Combinaison de la modélisation biophysique
et de marquages isotopiques

Study species 

This study is part of a larger program focusing on the population connectivity of the Humbug damselfish (Pomacentridae), Dascyllus aruanus, in the SWL of New Caledonia. Here, we therefore use the larval life history traits of this species as the basis for our exploration of larval dispersal patterns in the SWL. D. aruanus is an obligate coral-dwelling Pomacentridae, found exclusively in lagoon habitats (Allen 1991), where it lives in well-mixed waters among branching coral colonies in spatially discrete groups of 2 – 80 individuals (Sale 1972a, Holbrook et al. 2000, Cole 2002). Coral colonies provide protection from predators and substrate for laying benthic eggs Coates (1980b), Mizushima et al. (2000). D. aruanus adults are sedentary benthic spawners that breed on a lunar cycle throughout the year (Pillai et al. 1985). Spawning peaks in summer, during which each female can spawn several times at 1 week (personal observation) to 2 months (Mizushima et al. 2000) intervals. Eggs remain in benthic nests for 3 days after which hatchlings are released into the plankton where they disperse on average 3 weeks (mean planktonic larval duration -PLD- Thresher et al. (1989), Juncker et al. (2007), Soeparno et al. (2012) prior to settling on adult reef habitats. Newly hatched larvae have well developed sensorial abilities (Leis 2010) and pomacentrids larvae are able to swim actively against currents during the second half of their pelagic larval phase (Fisher 2005). Furthermore, field evidence indicates that late-stage larvae of coral reef fish can detect the presence of a reef at a range of at least 1 km (Leis et al. 1996). Thus the sensorial and swimming abilities which allow the orientation of competent larvae to suitable recruitment habitat (Sweatman 1983, Holbrook et al. 2000) may be present in D. aruanus larvae as early as the age of 11 days (i.e., half of mean PLD). 

Biophysical model 

Larval dispersal was simulated with a biophysical model using version 3.0 of the Lagrangian tool Ichthyop (Lett et al. 2008). Ichthyop is a three-dimensional (3D) particle-tracking model designed to study the effects of physical and biological factors on the transport and settlement of ichthyoplankton. The biophysical model is based on an offline forcing of an individual-based model (IBM) by a 3D hydrodynamic model. The hydrodynamic model used here is the high-resolution 3D Model for Applications at Regional Scales (MARS3D, Lazure and Dumas (2008)). . The model grid has a horizontal resolution of 500 m and 30 terrain following generalized sigma levels in the vertical dimension. This configuration is forced by realistic hourly winds at 4 km resolution obtained from the mesoscale Weather Research and Forecasting (WRF) model (Lefèvre et al. 2010). Simulated surface wind speeds and directions are very close to observations (Indice of Agreement IOA of 0.9 and RMSE < standard deviation of observations for both speed and direction, section 3.6, table 3.2). Realistic wind forcing corresponds to the years 2003-2004. This period is neutral regarding ENSO phases. Tides are included in MARS3D through a lateral forcing using the Oregon State University TPXO.6 tides solution (Egbert et al. 1994) for 8 tidal constituents. The TPXO tides solution is refined further within ADCIRC (ADvanced CIRCulation model, Luettich et al. (1992)), by using an unstructured and very fine resolution mesh (from 500 to 25 m in the SWL, Lefèvre, pers. comm.). Our larval transport IBM uses MARS3D model results covering the reproductive period of D. aruanus in New Caledonia i.e., from mid – September to late March. Outputs of MARS3D simulations were stored every 12 min as this time step is sufficient to account for the effects of tides. In the IBM, larvae are characterized by their latitude (°S), longitude (°E) and depth (m). Locations of individuals are updated every 5 min in three dimensions using the velocity fields stored from MARS3D interpolated in space and time via a forward-Euler integration scheme.

Simulations

In order to study the effect of wind regime on larval retention, we examined dispersal from a ∼ 500 m diameter patch reef located at the center of the SWL (called natal reef hereafter) where water residence time is close to the average of 11 days 68 / 219 (Ouillon et al. 2010). Location and extent of the natal reef and settlement areas in the SWL were defined as polygons based on GIS habitat maps provided by the atlas of coral reefs in New Caledonia (Andréfouët and Torres-Pulliza 2004). Since D. aruanus is ubiquitous in the SWL all reefs shallower than 20 m were considered as potential settlement habitats in the model. Hatching events (representing a release of 500 virtual larvae each) were simulated over the natal reef every 3 h over austral summer from mid-September 2003 to late March 2004 (i.e., 1320 simulations). For each release on the natal reef, larvae were randomly distributed throughout the water column from 0 to 20 m depth. Larvae from each hatching event were followed for up to 30 days (the largest value of PLD reported for D. aruanus (Soeparno et al. 2012)). We considered that larvae were initially transported passively by ocean currents during an 11-day precompetency period and that they became active afterwards with sensory and swimming capabilities that allowed them to detect and approach a settlement area. To do so, a non-explicit swimming behavior during the competency period was included by assuming that larvae could actively settle once at a given distance from a settlement reef. This distance was defined as 1 km (Leis et al. 1996). Any virtual larva located less than 1 km away from a settlement reef at any time between the end of the precompetency period and 30 days was then considered to have successfully settled. Simulations were run under two alternative hypotheses about natal homing. Under the first hypothesis, settlement was supposed to be driven by strict natal homing : settlement was only allowed on the natal reef, i.e., competent larvae were only able to settle on the natal reef. We will refer to this hypothesis as the “strict homing hypothesis”. Under the second hypothesis, settlement of competent larvae was allowed on the natal reef and in any other part of the SWL where suitable habitat for D. aruanus was available. We will refer to this hypothesis as the “no-homing hypothesis”. A longer precompetency period of 3 weeks (21 days) was also tested. A sensitivity analysis to the larval release depth was also conducted using simulations run for a precompetency period of 11 days for both hypotheses regarding natal homing. As larvae are characterized by their latitude (°S), longitude (°E) and depth (m) at each time step, it is possible to know the release depth of each recruited larvae at the end of the simulation. Five release depth intervals were tested from 0 to 20 m. All post-processing computations were done using R-2.15.0. 

Lagoon vs. reef retention

Simulation outputs were used to calculate larval retention at two different spatial scales. Larval retention was first computed for each simulation at the SWL scale. This retention, hereafter called “natal lagoon retention” (NLR), is defined as the ratio of the number of larvae released at the natal reef that settled on any of the settlement 69 / 219 Material and methods 51 reefs to the total number of larvae released at the natal reef. Larval retention was also computed for each hatching event at the natal reef scale. This local retention, hereafter referred to as “natal reef retention” (NRR), is defined as the ratio of the number of larvae released at the natal reef that settled back to that site to the total number of larvae released at the natal reef. Three retention time series extending from mid-September 2003 to early March 2004 were obtained : NLR under the no-homing hypothesis ; NRR under the no-homing hypothesis ; NRR under the strict-homing hypothesis. Note that under the strict-homing hypothesis NLR equals NRR. 

Cross-correlations

To study the link between wind conditions and simulated D. aruanus larval retention, we calculated cross-correlations between wind and retention time series with a maximum lag of 30 days. We extracted hourly meridional and zonal wind components from the WRF model at the closest grid point to the natal reef and converted them into an along-shelf (V-component) and cross-shelf (U-component) coordinate system, rotated 60° anti-clockwise from true north, with V positive towards the north-west (300°) and U positive onshore towards 30°. Time series of daily probability of occurrence of the four weather regimes defined by Lefèvre et al. (2010) were also used in a cross-correlation analysis with simulated retention time series. We used the Spearman rank-order correlation coefficient (hereafter Spear-R) because the simulated retention values were not normally distributed. Given that autocorrelation of time series increases the risk to consider that correlations between series are significant when they are not, we accounted for autocorrelation of all time series explicitly in judging the significance of correlations by adjusting the degrees of freedom following Pyper and Peterman (1998) and Botsford and Paulsen (2000). Auto-correlation timescales of order 2-8 days, depending on the simulation and time series examined, were identified and the effective degrees of freedom were corrected accordingly. This correction reduced the effective degrees of freedom and consequently substantially increased the value of correlation required for a significant result. All reported correlation coefficients are significantly different from zero at the 95% confidence level. 

Larval settlement maps

Larval settlement maps were plotted at the SWL scale from the simulations with the no-homing hypothesis. For each hatching event, the spatial distribution of settlers was computed for square grid cells of 0.01° spatial resolution. The proportional number of settlers (relative to the total number of released larvae = 500) in each grid cell was then averaged over all hatching events, as well as over two extreme decompositions of the simulation period : (1) hatching events whose precompetency period consisted of at 70 / 219 least 75% weather regime 1 (Strong SE Trade-wind, table 3.1) and (2) hatching events whose precompetency period consisted of at least 75% weather regime 4 (Subtropical SE wind, table 3.1). The centers of mass of settlement maps were then calculated as the weighted (by settlement) spatial average over grid cells. To enhance visibility, some results were converted to a logarithmic scale by calculating log10(Ni + 1) where Ni is the total number of larvae, over all simulations, settling in grid cell i. 

 Results 

 Time series 

The retention time series exhibit considerable temporal variability for both precompetency periods (fig. 3.2a and b). For an 11-day precompetency period, natal lagoon retention (NLR) ranges from 0% to 100% with a mean of 56.7% (± 26.4% SD). Natal Reef Retention (NRR) ranges from 0% to 42.6% with a mean of 6.7% (± 8.2% SD) under the strict-homing hypothesis. Under the no-homing hypothesis, NRR is considerably smaller, ranging from 0% to 23.0% with a mean of 1.4% (± 2.3% SD). The three simulated retention time series are highly positively correlated (Spear-R 0.63, 0.82, and 0.73, for the NLR and NRR when no homing, NLR and NRR when strict homing, and NRR when no homing and NRR when strict homing time series, respectively). At both spatial scales (natal reef and natal lagoon) and under both hypotheses (strict and no homing), mean retention decreases as the precompetency period increases. For a 21-day precompetency period, NLR ranges from 0% to 86.4% with a mean of 32.9% (± 20.9% SD). NRR ranges from 0% to 18.8% with a mean of 1.9% (± 3.0% SD) under the strict-homing hypothesis. Under the no-homing hypothesis, it ranges from 0% to 4.0% with a mean of 0.3% (± 0.5% SD). Wind conditions simulated by WRF model over austral summer 2003-2004 show two predominant states (fig. 3.2c) representing each about a third of the time series. Periods corresponding to trade winds with wind speed > 8 m s-1 and steady direction (mean = 110°N ± 23 SD) alternate with periods with weak wind speed (< 5 m s-1) and variable direction (mean = 160°N ± 88 SD). During trade wind episodes the along-shore wind component (V) is high towards North-West (mean = 8.6 m s-1 ± 2.0 SD) and the cross-shelf wind component (U) is oriented offshore (mean = -1.5 m s-1 ± 2.4 SD), whereas weak and variable wind episodes are characterized by a lower V (mean = 1.2 m s-1 ± 2.3 SD) and U component towards the coast (mean = 0.1 m s-1 ± 2.3 SD) (fig. 3.2c and d). Consistent with these conditions, wind regimes 1 (Strong SE Trade-wind) and 4 (Subtropical SE wind) are predominant during the study period (fig. 3.2e), occurring 35% and 43% of the study period, respectively. Regimes 2 (weak 71 / 219 Results 53 easterly circulation) and 3 (tropical SE wind) winds represent 8% and 13% of the study period, respectively. 

 Cross-correlations 

Natal lagoon retention (NLR) For an 11-day precompetency period, the time series of NLR shows significant negative cross-correlations with the along-shore component of wind (fig. 3.3a) between days 2 and 9 after release, with an absolute maximum correlation coefficient of 0.3 occurring at a lag of 5 days. For a 21-day precompetency period, correlations are consistently negative at about -0.2 for lags of 3-20 days, albeit only marginally significant at lags of 12-13 days and 18-19 days (correlation coefficient of ∼ -0.3) (fig. 3.3b). Cross-correlations with the cross-shelf component are not significant. For both precompetency period lengths, NLR is negatively correlated with the probability of weather regime 1 (fig. 3.4a and b) between days 2 and 5 after release with absolute maximum correlation coefficients of 0.4. NLR is not significantly correlated to regime 4, but correlation coefficients are generally positive (fig. 3.4c and d) for both precompetency period lengths. Cross-correlations with regimes 2 and 3 are not significant. Natal reef retention (NRR) Time series of NRR under the strict-homing hypothesis show significant negative cross-correlations with the along-shore component of wind for both precompetency period lengths (fig. 3.3a and b). Significant correlations occur between days 3 and 10 and between days 5 and 13 after release for precompetency period lengths of 11 and 21 days respectively, with absolute maximum correlation coefficients of 0.3 for both precompetency period lengths (fig. 3.3a and b). Under the no-homing hypothesis NRR is not significantly correlated to along-shore winds for a precompetency of 11 days, but correlation coefficients are generally negative at the beginning of the precompetency period (fig. 3.3a). For a precompetency of 21 days, the time series of NRR under the no-homing hypothesis shows significant negative cross-correlations with alongshore winds between 5 and 7 days (fig. 3.3b), with an absolute maximum correlation coefficient of 0.2. Cross-correlations with cross-shelf winds are not significant. For an 11-day precompetency period, NRR under the strict homing hypothesis is significantly and negatively correlated to regime 1 at day 4 with an absolute maximum correlation coefficient of 0.4 (fig. 3.4a) and NRR under the no-homing hypothesis is not significantly correlated to weather regime 1 at the beginning of the PLD, but the correlation coefficients are consistently negative (fig. 3.4a). For an 11-day precompetency 72 / 219 period, NRR is significantly and positively correlated with regime 1 between days 16 and 24 (no-homing hypothesis) and between days 17 and 19 (strict-homing hypothesis) with maximum correlation coefficients of 0.4 (fig. 3.4a). For a 21-day precompetency period with the no-homing hypothesis, NRR is significantly and negatively correlated with regime 1 between days 3 and 6, with an absolute maximum correlation coefficient of 0.4 (fig. 3.4b). For a 21-day precompetency period with the strict-homing hypothesis, NRR is significantly and positively correlated with regime 1 between days 21 and 24, with a maximum correlation coefficient of 0.4 (fig. 3.4b). For an 11-day precompetency period, NRR under the strict-homing hypothesis is significantly and positively correlated to regime 4 between days 1 and 5 after release with a maximum correlation coefficient of 0.5 (fig. 3.4c). Under the no-homing hypothesis NRR is not significantly correlated to regime 4, but correlation coefficients are generally positive (fig. 3.4c). For a 21-day precompetency period and under both homing hypotheses, NRR is significantly and positively correlated with regime 4 between days 4 and 5 after release with maximum correlation coefficients of 0.5 (fig. 3.4d). Cross-correlations with regimes 2 and 3 are not significant.