Dissertation - HQ

104 Vertical distribution during ontogeny
Advection scheme
eddies and fine scale vertical variations in the flow. In addition, the
large variations in weather conditions made it impossible to get a
continuous picture of the current throughout eight consecutive days.
Therefore, the Regional Ocean Modeling System (ROMS) was used to
simulate the flow around a cylindrical deep-ocean island, similar to, but
larger than, Tetiaroa (20 km diameter). The configuration was modified
from Dong et al. 219 . The incoming flow featured a vertical shear that
was parameterised according to ADCP measurements in the presumed
upstream region of Tetiaroa: 20 cm s -1 near the surface, 12 cm s -1 at
100 m, with a sigmoid decline. The simulation grid had dimensions
200 × 100 km × 500 m, with an horizontal resolution of 500 m, and 20
evenly spaced layers, which allowed to resolve mesoscale eddies shed
by the island.
Particles advection was performed off-line, with custom Fortran code.
The grid was restricted to the inner 150 × 80 km and the current
field was interpolated on a 250 m regular mesh through 4 th order
polynomial interpolation. In the vertical direction, only the first 100 m
were used, and divided in 5 layers. Simulated pre-flexion larvae were
simultaneously released at 100 points within a 5 km radius around
the island, at a depth of 25 m (the depth of maximum occurrence of
pre-flexion larvae in most families), on three occasions, contrasting
in terms of flow conditions. Larvae were advected using a first order
forward scheme, with a 1.30 h time step. While the advection scheme
did not allow all fine scale features of the flow to affect trajectories,
the same scheme was used for passive and vertically migrating larvae.
So it did not bias the comparison which was the primary focus of the
simulations here.
5.4 Results
5.4.1 Factors affecting vertical distribution
Predominant effect
of taxonomy
Circumstantial
differences for
sub-family taxa
An univariate regression tree was built to hierarchise the effect of
taxonomy, physical variables, and ontogeny on the location of the z cm,
computed per family and per stage, at each station. The first splits, robust
after cross-validation, show a strong effect of taxonomy (Figure 5.2).
Some families, such as Lutjanidae, Lethrinidae, and Holocentridae, were
systematically higher in the water column than others (Figure 5.3).
These two splits alone account for 23% of the variability (residual cross
validated error = 0.77). This result is confirmed by a very significant
difference between per-family z cms (Kruskal-Wallis, χ 2 = 211.43, df = 9,
p < 10 -16 ) which leaves no doubt, even though variances are different
(Fligner-Killeen, χ 2 = 29.7, df = 9, p = 0.0005).
Some differences between the distributions of sub-family taxonomic
groups are significant. For example, within the Serranidae, Epinephelini
were higher in the water column than Grammistini (Fligner-Killeen,
χ 2 = 1.1, df = 1, p = 0.3; Wilcoxon, W = 5, p = 0.002). There is also a