Guiana dolphin strandings are spatially concentrated and increasing in Brazil

Guiana dolphins (Sotalia guianensis) serve as critical environmental sentinels, yet they face escalating threats from human activities in coastal and estuarine ecosystems. Despite extensive data collection, integrated assessments of large-scale stranding patterns in this endangered species remain limited. This study provides a comprehensive spatiotemporal evaluation of strandings along 2,785 km of the Brazilian coast from 2020 to 2025, utilizing standardized data from four beach monitoring programs. We implemented a methodologically robust framework that combined kernel density estimation with extreme value statistics to identify persistent stranding hotspots and assess regional differences. Our results revealed a significant annual increase in strandings of approximately five percent, with the risk of these occurrences being spatially concentrated rather than evenly distributed. Significant regional variability was observed, with the highest stranding intensities and most extreme hotspots located in the Santos Basin, followed by the Sergipe-Alagoas and Campos-Espírito Santo regions. These identified hotspots coincide with heavily impacted estuarine complexes that host resident populations but currently lack adequate protection. By demonstrating that high-intensity stranding hotspots are often excluded from fully protected conservation units, this study highlights substantial gaps in marine management. These findings establish an empirical basis for spatial prioritization and targeted mitigation strategies, offering a quantitative framework to support national-scale conservation planning for the long-term persistence of this species.

Coastal and estuarine ecosystems rank among the most biologically productive environments, yet they are highly vulnerable to anthropogenic pressure. Urban expansion, vessel traffic, chemical and acoustic pollution, and intensifying economic activities profoundly alter the structural and functional dynamics of these systems, impacting organisms that rely on shallow coastal zones, bays, and estuaries for feeding, reproduction, and resting (Prosser et al., 2018; Kennish, 2021). Among these species, marine mammals serve as environmental sentinels because they respond sensitively to ecological disturbances and multiple forms of coastal degradation (Bossart, 2011; Fossi and Panti, 2017).

Stranding monitoring is a fundamental tool for assessing occurrence patterns, anthropogenic impacts, mortality, and spatial distribution of marine megafauna. When conducted systematically using standardized protocols, this type of monitoring enables the detection of temporal trends, episodic mortality events, disease outbreaks, fishery-related impacts and oceanographic phenomena, in addition to providing a continuous data record that is rarely achievable through open ocean observations (Alvarado-Rybak et al., 2020; Cantor et al., 2020; Lennon et al., 2025; Rodrigues et al., 2025; Wund et al., 2023) In Brazil, this activity has expanded substantially with the implementation of Beach Monitoring Programs (PMPs), established as mandatory requirements within the federal environmental licensing process for oil exploration, coordinated by the Brazilian Institute of the Environment and Renewable Natural Resources (IBAMA) (Prado et al., 2023; Rodrigues et al., 2025).

The Guiana dolphin (Sotalia guianensis) is among the species most frequently recorded by the Beach Monitoring Programs (PMPs) (Prado et al., 2023). This odontocete is widely distributed along the western Atlantic and strongly associates with estuarine and shallow coastal habitats (Santos and Rosso, 2007; Espinoza-Rodríguez et al., 2019; Lobo et al., 2021). This strong affinity for nearshore environments, which are often modified by intense anthropogenic use, renders the species particularly vulnerable to human-induced pressures. These threats include incidental bycatch in gillnets (de Moura et al., 2009; Di Beneditto, 2003; Zappes et al., 2009), habitat degradation, contamination, and infectious diseases outbreaks (de Moura et al., 2014; Flach et al., 2019; Gonzales-Viera et al., 2013). Consequently, S. guianensis is currently classified as “endangered” in Brazil (Farro et al., 2025) and accounts for a substantial proportion of all strandings documented along the Brazilian coast (Carvalho and Meirelles, 2020; Mayorga et al., 2020; Prado et al., 2023).

Although PMPs have accumulated an unprecedented volume of data in recent years (Prado et al., 2023), integrated analyses comparing spatial and temporal stranding dynamics across different regions remain scarce (Dudhat et al., 2022a; Petitguyot et al., 2025). This gap limits the ability to identify consistent regional patterns, compare relative risk levels, and support conservation strategies at the national scale. Biogeographical, oceanographic, and local abundance differences suggest that the stranding patterns of Guiana dolphins may vary substantially along the Brazilian coast (Carvalho and Meirelles, 2020; Cunha et al., 2020; Farro et al., 2025), potentially reflecting both true ecological variation and differences in monitoring effort or detection efficiency among projects. Comparative analyses using standardized PMP datasets are essential for addressing these gaps.

This study aimed to provide an integrated assessment of the spatiotemporal patterns of Guiana dolphin strandings along the Brazilian coast using standardized data from four PMPs. Our analysis encompasses regional differences in stranding intensity, the identification of spatial hotspots derived from kernel density estimation (KDE) and upper-tail metrics, and the characterization of seasonal and interannual variability based on effort-standardized indices and trend models. By synthesizing harmonized datasets across distinct coastal regions, this study establishes a quantitative framework for comparing risk profiles, delineating high-priority areas, and supporting conservation planning for S. guianensis.

Study area

The study encompassed the operational ranges of four Beach Monitoring Programs (PMPs) along the Brazilian coast (Figure 1), implemented under federal environmental licensing requirements and conducted using standardized and systematic terrestrial surveys, carried out by ground teams using 4x4 vehicles or on foot. Together, these programs continuously monitor approximately 2,785 km of coastline. The PMP BS (Santos Basin; BS) covers the coastal stretch from Laguna (Santa Catarina) to Saquarema (Rio de Janeiro), totaling approximately 1,500 km. The PMP BCES (Campos and Espírito Santo Basins; BCES) extends from Arraial do Cabo (Rio de Janeiro) to Conceição da Barra (Espírito Santo), covering roughly 700∑km. The PMP SEAL (SEAL) spans the region from Conde (Bahia) to Pontal do Peba (Alagoas), encompassing 270∑km. The PMP RNCE (RNCE) covers the coastline from Caiçara do Norte (Rio Grande do Norte) to Aquiraz (Ceará), with an approximate length of 325∑km (SIMBA, 2026). These regions represent distinct oceanographic, geomorphological, and anthropogenic pressures (de Sousa Pais et al., 2018; Lessa et al., 2018; Lopez Marulanda et al., 2026), enabling comparative analyses across coastal areas characterized by different environmental settings.

Monitoring is predominantly conducted daily in PMP SEAL, PMP BCES, and PMP BS, covering at least 90% of the coastline. The remaining areas, generally associated with limited access, are surveyed at lower frequency, typically weekly or biweekly. In contrast, PMP RNCE follows a more heterogeneous design, with 57.8% of the coastline monitored daily and the remainder surveyed at intervals of up to 20 days. To mitigate potential detection gaps in these lower-frequency areas, a community-based ‘Trigger and Registration’ protocol is implemented. This system allows for immediate notification by residents, which triggers an ad-hoc field response for data collection and georeferencing, ensuring that strandings are recorded regardless of the scheduled survey frequency. All records, whether from systematic surveys or community triggers, are standardized and integrated into the SIMBA database.

Stranding dataset

Stranding records of Sotalia guianensis from 2020 to 2025 were obtained from the Aquatic Biota Monitoring System (SIMBA - https://simba.petrobras.com.br/simba/web/), which consolidates primary data collected by Beach Monitoring Programs (PMPs). Compiled metadata included geographic coordinates, dates, localities, and additional field information. Quality control procedures involved removing duplicate entries, verifying inconsistent coordinates, and exclusion of records lacking reliable spatial positioning. Consequently, spatial analyses relied exclusively on valid georeferenced records.

In addition to spatial data, information on sex and life stage (categorized as adult, juvenile, or calf based on total body length and physical maturity) was extracted from the SIMBA metadata to characterize the demographic composition of strandings across regions.

Biological and anthropogenic data

Statistical Analysis of Demographic Data.

To evaluate differences in demographic structure, a Chi-square test of independence was used to compare life stage proportions among PMPs, followed by standardized residual analysis. Sex ratios within each region were tested for significant deviations from the expected 1:1 proportion using binomial tests.

Vessel Traffic Data.

To evaluate anthropogenic pressure, the Global Ship Density layer from the IMF World Seaborne Trade Monitoring System (Cerdeiro et al., 2020) was integrated into the spatial analysis. This raster dataset represents the aggregate of all ship categories based on AIS positions recorded between 2015 and 2021 at a 0.005° resolution (~500m), serving as a proxy for general maritime traffic intensity.

Stranding index per unit effort

The monthly stranding intensity was standardized using the Stranding Index per Unit Effort (I 100 /month), which quantifies the number of strandings per 100∑km of coastline monitored each month. This metric follows the same conceptual framework as the standardized Catch Per Unit Effort (CPUE) indices widely used in fisheries science, where effort standardization is essential to remove spatial and temporal heterogeneity and allow valid comparisons among regions and monitoring schemes. Such CPUE-based standardization approaches are well-established in the literature (Ricker, 1940).

The index is computed as shown in Equation 1, with each term defined as follows:

Equation 1 — I 100 /month.

Term definitions.

N — Number of strandings recorded each month.

km monitored — Total length of coastline (in km) effectively monitored by each PMP.

I 100 / m o n t h — Stranding Index per 100km per month (standardized rate of strandings relative to monitoring effort).

Spatial analysis

Kernel Density Estimation (KDE).

Spatial concentration areas of strandings were identified using Kernel Density Estimation (KDE) within a GIS framework (Silverman, 2018) (Silverman, 2018). Georeferenced records were converted into continuous raster surfaces representing spatial intensity at a spatial resolution of approximately 0.01° (~1 km). Pixel values were extracted after removing NA cells. Hotspots were defined using an objective quantile-based criterion, classifying pixels in the upper 5% of density values (quantile ≥ 0.95) as high-intensity areas. The distribution of the KDE values was summarized using boxplots based on quartiles and the interquartile range (IQR), applying the 1.5 × IQR rule to identify extreme values ((Tukey, 1977). To further highlight the critical concentration zones, the upper tail of the distribution was examined by isolating pixels above the elevated quantiles (Q95 and Q99). All spatial preprocessing, raster extraction, and quantile-based hotspot classification were performed in R (version 2024.12.1, RRID: SCR_001905) using the sf, terra, dplyr, and ggplot2 packages. Spatial data were processed using QGIS (QGIS 3.34.8, RRID: SCR_018507).

Tail-based Comparison Among Areas (quantile inference).

To assess the differences in extreme values among PMPs, the 0.95 and 0.99 quantiles of the kernel density distributions were directly compared across areas. To account for the spatial autocorrelation inherent in the KDE smoothing process, the nonparametric bootstrap procedure (1,000 resamples per PMP) resampled the original georeferenced stranding locations rather than the derived raster pixels (Efron and Tibshirani, 1994). At each iteration, density values were extracted at the resampled observation points to recalculate the quantiles. Paired bootstrap distributions were used to estimate confidence intervals and two-tailed p-values, and Holm correction was applied to adjust for multiple comparisons (Holm, 1979). All quantile computations and bootstrap analyses were conducted in R 2024.12.1 using dplyr and base R functions.

For the upper quantile and distributional graphical analyses, pixel-level density values extracted from each KDE-derived raster were compiled into a single dataset grouped by PMP. For the upper tail analysis, only pixels with density values above the Q95 threshold were included. For the full distribution analysis, all density values were used, and the upper tail was defined as values classified as outliers according to the 1.5 × IQR rule. This approach allowed us to identify both localized hotspots and broader spatial patterns. All statistical analyses and graphical processing were performed in R using the ggplot2 package.

Temporal analysis

Annual Temporal Trend.

The annual temporal trend in stranding counts was evaluated using Ordinary Least Squares (OLS) regression applied to both the aggregated annual series and each PMP-specific series. From these models, the slope coefficient, coefficient of determination (R²), and p-value were estimated (Draper and Smith, 1998) As a complementary nonparametric approach, the Kendall trend test was used to detect monotonic temporal patterns (Kendall and Gibbons, 1990). In addition, a log-linear model was applied to estimate the Annual Percent Change (APC) and its corresponding confidence intervals (Kim et al., 2000). Similarly, the negative binomial GLM confirmed this increasing trend, yielding an incidence rate ratio (IRR∑=∑1.052, 95% CI: 1.027–1.077, p < 0.001), corresponding to an approximate annual increase of 5.2%.

Annual Amplitude of Monthly Stranding Indices.

To quantify the intra-annual variability in stranding patterns across Beach Monitoring Projects (PMPs), we calculated the annual amplitude of the monthly stranding index for each PMP between 2020 and 2025. This metric reflects the magnitude of seasonal fluctuations by capturing the full range of monthly stranding activity within each year, defined as the difference between the highest and lowest monthly index values (A = I max - I min). Because zero-value months were retained, this measure captured the complete range of observed variation—from peak stranding periods to months with no detections—providing a biologically meaningful representation of seasonal dynamics. Finally, annual amplitude values were summarized using boxplots to display the median, interquartile range, and distribution shape for each PMP, enabling direct a comparison of seasonal fluctuation intensity and regional differences in stranding dynamics.

Use of AI-assisted tools

Data analysis was conducted using R scripts developed with the support of the Gemini 1.5 Pro large language model (Google). AI-assisted tools supported code development, debugging, and optimization for spatial and temporal analyses (including kernel density estimation, bootstrap-based quantile inference, and temporal trend modeling using OLS, log-linear models, and GLMs), as well as text checking and editing. The authors thoroughly reviewed, tested, and validated all AI-generated codes and text adjustments to ensure scientific accuracy and reproducibility.

From 2020 to 2025 (n = 6), the mean annual total number of strandings was 391.2 ± 16.7 (SE). Regional means were 233.0 ± 12.6 in BS, 107.0 ± 8.5 in BCES, 41.7 ± 4.0 in SEAL, and 9.5 ± 1.3 in RNCE. Counts were concentrated in BS and BCES, which together accounted for most records across the series, whereas SEAL and RNCE contributed smaller shares of the total. Interannual variation increased with count magnitude, with greater dispersion in the highest-count region and lower dispersion in regions where totals remained consistently low.

The demographic structure of stranded individuals revealed significant regional differences. The age structure differed significantly among PMPs (χ² = 15.23; df = 6; p = 0.019), although the effect size was small (Cramér’s V∑=∑0.065). Standardized residual analysis indicated that the BCES region had a significantly higher frequency of calves (residual = 2.48) than expected under independence. Regarding sex, a predominance of males was observed in all regions. However, significant deviations from the expected 1:1 sex ratio were confirmed only in SEAL (65.4% males; p < 0.001) and BS (59.6% males; p < 0.001), while differences in RNCE and BCES were not statistically significant (p > 0.05).

The hotspots (top 5% of density values) were concentrated in specific municipal clusters within each PMP area. In PMP BCES, key hotspots were confirmed in Vitória Bay and at the Doce River mouth, areas that also show significant overlap with high ship density. In PMP BS, the most intense stranding peak was identified in Paranaguá, followed by high-intensity hotspots in Sepetiba and Babitonga Bays, both of which are regions characterized by intense vessel traffic. In PMP SEAL, a prominent stranding peak was identified in the Sergipe River estuary, notably in a zone with lower maritime traffic compared to southern hotspots Finally, in PMP RNCE, stranding clusters were localized in Porto do Mangue and Tibau–Icapuí, where intense navigation is notably absent near the coast Hotspots were consistently concentrated in the same coastal areas, indicating a persistent spatial aggregation between 2020 and 2025 (Figure 2).

The analysis of the upper tail of the kernel density surfaces (95th and 99th percentiles) revealed strong heterogeneity among PMPs. BS exhibited the highest extreme values (Q95∑=∑23.06; Q99∑=∑52.59; maximum = 131.02), the widest tail range (Δ top = 107.96), and the largest spatial extent of hotspots (> Q95), reflecting significantly greater stranding intensity and spatial concentration. It was followed by SEAL and BCES, while RNCE presented the lowest upper-tail metrics (Table 1). Pairwise comparisons of the 95th percentile were highly significant for all PMP combinations (p < 0.001; Table 1), supporting a consistent ranking of hotspot intensity (BS > SEAL > BCES > RNCE). The graphical pattern, with Q95 thresholds progressively shifted to the right, matches the numerical results and indicates that the main regional differences are concentrated in the extreme-value domain, that is, hotspot zones (Figure 3).

Comparative boxplots of kernel density pixel values corroborate this pattern, showing strongly right-skewed distributions in all PMPs but with markedly different spreads and upper-tail behaviors. BS exhibited the broadest dispersion and the most pronounced upper tail, with numerous extreme outliers extending well beyond the interquartile range, consistent with dense and highly concentrated hotspot cores. SEAL and BCES showed intermediate variability, with wider interquartile ranges and moderate numbers of high-value outliers. In contrast, RNCE had the narrowest distribution and lowest overall values, with a limited spread and few extreme pixels. The median and interquartile patterns followed the same ordering observed in the upper-quantile analyses, reinforcing that regional contrasts are driven primarily by differences in extreme-density pixels rather than central tendencies (Figure 4). Pairwise quantile comparisons confirmed the significant differences among all PMPs (Table 2).

Trend analyses of the pooled dataset (all PMPs combined) indicated a significant overall increase in Guiana dolphin (Sotalia guianensis) strandings over time (Figure 5). The linear model showed a positive slope (slope = 19.74), strong explanatory power (R² = 0.81), and a significant effect (OLS p = 0.014). The nonparametric Kendall trend test corroborated this result, indicating a strong positive association (τ = 0.87, p = 0.017) between the two variables. Log-linear modeling produced a consistent pattern, with a significant annual percent change (APC∑=∑5.27% per year) and a 95% confidence interval that did not include zero (2.63% to 7.99%; p = 0.017). Similarly, the negative binomial GLM confirmed this increasing trend, yielding an incidence rate ratio (IRR∑=∑1.052, 95% CI: 1.027–1.077, p < 0.001), corresponding to an approximate annual increase of 5.2%.

The monthly standardized indices (strandings per 100∑km ± standard error) showed distinct seasonal profiles among PMPs (Figure 6). SEAL presented the highest indices in the first quarter of the year, peaking between February and March, followed by a marked decline in late autumn and winter and a secondary increase toward the end of the year. BS displayed a more gradual pattern, with lower values in the first semester and a progressive increase from mid-year onward, reaching higher values between August and October. BCES showed intermediate behavior, with moderate early year values, a mid-year dip, and a secondary peak in late winter–spring. In contrast, the RNCE consistently exhibited the lowest indices throughout the year, with minimal seasonal amplitude and near-zero values in some mid-year months. Although some monthly values overlapped, the overall patterns suggested clear seasonality in the BS, SEAL, and BCES, whereas the RNCE remained comparatively stable throughout the year.

The annual amplitude of the monthly stranding indices (per 100∑km) differed markedly among the four PMPs (Figure 6B), indicating strong regional contrasts in seasonal variability. SEAL exhibited the largest amplitudes (median = 3.33; range = 2.59–4.07), denoting pronounced seasonal peaks and the highest intra-annual fluctuation. BCES showed intermediate variability (median = 2.07; range = 1.29–3.43), broader than BS. The BS area presented low and highly consistent amplitudes (median = 1.50; range = 1.27–1.73), indicating stable intra-annual patterns. RNCE displayed the smallest amplitudes (median = 0.79; range = 0.63–0.95). Overall, the boxplots revealed a clear gradient of temporal variability from SEAL (highest) to RNCE (lowest), reflecting distinct regional dynamics.

Our results show that Guiana dolphin strandings along the Brazilian coast are spatially and temporally uneven, with clear regional differences in hotspot intensity and seasonal dynamics (Figures 2, 6). These spatial and seasonal patterns likely reflect multiple interacting drivers, including differences in local population density, gradients of anthropogenic pressure such as fishery interactions, vessel traffic, and coastal degradation, and geomorphological and oceanographic factors that affect carcass drift, transport, and deposition (Farro et al., 2025; IWC, 2024). The observed spatiotemporal heterogeneity in strandings is consistent with previous studies showing that stranding records reflect not only biological mortality but also coastal physical processes and regional threat regimes and should, therefore, be interpreted as a composite indicator of exposure and risk at the population level (Peltier et al., 2014a; Carvalho and Meirelles, 2020; Cunha et al., 2020; Almeida et al., 2024; Ruenes et al., 2024).

Systematic stranding monitoring is widely recognized as an efficient and cost-effective tool for obtaining information on cetacean species that are difficult to observe in the wild because of their coastal habits or elusive behavior (Pikesley et al., 2012; Prado et al., 2016; Dudhat et al., 2022a).

An analysis of the consolidated dataset from the four Beach Monitoring Programs (BS, BCES, SEAL, and RNCE), which cover extensive sedimentary basins along the Brazilian coast, revealed a concerning magnitude of impact. Between 2020 and 2025, these programs recorded an average of 391.2 ± 16.7 SE Guiana dolphin strandings per year within their combined monitoring areas. Because stranding records represent only a fraction of total mortality at sea (Prado et al., 2013; Authier et al., 2014; Peltier et al., 2014b), a sensitivity calculation using the commonly reported detection rate of 10–30% was performed. Applying the commonly cited detection rate of ten-30 percent, the projected annual true mortality for these regions ranged from ~1.300 to ~3.900 individuals. However, it is crucial to emphasize that this calculation is intended as a conservation illustration rather than a formal mortality estimate. A robust and precise quantification of true mortality would require the incorporation of region-specific data on carcass drift patterns, persistence times on the beach, and local detection probabilities, which are influenced by variable environmental factors such as currents, winds, and shoreline morphology (Prado et al., 2013; Authier et al., 2014; Peltier et al., 2014b). In addition to this absolute scale, our study identified a significant annual increase of 5.2% in the total recorded strandings between 2020 and 2025 (Figure 5). This statistically supported upward trend suggests that the phenomenon cannot be solely attributed to random fluctuations or sampling noise. Although gradual improvements in detection efficiency may account for a small portion of the increase, the consistent rise in these records indicates a notable escalation in recorded stranding intensity within this six-year observational window. Nevertheless, inferring definitive long-term demographic shifts from a six-year dataset requires caution, as this represents a relatively short observational window for understanding the complex population dynamics of these long-lived cetaceans. Despite this limitation, this progressive increase, combined with the species’ low reproductive capacity, characterized by late sexual maturity (five to eight years) and long intervals between births (two to three years) (de Oliveira Santos et al., 2001; Rosas and Monteiro-Filho, 2002), highlights the need for continued monitoring and evaluation of potential impacts on local populations. Although the total abundance of the species along the Brazilian coast remains unknown (Santos et al., 2010; Farro et al., 2025), the recurrent loss of hundreds of individuals is alarming, given that most studied populations consist of only a few hundred animals (Santos et al., 2010; de Moura et al., 2026).

This pattern reflects that the stranding risk for coastal cetaceans is not spatially uniform. Instead, recorded strandings tends to concentrate in discrete areas where critical habitat use overlaps with intense human activity (de Moura et al., 2023; IWC, 2024; Farro et al., 2025). Such spatial coincidence creates localized risk amplification, producing clusters of higher stranding intensity that shape the overall regional hierarchy.

The standardized spatial comparison revealed a robust hierarchy in stranding intensity among the monitored regions, with the Santos Basin exhibiting the strongest upper tails and highest extreme kernel density values, followed by SEAL, BCES, and RNCE (Figure 3 and 4). This ordering shows that the differences arise precisely in the most extreme fraction of the distribution, that is, in the hotspots, rather than in the central metrics. This finding suggests that stranding intensity risk is spatially concentrated instead of evenly distributed. Areas of intense human activity appear to overlap with the key habitats used by Guiana dolphins (de Moura et al., 2023; IWC, 2024; Farro et al., 2025). In the BS and BCES regions, the identified hotspots coincided with large estuarine and lagoon complexes recognized as intensive-use areas for the Guiana dolphin, including systems such as Paranaguá, and Sepetiba. These regions host some of the largest known resident populations of the species and simultaneously exhibit high levels of artisanal gillnet fishing, high ship density values identified in the shipping raster’s, and environmental degradation (Santos et al., 2010; IWC, 2024; de Moura et al., 2026; Guidio et al., 2026). Thus, the higher stranding intensity in these areas likely reflects not only greater local abundance but also increased exposure to lethal interactions, particularly gillnet entanglement, a pattern widely documented in the literature (Ott et al., 2002; Santos et al., 2010; Prado et al., 2016; Carvalho and Meirelles, 2020; Ruenes et al., 2022; Almeida et al., 2024; IWC, 2024; Farro et al., 2025).

The significantly higher frequency of calves recorded in the BCES region suggests that mortality may be disproportionately affecting the most vulnerable age classes. When considered alongside the elevated stranding intensity observed in this region, this demographic pattern is consistent with the hypothesis that cumulative anthropogenic pressures such as high vessel traffic intensity and habitat degradation may be contributing to increased mortality risk. A potential contributing factor is the long-term environmental legacy of the 2015 Fundão dam collapse, which released more than 50 million cubic meters of mining tailings that ultimately reached the Atlantic Ocean (Gomes et al., 2017). Notably, one of the principal stranding hotspots identified in the BCES region is located at the mouth of the Doce River, where the tailings first entered the marine environment. The resulting plume affected extensive estuarine and coastal ecosystems and overlapped with areas inhabited by resident cetacean populations (Giacomo et al., 2021; Amorim et al., 2022). Although a direct causal relationship between the Fundão disaster and Guiana dolphin mortality cannot be established from the present data, the persistence of environmental disturbances in these habitats, compounded by active shipping lane pressures, may represent an additional source of ecological stress for local populations. In contrast, the lower hotspot intensity observed in the RNCE region (Figures 3, 4) may result from a combination of factors, including smaller resident populations (Monteiro-Filho, 2025), a narrower continental shelf (Claudino-Sales et al., 2018), distinct hydrodynamics (Stramma and England, 1999), and potentially reduced coastal carcass retention. The absence of detectable seasonality (Figure 6) in this basin further supports the hypothesis that stressors operate there in a more constant manner and at a comparatively lower magnitude, in contrast to the more pulsed regimes observed in the southern and southeastern basins.

The hotspot identified in the PMP SEAL region corresponds to the Sergipe River estuary, a critical feeding and breeding area for Sotalia guianensis. Notably, this area lacks intense maritime traffic, which suggests that the high stranding intensity is primarily driven by other anthropogenic pressures, such as artisanal gillnet bycatch (Lima, 2010). Although some studies on habitat use and behavior exist (Nunes et al., 2014), the region still suffers from a severe data deficiency and is classified as a Management Unit lacking abundance estimation effort (IWC, 2024). In this context, continued long-term monitoring and studies on population parameters are essential to assess population viability and the impacts of anthropogenic pressures on these resident groups.

Additionally, a higher proportion of males was recorded among stranded individuals in the SEAL (65.4%) and BS (59.6%) regions. A comparable male-biased pattern has also been reported for the Paranaguá Estuarine Complex, located within the BS region, where Guidio et al. (2026) observed a male-to-female ratio of approximately 1.7:1 among stranded Guiana dolphins. While these findings point to a possible recurrence of male predominance in stranding datasets, caution is warranted when interpreting this result because stranding records may not accurately reflect the demographic composition of living populations. Additional long-term monitoring and demographic analyses are needed to determine whether the observed bias represents a consistent biological pattern or results from factors affecting stranding and detection processes.

The spatial patterns of Sotalia guianensis stranding reveal substantial gaps in the territorial protection of the species along the Brazilian coast. The areas of highest stranding intensity identified in this study largely coincide with coastal regions under strong anthropogenic pressure and limited coverage by fully protected areas. When present, these areas are predominantly sustainable use categories that restrict their ability to mitigate the impacts associated with fisheries, vessel traffic, and the degradation of estuarine habitats (Farina and Rennó Braga, 2019). These protection gaps may contribute to the high stranding intensity observed in regions occupied by resident populations with strong site fidelity and a high dependence on heavily impacted estuaries and bays, such as Paranaguá and Sepetiba (Lobo et al., 2021; de Moura et al., 2023; Cezimbra et al., 2025).

The analysis of tails and hotspots proved particularly suitable for asymmetric distributions, such as stranding density raster data. Upper quantiles described critical cores more effectively than global means, generating more informative metrics for spatial prioritization. This strategy, based on uppertail characterization and the identification of extreme values, is aligned with established approaches in extreme value statistics and spatial hotspot analysis (Anselin, 1995; Coles, 2001), yet it remains underexplored in studies of cetacean strandings and spatial distribution, where aggregated metrics and mean density estimates still predominate (Norman et al., 2012; Prado et al., 2016, Prado et al., 2023). The combination of systematic monitoring, extreme value statistics, and spatial analysis provides a methodologically robust framework for comparing stranding hotspots, and consequently offers substantial value for the conservation of coastal cetaceans.

Although PMPs provide valuable spatial data on Sotalia guianensis strandings, their contribution to conservation depends on the effective incorporation of this information into management actions (Parsons et al., 2015a; Reisinger et al.,). The persistence of major threats to coastal small cetaceans in Brazil, especially fishery interactions, habitat degradation, and vessel traffic, has been widely documented (Reeves et al., 2013; Nelms et al., 2021; Secchi et al., 2021; Farro et al., 2025). The absence of more specific regulatory measures, such as adjustments in fishing areas, improvements in maritime traffic management, or the expansion of fully protected areas, maintains resident populations in a state of chronic vulnerability and highlights a recurrent gap between monitoring efforts and effective conservation outcomes (Parsons et al., 2015a; Reisinger et al.,).

Due to their spatial scale and sampling regularity, the Beach Monitoring Programs represent one of the largest continuous marine stranding monitoring efforts worldwide and the most extensive in Brazil (SIMBA, 2026). With standardized daily surveys covering thousands of kilometers of coastline, the program has converted what was once largely opportunistic reporting into a robust dataset that is comparable across time and space (Prado et al., 2023; Baes et al., 2024). As a requirement for federal environmental licensing conducted by IBAMA, the PMP produces an unprecedented national dataset that supports the identification of stranding trends and critical hotspots for the Guiana dolphin (Sotalia guianensis) and provides a strong empirical basis for management strategies and marine conservation policies.