Introduction

The tropical epipelagic zone is the surface layer of oceans (0–200 m) within 30° of the equator (Parin 1968). Despite the exceptional magnitude of the tropical epipelagic zone, which covers roughly 37% of the open ocean (Longhurst and Pauly 1987), overall species diversity is lower than in other similar-sized marine areas (Angel 1993). The large geographic ranges of epipelagic fishes, remoteness of regions they occupy, and high mobility of their movements have resulted in a major gap in our basic understanding of epipelagic habitats. What determines the habitats of tropical epipelagic fishes, and how are these habitats best defined?

The eastern tropical Pacific Ocean (ETP) encompasses the waters of Central America from southern California (30°N) to Peru (20°S) and extends to approximately 150°W. This vast area supports some of the largest fisheries in the world (e.g., Engraulis ringens, Dosidicus gigas, Thunnus albacares), and has a distinct assemblage of marine mammals and seabirds susceptible to commercial fishing (Ballance et al. 2006). Long-term surveys and remotely sensed data have aided the identification of oceanographic partitions within the ETP, based largely on patterns of primary productivity (Pennington et al. 2006). These oceanographic partitions are associated with the ecology of various organisms [e.g., zooplankton (Fernández-Álamo and Färber-Lorda 2006), seabirds, and cetaceans (Ballance et al. 2006)]. However, the basic biology and ecology of other taxa in the ETP (e.g., epipelagic fishes) must be characterized to facilitate interpretation of mechanisms that influence species diversity in the region.

Flyingfishes are a particularly suitable model for investigating how oceanographic features may affect epipelagic fish habitats, because they are restricted to surface waters (Nesterov and Bazanov 1986) and do not make long-distance migrations (Parin 1968). Passive transport of early-life-stage individuals (Parin 1968), short generations [typically 1–2 years; (Steven et al. 1993)], and small home ranges [e.g., <500 km2 for Hirundichthys affinis; (Oxenford 1994)] indicate low migratory capabilities for flyingfishes. Remotely sensed oceanographic measurements can be directly associated with the surface layer habitats of flyingfishes. As a central component of epipelagic food webs throughout tropical and subtropical regions of the world oceans (Parin 1968; Rau et al. 1983), flyingfishes are important for advancing our overall understanding of mid-trophic marine ecology.

More than 50 species of exocoetids are currently recognized [22 species within the ETP; (Parin 1995)], making it one of the most species-rich lineages of epipelagic fishes. Morphological similarities among species and fast aerial gliding movements of individuals make flyingfishes difficult to identify in situ during observational surveys (Parin 1981, 1983; Oxenford et al. 1995; Churnside et al. 2009). Moreover, minor morphological differences between some species mean that increased scrutiny is required for accurate identification (Parin 1996, 1999; Parin and Belyanina 1998, 2002a, b; Parin and Shakhovskoy 2000).

In this study, we applied a combination of spatial analysis methods to a remarkably large dataset (11,125 identified and georeferenced flyingfish specimens collected throughout the ETP over a 21-year period) to characterize flyingfish habitats in the ETP. We first summarized the relative abundance of each species, and subsequently analyzed each of the 12 most abundant species using predictive distribution models based on 11 independent oceanographic variables. The overall species richness of flyingfishes in the ETP was also considered. Lastly, we estimated and plotted species diversity within our study area. This study provides a broad-scale description of flyingfish habitats, including representatives of every known genus and nearly half of known species.

Materials and methods

Study area, sampling, and specimen handling

Our study area comprised 20 million km2 of ocean within the ETP, extending from southern California to Peru, and offshore to roughly 150°W (Fig. 1a). From 1986 to 2007, the National Oceanic and Atmospheric Administration (NOAA) Southwest Fisheries Science Center (La Jolla, CA, USA) conducted 32 scientific research cruises in the ETP (Appendix). Flyingfishes were collected during these cruises, providing an unprecedented opportunity to study species distributions, relative abundance, and habitat preferences. Cruises were conducted between July and December (<1% in July; 20.4% in August; 27.4% in September; 28.4% in October; 22% in November; 1.7% in December). Specimens of all sizes (adults, juveniles, and larvae) were collected at 2491 sampling stations (Fig. 1a) along tracklines designed to systematically survey dolphin populations in the ETP. Flyingfish sampling stations began 1 h after sunset and usually lasted for 1 h. Adults were most common, and eggs were not sampled. To aid in specimen detection, one or two deck-mounted spotlights were used to illuminate 50 m2 of water next to the ship. Samples were collected by two researchers, each using a long-handled (8 m) dipnet with a basket 1 m in diameter and 1.5 m deep; net mesh size was <0.5 mm. Locality information (latitude and longitude) was recorded for each sampling locality using the shipboard navigation system.

Fig. 1
figure 1

a Flyingfish specimen collection localities from research cruises in the eastern tropical Pacific study area from 1986 to 2007. Black dots represent dipnet stations, and the gray polygon is an outline of the study area. b Map of the 12 oceanographic partitions within the eastern tropical Pacific study area drawn using ARCGIS, version 9.3.1 (ESRI, Redlands, CA, USA) and the descriptions from Pennington et al. 2006. Six oceanic partitions (NEC North Equatorial Current, CCTR Countercurrent Thermocline Ridge, NECC North Equatorial Countercurrent, EU Equatorial Upwelling, SEC South Equatorial Current, SPSG South Pacific Subtropical Gyre) and six coastal partitions (CC California Current, TEH Gulf of Tehuantepec, PAP Gulf of Papagayo, PAN Gulf of Panama, GAL Galapagos, PC Peru Current) were included

Full size image

After capture, all specimens were sacrificed using an ice bath and were subsequently frozen in seawater. At the end of each research cruise, specimens were thawed, fixed in 4% formalin, rinsed, and preserved in alcohol (70% isopropyl or ethyl). Specimens were identified to species using published species descriptions and regional field guides (both for the ETP and west-central Pacific) (Parin 1961, 1995, 1996, 1999; Parin and Belyanina 1998, 2002a, b; Parin and Shakhovskoy 2000; Shakhovskoy 2007; Shakhovskoy and Parin 2010). All specimens were catalogued at the Scripps Institution of Oceanography (USA), Los Angeles County Museum (USA), or Royal Ontario Museum (Canada).

Predictive species distribution models

MAXENT v.3.3.3 (Phillips et al. 2006; Phillips and Dudík 2008) is a widely used software program for producing predictive species distribution models. As inputs into MAXENT, we used two sets of data: (i) 11,125 identified specimens with cross-referenced “presence only” locality data (latitude and longitude), and (ii) oceanographic data for 11 variables measured within the ETP during the years 1997–2010 and published in Bio-ORACLE (Tyberghein et al. 2012). To ensure the robustness of these analyses, we considered only the 12 most abundant species, each with a minimum of approximately 100 collected individuals (Table 1).

Table 1 Numbers of flyingfish genera and species collected within the ETP study area
Full size table

Predictor data for 11 environmental variables were compiled by Tyberghein et al. (2012) and made available in Bio-ORACLE, a comprehensive source of oceanographic data that covers the ETP to a resolution of 5 arc minutes. These 11 variables were prioritized to ensure oceanographic and biological independence after scrutinizing a total of 23 variables and retaining only the variables that produced unique response curves. Specifically, the following parameters were used in MAXENT: a random seed, a random test percentage of 50, a regularization parameter of 1, a maximum of 10,000 iterations, a convergence threshold of 0.00001, and a maximum of 10,000 background points, with no test sample or bias file (Phillips et al. 2006; Phillips and Dudík 2008; Elith et al. 2011). To further enhance the model of each species, 10 bootstrap replicates were performed and summarized as an average. Models were additionally evaluated using the area under the curve (AUC) of the receiver operating characteristics (ROC) plot and standard deviation (SD), as well as jackknife variables of greatest influence when used either in isolation or after omission. The influence of each oceanographic variable within each species model was calculated using the percent of relative importance (% contribution) within MAXENT.

Species richness and diversity

To assess species richness within the ETP, we generated a map following the regional partitioning scheme proposed by Pennington et al. (2006), based on oceanographic processes that regulate primary productivity. In concordance with this scheme, and to avoid overlapping partitions, we grouped the Costa Rica Dome and Gulf of Papagayo. Specifically, we analyzed six coastal partitions and six oceanic partitions (Fig. 1b). We projected specimen data onto a map of the ETP using ARCGIS v.9.3.1 (ESRI, Redlands, CA, USA), and recorded whether a species was collected from within each of the 12 oceanographic partitions. A minimum of two individuals were required to consider a species present in any partition. We ranked each partition by the total number of flyingfish species it contained.

Flyingfish species diversity in the ETP was mapped using the following method: the study area was divided into a 1 decimal degree × 1 decimal degree grid and Shannon indices of diversity were calculated within each cell using DIVA-GIS, v.5.2 (Hijmans et al. 2005). Shannon Indices were projected onto a map of the ETP to visualize patterns of flyingfish species diversity throughout the study area.

Results

Relative species abundance

All 11,125 individual specimens were examined and identified to flyingfish genus and species (Table 1). A total of 7 genera and 25 species were collected from the ETP study area. The two most common genera were Exocoetus (3 species; n = 5423; 48.75% of all flyingfishes caught) and Cheilopogon (12 species; n = 2448; 22.0%). Exocoetus monocirrhus (n = 3578; 32.16%) and Exocoetus volitans (n = 1826; 16.44%) were the most common species, totaling nearly half of all collected individuals (n = 5404; 48.59%).

Maximum entropy

MAXENT results indicated that the null hypothesis of random distribution with respect to the variables considered was consistently rejected in the models of all species (AUC values >0.969; SD ≤ 0.003; Table 2) (Tittensor et al. 2010a). Mean sea surface temperature (sstmean) provided the greatest contribution to models for 11 of 12 flyingfish species analyzed (Table 2), and ranged from 19.2 to 41.7% (average of 30% for all species). Along with phosphate, salinity was second most important on average (13%) and the single most important variable for H. marginatus (22.9%), although sstmean ranked second (19.2%) in importance for the model of H. marginatus. Phosphate contributed 13% on average to the models of all species, while all other variables each contributed < 10% to models of abundant ETP flyingfish species (Table 2).

Table 2 A summary of MAXENT analysis results for each of the most abundant species (12 species; n ≥ 96)
Full size table

Comparisons between flyingfish genera revealed variability in the types of oceanographic data that contributed most to predictive distribution models within each genus. For example, the models of both Exocoetus species showed high contribution values for dissolved oxygen, salinity, and phosphate, while other values were much lower (Table 2; Fig. 2a, b). In contrast, models for Fodiator identified sea surface temperature, chlorophyll a, and calcite as having a strong influence on their predicted distributions (Table 2; Fig. 2c). Parexocoetus results did not show high contributions from chlorophyll a or calcite, but sea surface temperature and diffuse attenuation were clearly important (Table 2; Fig. 2d).

Fig. 2
figure 2

Predicted species distribution models for six abundant flyingfish species collected in the ETP study area, based on MAXENT analyses: a Exocoetus monocirrhus, b Exocoetus volitans, c Prognichthys sealei, d Prognichthys tringa, e Fodiator rostratus, f Parexocoetus brachypterus. Shaded areas correspond to regions of highest probability of occurrence as determined by average consensus after 10 bootstrap replicates for each species (warmer colors = higher probability of occurrence)

Full size image

For some flyingfish genera, predictive distribution models were similar across species (e.g., Exocoetus spp.; Table 2; Fig. 2a, b), while others included species with contrasting patterns (e.g., Cheilopogon spp., Hirundichthys spp.). For example, variables for Prognichthys species were different: sstmean influenced the model of Pr. sealei (41.7%) much more than it did for Pr. tringa (19.7%; Table 2), while the predicted distribution of Pr. tringa was more restricted to coastal waters than the model of Pr. sealei (Fig. 2c, d). Fodiator rostratus, P. brachypterus, and Cy. callopterus had similar distributions along coastal regions (Figs. 2e, f and 3a). Similarly, striking differences were found among the three Cheilopogon species analyzed; in particular, predictive distributions were large and widespread for Ch. dorsomacula and Ch. xenopterus, but restricted to coastal regions for Ch. papilio (Fig. 3b–d; Table 2). Hirundichthys species had different variables of greatest contribution from MAXENT analysis (salinity for H. marginatus, and sstmean for H. speculiger; Table 2), but similar predictive distribution maps (Fig. 3e–f).

Fig. 3
figure 3

Predicted species distribution models for six abundant flyingfish species collected in the ETP study area, based on MAXENT analysis results a Cypselurus callopterus, b Cheilopogon papilio, c Cheilopogon dorsomacula, d Cheilopogon xenopterus, e Hirundichthys marginatus, f Hirundichthys speculiger. Shaded areas correspond to regions of highest probability of occurrence as determined by average consensus after 10 bootstrap replicates for each species (Warmer colors = highest probability of occurrence)

Full size image

Species richness

Species richness varied across oceanographic partitions within the ETP and was highest within the warmest partitions (Table 3) of the North Equatorial Current (n = 16 species), North Equatorial Countercurrent and Gulf of Papagayo (n = 14 species). The cold waters of the South Pacific Subtropical Gyre and Peru Current each had very low species richness (n = 1 species).

Table 3 Species richness counts within each of the 12 oceanographic partitions sampled in this study
Full size table

Species diversity

The plot of species diversity resulting from the DIVA-GIS analysis revealed the highest diversity north of the Equator (0°), from 130°W toward shore and north into the Gulf of California (Shannon indices > 1.5; Fig. 4). Lower species diversity was found south of the Equator and toward the offshore boundaries of the study area (Shannon indices < 0.5; Fig. 4).

Fig. 4
figure 4

A projected interpolation of results from species diversity estimates (measured using the Shannon index within DIVA-GIS v.5.2; Hijmans et al. 2005). Light pink colors represent cells with lowest species diversity, and dark pink colors represent cells with highest species diversity. The gray polygon outlines the eastern tropical Pacific study area

Full size image

Discussion

Flyingfishes are found throughout tropical and subtropical regions of the world oceans, yet their habitats remain poorly characterized at regional scales. Further, no studies have attempted to define broad-scale patterns of flyingfish abundance, species richness, and diversity. Here, we analyzed an exceptionally large dataset for flyingfishes, with over 11,000 specimens systematically collected throughout a vast region in the ETP over 21 years, and employed spatial analysis techniques to describe patterns of relative species abundance, richness, and diversity, while identifying oceanographic variables that most correlate with the distributions of abundant species. Our study demonstrates that mean sea surface temperature is the best descriptive variable for the predictive distribution models of most flyingfish species in the ETP (11 out of 12 species analyzed). However, we found potentially important differences between the habitat models of some commonly collected species.

Flyingfish species abundance

Exocoetus was by far the most abundant and widely distributed genus in the ETP, followed by Cheilopogon. These two genera were disproportionately well-represented in our dataset. This pattern in Exocoetus has been described previously, as several authors considered them “dominant middle-sized nekton” in the Atlantic (Nesterov and Grudtsev 1981; Grudtsev et al. 1987), and worldwide (Parin and Shakhovskoy 2000). The occurrence of few abundant and many rare species within a defined area is a generalizable trend known to ecologists (the species abundance distribution; SAD), and remains consistent when applied to a variety of ecosystems and spatial scales (McGill et al. 2007). For adult flyingfishes in the eastern Caribbean, Oxenford et al. (1995) demonstrated high relative abundance of two species (52% Pa. hillianus and 47% H. affinis) and low abundance of Cy. cyanopterus (∼1%). In the much larger ETP study area investigated here, flyingfishes follow the SAD pattern, with few abundant species and numerous rare species. The exact mechanisms that determine SAD are not universal, but some correlations with this pattern have been observed, and may be relevant to ETP flyingfishes. For example, some authors have suggested that abundant species may have smaller body sizes (White et al. 2007), which is the case for Exocoetus [among the smallest of flyingfish species: max SL = 207 mm; (Parin and Shakhovskoy 2000)]. Nevertheless, other authors have noted that evolutionary history may be related to the abundance of species (Cadotte et al. 2010), and interestingly, the disproportionally abundant Exocoetus is one of the earliest diverging flyingfish genera (Lewallen et al. 2011).

Predictive distribution models for ETP flyingfishes

Mean sea surface temperature was most important for predictive distribution models in 11 of the 12 most abundant flyingfish species in the ETP. For marine fishes, water temperature can directly influence ontogenetic development (Watanabe and Yatsu 2004; Zwolinski et al. 2011), spawning (Weber and McClatchie 2010), and behavior (Block et al. 2001). Therefore, it is reasonable that this environmental characteristic may define fish habitats. For flyingfishes, gliding behavior may be energetically limited by water temperature, such that fast swimming speeds (10 m/s necessary for takeoff) are achieved only at temperatures above 20 °C (Davenport 1994). However, the exact relationship between temperature and gliding behavior has not been experimentally quantified. Our data suggest that changes to sea surface temperature, both short- and long-term, would have dramatic effects on the habitat boundaries and survival of many flyingfishes in the ETP.

Salinity had a high contribution (nearly 20%) in the models of four species (E. monocirrhus, E. volitans, H. marginatus, H. speculiger), suggesting that physiological constraints may limit the occurrence of some flyingfishes, although empirical evidence for such relationships are lacking. Also plausible are components of flyingfish ecology and/or water mass movements that may be directly correlated with the physical oceanography of the region. Indirect influences of flyingfish habitat need to be investigated in greater detail through experimental manipulations directed at determining physiological tolerance thresholds for each species. For example, phosphate was an important variable for the models of Ch. dorsomacula, Ch. xenopterus, and E. monocirrhus, but without necessary data on the potential effect of this nutrient on flyingfish species survival, this result is difficult to interpret.

Ph was notably lacking in contribution to ETP flyingfish species models (>2% on average), suggesting that ocean acidification may be a lesser threat to ETP flyingfish persistence than changes to sea surface temperature. Nevertheless, MAXENT analyses resulted in predictive models for the distribution of 12 abundant flyingfish species in the ETP that may correlate with pronounced biological diversity among flyingfishes, including egg types (e.g., buoyant, demersal, filamentous, smooth), larval morphology (e.g., single or multiple chin barbels, or no barbels), jaw morphology, spawning mode (e.g., aggregation behaviors), and two- vs. four-wing gliding strategies (Parin 1961; Collette et al. 1984). It remains possible that ph may indirectly alter flyingfish habitats over long time scales, for example by influencing their prey.

Flyingfish species richness in the ETP

Flyingfish species richness was highest in ETP partitions with relatively higher average sea surface temperatures. For example, the warm North Equatorial Current had the highest number of observed species, while only a single species was collected in each of the coldest oceanographic partitions (South Pacific Subtropical Gyre and Peru Current). In comparison to other oceanographic partitions within the ETP, the North Equatorial Current has low primary productivity and high sea surface temperature (Pennington et al. 2006). Conversely, the Peru Current is known as a primary productivity hotspot (Gutiérrez et al. 2011). Marine productivity hotspots are defined as regions of elevated chlorophyll and low sea surface temperatures that result from the upwelling of deep, nutrient-rich waters (Valavanis et al. 2004). These “productivity hotspots” are usually not correlated with increased species richness; on the contrary, they are dominated by few species with high abundance (Smith et al. 2011). Upwelling regions such as the California Current, Peru Current, and Benguela Current are known to support some of the largest single-species fisheries in the world (Smith et al. 2011). Our results for ETP flyingfishes are consistent with this trend, suggesting that primary productivity (chlorophyll a measurements) is not a good predictor of high flyingfish species diversity.

For marine organisms, a “kinetic energy” hypothesis was proposed to explain heterogeneous patterns of species richness at a global scale (Tittensor et al. 2010b). Because metabolic rates increase with temperature, it is expected that speciation rates, diversity, and thermal tolerance should lead to higher species richness in warmer regions (Rohde 1992; Currie et al. 2004; Clarke and Gaston 2006). Our results for ETP flyingfishes are consistent with this hypothesis. Other researchers have proposed that variance in sea surface temperature can be used to quantify the relative stability of a marine region and test the “climate stability hypothesis”, which predicts that stable climates have more species (Fraser and Currie 1996). Our results do not support this hypothesis, because dynamic (i.e., “unstable”) coastal partitions [e.g., Gulf of Papagayo and Gulf of Panama; (Pennington et al. 2006)] have relatively high flyingfish species richness. A lack of dissolved oxygen has further been proposed as a mechanism that limits species richness [termed the “stress hypothesis”; (Keeling et al. 2010)]. For ETP flyingfishes, species richness was lower in regions characterized by higher concentrations of dissolved oxygen (e.g., Peru Current), contradicting this hypothesis. Dissolved oxygen also provided moderate contributions to models of predicted species distributions (average values of approximately 8%; Table 2).

Flyingfish species diversity in the ETP

We found high species diversity nearshore in the ETP study area. Despite the relatively small sizes of the coastal partitions Gulf of Papagayo and Gulf of Panama, these had nearly as many observed species as the much larger North Equatorial Current and South Equatorial Current partitions (Table 3). This result was surprising, given that many flyingfishes are offshore specialists. However, the continental shelf is narrow in the ETP (Fiedler and Lavín 2006), meaning that offshore species are often found closer to shore; thus coastal partitions may capture the species diversity of both coastal and epipelagic communities. This hypothesis should be tested by comparing species diversity in regions with more extensive continental shelves, or by analyzing geographic partitions at finer scales.

Conclusion

This study revealed that Exocoetus, Cheilopogon, and Hirundichthys are the most abundant genera within the ETP. After modeling the predictive distributions of the 12 most abundant species (out of 25 total), we found differences in both inter- and intra-generic comparisons, which may reflect the unique biological characteristics of each species. Mean sea surface temperature was an important predictor of flyingfish habitats in the ETP, with high contributions to the models of nearly all abundant species. Species richness measures were higher in oceanographic features that are typically warmer, confirming that sea surface temperature is an important predictor of flyingfish abundance as well as distribution.