Data and code for: Pelagic sharks target long-lived, retentive anticyclonic eddies in the Northwest Atlantic Ocean

Open ocean ecosystems represent the largest habitat on Earth and are highly dynamic in time and space. Mesoscale eddies are a primary driver of this variability and serve a key structural role in ocean ecosystems. Eddies modulate marine biodiversity beyond their impacts on plankton, influencing many ecologically and commercially important predators that may preferentially occupy anticyclonic eddies. However, how animal-eddy interactions scale across predator species and the mechanistic drivers of these relationships remain an area of active research. We integrated satellite tracking data for sharks with observations of mesoscale eddies to determine how four shark species interact with eddies in the Gulf Stream region. Based on over 24,000 tracking days, we found that blue, white, and shortfin mako sharks selected for the cores of anticyclones while use of eddies by tiger sharks was less conspicuous. Some particularly large and long-lived anticyclones were occupied by tagged sharks for multiple weeks suggesting that these eddies may serve as hotspots for pelagic predators.

Methods
Methods of data collection/generation:

Satellite tagging and shark movements

We analyzed data from 108 deployments of satellite-linked electronic tags on four species of sharks in the North Atlantic: white shark (Carcharodon carcharias, n=10); tiger shark (Galeocerdo cuvier, n=21); shortfin mako (Isurus oxyrinchus, n=48); and blue shark (Prionace glauca, n=29). These tags provide relatively accurate (generally <10 km error; Douglas et al. 2012) satellite-based positions when the tagged individual comes to the surface, but do not provide any information on vertical habitat use. Resulting locations were processed with a Kalman filtering algorithm by Collecte Localisation Satellites (Lopez et al., 2014) and subsequently assigned error flags called location classes (LC): LC 3, <250 m; LC 2, 250-500 m; LC 1, 500-1500 m; LC 0, >1500 m for classes 3, 2, 1, 0. Additional classes A, B represent positions derived from less than four satellite messages which result in no estimates of spatial accuracy from CLS; however, recent work on several marine species and platforms suggests error for A, B classes is order 1-10 km and nearly always < 20 km (Lopez et al., 2014). Location class Z positions were considered invalid and removed from further analysis. Each individual’s tracking data was filtered using a speed filter (10 ms−1) to remove unrealistic locations and was divided into trajectories that allowed for data gaps of up to two days and required at least 20 locations per trajectory. These criteria were chosen to minimize interpolation in the modeling approach used to standardize the location data. The resulting filtered, individual trajectories were fitted in a continuous-time state-space modeling framework (foieGras package for R; Jonsen et al. 2020) to remove spurious locations and provide standardized, quality-controlled individual trajectories at 6-h time steps following previous work that indicates daily transmission rates of ∼1-4 positions per day for these species (Wiernicki et al., 2022). Resulting standardized trajectories were used to collocate shark movements to the mesoscale eddy field in the Gulf Stream study region, defined here to encompass 35-45ºN and 80-40ºW in the North Atlantic. Eddy diameters in the study area range from tens to hundreds of kilometers, therefore positional error inherent in the shark trajectories is relatively modest.

Null eddy use was quantified by using 20 correlated random-walk simulations per tagged shark. Individual simulated trajectories were constructed as a correlated-random walk that randomly draws from the observed turn angles and step lengths in the observed shark movement trajectories to generate simulated movement trajectories (adehabitatLT package for R; Calenge 2006). The use of observed trajectories to parameterize the simulations ensures realistic movements but with the hypothesized ecological mechanism, in this case eddy use, removed (Gotelli and Graves, 1996). Previous studies have used similar simulations to generate null models for habitat selection (e.g., Richard et al. 2013; Queiroz et al. 2016). To match the spatial bias in presence data, simulations were initiated at the tagging location for each individual and were constrained to realistic movements using bathymetry. Simulated tracks for each individual represent random eddy use based on the chance of encountering these features.

Eddy tracking

Eddies identified and tracked in daily maps of sea surface height were acquired from the Mesoscale Eddy Trajectory Atlas distributed by Aviso that describes daily tracks of coherent (i.e., retentive) mesoscale structures (eddies) based on maps of surface altimetry (Mason et al., 2014; Pegliasco et al., 2022). Eddies with lifetimes greater than four weeks are tracked daily based on their signatures in sea level anomaly (SLA) fields, and the Atlas includes both the latitude/longitude center and speed-based radius (R) of each feature throughout its lifetime, where R is the radius of a circle with an area equal to that enclosed by the streamline of the maximum circum-average geostrophic speed of the eddy (equivalent to eddy length scale Ls in Chelton et al. 2011). This radial eddy-centric distance is used to define the zones of an eddy relative to its center, assuming eddies are isotropic, where a normalized distance of zero is the eddy center, distances <1 indicate the core (i.e., interior) of the eddy, >1 and <2 indicate the periphery of the eddy, and >2 indicate waters outside the physical impact of the eddy (Gaube et al., 2017a). A custom meander filter was used to distinguish Gulf Stream meanders from the eddies of interest using net-zonal displacement (Gaube et al., 2017b; Braun et al., 2019).

We extracted polarity (cyclone or anticyclone), age (days), radius (R), average rotating speed at the contour of maximum circum-average speed of the eddy (hereafter, "speed"), and amplitude (i.e., the deviation in SLA from the surrounding waters) from the metrics reported in the eddy atlas (for further details, see https://doi.org/10.24400/527896/a01-2022.005) for each eddy matching real and simulated shark movement trajectories. We complemented this information by calculating eddy retention (d’Ovidio et al., 2013) and non-linearity (Chelton et al., 2011) metrics. Retention is a semi-Lagrangian metric that represents how long water parcels have been circulating within the core of an eddy (defined as area of negative Okubo-Weiss parameter, Okubo 1970; Weiss 1991). In this study, we used calculations of eddy retention carried out for the North Atlantic Aerosols and Ecosystems Study (Della Penna and Gaube, 2019) that relied on the altimetry-derived geostrophic currents distributed by the Copernicus Marine Environment Monitoring Service (CMEMS, http://marine.copernicus.eu). The non-linearity of an eddy was used to characterize the ability of these features to advect a trapped fluid parcel as it translates (Chelton et al., 2011). This is commonly represented as the nondimensional ratio U/c, where U is the maximum rotational speed (as described above) and c is the translation speed of the eddy estimated at each point along the eddy trajectory from centered differences of the (x, y) coordinates of successive centroid locations. Values of U/c > 1 imply that there is trapped fluid within the eddy interior that is advected with the eddy as the eddy translates.

See manuscript for complete details and associated references.