Translator Disclaimer
12 August 2016 Bycatch of Myliobatid Rays in the Central Mediterranean Sea: the Influence of Spatiotemporal, Environmental, and Operational Factors as Determined by Generalized Additive Modeling
Author Affiliations +
Abstract

Identification of the factors influencing the distribution of vulnerable species can be useful for predicting their occurrence at a local to regional scale and for identifying the most suitable measures of management and conservation. We used generalized additive models to assess the effects of spatiotemporal, environmental, and operational factors on the catches of two myliobatids: the Common Eagle Ray Myliobatis aquila and the Bull Ray Pteromylaeus bovinus. Fishing data were collected from commercial midwater trawlers operating in the north-central Adriatic Sea during 2006–2013. Presence/absence and abundance (CPUE) data were modeled separately, and eachmodel was then validated by using a test data set. The presence/absence and abundance of Common Eagle Rays and Bull Rays were mostly influenced by spatial (haul location) and temporal predictors. The major occurrences of Common Eagle Rays and Bull Rays were observed in the upper Adriatic Sea between late spring and early autumn. During winter, a southward shift in the catch was recorded for both species. In accordance with a significant effect of depth, Common Eagle Rays were more likely to be caught in hauls conducted between 10- and 60-m depths. The CPUEs of Common Eagle Rays and Bull Rays declined significantly with haul duration and net vertical opening. The validation procedure indicated that the predictive accuracy of the models was rather good. Giving new insight into the ecological requirements of Common Eagle Rays and Bull Rays, the results of this studymay contribute to the development of conservation strategies and can be used to direct future monitoring and research programs.

The Common Eagle Ray Myliobatis aquila and the Bull Ray Pteromylaeus bovinus are two benthopelagic elasmobranchs that occur in the eastern Atlantic from the southern North Sea and Ireland to Morocco and the Canary Islands, throughout the Mediterranean Sea, and along the coast of South Africa (Froese and Pauly 2015). Both species are primarily found in coastal and warm-temperate waters (down to 150–200-m depths), sometimes entering shallow lagoons and estuaries (El Kamel et al. 2009), but they also occur offshore (Serena 2005; Froese and Pauly 2015). Common Eagle Rays are usually recorded on sandy and muddy substrates, often in small groups swimming close to the bottom, where they feed almost exclusively on benthic invertebrates (mainly gastropods and bivalves; Capapé 1976; Jardas et al. 2004). In the Mediterranean Sea, the breeding period of the ovoviviparous Common Eagle Ray likely occurs between September and February (Notarbartolo di Sciara and Bianchi 1998). After a gestation period of 6–8 months, female Common Eagle Rays give birth to three to seven young (Serena 2005). Bull Rays are not confined to the bottom; rather, they are frequently seen close to the surface and in small groups, sometimes leaping from the water (Van der Elst 1988; Compagno et al. 1989). Bull Rays mainly feed on bottom-living crustaceans, gastropods, and bivalves, also occasionally consuming cephalopods and teleosts (Capapé 1977; Compagno et al. 1989). The Bull Ray is an ovoviviparous species that reproduces on a yearly cycle, with a gestation period of about 6 months and three to seven pups per litter (Compagno et al. 1989).

Like other large elasmobranchs, the Common Eagle Ray has suffered a dramatic decline in abundance and distributional range in recent decades, mainly due to overfishing and habitat degradation (Ferretti et al. 2005; Holtzhausen et al. 2009). Reported as a common species in the Bay of Biscay at the end of the last century, the Common Eagle Ray became very rare in the early 1990s (Quéro 1998). In the Gulf of Lion (Mediterranean Sea), a clear decreasing trend in Common Eagle Ray commercial landings from the demersal fishery during 1970–1995 indicated a rapid decline in their stocks (Aldebert 1997). There are currently no time series data with which to assess trends in other areas of the Mediterranean Sea, but population declines are likely to have occurred elsewhere; the International Union for Conservation of Nature (IUCN) designated the Common Eagle Ray as “near threatened” in the Mediterranean Sea (Holtzhausen et al. 2009; Malak et al. 2011).

For the Bull Ray, the IUCN global conservation assessment is noted as “data deficient” since the abundance status in many areas of the species' wide distributional range remains poorly known (Wintner 2006). In the Mediterranean Sea, the Bull Ray has been locally considered as potentially threatened due to its life history traits (a long-lived species with delayed maturity and low reproductive rates) and overfishing bycatch pressure (Zogaris and Dussling 2010). Vulnerability of Bull Rays to overfishing can be especially high in nearshore habitats, where this species frequently forages and aggregates in small schools (Seck et al. 2002; Serena 2005).

Despite a growing global concern about elasmobranch conservation and management, the lack of biological information and appropriate fisheries data sets for some vulnerable species at local and regional scales hinders assessment and conservation planning (Cavanagh and Gibson 2007). Valuable information on distribution and abundance patterns for Common Eagle Rays and Bull Raysmay be obtained by fishery-dependent surveys. In fact, several fishing gears and techniques (pelagic, midwater, and bottom trawls; trammel nets and gill nets; and bottom longlines) include Common Eagle Rays and Bull Rays as either target species or bycatch species (Serena 2005; Wintner 2006; Holtzhausen et al. 2009; Fortuna et al. 2010; Ferretti et al. 2013). Unfortunately, catch data from commercial fisheries for both species are frequently unavailable since these and other fish of low commercial value are usually discarded at sea by fishers (Mavrič et al. 2004; Capapè et al. 2008; Dulčić et al. 2008).

Since 2006, a project aimed at quantifying the accidental capture of large vertebrates (e.g., sharks, manta rays, sea turtles, and dolphins) in midwater trawl fisheries has been conducted in the north-central Adriatic Sea (Fortuna et al. 2010). Initial monitoring data on this fishing activity, which mainly targets anchovies, sardines, and mackerels, provided evidence for a substantial bycatch of Common Eagle Rays and Bull Rays. Based on those results, we attempted to identify the most important factors explaining the spatiotemporal pattern of Common Eagle Ray and Bull Ray catches in midwater trawl fisheries.

Relationships between fishing performance and environmental variables are frequently modeled by using linear regression techniques even though these relationships are often nonlinear (e.g., Bigelow et al. 1999). Linear regression techniques are straightforward in determining model parameters and in their interpretation, but they have a relatively restricted range of application and little flexibility (Chong and Wang 1997). To overcome these drawbacks, the generalized additive model (GAM) approach has been applied to modeling the catch rates of fishes as a function of spatial, temporal, and environmental variables (Bigelow et al. 1999;Walsh and Kleiber 2001; Damalas et al. 2007; Katsanevakis et al. 2009; Murase et al. 2009; Damalas and Megalofonou 2010; Maravelias et al. 2012; Drexler and Ainsworth 2013). The strength of GAMs, which are semiparametric extensions of generalized linear models, is their ability to deal with highly nonlinear and non-monotonic relationships between the response and a set of covariates (Hastie and Tibshirani 1986). In general, GAMs can be useful for (1) defining optimal conditions for a given species by using an array of environmental descriptors and (2) predicting a species' abundance or its likelihood of inhabiting a particular environment (Maravelias 1997; Stoner et al. 2001; Walsh and Kleiber 2001).

Identification of the factors that influence changes in the spatiotemporal distribution of fishes may have important implications for vulnerable species, such as the Common Eagle Ray and the Bull Ray, leading to the implementation of suitable management and conservation measures. In the present study, our main objective was to use a fishery-dependent data set to identify the most important environmental and operational factors explaining the spatiotemporal patterns in Common Eagle Ray and Bull Ray catches within the north-central Adriatic Sea. The underlying relationships between catch rates and the selected factors were analyzed via nonparametric GAMs. The predictive capabilities of the best-fitting models were then evaluated with a test data set (Valavanis et al. 2008).

METHODS

Study area and data collection.—Between 2006 and 2013, fishing data were collected from 47 midwater pair-trawling vessels that operated in the central and northern portions of the Adriatic Sea (Figure 1). Onboard observers monitored the trawlers for a total of 1,098 d. The nature of this fishery-dependent survey prevented the use of an experimental design with fixed sampling stations; nevertheless, some hauls were sporadically made at the same locations. Although the vessels worked throughout the year, the nets were only hauled out during daylight. Haul data, such as vessel and trawl net configuration, date, haul duration, vessel speed, and geographic coordinates, were recorded. From those data, we obtained spatial (haul location), temporal (month, season, and year), and operational (net vertical opening, haul duration, and vessel speed) descriptors.

FIGURE 1.

Map of the study area in the north-central Adriatic Sea, showing the spatial distribution of fishing effort and catches (CPUE, number of individuals per 106 m3) of (a) Common Eagle Rays and (b) Bull Rays during 2006–2012.

f01_382.jpg

Potentially relevant environmental factors were selected by considering some of the ecological traits of Common Eagle Rays and Bull Rays. Each of these rays is a thermophilic species that mainly occurs in tropical to warm-temperate coastal waters and sometimes enters brackish environments, such as semienclosed bays and lagoons (El Kamel et al. 2009; Zogaris and Dussling 2010). These habits prompted us to investigate the influence of four environmental variables: sea surface temperature, sea surface salinity, bottom depth, and primary production (chlorophyll concentration). Sea surface temperature, sea surface salinity, and chlorophyll concentration data for the study area over the study period were obtained from satellite-based and in situ observations developed over a mesh grid at a 0.0625° resolution from two European Union-funded projects, MyOcean andMyOcean2 (marine. copernicus.eu). Monthly average values were used for these variables.

Each haul was correlated to the sea surface salinity, sea surface temperature, chlorophyll concentration, and bathymetry based on the haul's exact date and coordinates. The CPUE, expressed as the number of fish caught divided by the swept volume (area of the trawl net opening multiplied by the distance towed), provided an estimate of fish abundance.

Statistical analysis and modeling.—The GAM techniques were used to evaluate the influences of a suite of spatiotemporal, environmental, and operational variables on the presence/ abundance of Common Eagle Rays and Bull Rays. The use of GAMs can be justified when there are suspected nonlinear relationships between multiple predictors and the response variable (Hastie and Tibshirani 1986). In GAM analyses, the expected value of the response variable Yi is related to the predictor Zmi according to the following general formulation:

e01_382.gif
where f is the link function, LP is the linear predictor, c is the intercept, sm is the one-dimensional smooth function of covariate Zm, and Zmi is the value of covariate m for the ith observation (Wood 2006).

The distribution of Common Eagle Ray and Bull Ray catches was skewed and included a large proportion of zero observations because these species were not specifically targeted by the midwater pair trawlers operating in the study area. Given the nature of the data, we applied the delta modeling procedure (Maunder and Punt 2004), which allowed us to model the probability of species presence and catch rates (CPUEs) separately. Models of presence/ absence data were obtained by using a binary response variable (coded 0 or 1) with a binomial error distribution and logit link. Abundance data were modeled by using CPUE as the response variable. Akaike's information criterion (AIC; Akaike 1973; Burnham and Anderson 2002) was used to select the best underlying error distribution (gamma, Gaussian, negative binomial, or Poisson) and link function. All models were fitted using thin-plate regression splines estimated by penalized iterative least squares, and the optimum degree of smoothing was defined by generalized cross validation (GCV; Wood 2006). To check for autocorrelation in the model residuals, Moran's I-statistic (Moran 1948) was computed by means of a specific tool in ArcGIS version 10.2 (ESRI, Redlands, California).

To partially address the spatial autocorrelation observed in the residuals of the presence/absence models, the effect of location was evaluated by using a bivariate smoothed predictor (i.e., the interaction between longitude and latitude; Table 1; Hollowed et al. 2012). Furthermore, α was set at 0.01 to compensate for potential overestimation of the significance level of the covariates.

Model fitting was accomplished by using the mgcv package (Wood 2006) in R version 3.0.2 (R Development Core Team 2013). Eleven predictor variables or covariates were considered for inclusion in the final models for each species: haul position (defined by latitude and longitude), year, season, month, bottom depth, sea surface temperature, sea surface salinity, chlorophyll concentration, net vertical opening, haul duration, and vessel speed. Applying a forward stepwise model building procedure, 1–11 covariates were added to each of the initial models (i.e., the null models, consisting only of the overall mean for each species). Removal of covariates followed the criteria proposed by Wood (2001). Finally, reduction in the GCV score was taken into account to select the “best” models (Wood 2006) for Common Eagle Rays and Bull Rays.

The significance of each term in the final models was assessed by use of an ANOVA F-ratio test. Prior to running the analyses, catch data were split into training and test sets: data collected between 2006 and 2012 were used for training, whereas those collected in 2013 were used for model validation. Model performance was evaluated according to the procedure described by La Mesa et al. (2015). The predictive accuracy of the presence/absence models was evaluated with a thresholdindependent receiver operating characteristic (ROC) plot (Fielding and Bell 1997; Guisan and Zimmermann 2000) and by estimating the area under the ROC curve (AUC) via the PresenceAbsence package in R (Freeman and Moisen 2008).

TABLE 1.

Spatial autocorrelation in model residuals, as estimated by Moran's I-statistic, for the models of Common Eagle Ray and Bull Ray presence/absence and abundance. Values of Moran's I were transformed to z-scores, with values ≥⃒ 1.96 or ≤⃒ 1.96 indicating significant spatial autocorrelation (P ≤⃒ 0.05).

t01_382.gif

RESULTS

During 8 years of monitoring, fishing data were collected from 11,014 hauls, which were unevenly distributed over the study area (Figure 1). The number of hauls per year ranged between 584 and 2,038 (Table 2). The proportion of successful hauls (i.e., hauls in which rays were caught) was 5.8% for Common Eagle Rays and 1.3% for Bull Rays, accounting for a total of 1,857 and 215 individuals, respectively. Yearly CPUEs ranged between 0.024 and 0.151 individuals/106 m3 for Common Eagle Rays and between 0.002 and 0.021 individuals/ 106 m3 for Bull Rays (Table 2).

Modeling of Species Occurrence (Presence/Absence Models)

The final model for the presence/absence of Common Eagle Rays in the catch encompassed 4 of the 11 main effects or covariates and took the following form: presence (link = logit) ∼ s(longitude, latitude) + s(month) + s(year) + s(depth). The final model for the presence/absence of Bull Rays included 3 of the 11 covariates: presence (link = logit) ∼ s(longitude, latitude) + s(year) + s(month). Temperature, salinity, and all of the operational factors were dropped based on the covariate removal criteria suggested by Wood (2006). Results from the forward stepwise GAM-fitting process are summarized in Table 3, along with results for the variability explained by the factors that were sequentially added to the models. The model for Common Eagle Rays explained 21.4% of the total deviance, and the ANOVA F-ratio test indicated that all terms were significant (Table 3). Haul position was the most important factor in terms of relative contribution to the total deviance explained (77.1%), followed by month (15.4%) and year (6.6%; Table 3). The spatial pattern of catches indicated that the most important area of Common Eagle Ray aggregation was located in the northern Adriatic Sea, namely between 44.5°N and 45.5°N and between 12.5°E and 13.5°E (Figure 2). An area of lesser aggregation was observed in the southeastern-most part of the investigated area (between 43.0°N and 43.5°N and between 14.5°E and 15.0°E; Figure 2). The probability of Common Eagle Ray presence was very low during winter and reached its maximum values between late spring and early autumn. At larger temporal scale, an increasing trend in catches was observed from 2009 onward (Figure 2). The lowest contribution to the total deviance (0.9%) was accounted for by depth. The probability of encountering Common Eagle Rays remained fairly constant in waters of 10–65 m depth, declining sharply on deeper grounds (Figure 2).

TABLE 2.

Summary of midwater pair-trawling effort (number of hauls), the geographic range of hauls, and the catches of Common Eagle Rays and Bull Rays in the north-central Adriatic Sea during 2006–2013 (CPUE = number of individuals caught per 106 m3).

t02_382.gif

The model for Bull Ray presence/absence explained 27.7% of the total deviance, and all terms had significant effects (Table 3). Haul position again showed the highest explanatory power (64.3% of the total deviance explained), followed by year (26.3%) and month (9.4%; Table 3). The spatial pattern of Bull Ray catches allowed us to identify a primary aggregation area, which was located in the northern part of the Adriatic Sea (between 44.5°N and 45.5°N and between 12.5°E and 13.5°E; Figure 3). The probability of encountering Bull Rays changed significantly at the monthly level (with two peaks in May–June and September–October) and across years (with one peak during 2007–2008; Figure 3).

TABLE 3.

Results of generalized additive model building for factors affecting the presence of Common Eagle Rays and Bull Rays in the north-central Adriatic Sea during 2006–2013. The F-values and P-values are results from an ANOVA F-ratio test between the model in the given row and the model in the previous row (GCV = generalized cross validation).

t03_382.gif

Modeling of Species Abundance (Positive-Catch Models)

Evaluation of AIC scores to compare abundance models assuming different error distributions indicated that a gamma distribution with a log link function provided the best performance (Table 4). The final model for the abundance (i.e., positive catch) of Common Eagle Rays included five covariates and took the following form: log2(CPUE + 1) ∼ s(longitude, latitude) + s(net vertical opening) + s(haul duration) + s (season) + s(depth). The final abundance model for Bull Rays included four covariates: log2(CPUE + 1) ∼ s(longitude, latitude) + s(net vertical opening) + s(haul duration) + s(month).

Table 5 provides information on the GAM-fitting process, the deviance explained, and ANOVA results. The final abundance model for Common Eagle Rays accounted for 34.6% of the total deviance, and all terms were significant (Table 5). The highest contributions to the total deviance explained were provided by haul position (36.7%), haul duration (28.3%), and net vertical opening (21.7%). The plot illustrating the effect of haul position revealed that Common Eagle Rays were mostly abundant in the northern portion of the Adriatic Sea (between 44.5°N and 45.4°N and between 12.7°E and 13.5°E; Figure 4). The CPUE of Common Eagle Rays declined significantly with haul duration and was also negatively affected by the net vertical opening (Figure 4). Minor contributions to the model were provided by season (8.1%) and depth (5.2%; Table 5). Minimum CPUE values were observed in winter, and maximum CPUEs were observed in summer. The CPUE did not significantly vary within the bathymetric range (i.e., 20–50 m) preferred by Common Eagle Rays, whereas the reduced density of data points below 50-m depth led to very large SEs (Figure 4).

FIGURE 2.

Generalized additive model-derived effects of covariates used to model the presence of Common Eagle Rays in the north-central Adriatic Sea during 2006–2013. In the bivariate predictor map, the solid line is the estimate of the smooth function, and the black dots indicate haul locations. In the univariate predictor plots, the solid line is the estimate of the smooth function, the dashed lines indicate 95% confidence bands, and the “rug” or bars on the x-axis show the relative density of data points.

f02_382.jpg

The final abundance model developed for Bull Rays explained 56% of the total deviance, and all terms were significant (Table 5). Haul position (34.3% of the total deviance explained) and haul duration (38.8%) were the most important covariates, followed by net vertical opening (17.9%) and month (9.1%). The spatial pattern of CPUEs indicated that Bull Rays were especially abundant in the northern portion of the investigated area (between 44.5°N and 45.5°N), with an eastward decreasing trend (Figure 5). A sharp decline in Bull Ray CPUE was observed over the entire ranges of haul duration and net vertical opening. The CPUE of Bull Rays also showed significant monthly fluctuations, reaching maximum values between June and September and in December (Figure 5).

Model Validation

The estimated values of AUC in the ROC plots were 0.84 for Common Eagle Rays and 0.78 for Bull Rays, indicating good predictive performance of the presence/absence models (Figure 6). The optimum probability threshold for model performance—occurring when sensitivity (percentage of true positives that are correctly predicted) equals specificity (percentage of true negatives that are correctly predicted)—was 0.10 for Common Eagle Rays and 0.01 for Bull Rays. Setting the probability threshold at these values, 75% of Common Eagle Ray samples and 71% of Bull Ray samples were correctly classified.

The model that was developed from the positive catches of Common Eagle Rays showed a moderate predictive power; the best-fitting regression line explained about 30% of the variability (r2 = 0.29). The model overpredicted small values of abundance and underpredicted large values, as suggested by the slope (0.27) and intercept (0.75; Figure 7a). Mismatch between the observed and predicted abundances of Common Eagle Rays was mostly located in the northernmost section (beyond latitude 45.3°N) of the study area (Figure 8). In contrast, the predictive ability of the abundance model for Bull Rays was very high, as suggested by the amount of variability (97%) explained by the regression line (r2 = 0.97; Figure 7b).

DISCUSSION

To identify the most important environmental and operational factors explaining the spatiotemporal pattern of Common Eagle Ray and Bull Ray catches in the north-central Adriatic Sea, we analyzed fishery-dependent data by separately modeling the probability of species presence and the positive catch rates. Fisherydependent data can be intrinsically biased by an unbalanced spatial distribution of effort and changes in gear configuration. Despite this, fishery-dependent data have often been used as a surrogate for species density and to explore the relationship between fish distribution and spatial, temporal, and environmental variables (Walsh and Kleiber 2001; Damalas et al. 2007; Megalofonou et al. 2009; Damalas and Megalofonou 2010).

FIGURE 3.

Generalized additive model-derived effects of covariates used to model the presence of Bull Rays in the north-central Adriatic Sea during 2006–2013. In the bivariate predictor map, the solid line is the estimate of the smooth function, and the black dots indicate haul locations. In the univariate predictor plots, the solid line is the estimate of the smooth function, the dashed lines indicate 95% confidence bands, and the “rug” or bars on the x-axis show the relative density of data points.

f03_382.jpg

TABLE 4.

Results comparing the use of different underlying error distributions for full generalized additive models (i.e., those including all predictors) of Common Eagle Ray and Bull Ray abundance (i.e., positive catches) in the north-central Adriatic Sea during 2006–2013. Models were compared by using Akaike's information criterion (AIC) scores.

t04_382.gif

The final models developed by GAM analyses for the presence/absence and abundance (CPUE) of Common Eagle Rays and Bull Rays included different sets of covariates. The explanatory power of the species abundance models was twofold higher than that of the presence/absence models. Spatial factors (i.e., those representing the geographic location of the hauls) and temporal factors played the predominant role in the presence/absence models, whereas environmental features were of minor importance. Higher occurrences of Common Eagle Rays and Bull Rays were observed in the upper Adriatic Sea off the estuaries of three main river systems in northern Italy (the Po, Adige, and Brenta rivers). This zone is extremely productive due to the riverine input of nutrients and specific oceanographic regimes; it is characterized by a great abundance of benthic (mollusks and crustaceans) and pelagic (anchovies) consumers (Pranovi and Link 2009), which are preferred prey for Common Eagle Rays and Bull Rays (Jardas et al. 2004; Dulčić et al. 2008). In the eastern Mediterranean Sea, Megalofonou et al. (2009) reported a northward increase in catches of Blue Sharks Prionace glauca, and those authors hypothesized a positive effect of productivity on the observed pattern.

TABLE 5.

Results of generalized additive model building for factors affecting the abundance (log2[CPUE + 1]) of Common Eagle Rays and Bull Rays in the north-central Adriatic Sea during 2006–2013. The F-values and P-values are results from an ANOVA F-ratio test between the model in the given row and the model in the previous row (GCV = generalized cross validation).

t05_382.gif

Common Eagle Rays were also present—but to a lesser extent—in the southeastern-most portion of the investigated area, where they only occurred during autumn and winter months. The most striking effect of season on the spatial patterns in Common Eagle Ray and Bull Ray catches was the southward shift during the winter. Shelf water temperatures in the northern Adriatic Sea during winter are lower than temperatures in the rest of the basin, mainly due to the Po River outflow (Böhm et al. 2003), reaching values (9–10°C) that presumably act as a limiting factor for the presence of Common Eagle Rays and Bull Rays. A similar temporal pattern of distribution was reported by Manfredi et al. (2010), who analyzed catch data on Common Eagle Rays collected from 2001 to 2007 during experimental bottom trawl surveys. In the summer, catches of Common Eagle Rays were limited to the northeastern Adriatic Sea, whereas they also extended to the central portion of the Adriatic Sea during autumn (Manfredi et al. 2010). In other Mediterranean areas (Languedoc, France; and Izmir Bay, Turkey), Common Eagle Rays were also principally caught during summer and autumn (Capapé et al. 2007; Gurbet et al. 2013).

FIGURE 4.

Generalized additive model-derived effects of covariates used to model the abundance (log2[CPUE + 1]) of Common Eagle Rays in the northcentral Adriatic Sea during 2006–2013. In the bivariate predictor map, the solid line is the estimate of the smooth function, and the black dots indicate haul locations. In the univariate predictor plots, the solid line is the estimate of the smooth function, the dashed lines indicate 95% confidence bands, and the “rug” or bars on the x-axis show the relative density of data points.

f04_382.jpg

FIGURE 5.

Generalized additive model-derived effects of covariates used to model the abundance (log2[CPUE + 1]) of Bull Rays in the north-central Adriatic Sea during 2006–2013. In the bivariate predictor map, the solid line is the estimate of the smooth function, and the black dots indicate haul locations. In the univariate predictor plots, the solid line is the estimate of the smooth function, the dashed lines indicate 95% confidence bands, and the “rug” or bars on the x-axis show the relative density of data points.

f05_382.jpg

The presence of Common Eagle Rays and Bull Rays in midwater pair-trawl catches showed significant variation at a yearly scale, although the two species exhibited opposing patterns. The occurrence of Common Eagle Rays reached its minimum value in the 2008 survey, whereas the occurrence of Bull Rays attained its maximum during that year. In the subsequent years of monitoring, there was an increasing trend in Common Eagle Ray catches, whereas the occurrence of Bull Rays declined progressively. Very few studies examining medium- to long-term changes in elasmobranch catch rates from commercial fisheries have reported data on Common Eagle Rays. Aldebert (1997) listed the Common Eagle Ray as being among the species that have declined continuously in the Gulf of Lion since 1970. Damalas and Vassilopoulou (2011) documented the presence of Common Eagle Rays in the Aegean Sea during bottom trawl surveys in 1995–2000 but not during the subsequent period of monitoring (2003–2006). Given the reported general decline of Common Eagle Ray populations in these other areas, the increasing occurrence of this species in the northern Adriatic Sea in more recent years would seem unexpected. On the other hand, we cannot ascertain whether the temporal patterns of Common Eagle Ray and Bull Ray presence observed during this relatively short monitoring period should be interpreted in terms of cyclic fluctuations or as part of long-term trends.

The effects of spatial and temporal descriptors on Common Eagle Ray and Bull Ray abundances closely resembled those identified for the species' occurrence. Among the investigated environmental predictors, only depth had a significant effect on the presence and abundance of Common Eagle Rays. In the study area, this species was recorded in waters of 10–110-m depth but was found most often in waters down to 60-m depth, corresponding with the results of previous investigations (Jardas 1984; Manfredi et al. 2010). The most striking differences between the presence/absence models and the abundance models stemmed from the operational factors, which had a significant influence only on species abundances. The observed decline in Common Eagle Ray and Bull Ray CPUEs with increases in haul duration and net vertical opening was likely determined by the rarity of the species, which contributed to the low likelihood of catching more than 1–3 individuals/haul. Consequently, an increase in haul duration or net size resulted in little to no influence on the number of individuals caught but generated a substantial increase in the swept volume, thereby negatively affecting the fishing yield (CPUE). Though apparently ineffective for reducing the bycatch of Common Eagle Rays and Bull Rays, limiting the midwater trawl fishing effort in terms of haul duration and net size could still be beneficial. In fact, the shortening of haul duration would greatly enhance the viability of fish that are caught and, in turn, their chance of survival after being discarded at sea. The release of these and other elasmobranch species soon after their capture is a common practice in many Mediterranean commercial fisheries (Capapé et al. 2008; Dulčić et al. 2008; Fortuna et al. 2010;Gurbet et al. 2013) and should be further encouraged by implementing activities that increase awareness of the “best practices” for mitigating the impact of bycatch on threatened species.

FIGURE 6.

Predictive performance of the presence/absence models for (a) Common Eagle Rays and (b) Bull Rays in the north-central Adriatic Sea, as measured by the receiver operating characteristic (ROC) plot and the area under the ROC curve (AUC). The point on the curve where sensitivity equals specificity (circle) and the relative threshold value are indicated.

f06_382.jpg

In this study, the model-fitting stage was followed by a validation procedure, which indicated that the presence/absence models for Common Eagle Rays and Bull Rays yielded good predictive accuracy. The model for Common Eagle Ray abundance data can also be considered reliable, although it leads to the overestimation of low abundance values. This is demonstrated by the correlation between the observed and predicted values. The abundance model for Bull Rays showed even better performance, but it should be tested again by using a larger validation data set to adequately assess its predictive ability. To develop models with higher predictive power and to extrapolate results to a wider spatial scale, further catch data should be obtained from other areas and other fishing gears. In addition, an effort should be made to test other factors that may potentially affect the distribution of Common Eagle Rays and Bull Rays, such as the presence and abundance of their preferred prey. In this manner, the predictive ability of the models can be improved.

FIGURE 7.

Relationship between predicted and observed abundances (log2 [CPUE + 1]) for (a) Common Eagle Rays and (b) Bull Rays in the northcentral Adriatic Sea.

f07_382.jpg

FIGURE 8.

Map of the study area in the north-central Adriatic Sea, depicting the spatial distribution of midwater pair-trawl fishing effort in relation to the (a) observed and (b) predicted abundances (CPUE, number of individuals per 106 m3) of Common Eagle Rays in the catch during 2013.

f08_382.jpg

The incidence of Common Eagle Ray and Bull Ray bycatch in midwater pair-trawling fisheries of the Adriatic Sea still remains unassessed since updated knowledge of the actual population abundances at this and larger geographic scales is lacking. However, the precautionary approach suggests the use of coordinated long-term monitoring and research programs to accurately quantify the impacts of fishing activity on Common Eagle Rays and Bull Rays in the Adriatic Sea and elsewhere in the Mediterranean Sea. In addition to species-specific fisheries-related data, further ecological information (e.g., fish age and growth, movement patterns, and potential nursery areas) will be crucial for the development of conservation targets and indicators. Research, monitoring, and information exchange should be undertaken through the involvement of scientific networks and in cooperation with relevant fisheries organizations.

ACKNOWLEDGMENTS

This work was funded by the Italian Ministry of Agriculture, Food, and Forestry (General Directorate of Fishery and Aquaculture) within the scientific and monitoring program on the implementation of Regulation (EC) Number 812/2004. Special thanks are due to the local coordinators: Otello Giovanardi, Antonio Mazzola, Sasa Raicevich, Antonello Sala, and Carola Vallini. We are also grateful to all of the independent observers who worked for the BYCATCH projects since 2006 and to Maria Luisa Cassese, Antonello Bruschi, and Valeria Pesarino for their support in the downloading and postprocessing of MyOcean environmental data.

REFERENCES

1.

Akaike, H. 1973. Information theory and an extension of the maximum likelihood principle. Pages 267–281 in B. N. Petrov and F. Csaki, editors. Second international symposium on information theory. Akademiai Kiado, Budapest. Google Scholar

2.

Aldebert, Y. 1997. Demersal resources of the Gulf of Lions (NW Mediterranean). Impact of exploitation on fish diversity. Vie Milieu 47:275–284. Google Scholar

3.

Bigelow, K. A., C. H. Boggs, and X. He. 1999. Environmental effects on Swordfish and Blue Shark catch rates in the U.S. North Pacific longline fishery. Fisheries Oceanography 8:178–198. Google Scholar

4.

Böhm, E., V. Banzon, E. D'Acunzo, F. D'Ortenzio, and R. Santoleri. 2003. Adriatic Sea surface temperature and ocean colour variability during the MFSPP. Annales Geophysicae 21:137–149. Google Scholar

5.

Burnham, K. P., and D. R. Anderson. 2002. Model selection and multimodel inference: a practical information-theoretic approach, 2nd edition. Springer, New York. Google Scholar

6.

Capapé, C. 1976. Étude du régime alimentaire de l'Aigle de mer, Myliobatis aquila (L., 1758) des côtes tunisiennes. [Study of the feeding regime of the Eagle Ray, Myliobatis aquila (L., 1758) of the Tunisian coasts.] Journal du Conseil International pour l'Exploration de la Mer 37:29–35. Google Scholar

7.

Capapé, C. 1977. Étude du régime alimentaire de la Mourine vachette, Pteromylaeus bovinus (Geoffroy Saint-Hilaire, 1817), (Pisces, Myliobatidae) des côtes tunisiennes. [Study of the feeding regime of the Bull Ray, Pteromylaeus bovinus (Geoffroy Saint-Hilaire, 1817), (Pisces, Myliobatidae) of the Tunisian coasts.] Journal du Conseil International pour l'Exploration de la Mer 37:214–220. Google Scholar

8.

Capapé, C., O. Guélorget, Y. Vergne, and J. P. Quignard. 2007. Reproductive biology of the Common Eagle Ray Myliobatis aquila (Chondrichthyes: Myliobatidae) from the coast of Languedoc (southern France, northern Mediterranean). Vie Milieu 57:1–6. Google Scholar

9.

Capapé, C., Y. Vergne, J. P. Quignard, and C. Reynaud. 2008. Biological observations on the Eagle Ray Myliobatis aquila (Chondrichthyes: Myliobatidae) off the Languedocian coast (southern France, northern Mediterranean). Annales Series Historia Naturalis 18:167–172. Google Scholar

10.

Cavanagh, R. D., and C. Gibson. 2007. Overview of the conservation status of cartilaginous fishes (chondrichthyans) in the Mediterranean Sea. International Union for Conservation of Nature, Gland, Switzerland. Google Scholar

11.

Chong, Y. S., and J. L. Wang. 1997. Statistical modeling via dimension reduction methods. Nonlinear Analysis: Theory, Methods, and Applications 30:3561–3568. Google Scholar

12.

Compagno, L. J. V., D. A. Ebert, and M. J. Smale. 1989. Guide to the sharks and rays of southern Africa. Struik Publishers, Cape Town, South Africa. Google Scholar

13.

Damalas, D., and P. Megalofonou. 2010. Environmental effects on Blue Shark (Prionace glauca) and Oilfish (Ruvettus pretiosus) distribution based on fishery-dependent data from the eastern Mediterranean Sea. Journal of the Marine Biological Association of the United Kingdom 90:467–480. Google Scholar

14.

Damalas, D., P. Megalofonou, and M. Apostolopoulou. 2007. Environmental, spatial, temporal and operational effects on Swordfish (Xiphias gladius) catch rates of eastern Mediterranean Sea longline fisheries. Fisheries Research 84:233–246. Google Scholar

15.

Damalas, D., and V. Vassilopoulou. 2011. Chondrichthyan bycatch and discards in the demersal trawl fishery of the central Aegean Sea (eastern Mediterranean). Fisheries Research 108:142–152. Google Scholar

16.

Drexler, M., and C. H. Ainsworth. 2013. Generalized additive models used to predict species abundance in the Gulf of Mexico: an ecosystem modeling tool. PLoS (Public Library of Science) One [online serial] 8:e64458. Google Scholar

17.

Dulčić, J., L. Lipej, M. O. Bonaca, R. Jenko, B. Grbec, O. Guélorget, and C. Capapé. 2008. The Bull Ray, Pteromylaeus bovinus (Myliobatidae), in the northern Adriatic Sea. Cybium 32:119–123. Google Scholar

18.

El Kamel, O., N. Mnasri, J. Ben Souissi, M. Boumaïza, M. M. Ben Amor, and C. Capapé. 2009. Inventory of elasmobranch species caught in the Lagoon of Bizerte (northeastern Tunisia, central Mediterranean). Pan-American Journal of Aquatic Science 4:383–412. Google Scholar

19.

Ferretti, F., R. A. Myers, P. Sartor, and F. Serena. 2005. Long-term dynamics of the chondrichthyan fish community in the upper Tyrrhenian Sea. International Council for the Exploration of the Sea, C.M. 2005/N:25, Copenhagen. Google Scholar

20.

Ferretti, F., G. C. Osio, C. J. Jenkins, A. A. Rosenberg, and H. K. Lotze. 2013. Long-term change in a meso-predator community in response to prolonged and heterogeneous human impact. Scientific Reports 3:1057. Google Scholar

21.

Fielding, A. H., and J. F. Bell. 1997. A review of methods for the assessment of prediction errors in conservation presence/absence models. Environmental Conservation 24:38–49. Google Scholar

22.

Fortuna, C. M., C. Vallini, E. Filidei, M. Ruffino, I. Consalvo, S. Di Muccio, C. Giona, U. Scacco, E. Tarulli, O. Giovanardi, and A. Mazzola. 2010. Bycatch of cetaceans and other species of conservation concern during pair trawl fishing operations in the Adriatic Sea (Italy). Chemistry and Ecology 26:65–76. Google Scholar

23.

Freeman, E. A., and G. Moisen. 2008. PresenceAbsence: an R package for presence-absence model analysis. Journal of Statistical Software 23: 1–31. Available:  www.jstatsoft.org/article/view/v023i11. (January 2015). Google Scholar

24.

Froese, R., and D. Pauly. 2015. FishBase. Available:  www.fishbase.org (April 2015). Google Scholar

25.

Guisan, A., and N. E. Zimmermann. 2000. Predictive habitat distribution models in ecology. Ecological Modelling 135:147–186. Google Scholar

26.

Gurbet, R., O. Akyol, E. Yalçin, and O. Özaydin. 2013. Discards in bottom trawl fishery in the Aegean Sea (Izmir Bay, Turkey). Journal of Applied Ichthyology 29:1269–1274. Google Scholar

27.

Hastie, T., and R. Tibshirani. 1986. Generalized additive models. Statistical Science 1:297–318. Google Scholar

28.

Hollowed, A. B., S. J. Barbeaux, E. D. Cokelet, E. Farley, S. Kotwicki, P. H. Ressler, C. Spital, and C. D. Wilson. 2012. Effects of climate variations on pelagic ocean habitats and their role in structuring forage fish distributions in the Bering Sea. Deep-Sea Research Part II Topical Studies in Oceanography 65–70:230–250. Google Scholar

29.

Holtzhausen, J.A., D.A. Ebert, F. Serena, , and C. Mancusi. 2009. Myliobatis aquila. International Union for Conservation of Nature, IUCN Red List of Threatened Species, version 2014.3. Available:  www.iucnredlist.org. (January 2015). Google Scholar

30.

Jardas, I. 1984. Horizontal and vertical distribution of benthos Selachia (Pleurotremata, Hypotremata) in the Adriatic. FAO (Food and Agriculture Organization of the United Nations) Fisheries Reports 290:95–108. Google Scholar

31.

Jardas, I., M. Šantić, and A. Pallaoro. 2004. Diet composition of the Eagle Ray, Myliobatis aquila (Chondrichthyes: Myliobatidae), in the eastern Adriatic Sea. Cybium 28:372–374. Google Scholar

32.

Katsanevakis, S., C. D. Maravelias, D. Damalas, A. P. Karageorgis, E. V. Tsitsika, C. Anagnostou, and C. Papaconstantinou. 2009. Spatiotemporal distribution and habitat use of commercial demersal species in the eastern Mediterranean Sea. Fisheries Oceanography 18:439–457. Google Scholar

33.

La Mesa, G., A. Annunziatellis, E. Filidei Jr. , and C. M. Fortuna. 2015. Modeling environmental, temporal and spatial effects on Twaite Shad (Alosa fallax) bycatches in the central Mediterranean Sea. Fisheries Oceanography 24:107–117. Google Scholar

34.

Malak, D. A., S. R. Livingstone, D. Pollard, B. A. Polidoro, A. Cuttelod, M. Bariche, M. Bilecenoglu, K. E. Carpenter, B. B. Collette, P. Francour, M. Goren, M. H. Kara, E. Massutí, C. Papaconstantinou, and L. Tunesi. 2011. Overview of the conservation status of the marine fishes of the Mediterranean Sea. International Union for Conservation of Nature, Gland, Switzerland. Google Scholar

35.

Manfredi, C., E. Ciavaglia, I. Isajlović, C. Piccinetti, and N. Vrgoč. 2010. Temporal and spatial distribution of some elasmobranchs in the northern and central Adriatic Sea. Biologia Marina Mediterranea 17:254–255. Google Scholar

36.

Maravelias, C. D. 1997. Trends in abundance and geographic distribution of North Sea Herring in relation to environmental factors. Marine Ecology Progress Series 159:151–164. Google Scholar

37.

Maravelias, C. D., G. Tserpes, M. Pantazi, and P. Peristeraki. 2012. Habitat selection and temporal abundance fluctuations of demersal cartilaginous species in the Aegean Sea (eastern Mediterranean). PLoS (Public Library of Scinece) One [online serial] 7:e35474. Google Scholar

38.

Maunder, M. N., and A. E. Punt. 2004. Standardizing catch and effort data: a review of recent approaches. Fishery Research 70:141–159. Google Scholar

39.

Mavrič, B., R. Jenko, T. Makovec, and L. Lipej. 2004. On the occurrence of the Pelagic Stingray, Dasyatis violacea (Bonaparte, 1832), in the Gulf of Trieste (northern Adriatic). Annales Series Historia Naturalis 14:181–186. Google Scholar

40.

Megalofonou, P., D. Damalas, M. Deflorio, and G. De Metrio. 2009. Modeling environmental, spatial, temporal, and operational effects on Blue Shark bycatches in the Mediterranean long-line fishery. Journal of Applied Ichthyology 25:47–55. Google Scholar

41.

Moran, P. A. P. 1948. The interpretation of statistical maps. Journal of the Royal Statistical Society: Series B 37:243–251. Google Scholar

42.

Murase, H., H. Nagashima, S. Yonezaki, R. Matsukura, and T. Kitakado. 2009. Application of a generalized additive model (GAM) to reveal relationships between environmental factors and distributions of pelagic fish and krill: a case study in Sendai Bay, Japan. ICES Journal of Marine Science 66:1417–1424. Google Scholar

43.

Notarbartolo di Sciara, G., and I. Bianchi. 1998. Guida degli squali e delle razze del Mediterraneo. [Guide to the sharks and rays of the Mediterranean Sea.] Muzzio Editore, Padova, Italy. Google Scholar

44.

Pranovi, F., and J. S. Link. 2009. Ecosystem exploitation and trophodynamic indicators: a comparison between the northern Adriatic Sea and southern New England. Progress in Oceanography 81:149–164. Google Scholar

45.

Quéro, J. 1998. Changes in the Euro-Atlantic fish species composition resulting from fishing and ocean warming. Italian Journal of Zoology 65:493–499. Google Scholar

46.

R Development Core Team. 2013. R: a language and environment for statistical computing. R Foundation for Statistical Computing, Vienna. Available:  www.R-project.org. (June 2016). Google Scholar

47.

Seck, A. A., Y. Diatta, A. Gueye-Ndiaye, and C. Capapé. 2002. Observations on the reproductive biology of the Bull Ray, Pteromylaeus bovinus (E. Geoffroy Saint-Hilaire, 1817) (Chondrichthyes: Myliobatidae) from the coast of Senegal (eastern tropical Atlantic). Acta Adriatica 43:87–96. Google Scholar

48.

Serena, F. 2005. Field identification guide to the sharks and rays of the Mediterranean and Black Sea. Food and Agriculture Organization of the United Nations, Species Identification Guide for Fishery Purposes, Rome. Google Scholar

49.

Stoner, A. W., J. P. Manderson, and J. P. Pessutti. 2001. Spatially explicit analysis of estuarine habitat for juvenile Winter Flounder: combining generalized additive models and geographic information systems. Marine Ecology Progress Series 213:253–271. Google Scholar

50.

Valavanis, V. D., G. J. Pierce, A. F. Zuur, A. Palialexis, A. Saveliev, I. Katara, and J. Wang. 2008. Modelling of essential fish habitat based on remote sensing, spatial analysis and GIS. Hydrobiologia 612:5–20. Google Scholar

51.

Van der Elst, R. 1988. A guide to the common sea fishes of southern Africa, 2nd edition. Struik Publishers, Cape Town, South Africa. Google Scholar

52.

Walsh, W., and P. Kleiber. 2001. Generalized additive model and regression tree analysis of Blue Shark (Prionace glauca) by the Hawaii-based longline fishery. Fishery Research 53:115–131. Google Scholar

53.

Wintner, S. P. 2006. Pteromylaeus bovinus. International Union for Conservation of Nature, IUCN Red List of Threatened Species, version 2014.3. Available:  www.iucnredlist.org. (January 2015). Google Scholar

54.

Wood, S. N. 2001. mgcv: GAMs and generalized ridge regression for R. R News 1:20–25. Google Scholar

55.

Wood, S. N. 2006. Generalized additive models: an introduction with R. Chapman and Hall/CRC Press, Boca Raton, Florida. Google Scholar

56.

Zogaris, S., and U. Dussling. 2010. On the occurrence of the Bull Ray Pteromylaeus bovinus (Chondrichthyes: Myliobatidae) in the Amvrakikos Gulf, Greece. Mediterranean Marine Science 11:177–184. Google Scholar
© Gabriele La Mesa, Aldo Annunziatellis, Elio Filidei Jr., and Caterina Maria Fortuna This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. The moral rights of the named author(s) have been asserted.
Gabriele La Mesa, Aldo Annunziatellis, Elio Filidei, and Caterina Maria Fortuna "Bycatch of Myliobatid Rays in the Central Mediterranean Sea: the Influence of Spatiotemporal, Environmental, and Operational Factors as Determined by Generalized Additive Modeling," Marine and Coastal Fisheries: Dynamics, Management, and Ecosystem Science 8(8), 382-394, (12 August 2016). https://doi.org/10.1080/19425120.2016.1167795
Received: 24 July 2015; Accepted: 15 March 2016; Published: 12 August 2016
JOURNAL ARTICLE
13 PAGES


SHARE
ARTICLE IMPACT
Back to Top