| Introduction

Understanding how sympatric marine species use and partition resources is critical in community ecology. When multiple functionally similar species co-exist, they may vary their ecological niche (i.e., resources and habitat) to limit interspecies competition for resources (Pianka 1976). Investigations of dietary niche can be important in understanding how resilient species might be to changes in their prey and environment. For example, some species may have broad dietary niches, potentially allowing them to adapt to changes in resource availability (e.g., Caputo et al. 2021). When consumers have a specialized diet or occupy a unique niche within the community, changes in the environment and prey abundance may negatively impact their ecology and have community-level consequences as they may have to use lower quality prey or travel farther to find resources (Piroddi et al. 2011). This may be especially problematic in areas with high human interactions, including coastal marine habitats. Therefore, it is critical to the conservation and management of communities to assess the trophic interactions of consumers to predict how changes in prey dynamics may affect food webs and ecosystems (Bowen 1997;Heithaus et al. 2008;Estes et al. 2016).

Investigating the foraging ecology of marine predators represents a major challenge in ecology, especially for cryptic species that are rarely observed and occur in remote areas where access to samples is logistically difficult. This is particularly true in areas where traditional dietary studies using stomach content analysis are not possible due to obstacles in the collection of samples (e.g., from strandings, incidental/targeted captures). Increasingly, studies on the trophic interactions of marine predators have relied on stable isotope analyses, as advancements in analytical tools over the past decade have improved our ability to reconstruct the diet of consumers based on predator-prey isotope data (e.g., Barros et al. 2010;Borrell et al. 2021;De Loizaga Castro et al. 2016). Stable isotopes within an animals' tissues reflect the primary producers in their foraging habitat (δ 13 C) and the trophic level of their prey (δ 15 N) (Newsome et al. 2010). These isotopes pass between prey and consumers in a predictable way called trophic enrichment, which can be generally approximated by a trophic enrichment factor (TEF). Advances in modeling of isotopic niche space, specifically SIBER metrics in R, allow us to effectively describe an animals' isotopic niche by drawing ellipses around the individual isotope values from a population or community. The area of these ellipses and how they overlap with the niche of other species can then be compared (Jackson et al. 2011) and used to assess how resources are partitioned and whether consumers play distinct or overlapping roles within a given community. While it remains challenging to identify specific prey in a consumers' diet from isotopes analysis, Bayesian mass-balance mixing models can allow researchers to make indirect inferences about prey consumed when prior knowledge of potential candidate prey is available. Though not as detailed as dietary information from stomach content analyses, these models estimate the contribution of potential prey sources to a consumers' diet using the stable isotope signatures from each (Stock et al. 2018). These tools allow researchers to investigate how species mitigate competition for resources, which is especially useful when many sympatric species co-exist.

The waters surrounding the volcanic island of Nosy Be, in northwestern Madagascar, are considered a global biodiversity hotspot, especially for marine megafauna (Cerchio et al. 2022). This is likely due to productive conditions from upwelling (Vianello et al. 2020) and the presence of a mosaic of marine habitats (e.g., mangroves, coral reefs, seagrass, continental slopes, and deep ocean; Obura et al. 2012). At least 27 species of cetaceans have been confirmed to occur in Madagascar, including 8 mysticete and 19 odontocete species. At least 17 species of odontocetes have been confirmed in north-western Madagascar (Cerchio et al. 2022). In coastal waters, Indo-Pacific bottlenose dolphin (Tursiops aduncus) and Indian Ocean humpback dolphin (Sousa plumbea) are the most frequently encountered species (Cerchio et al. 2015). Other species occur on reef banks, on the continental shelf and slope, and in deep oceanic waters of the region with sperm whales (Physeter macrocephalus), melon-headed whales (Peponocephala electra), short-finned pilot whales (Globicephala macrorhynchus), false killer whales (Pseudorca crassidens), spinner dolphins (Stenella longirostris), pantropical spotted (Stenella attenuata), and common bottlenose dolphins (T. truncatus) being among the most common species (Cerchio et al. 2022;Kiszka 2015).

In the coastal waters off Madagascar, T. aduncus and S. plumbea are threatened by a range of anthropogenic activities that are primarily related to artisanal fisheries bycatch and the growing tourism development in the region (Cerchio et al. 2009(Cerchio et al. , 2015;;Kiszka et al. 2009;Temple et al. 2018). S. plumbea are of particular concern as they have been declining across their range, primarily due to bycatch in coastal gillnets, and are now classified as "Endangered" on the IUCN Red List of Threatened Species (Braulik et al. 2015;Sharpe and Berggren 2019;Temple et al. 2018). While there are no data available on the trophic interactions of these two coastal dolphins in Malagasy waters, stomach content analyses in specimens from the western Indian Ocean suggests that they both feed on fish and cephalopod species in inshore waters (Karczmarski et al. 2000;Lane et al. 2014). For both species, their foraging habitat often overlaps with fishing activities in Madagascar as fishers often target inshore areas with beach seines, hand lines, and gillnets (see Gough et al. 2020). Additionally, several candidate prey for these dolphins (specifically Hemiramphus spp., Lethrinus spp., Scomberoides spp. Lutjanus spp. Carangoides spp.) were previously reported as main catch for small-scale fisheries in Madagascar (Le Manach et al. 2011). This may lead to direct (e.g., bycatch; Cerchio et al. 2015) and indirect impacts of fishing of the primary prey of coastal dolphins, which in turn, can affect the feeding success of these predators. To evaluate the potential indirect effects of artisanal fisheries on coastal dolphins in north-western Madagascar, obtaining dietary information is crucial.

Here, we used bulk carbon and nitrogen stable isotope analysis in skin tissues of coastal dolphins, T. aduncus and S. plumbea, and four other odontocetes found in the waters surrounding Nosy Be, Madagascar, to document their trophic interactions. We used niche metrics to understand the segregation/overlap of their trophic niches and modeled the prey preferences of T. aduncus and S. plumbea using Bayesian stable isotope mixing models, particularly to investigate the relative contribution of multiple potential prey taxa to their diet. Only the diet of T. aduncus and S. plumbea was modeled as they are of particular importance due to their potential overlap with extensive artisanal fisheries in this area (Gough et al. 2020;Razafindrakoto et al. 2009). We predicted that inshore reef prey species would make up the majority of the diet of S. plumbea given their coastal distribution, whereas T. aduncus would have a broader dietary niche, including both inshore reef and pelagic prey species (Cerchio et al. 2015;Kiszka, Oremus, et al. 2010;Kiszka, Simon-Bouhet, et al. 2010;Wang and Yang 2009). The broad aim of this study is to improve our understanding of the trophic ecology of the most common odontocete species in north-western Madagascar and to establish an important baseline for further investigations of how coastal fisheries and threatened coastal dolphin species (S. plumbea and T. aduncus) might overlap in this region.

| Methods

| Sample Collection

Biopsy samples were collected during boat-based surveys in the waters surrounding the Nosy Be, Nosy Iranja, and Nosy Mitsio archipelagos, Madagascar (approximately centered at 13.32° S, 48.26° E) each year from 2008 to 2014 (Figure 1). The sampling period varied among years but generally was during spring months (September through November) except in 2011, in which sampling was conducted during winter months (July and August) (Table 1). Spatial effort also varied among years, with 2008 to 2010 covering primarily the nearshore coastal waters around Nosy Be, Nosy Komba, and the mainland of Madagascar (Figure 1). In 2011, sampling was extended to the shelf waters of the Nosy Mitsio archipelago, and from 2012 to 2014, sampling primarily occurred in the deep-water slope habitat and corresponding shelf waters off Nosy Iranja and Nosy Be, with less effort in nearshore coastal habitat. Consequently, cetaceans that were sampled varied across years, with 2008-2011 focused exclusively on coastal dolphins, and 2012-2014 focused more on deep-water and oceanic species (Table 1). Greater detail on effort can be found in Cerchio et al. (2015). Only adults were sampled and were distinguished based on body length. In total, six different species of odontocete were sampled: Sousa plumbea, Tursiops aduncus, Stenella longirostris, Stenella attenuata, Globicephala macrorhynchus, and Physeter macrocephalus (Table 1). When cetaceans were encountered, small dolphins (i.e., S. plumbea, T. aduncus, S. longirostris, and S. attenuata) were sampled using a compressed air rifle with adjustable pressure valve (DanInject JM Special 25 with a 13 mm barrel, 0-25 bar pressure) and custom-made darts equipped with 20 × 5 mm biopsy tips, similar to previous research (e.g., Kiszka, Oremus, et al. 2010;Kiszka, Simon-Bouhet, et al. 2010;Noren and Mocklin 2012;Sinclair et al. 2015). G. macrorhynchus were sampled with a 150 lb crossbow (Barnett Wildcat) and custom-made bolts equipped with 25 × 8 mm biopsy tips, and P. macrocephalus were sampled by collecting sloughed skin at the surface of the ocean from encountered animals. Dolphins were biopsied below the dorsal fin when sufficiently close (approximately 2-10 m) to the research boat. Biopsy sampling was conducted under scientific permits issued to SC and/or NA each year by the Madagascar Ministère de la Pêche et des Ressources Halieutiques. Cetacean skin was kept frozen at -20°C until processed.

Samples of potential prey items of coastal dolphins (i.e., S. plumbea and T. aduncus) were obtained from local fish markets in Nosy Be in June-July 2019 in the villages of Hellville and Ambatoloaka (Figure 1). Fishers declared that fish and cephalopods sold were all collected in the coastal waters of Nosy Be and neighboring islands. Prey were chosen based on previous stomach content research into the feeding ecology of these two dolphin species (e.g., Amir et al. 2005;Baldwin et al. 2004;Barros and Cockcroft 2014;Browning et al. 2014;Kaiser 2012;Karczmarski et al. 2000;Ross 1984;Sekiguchi et al. 1992), particularly from other regions in the western Indian Ocean (South Africa, the Arabian Sea, and Tanzania). As fisheries in this region target a wide range of demersal and pelagic species in reef, seagrass, and other coastal marine habitats, we assumed that the prey available at fish markets is representative of what is available in the ecosystem, and we included potential prey from the expected size range for both Indian Ocean humpback and Indo-Pacific bottlenose dolphins. Muscle samples from fish were removed from the dorsal area and tissue from cephalopods was taken from the mantle and put in ethanol before being frozen at -20°C until processed. Ethanol preservation may impact fish muscle isotope values, specifically depleting carbon-13; however, species-specific correction factors would be needed which are not available for the prey used here (Kelly et al. 2006). Given that prey used in mixing models are averaged across species, including the standard deviation, the interpretation of the results is likely not influenced (Kelly et al. 2006;Arrington and Winemiller 2002;Olin et al. 2014).

As prey and consumers were sampled at different times, the potential impact of the Suess effect was considered. The increase of atmospheric CO 2 since the industrial revolution has caused a decrease in atmospheric δ 13 C (as fossil fuels are depleted in 13C), which has in turn decreased the δ 13 C of oceanic dissolved organic carbon (DIC) (Friedli et al. 1986;Keeling et al. 2013), known as the Suess effect. Based on the annual average decrease in δ 13 C in the Indian Ocean at the latitude of our study (Sonnerup et al. 2000), the δ 13 C for each dolphin sample was adjusted by -0.017/year to match 2019 levels, that is, when prey samples were collected.

| Stable Isotope Analysis

Stable isotope analysis was completed at the Center for Aquatic Chemistry and Environment in the Institute of Environment (Florida International University, North Miami, FL). Muscle and skin samples were dried, homogenized into a fine powder, and lipid-extracted prior to analysis to allow us to compare samples, as lipids are 13 C depleted in a non-uniform way (DeNiro and Epstein 1978). Lipids were extracted by agitating muscle and skin tissues in a 2:1 chloroform: methanol mixture for 1 min with a solvent volume 5-times greater than the sample, after which the samples were left at room temperature for 1 h, centrifuged, and the supernatant was removed. This process may also exclude the effect of ethanol (Kiszka et al. 2014). After repeating this procedure two more times, each sample was rinsed in deionized water, dried, and 0.4-0.5 mg of sample added to a 4 × 6 mm tin capsule for stable isotope analysis using a ThermoFinnigan Delta V isotope ratio mass spectrometer (IRMS) coupled with a NA 1500 Ne elemental analyzer. Analytical reproducibility was based on replicates of internal standards including bovine liver (NBS standard reference material) and glycine (Alfa Aesar); variation among standards was 0.07‰ and 0.08‰ for δ 13 C and δ 15 N, respectively. The mean C:N values from analyzed tissues were less than 4, indicating adequate lipid extraction (Lesage et al. 2010). Isotopic ratios (R) are reported in the standard delta (δ) notation relative to the international standards of Vienna Pee Dee belemnite (δ 13 C) and atmospheric nitrogen (δ 15 N) using the following equation: where X is 13 C or 15 N and R is the isotope ratio 13 C/ 12 C or 15 N/ 14 N (Peterson and Fry 1987).

| Data Analysis

Data were tested for normality using inspection of Q-Q plots and histograms, and Shapiro-Wilks tests, and for homogeneity of variance using Levene's test. Reciprocal transformation of δ 15 N

TABLE 1 | Number of samples and period of sampling across years for odontocete species in north-western Madagascar.

2008 2009 2010 2011 2012 2013 2014 Total First sample 8-September 3-September 16-October 10-July 8-November 3-November 25-October Last sample 16-September 14-September 19-October 13-August 14-December 13-November 10-November Sousa plumbea (SP) 3 4 1 8 6 3 25 Tursiops aduncus (TA) 2 4 1 1 2 3 1 14 Stenella attenuata (SA) 4 2 6 Stenella longirostris (SL) 13 7 9 29 Globicephala macrorhynchus (GM) 3 2 5 Physeter macrocephalus (PM) 7 7 Total 5 8 2 9 28 22 12

values was performed where necessary to satisfy the normality assumption. Differences among species were explored for δ 13 C and δ 15 N using a MANOVA. Tukey's HSD post hoc tests were performed to investigate pairwise comparisons. Temporal variation (i.e., the effect of sampling year and month) in isotope values was investigated for species with N > 6 (i.e., S. plumbea, T. aduncus, and S. longirostris) using ANOVAs. Statistical analyses were performed in R (v. 4.0.0). All values are presented as mean ± SD.

Isotopic niche width was then calculated for each species using SIBER (Stable Isotope Bayesian Ellipses in R) metrics in R (v. 4.0.0; Jackson et al. 2011;R Core Team 2017). Bayesian ellipses were drawn around a set proportion of the bivariate data (40% in this case) using 10 5 iterations, burn-in rate of 1000, and thinned by 10 (Jackson et al. 2011), and Bayesian standard ellipse area (SEA B ; units = ‰ 2 ) was calculated and corrected for small sample sizes by using a posteriori randomly replicated sequences (SEA C ), applying a two-dimensional correction (Jackson et al. 2011). The overlap in these 40% SEA B was calculated between each set of species separately using the Bayesian overlap function, where 100% indicates completely overlapping ellipses and 0% indicates entirely distinct niches (Jackson et al. 2011).

For coastal dolphins, (i.e., S. plumbea and T. aduncus), Bayesian mass-balance stable isotope mixing models were built using the "MixSIAR" package for R (Stock et al. 2018) to estimate the relative contribution of potential prey to the diet of coastal dolphins. These models use a TEF to account for enrichment in each isotope between prey and predator. The TEFs used here were calculated using feeding experiments on common bottlenose dolphin. These were based on dolphin skin samples and took place over 350 days, ensuring isotopes stabilized in the tissues before collecting samples for the TEF calculation (Giménez et al. 2016). These have been used in previous research on small cetaceans (e.g., Caputo et al. 2021), as taxonomically close species are known to have similar TEF values (Giménez et al. 2016). The TEFs used here were 1.01‰ ± 0.37‰ for δ 13 C and 1.57‰ ± 0.52‰ for δ 15 N. The appropriateness of prey groups for coastal dolphin diet and the TEFs used here was evaluated by determining the likelihood that prey groups were included in a simulated dolphin mixing polygon, as per Smith et al. (2013). Mean (±SD) isotope values from muscle were calculated from potential prey species in the Nosy Be region (Table 2). The prey were grouped into five categories using k-means cluster analysis in R based on similarities of their δ 15 N and δ 13 C values, trophic level, and habitat (i.e., reef planktivores, inner reef mesopredators, outer reef mesopredators, pelagic planktivores, and pelagic mesopredators; Table 2). The optimal number of clusters was determined using the "Elbow Method," which plots the withincluster sum of squares, a distanced based metric which measures the sum of squares distance from each point to its assigned center, against the number of clusters creates clusters based on minimizing within-cluster variance (see Yuan and Yang 2019). Trophic level, habitat, and diet for each prey item were taken from FishBase (Froese and Pauly 2024). Mixing models were formulated with three Markov chain Monte Carlo chains of 300,000 draws and a burn-in of 200,000 draws, that were considered to have converged when they passed the Gelmen-Rubin and Geweke Diagnostics (Gelmin-Rubin = all variables < 1.01; Geweke = < 5% variables outside of ±1.96 in each chain; following Stock and Semmens 2016).

| Results

In total, 86 adult odontocete samples from six different species were collected and processed for δ 15 N and δ 13 C (Table 3). Both δ 15 N (df = 5, F = 17.4, p < 0.0001) and δ 13 C (df = 5, F = 41.6, p < 0.0001) differed among species. P. macrocephalus had the highest δ 15 N values compared to all other species, whereas S. plumbea had the lowest δ 15 N values compared to all other species, except T. aduncus and S. longirostris (Table 3, Figure 2). S. plumbea had significantly higher δ 13 C values compared to all other species, whereas S. longirostris had lower δ 13 C values compared to all other species, except P. macrocephalus and S. attenuata (Table 3, Figure 2). We found no effect of sampling year (S. plumbea: δ 13 C: df = 5, F = 1.67, p = 0.20, δ 15 N: df = 5, F = 1.27, p = 0.32; T. aduncus: δ 13 C: df = 6, F = 0.35, p = 0.58, δ 15 N: df = 6, F = 4.1, p = 0.1) or month (S. plumbea: δ 13 C: df = 2, F = 1.36, p = 0.28, δ 15 N: df = 2, F = 1.17, p = 0.34; T. aduncus: δ 13 C: df = 4, F = 1.22, p = 0.38, δ 15 N: df = 4, F = 0.52, p = 0.72) on δ 13 C or δ 15 N for both coastal dolphin species. The δ 15 N values for S. longirostris were significantly lower in 2012 than in 2013 or 2014, but actual values were < 1‰ on average and there was unequal sampling bias.

| Stable Isotope Ellipses and Niche Width

The isotopic niche of T. aduncus was the largest (Table 4) compared to all studied species (Table 4), primarily overlapping with G. macrorhynchus (22.0%; Table 4). P. macrocephalus had a significantly smaller ellipse area (SEA B mode = 0.23) than all other species (Table 4) that segregated from all other species (Table 4). The overlap area among the niche of S. plumbea and T. aduncus was 2.9% (Table 4), and the isotopic niche of S. plumbea was segregated from all other species.

| Bayesian Mixing Models

Based on mixing polygons, one S. plumbea and three T. aduncus fell outside the 95% contour and were removed from the mixing models (see Figure 3 and Figure 4). Reef planktivores made up the largest proportion of S. plumbea's diet (38.9% ± 8.7%), followed by inner reef mesopredators (20.5% ± 10.7%) (Figure 5). Outer reef mesopredators (16.9% ± 11.2%) and pelagic planktivores (15.3% ± 9.1%) also appeared to be marginally important prey items for S. plumbea, whereas pelagic mesopredators (8.3% ± 6.2%) were less important. A wider variety of prey contributed more equally to the diet of T. aduncus, including outer reef mesopredators (31.1% ± 25.1%), pelagic planktivores (22.8% ± 19.7%), inner reef mesopredators (17.8% ± 13.6%), reef planktivores (15.7% ± 9.8%), and pelagic mesopredators (12.6% ± 13.2%).

| Discussion

Assessing the foraging ecology and niche dynamics of marine predators is critical to understanding their trophic role and better predicting how community changes (predators and/or prey abundance) will affect predator-prey relationships within ecosystems. This is the first study to investigate resource use and partitioning in cetaceans in Madagascar and to investigate the ecological role of coastal dolphins within this odontocete community. Analysis of the trophic niche based on isotope values in dolphin skin provides clear evidence that distinguishes the trophic niches of coastal dolphins, S. plumbea and T. aduncus, from each other and the odontocete community. These results are consistent with other dietary studies elsewhere in the western Indian Ocean, which confirm that both coastal dolphins forage closer inshore on coastal prey (Gross et al. 2009;Browning et al. 2014;Karczmarski et al. 2000;Stensland et al. 2006;Amir et al. 2005).

Despite the importance of information that can be gained from stable isotope analysis, limitations need to be considered when interpreting results. Firstly, the difference in sample size between the coastal dolphin species and limited samples (n < 6) from some of the odontocete species may affect the results obtained. It is also important to note that the sex of animals is unknown. The standard ellipses corrected for small sample size were used here to improve comparability between species as sample size differs (Jackson et al. 2011). Secondly, stable isotope mixing models used to investigate the diet of coastal dolphins are highly sensitive to the potential prey put into the models and the TEFs used. Models ideally should be constructed using a priori knowledge from stomach contents or previous knowledge where possible, and the models assume that all important prey species are included in the prey sampling (Smith et al. 2013); however, such data were not available here to confirm our selection of prey species and thus we could not be certain that all assumptions have been satisfied. Therefore, it is possible that important prey items for dolphins in our study area are missing from our model, which is particularly problematic for generalist consumers, as a larger variety of prey items may contribute to their diet. Generalist consumers present an additional challenge when interpreting mixing model outputs, as many prey species may contribute to the diet leading to high variance. Similarly, for specialist consumers, if their main prey item is missing from the mixing model, the results would not be representative of their diet. To reduce these potential limitations, we used the best available knowledge on the diet of these species in the western Indian Ocean region and referred to diet as groupings or guilds rather than identifying specific species of prey preferred by each dolphin consumer. In addition, no TEF was available for either species, but the TEF used here was calculated from diets of T. truncatus and similar species should have similar TEFs, making it appropriate for our models (Giménez et al. 2016). Ethanol preservation can also affect 𝛿 15 N results (Kiszka et al. 2014), and both consumer and prey samples were preserved using ethanol and should be considered a caveat. Despite these caveats, we believe this study to be of value as a first assessment of these trophic relationships. This is particularly true given that the results complement what is known about the diet and habitat use of both coastal dolphin species, giving us confidence in our prey selection and final mixing model.

The isotopic niche of S. plumbea and, to some extent, T. aduncus did not overlap with oceanic species (i.e., P. macrocephalus, G. macrorhynchus, S. longirostris, and S. attenuata), consistent with what is known about the feeding ecology of either species groups. For instance, P. macrocephalus had the highest relative trophic level of all studied odontocetes, with no overlap in isotopic niche with the other species. Globally, P. macrocephalus feeds on high trophic level meso-and bathypelagic cephalopods (e.g., Barros 2003;Jaquet et al. 2007;Judkins et al. 2015;Kawakami 1980;Pauly et al. 1998;Rendell et al. 2004), particularly in tropical and subtropical regions. G. macrorhynchus are also deep-diving species that forage on mesopelagic cephalopods (Hacker 1986;Hernández-García and Martin 1994); however, their isotopic niche did not overlap with P. macrocephalus. G. macrorhynchus had a broader niche area, including highly variable δ 15 N values, ranging from 11.9‰-14.6‰. This suggests that they forage on a broad range of prey species and/or on a wide range of size classes of the same species. Their δ 13 C values appear elevated for an oceanic species, overlapping mostly with coastal T. aduncus. However, this is potentially due to depth gradients in δ 13 C values (Hobson 1999), where deep-water prey species often exhibit enriched δ 13 C values due to detrital carbon sources on the ocean floor (Kiszka, Oremus, et al. 2010;Kiszka, Simon-Bouhet, et al. 2010). Other studies on the trophic ecology of G. macrorhynchus also suggest that these animals forage in deep FIGURE 4 | Mixing polygon for Sousa plumbea and Tursiops aduncus predators (black dots) and their prey (white x), with a TEF from (Giménez et al. 2016) (δ 15 N = 1.57‰ ± 0.52‰, δ 13 C = 1.01‰ ± 0.37‰). Individuals falling outside the mixing polygon (indicated in red) were removed from mixing models. The scale bar is the probability that prey fall within the consumer diet, with contours at the 5% level (outermost contour) and at every 10% level.

waters on a variety of cephalopods (Hacker 1986;Hernández-García and Martin 1994;Kiszka, Oremus, et al. 2010;Kiszka, Simon-Bouhet, et al. 2010;Monteiro et al. 2017).

The isotopic niche of both Stenella species was depleted in δ 13 C values and largely segregated from coastal dolphins (< 11% overlap). While these species usually use coastal or reef-associated habitats to rest and socialize (Tyne et al. 2015;Kiszka, Simon-Bouhet, et al. 2011), isotope niche data presented here suggest they forage on oceanic prey. S. attenuata and S. longirostris are known to co-occur in much of their global range, often forming mixed species groups (Gross et al. 2009;Kiszka, Perrin, et al. 2011;Norris et al. 1994;Cerchio et al. 2022). However, these mixed-species groups are thought to form to reduce predation risk and do not seem to provide foraging benefits (Perrin 2009;Kiszka, Perrin, et al. 2011). Previous research in tropical ecosystems suggests that S. attenuata forages in epipelagic waters closer to shore whereas S. longirostris forages on offshore pelagic fish and cephalopods (Perrin et al. 1973;Dolar et al. 2003;Norris et al. 1994), which is consistent with the isotopic niches found here. S. longirostris was the most depleted of all species in terms of δ 13 C values, indicating that they feed farther from shore.

Here, the limited (6.4%) overlap in their dietary niche demonstrates that these two Stenella species forage in distinct foraging habitats and on different prey.

Despite their co-occurrence in coastal waters of this region, our results demonstrate that coastal dolphins (S. plumbea and T. aduncus) clearly partition resources and foraging habitats in this area, and may play distinct roles in the coastal food webs. S. plumbea had significantly enriched δ 13 C values, foraging further inshore than T. aduncus, which is consistent with studies on their spatial distribution in Madagascar (Cerchio et al. 2015) and at other coastal locations in the southwest Indian Ocean (Karczmarski et al. 2000;Stensland et al. 2006). T. aduncus had the largest isotopic niche areas, indicating that they feed across a wider variety of habitats and prey than the other odontocetes, including S. plumbea.

Bayesian mass-balance mixing models provided valuable information on the diet of each coastal dolphin species. While S. plumbea had a more specialized diet, with reef-associated planktivores contributing mostly to their diet, T. aduncus had similar contributions (> 12%) from all prey groups to their diet, with a large amount of variation (i.e., high standard deviation) indicating more individual variation than for S. plumbea. Importantly, reef planktivores contributed the most to the diet of S. plumbea (38.9%) and some of the diet of T. aduncus (15.7%

). Halfbeak (Hemiramphus far) is a common reef planktivore in these waters and is among the top six most important taxa caught from 1950 to 2008 by artisanal fisheries in Madagascar (when catches were last recorded), consisting of 20,991 t of fish in 2008 (Le Manach et al. 2011). Artisanal fishing catches were estimated at 93,000 t a year in 2008 and have likely increased even further due to human population increase and the increase in protein demand in Madagascar (Le Manach et al. 2011; Barnes-Mauthe et al. 2015). Artisanal fishing is socioeconomically important in Madagascar. In communities off western Madagascar (including Nosy Be), 82% of households depend on fishing for their income and 99% of food protein comes from fish (Barnes-Mauthe et al. 2015; Le Manach et al. 2012). The continued increase in fishing effort has led to evidence of overfishing in coastal waters, including a higher fishing-related mortality rate than the natural rate, a large proportion of non-mature fish being caught, and a low proportion of large individuals (Le Manach et al. 2011). Interactions with artisanal fisheries, through overfishing and/or bycatch, are a main threat to S. plumbea populations throughout their range (Temple et al. 2018; Kiszka et al. 2017; Cerchio et al. 2009; Cerchio et al. 2015;Braulik et al. 2015;Sharpe and Berggren 2019), but also to T. aduncus (Amir et al. 2005;Cerchio et al. 2015, Braulik et al. 2015;Christiansen et al. 2010;Kiszka et al. 2017). Coastal dolphins and fishing activities co-occur in north-western Madagascar, and throughout most of the range of both species.

Although the magnitude of bycatch in coastal fishing gear is unknown in north-western Madagascar, incidental captures in the region have been documented both through direct observation and reported in fisher interview surveys, and therefore coastal dolphins are likely to be impacted (Cerchio et al. 2015). In addition, competition with fisheries in reef-associated and other coastal habitats, where there is a proven decline of fish stocks in the region (Gough et al. 2020;Le Manach et al. 2011), may also have an impact on coastal dolphin populations, at least to some extent. It becomes critical to further investigate competitive interactions between coastal dolphins and artisanal/subsistence fisheries in north-western Madagascar and throughout the western Indian Ocean region, where humpback dolphins are still declining (Braulik et al. 2015).