ABSTRACT Cetaceans are widely regarded as sentinels of marine ecosystem health, yet their monitoring in the Indian Ocean remains challenging because conventional survey techniques often lack adequate spatial and temporal resolution. In this study, we developed and evaluated a deep learning‐based detection framework for cetaceans in Indian waters using high‐resolution imagery collected during research cruises conducted between 2024 and 2025. The framework employs the YOLOv11 object detection architecture, trained and validated on field‐based images of three representative species: Balaenoptera musculus (blue whale), Globicephala macrorhynchus (short‐finned pilot whale), and Stenella longirostris (spinner dolphin). To improve detection reliability, the pipeline integrates ensemble‐ and rule‐based post‐processing strategies, including weighted box fusion (WBF) and a novel adaptive distance‐aware refinement and intelligent fusion (ADRIF) algorithm designed to reduce duplicate detections and false positives in complex marine science applications. The optimized framework achieved a mean average precision (mAP@0.5) of 81.6% and an F 1‐score of 80.05% on the test dataset, demonstrating robust detection performance under variable oceanographic conditions. These results highlight the potential of deep learning–based approaches to enhance the efficiency and consistency of visual cetacean detection from vessel‐based imagery. The proposed framework provides a reproducible methodological foundation for integrating computer vision tools into marine mammal research and future large‐scale monitoring efforts.
ABSTRACT The waters surrounding the southern Iberian Peninsula are a key conservation area for the long‐finned pilot whale ( Globicephala melas melas ), particularly in the pelagic regions of the Strait of Gibraltar, the Alboran Sea, and the Gulf of Vera, where upwelling, complex bathymetry and frontal dynamics enhance productivity and prey availability, attracting cetaceans year‐round. This study presents the first published satellite tracking data for this species in the Mediterranean Sea, with six individuals tracked between March and November 2011 (21–93 days; 235–410 filtered locations per individual). We used boosted regression trees to model habitat suitability and identify the environmental factors associated with the distribution of long‐finned pilot whales. To evaluate the relative importance of physiographic features and oceanographic conditions, we fitted three models using static predictors, dynamic predictors, and both combined. The combined model showed the highest predictive accuracy (AUC = 0.87), followed by the static‐only (0.86) and dynamic‐only models (0.73). Habitat suitability was primarily driven by bathymetry and distance to shore, with whales preferentially using the continental slope (500–1500 m depth) near the shelf break. Among oceanographic variables, eastward surface currents were positively associated with habitat suitability, likely reflecting Atlantic inflow sustaining frontal systems and prey aggregation in the Alboran Sea. These findings can inform conservation planning and prioritization within existing protected areas.
ABSTRACT Cetacean epibionts offer noninvasive insights into host health, physiology, and spatial ecology, yet quantitative assessments in wild populations remain scarce. Although vessel‐based observations may be limited by field of view and analyst input, unoccupied aerial vehicles (UAVs; drones) are increasingly used to monitor cetacean health indices, enhancing observational availability and reproducibility. However, the suitability of drones for quantifying epibiota has not been thoroughly evaluated. Implementing a standardized workflow, we compared dorsal proportional epibiont coverage estimates derived from drone‐based imagery and traditional vessel‐based techniques across reproductive classes of humpback whales ( Megaptera novaeangliae ). We analyzed 32 paired datasets using pixel‐based methods, incorporating semiautomated graph‐cut segmentation and iterative color‐thresholding. Drone‐derived estimates of epibiont coverage consistently exceeded vessel‐based measures, which averaged 58% of the relative coverage estimated from corresponding drone imagery, partly attributable to variable lighting during image capture. Nevertheless, 94% of paired observations differed by less than 5% in absolute coverage, indicating strong agreement. While refinement of flight parameters, image resolution, and further validation are recommended, we demonstrate that drones can provide a reliable source of quantitative epibiont data, with potential applicability across diverse epibiont–host systems, benefiting cetacean population health monitoring programs and informing conservation strategies, particularly for elusive or understudied species.
ABSTRACT North Pacific humpback whales are a celebrated conservation success, rebounding from extensive commercial whaling. Yet they remain vulnerable to the impacts of climate change, a multifaceted stressor that can exacerbate existing threats. During 2013–2016, a Northeast Pacific marine heatwave severely reduced forage availability in their feeding grounds, concurrent with diminished whale recruitment and increased mortality. We analyzed photo‐identification data of whales that visited the Main Hawaiian Islands from 2002 to 2023 to estimate apparent survival, abundance, and distinct population segment (DPS) growth rates. Survival was high prior to (0.987; SE = 0.003), low during (0.684; SE = 0.022), and higher, but still reduced, after (0.956; SE = 0.006) the heatwave. Annual breeding season abundance estimates averaged 11,617 (SD = 1914) and included a striking 37% decrease from 2014/15 to 2021/22. While the average annual DPS growth rate was stable (1.00) over 20 breeding seasons, shorter periods included high growth (1.09) and steep decline (0.783). Depressed survival and abundance estimates may signal lingering heatwave effects or adjustment to current carrying capacity. Climate change predictions include more frequent, intense, and longer heatwaves; thus, this DPS serves as a crucial ecosystem‐health indicator. Continued research and monitoring are essential to support resiliency against future climate perturbations.
ABSTRACT Cetaceans are prone to facing deadly situations near coastal areas, from potential vessel collisions to natural stranding events. In particular situations, animals produce distress calls that may signal a pending negative situation. In such situations, stranding response networks and authorities could benefit from an alert system using real‐time passive acoustic monitoring, which requires adequate call identification and to set a reliable alarm to trigger a response. Therefore, an alarm system for S10 gray whale putative distress calls is presented. The system, trained with a set of 1177 two‐second segments with S10 calls, uses a Resnet‐18 convolutional neural network and transfer learning to identify the S10 call with a precision of 0.9891. Once the S10 call is confirmed by redundancy within a time frame, an alarm is sent through Short Message Service to members of the local stranding response network. The low‐cost system is easy to replicate and upgrade by adding more models, and BirdNET front end facilitates its widespread use.
ABSTRACT Multiple killer whales of the Iberian subpopulation interact with boats by establishing physical contact through bumping and pushing. First reported broadly in 2020, this disruptive behavior has since caused considerable damage to hundreds of vessels. Even though there are various hypotheses for what drives these interactions, the cause and purpose of this behavior are not yet understood. Here, we report on seven events between 2018 and 2025 where killer whales made direct physical contact with a research vessel in southern Portugal. Out of 26 identified individuals, 8 interacted with our rigid‐hulled inflatable boat: 2 adult females and 6 juveniles of both sexes, from 5 recognized matrilines. Notably, one juvenile interacted on two occasions. Underwater footage revealed a variable nature of interactions, ranging from displays of curiosity to agonistic behaviors. In all encounters, feeding/foraging was observed. We documented an interaction that predates official reports by 2 years and recorded a new individual participating in this behavior, highlighting its continued spread through social transmission. In our observations, keeping the vessel stationary did not reliably decrease interaction intensity, while accelerating caused apparent loss of interest. However, improved mitigation measures are needed to prevent harm to humans and killer whales, securing long‐term coexistence.
ABSTRACT The behavior, distribution, and life history of sei whales ( Balaenoptera borealis ) are largely unknown in the North Pacific. We analyzed data from three bottom‐mounted acoustic recorders deployed 25 (shelf; depth = 300 m), 45 (slope; depth = 630 m), and 65 (abyssal; depth = 2860 m) nautical miles off of Newport, Oregon, USA that recorded continuously from October 2021 to December 2022. Acoustic data were visually and aurally reviewed in search of sei whale calls; putative calls were annotated and assigned confidence levels based on existing descriptions in the literature. High confidence sei whale calls were characterized as broadband downsweeps with an average frequency range of 103–30 Hz and a duration of 2.73 s. Calls occurred in groups ranging from 1 to 32 calls with an average of 9 calls per bout. We document a strong peak in putative sei whale calls in October and November and a significant diel pattern (Hermans‐Rasson Test, p = 0.001) with decreasing effect size at more offshore locations. Our findings indicate that sei whales may be present in this region more frequently than previously thought, providing key information on sei whale occurrence in the North Pacific to guide future conservation management efforts.
ABSTRACT The pygmy ( Kogia breviceps ) and dwarf ( Kogia sima ) sperm whales are cryptic cetaceans characterized by inconspicuous surfacing behavior, small group sizes, and deep‐diving habits, making visual detection particularly challenging. Here, we present the first broad‐scale passive acoustic monitoring (PAM) assessment of Kogia spp. occurrence along the Brazilian coast, spanning both the Equatorial Atlantic Ocean (EAO) and Southwestern Atlantic Ocean (SWAO). Across 25 survey campaigns, 504 survey days, and more than 6441 h of acoustic recordings, 94 acoustic detections were identified, predominantly within the Brazilian Equatorial Margin (BEM), resulting in substantially higher detection‐per‐unit‐effort values in the EAO than in the SWAO. Two rare visual‐acoustic encounters of Kogia sima were also documented, including one detection incorporated into the dataset. A total of 2796 echolocation clicks were analyzed, exhibiting the characteristic narrow‐band high‐frequency acoustic profile of Kogia spp. (mean peak frequency = 126.2 ± 3.8 kHz; mean −3 dB bandwidth = 4.7 ± 2.56 kHz; mean inter‐click interval = 141 ± 97 ms). Detections occurred across a broad bathymetric range (9–4411 m), and kernel density analyses revealed spatial heterogeneity in occurrence patterns, with recurrent detection areas concentrated within the BEM. Collectively, these findings demonstrate the effectiveness of PAM for studying elusive deep‐diving cetaceans and provide an important baseline for future ecological monitoring, impact assessment, and conservation efforts targeting Kogia spp. in the western Atlantic.
ABSTRACT Machine learning (ML), particularly the availability of deep neural networks, has enabled the proliferation of modern artificial intelligence (AI) and has revolutionized data processing capacity in marine mammal science. The development and application of these systems has delivered 1000‐fold increases in throughput for applications such as photo identification and orders of magnitude increases in performance for passive acoustic detection of calls. Some AI systems now exceed the performance of human experts. As marine environments face unprecedented and compounding pressures from climate change, habitat degradation, and anthropogenic disturbance, these technological advances offer powerful promise. However, the success of AI applications depends less on algorithm sophistication than on the underlying information systems that enable their functionality. Here, we argue that marine mammal scientists must approach AI development with intentional agency—as captains rather than crew—emphasizing system‐level design, interpretability, human oversight, and collaborative frameworks. We examine the spectrum of AI capabilities today, from pattern recognition where ML excels to context judgments and value judgments which must remain defined and decided by humans. By designing AI systems that center domain expertise, prioritize accessibility, and facilitate community engagement, the marine mammal science community can harness computational power while maintaining scientific rigor and advancing equitable conservation outcomes.
ABSTRACT Two coastal dolphin species, the common bottlenose dolphin ( Tursiops truncatus —Tt) and the short‐beaked common dolphin ( Delphinus delphis —Dd), co‐occur along the southern Israeli Mediterranean coast, where the critically endangered Dd population faces ecological pressures, potentially including competitive exclusion by that of the Tt population. In this context, reliable acoustic differentiation between the two species is needed to supplement visual differentiation. This study aimed to assess whether species‐specific whistle characteristics could be used to acoustically distinguish between the two taxa using a novel manual contour extraction tool, Sound Annotator, combined with machine learning classification. A total of 940 high‐quality whistles (705 Dd, 235 Tt) from field recordings collected between 2018 and 2024 were available for analysis. Sixteen frequency, temporal, and shape parameters were derived from the extracted contours and analyzed with multiple classification models. Principal component analysis revealed high acoustic overlap, yet the XGBoost classifier effectively classified individual whistles, with minimum, median, and beginning frequency emerging as the most informative features. This framework facilitates non‐invasive, species‐specific monitoring of dolphin populations in challenging visual conditions, particularly at night, providing a foundation for conservation‐focused acoustic monitoring in the eastern Mediterranean.
Knowledge of mother-calf pairs, migratory patterns, female reproductive cycles, and birth-year calf development are all essential for understanding risks in these critical life stages and monitoring little-known dwarf minke whale (Balaenoptera acutorostrata) populations. Rare great barrier reef (GBR) sightings of mother-calf pairs accumulated by the Minke Whale Project from 2003 to 2023 plus sporadic southeastern Australian sightings were analysed. Imagery was used to identify individual whales, and calf length and breathing rate data were collected. Mother-calf spatiotemporal distribution, calf development, and calving intervals of resighted females were analysed. From > 20 years accumulated data, 215 of 293 (73.4%) sightings were from the GBR austral winter aggregation area, and 76 (26.6%) along the southeastern coastline. Sightings shifted north until late May and south from August, with no spatial shifting over the June-July aggregation period. Combining calf length and breathing rate data, this suggests calving occurs in southern regions during the northbound migration, and a GBR calving ground is unlikely. Limited resightings of mother-calf pairs confirm some remain in the GBR for extended periods, and calf morphometric data suggests a doubling in length over the first year. Calving intervals of 1-3 years were recorded, using resighting histories of GBR females. This research highlights the need for investigating dwarf minke spatiotemporal patterns of migration, particularly of mothers with calves, to better inform population monitoring, risk management, and the development of predictive models.
Haul-out behavior is a key component of the behavioral ecology of pinnipeds and serves essential resting functions. Various intrinsic and extrinsic factors have been shown to determine the occurrence of haul-out events. However, in the Baltic Sea, a nontidal, semi-enclosed shelf sea, the drivers of haul-out behavior in seals are not well characterized, especially in juveniles. In this study, 14 rehabilitated juvenile gray seals (Halichoerus grypus) were equipped with Argos satellite transmitters to enable post-release monitoring and to investigate the spatial and temporal aspects of haul-out behavior. The animals dispersed widely in the Baltic Sea and hauled out along coastal areas that largely overlapped with sites used by wild conspecifics. Seals tended to haul out during evening and night hours, with peak activity occurring after sunset. Both the probability and duration of haul-out events showed a tendency to increase with longer nights. No seasonal patterns or effects of sex or release year were identified. The association between haul-out behavior and nighttime suggests a possible link to prey activity patterns but does not preclude additional factors such as physiological rhythms, potential effects of rehabilitation, or inherited behavioral tendencies shaped by historical hunting pressure. Our findings indicate that such patterns may emerge early in life, independent of prior experience.
ABSTRACT Boat strikes injure and kill large numbers of Florida manatees ( Trichechus manatus latirostris ) each year. Management actions taken to mitigate this threat can be potentially enhanced by consideration of manatee cognitive processes. The role of cognition in marine mammal conservation has received little consideration in the literature. Here, we integrate discoveries from laboratory research on auditory sensory processes and perception with findings from field studies of how manatees respond to motorized vessels and the effect of the ambient soundscape on response times to boat noise. We incorporate that information into recommendations for manatee management to further reduce the risk of collisions. We also suggest lines of research to investigate other relevant cognitive issues—including attention, learning, memory, and decision‐making—to improve conservation efforts and understanding of manatee behavior.
