Cetacean acoustic occurrence in the Southern Ocean from biologged southern elephant seals

The Southern Ocean is a crucial habitat for multiple species of cetaceans, from mysticete species to odontocetes such as sperm whales and delphinids. However, their presence at high latitudes is still poorly known due to the remoteness of the area, with little literature on delphinid presence and sparse sampling along mysticete migration routes. At the end of their breeding season, southern elephant seals leave their haul-out sites and set out on a foraging trip, covering vast expanses of the Southern Ocean during the austral spring. During these trips, elephant seals rest by entering drift phases, where their speed is minimal. These drift phases can act as listening points of opportunity to analyze sound sources across all frequency ranges. Between 2017 and 2020, nine female elephant seals (3 from the Península Valdés population and 6 from the Kerguelen population) equipped with DTAG4 (Digital Acoustic Recording Tag) devices recorded the underwater ambient soundscape continuously (samplerate of either 32000 Hz or 38400 Hz). This study aimed at using drift phase recordings to assess cetacean acoustic occurrence, providing unique and complementary data to existing observations. Results show that multiple different cetacean species, both mysticetes and odontocetes, can be monitored, with high bioacoustic activity being found in some datasets. The annotation of the gathered recordings thereby offers a remarkable snapshot of cetacean acoustic presence in the Southern Ocean, giving new insights into their habitat use during spring. Results show acoustic presence of antarctic blue whales and humpack whales around the polar front and subantarctic front respectively in the Indian sector of the Southern Ocean and increased presence of sperm whales along important bathymetric gradients in the Atlantic sector of the Southern Ocean. These promising results emphasize the potential of biologged animals to monitor other species of interest, with the added benefit to sample across considerable areas and in harsh environments. This study paves the way for new research such as modeling cetacean acoustic presence using co-located environmental data collected by the tags.

Since the end of commercial whaling, significant efforts have been directed toward assessing the conservation status and distribution of migrating cetacean species, mainly sperm whales and mysticetes (baleen whales), with key progress made in understanding their habitat preferences (Savoca et al., 2024; Michel et al., 2025). These advances also extend to delphinid species (Janik, 2009; Thiebot and Weimerskirch, 2013), whose populations face similar conservation challenges and whose distribution and ecological roles are still poorly known today. For some odontocete species (toothed whales including delphinids and sperm whales) for instance, recent studies provide important insight into their foraging behavior and interactions with fisheries (Miller et al., 2004; Kyhn et al., 2010; Richard et al., 2022). At broader spatial scales, several studies have emphasized the ecological importance of oceanographic fronts, where enhanced productivity supports higher trophic levels (Bost et al., 2009). In particular, increased occurrences of species such as blue whales and sperm whales have been associated with strong sea surface temperature gradients (Etnoyer et al., 2006; Gannier and Praca, 2007). Overall, these developments highlight substantial progress in cetacean research, linking species distribution, abundance and behavior to key environmental drivers.

These advances have been facilitated by the ongoing development and diversification of monitoring platforms. Multiple methods now exist to survey cetaceans, including both fixed and mobile approaches (Nowacek et al., 2016). Earlier fixed methods include opportunistic analyses of stranded animals (Waerebeek et al., 2010) and land-based observations. Over the years, mobile detection techniques have been developed, including boat-based observation protocols, aerial surveys from planes or unmanned aerial vehicles, and satellite imagery (Friedlaender et al., 2006, Friedlaender et al., 2011; Williams et al., 2014; Gray et al., 2019; Cubaynes and Fretwell, 2022; Herr et al., 2022; Andrews-Goff et al., 2024). In the past decades, another observational method, underwater Passive Acoustic Monitoring (PAM) has increasingly emerged as an effective and non-invasive method to detect the presence of cetaceans and to assess their seasonal occurrence, using both fixed and mobile deployments (Mellinger et al., 2007; Dréo et al., 2019; van Geel et al., 2022; Estabrook et al., 2025). While surface-based observational methods are limited by favorable light and weather conditions and are only possible when animals surface, PAM is not subject to these limitations. For instance, fixed-moorings enable acoustic monitoring of cetaceans over large temporal scales without being limited by field conditions and depth (Torterotot et al., 2020; Shabangu and Kowarski, 2022). With recent technological developments, mobile PAM platforms are becoming an alternative solution to assess cetacean occurrence over large spatial scales. These platforms, such as gliders (Baumgartner and Fratantoni, 2008; Torterotot et al., 2022), drifting buoys (McCullough et al., 2021; Almunia et al., 2025) or biologging tags (Saddler et al., 2017; Clayton et al., 2023; Noda et al., 2024) enable population monitoring, real-time detections and linking acoustic occurrence, or the assessment of a species’ presence inferred from acoustic data, with oceanographic conditions.

However, assessing cetacean occurrence, whether using PAM platforms or more traditional methods, is particularly difficult in high latitudes such as in the Southern Ocean. The remoteness of this ocean and its harsh weather conditions lead to costly and challenging deployments. High latitudes, mainly offshore and along cetacean migration routes, are therefore heavily undersampled. Furthermore, the multiple sound sources and environmental noises picked up by the hydrophone, sometimes overlapping in frequency, can hinder the analysis of specific signals.

The Southern Ocean used to host the largest mysticete populations before being heavily depleted during the 20 th century by the whaling industry. Being able to monitor their presence is therefore critical to assess their recovery. In that context, PAM may help better understand their occurrence and distribution within the Southern Ocean, as well as changes in their acoustic abundance. This ocean, with few shipping lanes and continental masses, contains little to no parasitic noises that would mask cetacean vocalizations. Recent PAM research has for instance confirmed the presence of multiple cetacean species in the subantarctic, with different breeding and foraging grounds being identified (Calderan et al., 2021; Casey et al., 2022). However, while some fixed PAM deployments exist, they remain insufficient to adequately sample the vast expanse of the Southern Ocean. Achieving fine spatial acoustic sampling of the area with such deployments would require substantial effort.

The use of marine animals as autonomous sampling platforms has emerged as an increasingly valuable way to gather biological and oceanographic data in regions that are otherwise difficult to access. These systems provide continuous, high-resolution measurements of environmental variables such as temperature, salinity, and ecosystem dynamics, complementing traditional observation methods and filling critical data gaps, especially in hard to access areas such as high latitudes (Charrassin et al., 2008; Iwata and Akamatsu, 2025). Most recently, these observations have been integrated into global networks such as the Global Ocean Observing System, significantly enhancing ocean processes monitoring capacities, and helping in the improvement of climate predictions while supporting sustainable ocean management (McMahon et al., 2021). More specifically, PAM recordings obtained through biologging offer a valuable opportunity to complement existing observations and bridge key gaps in current cetacean studies. Previous research on tagged green sea turtles has already shown the potential of PAM from biologging to assess cetacean acoustic presence in coastal areas of Japan through soundscape classification (Noda et al., 2024). Generalizing PAM deployments on different tagged species would therefore enable sampling in remote areas of high interest to key species.

Southern Elephant Seals (SES) have been used for over twenty years to sample the Southern Ocean, particularly in the Indian and Atlantic sectors, collecting unique environmental and behavioral data (Roquet et al., 2017; McMahon et al., 2025). For instance, accelerometer and magnetometer data allow to reconstruct the elephant seal’s diving behavior and trajectories. More recently, SES were equipped with acoustic tags recording the soundscape continuously. This data has enabled the observation of in-situ met-ocean variables such as surface wind speed (Cazau et al., 2017; Martinez et al., 2023; Gros-Martial et al., 2025) or rainfall rate (Gros-Martial, 2025). Furthermore, flow noise, which is the sound generated by water turbulence around the hydrophone membrane, is the main noise source at lower frequencies (below 3000 Hz), and gives indications about the individual’s behavior and swimming speed. Acoustic data can therefore be used to detect drift phases, where individuals glide passively with no swimming activity, with little flow noise being present in the recordings (Gros-Martial et al., 2025). Finally, PAM data can be effectively used to detect and classify biophonic sources, including cetacean vocalizations. While whale calls have previously been identified in PAM data from biologged elephant seals (Fletcher et al., 1996), no attempts to annotate species-specific calls to monitor cetacean populations and distributions have been made.

PAM observations from tagged SES are possible due to the acoustic behavior of cetacean species that frequent areas where SES forage. Throughout the Southern Ocean, Antarctic Blue Whales (ABW, Balaenoptera musculus), Pygmy Blue Whales (PBW, Balaenoptera musculus brevicauda), Fin Whales (FW, Balaenoptera physalus), Antarctic Minke Whales (MW, Balaenoptera bonaerensis), Southern Right Whales (SRW, Eubalaena australis), Humpback Whales (HW, Megaptera novaeangliae), and Sei Whales (SW, Balaenoptera borealis) (see Table 1) make stereotyped vocalizations across different frequency ranges that are well documented (Payne, 1986; Samaran et al., 2013; Shabangu et al., 2020; Cerchio and Weir, 2022; Casey et al., 2022). These calls are mostly low frequency (below 1000 Hz) such as the FW 20 Hz pulse signal or the Unit-A of ABW Z-calls at 27 Hz. Similarly, odontocetes produce high frequency signals, such as clicks, whistles, and burst-pulse sounds. For instance, sperm whales (Physeter macrocephalus) emit identifiable clicks with considerable source levels that can be used to infer their behavior (Miller et al., 2004). These clicks are usually in the frequency range between 400 Hz and 15000 Hz with a peak around 2000 Hz for males and between 1200 Hz and 3000 Hz for females, depending on the detected individual as well as recording properties (Goold and Jones, 1995). Similarly, while species-specific delphinid acoustic activities are sometimes poorly known, especially for offshore Southern Ocean species, the current literature indicates usual frequency ranges. Delphinid whistles are tonal signals whose fundamental frequency is between 800 Hz and 28500 Hz, while echolocation sounds include broad-band calls between 100 Hz and 30000 Hz for killer whales (Orcinus orca) and usually between 200 Hz and 130000 Hz for other delphinids (Evans, 1973; Janik, 2009; Ollier et al., 2023).

Using nine PAM datasets from biologged SES that were tagged with oceanographic beacons and biologgers including a hydrophone, along with measurements of acceleration, magnetometry, light and pressure, this study aims at assessing the potential of tagged SES to gather complementary observations of cetaceans in the Southern Ocean. By extracting drift phases, during which minimal flow noise generated by animal movement is produced, a high-quality dataset of listening points is curated along the SES track to identify and detect biophonic sources. This approach enables data collection across vast and previously under-sampled regions of the Southern Ocean. We hypothesize that these opportunistic listening points enable the annotation of a wide range of cetacean vocalizations, from low-frequency mysticete calls to higher-frequency sperm whale clicks and delphinid whistles while minimizing masking by flow noise. SES could therefore become effective mobile platforms for studying both mysticetes and odontocetes in remote offshore areas where monitoring has been minimal, while providing broad spatial coverage of ecologically important fronts, foraging grounds, and migration routes (Mackintosh, 1965; Bost et al., 2009; Rivière et al., 2019). Accordingly, this study aims to evaluate the effectiveness of SES as acoustic monitoring platforms by assessing their ability to reliably detect cetacean presence, identify which species can be recorded, and determine the overall potential of this approach for future large scale monitoring. We suppose that SES, by searching for areas of high biological richness during austral spring, will offer unique data on the acoustic presence of cetaceans, especially along baleen whale migration routes (McConnell et al., 1992; Waerebeek et al., 2010; Guinet et al., 2014). The resulting recordings offer valuable insights into species presence and habitat use, helping to fill critical knowledge gaps and complement existing observation methods, while highlighting the promise of biologging for cetacean research in remote environments.

Fieldwork and instrument deployment

Between 2017 and 2020, DTAG4 (Digital Acoustic Recording Tag) devices (Johnson and Tyack, 2003) were head-mounted on nine female elephant seals, three in Península Valdés, Argentina, and six in Kerguelen Islands. Tags were deployed only on female southern elephant seals as they have a higher site fidelity than males, thereby increasing the odds of recollecting the tag during the moulting period in January. Fieldwork was conducted in Kerguelen under authorization APAFIS #42011–2023032323039910 v4, granted in accordance with the French Rural and Maritime Fishing Code (articles R.214–87 to R.214126). Permits for fieldword in Península Valdés were granted by the Dirección de Fauna y Flora Silvestre de la Provincia de Chubut (Disposición N° 21/2018-DFyFS-MP, Disposición N° 68/2019-DFyFS-MP) and Subsecretaría de Conservación y Areas Protegidas de Chubut (Disposición N° 61-SsCyAP/18 and Disposición N° 063/2019-SsCyAP). The tags were deployed during the breeding season, in austral spring, between September and November, and gathered the following year during the moulting season around January. The different trajectories of each tagged individual, obtained from GPS data, can be seen Figure 1.

Tagged individuals were anesthetized using intravenous tiletamine and zolazepam 1:1 (McMahon et al., 2000). Instruments were then deployed on their head using a resin (two-component epoxy) attached to their fur (Charrassin et al., 2008). The protocol lasted around thirty minutes, with about ten to fifteen minutes between the anesthesy and the deployment, and was conducted by a veterinarian. After the procedure, tagged individuals were then monitored to ensure they showed no apparent signs of distress and resumed normal behaviour. The females that were tagged were either returning to sea after weaning or were at the end of the lactation period. During handling, particular care was taken to ensure that the pup remained close to its mother and that no separation occurred during or after the procedure.

The tags logged depth every 0.02 s and recorded the surrounding soundscape continuously with a sampling frequency of 32000 Hz or 38400 Hz depending on the deployment (see Table 2), enabling an analysis of frequencies up to 16000 Hz and 19200 Hz respectively. The hydrophones have a high-pass filter with a cutoff frequency of 20 Hz. Each hydrophone was positioned towards the front of the animal as the first aim of these deployments was to assess the animal’s swimming speed. The loggers were also equipped with an accelerometer sampling at 200 Hz used to qualify the individual’s diving behavior. Furthermore, a satellite data relayed logger equipped with a CTD sensor (CTD hereafter) from the Sea Mammal Research Unit was deployed on the back of the seal, logging temperature and salinity data continuously at 0.5 Hz. Figure 2 shows how the tags were deployed on the individuals.

Drift phase dataset

Drift phases are portions of dives where SES rest in a falling-leaf trajectory with low vertical speed (Mitani et al., 2010). This decreases the intensity of flow noise considerably, which usually masks low frequency sources up to 3000 Hz (Figure 3). In order to annotate low frequency cetacean calls, these drift phases were extracted to create a smaller dataset (Table 2) with audio of higher quality where signals of interest are not masked by flow noise. Subsampling the original dataset therefore allowed for a more precise annotation of mysticete, sperm whale and delphinid vocalizations.

SES are known to search for areas of high foraging interest along fronts (Rivière et al., 2019) and with regards to bathymetry (McConnell et al., 1992). This results in a non-random distribution of drift dives, with individuals foraging and therefore sampling particular areas more intensively. Therefore, annotation results were considered with regards to two large scale oceanographic features: i) the different water masses delimited by the Antarctic Circumpolar Current (ACC) fronts, namely the Southern ACC Front, Polar Front and Subantarctic Front derived from thermohaline profiles (see Section 2.4) and ii) bathymetry as given by the GEBCO Bathymetric Compilation Group 2026, 2026). Areas in which SES were just transiting are also sampled, but less extensively.

Drift phases were detected using an unsupervised learning algorithm based on power spectral densities of the acoustic signal. Acoustic data from each dive were separated and cropped up to 15 min, with the drift phase taking place during the descent portion of a dive due to the high body density of SES after the breeding season. In order to reduce the dimensions of the spectrogram, UMAP (Uniform Manifold Approximation and Projection for dimension reduction (McInnes et al., 2020)) was applied on spectrogram representations of each dive. Clusters were identified in this new embedding space using the HDBSCAN algorithm (Hierarchical Density-Based Spatial Clustering of Applications with Noise). As drift phases are characterized by multiple samples with little to no flow noise, dives containing drift phases formed a clear cluster. This method was validated using manual verification of detected drift dives using the depth profile and further comparison with an existing detection method based on acceleration and vertical speed (Richard et al., 2014). Acoustic recordings from the drift phases were then extracted from the drift dive cluster. Drift phases usually last between 3 to 8 min, detections of drift phases shorter than 3 min were therefore removed from the dataset in accordance with (Dragon et al., 2012). A more detailed explanation of drift phase detection is given in Supplementary Materials. Only drift phases were used for the entire study, for both acoustic presence and the model.

Annotations

Each detected drift phase was visually examined by two expert annotators. The annotations were performed through visual inspection of spectrograms and were reported at the file level as positive or negative (1 or 0) for each label of interest. The annotation process for the 109 hours and 11 minutes (Table 2) of recordings required an average of 24 hours per annotator. To further consolidate the results and increase confidence in the annotations, a third annotator was introduced to review and resolve disagreements between the initial two annotators. This validation step follows common practices in the literature for expert annotation of PAM data (Miller et al., 2022; Leroy et al., 2018b), where adjudication by an additional expert is used to ensure higher reliability of the final labels.

Inter-annotator agreement was measured using the κ metric (Cohen, 1960). Following the methodology highlighted in Dubus et al. (2023), and because the annotations were performed as weak labels at the species level, only Fleiss’ κ F (Fleiss, 1975) was used in this study. Fleiss’ κ extends Cohen’s metric to multiple annotators and evaluates the level of agreement while accounting for the agreement expected by chance given the distribution of annotations across categories. Its value ranges from −1 to 1, with higher values indicating stronger agreement between annotators. Figure 4 illustrates the difficulty of the annotation task through three spectrograms of a similar vocalization, namely a 27 Hz ABW chorus, detailed at the end of the section.

Drift phases were annotated using spectrograms of varying frequency ranges computed using different window sizes to better detect low-frequency calls produced by mysticetes (spectrogram in the 0 − 200 Hz range with NFFT = 16384 and 0 − 1000 Hz with NFFT = 8192), and higher-frequency signals characteristic of delphinids, sperm whales and mysticete middle-frequency calls (spectrogram in the 200 Hz to 16000−19200 Hz range with NFFT = 1024). After confirmation from literature and existing catalogs of cetacean vocalizations, baleen whale species, sperm whales and delphinids were annotated. Delphinid whistles, buzzes and clicks were annotated but not attributed to a species due to the diversity of possible sources and the lack of literature on their vocalizations in the Southern Ocean. Species-specific vocalizations of baleen whales were identified when possible. The list of expected whale species and their associated acronyms can be found in Table 1.

Throughout this study, acoustic occurrence refers to the assessment of species presence using PAM data, specifically the presence of species-specific vocalizations in the drift-phase dataset. Because it relies on vocal activity rather than direct observation, acoustic occurrence may differ from species occurrence inferred through other monitoring approaches. Three complementary metrics were derived to characterize acoustic occurrence: i) the number of drift phases containing at least one confirmed vocalization, ii) the proportion of drift phases containing at least one confirmed vocalization (i.e., the number of drift phases with annotations divided by the total number of drift phases), and iii) the detection density defined below.

For each day containing at least one annotation, detection density was computed as the ratio of the cumulative duration of phases with confirmed vocalizations to the total duration of all drift phases detected that day for a single individual. If { j: j = 1, 2, …, K } corresponds to the K dives with an annotation for one specific species during a day and { i: i = 1, 2, …, N } the number of drift phases detected during the same day, the detection density D is:

with Δ T the duration of each drift phase.

These metrics capture different aspects of acoustic occurrence accounting for variability in the sampling effort. The number of drift phases with annotations reflects an absolute acoustic occurrence but is influenced by the total number of drift phases recorded on a given day which varies throughout the SES foraging trip. Expressing this acoustic occurrence as a proportion normalizes for differences in the number of sampling opportunities, thereby reducing biases associated with heterogeneous drift phase distributions among days and individuals. Detection density further accounts for differences in drift phase duration by quantifying the fraction of total listening time during which cetaceans were acoustically present enabling diel comparison between tagged individuals and annotated species. Together, these complementary metrics provide a more robust characterization of acoustic occurrence.

For mysticetes, most calls show long-term patterns that are more easily identifiable on spectrograms spanning one hour or more. ABW and FW make stereotyped calls, between 20 Hz and 30 Hz (Shabangu et al., 2020) at a high source level and are commonly found in the study area. ABW frequently emit Z-calls, a signal lasting between 18 to 26 seconds, and repeated every 70 to 80 seconds (Ljungblad et al., 1998). The upper portion of the call, Unit-A, can be detected remotely through a 27 Hz chorus (Širović et al., 2009; Samaran et al., 2013). The ABW 27 Hz chorus shows diel variations yet continuous presence throughout the day during spring (Shabangu et al., 2020). Similarly, FW pulses at 20 Hz are emitted in bouts that can last 32.5 hours, with a pulse being produced once every 7 to 26 second (Watkins et al., 1987). The main SRW call in the South Georgia Islands foraging grounds is a 100 Hz to 150 Hz upsweep (Calderan et al., 2021). While far from the study area, these islands are similarly located with regards to the ACC fronts and provide a useful case for comparison. In the Falkland Islands, acoustic studies of SW have identified low-frequency calls such as double downsweep signals centered at 60 Hz, as well as mid-frequency songs at around 2000 Hz (Cerchio and Weir, 2022). Overall, seasonal patterns show presence of HW, FW (20 Hz and 40 Hz pulses) as well as ABW (Z-calls), SW, PBW and MW.

Functional oceanographic domains

Functional Oceanographic Domains (FODs) describe regions that share the same temperature and salinity (TS hereafter) conditions, taking into account vertical dimension and allowing TS profile shape to be characterized. FODs were identified using Copernicus Marine Environment Monitoring Service (CMEMS) temperature and salinity composite products 1. Daily modelled TS profiles ranging from 20 m to 300 m depth were selected around Kerguelen Islands (Península Valdés), covering rectangular maps (1/12° horizontal resolution) ranging in latitude from 40°S to 60°S (37°S to 57°S) and in longitude from 45°W to 95°W (80°E to 45°E). These daily maps encompass the SES post-breeding foraging trips, from October to December of each year with deployments. The resulting sample of approximately 40 millions (Kerguelen + Península Valdés) TS profiles was analyzed using the methodology from Fonvieille et al. (2023). A bivariate principal component analysis for functional data (FPCA; Wang et al. (2016)) was performed to decompose the thermohaline vertical structure into modes of variability (Pauthenet et al., 2017). Clusters of TS profiles were then identified, both study areas combined, using a model-based clustering approach (Bouveyron et al., 2019), which was applied to the first three principal coordinates of the FPCA (e.q. the magnitude of a given mode of variability). Model-based clustering allows observations to be assigned a probability for each cluster. Only profiles with a probability threshold above 0.75 were assigned to an FOD. The rest were assigned to FOD 0.

The final step consisted in projecting SES TS profiles onto FPCA maps constructed from the model TS profiles, in order to attribute a FOD to each dive (Wang et al., 2016; Fonvieille et al., 2023). This method requires all profiles (both model and in-situ) to be defined within the same depth range. This depth range was chosen to on one hand maximise TS vertical variability inclusion, and on the other hand minimise SES profiles exclusion as SES perform shallower dives, especially at night. Profiles for which the SES did not dive deeper than 300 m were therefore excluded in this procedure. Model estimated FODs were assigned to such dives if a profile was available less than 6 km away, in alignment with the CMEMS spatial resolution. This method can accurately identify the frontal areas and specific water masses of the Southern Ocean as it uses a large number of TS model profiles covering a wide spatial range to construct the FOD clusters. Using only SES TS profiles to construct FOD clusters would introduce bias, as SES-sampled areas would be influenced by their foraging behaviour. Further methodological details on bivariate FPCA procedures can be found in Fonvieille et al. (2023); Nerini et al. (2022) and Pauthenet et al. (2017).

Generalized linear model

A generalized linear model with a binomial error distribution and a logit link function was used to explore the relationship between cetacean acoustic presence per species and oceanographic environments. The response variable was coded as a binary outcome representing the presence or absence of cetacean vocalizations for each drift phase. Fixed effects included the FODs as well as the bathymetric gradient computed using the GEBCO Bathymetric Compilation Group 2026, 2026 data. Bathymetric slope was derived from the GEBCO Bathymetric Compilation Group 2026, 2026 bathymetric grid after successive median filtering (scipy’s median filter) and spatial subsampling to reduce small-scale noise and computational cost and to keep only the large scale bathymetric gradient, such as the transition between the continental plateau and the Argentine Basin. The bathymetry was first smoothed with a 25 × 25 cell median filter and downsampled by a factor of 10, then smoothed with a 15 × 15 filter and downsampled by a factor of 8, and finally smoothed with an 11 × 11 filter and downsampled by a factor of 6, essentially subsampling the GEBCO grid down to a 2° resolution. The local bathymetric gradient was then computed in the latitudinal and longitudinal directions, converted from degrees to meters using latitude-dependent scaling, combined as the gradient magnitude, and interpolated along each seal track at the start time of each drift phase.

Frontal areas, or profiles corresponding to FOD 0, were removed from the dataset to fit the model, as the different ACC fronts were pooled in this cluster. While excluding annotations assigned to FOD 0 lowered the number of model inputs and could impact its statistical significance, it reduced the risk of associating a detection from a different FOD than the one occupied by the SES, due to the long range of cetacean vocalizations, thereby limiting spatial misclassification. Prior to analysis, predictor variables were inspected for collinearity and standardized to facilitate model convergence. Parameter estimates were obtained using maximum likelihood estimation. The model was only applied to species with enough annotations.

Drift phase dataset

The drift phase dataset is shown in Figure 5. Drift phase distribution varied between individuals and between the two study areas.

Very few drift phases took place in the first few days after individuals left their haul-out site. The drift phase dataset was therefore skewed towards open ocean waters with little sampling in coastal areas of Kerguelen and Península Valdés. The number of drift dives per day ranged between 0 and a maximum of 20, with an average of just under 4. Final dataset size can be seen Table 2.

Figure 5 shows that drift phases were distributed across a vast region of both the Atlantic and Indian sectors of the Southern Ocean, giving the opportunity to study cetacean presence across different frontal areas and environments. The nine individuals sampled areas ranging from the northern boundary of the ACC to the Southern ACC Front, visiting diverse bathymetric regions, all during the spring season.

Inter-annotator variability

Table 3 shows the inter-annotator agreement measured using Fleiss’ κ F before validation by the third annotator. Most vocalization categories exhibited substantial agreement between annotators, with values generally above 0.50. In particular, SW vocalizations achieved the highest agreement score (κ F = 0.83), while HW, ABW, delphinid, and sperm whale annotations showed high agreement levels. Lower agreement was observed for FW and MW vocalizations, with intermediate κ F values of 0.51 and 0.60, respectively. SRW vocalizations presented the lowest agreement (κ F = 0.37), indicating a higher level of ambiguity in their annotation.

Overall, these results suggested that the annotation protocol yielded generally consistent labels across most species categories, with variability comparable to that reported in the literature for expert annotation of cetacean signals in PAM data (Dubus et al., 2023), while also highlighting specific vocalization types that remained more challenging to annotate reliably.

Cetacean acoustic occurrence

In total, six species of mysticetes were annotated, with four mysticetes acoustically present both around Kerguelen and off Península Valdés, the ABW, FW, MW and SRW (Table 4). The ABW was the most annotated species with 134 drift phases containing vocalizations. HW were recorded in the Indian sector of the Southern Ocean, while SW were exclusively found in recordings from individuals from the Península Valdés population. Recordings from the Kerguelen population revealed a high acoustic occurrence of mysticetes, particularly ABW and HW, and to a lesser extent FW, whereas recordings from the Península Valdés population showed comparatively low acoustic occurrence across species. Although an important diversity of species were identified in the latter region, SW exhibited the highest occurrence amongst baleen whale species, with 2.8% of drift phase recordings containing vocalizations (Table 4).

Sperm whale acoustic occurrence was comparatively high in recordings from both populations, with vocalizations found in 4.3% and 7.0% of drift phases for the Kerguelen and Península Valdés populations respectively. Similarly, delphinid vocalizations were annotated in 1.3% and 2.5% of drift phases, respectively.

Average detection densities varied between 16.1% and 79%, and was high (above 33%) for ABW, delphinids and sperm whales in both populations. Overall, recordings from the Indian sector of the Southern Ocean exhibited higher acoustic occurrences of mysticete species, whereas recordings from the Atlantic sector showed a comparatively greater proportion of drift phases with odontocete annotations.

Annotation locations

For SES from the Kerguelen population, only a few drift phases were detected when the SES females were transiting across the Kerguelen Plateau (Figure 5a), limiting the possibility of acoustically detecting cetaceans in this area. Distinctions in acoustic presence as a function of bathymetry were therefore unclear. Figure 6a shows an important diversity of species in the Indian sector of the Southern Ocean including HW, ABW and FW as well as an important acoustic occurrence of sperm whales and delphinids.

SES from the Península Valdés population travel through areas with distinct bathymetric features including the extended shallow Patagonian Shelf and the abrupt slope towards the Argentine Basin (Figure 5b). Different species’ acoustic presence were found to be related to these distinctive features (Figure 6b). The shelf region was characterized by multiple occurrences of SW and one annotation of ABW (northern most annotation of the shelf, on the same day as a SW annotation), with no annotation of either species in deeper waters. Sperm whales were only found along the shelf break and over the steep bathymetric gradients (Table 5). Figure 6b also highlights cetacean presence in the Argentine Basin, with delphinids being acoustically present on multiple occasions. Calls of SRW and MW were also recorded in this offshore area.

For both study areas, no PBW vocalizations were found, however multiple biological signals were found that could not be attributed to a specific species.

Functional oceanographic domains

Figure 7 shows the different FODs obtained from the daily model data after clustering the principal coordinates of the FPCA. The areas where profiles could not be clustered into FODs, and were therefore assigned to FOD 0, corresponded to dynamic frontal areas and were associated to the ACC fronts. More specifically, comparing the FODs with the ACC fronts as defined by Orsi et al. (1995) highlighted that FOD 1 and FOD 2 were separated by the Southern ACC Front, while the region between FOD 2 and FOD 3 corresponded to the Polar Front. Finally, the zone between FOD 3 and FOD 4 was linked to the Subantarctic Front. FOD 5, north of the STF, corresponded to warmer subtropical water masses, while FOD 6 was associated to fresher subantarctic waters advected equatorward by the Humboldt Current system. These associations between FODs and fronts are coherent with previous results from Pauthenet et al. (2017) and Fonvieille et al. (2023). Detailed TS profiles from each FOD can be found in Supplementary Materials. Projecting SES TS profiles therefore made it possible to identify the water mass or frontal areas in which SES were located at the time a cetacean was present.

Combining Figures 6–8 highlights different cetacean acoustic occurrence distributions with regards to fronts. In the Indian sector of the Southern Ocean, HW were found mainly North of the Subantarctic Front, in FOD 3 and 5, near Crozet Island. Annotations of FW calls showed acoustic occurrence across multiple fronts, with high detection density in frontal areas. ABW were predominantly located around the Polar Front, either in this frontal area or in FOD 2 and 3.

Across both study areas, SRW and MW were annotated in different fronts and in frontal areas, with both species found in FOD 5. However too few drift phases contained annotations of these species calls to generalize. Sperm whales were annotated in recordings from most tagged individuals, regardless of the water mass, with a substantial proportion of occurrences recorded in the KER18–1 dataset, particularly in FOD 3 characterized by high ABW acoustic occurrence in the Kerguelen drift phase dataset and in frontal areas of the Atlantic sector of the Southern Ocean. Similarly, delphinids, with an acoustic occurrence only East of Kerguelen, were present both north and south of the Polar Front and found in FOD 4 in the Argentine Basin.

Annotations of SW and ABW over the continental plateau of Argentina could not be associated to any water mass due to the shallow bathymetry of the area.

Generalized linear model

Table 5 presents the significant results of the generalized linear model using bathymetric gradient and FODs as fixed effects. Three results were significant.

First, in the Indian sector of the Southern Ocean, the FOD number had a strong positive influence on HW acoustic presence, indicating that this species was more frequently found in warmer waters at lower latitudes during the spring season (β = 0.645, p < 0.001). In contrast, ABW had a higher probability to be acoustically present in higher latitude water masses (β = −0.565, p < 0.001). Sperm whales in the Atlantic sector of the Southern Ocean were found along important bathymetric gradients (β = 0.612, p < 0.001). No significant effects of either bathymetric gradient or FOD number were found for FW, sperm whale, or delphinid acoustic occurrences in either study area.

Using drift phases as listening point

Drift phases serve as natural listening points on account of the enhanced quality of their associated spectrogram. In contrast, annotating low-frequency vocalizations during the active swimming phases of dives is considerably more challenging, as these signals may be entirely masked by flow-induced noise.

However, focusing solely on drift phases introduces certain biases that must be carefully considered. The main bias stems from the non-random distribution of drift phases during an SES’ foraging trip (Figure 5). The number of drift phases per day is positively correlated with foraging effort and intensity, meaning that more drift phases occur in water masses with high biological richness (Piot, 2024). This skews cetacean acoustic occurrence to be higher in areas where SES forage more. This issue is addressed by comparing the detection density D rather than the number of drift phases with annotations.

Moreover, SES do not start entering drift phases directly until multiple days after they left their haul-out site. The first drift phases tend to start several hundred kilometers away from the departure location and generally beyond the continental shelf. This spatio-temporal gap implies a masking of the ecological traits characteristic of the species targeted. First, ecological representativity of cetacean abundance is biased toward offshore species, whose abundance is usually lower compared to that of shallow coastal delphinid species, meaning that annotation frequency in this dataset cannot be read as a proxy for true relative abundance across taxa. Moreover, this temporal bias introduces a further bias in the seasonal variation representativity of a wide range of cetacean species, with their foraging or reproductive phase not being sampled. For example, this is the case for the Southern Right Whale (SRW) where only foraging-related upsweeps were annotated while mating-associated gunshot calls remained entirely absent. This can be explained by SES leaving their haul-out site near the beginning of the SRW reproductive season around Penı́nsula Valdés. Therefore, cetacean presence in coastal areas can difficultly be assessed from the drift phase dataset considered here.

However, some species like the dusky dolphins (Aethalodelphis obscurus), which are known to frequent shallow coastal waters (Garaffo et al., 2007; Garaffo et al., 2011), could be detected in the first few hours of each dataset by annotating all dives and not focusing solely on drift phases. Other species, like the Commerson’s dolphins (Cephalorhynchus commersonii), could also be detected in both study areas. However, this would also necessitate a higher sampling rate for the tags due to the high frequency (110000 Hz) of their clicks (Martin et al., 2021). The high frequency vocalizations of such species are not covered by flow noise and could be annotated in the entirety of the PAM datasets. This approach would enhance monitoring of delphinid species and sperm whales, whose clicks and whistles exceed 3000 Hz and can remain detectable even during active swimming phases, as killer whale vocalizations have previously been detected in such data (Gros-Martial et al., 2025).

While restricting the analysis to drift phases allows for a rapid annotation of both mysticetes and odontocetes species, it introduces an irregular duty cycle in the otherwise continuous recordings, generating biases especially for short distance propagating clicks and whistles. This inherent duty cycle is less critical for mysticetes, whose low frequency vocalizations are constructed as long stereotyped songs and propagate over greater distances and can therefore be detected from further away. For instance, in the case of Z-calls, the impact of duty cycles was shown to be negligible (Michel et al., 2024).

Finally, as shown Figure 3, the listening points used for annotation contain low frequency harmonics due to flow noise in the 80 Hz to 110 Hz frequency bandwidth (visible Figure 3). This makes the annotation of baleen whale calls, such as Sei Whale (SW) and Fin Whale (FW) upsweeps, Antarctic Blue Whale (ABW) D-calls (frequency-modulated downsweeps) or Minke Whale (MW) calls challenging, both due to the similar frequency range but also the analogous shape. Placing the hydrophone toward the back of the animal is expected to improve signal quality and reduce the presence of these flow-induced harmonics, potentially resulting in better detectability for some baleen whale vocalizations. Because flow noise is generated by pressure fluctuations as water moves over the hydrophone, positioning the sensor in a more downstream region characterized by reduced turbulence may limit its exposure to high-velocity flows and thereby decrease noise contamination (von Benda-Beckmann et al., 2016; Fiore et al., 2017). Positioning the hydrophone in this manner is not expected to have a significant influence on other sound sources.

Modelling cetacean presence with environmental variables

The current generalized linear model captures broad scale relationships between acoustic occurrences for specific species and different oceanographic structures. These first promising results, with significant relationships between the distribution of ABW and Humpback Whale (HW) and in-situ thermohaline profiles, opens the door to more complex models. Because the tags log other oceanographic variables, additional environmental and behavioural predictors could further improve model performance and interpretability. In particular, variables such as bioluminescence, chlorophyll- a proxy, or the tagged individual’s number of prey capture attempts could provide valuable information on the biological richness of the water masses through prey availability and trophic interactions, and were used in a previous preliminary study (Gros-Martial et al., 2025). These in-situ observations could be important drivers of cetacean distribution and would increase our understanding of their habitat use and migration patterns with regards to fine scale oceanographic and biological conditions rather than environmental data from models.

What’s more, while profiles with low FOD attribution probabilities were aggregated into a broad frontal area category, explicitly separating them into the different ACC fronts (Subantarctic Front, Polar Front, and Southern ACC Front) would increase the number of acoustic occurrences used to fit the model and improve its explanatory power. This would improve the ability of the model to distinguish species specific associations with particular water masses and frontal systems.

New contributions to PAM for cetacean monitoring

This study addresses cetacean presence within two considered regions with a focus on spring, a season of ecological importance as whales are migrating between breeding and foraging grounds. Results are compared to existing studies, summarized in Table 6, which offer useful comparison data. However, these studies are limited either by their sample rate that restricts analyses to mysticetes, or by locations that differ relative to Southern Ocean fronts and therefore in species presence and acoustic behavior at a given time. The extensive spatial coverage provided by roaming SES offers valuable insights into species behavior during this season, a period of major ecological significance characterized by both whale migration and phytoplankton bloom, leading to enhanced ocean productivity (Ardyna et al., 2017). The acoustic data provide a valuable complement to existing studies by enabling fine-scale temporal analyses at both study sites, while also facilitating the monitoring of sperm whales and delphinids, whose habitat use remains poorly understood and has so far been investigated primarily through studies focusing on depredation (Richard et al., 2022; Miller et al., 2024). Very few PAM platforms for the study of cetaceans have previously been deployed in the two study areas - the Polar Front zone around Kerguelen Island, and the Argentine Basin to the subantarctic zone of the Atlantic Ocean.

Mysticetes

Acoustic monitoring context

Mysticetes have been extensively studied in regions frequented by the two SES populations. In the southern Kerguelen Plateau, an AAD-MAR hydrophone deployed near Heard Island at 1050 m depth and sampling at 12000 Hz has enabled monitoring of both sperm whales and mysticetes (Miller et al., 2024). In parallel, the OHA-SIS-BIO network has provided continuous low-frequency recordings (up to 120 Hz) from 2010 to 2023, supporting long-term monitoring of baleen whales (Royer, 2009; Leroy et al., 2018a; Torterotot et al., 2020). Two hydrophones from this network, WKER and ELAN, are located within the present study area.

In the South Atlantic Ocean, baleen whale acoustic studies have mainly focused on localized or coastal deployments. For example, hydrophones around the Falkland Islands recorded at 24000 Hz (Cerchio and Weir, 2022), while sonobuoys near South Georgia sampled at 48000 Hz (Calderan et al., 2021).

Antarctic blue whale (ABW) occurrence

At WKER and ELAN, the OHASIS-BIO hydrophones, ABW were detected between 2017 and 2022, with near year-round presence and a marked peak beginning in September, sometimes extending through spring. Data from the OHA-SIS-BIO network indicate that ABW can also occur north of the Subtropical Front during spring (Torterotot et al., 2020). Tag-derived SES acoustic occurrence shows similar temporal patterns but emphasizes higher occurrence near the Polar Front and only limited presence north of the Subantarctic Front. Together, these results suggest reduced ABW acoustic presence at lower latitudes during austral spring.

Fin whale (FW) occurrence

FW acoustic occurrence around Kerguelen, estimated from the drift phase dataset, is similar to results from Aulich et al. (2022) with presence off Heard Island, on the Kerguelen Plateau, in the middle of October. FW 20 Hz pulses were otherwise detected around Australia during the austral winter and around Antarctica in the austral fall. However, as their acoustic presence was mainly associated with frontal areas, no significant results emerged between their occurrence and FODs. Defining specific frontal areas, instead of combining them into FOD 0, could shed light on spatial distribution.

Antarctic minke whale (MW) occurrence

Antarctic minke whales were only sporadically present in the drift-phase dataset, typically as single-day events. This species is known to expand its latitudinal range during spring, contrasting with a more coastal Antarctic distribution during the rest of the year (Bombosch et al., 2014). Annotations from individual KER17–1 indicate that MW can be found further north than the Subantarctic Front during this period. However, estimating MW call detection ranges would improve distribution assessments.

Southern right whale (SRW) occurrence

Similarly to MW, SRW were only occasionally found in the drift phase dataset. The annotations likely reflect early seasonal presence at high latitudes as subantarctic islands, such as Crozet and Kerguelen, are known foraging grounds in late spring and summer (Sprogis et al., 2024). Satellite tracking shows that SRW migrate south and east during November and December toward subantarctic feeding grounds in the Argentine Basin (Zerbini et al., 2016). These individuals originate from Pen´ınsula Valdes,´ where Golfo Nuevo and Golfo San Jose´ provide sheltered breeding habitats (Payne, 1986). Their vocal behavior includes gunshot calls associated with mating (Parks et al., 2005) that were not found in the drift phase dataset, upsweeps linked to foraging (Calderan et al., 2021), which were recorded in the Argentine Basin, a known feeding and transit area. With the first drift phases occurring only after a couple of days at sea, vocalizations linked to mating behavior might be omitted after curation of the dataset.

Humpback whale (HW) occurrence

A high acoustic occurrence of HW calls was recorded north of the Subantarctic Front for individual KER17 1. This is consistent with migratory movements from low-latitude breeding grounds to high-latitude feeding areas. This interpretation aligns with acoustic observations of HW near South African coasts in September–October and in Antarctic waters from March to May (Shabangu and Kowarski, 2022).

Sei whale (SW) occurrence

Previous studies have shown that, around the Falkland Islands, SW are present from early December onward and persisted throughout summer, with downsweeps dominating early in the season and songs or L-calls appearing mainly from March to June (Cerchio and Weir, 2022). This seasonal pattern is consistent with the present study, where low-frequency downsweeps were annotated in October and November with no L-calls being found, likely reflecting southward migration toward foraging grounds. SES data further indicate high SW acoustic occurrences over the Patagonian Shelf and minimal presence in the Argentine Basin, possibly due to track differences, with PV18–1 following a similar southward migration route than SW.

Pygmy blue whale (PBW) absence

Pygmy Blue Whales (PBW) were regularly detected in the OHA-SIS-BIO dataset at WKER and ELAN during fall and early summer. Their absence in the SES drift-phase dataset reinforces this result and suggests that this species does not frequent subantarctic waters during spring.

Community structure and latitudinal segregation

Overall, acoustic occurrences align well with previous studies, showing comparable acoustic diversity and occurrence around subantarctic islands. The relative proportions are similar, with dominant occurrences of HW, ABW, and FW, and only limited MW annotations.

A clear latitudinal structuring emerges with i) ABW concentrated near the Polar Front, ii) FW occurring between the Subantarctic and Polar Fronts, and iii) HW mainly annotated north of the Polar Front, and confirmed by the effect of FOD on their acoustic occurrence, with a positive and negative relationship for HW and ABW respectively. These patterns likely reflect differences in migration timing, which is coherent with Southern Indian Ocean and Antarctic presence from other studies (Aulich et al., 2022; Torterotot et al., 2022). This interpretation is consistent with Mackintosh (1965), who described a temporal migration lag in baleen whales, with larger species reaching higher latitudes earlier (ABW then FW then HW then SW). Minke whales appear to be an exception, lacking clear migratory timing. This framework may also explain the absence of SW annotation in the Indian sector of the Southern Ocean, as individuals may still remain at lower latitudes during the study period.

Overall, acoustic occurrences are consistent with existing literature across both the Southern Indian and South Atlantic Oceans. They confirm a strong seasonal structuring of baleen whale presence with clear latitudinal segregation linked to oceanographic fronts (or FODs), and species-specific migration timing shaping acoustic distributions. These results reinforce the importance of combining fixed hydrophone networks and animal-borne sensors to better resolve spatial and temporal patterns in baleen whale ecology.

Sperm whales

Sperm whale clicks were annotated in seven of the nine datasets, with occurrences along the Argentine continental slope and in the Indian sector of the Southern Ocean (Figure 6).

Indian sector of the Southern Ocean

Acoustic occurrences in October and November between Crozet and Kerguelen Islands are consistent with photo-identification data collected from Patagonian toothfish fisheries, which confirmed the presence of male sperm whales in the area (Labadie et al., 2018). They show a clear spatial gradient, with a higher number of annotations in the southern part of the study area, along the Polar Front, and only limited presence near the Subantarctic Front. This distribution suggests that the individuals are likely migrating males moving southward toward Antarctic waters, where sperm whales are acoustically present during summer but largely absent in spring (Miller and Miller, 2018).

Atlantic sector of the Southern Ocean

In the Atlantic sector of the Southern Ocean, visual surveys report sperm whale presence northeast of Península Valdés and between the Falkland Islands and South Georgia (Waerebeek et al., 2010). The present study shows consistent patterns, with acoustic occurrence concentrated around the Subantarctic Front, particularly over the continental slope. This spatial association highlights the ecological importance of frontal systems and steep bathymetric gradients as foraging habitats. Similar patterns have been observed in the North Atlantic, where sperm whale detection densities are higher around oceanic fronts and eddies (Virgili et al., 2019).

Acoustic behaviour and methodological limitations

Sperm whales primarily emit clicks for echolocation and foraging, but also produce other vocalizations, including slow clicks, buzzes, and codas. Among these, buzzes are specifically associated with prey capture attempts (Miller et al., 2004). In this study, most annotations correspond to clicks, and vocalization types were not differentiated. Incorporating signal classification in future analyses could refine ecological interpretations, particularly by clarifying relationships between foraging behaviour, vocal activity, and environmental features such as fronts and eddies. Overall, sperm whale acoustic occurrence from the drift phase dataset are consistent with previous visual and acoustic studies across both ocean basins. They emphasize a strong association with major oceanographic fronts, preferential use of continental slope habitats and likely seasonal southward migration of male individuals. Further differentiation of vocalization types would provide valuable insights into behavioural states and habitat use.

Delphinids

PAM studies targeting delphinids in offshore areas of the Southern Ocean remain limited with most available data coming from coastal and opportunistic boat-based studies.

Indian sector of the Southern Ocean

Short-term acoustic recordings around the Crozet Islands, using hydrophones deployed on demersal longlines, have primarily focused on depredation by killer whales (Richard et al., 2022). In the Kerguelen Islands and across the Kerguelen Plateau, early sighting and stranding records report occasional presence of multiple species, including strap-toothed whales, killer whales, pilot whales, hourglass dolphins, dusky dolphins, the spectacled porpoise, and the endemic Commerson’s dolphin (Robineau, 1989). More recent ship-based observations from the Marion Dufresne confirm the presence of southern right whale dolphins, a circumpolar pelagic species inhabiting subantarctic waters (Lipsky, 2009). Additional evidence includes mass strandings of pilot whales on Kerguelen shores (Fontaine et al., 2015) and observations from Patagonian toothfish fisheries, which indicate their occurrence along the shelf break.

Atlantic sector of the Southern Ocean

A similar diversity of delphinids has been documented in the Argentine Basin and over the Patagonian Shelf. Acoustic monitoring has confirmed the presence of Peale’s dolphins in shallow waters through detection of clicks and whistles (Martin et al., 2024; Malige et al., 2025). Further north, PAM studies off Península Valdés have targeted Franciscana dolphins (Giardino et al., 2025). Coastal observations in Argentina report frequent sightings of dusky and Commerson’s dolphins around Península Valdés (Garaffo et al., 2007, Garaffo et al., 2011), while Risso’s dolphins have also been documented along South American coasts (Boy et al., 2011; Riccialdelli et al., 2011). Offshore, research cruises have recorded hourglass dolphins, southern right whale dolphins, and pilot whales near the continental slope (Waerebeek et al., 2010), however, hourglass dolphins are usually observed at higher latitudes (Dellabianca et al., 2012).

More broadly, the Southern Ocean and subantarctic regions also host other odontocetes, including pygmy sperm whales, false killer whales, bottlenose dolphins, potentially southern bottlenose whales, and several beaked whale species (Waerebeek et al., 2010).

Acoustic behaviour and methodological limitations

The present results confirm that delphinids are present in offshore environments in both the Atlantic and Indian sectors of the Southern Ocean. They also highlight the potential of SES borne acoustic recorders to improve knowledge of their distribution and habitat use.

Although species-specific identification was not achieved in this study, expanding the number of deployments could help in the development of an acoustic catalogue of high-latitude delphinid vocalizations. Given that delphinids produce relatively stereotyped clicks and whistles, future work could enable species discrimination, leading to more precise assessments of spatial distribution and ecological preferences in these poorly studied offshore environments.

Overall, despite sparse dedicated PAM studies, multiple independent data sources (acoustics, sightings, and strandings) converge to show a diverse and widespread delphinid community across subantarctic and offshore Southern Ocean regions. PAM data from tagged SES represents a promising approach to bridge current observational gaps and refine our understanding of these species.

Limitations and potential of biologging for cetacean monitoring

This study should be seen as a proof of concept relying on biologging data from one species for the purpose of studying other species of interest relatively uncommon and mainly inaccessible to sampling. For instance, during the Detection, Classification, Localisation and Density Estimation 2024 conference 2, only 5.5% of presentations used biologging data, all of which aimed at studying vocalizations from the tagged species and individual. However, PAM data from biologging presents significant opportunities as demonstrated by Noda et al. (2024), with deep learning models producing real-time classification of the soundscape, from geophony, anthrophony to biophony, but also for the identification of multiple species ranging from fish to cetaceans. While the scope of this research differed, the long-run objective is to obtain real-time and species-specific monitoring of cetaceans in the Southern Ocean.

Elephant seals as ideal PAM platforms

Given that SES, after weeks of fasting, forage in the pelagic zone, constantly searching for areas of high foraging interest, they make remarkable platforms to acoustically monitor cetaceans that forage or transit in the same areas. Furthermore, SES are believed to produce little to no vocalizations underwater. The presence of click trains found in northern elephant seals acoustic dataset is so far the only proposed underwater vocalization for the species (Burgess et al., 1998) but was not found here. Therefore, geophonic and biophonic sources are the principal sources of noise during drift phases, in areas with little anthropogenic activities. Transient biophonic sources found in the acoustic datasets can therefore be easily identified, whether through manual annotation or supervised and unsupervised algorithms.

Sample size

The relatively limited sample size, with only nine SES tagged with hydrophones, limits the generalisability of cetacean acoustic distribution patterns. Firstly, while SES behavior enables large spatial coverage, each point is effectively sampled only once, and only large scale analysis is possible. Furthermore, as cetacean acoustic occurrence depends on vocal activity, the limited number of individuals may lead to uneven detection densities across the study area, biasing apparent distribution patterns toward specific SES positions. Secondly, using data from multiple years into a single analytical framework introduces additional uncertainty, as it assumes that cetacean migration and habitat use is constant from year to year. However environmental variability and prey availability could lead to interannual shifts in distribution and acoustic detectability. These limitations indicate that cetacean distribution interpretation should be made with caution and underscore the need for additional deployments to improve spatial and temporal coverage.

Additional deployments, especially on females, would increase temporal and spatial coverage within the current study areas, whereas tagging male SES would extend spatial coverage to the Antarctic continent. Tagging both male and female SES after the breeding season would effectively provide a larger snapshot of the Southern Ocean during the migration period, improving the latitudinal interpretation of acoustic occurrence. Long-term deployment strategies could focus on improving temporal coverage by extending battery life, reducing sampling rate, or deploying tags on SES during the moulting season to document cetacean acoustic occurrence during the austral fall and winter periods.

Ethical costs

Tagging SES raises important ethical considerations related to human intervention and the potential disturbance of the animals during the mating season. Although biologging provides valuable ecological and behavioural data, it is necessary to ensure that deployments minimize animal disturbance and do not alter natural behaviour. Before concluding that biologged SES constitute an effective PAM tool, the limited sampling window achieved here must be weighed against the logistical effort and ethical costs of this approach, which requires capturing, instrumenting, and recapturing reproducing females. By comparison, the OHA-SIS-BIO network provides continuous temporal coverage across nine sites in the Southern Indian Ocean, while gliders offer a comparable mobile alternative without the welfare cost of instrumenting a live animal. Despite these two points against biologging, the approach is justified by its opportunism relative to the in-situ measurements already being collected by the biologged individuals: elephant seals contribute to more than 80% of high frequency temperature and salinity profiles collected south of 60°S and 95% within the pack ice zone (McMahon et al., 2025). PAM sensors are therefore added opportunistically alongside the oceanographic tags already deployed, making the additional instrumentation cost relatively limited. PAM recording on elephant seals should therefore be regarded as a complementary source of data to an already existing and important biologging infrastructure, particularly well-suited to linking both low and high frequency vocalizations to co-localized in-situ habitat features in the biologically important areas sampled by SES.

Detection range

Propagation distance varies importantly between species, with different source levels and vocalization frequencies (see Section 2.3). For instance, ABWs and FWs can be heard up to 1700 km in spring (Shabangu et al., 2020) while HW songs have a maximum propagation distance estimated between 18 km and 45 km (Shabangu and Kowarski, 2022). Similarly, in coastal areas, SW calls have a detection distance between 10 km and 15 km (Baumgartner et al., 2008), while northern right whales upcalls have a supposed detection distance between 9 km (Clark et al., 2010) and 40 km (Thode et al., 2017), which can be extended to SRW upcalls. Sperm whales have similar detection distances and produce highly directional clicks that can be detected up to 30 km away (Leaper et al., 2000). No study exists about communication spaces for many delphinid species present in the study areas, such as hourglass dolphins or southern right whale dolphins. Studies in coastal areas estimate that bottlenose dolphin whistles may have detection ranges up to 25 km for low frequency whistles (under 10000 Hz) and below 4 km for high frequency whistles (Janik, 2000). Other studies report that Risso’s dolphin clicks can be detected up to 4.1 km (Frasier et al., 2016) while killer whale calls reach distances between 5 km and 16 km depending on the type of whistles and their pulsed calls have an estimated detection range of just under 8 km (Miller, 2006; Mouy et al., 2025). While these detection distances vary depending on sea state, bathymetry, the frequency of the signal and the individual’s depth, they remain good approximations of each species’ detection range.

The short windows of analysis provided by the drift phases, lasting between 3 to 8 minutes in general, makes the identification of both the FW 20 Hz pulse and the ABW Z-call difficult, with only a handful of calls possibly visible per phase due to the periodicity mentioned in Section 2.3. However, these signals were present in most of the data from two SES from the Kerguelen population. More generally, distant baleen whale vocalizations were harder to annotate, explaining the differences between the two annotators through the κ F metric. Delphinid clicks and whistles, although limited in propagation range, were often clearly visible, leading to a high κ F metric and strong agreement between the annotators. While dolphin vocalizations, whether buzzes, whistles or clicks, do not compose songs, they are emitted with lower intervals than most baleen whale calls, making their annotation during short drift phases easier and more certain.

Furthermore, detection ranges introduce an inherent spatial uncertainty, with annotations reflecting the SES’ position rather than the true location of the cetacean. Consequently, the oceanographic conditions (FOD and bathymetry) assigned to an annotation should be interpreted as those of the recording platform rather than the exact location of the vocalizing animal. These potential position errors must be considered when interpreting results and represent a source of error in the generalized linear model. Model associations should therefore be interpreted at broad regional rather than individual scales. This limitation is particularly relevant in frontal areas, where SES forage near boundaries between FODs. Further work should aim at estimating the distance between the cetacean and the SES by integrating received levels with source level estimates from the literature as a proxy for the distance of the annotated call from the hydrophone. This approach would help refine spatial interpretation and possible cetacean habitats, and is particularly important to accurately model cetacean acoustic presence in relation to in-situ environmental conditions.

Most PAM studies focus on restricted areas, limited by the detection area of fixed hydrophones. Similarly to gliders, biologged SES facilitate the sampling of vast portions of the poorly sampled Southern Ocean, which used to host the largest baleen whale populations. This study’s annotations make up the first assessment of cetacean occurrence from biologging data in the Southern Ocean, demonstrating that biologged southern elephant seals provide a valuable complementary PAM platform for remote Southern Ocean monitoring, while future studies combining concurrent visual, satellite and fixed acoustic observations will be necessary to fully quantify spatial uncertainty. These first results open the door to further research to assess cetacean behavior with regards to oceanographic conditions, effectively leading to a better characterization of their habitat.

Combined with in-situ data from the tags, logging temperature, salinity, chlorophyll- a, and bioluminescent flashes, the acoustic presence of cetaceans could be modeled in relation to environmental variables with unprecedented resolution. The available data therefore provides an opportunity to be the first co-localized dataset of PAM annotations of cetaceans with in-situ environmental variable profiles, providing a novel and promising approach for cetacean conservation. Other variables could include PAM based wind speed observations or the individual’s prey capture attempts as a proxy for the biological richness of the water mass. On the longer term, acoustic monitoring via biologged SES may contribute to improve current assessments of the occurrence of cetaceans, especially baleen whales, in order to better monitor their recovery.