Present address: R. S. Schick, University Program in Ecology, Duke University, Durham, NC 27708, USA
Inclusion of prey data improves prediction of bluefin tuna (Thunnus thynnus) distribution
Version of Record online: 28 JAN 2009
© 2009 The Authors.
Volume 18, Issue 1, pages 77–81, January 2009
How to Cite
SCHICK, R. S. and LUTCAVAGE, M. E. (2009), Inclusion of prey data improves prediction of bluefin tuna (Thunnus thynnus) distribution. Fisheries Oceanography, 18: 77–81. doi: 10.1111/j.1365-2419.2008.00499.x
- Issue online: 28 JAN 2009
- Version of Record online: 28 JAN 2009
- Received 17 February 2008 Revised version accepted 10 December 2008
- Atlantic herring;
- bluefin tuna;
- Clupea harengus;
- Gulf of Maine;
- prey data;
- sea surface temperature fronts;
- spatial correlation;
- Thunnus thynnus
We examined the distribution of Atlantic bluefin tuna (Thunnus thynnus) in the Gulf of Maine, Northwest Atlantic Ocean, from 17 to 23 August 1995, in relation to physical and biological parameters. Specifically, we fit a binomial GLM to the bluefin tuna presence–absence data and predictor variables that include: sea surface temperature (SST), ocean depth, distance to an SST front, time-lagged density of SST fronts, and an interpolated surface of Atlantic herring (Clupea harengus) density. In addition, we use simple and partial Mantel tests to examine whether bluefin tuna presence–absence data are significantly associated with these predictors, once spatial autocorrelation is accounted for. Results suggest that the distribution of bluefin tuna significantly correlated with herring density (z = 3.525, P = 0.000424), and that inclusion of biological variables results in a more parsimonious model. Mantel tests results indicate that bluefin tuna abundance is significantly correlated with herring density after the effect of spatial structure is removed (Mantel r = 0.043, P < 0.019).
A comprehensive understanding of the distribution of pelagic species such as tunas, sharks and billfish depends on data describing their movements, distribution and abundance, environment and prey associations. Having all of this information would be optimal but, in reality, it is difficult to generate location data for a species of interest, and even harder to collect or generate these data for associated prey species. Fortunately, with new remote sensing technologies and GIS it has become easier to document the physical template for pelagic species at meaningful spatial and temporal scales. Scientists and managers can map and integrate ecological information from the co-location of pelagic species and their environment. As part of the species-environment relationship for pelagic species, theory suggests that sea surface temperature fronts represent oceanic areas where shear forces aggregate prey, which in turn may concentrate predators (Olson and Backus, 1985; Olson et al., 1994). Studies have empirically explored this relationship (Laurs et al., 1984; Fiedler and Bernard, 1987; Podestáet al., 1993; Royer et al., 2004; Schick et al., 2004), and while Fiedler and Bernard (1987) were able to document differential stomach contents in tunas closer to fronts, Podestáet al. (1993) and Schick et al. (2004) found that fronts were a good, if imperfect, predictor of the distribution of pelagic species. Royer et al. (2004) found tunas spatially correlated with thermal and ocean color fronts. Where possible, these studies noted that the addition of prey into the models would be a useful analytical exercise. To complete such an exercise, here we use both remotely sensed data and in situ prey data to explore the quantitative relationship for bluefin and their environment.
Recently, the number and distribution of prey patches in the Gulf of Maine (GOM) were predicted based on a two-dimensional analysis of movements of individual bluefin tuna (Gutenkunst et al., 2006) traveling in schools tracked by acoustic methods (Lutcavage et al., 2000). Presumably, the putative prey would also be located at fronts and bathymetric features. Although much attention has been paid to the distribution and abundance of herring in the GOM (Sinclair and Iles, 1985; Overholtz, 2006), and although herring are the bluefin tuna’s preferred prey there (Crane, 1936; Chase, 2002; Estrada et al., 2005; Golet et al., 2007), relatively little is known about their co-occurrence, or what role each plays in the GOM ecosystem. Because of this lack of knowledge and because of recent apparent declines in both bluefin tuna and their fattening and condition factors (Golet et al., 2007), as well as declines in inshore Atlantic herring biomass in the GOM, there is compelling need for fisheries managers to better understand the relationship between bluefin tuna and herring. For example, see for example the discussion generated by the ‘Variation in the Distribution of Bluefin Tuna in the Gulf of Maine’ session at the Maine Fishermen’s Forum, March 5th, 2005, Rockland, ME, USA (http://www.fishermensvoice.com/archives/herringturfwar.html, last accessed 2 January 2009).
We present a unique glimpse into the co-occurrence of both species and attempt to assess the statistical significance of their observed distributions. With data on the distribution of prey and predator, herring and bluefin tuna, we use a binomial GLM and a partial Mantel test to test whether the inclusion of prey data results in a better model fit for bluefin tuna. In addition, we examine the implications of our results for fisheries management and the need for a better understanding of top mobile predators and impacts of ecosystem shifts in the GOM ecosystem.
We added data on prey distribution to the physical model built and tested in Schick et al. (2004). The prey distribution layer was an interpolated surface of Atlantic herring density. To develop the herring layer we started with point locations that represented trawl stations from a mid-water trawl conducted by NOAA Fisheries, Northeast Fisheries Science Center (Dr Michael Fogarty, pers. commun.; see http://www.nefsc.noaa.gov/sos/ and Clark (1998) for further descriptions of NOAA Fisheries, Northeast Fisheries Science Center’s monitoring and assessment efforts). A total of 82 trawls were conducted in the GOM at different times between 17 August and 23 August 1995. At each of these trawl stations several direct and expanded statistics on catch were tabulated. We chose the ‘expanded weight of catch’ statistic for Atlantic herring. From these points we used universal kriging (ver Hoef, 1993) to interpolate a surface of herring density (Fig. 1). We generated the kriged surface using a spherical model for the semivariogram. Following methods previously outlined (Schick et al., 2004) we extracted prey density values for both tuna presence and absence locations that occurred during the same time period that the trawls were conducted. Interpolations and extractions were performed with ArcInfo® workstation version 9.0 GIS software (ESRI, Redlands, CA, USA).
With a complete database of bluefin presence–absence and accompanying values for each of the physical and biological predictor variables, we used two statistical methods (GLM and partial Mantel test) to test whether (i) bluefin tuna were significantly correlated with the herring distribution, and (ii) the inclusion of prey resulted in a more parsimonious model (as determined by Akaike’s information criterion, AIC). For the GLM we fit bluefin tuna presence–absence data to a full model that, in addition to herring distribution data, included the physical variables sea surface temperature (SST), distance to an SST front, time-lagged density of SST fronts, bottom depth and bottom slope (for details see Schick et al., 2004). From the full model we ran a forwards and backwards stepwise variable selection procedure that used AIC to determine the optimal model. We re-fit the resulting model to the data and summarized the significance of each of the predictor variables. In addition, for comparison we fit the final model from Schick et al. (2004) to this temporal subset of data. That is, we tested the presence–absence model for this brief snapshot of data against the same set of variables in the 2004 model (tuna presence versus depth, temperature, distance to fronts, frontal density, and slope). Because each of the variables in these models is spatially autocorrelated, we used a partial Mantel test to account for this autocorrelation in a test comparing bluefin tuna presence–absence and herring density.
We explored two additional models: (i) herring density versus the environment; and (ii) tuna school size versus the environment. The first model was a GLM of interpolated herring catch against each of the five environmental variables (depth, distance to front, frontal density, sea surface temperature, and slope). We performed a step-wise selection to determine the final model. The second model was a Poisson GLM of tuna school size versus the environment. Here we used the same environmental variables as the bluefin presence–absence model, and performed step-wise variable selection.
During the study period bluefin tuna were clumped around an area of high herring density (Fig. 1). While aerial surveys for bluefin tuna covered a broader spatial and temporal extent than the herring data presented here (Schick et al., 2004), we only extracted prey data for bluefin schools that occurred at the same spatial and temporal extent of the herring data. Forty-four schools of bluefin tuna were located, representing a total of 2114 individuals. This observed distribution results in a statistically significant relationship between bluefin tuna and herring (Table 1).
|Distance to front||−0.0000||0.0000||−0.18||0.8551|
The inclusion of herring density in the model resulted in a final model with lower AIC than one comprising only physical variables; AIC for the final model from Schick et al. (2004) applied to these data: 423.96; versus AIC for the model that includes prey: 412.46. The final model depicted bluefin presence as a linear relationship with herring density, SST, and bottom depth (Table 2). In a less parsimonious model that included SST fronts, herring density and an interaction term between these two variables, there was a significant relationship to bluefin tuna presence (P < 0.0355). After controlling for spatial structure, a significant relationship existed between bluefin presence–absence and herring density (Mantel r = 0.043, P < 0.019).
Results from the model of herring density versus the environment indicate the herring are strongly correlated with fronts. Interpolated herring values are negatively correlated with distance to front (P << 0.0001), and positively correlated with frontal density (P << 0.0001). Unlike the presence–absence model, the final model selected for tuna school size versus the environment included all variables. Larger schools are found farther from prey, and in cooler waters (temperature: β = −0.03, P << 0.001; prey: β = −0.001, P << 0.001).
Many retrospective analyses on pelagic fishes and their environment have relied on remotely sensed variables to help decipher what drives their distribution patterns. As the assembly of these physical datasets is relatively easy, the physical variables used (e.g., SST, ocean color, temperature and color fronts) are often used in place of more proximate (but harder to record) variables. For example, bluefin tuna might not be in search of fronts themselves, but the prey that are more likely to be found at these fronts (Olson et al., 1994; Lutcavage et al., 2000; Brill et al., 2002). From a data gathering and processing standpoint, it may be easier to use fronts as a proxy for prey location, but actual data on the distribution of prey are preferable, especially when ecosystem relationships and trophic structure are of interest. Our results suggest that bluefin tuna co-occur with herring distributions in areas of the Gulf of Maine. However, this finding needs to be viewed with appropriate caution, because the results presented here represent a brief ‘snapshot’ (1 week) of the distributions of highly mobile pelagic species. Concluding that bluefin tuna will always be found in areas of high herring density would be over-reaching, because this is not always the case (Golet et al., 2007). Nevertheless, bluefin tuna enter the GOM to feed, and the inclusion of herring was both highly significant and resulted in a better model fit.
Inclusion of prey data resulted in a final model that excluded both indices of SST fronts that were significant in the final model presented in Schick et al. (2004). Given previous analysis of frontal relationships, the lack of a significant relationship in the model was surprising because it suggests that their role may be incidental to the role of prey or temperature distribution. For example, in both Podestáet al. (1993) and Schick et al. (2004), fronts were significant, but they did not explain all the variance observed in the distribution of swordfish and bluefin tuna, respectively. Although we do not expect fronts to be the sole patterning agent for herring and bluefin tuna, information on prey distribution is a significantly better predictor than the location and density of SST fronts. Our results failed to reveal a direct linear link between fronts, herring and bluefin tuna; however, two items point to synergistic effects among biotic and abiotic factors: (i) the significant relationship between bluefin presence and the interaction term for fronts and herring; and (ii) the significant positive relationship between frontal density and herring density. A full understanding of this trophic structure would require larger datasets over longer time frames as well as more information on the distribution of prey items for herring. Lastly, the significant results for the GLM on school size were not surprising, but the direction of the relationship was. A priori we expected larger schools to be in locations with higher herring density, but this was the opposite of what we observed. This may be an artifact of the small sample size, as it is clear from both the spatial locations (Fig. 1) and the presence–absence results (Table 2) that bluefin are seen in waters with higher herring density. Alternatively, there may be other aggregation drivers (e.g., lunar phase, tidal effects, social meeting points) acting separately from the foraging force (Lutcavage and Kraus, 1995; Lutcavage et al., 2000).
The role of prey or forage species in marine ecosystems has been documented in recent theoretical research (Walters et al., 2005). In their modeled ecosystems, Walters et al. (2005) found that application of Maximum Sustainable Yield policies leads to ecosystem degradation and a loss in top predators. Although these results are theoretical, they call into question the role of single species assessment in the GOM, and point to a need for additional research that examines the role of each of these fisheries within the larger GOM ecosystem. For example, speculation exists among fishers that increased landings of herring in the GOM have resulted in a regional depletion of inshore herring biomass, and may be spatially and temporally linked to reduced bluefin landings. The New England Fishery Management council recently restricted the mid-water trawl fishery for herring owing to declining herring biomass (http://www.nefmc.org/press/press_releases/HerringPR106.pdf, last accessed 2 January 2009). Whether the regional (i.e., inshore) reduction in herring biomass produced the recent declines in the bluefin fishery remains to be seen, but at the very least it is clear that when we are able to statistically test that the distribution of bluefin tuna is related to their prey, bluefin are significantly correlated with elevated densities of herring.
We thank Mike Fogarty and Joan Palmer of NOAA Fisheries’ Northeast Fisheries Science Center for access to the prey data, and for their patience in answering questions about said dataset. We also thank John Field and Tom Pearson of NOAA Fisheries, Southwest Fisheries Science Center, Fisheries Ecology Division, as well as François Royer of University of New Hampshire’s Large Pelagics Research Lab, for helpful comments on earlier versions of this manuscript.
- 2002) Horizontal and vertical movements of juvenile North Atlantic Bluefin Tuna (Thunnus thynnus) in the western North Atlantic determined using ultrasonic telemetry, with reference to population assessment by aerial surveys. Fish. Bull. 100:155–167. , , et al. (
- 2002) The diet of bluefin tuna (Thunnus thynnus) off the coast of Massachusetts. Fish. Bull. 100:168–180. (
- 1998) Status of Fishery Resources off the Northeastern United States for 1998. Technical Report 115. Woods Hole, MA: NOAA-NMFS-NEFSC. (
- 1936) Notes on the biology and ecology of giant tuna Thunnus thynnus L. Zool. N. Y. 212:207–212. (
- 2005) Diet and trophic position inferred from stable carbon and nitrogen isotopes of Atlantic bluefin tuna (Thunnus thynnus). Mar. Biol. 147:37–45. , and (
- 1987) Tuna aggregation and feeding near fronts observed in satellite imagery. Cont. Shelf Res. 7:871–881. and (
- 2007) Decline in condition of northern bluefin tuna (Thunnus thynnus) in the Gulf of Maine. Fish. Bull. 105:390–395. , , and (
- 2006) Inferring resource distributions from Atlantic bluefin tuna movements: an analysis based on net displacement and length of track. J. Theor. Biol. 245:243–257. , , and (
- 1993) Universal kriging for ecological data. In: Environmental Modeling with GIS. M.Goodchild, B.Parks & L.Steyaert (eds) Oxford: Oxford University Press, pp. 447–453. (
- 1984) Albacore tuna catch distribution relative to environmental features observed from satellites. Deep-Sea Res. 31:1085–1099. , and (
- 1995) The feasibility of direct photographic aerial assessment of giant bluefin tuna in New England waters. Fish. Bull. 93:495–503. and (
- 2000) Tracking adult North Atlantic bluefin tuna (Thunnus thynnus) in the northwestern Atlantic using ultrasonic telemetry. Mar. Biol. 137:347–358. , , , , and (
- 1985) The concentrating of organisms at fronts: a cold-water fish and a warm-core Gulf Stream ring. J. Mar. Res. 43:113–137. and (
- 1994) Life on the edge: marine life and fronts. Oceanography 7:52–60. , , , , and (
- 2006) Estimates of consumption of Atlantic herring (Clupea harengus) by bluefin tuna (Thunnus thynnus) during 1970–2002: an approach incorporating uncertainty. J. Northw. Atl. Fish. Sci. 36:55–63. (
- 1993) Exploring the association between swordfish catch rates and thermal fronts on U.S. longline grounds in the western North Atlantic. Cont. Shelf Res. 13:253–277. , and (
- 2004) Association between bluefin tuna schools and oceanic features in the western Mediterranean. Mar. Ecol. Prog. Ser. 269:249–263. , and (
- 2004) Bluefin tuna (Thunnus thynnus) distribution in relation to sea surface temperature fronts in the Gulf of Maine (1994–96). Fish. Oceanogr. 13:225–238. , and (
- 1985) Atlantic herring (Clupea harengus) distributions in the Gulf of Maine–Scotian Shelf area in relation to oceanographic features. Can. J. Fish. Aquat. Sci. 42:880–887. and (
- 2005) Possible ecosystem impacts of applying MSY policies from single-species assessment. ICES J. Mar. Sci. 62:558–568. , , and (