Whistle characteristics and species classification algorithm for delphinids in the Southwestern Atlantic Ocean

ABSTRACT

This study investigates the whistle characteristics of six delphinid species in the Southwestern Atlantic Ocean and develops a classification algorithm to distinguish among them. Acoustic recordings were collected along the coast of São Paulo for common dolphins (Delphinus delphis), orcas (Orcinus orca), Atlantic spotted dolphins (Stenella frontalis), and common bottlenose dolphins (Tursiops truncatus). Recordings of spinner dolphins (Stenella longirostris) were obtained from the Fernando de Noronha Archipelago, while Guiana dolphins (Sotalia guianensis) were recorded in the Cananéia Estuary, in southern São Paulo. A total of 1,797 whistles were analyzed using seven acoustic parameters: duration (ms), and minimum, maximum, delta, center, initial, and final frequencies (kHz). These acoustic features were classified using a Random Forest algorithm, which achieved an overall classification accuracy of 55%. The highest accuracy was observed for D. delphis (73.6%), followed by S. guianensis (65%) and S. frontalis (55%). These findings represent a meaningful step toward the development of whistle classification tools and support the potential use of passive acoustic monitoring for long-term studies of delphinid populations in the Southwestern Atlantic Ocean.

Keywords:
Dolphins; Species identification; Whistle repertoire; Passive acoustic monitoring

INTRODUCTION

Delphinids produce tonal, frequency-modulated whistles that constitute the acoustic repertoire of each species or population, primarily serving social communication (Janik and Slater, 1998). Whistle repertoires vary geographically, with local characteristics reported for common dolphins (Delphinus delphis), orcas (Orcinus orca), Atlantic spotted dolphins (Stenella frontalis), Guiana dolphins (Sotalia guianensis), spinner dolphins (Stenella longirostris), and common bottlenose dolphins (Tursiops truncatus) (Ansmann et al., 2007; Azevedo et al., 2007; Baron et al., 2008; Bazúa-Durán and Au, 2002, 2004; Gannier et al., 2020; Lehnhoff et al., 2025; Papale et al., 2014; Petrella et al., 2011).

Whistle characteristics are influenced by behavioral context, social interactions, group size and composition, interspecific associations, and behavioral state (Ali Ahmed et al., 2025; Heiler et al., 2016; May-Collado, 2010; Quick and Janik, 2008; Rendell et al., 1999). Variation within and among populations can also result from genetic divergence, environmental factors, cultural transmission, ambient noise, and anthropogenic influences (Ansmann et al., 2007; Janik and Slater, 2000; Jensen et al., 2009; May-Collado, 2013; Papale et al., 2015; Samarra et al., 2015). Studying these features enhances understanding of population-specific repertoires and ecological patterns.

Delphinid whistles are highly variable and often overlap in frequency, which complicates species classification (Ali Ahmed et al., 2025; Amorim et al., 2019, 2022; Baron et al., 2008; Gannier et al., 2020; Lammers et al., 2003; Lima et al., 2016; Machado et al., 2024; Oswald et al., 2003, 2004, 2007). Nevertheless, discrimination based on whistle parameters has been achieved using discriminant function analysis (DFA), classification and regression trees (CART), or combined approaches (Amorim et al., 2019; Bonhoeffer et al., 2025; Gannier et al., 2010; Lima et al., 2016; Oswald et al., 2003, 2004, 2007; Rendell et al., 1999).

Understanding whistle parameters of delphinids in the Southwestern Atlantic Ocean (SAO) is critical for developing species-level classification algorithms and implementing passive acoustic monitoring programs (Erbs et al., 2017; Gillespie et al., 2013; Rankin et al., 2016). The São Paulo coast is subject to intense anthropogenic activity, including shipping, fishing, and recreational boating, with major ports at Santos and São Sebastião and the nearby “Terminal Petrolífero Almirante Barroso” (CODESP, 1992; Figueiredo et al., 2017; Santos et al., 2010; Zanardi-Lamardo et al., 2013). Effective classification methods are therefore essential to support management and conservation efforts. This study aims to describe whistle parameters and develop a classification algorithm for six delphinid species inhabiting shallow waters of the SAO: D. delphis, O. orca, S. frontalis, S. guianensis, S. longirostris, and T. truncatus.

METHODS

Acoustic recordings of D. delphis, O. orca, S. frontalis and T. truncatus were made during oceanographic cruises conducted between 2012 and 2015 along the coast of São Paulo state (between 23-25°S and 45-48°W, detailed information on the observation effort is described in Paschoalini and Santos, 2020; Table 1, Figure 1a). Recordings of S. longirostris were made in the Fernando de Noronha Archipelago (3°50’S, 32°26’W; Table 1, Figure 1b). Recordings of S. guianensis were made in the Cananéia Estuary (24°58’S, 47°50’W; Table 1, Figure 1c), São Paulo state.

Table 1
Acoustic recordings of delphinids from the Southwestern Atlantic Ocean. The date (month and year) is presented in relation to georeferenced position (South = S; West = W), duration of the recordings (minutes and seconds, M:S), the species recorded, and the visually estimated school size.

Date		Position		Recording	Species	School size
Month	Year	S	W	(M:S)
Jun	2015	23°28’	44°45’	12:28	Delphinus delphis	120
Dec	2012	23°58’	45°33’	21:54	Orcinus orca	18
Apr	2018	24°54’	47°48’	21:13		4
Apr	2018	24°54’	47°50’	09:17	Sotalia guianensis	6
Apr	2018	24°55’	47°51’	12:27		5
Dec	2014	24°8’	46°16’	11:00	Stenella frontalis	60
Dec	2014	24°4’	45°57’	09:01		30
Jan	2015	24°41’	47°6’	05:00		12
Jan	2015	24°30’	46°44’	05:03		50
Feb	2015	24°10’	46°15’	08:45		20
Apr	2019	23°59’	45°34’	12:29		9
Jun	2019	24°8’	45°44’	07:03		8
Oct	2019	23°56’	45°31’	42:17		70
May	2017	3°51’	32°27’	08:42		30
May	2017	3°50’	32°25’	07:01	Stenella longirostris	20
May	2017	3°49’	32°24’	34:41		30
Nov	2014	24°15’	46°13’	04:50	Tursiops truncatus	35
Dec	2014	23°49’	45°24’	05:22		80
Jun	2015	23°30’	44°59’	12:34		25
Jan	2019	23°53’	45°29’	25:27		8

Figure 1
Maps of the surveyed areas to record delphinid whistles along the coast of São Paulo state (a): common dolphins (Delphinus delphis), orcas (Orcinus orca), Atlantic spotted dolphins (Stenella frontalis) and common bottlenose dolphins (Tursiops truncatus); Fernando de Noronha Archipelago (b): spinner dolphins (Stenella longirostris); and the Cananéia Estuary (c): Guiana dolphins (Sotalia guianensis).

Recording sessions were performed opportunistically, using a research boat smaller than 15 m in length, with the boat engine off, and in the absence of other boats besides the research boat, under similar weather conditions (Beaufort Sea state ≤ 2), and using the same acoustic recording methodology. At each sighting, the research boat approached the sighted group for species identification, group size estimation, and acoustic recording. Groups were defined as two or more dolphins in which each individual was within 10 m of at least one other member of the group (Quick and Janik, 2008). Given the difficulty in evaluating the behavioral state of recorded individuals with visual observations limited by the water surface, the repertoire of whistles was recorded independently of behavioral activity (e.g., travelling, foraging and/or feeding, and socializing). Only single-species groups of delphinids were recorded within the range of the hydrophone, with no other species known to whistle detected. The recordings were conducted using a broadband system that included an omnidirectional hydrophone (HTI-96 MIN, sensitivity: −201 dBV µPa,±3 dB from 2 Hz to 30 kHz). According to Oswald et al. (2004), a minimum upper bandwidth of 24 kHz is necessary to accurately capture the fundamental contours of whistles. The hydrophone was connected to a Sony PCM-M10 acoustic recorder, which operated at a 96 kHz sampling rate and 16-bit depth.

The extraction of time-frequency contours from whistles was performed using ROCCA (Oswald et al., 2007) and spectrograms (Hann window of a 1024-point Fast Fourier Transform with 60% overlap). Only whistles that were clear in overall contour shape and did not overlap with other whistles were randomly chosen for analysis. Based on previous whistle description studies (e.g., Bazúa-Durán and Au, 2002; Oswald et al., 2007), the following parameters were extracted from the whistle contour: frequency (kHz) minimum, maximum, delta (maximum - minimum), center (minimum + (maximum - minimum) / 2), initial, and final frequency, and duration (ms). The average, standard deviation, minimum and maximum values of the acoustic parameters were also estimated. Acoustic parameters were used to generate box plots in the statistical software R (R Core Team, 2019).

The mean, standard deviation, minimum, and maximum values of the acoustic parameters reported in previous studies-including minimum, maximum, delta, center, initial, and final frequencies-were used to generate box plots, allowing for comparison of medians and quartiles across studies. For each species, the average whistle frequency parameters were obtained from the following locations: D. delphis from the Atlantic Ocean, Pacific Ocean, and Mediterranean Sea; O. orca from the Atlantic Ocean, Pacific Ocean, and Norwegian Sea; S. guianensis from the Cananéia and Guaraqueçaba estuaries, Sepetiba and Guanabara Bays, and the Atlantic coast of Brazil; S. frontalis from the Southwestern Brazilian coast, São Miguel Island, Canary Archipelago, and the Gulf of Mexico; S. longirostris from the Fernando de Noronha Archipelago, Southwestern Atlantic Ocean, and Hawaiian Archipelago; and T. truncatus from the Atlantic and Pacific Oceans. The studies consulted are presented in Table 2.

Table 2
Whistle frequency parameters gathered from previous studies of D. delphis (Dd), O. orca (Oo), S. guianensis (Sg), S. frontalis (Sf), S. longirostris (Sl) and T. truncatus (Tt) from different occurrence areas. For each species, a code (letters from B to K) abbreviates the citation of each investigation used to provide the box plots presented in Figure 5.

Species	Code	Reference	Location
Dd	B	Papale et al., 2014	Atlantic Ocean
	C	Papale et al., 2015
	D	Amorim et al., 2019
	E	Gannier et al., 2020
	F	Machado et al., 2024
	G	Oswald et al., 2003	Pacific Ocean
	H	Oswald et al., 2007
	I	Petrella et al., 2011
	J	Ansmann et al., 2007	Mediterranean Sea
	K	Azzolin et al., 2019
Oo	B	Andriolo et al., 2015	Atlantic Ocean
	C	Amorim et al., 2019
	D	Machado et al., 2024
	E	Thomsen et al., 2001	Pacific Ocean
	F	Samarra et al., 2010	Norwegian Sea
	G	Samarra et al., 2015
Sg	B	Azevedo Van Sluys, 2005	Cananéia Estuary
	C	Pivari & Rosso, 2005
	D	Deconto & Monteiro-Filho, 2016
	E	Azevedo & Van Sluys, 2005	Guaraqueçaba
	F	Erber & Simão, 2004	Sepetiba Bay
	G	Andrade et al., 2015
	H	Azevedo & Simão, 2002	Guanabara Bay
	I	Azevedo & Van Sluys, 2005
	J	Andrade et al., 2015
	K	Machado et al., 2024	Atlantic Ocean
Sf	B	Azevedo et al., 2010	Southwestern Brazilian coast
	C	Lima et al., 2016
	D	Amorim et al., 2019
	E	Machado et al., 2024
	F	Gannier et al., 2020	São Miguel Island
	G	Papale et al., 2015	Canary Archipelago
	H	Baron et al., 2008	Gulf of Mexico
Sl	B	Camargo et al., 2006	Fernando de Noronha Archipelago
	C	Moron et al., 2015
	D	Amorim et al., 2019	Southwestern Atlantic Ocean
	E	Machado et al., 2024
	F	Bazúa-Durán & Au, 2002	Hawaiian Archipelago
	G	Lammers et al., 2003
	H	Oswald et al., 2003
	I	Oswald et al., 2007
Tt	B	Lima et al., 2016	Atlantic Ocean
	C	Amorim et al., 2019
	D	Lima et al., 2020 (Tramandaí)
	E	Lima et al., 2020 (Rio de Janeiro)
	F	Lima et al., 2020 (Uruguay)
	G	Machado et al., 2024
	H	Oswald et al., 2003	Pacific Ocean
	I	Oswald et al., 2007

Random Forest (RF) was used to examine the capacity to classify whistles of delphinids from the SAO. RF is a machine learning algorithm-a supervised classification method that can model complex interactions among parameters, measure error rate, and provide stability and classification accuracy (Cutler et al., 2007; Siroky, 2009). The model creates a series of classification trees that are not influenced by one another when constructed and are robust to overfitting (Breiman, 2001). RF has demonstrated excellent performance in bioacoustic studies (e.g., Webster et al., 2016; Yang et al., 2020; Rubbens et al., 2023).

Two-thirds of the dataset, comprising data from various schools, were randomly allocated to the training sets of RF algorithms (e.g., Shamir et al., 2014; Webster et al., 2016). The remaining data, consisting of delphinid species whistles, were used as testing sets to evaluate whistle characteristics. Whistle parameters were used as predictor variables and the initials of the Latin names of delphinid species were the response parameters. During the training phase, out-of-bag (OOB) observations, data instances excluded from the bootstrap sample used to construct each individual decision tree, serve as internal validation data for that tree. The OOB error is calculated by aggregating the prediction discrepancies across all trees using their respective OOB samples (Breiman, 2001; Liaw and Wiener, 2002; Siroky, 2009). This metric offers a robust and unbiased estimate of the model’s generalization performance for each species, functionally analogous to k-fold cross-validation but computationally more efficient within the RF framework. We used the statistical software R (R Core Team, 2019) to conduct the RF analysis, and graphical representations.

The training data were incorporated into the model through bootstrap resampling, and at each node, a random subset of predictor variables was assessed to identify the most informative split (Siroky, 2009). For the best accuracy, the number of forecast variables to be searched for each node (mtry) and the number of replicates of trees (ntree) were defined by (tuneRf), also from the RF package (Liaw and Wiener, 2002). A sufficiently large ntree value was chosen that provided robust and stable results (Strobl et al., 2009). The parameters for this particular model were set at 3,000 ntree and 4 mtry.

The mean decrease Gini (MDG) was implemented to estimate the most important predictive parameters for whistle classification. This metric is based on a weighted mean of the improvement of individual trees based on the inclusion of each variable as a predictor (Breiman, 2001; Liaw and Wiener, 2002). For each tree in the forest, the prediction error was calculated on the OOB data. Then, for each parameter, the same calculation was performed using a random permutation of the value of that parameter. Finally, for each parameter, the differences in prediction errors were averaged over all trees (Liaw and Wiener, 2002).

To validate the predictive power of the RF algorithm, one third of the original dataset was randomly selected and set aside for use as the validation dataset. These data were not involved in the training of the model. Predictions were made using the predict function in the RF package (Liaw and Wiener, 2002). This function classified the data based on the complex relationships among the predictor variables as determined by the RF model. The predict function assigned an expected species to each set of whistle parameters, and these predictions were then compared to the whistle species recorded.

RESULTS

A total of 1,797 whistles were analyzed from 4 h 36 min of recordings, comprising 649 whistles from S. frontalis (1 h 40 min 38 s), 448 whistles from D. delphis (12 min 28 s), 267 from S. guianensis (42 min 57 s), 251 from S. longirostris (50 min 24 s), 92 from O. orca (21 min 54 s), and 90 from T. truncatus (35 min 39 s, Table 1). Data from D. delphis and O. orca were obtained from a single sighting (Table 1).

The overview of descriptive data for each species is presented in Table 3. The average center frequency of delphinid whistles analyzed was 4.9 kHz for O. orca, 11 kHz for S. frontalis, 11.7 kHz for D. delphis, 12.6 kHz for S. longirostris, 12.7 kHz for T. truncatus, and 14.8 kHz for S. guianensis (Table 3 and Figure 2). T. truncatus and S. longirostris emitted whistles with longer duration (982.5 ms and 817.1 ms respectively) than D. delphis and O. orca (736.4 ms and 744.5 ms respectively), followed by S. frontalis (528.1 ms) and S. guianensis (327.4 ms; Table 3 and Figure 3). Figure 2 depicts the frequency range for each parameter with its median and quartiles, and Figure 3 shows the duration range of delphinid whistles from the SAO.

Table 3
Whistle parameters of Delphinus delphis (Dd), Orcinus orca (Oo), Sotalia guianensis (Sg), Stenella frontalis (Sf), Stenella longirostris (Sl) and Tursiops truncatus (Tt) gathered in the present study. A sample of whistles (n) analyzed for each species is shown along with mean±standard deviation for the whistle parameters: frequency (kHz) minimum (FMin), maximum (Fmax), delta (FDelta), center (FCenter), initial (FInitial), final (FFinal), and duration (ms).

	n	FMin	FMax	FDelta	FCenter	FInitial	FFinal	Duration
Dd	448	9.22±2.50	14.29±3.14	5.07±2.56	11.76±2.53	11.24±3.49	12.30±3.65	744.46±423.11
Oo	92	4.50±2.02	5.42±2.36	0.92±0.97	4.96±2.14	5.01±2.31	4.91±2.18	736.36±360.00
Sg	267	9.83±3.35	19.79±5.75	9.96±6.62	14.81±3.34	10.22±3.85	19.72±5.76	327.37±150.00
Sf	649	8.15±3.02	13.95±4.73	5.79±3.62	11.03±3.55	9.18±3.77	12.47±4.86	528.07±320.00
Sl	251	9.57±2.27	15.61±2.88	6.04±2.89	12.59±2.15	11.06±3.47	13.73±3.73	817.15±300.00
Tt	90	9.22±2.30	16.13±4.74	6.91±4.18	12.67±3.08	12.77±5.34	11.56±3.90	982.52±210.00

Figure 2
Descriptive statistics (median and quartiles of frequency parameters) of delphinid whistles: Delphinus delphis, Stenella frontalis, Orcinus orca, Stenella longirostris, Sotalia guianensis, and Tursiops truncatus.

Figure 3
Descriptive statistics (median and quartiles) of the duration of delphinid whistles: Delphinus delphis (Dd), Orcinus orca (Oo), Stenella frontalis (Sf), Sotalia guianensis (Sg), Stenella longirostris (Sl), and Tursiops truncatus (Tt).

The Random Forest (RF) model achieved a predicted classification accuracy of 56.7% for delphinid whistles. Figure 4 presents the Random Forest learning curves with their corresponding out-of-bag (OOB) errors, as well as the ROC curves and AUC values for each species. The learning curves stabilized before reaching 2,000 trees (Figure 4a). The ROC analyses indicated good predictive performance, with AUC values exceeding 0.70 (Figure 4b), demonstrating that the models performed well in predicting whistles. Additionally, the model provided insights into the relative importance of each acoustic parameter for species classification based on the Mean Decrease in Gini (MDG) index. The parameters were ranked as follows: whistle duration (190.3 MDG), final frequency (140.3 MDG), initial frequency (125.2 MDG), delta frequency (122.1 MDG), maximum frequency (115.2 MDG), minimum frequency (100.9 MDG), and center frequency (97.6 MDG).

Figure 4
Learning curves and corresponding OOB errors calculated for the 10,000 trees of the Random Forest (a), and ROC curves and associated AUC for each species (b): Dd, (D. delphis), Oo (O. orca), Sf (S. frontalis), Sg (S. guianensis), Sl (S. longirostris), and Tt (T. truncatus).

Among the species analyzed, S. guianensis achieved the highest correct classification rate, followed by D. delphis, S. frontalis, S. longirostris, T. truncatus, and O. orca. Notably, 82.3% of O. orca whistles were misclassified as S. frontalis (Table 4). The prediction voting confusion matrix from the RF model is detailed in Table 4. In terms of accuracy relative to the validation dataset, the RF model correctly classified 55% of the corresponding whistles. Species-specific identification accuracy was as follows: 73.6% for D. delphis, 64.9% for S. guianensis, 54.7% for S. frontalis, 42.2% for S. longirostris, 23.3% for T. truncatus, and only 16.7% for O. orca. Furthermore, a substantial proportion (73.3%) of O. orca whistles were misidentified as S. frontalis (Table 5).

Table 4
Confusion matrix of prediction voting (absolute and %) of the random forest algorithm, showing Delphinus delphis (Dd), Orcinus orca (Oo), Sotalia guianensis (Sg), Stenella frontalis (Sf), Stenella longirostris (Sl), Tursiops truncatus (Tt), and OOB (out-of-bag error estimate in %).

	n	Dd	Oo	Sf	Sg	Sl	Tt
Dd	300	186 (62)	4 (1.3)	61 (20.3)	8 (2.7)	37 (12.4)	4 (1.3)
Oo	62	3 (4.8)	8 (12.9)	51 (82.3)	0	0	0
Sf	435	60 (13.8)	53 (12.2)	265 (60.9)	37 (8.5)	19 (4.4)	1 (0.2)
Sg	179	5 (2.8)	0	38 (21.1)	134 (74.9)	1 (0.6)	1 (0.6)
Sl	168	51 (30.3)	0	40 (23.8)	4 (2.4)	69 (41.1)	4 (2.4)
Tt	60	15 (25)	0	10 (16.7)	2 (3.3)	12 (20)	21 (35)

Table 5
Confusion matrix of validation voting (absolute value and %) of the random forest algorithm, highlighting the correct classification of Delphinus delphis (Dd), Orcinus orca (Oo), Stenella frontalis (Sf), Sotalia guianensis (Sg), Stenella longirostris (Sl), and Tursiops truncatus (Tt).

	n	Dd	Oo	Sf	Sg	Sl	Tt
Dd	148	109 (73.6)	0	22 (14.9)	0	14 (9.5)	3 (2.0)
Oo	30	3 (10)	5 (16.7)	22 (73.3)	0	0	0
Sf	214	43 (20.1)	23 (10.7)	117 (54.7)	16 (7.5)	15 (7)	0
Sg	88	3 (3.4)	1 (1.1)	25 (28.4)	57 (64.9)	1 (1.1)	1 (1.1)
Sl	83	28 (33.7)	2 (2.4)	18 (21.7)	0	35 (42.2)	0
Tt	30	12 (40)	1 (3.3)	5 (16.7)	1 (3.3)	4 (13.3)	7 (23.3)

DISCUSSION

The average center frequency of delphinid whistles recorded in the South Atlantic Ocean (SAO) ranged from 4.9 kHz in O. orca to 14.8 kHz in S. guianensis. Notably, the whistle frequency ranges exhibited substantial overlap among species, suggesting potential acoustic niche sharing. The average center frequency whistles of D. delphis, S. frontalis, S. longirostris, and T. truncatus overlap (Table 3, Figure 2). The parameter descriptions for D. delphis and O. orca may not fully represent the complete whistle repertoire of the observed populations, as they were derived from a single sighting and likely reflect a limited range of behavioral contexts. In this study, we present a preview of the whistle repertoire of these species. Considering that behavior can influence species whistle repertoire and consequently interspecific classification (May-Collado, 2010), new sightings, recording of different behavioral states and description efforts of the whistle repertoire of these species will refine the quantitative knowledge and possibly improve the classification accuracy of the RF algorithm presented here.

The overlapping ranges of whistle parameters pose a significant challenge for the development of accurate classification algorithms (Table 3, Figures 2 and 3). Previous studies had already reported the overlap of frequencies for delphinid whistles (e.g., Amorim et al., 2019; Baron et al., 2008; Gannier et al., 2020; Lammers et al., 2003; Lima et al., 2016; Machado et al., 2024; Oswald et al., 2003, 2004, 2007). The frequency whistle ranges from this study and previous studies are shown in Figure 5. The variation in whistle parameters among different species or populations is likely the result of physiological or environmental factors such as body size, ambient background noise (Rendell et al., 1999), or bandwidth limitation of the frequency range studied (e.g., Pivari and Rosso, 2005). Broadband recording systems provide the possibility to understand the complete whistle structure of dolphin species (May-Collado and Wartzok, 2009). The differences in frequency range observed in this study (Figure 5) compared to previous studies may be attributed to variations in sampling rates and the fact that most recordings were conducted in shallow waters, except S. longirostris, which was recorded in the vicinity of the Fernando de Noronha Archipelago.

Figure 5
Whistle frequency, with median and quartiles of frequency parameters of delphinids from different occurrence areas, showing box plots with data gathered by the present study (A) and from previous studies (see Table 2 for access codes) of D. delphis from the Atlantic Ocean (AO), the Pacific Ocean (PO), and the Mediterranean Sea (MS); of O. orca from the Atlantic Ocean (AO), the Pacific Ocean (PO) and the Norwegian Sea (NS); of S. guianensis from the Cananéia Estuary (CE), Guaraqueçaba estuary (GUA), Sepetiba Bay (SB), Guanabara Bay (GB) and from the Atlantic Ocean (AO); of S. frontalis from the Southwestern Brazilian coast (SBC), São Miguel Island (SMI), the Canary Archipelago (CA) and the Gulf of Mexico (GM); of S. longirostris from the Fernando de Noronha Archipelago (FNA), the Southwestern Atlantic Ocean (SAO) and the Hawaiian Archipelago (HA); and of T. truncatus from the Atlantic Ocean (AO) and the Pacific Ocean (PO).

The D. delphis individuals recorded in this study produced whistles with a lower frequency range compared to other groups previously described in the Atlantic Ocean (Amorim et al., 2019; Gannier et al., 2020; Machado et al., 2024; Papale et al., 2014, 2015). This difference may be attributed to the limited dataset, as recordings were obtained from a single observed school, potentially restricting the variation in center frequency. Whistle frequencies from the Pacific Ocean (Oswald et al., 2003, 2007; Petrella et al., 2011) and Mediterranean Sea (Ansmann et al., 2007; Azzolin et al., 2019) showed region-specific characteristics, but the variability in the frequency span was similar to whistles of D. delphis from the Atlantic Ocean (Figure 5). This species is considered to have high abundance with a worldwide distribution (Perrin, 2002) and predominantly inhabits deeper waters of the continental shelf, occasionally visiting highly productive waters such as the coast of São Paulo state (e.g., Figueiredo et al., 2020; Santos et al., 2017, 2019). Whistle frequency parameters of D. delphis vary in relation to the presence of anthropogenic and natural noise (Papale et al., 2015), and attention should be drawn to data gathered along the highly impacted area of the coast of São Paulo (e.g., CODESP, 1992; Figueiredo et al., 2017; Santos et al., 2010; Zanardi-Lamardo et al., 2013). Areas with high noise pollution have cumulative effects that reduce habitat quality and consequently their use and preference by cetacean populations (Jensen et al., 2009).

Whistles of O. orca reported here were similar to those obtained by previous studies from the Atlantic Ocean (e.g., Andriolo et al., 2015; Amorim et al., 2019; Machado et al., 2024) and from the Pacific Ocean (e.g., Thomsen et al., 2001; Figure 3). Whistles with frequency range above 20 kHz were reported for O. orca from oceanic waters of the Atlantic Ocean (Andriolo et al., 2015) and from the Norwegian Sea (Samarra et al., 2010, 2015; Figure 5). As our recorder equipment could not capture acoustic signals above 48 kHz, we did not report upper limit frequencies as reported by Andriolo et al. (2015), Machado et al. (2024), and Samarra et al. (2010, 2015), who used recorders with higher frequency sensitivity. According to Thomsen et al. (2001), O. orca whistles are generally longer in duration compared to those of other delphinids. In the present study, T. truncatus and S. longirostris produced whistles with comparatively longer duration ranges (Table 3, Figure 3) than other species described by Thomsen et al. (2001). However, as recordings were obtained from a single observed school, the range of center frequencies may be limited due to the restricted dataset.

O. orca is a cosmopolitan species found across all ocean basins, with a preference for temperate and coastal regions, particularly areas of high biological productivity (Ford, 2009; Heyning and Dahlheim, 1988). In the Southern Hemisphere, migratory behavior has been reported, likely associated with ecological, physiological, and biological drivers (Durban and Pitman, 2012; Pitman et al., 2019). Along the Brazilian coast, sightings are sporadic but documented (Amorim et al., 2019; Dalla Rosa and Secchi, 2007; Lodi and Hetzel, 1998; Machado et al., 2024; Pinedo et al., 2002; Santos and Netto, 2005), including off the coast of São Paulo State, where a seasonal pattern has been noted, with increased occurrence during the austral summer months (November-February) (Figueiredo et al., 2020; Santos and Silva, 2009; Santos et al., 2010, 2019; Siciliano et al., 1999).

The whistle frequency range of S. guianensis from Cananéia Estuary (CE), which houses a population of ~400 individuals (Mello et al., 2019), was diversified among groups in the same area (Azevedo and Van Sluys, 2005; Deconto and Monteiro-Filho, 2016; Pivari and Rosso, 2005; Figure 5). Geographic and intraspecific differences in whistle parameters (Figure 5) were observed among populations from the Guaraqueçaba Estuary (Azevedo and Van Sluys, 2005), Sepetiba Bay (Andrade et al., 2015; Erber and Simão, 2004), Guanabara Bay (Andrade et al., 2015; Azevedo and Simão, 2002; Azevedo and Van Sluys, 2005) and along the coast of the Atlantic Ocean (Machado et al., 2024). Whistle characteristics may vary based on behavioral activities, social interactions, ambient background noise and anthropogenic influence (Deconto and Monteiro-Filho, 2016; May-Collado, 2013). Due to intra- and inter-regional variability, improving the accuracy of RF models requires further efforts to record and characterize S. guianensis whistles across their distribution range. These new observations should be incorporated into the algorithm’s training and validation datasets to enhance model performance.

The whistle frequency range of S. frontalis reported here showed that the distributions were similar to recordings gathered from Ilha Grande Bay, Rio de Janeiro (Azevedo et al., 2010; Lima et al., 2016; Figure 5). This pattern of frequency may represent a connectivity between these populations. The highest whistle frequencies were recorded from populations on the southeastern Brazil coast (Amorim et al., 2019), São Miguel Island (Gannier et al., 2020), Canary Archipelago (Papale et al., 2015) and Gulf of Mexico (Baron et al., 2008; Figure 5). S. frontalis is an endemic species of tropical and warm-temperate waters of the Atlantic Ocean (Perrin, 2002) and is the only member of the genus Stenella that is frequently observed in shallow waters of Brazil (Moreno et al., 2005), including in coastal waters of São Paulo state (e.g., Santos et al., 2017), where it has been reported as one of the commonest resident cetacean species (Figueiredo et al., 2020). Consequently, the RF algorithm may be useful in long-term monitoring programs for this species in the SAO.

A broader frequency span of S. longirostris whistles was registered in this study compared to previous studies conducted in the Fernando de Noronha Archipelago (Camargo et al., 2006; Moron et al., 2015). Geographic and intraspecific differences based on whistle parameters (Figure 5) were observed between studies from the Fernando de Noronha Archipelago compared to the SAO (Amorim et al., 2019; Machado et al., 2024) and the Hawaiian Archipelago (Bazúa-Durán and Au, 2002; Lammers et al., 2003; Oswald et al., 2003, 2007). In the SAO, sightings of S. longirostris have been made in oceanic waters near islands or banks, such as the Archipelago of Fernando de Noronha where there is a resident population, and on the outer continental shelf and beyond the continental slope (Amorim et al., 2019; Machado et al., 2024; Secchi and Siciliano, 1995). They are rarely observed in shallow waters (Amaral et al., 2015; Moreno et al., 2005), where at least one incidental capture was reported (Santos and Ditt, 2009). Thus, the RF algorithm may be a useful tool to identify occasional whistles in shallow waters of the SAO where the species is not commonly reported (e.g., Santos et al., 2010).

The whistle frequency range of T. truncatus presented here was similar to recordings from the SAO (Amorim et al., 2019), coast of Rio de Janeiro (Lima et al., 2020), and Guanabara Bay (Lima et al., 2016; Figure 5). Populations from the Hawaiian Archipelago, Pacific Ocean (Oswald et al., 2003, 2007) showed similar frequency variability (Figure 5). Discrimination between the data from this study and from elsewhere may be due to the whistle characteristics described by this study being obtained in shallow waters. T. truncatus is a cosmopolitan species distributed in tropical and temperate waters (Wells and Scott, 2017), including the coastal waters of São Paulo state (Figueiredo et al., 2020; Paschoalini and Santos, 2020; Santos et al., 2017, 2019). According to Connor et al. (2000), T. truncatus lives in complex, fission-fusion societies that encourage individual-based communication. Beyond this, features such as group size and composition are also known to influence the acoustic properties of whistles (Heiler et al., 2016; Quick and Janik, 2008) and contribute to region-specific characteristics and variation between groups and populations.

Although neural network-based approaches, including convolutional and recurrent architectures, have shown high performance in a range of acoustic classification tasks, their successful application typically requires large, well-annotated datasets, as well as increased computational complexity and extensive hyperparameter tuning. In the present study, a Random Forest classifier was selected due to its robustness when applied to moderate-sized datasets, its strong generalization ability with relatively limited tuning, and its computational efficiency. Moreover, this approach provides explicit measures of feature importance, facilitating interpretation of which whistle parameters contribute most strongly to class discrimination. In marine acoustic studies, where the availability of large labeled datasets is often constrained, these characteristics are particularly advantageous. Consistent with the framework outlined by Rubbens et al. (2023), the methodological choice adopted here reflects a balance between classification performance, methodological transparency, and the interpretability of acoustic and ecological inferences.

The RF model is a standardized method able to classify whistles from delphinids of the SAO and to list the most important parameters for classification, helping to understand the relative importance of covariates, the effect of adding more parameters on out-of-bag test set error, as well as proximities between observations and marginal effects. The application of RF analysis has been employed for species classification based on acoustic parameters (Ali Ahmed et al., 2025; Amorim et al., 2022; Gillespie et al., 2013; Machado et al., 2024), for classifying whistles emitted by populations of Pseudorca crassidens (Barkley et al., 2019), and for the classification of whistles of D. delphis and D. bairdii to evaluate interspecific differences in whistle contours (Oswald et al., 2021).

In this study, the RF model correctly classified 55% of delphinid whistles (Table 5). However, classification performance varied considerably among species. The lowest accuracy rates were associated with T. truncatus and O. orca, which were represented by the smallest sample sizes. In both cases, only a single group was recorded, likely limiting the range of whistle characteristics (particularly center frequency) and, as a result, reducing the model’s classification performance for these species (Tables 4 and 5). This highlights the importance of expanding the dataset through additional recordings, especially in underrepresented regions of the SAO, to improve model accuracy.

Interestingly, despite O. orca presenting a substantially lower center frequency compared to the other species analyzed, its whistles were often misclassified as those of S. frontalis. The MDG analysis indicated that center frequency was the least influential variable in the classification process, suggesting that this parameter alone may not be sufficient to distinguish certain species pairs. Additionally, O. orca and S. frontalis shared overlapping distributions for several whistle parameters (Table 3; Figures 2 and 3), which may explain the confusion observed in the RF model’s predictions (Tables 4 and 5). This misclassification pattern suggests that center frequency alone provides limited discriminatory information, consistent with its low MDG ranking, whereas temporal parameters such as whistle duration may capture additional species-specific structure that improves discrimination when spectral overlap is pronounced.

The accuracy of the RF model in distinguishing whistles emitted by different species appears to be influenced by multiple interacting factors. These include: (i) intrinsic acoustic similarities among species; (ii) variability in whistle structure due to individual differences, geographic regions, and behavioral states; (iii) limitations resulting from small or imbalanced sample sizes; (iv) inherent constraints of the RF algorithm; and (v) the choice of frequency-related parameters, which may not fully capture species-specific acoustic characteristics. These findings highlight the need for more comprehensive and geographically diverse datasets, as well as the potential benefits of incorporating additional acoustic features (e.g., modulation patterns, harmonics) to improve species classification.

Most whistle-based classification studies of delphinids have achieved correct classification rates above chance, although values vary depending on species composition, dataset size, and analytical approach. Early efforts reported moderate performance in the Mediterranean (e.g., ~62.9%; Gannier et al., 2010) and Pacific Ocean (e.g., 41-52%; Oswald et al., 2003, 2004, 2007). More recent machine-learning applications have improved these benchmarks, such as Machado et al. (2024), who obtained robust classification performance for multiple delphinids in the Western South Atlantic using Random Forest models, and Ali Ahmed et al. (2025), which demonstrated effective discrimination of whistles among three oceanic dolphin species in the Comoros. Additionally, implementations of advanced deep-learning frameworks have shown high classification accuracy and scalability (e.g., Bonhoeffer et al., 2025 with orcAI for killer whales). In the Southwestern Atlantic, Lima et al. (2016) and Amorim et al. (2019, 2022) reported correct classification rates of 72.5%, 59.3%, and 87%, respectively, using whistle parameters with RF classifiers.

In the present study, D. delphis and S. guianensis exhibited the highest correct classification scores under the RF model, followed by S. frontalis and S. longirostris. Notably, species-specific classification performance for S. frontalis and S. guianensis exceeded those previously reported for the SAO, where S. frontalis achieved as low as 4.6% and S. guianensis ~70.7% correct classification in earlier analyses (Amorim et al., 2019; Lima et al., 2016). Given that S. guianensis is one of the most frequently encountered cetacean species in Brazilian coastal and estuarine waters (Silva et al., 2010), these results suggest that Random Forest classifiers may offer a robust and scalable approach for long-term passive acoustic monitoring programs across the Southwestern Atlantic Ocean.

CONCLUSION

Reliable and replicable measurements of odontocete whistle characteristics are essential to expand the application of passive acoustic monitoring to broader research and conservation efforts. Consequently, there is a global demand for algorithms that deliver low classification error rates. The acoustic parameters analyzed and the RF algorithm presented in this study represent a significant step toward the development of whistle classification tools and resources that support the implementation of long-term monitoring programs for delphinids in the Southwestern Atlantic Ocean.

DATA AVAILABILITY STATEMENT

All data are available from the corresponding author upon reasonable request.

ACKNOWLEDGMENTS

We thank the crew of the R/V Alucia/Ocean X for all support at the Fernando de Noronha Archipelago. Instituto Chico Mendes/ICMBio São Sebastião helped during data collection at the northern coast of São Paulo state. We also acknowledge Julie N. Oswald for her assistance and suggestions with data analysis. Victor Paschoalini provided the map. Thanks to Bill Rossiter and the Cetacean Society International for supporting DDB attendance at the 22^nd Biennial Conference on the Biology of Marine Mammals.

REFERENCES

• Ali Ahmed, R., Zanoli, A., Scebba, F., Abdou Rabi, F., Ben Ali Abdallah, Y., Ferri, S., Caruso, F., Giacoma, C. & Favaro, L. 2025. Comparative analysis of the whistles of three oceanic dolphins in the Comoros. Marine Mammal Science, 42(1), e70062. DOI: https://doi.org/10.1111/mms.70062
  » https://doi.org/10.1111/mms.70062
• Amaral, K. B., Alves, D. J., Heinzelmann, L., Borges-Martins, M., Siciliano, S. & Moreno, I. B. 2015. Ecological niche modeling of Stenella dolphins (Cetartiodactyla: Delphinidae) in the southwestern Atlantic Ocean. Journal of Experimental Marine Biology and Ecology, 472, 166-179. DOI: https://doi.org/10.1016/j.jembe.2015.07.013
  » https://doi.org/10.1016/j.jembe.2015.07.013
• Amorim, T. O. S., Castro, F. R., Moron, J. R., Duque, B. R., Di Tullio, J. C., Secchi, E. R. & Andriolo, A. 2019. Integrative bioacoustics discrimination of eight delphinid species in the western South Atlantic Ocean. PLoS ONE, 14(6), e0217977. DOI: https://doi.org/10.1371/journal.pone.0217977
  » https://doi.org/10.1371/journal.pone.0217977
• Amorim, T. O. S., Castro, F. R., Ferreira, G. A., Neri, F. M., Duque, B. R., Mura, J. P. & Andriolo, A. 2022. Acoustic identification and classification of cetaceans in the Brazilian marine area affected by an environmental disaster. The Journal of the Acoustical Society of America, 152(6), 3204-3215. DOI: https://doi.org/10.1121/10.0016358
  » https://doi.org/10.1121/10.0016358
• Andrade, L. G., Lima, I. M. S, Macedo, H. da S., Carvalho, R. R., Lailson-Brito Jr., J., Flach, L. & Azevedo, A. de F. 2015. Variation in Guiana dolphin (Sotalia guianensis) whistles: using a broadband recording system to analyze acoustic parameters in three areas of southeastern Brazil. Acta Ethologica, 18, 47-57. DOI: https://doi.org/10.1007/s10211-014-0183-7
  » https://doi.org/10.1007/s10211-014-0183-7
• Andriolo, A., Reis, S. S., Amorim, T. O. S., Sucunza, F., Castro, F. R., Maia, Y. G., Zerbini, N. A., Bortolotto, G. A. & Rosa, L. D. 2015. Killer whale (Orcinus orca) whistles from the western South Atlantic Ocean include high frequency signals. The Journal of the Acoustical Society of America, 138(3), 1696-1701. DOI: https://doi.org/10.1121/1.4928308
  » https://doi.org/10.1121/1.4928308
• Ansmann, I. C., Goold, J. C., Evans, P. G. H., Simmonds, M. & Keith, S. G. 2007. Variation in the whistle characteristics of short-beaked common dolphins, Delphinus delphis, at two locations around the British Isles. Journal of the Marine Biological Association of the United Kingdom, 87(1), 19-26. DOI: https://doi.org/10.1017/S0025315407054963
  » https://doi.org/10.1017/S0025315407054963
• Azevedo, A. F., Flach, L., Bisi, T. L., Andrade, L. G., Dorneles, P. R. & Lailson-Brito, J. 2010. Whistles emitted by Atlantic spotted dolphins (Stenella frontalis) in southeastern Brazil. The Journal of the Acoustical Society of America, 127(4), 2646-2651. DOI: https://doi.org/10.1121/1.3308469
  » https://doi.org/10.1121/1.3308469
• Azevedo, A. F., Oliveira, A. M., Rosa, L. D. & Lailson-Brito, J. 2007. Characteristics of whistles from resident bottlenose dolphins (Tursiops truncatus) in southern Brazil. The Journal of the Acoustical Society of America, 121(5), 2978-2983. DOI: https://doi.org/10.1121/1.2713726
  » https://doi.org/10.1121/1.2713726
• Azevedo, A. F. & Simão, S. M. 2002. Whistles produced by marine tucuxi dolphins (Sotalia fluviatilis) in Guanabara Bay, southeastern Brazil. Aquatic Mammals, 28(3), 261-266.
• Azevedo, A. F. & Van Sluys, M. 2005. Whistles of tucuxi dolphins (Sotalia fluviatilis) in Brazil: Comparisons among populations. The Journal of the Acoustical Society of America, 117(3), 1456-1464. DOI: https://doi.org/10.1121/1.1859232
  » https://doi.org/10.1121/1.1859232
• Azzolin, M., Gannier, A., Papale, E., Buscaino, G., Mussi, B., Ardizzone, G., Giacoma, C. & Pace, D. S. 2019. Whistle variability of the Mediterranean short beak common dolphin. Aquatic Conservation: Marine and Freshwater Ecosystems, 31(S1), 36-50. DOI: https://doi.org/10.1002/aqc.3168
  » https://doi.org/10.1002/aqc.3168
• Barkley, Y., Oleson, E. M., Oswald, J. N. & Franklin, E. C. 2019. Whistle classification of sympatric false killer whale populations in Hawaiian waters yields low accuracy rates. Frontiers in Marine Science, 6, 645. DOI: https://doi.org/10.3389/fmars.2019.00645
  » https://doi.org/10.3389/fmars.2019.00645
• Baron, S. C., Martinez, A., Garrison, L. P. & Keith, E. O. 2008. Differences in acoustic signals from delphinids in the western North Atlantic and northern Gulf of Mexico. Marine Mammal Science, 24(1), 42-56. DOI: https://doi.org/10.1111/j.1748-7692.2007.00168.x
  » https://doi.org/10.1111/j.1748-7692.2007.00168.x
• Bazúa-Durán, C. & Au, W. W. L. 2002. The whistles of Hawaiian spinner dolphins. The Journal of the Acoustical Society of America, 112(6), 3064-3072. DOI: https://doi.org/10.1121/1.1508785
  » https://doi.org/10.1121/1.1508785
• Bazúa-Durán, C. & Au, W. W. L. 2004. Geographic variations in the whistles of spinner dolphins (Stenella longirostris) of the Main Hawaiian Islands. The Journal of the Acoustical Society of America, 116(6), 3757-3769. DOI: https://doi.org/10.1121/1.1785672
  » https://doi.org/10.1121/1.1785672
• Bonhoeffer, S., Selbmann, A., Angst, D. C., Ochsner, N., Miller, P. J. O., Samarra, F. I. P. & Baumgartner, C. D. 2025. orcAI: a machine learning tool to detect and classify acoustic signals of killer whales in audio recordings. Marine Mammal Science, 42(1), 1-14. DOI: https://doi.org/10.1111/mms.70083
  » https://doi.org/10.1111/mms.70083
• Breiman, L. 2001. Random forest. Machine Learning, 45, 5-32. DOI: https://doi.org/10.1023/A:1010933404324
  » https://doi.org/10.1023/A:1010933404324
• Camargo, F. S., Rollo Jr., M. M., Giampaoli, V. & Bellini, C. 2006. Whistle variability in South Atlantic spinner dolphins from the Fernando de Noronha Archipelago off Brazil. The Journal of the Acoustical Society of America, 120(6), 4071-4079. DOI: https://doi.org/10.1121/1.2359704
  » https://doi.org/10.1121/1.2359704
• CODESP (Companhia Docas do Estado de São Paulo). 1992. Port of Santos, Brazil - 100 years (1892-1992), Companhia Docas do Estado de São Paulo, (Special Issue), 1-12.
• Connor, R. C., Wells, R. R., Mann, J. & Read, A. J. 2000. The bottlenose dolphin: social relationships in a fission-fusion society, In: Mann, J., Connor, R. C., Tyack, P. L. & Whitehead, H. (pp. 91-126). Cetacean societies: field studies of dolphins and whales. Chicago: University of Chicago Press, Chicago.
• Cutler, D. R., Edwards Jr, T. C., Beard, K. H., Cutler, A., Hess, K. T., Gibson, J. & Lawler, J. J. 2007. Random forests for classification in ecology. Ecology, 88(11), 2783-2792. DOI: https://doi.org/10.1890/07-0539.1
  » https://doi.org/10.1890/07-0539.1
• Dalla Rosa, L. & Secchi, E. R. 2007. Killer whale (Orcinus orca) interactions with the tuna and swordfish longline fishery off southern and south-eastern [Brazil: A comparison with shark interactions. Journal of the Marine Biological Association of the United Kingdom, 87(1), 135-140. DOI: https://doi.org/10.1017/S0025315407054306
  » https://doi.org/10.1017/S0025315407054306
• Deconto, L. S. & Monteiro-Filho, E. L. A. 2016. Day and night sounds of the Guiana dolphin, Sotalia guianensis (Cetacea: Delphinidae) in southeastern Brazil. Acta Ethologica, 19, 61-68. DOI: https://doi.org/10.1007/s10211-015-0223-y
  » https://doi.org/10.1007/s10211-015-0223-y
• Durban, J. W. & Pitman, R. L. 2012. Antarctic killer whales make rapid round-trip movements to subtropical waters: evidence for physiological maintenance migrations? Biology Letters, 8, 274-277. DOI: http://doi.org/10.1098/rsbl.2011.0875
  » http://doi.org/10.1098/rsbl.2011.0875
• Erber, C., Simão, S. M. 2004. Analysis of whistles produced by the Tucuxi dolphin Sotalia fluviatilis from Sepetiba Bay, Brazil. Anais da Academia Brasileira de Ciências, 76(2), 381-385. DOI: https://doi.org/10.1590/S0001-37652004000200029
  » https://doi.org/10.1590/S0001-37652004000200029
• Erbs, F., Elwen, S. H. & Gridley, T. 2017. Automatic classification of whistles from coastal dolphins of the southern African subregion. The Journal of the Acoustical Society of America, 141(4), 2489-2500. DOI: https://doi.org/10.1121/1.4978000
  » https://doi.org/10.1121/1.4978000
• Figueiredo, G. C., Amaral, K. B. & Santos, M. C. de O. S. 2020. Cetaceans along the southeastern Brazilian coast: Occurrence, distribution and niche inference at local scale. PeerJ, 8, e10000, DOI: https://doi.org/10.7717/peerj.10000
  » https://doi.org/10.7717/peerj.10000
• Figueiredo, G. C., Santos, M. C. de O., Siciliano, S. & Moura, J. F. 2017. Southern right whales (Eubalaena australis) in an urbanized area off the southwestern Atlantic Ocean: update records and conservation issues. Aquatic Mammals, 43(1), 52-62. DOI: https://doi.org/10.1578/AM.43.1.2017.52
  » https://doi.org/10.1578/AM.43.1.2017.52
• Ford, J. K. B. 2009. Killer whale: Orcinus orca. In: Perrin, W. F., Würsig, B. & Thewissen, J. G. M. Encyclopedia of marine mammals (pp. 60-657). California: Academic Press.
• Gannier, A., Fuchs, S., Gannier, A., Fernandez, M. & Azevedo, J. M. N. 2020. Dolphin whistle repertoire around São Miguel (Azores): Are you common or spotted? Applied Acoustics, 161, 107169. DOI: https://doi.org/10.1016/j.apacoust.2019.107169
  » https://doi.org/10.1016/j.apacoust.2019.107169
• Gannier, A., Fuchs, S., Quèbre, P. & Oswald, J. N. 2010. Performance of a contour-based classification method for whistles of Mediterranean delphinids. Applied Acoustics, 71(11), 1063-1069. DOI: https://doi.org/10.1016/j.apacoust.2010.05.019
  » https://doi.org/10.1016/j.apacoust.2010.05.019
• Gillespie, D., Caillat, M., Gordon, J. & White, P. 2013. Automatic detection and classification of odontocete whistles. The Journal of the Acoustical Society of America, 134(3), 2427-2437. DOI: https://doi.org/10.1121/1.4816555
  » https://doi.org/10.1121/1.4816555
• Heiler, J., Elwen, S. Hl, Kriesell, H. J. & Gridley, T. 2016. Changes in bottlenose dolphin whistle parameters related to vessel presence, surface behaviour and group composition. Animal Behaviour, 117, 167-177. DOI: https://doi.org/10.1016/j.anbehav.2016.04.014
  » https://doi.org/10.1016/j.anbehav.2016.04.014
• Heyning, J. E. & Dahlheim, M. E. 1988. Orcinus orca. Mammalian Species, 304, 1-9. DOI: https://doi.org/10.2307/3504225
  » https://doi.org/10.2307/3504225
• Janik, V. M. & Slater, P. J. B. 1998. Context-specific use suggests that bottlenose dolphin signature whistles are cohesion calls. Animal Behaviour, 56(4), 829-838. DOI: https://doi.org/10.1006/anbe.1998.0881
  » https://doi.org/10.1006/anbe.1998.0881
• Janik, V. M. & Slater, P. J. B. 2000. The different roles of social learning in vocal communication. Animal Behaviour, 60(1), 1-11. DOI: https://doi.org/10.1006/anbe.2000.1410
  » https://doi.org/10.1006/anbe.2000.1410
• Jensen, F. H., Bejder, L., Wahlberg, M., Soto, N. A., Johnson, M. & Madsen, P. T. 2009. Vessel noise effects on delphinid communication. Marine Ecology Progress Series, 395, 161-175.
• Lammers, M. O., Au, W. W. L. & Herzing, D. L. 2003. The broadband social acoustic signaling behavior of spinner and spotted dolphins. The Journal of the Acoustical Society of America, 114(3), 1629-1639. DOI: https://doi.org/10.1121/1.1596173
  » https://doi.org/10.1121/1.1596173
• Lehnhoff, L., Glotin, H., Gall, Y. L., Menut, E., Peltier, H., Pochat, A., Pochat, K., Van Canneyt, O. & Mérigot, B. 2025. Whistles characterisation using artificial intelligence reveals responses of short-beaked common dolphins to a bio-inspired acoustic mitigation device for fishing nets. Scientific Reports, 15, 40618. DOI: https://doi.org/10.1038/s41598-025-24256-5
  » https://doi.org/10.1038/s41598-025-24256-5
• Lima, I. M. S., Andrade, L. G., Bittencourt, L., Bisi, T. L., Flach, L., Lailson-Brito, J. & Azevedo, A. F. 2016. Whistle Comparison of four delphinid species in Southeastern Brazil. The Journal of the Acoustical Society of America, 139(5), 124-127. DOI: https://doi.org/10.1121/1.4947310
  » https://doi.org/10.1121/1.4947310
• Lima, I. M. S, Venuto, R., Menchaca, C., Hoffmann, L. S., Rosa, L. D., Genoves, R., Fruet, P. F., Milanelli, A., Laporta, P., Tassino, B., Bueno, S. M., Freitas, T. R. O., Bittencourt, L., Lailson-Brito Jr, J. & Azevedo, A. F. 2020. Geographic variation in the whistles of bottlenose dolphins (Tursiops spp.) in the southwestern Atlantic Ocean. Marine Mammal Science, 36(3), 1058-1068. DOI: https://doi.org/10.1111/mms.12690
  » https://doi.org/10.1111/mms.12690
• Liaw, A. & Wiener, M. 2002. Classification and regression by random forest. R News, 2, 18-22.
• Lodi, L. & Hetzel, B. 1998. Orcinus orca (Cetacea; Delphinidae) in coastal waters of Rio de Janeiro State. Bioikos, 12(1), 46-54.
• Machado, R. B., Amorim, T. O. S., Castro, F. R. & Andriolo, A. 2024. Use of whistles for acoustic classification of delphinids (Odontoceti: Delphinidae) in the Western South Atlantic Ocean. The Journal of the Acoustical Society of America, 156(2), 1070-1080. DOI: https://doi.org/10.1121/10.0028170
  » https://doi.org/10.1121/10.0028170
• May-Collado, L. J. 2010. Changes in whistles structure of two dolphin species during interspecific associations. Ethology, 116(11), 1065-1074. DOI: https://doi.org/10.1111/j.1439-0310.2010.01828.x
  » https://doi.org/10.1111/j.1439-0310.2010.01828.x
• May-Collado, L. J. 2013. Guyana dolphins (Sotalia guianensis) from Costa Rica emit whistles that vary with surface behaviors. The Journal of the Acoustical Society of America, 134(4), 359-365. DOI: https://doi.org/10.1121/1.4818938
  » https://doi.org/10.1121/1.4818938
• May-Collado, L. J. & Wartzok, D. 2009. A characterization of Guyana dolphin (Sotalia guianensis) whistles from Costa Rica: the importance of broadband recording systems. The Journal of the Acoustical Society of America, 125(2), 1202-1213. DOI: https://doi.org/10.1121/1.3058631
  » https://doi.org/10.1121/1.3058631
• Mello, A. B., Molina, J. M. B., Kajin, M. & Santos, M. C. O. 2019. Abundance estimates of Guiana dolphins (Sotalia guianensis; Van Bénéden, 1864) inhabiting an estuarine system in southeastern Brazil. Aquatic Mammals, 45(1), 56-65. DOI: https://doi.org/10.1578/AM.45.1.2019.56
  » https://doi.org/10.1578/AM.45.1.2019.56
• Moreno, I. B., Zerbini, N. A., Danilewicz, D., Santos, M. C. de O., Simões-Lopes, P. C., Lailson-Brito Jr., J. & Azevedo, A. F. 2005. Distribution and habitat characteristics of dolphins of the genus Stenella (Cetacea: Delphinidae) in the southwest Atlantic Ocean. Marine Ecology Progress Series, 300, 229-240. DOI: https://doi.org/10.3354/meps300229
  » https://doi.org/10.3354/meps300229
• Moron, J. R., Amorim, T. O. S., Sucunza, F. & Castro, F. R., Rossi-Santos, M. & Andriolo, A. 2015. Spinner dolphin whistle in the Southwest Atlantic Ocean: is there a geographic variation? The Journal of the Acoustical Society of America, 138(4), 2495-2498. DOI: https://doi.org/10.1121/1.4931900
  » https://doi.org/10.1121/1.4931900
• Oswald, J. N., Barlow, J. & Norris, T. F. 2003. Acoustic identification of nine delphinid species in the eastern tropical Pacific Ocean. Marine Mammal Science, 19(1), 20-37. DOI: https://doi.org/10.1111/j.1748-7692.2003.tb01090.x
  » https://doi.org/10.1111/j.1748-7692.2003.tb01090.x
• Oswald, J. N., Rankin, S. & Barlow, J. 2004. The effect of recording and analysis bandwidth on acoustic identification of delphinid species. The Journal of the Acoustical Society of America, 116(5), 3178-3185. DOI: https://doi.org/10.1121/1.1804635
  » https://doi.org/10.1121/1.1804635
• Oswald, J. N., Rankin, S., Barlow, J. & Lammers, M. O. 2007. A tool for real-time acoustic species identification of delphinid whistles. The Journal of the Acoustical Society of America, 122(1), 587-595. DOI: https://doi.org/10.1121/1.2743157
  » https://doi.org/10.1121/1.2743157
• Oswald, J. N, Walmsley, S. F., Casey, C., Fregosi, S., Southall, B. & Janik, V. M. 2021. Species information in whistle frequency modulation patterns of common dolphins. Philosophical Transactions of the Royal Society B, 376(1836), 20210046. DOI: https://doi.org/10.1098/rstb.2021.0046
  » https://doi.org/10.1098/rstb.2021.0046
• Paschoalini, V. U. & Santos, M. C. O. 2020. Movements and habitat use of bottlenose dolphins, Tursiops truncatus, in south-eastern Brazil. Journal of the Marine Biological Association of the United Kingdom, 100(4), 651-662. DOI: https://doi.org/10.1017/S0025315420000387
  » https://doi.org/10.1017/S0025315420000387
• Papale, E., Azzolin, M., Cascão, I., Gannier, A., Lammers, M. O., Martin, V. M., Oswald, J., Perez-Gil, M., Prieto, R., Silva, M. A. & Giacoma, C. 2014. Macro- and micro-geographic variation of short-beaked common dolphin’s whistles in the Mediterranean Sea and Atlantic Ocean. Ethology Ecology & Evolution, 26(4), 392-404. DOI: https://doi.org/10.1080/03949370.2013.851122
  » https://doi.org/10.1080/03949370.2013.851122
• Papale, E., Gamba, M., Perez-Gil, M., Martin, V. M. & Giacoma, C. 2015. Dolphins adjust species-specific frequency parameters to compensate for increasing background noise. PLoS ONE, 10(4), e0121711. DOI: https://doi.org/10.1371/journal.pone.0121711
  » https://doi.org/10.1371/journal.pone.0121711
• Perrin, W. F. 2002. Common dolphins Delphinus delphis, D. capensis, and D. tropicalis. In: Perrin, W. F., Würsig, B. & Thewissen, J. G. M. Encyclopedia of marine mammals (pp. 245-248). California: Academic Press.
• Petrella, V., Martinez, E., Anderson, M. G. & Stockin, K. A. 2011. Whistle characteristics of common dolphins (Delphinus sp.) in the Hauraki Gulf, New Zealand. Marine Mammal Science, 28(3), 479-496. DOI: https://doi.org/10.1111/j.1748-7692.2011.00499.x
  » https://doi.org/10.1111/j.1748-7692.2011.00499.x
• Pinedo, M. C., Ploacheck, T., Barreto, A. S. & Lammardo, M. P. 2002. A note on vessel of opportunity sighting surveys for cetaceans in the shelf edge region off the southern coast of Brazil. Journal of Cetacean Research Management, 4(3), 323-329. DOI: https://doi.org/10.47536/jcrm.v4i3.846
  » https://doi.org/10.47536/jcrm.v4i3.846
• Pitman, R. L., Durban, J. W., Joyce, T., Fearnbach, H., Panigada, S. & Lauriano, G. 2019. Skin in the game: Epidermal molt as a driver of long-distance migration in whales. Marine Mammal Science, 36(2), 565-594. DOI: https://doi.org/10.1111/mms.12661
  » https://doi.org/10.1111/mms.12661
• Pivari, D. & Rosso, S. 2005. Whistles of small groups of Sotalia fluviatilis during foraging behavior in southeastern Brazil. The Journal of the Acoustical Society of America, 118(4), 2725-2731. DOI: https://doi.org/10.1121/1.2033569
  » https://doi.org/10.1121/1.2033569
• Quick, N. J. & Janik, V. M. 2008. Whistle rates of wild bottlenose dolphins (Tursiops truncatus): Influences of group size and behavior. Journal of Comparative Psychology, 122(3), 305-311. DOI: https://doi.org/10.1037/0735-7036.122.3.305
  » https://doi.org/10.1037/0735-7036.122.3.305
• R Core Team. 2019. R: A language and environment for statistical computing, In:. R Foundation for statistical computing (pp. 915-919), Vienna: R Foundation for statistical computing.
• Rankin, S., Archer, F., Keating, J. L., Oswald, J. N., Oswald, M., Curtis, A. & Barlow, J. 2016. Acoustic classification of dolphins in the California current using whistles, echolocation clicks, and burst pulses. Marine Mammal Science, 33(2), 520-540. DOI: https://doi.org/10.1111/mms.12381
  » https://doi.org/10.1111/mms.12381
• Rendell, L. E., Matthews, J. N., Gill, A., Gordon, J. C. D. & Macdonald, D. W. 1999. Quantitative analysis of tonal calls from five odontocete species, examining interspecific and intraspecific variation. Journal of Zoology, 249(4), 403-410. DOI: https://doi.org/10.1111/j.1469-7998.1999.tb01209.x
  » https://doi.org/10.1111/j.1469-7998.1999.tb01209.x
• Rubbens, P., Brodie, S., Cordier, T., Barcellos, D. D., Devos, P., Fernandes-Salvador, J. A., Fincham, J. I., Gomes, A., Handegard, N. O., Howell, K., Jamet, C., Kartveit, K. H., Moustahfid, H., Parcerisas, C., Politikos, D., Sauzède, R., Sokolova, M., Uusitalo, L., Van den Bulcke, L., van Helmond, A. T. M., Watson, J. T., Welch, H., Beltran-Perez, O., Chaffron, S., Greenberg, D. S., Kühn, B., Kiko, R., Lo, M., Lopes, R. M., Möller, K. O., Michaels, W., Pala, A., Romagnan, J.-B., Schuchert, P., Seydi, V., Villasante, S., Malde, K. & Irisson, J.-O. 2023. Machine learning in marine ecology: an overview of techniques and applications. ICES Journal of Marine Science, 80, 1829-1853. DOI: https://doi.org/10.1093/icesjms/fsad100
  » https://doi.org/10.1093/icesjms/fsad100
• Samarra, F. I. P., Deecke, V. B., Vinding, K., Rasmussen, M. H., Swift, R. J. & Miller, P. J. O. 2010. Killer whales (Orcinus orca) produce ultrasonic whistles. The Journal of the Acoustical Society of America, 128(5), 205-210. DOI: https://doi.org/10.1121/1.3462235
  » https://doi.org/10.1121/1.3462235
• Samarra, F. I. P., Deecke, V. B., Simonis, A. E. & Miller, P. J. O. 2015. Geographic variation in the time-frequency characteristics of high-frequency whistles produced by killer whales (Orcinus orca). Marine Mammal Science, 31(2), 688-706. DOI: https://doi.org/10.1111/mms.12195
  » https://doi.org/10.1111/mms.12195
• Santos, M. C. de O. & Netto, D. F. 2005. Killer whale (Orcinus orca) predation on a franciscana dolphin (Pontoporia blainvillei) in Brazilian waters. The Latin American Journal Aquatic Mammals, 4(1), 69-72. DOI: https://doi.org/10.5597/lajam00072
  » https://doi.org/10.5597/lajam00072
• Santos, M. C. de O. & Ditt, E. H. 2009. Incidental capture of a spinner dolphin, Stenella longirostris (Gray, 1828), in a shark gillnet off Southeastern Brazil. The Latin American Journal Aquatic Mammals, 7(1-2), 97-99. DOI: http://dx.doi.org/10.5597/lajam00143
  » http://dx.doi.org/10.5597/lajam00143
• Santos, M. C. O. & Silva, E. 2009. Records of a male killer whale (Orcinus orca) off southeastern Brazil. Brazilian Journal of Oceanography, 57(1), 65-68. DOI: https://doi.org/10.1590/S1679-87592009000100007
  » https://doi.org/10.1590/S1679-87592009000100007
• Santos, M. C. O., Siciliano, S., Vicente, A. F. C., Alvarenga, F. S., Zampirolli, É., Souza, S. P. & Maranho, A. 2010. Cetacean records along São Paulo state coast, southeastern Brazil. Brazilian Journal of Oceanography, 58(2), 123-142. DOI: https://doi.org/10.1590/S1679-87592010000200004
  » https://doi.org/10.1590/S1679-87592010000200004
• Santos, M. C. O., Figueiredo, G. C. & Van Bressem, M. F. 2017. Cetaceans using the marine protected area of “Parque Estadual Marinho da Laje de Santos”, Southeastern Brazil. Brazilian Journal of Oceanography, 65(4), 605-613. DOI: https://doi.org/10.1590/S1679-87592017130606504
  » https://doi.org/10.1590/S1679-87592017130606504
• Santos, M. C. O., Laílson-Brito, J., Flach, L., Oshima, J. E. F., Figueiredo, G. C., Carvalho, R. R., Ventura, E. S., Molina, J. M. B. & Azevedo, A. F. 2019. Cetacean movements in coastal waters of the southwestern Atlantic Ocean. Biota Neotropica, 19(2), e20180670. DOI: https://doi.org/10.1590/1676-0611-BN-2018-0670
  » https://doi.org/10.1590/1676-0611-BN-2018-0670
• Secchi, E. R. & Siciliano, S. 1995. Comments on the Southern range of the spinner dolphin (Stenella longirostris) in the western South Atlantic. Aquatic Mammals, 21, 105-108.
• Siciliano, S., Laílson-Brito, J. & Azevedo, A. F. 1999. Seasonal occurrence of killer whales (Orcinus orca) in waters of Rio de Janeiro, Brazil. International Journal of Mammalian Biology, 64, 251-255.
• Silva, V. M. F., Fettuccia, D., Rodrigues, E. S., Edwards, H., Moreno, I. B., Moura, J. F., Wedekin, L. L., Bazzalo, M., Emin-Lima, N. R., Carmo, N. A. S., Siciliano, S. & Utreras, V. B. 2010. Report of the working group on distribution, habitat, characteristics and preferences, and group size. The Latin American Journal of Aquatic Mammals, 8(1-2), 31-38. DOI: https://doi.org/10.5597/lajam00151
  » https://doi.org/10.5597/lajam00151
• Siroky, D. S. 2009. Navigating random forest and related advances in algorithmic modeling. Statistics Surveys, 3, 147-163. DOI: https://doi.org/10.1214/07-SS,033
  » https://doi.org/10.1214/07-SS,033
• Strobl, C., Malley, J. & Tutz, G. 2009. An introduction to recursive partitioning: Rationale, application, and characteristics of classification and regression trees, bagging, and random forests. Psychological Methods, 14(4), 323-348. DOI: https://doi.org/10.1037/a0016973
  » https://doi.org/10.1037/a0016973
• Thomsen, F., Franck, D. & Ford, J. K. B. 2001. Characteristics of whistles from the acoustic repertoire of resident killer whales (Orcinus orca) off Vancouver Island, British Columbia. The Journal of the Acoustical Society of America, 109(3), 1240-1246. DOI: https://doi.org/10.1121/1.1349537
  » https://doi.org/10.1121/1.1349537
• Webster, T. A., Dawson, S. M., Rayment, W. J., Parks, S. E. & Van Parijs, S. M. 2016. Quantitative analysis of the acoustic repertoire of southern right whales in New Zealand. The Journal of the Acoustical Society of America, 140(1), 322-333. DOI: https://doi.org/10.1121/1.4955066
  » https://doi.org/10.1121/1.4955066
• Wells, R. S. & Scott, M. D. 2017. Common bottlenose dolphin, Tursiops truncatus, In: Perrin, W. F., Würsig, B. & Thewissen, J. G. M. Encyclopedia of marine mammals (pp. 249-255). California: Academic Press, California.
• Yang, L., Sharpe, M., Temple, A. J., Jiddawi, N., Xu, X. & Berggren, P. 2020. Description and classification of echolocation clicks of Indian Ocean humpback (Sousa plumbea) and Indo-Pacific bottlenose (Tursiops aduncus) dolphins from Menai Bay, Zanzibar, East Africa. PLoS ONE, 15(3), e0230319. DOI: https://doi.org/10.1371/journal.pone.0230319
  » https://doi.org/10.1371/journal.pone.0230319
• Zanardi-Lamardo, E., Bícego, M. C. & Weber, R. R. 2013. The fate of an oil spill in São Sebastião channel: a case study. Brazilian Journal of Oceanography, 61(2), 93-104. DOI: https://doi.org/10.1590/S1679-87592013000200002
  » https://doi.org/10.1590/S1679-87592013000200002

• AI USE DISCLOSURE
  The authors declare that generative artificial intelligence (AI) tools as ChatGPT were used exclusively to refine the English language. The content was carefully reviewed by the authors, who are fully responsible for the final version of the manuscript.
• FUNDING
  Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - Brazil (CAPES) - Finance Code 001; Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq) Grant/Award n° 163148/2015-5, Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP) Processo n° 2011/51543-9, Processo n° 2011/50317-5, and Alucia/Ocean X.

Edited by

• Editor:
  Rubens Lopes

Publication Dates

• Publication in this collection
  31 July 2026
• Date of issue
  2026

History

• Received
  01 Oct 2025
• Accepted
  22 Mar 2026