Quasi-planktonic behavior of foraging top marine predators

Monitoring marine top predators is fundamental for assessing the health and functioning of open ocean ecosystems. Although recently tracking observations have substantially increased, factors determining the horizontal exploration of the ocean by marine predators are still largely unknown, especially at the scale of behavioral switches (1–100 km, days-weeks). It is commonly assumed that the influence of water movement can be neglected for animals capable of swimming faster than the current. Here, we challenge this assumption by combining the use of biologging (GPS and accelerometry), satellite altimetry and in-situ oceanographic data (ADCP and drifting buoys) to investigate the effect of the mesoscale ocean dynamics on a marine predator, the southern elephant seal. A Lagrangian approach reveals that trajectories of elephant seals are characterized by quasi-planktonic bouts where the animals are horizontally drifting. These bouts correspond to periods of increased foraging effort, indicating that in the quasi-planktonic conditions energy is allocated to diving and chasing, rather than in horizontal search of favourable grounds. These results suggest that mesoscale features like eddies and fronts may act as a focal points for trophic interactions not only by bottom-up modulation of nutrient injection, but also by directly entraining horizontal displacements of the upper trophic levels.


Supplementary information
Details about the Quasi-Planktonicity Index algorithm The QPI compares a section of a seal's trajectory with the movement of a numerical passive tracer that is determined only by the horizontal currents. In the ideal case of a perfectly determined velocity field, if a trajectory is purely passive, the numerical passive tracer should perfectly match the real trajectories. However, in reality there are two sources of error that have to be dealt with: • an uncertainty on the initial condition of the velocity field: the spatial resolution of altimetry is considered to be comparable to its grid spacing, i.e., 1/3 • , hence when we initialize a numerical drifter we may actually initialize it with a mismatch of 1/3 • in respect to the velocity field, • an underestimation of the horizontal velocities: altimetry observations are taken along satellite tracks and are then interpolated together for providing a gridded product. The interpolation procedure smooths the signal and may underestimate real velocities, resulting in our case into a delay of the simulated trajectories which lag behind real ones and in turn, into a spurious mismatch during the comparison.
• ageostrophic components: by definition, ageostrophic components of the velocity fields do not appear as a signal on the Sea Surface Height and therefore cannot be observed by satellite altimetry.
The algorithm to compute the QPI aims at mitigating the effect of these sources of error. To compute QPI, we perform the following steps: 1. sample the elephant seal's trajectory X(t) = (X(t), Y (t)) with a 6-hours frequency.
2. for each day t 0 we initialize around the location X(t 0 ) a set of j initial conditions x j (t 0 ) = (x j (t 0 ), y j (t 0 )). They represent the initial conditions of a set of synthetic trajectories D r ( where r indicates the radius of the disk.
3. advect the initial conditions for a time t max = N + t buf f er .
4. for each elephant seal's locations between t 0 and t 0 +N : where dist refers to en Eulerian distance computed on the non-regular latitude-longitude grid.
5. compute the QP I as the mean distance between the closest simulated trajectory (shadow trajectory) and the real one This algorithm limits the effect of the uncertainty on the initial location of the altimetric velocity in respect to the location of the real trajectory (that for the case of elephant seals and SVP drifters we consider with no error, given the high resolution of GPS tracks), by advecting an ensemble of numerical trajectories whose radius r is chosen in relation to the resolution of the altimetry data: in this study we used r = 0.3 • .
To compensate for a lag in the simulated trajectory we introduce the pseudodistance defined in step 4 instead of an Eulerian step-by-step distance. Indeed, even if a simulated and a measured trajectory are very close, if the velocity field is underestimated, the Eulerian distance would increase and we would not identify a low value of the diagnostic. Therefore, we advect the simulated trajectory for a t max that is not just equal to the number of steps we use for  the comparison (N ), but we introduce a buffer (in this study buf f er = 4 days), so that we make sure that we are compensating for all the effects of the delay. We then ensure that for each position of the real trajectory, the distance is computed with the closest point of the simulated trajectory, and not the point that corresponds to the same instant. Finally, we addressed the presence of possible ageostrophic components by validating the QPI (i.e., computing it for SVP drifters) also adding to satellitederived currents the Ekman components derived by wind re-analysis. Figure SI 4 displays the distributions of the QPI computed for SVP drifters using different altimetry products. Even if by using a regional Ekman-corrected altimetry the simulated trajectories have Lagrangian properties that are more similar to the SVP ones, as detailed in Ref [1], the changes in the QPI distribution does not change qualitatively the result.
When computing this algorithm, there are few parameters that can be tuned. If r is constrained by the resolution of the velocity field, the choice of N is relatively flexible and it is related to the scale of the patterns we want to identify. As in this study we are interested in labelling bouts of trajectories with a resolution high enough to distinguish behavioral switches between extensive and intensive foraging (typically of of few days) we used a value of N = 4 days = 16 steps. The QP I can be computed with different frequencies: in this study we sampled the trajectories every 6 hours, but when comparing the QP I with the attempt capture rate, we used a daily resolution to integrate for the effects of the daynight cycle of the attempt capture rate.   and attempt capture rate measurements (c) along an elephant seal's trajectory. Note that the colorscale in (c) is reversed. The patterns along this and the other trajectories of the attempt capture rate, heading velocity and the QPI include, in agreement with previous observations [2] of their foraging habit, an inbound and an outbound phase of the trajectory, with a lower attempt capture rate and high QPI, with an intensive foraging and low QPI phase in between. Figure 4: Distributions of the QPI for SVP (real) drifters computed using different altimetry products: a) geostrophic global product, c) Ekman-corrected regional product and e) geostrophic regional product. Using different products does not alter significantly the shape and the extent of the distribution, yet differences in the distributions can be observed in the tails, as displayed in b), d) and f).