Gulf Stream Separation

Gulf Stream Separation in General Circulation Models

Climate models currently struggle with the more traditional, coarse ( O(100 km) ) representation of the ocean. In these coarse ocean simulations, western boundary currents are notoriously difficult to model accurately. The modeled Gulf Stream is typically seen exhibiting a mean pathway that is north of observations, and is linked to a warm sea-surface temperature bias in the Mid-Atlantic Bight. Although increased resolution ( O(10 km) ) improves the modeled Gulf Stream position, there is no clean recipe for obtaining the proper pathway. 

The 70 year history of literature on the Gulf Stream separation suggests that we have not reached a resolution on the dynamics that control the current’s pathway just south of the Mid-Atlantic Bight. Without a concrete knowledge on the separation dynamics, we cannot provide a clean recipe for accurately modeling the Gulf Stream at increased resolutions. Further, any reliable parameterization that yields a realistic Gulf Stream path must express the proper physics of separation. 

The goal of this project was to determine what controls the Gulf Stream separation. To do so, I examined the results of a model intercomparison study and a set of numerical regional terraforming experiments. It is argued that the separation is governed by local dynamics that are most sensitive to the steepening of the continental shelf, consistent with the topographic wave arrest hypothesis of Stern (1998). A linear extension of Stern’s theory is provided, which illustrates that wave arrest is possible for a continuously stratified fluid.


J. Schoonover, W.K. Dewar, N. Wienders, and B. Deremble. Local Sensitivities of the Gulf Stream Separation. J. Phys. Oceanogr., 47:353–373, 2017

J. Schoonover et al. North Atlantic Barotropic Vorticity Balances and the Gulf Stream Separation in Numerical Models. J. Phys. Oceanogr., 46:289–303,


Gulf Stream Separation Dynamics (FSU Dissertation)

FSU RCC - Gulf Stream Separation Explored Through HPC

Model Runs

We are actively working through creating a series of downscaled runs focused on the Gulf Stream along the eastern US seaboard and downstream of its separation from the continental slope near Cape Hatteras, NC. Each execution of the MITgcm is assigned a simulation id and a simulation phase. 

The simulation id corresponds to the regional mesh configuration, boundary and atmospheric input deck sources, and the target subgridscale parameterization and time-stepping parameters. For each simulation id, the model is transitioned through multiple phases. Early in a simulation, the model interior goes through a period of adjustment where gravity waves and other spurious noise propagates through the numerical ocean. During these early phases, the time steps are usually smaller and the viscosity and diffusivity is artificially larger in order to quickly damp out the noise while maintaining numerical stability. We gradually change these parameters through various phases in the simulation to reach the target parameters for production runs. Once we reach production runs, multi-year model runs are executed to create sufficiently long datasets for driving the next down-scaling run.

Simulation ID : mitgcm50-z75

Simulation ID : mitgcm100-z75

Simulation Monitoring

As the MITgcm runs, we are regularly parsing the monitoring statistics and computing other time-series output. This data is posted to Big Query and visualized in Data Studio. This allows us to keep track of simulations running on Google Cloud Platform, NCAR's Cheyenne supercomputer, and a departmental cluster at Florida State University. 

The dashboard below is an interactive panel. You can change the date range and the simulation id and phase