Acoustic signal alteration in pantropical spotted and spinner dolphins exposed to seismic airgun noise

Human activities are significantly altering the marine soundscape. Seismic exploration generates high-intensity sounds that propagate over large spatial and temporal scales and can adversely affect cetaceans, which rely on sound for communication, foraging, navigation, and survival. This study investigates the acoustic responses of pantropical spotted dolphins (Stenella attenuata) and spinner dolphins (Stenella longirostris) to seismic airgun noise in the Southwestern Atlantic Ocean between 2017 and 2026. The effects of seismic prospection activity noise on the acoustic parameters of both species were assessed using permutation tests, Principal Coordinates Analysis (PCoA), and a Random Forest Algorithm (RFA). Classification accuracy within the RFA increased during active seismic conditions due to a reduction in acoustic parameter variance, a trend mirrored by clearer PCoA cluster segregation for echolocation clicks. Whistles and click trains analysed under active and inactive airgun conditions revealed species-specific response patterns. S. attenuata reduced whistle frequency parameters and duration, narrowing its overall bandwidth, whereas S. longirostris increased whistle duration. Conversely, for click trains, both species exhibited a significant reduction in inter-click intervals alongside a narrower distributional range of click peak frequency values. These findings indicate that exposure to seismic noise elicits complex, multi-parametric adjustments that alter signal features. Beyond documenting these patterns, this study demonstrates that machine learning frameworks are effective tools for diagnosing shifts in cetacean acoustic profiles under anthropogenic disturbance. Recognising that distinct parameters within whistles and clicks respond differently under acoustic pressure is necessary for ensuring the accuracy of acoustic monitoring used during mitigation frameworks. Incorporating these variations into regulatory guidelines prevents automated detection errors, improving the management and protection of marine species within industrialised soundscapes.

Human activities are changing the acoustic marine environment. The noise produced by these activities has been considered one of the major threats to marine fauna (Payne and Webb, 1971; McCauley et al., 2003; Duarte et al., 2021). Anthropogenic noise has also gained recognition as an international pollutant and its importance has been recognized in the conservation of ecosystems (Popper and Hawkins, 2019; Ferrier-Pagès et al., 2021; Pichegru et al., 2022). The adverse effects of noise on wildlife range from behavioural changes, such as reduced foraging and resting and changes in group cohesion (Kight and Swaddle, 2011; Herbert-Read et al., 2017; Southall et al., 2024), to physiological responses, such as higher levels of stress hormones to changes in species distribution, community richness and composition (Duarte et al., 2015; Kunc and Schmidt, 2019). Additionally, high levels of both acute and chronic noise can mask acoustic signals, affecting the perception of intra- and/or interspecific signals of interest, jeopardizing the animals’ ability to detect signals from their prey/predator, hindering territorial defence, mate attraction, and other critical social interactions (Brumm et al., 2004; Sousa-Lima and Clark, 2008 and Sousa-Lima and Clark, 2009, Chan et al., 2010).

In the ocean, seismic exploration represents one of several activities that can potentially be harmful to the environment and marine life (Weilgart, 2007; Weilgart, 2013). Seismic airguns, associated with seismic exploration, generate repetitive high-intensity sounds that propagate over vast areas for long periods of time (Nieukirk et al., 2004; Schoeman et al., 2022). Several key studies highlight the effects of these sound sources on cetaceans, which are highly dependent on sound to perform various essential functions for their survival, such as feeding and navigation, and for their reproduction (Au, 2000; Au and Hastings, 2008; Southall et al., 2021). The documented effects include auditory disturbances, increase, reduction or cessation of vocal behaviours, abandonment of feeding areas, changes in diving patterns, decreased surface time, and interference with foraging behaviours (e.g., Finneran et al., 2002; 2015; Gordon et al., 2003; Stone and Tasker, 2006; Di Iorio and Clark, 2010; Castellote et al., 2012; Monaco et al., 2016; Kavanagh et al., 2019). Concerns about the potentially fatal impacts of marine seismic surveys on marine mammals have been raised, including by at-sea observations (e.g., ice entrapment, postural instability and akinesia) and records of stranding events (Gray and Van Waerebeek, 2011; Heide-Jørgensen et al., 2013; Castellote and Llorens, 2016; McGeady et al., 2016).

The genus Stenella Gray, 1866 comprises 5 species of dolphins distributed worldwide in tropical, subtropical and temperate waters (Fertl et al., 2003; Jefferson et al., 2008; Moreno et al., 2005; Perrin, 2009). From those, the pantropical spotted dolphin (Stenella attenuata) and the spinner dolphin (S. longirostris) are commonly recorded co-occurring during seismic surveys (Parente et al., 2007; Holst et al., 2017). Both species are known to produce whistles, click trains, and burst-pulsed sounds (Moron et al., 2015; Silva et al., 2016; Gong et al., 2019; Smith et al., 2019; MaChado et al., 2024).

To date, few studies have addressed how seismic noise affects small odontocetes (e.g. Gray and Van Waerebeek, 2011; Pirotta et al., 2014; Sarnocińska et al., 2020; Maciel et al., 2026) and little is known about whether these species’ acoustic behaviour is maintained or altered during exposure to seismic prospection sound sources.

Seismic airguns produce source levels that are typically 60–90 dB higher than those generated by motorboats, with commercial airgun arrays emitting peak source levels (0-peak) of approximately 220–230 dB re 1 µPa for individual guns, with array source levels approaching 260 dB re 1 µPa at 1 m (Caldwell and Dragoset, 2000; Hermannsen et al., 2015).Measurements from seismic operations in Brazilian waters reporting maximum near-field levels of approximately 244.5 dB re 1 µPa (Fontana and Boukhanfra, 2018) The sound useful for prospection propagates downwards, some acoustic energy is radiated laterally from the source, generating a complex radiation pattern (Harris et al., 2017). The marine mammal received signal’s complexity can be increased by multiple propagation paths, which can create long reverberant waveforms of several seconds at extended ranges (Gordon et al., 2003; Nieukirk et al., 2004).

Given the significant disparity in source levels, it is plausible to expect that S. attenuata and S. longirostris would not only respond similarly to seismic airgun noise but potentially exhibit more pronounced behavioural alterations compared to their reactions to the noise from vessels. Studies on fin whale (Balaenoptera physalus) have shown that these whales undergo more substantial behavioural and vocal modifications in response to seismic airgun noise than to shipping noise, suggesting a reduced tolerance to the impulsive and intense characteristics of airgun signals (Castellote et al., 2012). This heightened sensitivity to seismic noise could similarly affect small odontocetes, including Stenella species, in a comparable or more acute manner.

The Southwestern Atlantic Ocean hosts an expanding oil and gas industry alongside a diverse cetacean community, making it a critical area for assessing the impacts of high-intensity seismic surveys. Despite the widespread use of these industrial sound sources, knowledge gaps persist regarding the precise mechanisms underlying individual cetacean responses and their subsequent population-level consequences. In Brazil, regulatory measures are in place to mitigate the impacts of seismic prospecting. However, to establish acoustic exposure guidelines and enhance current mitigation measures, the Licensing Environmental Agency requires scientific support to understand the animals’ responses at the individual and species levels (Monaco et al., 2016; Vilardo and Barbosa, 2018). Field studies on free-ranging marine animals are essential for understanding the impact of seismic noise, offering ecological realism and generating results that are relevant for regulatory decisions and effective conservation policies (Gomez et al., 2016; Affatati and Camerlenghi, 2023; Wedekin et al., 2026). The present study assessed how different species of the genus Stenella respond to seismic activities by investigating their acoustic behaviour during exposure to seismic prospecting activities in the Southwestern Atlantic Ocean.

Data collection

Acoustic data were collected between 2017 and 2026 through distinct cruises in the Southwestern Atlantic Ocean (Table 1). The acoustic detections between 2017 and February 2022 occurred aboard seismic support vessels (i.e., separate from the source vessels towing the airgun arrays) during commercial seismic survey campaigns conducted in Brazilian oceanic waters. Crucially, under the regulations enforced by the Brazilian Institute of Environment and Renewable Natural Resources (IBAMA), concurrent seismic operations in Brazilian waters are strictly managed through a mandatory ‘Joint Operations Strategy’ (Estratégia de Operações Conjuntas). This regulatory framework mandates a minimum spatial separation distance of 60 km between active seismic vessels operating concurrently (IBAMA, 2022; Viana and de Mello, 2022). This strict buffer zone ensures that the acoustic data collected during each campaign aboard seismic support vessels were not confounded by overlapping high-intensity impulsive noise from other nearby industrial seismic surveys. Species identification was performed by experienced researchers who photographed animals during visual monitoring efforts for confirmation, identifying specific diagnostic characteristics of each species. Acoustic recordings were obtained using a 300 meter long towed array (AUSET ®), consisting of four hydrophones at its end, equipped with a high-pass filter (0.499 kHz), coupled to a system used to record acoustic signals composed of a SailDaq or Roland octa-capture digitizer (sampling frequency of 500kHz/24 or 16 bits and 196kHz/24 bits, respectively) connected to a computer on board. The recordings from August 2022 and March 2023 occurred aboard the tugboat Urano using two Benthowave Instrument hydrophones arranged in a vertical line with a separation distance of 1 meter and protected by a stainless-steel cage. A SoundDevices/MixPre-6-II recorder was employed for digitizing the recorded acoustic data, with a sampling frequency of 192 kHz and 16-bit resolution. Acoustic data collections were conducted concurrently with the recording of geographic coordinates, using the PAMGuard software (Gillespie et al., 2008). Acoustic data derived from eight distinct cruises, documenting a total of 49 encounters with the specified target species, were used for analysis (Figure 1, Table 1). The acoustic monitoring of the animals was conducted during periods with and without seismic airgun activity – airgun active (on) and inactive (off), respectively. The occurrence of airgun operations was verified with the seismic vessel crew and confirmed through spectrogram analysis. All recordings were conducted at distinct times, with concurrent visual sightings for species confirmation, capturing vocalising animals either with the airgun on or with the airgun off. The only exception is one encounter with S. longirostris, where the same group was observed both in the presence and absence of airgun activity (Table 1). To mitigate the potential for capturing signals that could represent a transitional response—shifting from on to off conditions—a five-minute buffer was implemented following the cessation of airgun activity. This precaution was designed to minimize any latency effects in the animals’ acoustic behaviour.

Upon visual detection of a cetacean group by onboard observers, pertinent information, such as geographical coordinates and group size was systematically documented during daylight hours. Marine mammal observers accomplished species confirmation by analysing photographs captured during sightings and verifying species-specific diagnostic characteristics.

Acoustic analysis

Biological acoustic activity periods were identified through both visual and aural inspection using Raven Pro 1.6 (Cornell Laboratory of Ornithology, NY). Only acoustic files accompanied by sighting data of the identified species were included in the analyses. Whistles and click trains were manually selected from spectrograms (using a Hann window of 1843 points, DFT of 2048 points, and 70% overlap). The following whistle parameters were extracted: low frequency, high frequency, delta frequency (i.e., the difference between high and low frequencies), centre frequency, peak frequency, beginning and ending frequencies, and whistle duration. Selection criteria were based on a signal-to-noise ratio (≥ 10 dB) for whistles exhibiting a clear contour shape. Heavily overlapped whistles, in which individual contours could not be reliably distinguished, were excluded because parameter extraction under these conditions is susceptible to masking effects and ambiguous measurements, compromising species-level comparisons (Oswald et al., 2021; Scaradozzi et al., 2024). For those partially overlapping whistles, a manual validation procedure was specifically applied to the energy-based parameters (e.g., peak time, peak frequency, and centre frequency): Only measurements whose coordinates coincided with the target contour were retained, ensuring that the extracted values originated from the selected whistle rather than from adjacent overlapping signals. Measurements were confined to the fundamental frequency of whistle contours.

Measurements were confined to the fundamental frequency of whistle contours.

For click trains, we utilized PAMGuard (Gillespie et al., 2008) to extract click sequences, and the PAMpal package v. 0.12.6 (Sakai, 2023) in R v.4.1.0 (CRAN Project) to compute the following parameters: peak frequency, 3 dB bandwidth, 10 dB bandwidth, and inter-click interval (ICI). PAMpal settings incorporated an 8 kHz Butterworth high-pass filter and a FFT window length of 2.5 ms for use in calculations. S. attenuata clicks from 2017 were excluded from analysis due to the lower sampling frequency (192 kHz) (Table 1).

Statistical analysis

To assess the acoustic behavioural response of animals exposed to the seismic sound source, the whistles and clicks from the two seismic airgun contexts – airgun on and off - were analysed. Initially, we evaluated the effects of different seismic activity contexts on the performance of Random Forest models. Random Forest was selected due to its robustness in handling complex and high-dimensional datasets, low susceptibility to overfitting, and ability to model non-linear relationships, making it particularly suitable for ecological and bioacoustic applications (Barkley et al., 2019; Amorim et al., 2022; MaChado et al., 2024). To visualise patterns of acoustic similarity and multivariate group distribution across seismic contexts, a Principal Coordinates Analysis (PCoA) was performed on the acoustic parameters extracted from whistles and clicks. Prior to analysis, all variables were standardised to eliminate the influence of differing measurement scales, and pairwise Euclidean distances were calculated among observations. To enhance cluster clarity and reduce data density, only the 50% of records exhibiting the shortest Euclidean distances to their respective group centroids were retained for the final ordination. All statistical analyses were conducted using R and Python software environments.

To further investigate potential behavioural responses, acoustic parameters were compared between airgun-on and airgun-off conditions using a permutation test implemented in the lmPerm package (Wheeler and Torchiano, 2016), based on 100,000 permutations. This non-parametric approach evaluates whether differences between groups exceed those expected by chance and does not require assumptions of normality, making it appropriate for the characteristics of our dataset. Statistical significance was assessed at α = 0.05. Additional details on permutation-based inference can be found in Efron and Tibshirani (1993).

To verify variation at the species level, rather than focusing on specific populations or individual differences, we utilised acoustic data collected from different locations, presumed to represent distinct populations, and across multiple years. By incorporating data from various basins, seasons, and years, we also minimised the influence of behavioural variability.

A total of 790 whistles (395 during airgun activity and 395 without airgun activity) and 1236 clicks (618 during airgun activity and 618 without airgun activity) were analysed for both S. attenuata and S. longirostris. For inter-click interval (ICI) analyses, 897 measurements were obtained for S. attenuata and 63 for S. longirostris. Descriptive statistics for all analysed acoustic parameters are provided in the Supplementary Material.

For clicks, the model achieved an overall accuracy of 64.73% (95% CI = 61.17–68.18). The highest classification accuracies were observed for SI_ON (72.4%), Sa_ON (71.4%), and SI_OFF (63.8%), whereas Sa_OFF exhibited the lowest classification success (51.4%). Misclassifications occurred predominantly between ON and OFF seismic contexts within the same species (Figure 2A). Consistent with these results, the PCoA revealed a clear separation among the four groups, with the first two axes explaining 94.4% of the total variation (PCoA1 = 76.9%; PCoA2 = 17.5%) (Figure 2B). For whistles, the model achieved an overall accuracy of 47.88% (95% CI = 43.29–52.50). The highest classification accuracies were recorded for SI_ON (68.6%) and SI_OFF (48.3%). For S. attenuata, the Sa_ON class exhibited a higher classification success rate (39.8%) than Sa_OFF (34.7%). Similar to the pattern observed for clicks, misclassifications occurred primarily between ON and OFF seismic contexts within the same species (Figure 2C). The PCoA based on whistle parameters showed greater overlap among groups, with the first two axes explaining 68.5% of the total variation (PCoA1 = 38.9%; PCoA2 = 29.6%) (Figure 2D).

Comparative analyses of echolocation click acoustic parameters in Stenella attenuata and Stenella longirostris revealed statistically significant differences between exposure contexts (p < 0.05) for the measured parameters (Figure 3; Table 2). For both species, spectral parameters related to peak frequency, −3 dB bandwidth, and −10 dB bandwidth showed higher values during periods of active airgun operation (airgun ON). ICI values reduced across the exposure contexts for both species. Additionally, both species exhibited broader frequency distributions for peak frequency parameters during periods without airgun activity (airgun OFF) (Figure 3). Comparative analyses of whistle acoustic parameters revealed statistically significant differences between exposure contexts (p < 0.05): for Stenella attenuata, spectral and temporal parameters showed a reduction in high frequency, delta frequency, and duration, alongside an increase in low frequency during periods of active airgun operation (Figure 4). Conversely, Stenella longirostris exhibited an increase in whistle duration and high frequency under the same active seismic conditions. Both species increased their low frequency during the airgun ON context (Table 2).

Permutation analyses, assessing significant differences in acoustic parameters under the different seismic airgun contexts (on vs. off), demonstrated that odontocetes of the species Stenella attenuata and S. longirostris adjust their acoustic parameters in response to the airgun status during seismic surveys. The acoustic adjustment as a compensatory response to noise is reported across different cetacean species and the most documented strategies are changes in the amplitude, alteration of vocalizations temporal patterns, and shifts in frequency values (e.g. Sousa-Lima and Clark, 2008; La Manna et al., 2013; Radtke et al., 2023).

In this study, we observed the reduction of a variety of frequency parameters of S. attenuata whistles during exposure of noise from airguns. When analysing Indo-Pacific bottlenose dolphin (Tursiops aduncus) whistles from three locations, Morisaka et al. (2005) discovered that the lowest frequency whistles were produced in the noisiest area, a finding consistent with other research showing that dolphin whistle frequencies become lower in the presence of boat noise (May-Collado and Quiñones-Lebrón, 2014). As acoustic signals travel through the environment, they are subjected to degradation, with signal propagation particularly constrained by frequency-dependent attenuation (Brumm and Slabbekoorn, 2005). Since higher frequencies are less efficient for long-distance communication (Wiley and Richards, 1978), animals may reduce the frequency of contact calls in the presence of noise to enhance their communication range. Given that whistles are known to function as contact calls (e.g.: Janik and Sayigh, 2013, Andriolo et al, 2015; Viana et al., 2022), a decrease in the frequency of the whistle, as observed in S. attenuata, becomes plausible to optimize communication among conspecifics in a noisy environment.

Both S. attenuata and S. longirostris exhibited a significant increase (p < 0.05) in the low frequency of their whistle. For S. attenuata shift, coupled with the reduction in both high frequency and delta frequency, demonstrates that the species constricts its acoustic bandwidth in the presence of seismic airgun noise, potentially as a response to the restricted acoustic space free from potential masking available in these contexts. This bandwidth constriction is further supported by the click peak frequency results during airgun activity for both species as they exhibited a much narrower distributional range. This suggests that the presence of airgun noise exerts a selective pressure on the animals’ acoustic behaviour, limiting their peak frequency usage. These results are an indication that the use of acoustic parameters is adjusted to encode information when exposed to anthropogenic noise. In its absence, the animals are able to communicate using a broader range of acoustic signals compared to when such noise is present. This restriction of acoustic space is further supported by the machine learning results. The RFA yielded lower misclassification rates for both clicks and whistles during active seismic conditions across both species. This pattern may reflect a reduction in behavioural and acoustic variability, as animals under baseline conditions (airgun off) likely exhibit a broader and less constrained range of signal frequencies and durations. In contrast, exposure to high-intensity anthropogenic noise may constrain signal production, resulting in more stereotyped acoustic patterns and reduced use of the available acoustic niche. Because acoustic parameters occupied a narrower and less variable distribution during airgun activity, overlap among classes decreased, improving the discriminatory performance of the RFA model. Within the Random Forest framework, this reduction in parameter variability increases class separability, explaining the improved classification performance under active seismic conditions. This adjustment demonstrates acoustic plasticity in these species in response to disturbance, while highlighting how anthropogenic noise constrains their acoustic occupancy within the soundscape.

This capacity for algorithmic models to successfully segregate disturbed versus undisturbed acoustic states in Stenella attenuata aligns with previous ethoacoustic research. For instance, Poupard et al. (2019) demonstrated that multivariate dimensionality reduction and clustering techniques could effectively isolate distinct acoustic clusters in this species when exposed to anthropogenic noise from whale-watching vessels. While highly overlapping signals are biologically relevant communication components, they could introduce parameter masking that hinder reliable feature extraction (Fearey et al., 2019; Oswald et al., 2021) and increase intra-class variability, reducing the classification accuracy (Ali Ahmed et al., 2026; Oswald et al., 2021). Consequently, these signals were not included in the analyses but retained in the broader acoustic database. Future studies should investigate these complex vocal interactions using analytical frameworks designed for group-level communication, providing perspectives on how anthropogenic noise influences coordination, information exchange, and social cohesion in delphinids.

The structural and physical properties of odontocete acoustic signals are constrained by craniofacial anatomy and the asymmetric arrangement of nasal soft tissues involved in sound production (Laeta et al., 2021). These constraints differ between echolocation clicks and communication whistles that have distinct biological functions. Echolocation clicks are short, directional signals optimized for navigation and prey detection, whereas whistles are primarily used for social communication and generally exhibit greater temporal and spectral flexibility. These functional differences were reflected in the multivariate analyses: under the airgun off context, click parameters from both species occupied a largely overlapping region of the PCoA space, consistent with the relatively low classification accuracy achieved by the RFA model (51.4% for Stenella attenuata and 63.8% for S. longirostris). During seismic prospection operations, click parameters became more clearly separated between species, resulting in improved classification accuracy (71.4% and 72.4%, respectively). This pattern suggests that exposure to seismic noise was associated with a reduction in within-species variability and the emergence of more distinct species-specific click profiles. Compared with clicks, whistle parameters occupied a broader and more overlapping acoustic space, consistent with their communicative function and greater behavioural flexibility. The cluster of S. attenuata under “airgun off” encompassed a wide range of frequency and duration combinations, indicating substantial variation in whistle structure under baseline conditions. Under “airgun on”, this species showed a moderate contraction of frequency-related parameters, reflected in a modest increase in classification accuracy (39.8%). In contrast, S. longirostris exhibited a larger shift in whistle structure during seismic activity, with classification accuracy increasing from 48.3% to 68.6%. As a result, whistles produced during airgun exposure occupied a more distinct region of the multivariate space than those recorded under baseline conditions. Taken together, these findings indicate that clicks and whistles changed differently to seismic noise exposure. While click responses were characterized by increased separation between species within a relatively constrained acoustic space, whistles reflected species-specific adjustments within a more flexible communication system. These results suggest that the functional role of a signal influences how odontocetes modify acoustic behaviour in response to anthropogenic noise.

S. attenuata reduced its whistle duration during airgun exposure, whereas S. longirostris exhibited the opposite response by increasing whistle duration. Such contrasting compensatory strategies are not uncommon across taxa; for instance, while some avian species increase song or call duration in response to anthropogenic noise, others exhibit a reduction in overall duration by compressing their inter-note intervals (León et al., 2014; Gomes et al., 2022). A similar pattern of interspecific variation is observed in anurans. In their review, Gomes et al. (2022) noted a general tendency among amphibians to decrease this temporal parameter under noisy conditions. However, the advertisement calls of Boana leptolineata did not vary linearly; rather, call length either increased or decreased depending strictly on the order of noise intensity to which the individuals were exposed (Caorsi et al., 2017). Reduction of whistle duration in the presence of noise was also described in the Guiana dolphin (Sotalia guianensis) (Bittencourt et al., 2017) and Indo-pacific bottlenose dolphin (Marley et al., 2017). Black-fronted Titi monkeys also presented reduced call duration when recorded in noisy areas (Duarte et al., 2018). In another primate, the black tufted-ear marmoset, longer calls were found in the noisiest areas, a similar result was described for killer whale calls and for bottlenose dolphin whistles in the presence of boat noise (Foote et al., 2004; May-Collado and Wartzok, 2008; May-Collado and Quiñones-Lebrón, 2014). In the seismic prospection activity context, by emitting briefer acoustic signals, individuals ensure that the entire communicative unit is deployed and completed strictly within the silent inter-pulse intervals, thereby avoiding partial masking by the subsequent airgun blast. Rather than evading the noise window, this species potentially maximizes signal energy and detection probability over time. In a disturbed soundscape, extending the duration of a vocalization increases the likelihood that at least a fragment of the acoustic message will span across a period of lower ambient interference and be perceived by conspecifics. This dichotomy underscores that even closely related delphinids exposed to similar pressures can display contrasting anti-masking mechanisms to maintain acoustic cohesion.

The observed reduction in inter-click interval for both species during active airgun seismic source conditions may be attributed to the animal’s attempt to reduce the likelihood of overlap between its clicks and the sound generated by the seismic airgun, therefore, mitigating acoustic masking. León et al. (2014) described the reduction of the intervals between canaries’ song notes under noisy conditions to maintain the entire information where the “gaps between elements” are minimized instead of reducing the actual number of notes. The reduction of the inter-click interval can be a strategy to maintain the number of clicks from a certain train without information loss. Although delphinid clicks are mostly attributed as foraging signals, recent work demonstrated these vocalizations may function as communicative social signals (Gallo et al., 2023). Therefore, changes in the number of clicks can be important from a communicative perspective, as each click (or its sequence) may carry different information. Given the highly predictable, intermittent nature of industrial airgun arrays—which discharge high-intensity impulsive noises at fixed intervals (typically every 10–15 seconds)—the dolphins are presented with regular, brief windows of relative quiet between blasts. To avoid the severe energetic masking of faint returning echoes by the primary seismic pulse, these odontocetes appear to adopt an anti-masking strategy akin to the ‘auditory glimpsing’ mechanism originally described by Cooke (2006) and documented in other taxa by Vélez and Bee (2011). By decreasing their ICI, the animals efficiently pack a denser sequence of echolocation queries into these silent inter-pulse gaps, maximizing sensory update rates before the onset of the next acoustic disturbance. Reducing the inter-click interval without reducing the number of clicks could be a strategy to maintain the entire message in a noisy environment, theoretically at a higher energy cost to maintain vocal performance.

With exception of whistle duration, both S. attenuata and S. longirostris exhibited similar changes in whistle and click parameters in response to airgun noise. Regarding their acoustic plasticity, Gruden et al. (2016), when comparing these two Stenella species oscillatory whistles reported that S. longirostris showed more oscillations than S. attenuata: whistles with four or more oscillations occurring only in S. longirostris. Furthermore, Pires et al. (2021) found the ascending-descending (one oscillation) as the most predominant S. attenuata whistle modulation category. Ecological divergences among species may subject them to distinct environmental stimuli and adaptive challenges, influencing the evolution of their vocal strategies (Rendell et al., 1999; Lima et al., 2016; Rege-Colt et al., 2023). The acoustic niche hypothesis suggests that a species can adjust the frequency and/or temporal aspects of its sound emissions to prevent acoustic masking caused by signal overlap (Krause, 1993; Mossbridge and Thomas, 1999; Duarte et al., 2019). Therefore, this contrasting behavioural strategy regarding whistle duration found may be a consequence of interspecific acoustic differences.

Investigations by Maciel et al. (2026) within the Campos and Santos basins documented elevated whistle frequencies and extended durations for Stenella attenuata during seismic operations, suggesting these modifications serve as adaptive mechanisms for communication in disturbed marine soundscapes. Conversely, the S. attenuata individuals monitored in this research demonstrated a reduction in these parameters when exposed to airgun activity. This observed divergence in acoustic results likely stems from specific exposure protocols, variations in local environmental conditions, or the distinct propagation characteristics of the acoustic field. Collectively, these results underscore the context-dependent nature of cetacean vocal responses to anthropogenic noise and emphasize the necessity for further investigation across diverse geographical regions and varied regimes of exposure.

Finally, the observed acoustic response to noise between S. longirostris and S. attenuata could be associated with evolutionary history. Behavioural plasticity results from a complex interaction between genetic basis, individual learning, and cultural transmission (Mery and Burns, 2010; Bendesky and Bargmann, 2011). Heritable genetic mechanisms influence morphological, physiological, and neurological processes, shaping how species respond to environmental challenges (Niepoth and Bendesky, 2020).

Attention should be given to smaller delphinid species like those from the genus Stenella regarding the impact of seismic activity. In this study, we found frequency and time shifts in whistles and clicks on both studied Stenella species when the airgun operations were on. Weir (2008), when studying the reactions of the humpback whales (Megaptera novaeangliae), sperm whales (Physeter macrocephalus) and Atlantic spotted dolphins (Stenella frontalis) to airgun noise, described S. frontalis as the species “with the most marked overt response” of the three cetacean species examined: the response occurred at a significantly greater distance from the airgun array during full-array operations compared to guns-off periods and positive-approach behaviour was observed exclusively during guns-off periods. Abnormal locomotion was observed in one individual of S. attenuata located 600 m ahead of an airgun array presumably experiencing acoustic distress. The authors reported a catatonic-like state of akinesia, the animal was lifting its head above the surface for 5 minutes and displayed incremental postural instability and apparent exhaustion ultimately sinking almost motionless near the airgun array (Gray and Van Waerebeek, 2011).

A key study by Bain and Williams (2006) highlighted that marine mammal species that share analogous hearing sensitivity exhibited diverse behavioural responses to airgun noise. Considering that a species’ auditory sensitivity is often tuned to the frequencies used in its communication, a trade-off between optimizing call transmission effectiveness and receiver perception efficiency in the presence of noise is expected (Patricelli and Blickley, 2006; Wood and Yezerinac, 2006). Evidence from other cetaceans and marine mammals indicates that while behavioural adjustments occur in response to anthropogenic noise, such shifts incur severe energetic costs that may fail to fully offset long-term impacts on individual fitness and population viability (Gallagher et al., 2021; Perez‐Marrufo et al., 2026). These behavioural responses may cost the dolphins energetically, such trade-offs imposed by noise can lead to decreased survival and reproductive success at individual and population levels (Nabi et al., 2018; Erbe et al., 2019; Noren et al., 2012; Tennessen et al., 2024.

In conclusion, this study demonstrates that Stenella attenuata and S. longirostris modify both whistle and click characteristics in the presence of seismic airgun noise. By combining passive acoustic monitoring, machine learning classification, multivariate analyses, and univariate permutation tests we show that anthropogenic noise can alter acoustic parameters commonly used for species identification and behavioural assessment. These findings have direct implications for conservation monitoring and environmental impact assessments as responses to anthropogenic noise may influence the performance of automated classification systems. Specifically, our results indicate that automated PAM systems should be trained using recordings obtained under both baseline and noise-exposed conditions, since anthropogenic noise modifies the acoustic parameters used for species classification. Incorporating this noise-related acoustic variability into passive acoustic monitoring frameworks is therefore critical to ensure that automated detection systems remain reliable under both baseline and anthropogenically disturbed conditions. Beyond documenting species-specific responses to seismic noise, this study provides a framework for evaluating anthropogenic impacts in situations where visual observations are limited or unavailable. Additionally, while contemporary investigations have confirmed the bioacoustic and visual occurrence of data-deficient megafauna like beaked whales throughout the Brazilian Equatorial Margin, considerable knowledge deficits persist, obstructing the advancement of efficacious conservation measures in this territory (MaChado et al., 2025; Rodrigues‐Soares et al., 2026). Given the accelerating industrial interest in seismic exploration and petroleum extraction across this sector, the empirical baseline provided by our study is necessary for establishing proactive mitigation guidelines in this rapidly developing marine frontier.

The most realistic conditions for sound exposure will be achieved through studies conducted during actual seismic operations, which incorporate real-time and authentic sound sources, while logistically and financially challenging, they can capture natural behaviours and ecological interactions in representative environments (Gomez et al., 2016; Affatati and Camerlenghi, 2023). A central consideration is whether vocalization adjustments ultimately translate into reductions in foraging efficiency or social cohesion, as these trade-offs reflect a rebalancing of the energy budget from a behavioural ecology perspective; if such adjustments incur cumulative energetic or social costs, they may compromise long-term fitness and population viability.

To develop effective noise mitigation strategies, future studies must consider the processes underlying behavioural responses, including the context of exposure, sound source characteristics, proximity to the receiver, bathymetry, and species-specific factors such as sex, age, reproductive state, prior experience, social structure and behavioural conditioning (Wartzok et al., 2003; Nowacek et al., 2007; Weilgart, 2007; Ellison et al., 2011; Jolliffe et al., 2025). Inaccurate predictions may result from only considering more generalizing behavioural responses across broad taxonomic groups and sound categories (Southall et al., 2021). As anthropogenic noise continues to intensify worldwide, its overlap with biologically important habitats is expected to grow with shifting species distributions driven by population recovery, climate change, and industrial expansion, underscoring the urgent need for robust and adaptive management measures (Erbe et al., 2018, Erbe et al., 2019b; Guazzo et al., 2020). Yet, there remains no universally accepted definition of best practice for underwater noise management, with approaches varying substantially across countries, regions, species, and contexts (Jolliffe et al., 2025). These findings reinforce the utility of machine learning frameworks as tools for diagnosing shifts in cetacean acoustic profiles under anthropogenic pressure. Although the present opportunistic dataset was not designed to quantify response thresholds as a function of exact distance, the observed acoustic adjustments demonstrate that behavioural responses can be detected using PAM, providing a basis for future studies aimed at defining exposure-response relationships and refining mitigation measures such as exclusion zones. documenting both natural repertoires and noise-induced alterations allows management programmes to track short- and long-term shifts in soundscape ecology. This comprehensive acoustic mapping is vital for assessing cumulative impacts and adjusting the intensity of mitigation measures based on the severity of the observed behavioural changes within increasingly industrialised marine environments.