Whale Calls Research Articles

Reinforcement learning-based framework for whale rendezvous via autonomous sensing robots.

Rendezvous with sperm whales for biological observations is made challenging by their prolonged dive patterns. Here, we propose an algorithmic framework that codevelops multiagent reinforcement learning-based routing (autonomy module) and synthetic aperture radar-based very high frequency (VHF) signal-based bearing estimation (sensing module) for maximizing rendezvous opportunities of autonomous robots with sperm whales. The sensing module is compatible with low-energy VHF tags commonly used for tracking wildlife. The autonomy module leverages in situ noisy bearing measurements of whale vocalizations, VHF tags, and whale dive behaviors to enable time-critical rendezvous of a robot team with multiple whales in simulation. We conducted experiments at sea in the native habitat of sperm whales using an "engineered whale"-a speedboat equipped with a VHF-emitting tag, emulating five distinct whale tracks, with different whale motions. The sensing module shows a median bearing error of 10.55° to the tag. Using bearing measurements to the engineered whale from an acoustic sensor and our sensing module, our autonomy module gives an aggregate rendezvous success rate of 81.31% for a 500-meter rendezvous distance using three robots in postprocessing. A second class of fielded experiments that used acoustic-only bearing measurements to three untagged sperm whales showed an aggregate rendezvous success rate of 68.68% for a 1000-meter rendezvous distance using two robots in postprocessing. We further validated these algorithms with several ablation studies using a sperm whale visual encounter dataset collected by marine biologists.

DeepWhaleNet: Climate Change-Aware FFT-Based Deep Neural Network for Passive Acoustic Monitoring

Climate change poses severe risks to the survival of many whale populations, whose habitats and migration patterns are affected by environmental changes. To detect these whales effectively, especially in deep-sea environments, we need to use AI-based techniques to handle the acoustic diversity and variability of different species. However, current methods for whale detection are based on pre- and post-processing steps that reduce their efficiency and generalizability. To address this issue, we present DeepWhaleNet, a novel deep-learning model that automates whale detection in Underwater Passive Acoustic Monitoring datasets. DeepWhaleNet simplifies the detection process by extracting relevant features from raw log-power spectrograms and helps protect these threatened species by supporting conservation efforts. Our model uses a larger short-time Fourier transform as input and a custom ResNet-18 architecture for classification, which enables it to separate whale sounds from noise and capture their temporal and spectral characteristics. We evaluate the performance of DeepWhaleNet and show that it surpasses state-of-the-art methods, achieving an 8.3% improvement in the F-1 score and 21% higher average precision of binary relevance than the baseline method. Moreover, our model demonstrates its versatility and suitability for species-specific retrieval problems through an ablation study on multi-label retrieval problems and a 99.1% recall for Blue Whales.

Continuous active sonar may not fix noise pollution of killer whale habitats

Although quieter, continuous active sonar still interferes too much with killer whale communication.

Functional data analysis to describe and classify southern resident killer whale calls

The Southern Resident Killer Whale (SRKW) is an endangered population of whales found in the northeast Pacific. They have a vocal dialect unique from other killer whales, having a repertoire of distinct stereotyped calls. A framework for distinguishing SRKW call types using the frequency traces of the amplitude ridges from their spectrograms (termed frequency ridges) is proposed. The first step is the extraction of these ridges of SRKW calls using an Sequential Monte Carlo approach. Next, they are converted into functional data using B-spline functions. They are analyzed with a functional principal component (FPC) analysis to characterise the intrinsic variability of frequency ridges within a call type. The FPCs are able to capture the general patterns in the frequency ridges of the different SRKW call types. The FPCs are also used as the basis for call classification. Using a cross-validation procedure to assess the robustness of the classification, this framework proves to be successful for classification with some call types having an F1-score ≥80%, but other calls less well discriminated. On balance, this approach showed reasonable performance given the small sample size available, and provides a useful contribution towards the development of a universal tool for call classification.

Auditory masking of tonal and conspecific signals by continuous active sonar, amplitude modulated noise, and Gaussian noise in killer whales (Orcinus orca).

Continuous active sonar is thought to mitigate severe acoustic impacts due to its lower sound pressure level compared to pulsed active sonar typically used by world navies. However, due to its almost continuous duty cycle, continuous active sonar could have a higher potential for auditory masking. Here, we evaluate the auditory masking potential of several noise types including a recording of continuous active sonar, amplitude modulated noise, and Gaussian noise, on signal detection in two killer whales. Signals were either a 1.5 kHz pure tone or a recording of a broadband burst-pulse killer whale call. For the 1.5 kHz tone, all noise types resulted in statistically significant masking, however, there was a release from masking of approximately 13 dB for the amplitude-modulated noise. When the killer whale call was the signal, the whales employed an off-frequency listening strategy where the whales were able to detect frequency components of the signal that did not directly overlap with the noise. However, this strategy was less useful for the continuous active sonar noise due to its broadband harmonic structure. Continuous active sonar has spectral features that considerably overlap with those of killer whale calls, making this type of noise an effective auditory masker.

Diverse baleen whale acoustic occurrence around two sub-Antarctic islands: A tale of residents and visitors

Knowledge on the occurrence and behaviour of baleen whales around sub-Antarctic regions is limited, and usually based on short, seasonal sighting research from shore or research vessels and whaling records, neither of which provide accurate and comprehensive year-round perspectives of these animals’ ecology. We investigated the seasonal acoustic occurrence and diel vocalizing pattern of baleen whales around the sub-Antarctic Prince Edward Islands (PEIs) using passive acoustic monitoring data from mid-2021 to mid-2023, detecting six distinct baleen whale songs from Antarctic blue whales, Madagascan pygmy blue whales, fin whales, Antarctic minke whales, humpback whales, and sei whales. Antarctic blue and fin whales were detected year-round whereas the other species’ songs were detected seasonally, including a new Antarctic minke whale bio-duck song sub-type described here for the first time. Antarctic minke and sei whales were more vocally active at night-time whereas the other species had no clear diel vocalizing patterns. Random forest models identified month and/or sea surface temperature as the most important predictors of all baleen whale acoustic occurrence. These novel results highlight the PEIs as a useful habitat for baleen whales given the number of species that inhabit or transit through this region.

Beyond counting calls: estimating detection probability for Antarctic blue whales reveals biological trends in seasonal calling

We explore the utility of estimating the density of calls of baleen whales for better understanding acoustic trends over time. We consider as a case study stereotyped ‘song’ calls of Antarctic blue whales (Balaenoptera musculus intermedia) on their Antarctic feeding grounds over the course of a year-long, continuous recording from 2014. The recording was made in the Southern Ocean from a deep-water autonomous hydrophone moored near the seafloor in the Eastern Indian sector of the Antarctic. We estimated call density seasonally via a Monte-Carlo simulation based on the passive sonar equation, and compared our estimates to seasonal estimates of detection rate, which are commonly reported in acoustic studies of Antarctic blue whales. The resulting seasonal call densities at our Antarctic site were strongly influenced by seasonally varying noise levels, which in turn yielded seasonal differences in detection range. Incorporating the seasonal estimates of detection area into our analysis revealed a pattern of call densities in accord with historic (non-acoustic) knowledge of Antarctic blue whale seasonal distribution and migrations, a pattern that differed from seasonal detection rates. Furthermore, our methods for estimating call densities produced results that were more statistically robust for comparison across sites and time and more meaningful for interpretation of biological trends compared to detection rates alone. These advantages came at the cost of a more complex analysis that accounts for the large variability in detection range of different sounds that occur in Antarctic waters, and also accounts for the performance and biases introduced by automated algorithms to detect sounds. Despite the additional analytical complexities, broader usage of call densities, instead of detection rates, has the potential to yield a standardized, statistically robust, biologically informative, global investigation of acoustic trends in baleen whale sounds recorded on single hydrophones, especially in the remote and difficult to access Antarctic.

Tracking Whale Calls in the Lower St. Lawrence Seaway at Land Seismometers

The Lower St. Lawrence Seaway (LSLS) is critical to Canada’s economy both as part of a major marine shipping corridor and a site of intensive fishing. Every year, fin whales and blue whales frequent the LSLS feeding ground. Understanding the mechanisms driving whale habitat usage is key for making informed decisions on shipping and fishing, reducing whale collision risks and mitigating noise pollution. We detect whales in the LSLS with land seismometers by using a method that relies on the intervals of the regularly repeating low frequency calls. The resulting catalogue contains 14,076 fin whale detections and 3,739 blue whale detections between February 2020 and January 2022. These detections follow the overall pattern of hydrophones, with most detections from fall to early winter in the Estuary and until mid-winter/spring in the Gulf. High detection rates in the Northwest Gulf throughout the winter months demonstrate that this region is potentially utilized year-round. This labelled catalogue may be suitable for developing a deep learning-based whale call detection algorithm. Making use of seismometers and deep learning can increase whale monitoring coverage within the LSLS and elsewhere.

BASSA: New software tool reveals hidden details in visualisation of low-frequency animal sounds.

The study of animal sounds in biology and ecology relies heavily upon time-frequency (TF) visualisation, most commonly using the short-time Fourier transform (STFT) spectrogram. This method, however, has inherent bias towards either temporal or spectral details that can lead to misinterpretation of complex animal sounds. An ideal TF visualisation should accurately convey the structure of the sound in terms of both frequency and time, however, the STFT often cannot meet this requirement. We evaluate the accuracy of four TF visualisation methods (superlet transform [SLT], continuous wavelet transform [CWT] and two STFTs) using a synthetic test signal. We then apply these methods to visualise sounds of the Chagos blue whale, Asian elephant, southern cassowary, eastern whipbird, mulloway fish and the American crocodile. We show that the SLT visualises the test signal with 18.48%-28.08% less error than the other methods. A comparison between our visualisations of animal sounds and their literature descriptions indicates that the STFT's bias may have caused misinterpretations in describing pygmy blue whale songs and elephant rumbles. We suggest that use of the SLT to visualise low-frequency animal sounds may prevent such misinterpretations. Finally, we employ the SLT to develop 'BASSA', an open-source, GUI software application that offers a no-code, user-friendly tool for analysing short-duration recordings of low-frequency animal sounds for the Windows platform. The SLT visualises low-frequency animal sounds with improved accuracy, in a user-friendly format, minimising the risk of misinterpretation while requiring less technical expertise than the STFT. Using this method could propel advances in acoustics-driven studies of animal communication, vocal production methods, phonation and species identification.

Variation in glider-detected North Atlantic right, blue, and fin whale calls in proximity to high-traffic shipping lanes

Variation in glider-detected North Atlantic right, blue, and fin whale calls in proximity to high-traffic shipping lanes

The 2019 Marine Heatwave at Ocean Station Papa: A Multi‐Disciplinary Assessment of Ocean Conditions and Impacts on Marine Ecosystems

AbstractIn the past decade, two large marine heatwaves (MHWs) formed in the northeast Pacific near Ocean Station Papa (OSP), one of the oldest oceanic time series stations. Physical, biogeochemical, and biological parameters observed at OSP from 2013 to 2020 are used to assess ocean response and potential impacts on marine life from the 2019 northeast Pacific MHW. The 2019 MHW reached peak surface and subsurface temperature anomalies in the summertime and had both coastal, impacting fisheries, and offshore consequences that could potentially affect multiple trophic levels in the Gulf of Alaska. In the Gulf of Alaska, the 2019 MHW was preceded by calm and stratified upper ocean conditions, which preconditioned the enhanced surface warming in late spring and early summer. The MHW coincided with lower dissolved inorganic carbon and higher pH of surface waters relative to the 2013–2020 period. A spike in the summertime chlorophyll followed by a decrease in surface macronutrients suggests increased productivity in the well‐lit stratified upper ocean during summer 2019. More blue whale calls were recorded at OSP in 2019 compared to the prior year. This study shows how the utility of long‐term, continuous oceanographic data sets and analysis with an interdisciplinary lens is necessary to understand the potential impact of MHWs on marine ecosystems.

From citizen science to AI models: Advancing cetacean vocalization automatic detection through multi-annotator campaigns

Continuous underwater Passive Acoustic Monitoring (PAM) has emerged as a strong tool for cetacean research. To handle the vast volume of collected data, it is essential to employ automated detection and classification methods. The recent advancement of deep learning, involving model training and testing, requires a large amount of labeled data. These labels are derived through the manual annotation of audio files often reliant on human experts. Based on an annotation campaign focusing on blue whale calls in the Indian Ocean involving 19 novice annotators and one expert in bioacoustics, this study explores the integration of novice annotators in marine bioacoustics research, through citizen science programs, which could drastically increase the size of labeled datasets and enhance the performance of detection and classification models. The analysis reveals distinctive annotation profiles influenced by the complexity of vocalizations and the annotators' strategies, ranging from conservative to permissive. To address the challenges of annotation discrepancies, Convolutional Neural Networks (CNNs) are trained on annotations from both novices and the expert. The results show variations in model performance. Our work highlights the importance of annotation guidelines encouraging a more conservative approach to improve overall annotation quality. In an effort to optimize the potential of multi-annotation and mitigate the presence of noisy labels, two annotation aggregation methods (majority voting and soft labeling) are proposed and tested. The results demonstrate that both methods, particularly when a sufficient number of annotators are involved, significantly improve model performance and reduce variability: the standard deviation of the area under PR and ROC curves fall under 0.02 for both vocalizations with 13 aggregated annotators, while it was at 0.17 and 0.21 for the Blue Whale Dcalls and 0.05 and 0.04 for the SEIO PBW vocalizations with all annotators separately. Moreover, these aggregation methods enable the training of models using non-expert annotations that achieve performance of models trained with expert annotations. These findings suggest that crowdsourced annotations from novice annotators can be a viable alternative to expert annotations.

A decade of change and stability for fin whale song in the North Atlantic

Fin whale song is a ubiquitous low-frequency pulsing that has been recorded in every ocean basin. In recent years, the complexity of fin whale song has been realized. In the North Atlantic, this song is made up of two low-frequency note types (A and B notes) and one higher-frequency note type (upsweeps). The A and B notes are produced with both singlet and doublet inter-note intervals. These song patterns are interwoven throughout an individual’s song and have been observed to shift over time, both gradually and suddenly. Like many baleen whales, the population size of fin whales is poorly defined. Passive acoustic monitoring has been suggested as a way to estimate abundance if a cue rate or calling rate can be determined. In this study, we used recordings from 119 fin whale tracks from 2013–2023 generated from passive acoustic recorders in the North Atlantic to define fin whale song patterns in this ocean basin, identify changes in these song patterns over time, and calculate cue rates. The fin whales recorded sang in 4 distinct patterns. Inter-note intervals were 25 s for A-A singlets, 13 s for B-B singlets, 12/20 s for B-B doublets, and 10/11 s for A-B doublets. B-B doublet inter-note intervals significantly increased from May 2019–April 2023 and A-B inter-note intervals showed an intra-annual increasing trend that reset each summer. Fin whale song in the North Atlantic also often includes higher frequency upsweep notes immediately prior to B notes. While A and B note peak frequency was steady over these years, upsweep peak frequency significantly decreased, continuing a similar trend that has been observed over 30 years. Despite these changes, the dominant song pattern remained consistent and cue rate for actively singing whales was relatively stable, which is in contrast to patterns observed in the North Pacific. Many of these nuances in fin whale singing behavior are newly described herein, and these findings are vital for monitoring baseline behavior, behavioral trends over time, and ultimately estimating abundance of a poorly understood species.

Humpback whale (Megaptera novaeangliae) social calls in southern British Columbia

AbstractHumpback whale (Megaptera novaeangliae) nonsong vocalizations, or social calls, are much more poorly understood than humpback song. We examined humpback whale social calls from a foraging ground in southern British Columbia (BC) and developed a catalog for humpback social calls in BC. We tagged four humpback whales on the eastern edge of Swiftsure Bank, BC, in early September 2020, with a passive acoustic and movement tag. We manually classified 32 call types in our data set based on comparisons with published classifications of humpback social calls. Many of the calls identified in our data set had similar characteristics to calls from other locations. We also used two statistical classification methods, a cluster analysis and a random forest. The cluster analysis grouped 20 of these call types into four categories, and the random forest classifier was able to accurately classify all 20 call types 87.6% of the time. This study fills a geographical gap of humpback whale social calls on foraging grounds and is a first step towards categorizing the social calls of humpback whales in BC.

Contextual and combinatorial structure in sperm whale vocalisations

Sperm whales (Physeter macrocephalus) are highly social mammals that communicate using sequences of clicks called codas. While a subset of codas have been shown to encode information about caller identity, almost everything else about the sperm whale communication system, including its structure and information-carrying capacity, remains unknown. We show that codas exhibit contextual and combinatorial structure. First, we report previously undescribed features of codas that are sensitive to the conversational context in which they occur, and systematically controlled and imitated across whales. We call these rubato and ornamentation. Second, we show that codas form a combinatorial coding system in which rubato and ornamentation combine with two context-independent features we call rhythm and tempo to produce a large inventory of distinguishable codas. Sperm whale vocalisations are more expressive and structured than previously believed, and built from a repertoire comprising nearly an order of magnitude more distinguishable codas. These results show context-sensitive and combinatorial vocalisation can appear in organisms with divergent evolutionary lineage and vocal apparatus.

Antarctic sonobuoy surveys for blue whales from 2006-2021 reveal contemporary distribution, changes over time, and paths to further our understanding of their distribution and biology

Seven passive acoustic surveys for marine mammal sounds were conducted by deploying sonobuoys along ship tracks during Antarctic voyages spanning years 2006-2021. These surveys included nearly 330° of longitude throughout Antarctic (south of 60°S) and sub-Antarctic (between 50-60°S) latitudes. Here, we summarise the presence of calls from critically endangered Antarctic blue whales (Balaenoptera musculus intermedia) detected on all seven of these surveys. We describe and compare the spatial distribution of detections of three different types of Antarctic blue whale calls: unit-A, Z-calls, and D-calls. Three sets of voyages partially overlapped spatially but in different years, providing three regions (Indian Sector, Dumont d’Urville Sea, Ross Sea) to investigate differences over time for these three different call types. The proportion of sonobuoys with calls present was significantly higher in the more recent years for seven of the 15 combinations of years, regions, and call type. The proportion of sonobuoys with calls present was significantly lower only for one of the 15 combinations (unit A in the Ross Sea between 2015 vs 2017), and not significantly different for the remaining seven pairwise comparisons. We discuss possible explanations for these observations including: differences in probability of detection, whale behaviour, whale distribution, and abundance. These explanations are not mutually exclusive and cannot yet be resolved without application of complex analytical methods and collection of additional data. Lastly, we discuss future work that could help clarify the contributions of each of these potential drivers of acoustic detection. We propose continued acoustic data collection, application of new analytical methods, and collection of other synergistic data from Antarctic blue whales on their feeding grounds as a basis for future work on this species. This could provide a cost effective and holistic means of monitoring their status after the effects of 20th century industrial whaling, as well as their responses to natural and anthropogenic changes to their main prey, Antarctic krill, and a changing climate.

Generating Synthetic Sperm Whale Voice Data Using StyleGAN2-ADA

The application of deep learning neural networks enables the processing of extensive volumes of data and often requires dense datasets. In certain domains, researchers encounter challenges related to the scarcity of training data, particularly in marine biology. In addition, many sounds produced by sea mammals are of interest in technical applications, e.g., underwater communication or sonar construction. Thus, generating synthetic biological sounds is an important task for understanding and studying the behavior of various animal species, especially large sea mammals, which demonstrate complex social behavior and can use hydrolocation to navigate underwater. This study is devoted to generating sperm whale vocalizations using a limited sperm whale click dataset. Our approach utilizes an augmentation technique predicated on the transformation of audio sample spectrograms, followed by the employment of the generative adversarial network StyleGAN2-ADA to generate new audio data. The results show that using the chosen augmentation method, namely mixing along the time axis, makes it possible to create fairly similar clicks of sperm whales with a maximum deviation of 2%. The generation of new clicks was reproduced on datasets using selected augmentation approaches with two neural networks: StyleGAN2-ADA and WaveGan. StyleGAN2-ADA, trained on an augmented dataset using the axis mixing approach, showed better results compared to WaveGAN.

Effects of duty cycle on passive acoustic monitoring metrics: The case of blue whale songs.

Long-term fixed passive acoustic monitoring of cetacean populations is a logistical and technological challenge, often limited by the battery capacity of the autonomous recorders. Depending on the research scope and target species, temporal subsampling of the data may become necessary to extend the deployment period. This study explores the effects of different duty cycles on metrics that describe patterns of seasonal presence, call type richness richness, and daily call rate of three blue whale acoustics populations in the Southern Indian Ocean. Detections of blue whale calls from continuous acoustic data were subsampled with three different duty cycles of 50%, 33%, and 25% within listening periods ranging from 1 min to 6 h. Results show that reducing the percentage of recording time reduces the accuracy of the observed seasonal patterns as well as the estimation of daily call rate and call call type richness. For a specific duty cycle, short listening periods (5-30 min) are preferred to longer listening periods (1-6 h). The effects of subsampling are greater the lower the species' vocal activity or the shorter their periods of presence. These results emphasize the importance of selecting a subsampling scheme adapted to the target species.

Global conservation genomics of blue whales calls into question subspecies taxonomy and refines knowledge of population structure

AbstractBlue whales (Balaenoptera musculus) are the largest living animal and, like other baleen whales, became endangered due to whaling. Here, we used population genomics to infer the number, distribution and other characteristics of subspecies and populations. We used the largest DNA dataset in blue whales, both in terms of genomic markers (16,661 SNPs and mtDNA) and geographic coverage (n = 276 for SNPs; n = 531 for mtDNA). We found greatest divergence among the eastern Pacific, Indo‐western Pacific and Antarctic blue whales. There were indications that natural selection in different environments promoted divergence among these groupings. Within these regions, there was divergence between the eastern North and eastern South Pacific, and among the eastern Indian Ocean, the western South Pacific and the northern Indian Ocean. There was no divergence within the Antarctic. These findings are consistent with the current classification of Antarctic and Indo‐western Pacific blue whales in the Southern Hemisphere as different subspecies but call into question the subspecies taxonomy of eastern Pacific blue whales. The study shows that opposite breeding seasons on either side of the equator do not necessarily inhibit connectivity across the equator, and reinforces that population structure needs to be well understood to conserve the diversity within species.

Comparing the underwater soundscape of the Hawaiian Islands Humpback Whale National Marine Sanctuary and potential influences of the COVID-19 pandemic

Passive acoustic monitoring is an effective technique for long-term monitoring of the soundscape in marine protected areas. Ocean noise is a key concern for the U.S. Office of National Marine Sanctuaries and has been identified as a research priority. The Sanctuary Soundscape Monitoring Project (“SanctSound”) was implemented to support efforts to address ocean noise across seven U.S. sanctuaries using a comprehensive and standardized approach. In this study, acoustic recordings were collected in the Hawaiian Islands Humpback Whale National Marine Sanctuary during the humpback whale seasons (November-May) from 2018-2022. Data encompassed 14 deployments across four sites in the main Hawaiian Islands: Hawaiʻi, Maui, O’ahu, and Kauaʻi. The soundscape was dominated by biological sources, most prominently the seasonal detection of humpback whale song. Third octave level monthly medians ranged from 70.4-105 dB re 1 µPa across sites with distinct peaks from January to April particularly at both Hawaiʻi and Maui sites. Overall, we reported relatively low vessel detection rates, with Maui having the highest daily average of vessel detections (x = 19.16). No COVID-19 impact could be observed acoustically using soundscape metrics which was likely due to the dominance of humpback whale chorusing. However, vessel detections and AIS data revealed a reduction in vessel activity after the onset of the pandemic at the Maui and Hawaiʻi sites. This study demonstrates that standardized metrics are a useful tool for obtaining long-term, baseline soundscape levels to understand the various contributions to the underwater soundscape and potential changes within marine protected areas in Hawaiʻi.