A review of big data analysis methods for baleen whale passive acoustic monitoring

AimSpecies distribution models (SDMs) are essential tools in ecology and conservation. However, the scarcity of visual sightings of marine mammals in remote polar areas hinders the effective application of SDMs there. Passive acoustic monitoring (PAM) data provide year‐round information and overcome foul weather limitations faced by visual surveys. However, the use of PAM data in SDMs has been sparse so far. Here, we use PAM‐based SDMs to investigate the spatiotemporal distribution of the critically endangered Antarctic blue whale in the Weddell Sea.LocationThe Weddell Sea.MethodsWe used presence‐only dynamic SDMs employing visual sightings and PAM detections in independent models. We compared the two independent models with a third combined model that integrated both visual and PAM data, aiming at leveraging the advantages of each data type: the extensive spatial extent of visual data and the broader temporal/environmental range of PAM data.ResultsVisual and PAM data prove complementary, as indicated by a low spatial overlap between daily predictions and the low predictability of each model at detections of other data types. Combined data models reproduced suitable habitats as given by both independent models. Visual data models indicate areas close to the sea ice edge (SIE) and with low‐to‐moderate sea ice concentrations (SIC) as suitable, while PAM data models identified suitable habitats at a broader range of distances to SIE and relatively higher SIC.Main ConclusionsThe results demonstrate the potential of PAM data to predict year‐round marine mammal habitat suitability at large spatial scales. We provide reasons for discrepancies between SDMs based on either data type and give methodological recommendations on using PAM data in SDMs. Combining visual and PAM data in future SDMs is promising for studying vocalized animals, particularly when using recent advances in integrated distribution modelling methods.

Passive acoustic monitoring (PAM) data collection has been growing exponentially, resulting in petabytes of data that document ocean soundscapes, how they change over time, and what animals use these ecosystems at varying timescales. Efficiently extracting this critical information and comparing it to other datasets in the context of ecosystem-based management is a Big Data challenge that traditional desktop processing methods cannot address. The curation, management, and dissemination of PAM datasets is another challenge in need of collaborative progress. To meet these exigencies, a multi-agency funded Sound Cooperative (SoundCoop) project is building community-focused, national cyberinfrastructure capability for PAM data to promote improved, scalable and sustainable accessibility and applications for management and science. Driven by partnerships and framed by four case studies, the SoundCoop has established guidance on the standardized processing of sound level metrics using free software toolkits and begun developing core cyberinfrastructure components that future PAM projects can leverage. U.S. and international scientists contributed PAM data collected across 10 long-term monitoring projects to operationalize the production of hybrid-millidecade spectra across a diversity of labs/instruments. Collectively, the contributed data demonstrate the value of standardized processing that enables the creation of comparable results from disparate monitoring efforts.

Integration of passive acoustic monitoring data into OBIS-SEAMAP, a global biogeographic database, to advance spatially-explicit ecological assessments

Navy training events involving the use of explosives pose a potential threat to marine mammals. This study used passive acoustic and visual monitoring data to evaluate marine mammals’ behavioral responses to noise from explosive events. Monitoring was conducted during five training events in the Virginia Capes (VACAPES) Range Complex during August/September of 2009–2012. Passive acoustic monitoring methods ranged from a single hydrophone to an array of sonobuoys monitored in real time. Visual monitoring effort over the five events totaled approximately 34 h (day before events: 10.1 h; days of events: 22.3 h; day after events: 1.5 h), yielding a total of 27 marine mammal sightings. Approximately 54 h of acoustic data were collected before, during, and after the five events. Behavioral changes were evaluated based on analysis of vocalizations detected before, during, and after explosions and concurrent data from visual sightings. For time periods with both visual and acoustic monitoring data, detection methods were compared to evaluate effectiveness. Continuing use and evaluation of both visual and passive acoustic methods for monitoring of explosive training events will improve our knowledge of potential impact resulting from explosive events and help improve management and conservation of marine mammals.

The southeast Pacific (SEP) contains the home ranges of several migratory large whale species, determined largely based on research in coastal waters. These whales’ pelagic seasonal residency is unknown. The Juan Fernandez Archipelago (JFA) is an offshore island in the SEP where passive acoustic monitoring (PAM) data collection is ongoing at the HA03 hydroacoustic station (sample rate of 250 Hz) maintained by the Preparatory Comprehensive Nuclear-Test-Ban Treaty Organization. Six years (2007–2009 and 2014–2016) of PAM data from HA03 were examined for the acoustic presence of Antarctic, Chilean (southeast Pacific 1, southeast Pacific 2), and southeast Indian Ocean blue whale song types; the fin whale 20-Hz song; and minke whale vocalizations. The weekly presence or absence of these six vocalization types was annotated manually by an expert analyst and then averaged across all years. For each vocalization type, the number of weeks per month with presence was averaged over all years. Consistently, we found austral wintertime presence of Antarctic blue, fin, and minke whales; and the year-round presence of Chilean blue whales. Southeast Indian Ocean blue whales were also detected, but very rarely. We discuss the possible seasonal movements of each species or acoustic group in the offshore SEP. This is the first year-round multispecies study of baleen whale in the offshore SEP and provides valuable information for understanding the migrations of endangered baleen whales in this region, highlighting the importance of offshore areas as hotspots for baleen whale biodiversity.

ABSTRACTAutomated detection of acoustic signals is crucial for effective monitoring of sound‐producing animals and their habitats across ecologically relevant spatial and temporal scales. Recent advances in deep learning have made these approaches more accessible. However, few deep learning approaches can be implemented natively in the R programming environment; approaches that run natively in R may be more accessible for ecologists. The “torch for R” ecosystem has made deep learning with convolutional neural networks (CNNs) accessible for R users. Here, we evaluate a workflow for the automated detection and classification of acoustic signals from passive acoustic monitoring (PAM) data. Our specific goals include (1) present a method for automated detection of gibbon calls from PAM data using the “torch for R” ecosystem, (2) conduct a series of benchmarking experiments and compare the results of six CNN architectures; and (3) investigate how well the different architectures perform on data sets of the female calls from two different gibbon species: the northern gray gibbon (Hylobates funereus) and the southern yellow‐cheeked crested gibbon (Nomascus gabriellae). We found that the highest‐performing architecture depended on the species and test data set. We successfully deployed the top‐performing model for each gibbon species to investigate spatial variation in gibbon calling behavior across two grids of autonomous recording units in Danum Valley Conservation Area, Malaysia and Keo Seima Wildlife Sanctuary, Cambodia. The fields of deep learning and automated detection are rapidly evolving, and we provide the methods and data sets as benchmarks for future work.

Passive acoustic monitoring is increasingly being used as a cost‐effective way to study wildlife populations, especially those that are difficult to census using conventional methods. Burrow‐nesting seabirds are among the most threatened birds globally, but they are also one of the most challenging taxa to census, making them prime candidates for research into such automated monitoring platforms. Passive acoustic monitoring has the potential to determine presence/absence or quantify burrow‐nesting populations, but its effectiveness remains unclear. We compared passive acoustic monitoring, tape‐playbacks andGPStracking data to investigate the ability of passive acoustic monitoring to capture unbiased estimates of within‐colony variation in nest density for the Manx ShearwaterPuffinus puffinus. Variation in acoustic activity across 12 study plots on an island colony was examined in relation to burrow density and environmental factors across 2 years. As predicted fewer calls were recorded when wind speed was high, and on moon‐lit nights, but there was no correlation between acoustic activity and the density of breeding birds within the plots as determined by tape‐playback surveys. Instead, acoustic indices correlated positively with spatial variation in the in‐colony flight activity of breeding individuals detected byGPS. Although passive acoustic monitoring has enormous potential in avian conservation, our results highlight the importance of understanding behaviour when using passive acoustic monitoring to estimate density and distribution.

The urgent need for effective wildlife monitoring solutions in the face of global biodiversity loss has resulted in the emergence of conservation technologies such as passive acoustic monitoring (PAM). While PAM has been extensively used for marine mammals, birds, and bats, its application to primates is limited. Black-and-white ruffed lemurs (Varecia variegata) are a promising species to test PAM with due to their distinctive and loud roar-shrieks. Furthermore, these lemurs are challenging to monitor via traditional methods due to their fragmented and often unpredictable distribution in Madagascar's dense eastern rainforests. Our goal in this study was to develop a machine learning pipeline for automated call detection from PAM data, compare the effectiveness of PAM versus in-person observations, and investigate diel patterns in lemur vocal behavior. We did this study at Mangevo, Ranomafana National Park by concurrently conducting focal follows and deploying autonomous recorders in May-July 2019. We used transfer learning to build a convolutional neural network (optimized for recall) that automated the detection of lemur calls (57-h runtime; recall = 0.94, F1 = 0.70). We found that PAM outperformed in-person observations, saving time, money, and labor while also providing re-analyzable data. Using PAM yielded novel insights into V. variegata diel vocal patterns; we present the first published evidence of nocturnal calling. We developed a graphic user interface and open-sourced data and code, to serve as a resource for primatologists interested in implementing PAM and machine learning. By leveraging the potential of this pipeline, we can address the urgent need for effective primate population surveys to inform conservation strategies.

Various methods for estimating animal density from visual data, including distance sampling (DS) and spatially explicit capture-recapture (SECR), have recently been adapted for estimating call density using passive acoustic monitoring (PAM) data, e.g., recordings of animal calls. Here we summarize three methods available for passive acoustic density estimation: plot sampling, DS, and SECR. The first two require distances from the sensors to calling animals (which are obtained by triangulating calls matched among sensors), but SECR only requires matching (not localizing) calls among sensors. We compare via simulation what biases can arise when assumptions underlying these methods are violated. We use insights gleaned from the simulation to compare the performance of the methods when applied to a case study: bowhead whale call data collected from arrays of directional acoustic sensors at five sites in the Beaufort Sea during the fall migration 2007–2014. Call detections were manually extracted from the recordings by human observers simultaneously scanning spectrograms of recordings from a given site. The large discrepancies between estimates derived using SECR and the other two methods were likely caused primarily by the manual detection procedure leading to non-independent detections among sensors, while errors in estimated distances between detected calls and sensors also contributed to the observed patterns. Our study is among the first to provide a direct comparison of the three methods applied to PAM data and highlights the importance that all assumptions of an analysis method need to be met for correct inference.

ABSTRACTIn this paper, we present an automatic method, without human supervision, for the detection and classification of blue whale vocalizations from passive acoustic monitoring (PAM) data using Hidden Markov Model technology implemented with a state-of-the-art machine learning platform, the Kaldi speech processing toolkit. 157.5 hours of PAM data were annotated for model training and testing, selected from a dataset collected from the Corcovado Gulf, Chilean Patagonia in 2016. The system obtained produced 85.3% accuracy for detection and classification of a range of different blue whale vocalizations. This system was then validated by comparing its unsupervised detection and classification results with the published results of southeast Pacific blue whale song phrase (‘SEP2’) via spectrogram cross-correlation, involving a dataset collected with a different hydrophone instrument. The proposed system led to a reduction in the root mean square error relative to published results as high as 80% when compared with comparable methods employed elsewhere. This is a significant step in advancing the monitoring of endangered whale populations in this region, which remains poorly covered in terms of PAM and general ocean observation. With further training, testing and validation, this system can be applied to other target signals and regions of the world ocean.

Marine passive acoustic monitoring (PAM) has produced petabytes of data that are used by researchers, resource managers, industry, and regulators to understand how marine animals use sound and the impacts of anthropogenic noise on species and ecosystems throughout the global ocean. These big data provide unprecedented opportunities to study underwater soundscapes but also enormous challenges to efficiently extract information. To address these challenges, the Sound Cooperative (SoundCoop) project built community-focused cyberinfrastructure to promote improved, scalable and sustainable processing and access of marine PAM data for management, science, industry and military applications. Driven by cross-institutional participation representing a diversity of data collection methods and conditions, SoundCoop established guidance for standardized processing of ocean sound level metrics using freeware software toolkits and developed core tools and processes that support open science. Four comparative analyses that connect PAM monitoring efforts and integrate non-acoustic data illustrate how interoperable sound level metrics support a more coherent and synoptic perspective on ocean soundscapes using methods that current and future PAM projects can leverage. Such a framework around PAM big data offers the opportunity to revolutionize large-scale marine ecology and oceanography in similar ways to other transformative approaches for understanding environmental patterns and processes at global scales.

Passive Acoustic Monitoring (PAM) has emerged as a pivotal technology for wildlife monitoring, generating vast amounts of acoustic data. However, the successful application of machine learning methods for sound event detection in PAM datasets heavily relies on the availability of annotated data, which can be laborious to acquire. In this study, we investigate the effectiveness of transfer learning and active learning techniques to address the data annotation challenge in PAM. Transfer learning allows us to use pre-trained models from related tasks or datasets to bootstrap the learning process for sound event detection. Furthermore, active learning promises strategic selection of the most informative samples for annotation, effectively reducing the annotation cost and improving model performance. We evaluate an approach that combines transfer learning and active learning to efficiently exploit existing annotated data and optimize the annotation process for PAM datasets. Our transfer learning observations show that embeddings produced by BirdNet, a model trained on high signal-to-noise recordings of bird vocalisations, can be effectively used for predicting anurans in PAM data: a linear classifier constructed using these embeddings outperforms the benchmark by 21.7%. Our results indicate that active learning is superior to random sampling, although no clear winner emerges among the strategies employed. The proposed method holds promise for facilitating broader adoption of machine learning techniques in PAM and advancing our understanding of biodiversity dynamics through acoustic data analysis.

North Atlantic right whale monitoring in Roseway Basin, Canada, is primarily based on short‐term (<14 d) visual surveys conducted during August–September. Variability in survey effort has been the biggest limiting factor to studying changes in the population's occurrence and habitat use. Such efforts could be enhanced considerably using passive acoustic monitoring (PAM). We sought to determine if variation in whale presence, relative abundance, demography, and/or behavior (estimated through visual surveys) could be explained by variation in three right whale call types in this habitat. A generalized linear model was fit to 23 d of concurrent PAM and visual monitoring during four summers within the Roseway Basin Right Whale Critical Habitat boundaries. The model revealed significant positive relationships between relative abundance, call counts and presence of surface‐active group behavior. PAM can refine daily right whale presence estimates. While visual observations (n= 23 d) implied a 40% decline in right whale presence during 2014–2015 relative to 2004–2005, PAM data (n= 211 d) showed right whales were present between 71%–85% of survey days throughout all years analyzed. We demonstrate that PAM is a useful tool to extend periods of right whale monitoring, especially in areas where visual monitoring efforts may be limited.

Successful conservation and management of marine top predators rely on detailed documentation of spatiotemporal behavior. For cetacean species, this information is key to defining stocks, habitat use, and mitigating harmful interactions. Research focused on this goal is employing methodologies such as visual observations, tag data, and passive acoustic monitoring (PAM) data. However, many studies are temporally limited or focus on only one or few species. In this study, we make use of an existing long‐term (2009–2019), labeled PAM data set to examine spatiotemporal patterning of at least 10 odontocete (toothed whale) species in the Hawaiian Islands using compositional analyses and modeling techniques. Species composition differs among considered sites, and this difference is robust to seasonal movement patterns. Temporally, hour of day was the most significant predictor of detection across species and sites, followed by season, though patterns differed among species. We describe long‐term trends in species detection at one site and note that they are markedly similar for many species. These trends may be related to long‐term, underlying oceanographic cycles that will be the focus of future study. We demonstrate the variability of temporal patterns even at relatively close sites, which may imply that wide‐ranging models of species presence are missing key fine‐scale movement patterns. Documented seasonal differences in detection also highlights the importance of considering season in survey design both regionally and elsewhere. We emphasize the utility of long‐term, continuous monitoring in highlighting temporal patterns that may relate to underlying climatic states and help us predict responses to climate change. We conclude that long‐term PAM records are a valuable resource for documenting spatiotemporal patterns and can contribute many insights into the lives of top predators, even in highly studied regions such as the Hawaiian Islands.

Passive acoustic monitoring (PAM) provides new opportunities for assessing bird abundance and habitat preferences, yet its performance relative to traditional point-count surveys (PCO) remains insufficiently tested, especially for quiet and inconspicuous forest passerines. We compared the vocal activity and habitat associations of the Goldcrest Regulus regulus and Firecrest Regulus ignicapilla in a temperate forest ecosystem using PAM-derived and PCO-based indices. Across 30 monitoring points in the Romincka Forest (Poland), PAM yielded >33,000 recorded songs and revealed strong spatial variation in both species. Vocal activity measures obtained from PAM correlated positively with PCO detections and territories, confirming the reliability of PAM as a complementary abundance indicator. Goldcrest vocal activity showed a strong positive association with the proportion of coniferous trees—especially spruce—and with local tree-species richness, reflecting the species’ affinity for structurally diverse conifer-dominated stands. In contrast, Firecrest abundance was unrelated to forest structure in PAM data, while PCO detections indicated avoidance of pine and lower activity in species-rich stands. No significant relationship with stand age was observed for either species. The weak interspecific correlations in activity parameters highlight their distinct ecological niches despite overlapping ranges. Based on PCO Goldcrests proved to be more abundant, with a territorial ratio of 3:2 compared to Firecrests. Our study demonstrates that PAM effectively captures variation in abundance and habitat selectivity of both Regulus species and provides a scalable, efficient complement to traditional surveys in temperate forest ecosystems.