24.10.2014 Views

Dissertation - HQ

Dissertation - HQ

Dissertation - HQ

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

150 Oceanography vs. behaviour<br />

controls increase significantly: from 140 · 170 · 3 · 6 = 176,400 to<br />

300 · 160 · 5 = 240,000 for the state and from seven to hundreds for<br />

controls. On the other hand, stochasticity sources are removed: there<br />

is no explicit energy budget anymore, so no feeding, and survival is<br />

not represented (i.e. is considered homogenous spatially). So the use of<br />

transition matrices is inappropriate because they would be very large<br />

but only one final state would be reachable from any given initial state<br />

Loops instead of and decision. Therefore, gain is computed at each state inside a loop,<br />

matrix computation and transition matrices are never built. The loop is coded in Fortran 90<br />

and compile time vectorisation (using Intel Fortran Compiler) as well<br />

as local parallelisation (through OpenMP) accelerate the process. The<br />

extraction/interpolation of current speeds and computation of optimal<br />

decisions takes between one and two hours for the set of parameters<br />

specified above, on a cluster node with four 2.33 GHz double core CPUs.<br />

Both optimal decision and end points after advection are stored in a<br />

NetCDF file, so the advection does not have to be done again for the<br />

forward computation of trajectories. Computing trajectories is therefore<br />

virtually instantaneous and can be done in an interpreted language (R<br />

in this case).<br />

6.4.4 Large impact of swimming<br />

Swimming<br />

greatly enhances<br />

self-recruitment<br />

First, the impact of swimming is assessed by comparing trajectories of<br />

passive and active larvae in the same situation (identical release sites and<br />

date, same PLD). Figures 6.15 and 6.16 highlight the tremendous impact<br />

of swimming, even for slow swimming larvae, in both configurations<br />

used here. When larvae are treated as passive particles, most of them<br />

are advected away from their release location. In the island case, passive<br />

retention in regions of weak flow is almost never sufficient to retain<br />

particles for the whole larval phase. In the promontory case, where<br />

backward flow is more stable behind the cape, a small percentage of<br />

larvae can be passively retained and self-recruit. These are just five<br />

or ten trajectories, in one flow situation, but they are representative<br />

of the overall magnitude of the impact of swimming. The effect is<br />

estimated quantitatively by computing the percentage of recruiting (i.e.<br />

self-recruiting) larvae starting from the promontory or the island at<br />

three release dates (Table 6.2). In all cases, self-recruitment is quasiimpossible<br />

for passive particles but swimming shifts the regime to a<br />

situation where most larvae can self-recruit.<br />

Table 6.2 Mean percentage of successful trajectories for the coral reef fish P.<br />

amboinensis and a temperate fish, in the island and promontory configurations.<br />

Coral-Reef Temperate<br />

passive active passive active<br />

Island 0 95 0 45<br />

Promontory 2 95 1 72

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!