co-operation between research groups to build up large datasets, the use of platforms of opportunity to study beaked whales in areas where little research has previously been undertaken and maximising the amount of information that can be learned from each possible source of data, such as stranded animals, through co-ordinated national and international research programmes.

A New Acoustical Autonomous Method for Identifying Endangered Whale Calls: A Case Study of Blue Whale and Fin Whale.

In this paper, we study to improve acoustical methods to identify endangered whale calls with emphasis on the blue whale (Balaenoptera musculus) and fin whale (Balaenoptera physalus). A promising method using wavelet scattering transform and deep learning is proposed here to detect/classify the whale calls quite precisely in the increasingly noisy ocean with a small dataset. The performances shown in terms of classification accuracy (>97%) demonstrate the efficiency of the proposed method which outperforms the relevant state-of-the-art methods. In this way, passive acoustic technology can be enhanced to monitor endangered whale calls. Efficient tracking of their numbers, migration paths and habitat become vital to whale conservation by lowering the number of preventable injuries and deaths while making progress in their recovery.

Baleen whales in the Scotia Sea during January and February 2003

Different species of baleen whales display distinct spatial distribution patterns in the Scotia Sea during the austral summer. Passive acoustic and visual surveys for baleen whales were conducted aboard the RRS James Clark Ross in the Scotia Sea and around South Georgia in January and February 2003. Identified calls from four species were recorded during the acoustic survey including southern right (Eubalaena australis), blue (Balaenoptera musculus), fin (B. physalus) and humpback whales (Megaptera novaeangliae). These acoustic data included up calls made by southern right whales, downswept D and tonal calls by blue whales, two possible types of fin whale downswept calls and humpback whale moans and grunts. Visual detections included southern right, fin, humpback and Antarctic minke whales (B. bonaerensis sp.). Most acoustic and visual detections occurred either around South Georgia (southern right and humpback whales) or south of the southern boundary of the Antarctic Circumpolar Current (ACC) and along the outer edge of the ice pack (southern right, blue, humpback and Antarctic minke whales). Fin whales were the exception, being the only species acoustically and visually detected primarily in the central Scotia Sea, along the southern ACC front. In addition to identifiable calls from these species, two types of probable baleen whale calls were detected: 50Hz upswept and pulsing calls. It is proposed that minke whales may produce the pulsing calls, based on their similarities with minke whale calls recorded in the North Atlantic Ocean. There was an overlap between locations of fin whale sightings and recordings and locations of 50Hz upswept calls in the central Scotia Sea, but these calls were most similar to calls attributed to blue whales in other parts of Antarctica. More study is required to determine if baleen whales produce these two call types, and if so, which species. The efficiency of acoustics and visual surveys varied by species, with blue whales being easier to detect using acoustics, Antarctic minke whales being best detected during visual surveys and other species falling in between these two extremes.

Biogeographic characterisation of blue whale song worldwide: using song to identify populations

Blue whale songs provide a measure for characterising worldwide blue whale population structure. These songs are divided into nine regional types, which maintain a stable character. Five of the nine song types have been recorded over time spans greater than 30 years showing no significant change in character. The nine song types can be divided into those containing only simple tonal components (high latitude North Pacific, North Atlantic and Southern Ocean song types), those comprised of complex pulsed units in addition to the tonal components (Pacific Ocean margin song types from California, Chile and New Zealand), and those which have the greatest complexity of all and the longest cycling times (Indian Ocean song types from Sri Lanka, Fremantle and Diego Garcia). We suggest that temporally stable differences in song provide another characteristic for comparison with genetic and morphological data when defining blue whale populations. Furthermore, as Mellinger and Barlow (2003) recommend, when there is a lack of other data or lack of clarity in other data sets, evidence of distinct differences in songs between areas should be used as a provisional hypothesis about population structure when making management decisions. Worldwide study is needed to better understand the various populations and subspecies within species like the blue whale that have large geographic distributions and have both migrating and resident populations.

Automated extraction of baleen whale calls based on the pseudo-Wigner–Ville distribution

Baleen whales produce a wide variety of frequency-modulated calls. Extraction of the time-frequency (TF) structures of these calls forms the basis for many applications, including abundance estimation and species recognition. Typical methods to extract the contours of whale calls from a spectrogram are based on the short-time Fourier transform and are, thus, restricted by a fixed TF resolution. Considering the low-frequency nature of baleen whale calls, this work represents the contours using a pseudo-Wigner-Ville distribution for a higher TF resolution at the cost of introducing cross terms. An adaptive threshold is proposed followed by a modified Gaussian mixture probability hypothesis density filter to extract the contours. Finally, the artificial contours, which are caused by the cross terms, can be removed in post-processing. Simulations were conducted to explore how the signal-to-noise ratio influences the performance of the proposed method. Then, in experiments based on real data, the contours of the calls of three kinds of baleen whales were extracted in a highly accurate manner (with mean deviations of 5.4 and 0.051 Hz from the ground-truth contours at sampling rates of 4000 and 100 Hz, respectively) with a recall of 75% and a precision of 78.5%.

Source separation and arrival angle estimation using a single acoustic vector sensor

Extracting individual time signatures from two angularly-separated sources overlapping in time and bandwidth typically requires coherent array processing on multiple hydrophones, a process whose performance is degraded by limited aperture and spatial aliasing. Vector sensors, which measure both acoustic pressure and particle velocity, can measure an acoustic field’s directionality in a single, compact sensor package. Standard beamforming techniques can be applied to the outputs of a vector sensor to suppress one source’s signal, but its efficacy is limited by the angular insensitivity of the resulting cardioid beam pattern. Here we use fundamental physical principles to demonstrate that measurements on a 2-D vector sensor can completely determine the azimuths, amplitudes, and relative phases of two horizontal plane waves of the same frequency, achieving source separation and direction-of-arrival estimates for both. Interference between the waves creates a reactive intensity whose orientation is perpendicular to the bisector of the two wavenumber vectors. This observation leads to closed-form inversion formulas. Data examples of the technique are illustrated using overlapping humpback whale songs off Hawaii, and on weak airgun and bowhead whale signals embedded within intense directional noise from a nearby drilling site in the Beaufort Sea. [Work sponsored by ONR TFO.]

Regularities in spectral entropy variations within humpback whale song sessions

Consecutive songs produced by humpback whales recorded off the coast of Hawaii show recurrent patterns of acoustic variation. Analyses of spectral entropy, a measure of the uniformity of energy distribution in the frequency domain, revealed that spectral entropy cycles in predictable patterns within song sessions. At time scales spanning tens of minutes, variations in spectral entropy appear to track changes in received intensity levels, which are correlated with song duration and may reflect the diving behavior of singers. Intra-individual variations in spectral entropy suggest that singing humpback whales may also vocally control entropic features of songs, both by gradually modulating the acoustic characteristics of units and by varying the amount of time they spend producing specific patterns of units. Preliminary analyses suggest that variations in spectral entropy within song cycles are indicative of the amount of time and energy singers spend producing constant-frequency (CF) elements versus more frequency-modulated/broadband elements. CF elements of units are generally more difficult to spatially localize, but easier to detect over long distances, suggesting that singing humpback whales may be varying the production of consecutive songs in ways that affect auditory spatial processing.

Adaptive signal adjustments in humpback whale songs

The use of vocal flexibility by humpback whales (Megaptera noavaengliae) often described as “singing” is a highly sophisticated communicative process, possessing a multilayered hierarchical structure across the various acoustical parameters. A deeper spectral analysis of songs made available through Google's Pattern Radio website studied in the month of February 2015 display several adaptive signal adjustments in humpback whale songs. As single singer variations in song properties become more clear, multi-singer variations in song properties remain hypothetical. This raises an important question of incorporating a temporal analysis approach to determine the multi-singer variations in song properties.

I know what you did last winter: Bowhead whale anomalous winter acoustic occurrence patterns in the Beaufort Sea, 2018–2020

Global warming is affecting the Arctic at a faster pace than the rest of the world, causing an urgent need to monitor ecosystems to detect possible climate-related changes. To this end, five passive acoustic datasets were recorded in the southern Amundsen Gulf (eastern Beaufort Sea) from September 2018 to September 2020 and analyzed for Bering-Chukchi-Beaufort (BCB) bowhead whale calls using a combination of automated and manual detections. Results indicate a large variation in bowhead occurrence patterns between the two years. For 2018–2019, we obtained the first evidence of bowheads overwintering in what is typically their summer foraging ground. Examination of the following year’s recordings sheds light on whether this interruption in bowhead annual migration was an anomaly or part of an ongoing phenological shift due to climate change. Time series of remotely sensed sea ice concentration at the study area were considered over the last seven years in interpreting differences in migratory behavior of the whales. Statistical quantification of seasonal patterns and habitat preferences of bowheads, based on the 2018–2019 acoustic data, are presented to provide context to BCB bowhead ecology. Passive acoustic monitoring is an indispensable tool in discerning whale responses to a changing ocean in the Arctic.

North Atlantic right whale calls monitoring: Signal processing meets deep learning

The monitoring of marine mammals, such as North Atlantic Right whales (NARW), via their acoustic signatures, can contribute to the protection of these endangered species. The monitoring via passive acoustics, however, may deteriorate, due to the propagation properties of the underwater channel, whereas these signals are susceptible to noise masking and/or mixing as they travel long distances. A day (July 30th, 2018) of NARW annotated recordings (Provided by DCLDE), including seven imbalanced call types, were explored for detection, classification, and density estimation purposes. The call types involved right whale upcall, gunshot, moan, scream, minke whale pulse train, low-frequency pulse, and low-frequency down sweep. Classifying these imbalanced call types represented a real challenge for classical signal processing approaches and conventional deep learning networks (e.g.,Convolutional Neural Networks). The remedy to these challenges involved data preprocessing (spectrogram cleaning/sifting) and the use of a state-of-the-art deep learning transform, self-supervised, that accounts for the imbalances in the dataset. The preprocessed spectrograms boosted the classification score at the deep learning network. The accuracy results for the training set exceeded 90% while the accuracy scored 81% for the testing set.