Modeling turbulent dispersion on the North Flank of Georges Bank using Lagrangian Particle Methods

Circulation and transport at the North Flank of Georges Bank are studied using a data-assimilative 3-D model of frontal dynamics under stratified, tidally energetic conditions over steep topography. The circulation model was used in real-time during a cross-frontal transport study. Skill is evaluated retrospectively, relative to CTD, ADCP, drifter, and fluorescent dye observations. Hydrographic skill is shown to be retained for periods of weeks, requiring only initialization from routine surveys and proper atmospheric heating subsequently. Transport skill was limited by the wind stress input; real-time forecast winds taken from an operational meteorological model produced cross-isobath Ekman transport which was not observed locally. Retrospective use of observed local wind stress removed this cross-frontal bias. The contribution of tidal-time motion to the dispersion of a passive tracer is assessed using an ensemble of passive particles. The particle release simulates an at-sea dye injection in the pycnocline, which is followed for four days. Non-advective vertical tracer transport is represented as a random walk process sensitive to the local eddy diffusivity and its gradient, as computed from the turbulence closure. Non-advective horizontal tracer transport is zero for these ensembles. Computations of ensemble variance growth support estimates of (Lagrangian) horizontal dispersion. Off-bank, ensembles are essentially non-diffusive. As an ensemble engages the mixing front, its vertical diffusivity rises by 3 orders of magnitude, and horizontal spreading occurs in the complex front. The resultant horizontal dispersion is estimated from the ensemble variance growth, in along-bank and cross-bank directions. It is partitioned, roughly, between that contributed by 3-D advection alone, and that initiated by vertical diffusion. Engagement in the mixing front occurred in the forecast ensemble as a result of Ekman drift produced by an erroneous wind prediction. In the hindcast, observed wind left the ensemble non-diffusive and compact, advecting parallel to the mixing front and experiencing some advective shear dispersion. Lagrangian dispersion is event-specific and both simulations here represent credible events with dramatically different ecological outcomes. The skill metrics used are less sensitive, indicating that metrics tailored to surface-layer phenomena would be more appropriate in a data-assimilative context. The hindcast is closer to truth, based on first principles (better information). The level 2.5 closure used is realistic in the ocean interior; the near-surface processes need further refinement, especially as both surface- and bottom-generated turbulence affect these events strongly. (c) 2005 Elsevier Ltd. All rights reserved.