Amélioration des méthodes de prévision de dérive et de dispersion d'objets flottants à la surface de l'océan

RESUME: Cette etude a pour but de developper des methodes pour ameliorer les previsions de derive et de dispersion des objets flottants dans l'estuaire et le golfe du Saint-Laurent (EGSL). La precision de ces previsions depend dans une large mesure, de la qualite des previsions de courants oceaniques, de vent et de vagues, mais egalement du modele de derive utilise. Le modele de derive le plus utilise de maniere operationnelle au Canada et ailleurs se base sur les previsions de courants proches de la surface fournis par la cellule superieure du modele de circulation auquel on ajoute un terme de correction proportionnel a la vitesse du vent proche de la surface. Une telle formulation suppose que ce terme de correction inclut de nombreux processus, y compris la derive due au vent, le cisaillement vertical du courant oceanique non resolu, la derive induite par les vagues et les processus sous-meso-echelles mal resolus par le modele de circulation a faible resolution horizontale. Pourtant, ces trois derniers processus ne sont pas necessairement lies lineairement a la vitesse du vent local. En consequence nous avons tente de prendre en compte de facon explicite ces trois processus. Le cisaillement a ete resolu en extrapolant les courants proches de la surface (fournis par le modele oceanique regional GSL5km) a la surface en supposant la dynamique d'Ekman. La derive de Stokes a ete incluse dans le modele de derive. Celle-ci a ete simulee par le modele WAVEWATCH III (WW3). Les donnees atmospheriques proviennent du systeme de prevision deterministe regional (\emph{Regional Deterministic Prediction System}, RDPS) de 35~km de resolution centre sur l'EGSL. Les processus sous-meso-echelles sont pris en compte via l'augmentation de la resolution horizontale du modele de circulation (STLE400m, couvrant uniquement l'Estuaire du Saint-Laurent). Les performances des modeles de derive implementes ont ete evaluees a l'aide des observations de bouees derivantes deployees dans l'EGSL. En ajoutant un terme de correction proportionnel au vent au courant extrapole, les modeles de derive implementes ont apporte une reduction de la distance de separation relativement au modele standard allant de 25\% a 35\% sur differents delais de derive de 3h a 72h. En remplacant le terme de correction du vent par la derive de Stokes cette reduction d'erreur atteint 40\%. Mieux encore, on obtient une bonne prevision de derive en augmentant la resolution horizontale du modele de circulation, en conservant la methode d'extrapolation du courant proche de la surface en surface puis en incluant la derive de Stokes. Cela conduit effectivement a un score de competence pouvant atteindre 0.90 sur une periode de derive de 65h. Deux cas differents de deploiement des bouees ont ete envisages pour etudier la dispersion. Les indicateurs de dispersion simules a deux resolutions differentes sont compares a ceux calcules a partir des observations. Dans le premier cas d'etude, la dispersion relative et absolue sont soit surestimees ou sous-estimees par les modeles de derive tandis que dans l'autre cas, le modele de derive utilisant le courant extrapole a basse resolution horizontale sous-estime severement ces deux indicateurs de dispersion. L'augmentation de la resolution horizontale a permis une amelioration de la prevision de la dispersion. La diffusivite relative et la dispersion relative montrent deux regimes regime de dispersion. Un regime de dispersion non-local pour des separations plus petites que 3.5~km pour le premier deploiement d'un ensemble de 9 deriveurs et 500~m pour le deuxieme deploiement d'un ensemble de 8 deriveurs. Au-dela de ces distances la dispersion est locale. -- Mot(s) cle(s) en francais : prevision, derive de surface, couche d'Ekman, bouees derivantes, extrapolation, derive de Stokes, sous-meso-echelles, dispersion. -- ABSTRACT: The general objective of this study is to improve surface drift and dispersion forecasts methods for floating objects in the Estuary and Gulf of St.~Lawrence (EGSL). The accuracy of these forecasts depends to a large extent on the quality of ocean currents, winds and waves forecasts, but also on the drift model used. The standard Eulerian leeway drift model used in most operational systems in Canada and elsewhere considers near-surface currents provided by the top grid cell of the ocean circulation model and a correction term proportional to the near-surface wind. Such formulation assumes that the wind correction term accounts for many processes including windage, unresolved ocean current vertical shear, wave-induced drift, and submesoscale processes poorly resolved by the low horizontal resolution ocean model. However, these processes are not necessarily linearly related to the local wind velocity. We have attempted to take into account explicitly these three processes. The shear was resolved by extrapolating near-surface currents (provided by the regional ocean model GSL5km) to the surface assuming Ekman dynamics. The Stokes drift computed by the WAVEWATCH III model (WW3) was included in the drift model. The atmospheric data come from the Regional Deterministic Forecast System (RDPS) with a 35-km resolution centered on the EGSL. Submesoscale processes are taken into account by increasing horizontal resolution of the ocean model (STLE400m, covering only the St.~Lawrence Estuary). The performances of the implemented drift models were evaluated using observations of drifting buoys deployed in the EGSL. By adding a wind correction term to the extrapolated currents, the implemented drift models brought reductions of the separation distances relative to the control model reaching 25\% to 35\% over different drift times ranging from 3h to 72h. By replacing the wind correction term with the Stokes drift, this reduction of separation distance reaches 40\%. Better still, a good drift forecast is obtained by increasing the horizontal resolution of the circulation model, while keeping the extrapolation method of the near-surface current to the surface and including the Stokes drift. This effectively leads to a skill score of up to 0.90 over a long drift period of 65 hours. Two different cases of deployment of buoy clusters were conducted to investigate the dispersion. The simulated dispersion indicators at two different resolution of the ocean model are compared to those computed from the observations. In the first case of study, the relative and absolute dispersion are either overestimated or underestimated by the drift models while, in the other case, the drift model using the extrapolated current at low horizontal resolution severely underestimates these two dispersion indicators. The dispersion was well estimated when increasing the horizontal resolution of the ocean model. The relative dispersion and relative diffusivity show two different dispersion regimes. A non-local dispersion regime for separations smaller than 3.5~km for the first cluster of drifters and 500~m for the second cluster of drifters. Beyond these scales the dispersion is local. -- Mot(s) cle(s) en anglais : surface drift forecast, Ekman layer, drifting buoys, shear extrapolation, Stokes drift, submesocale, dispersion.