PARALLEL OCEAN GENERAL-CIRCULATION MODELING

We have developed a global ocean model on the massively parallel CM-2 Connection Machine based on the Bryan-Cox-Semtner ocean general circulation model. This paper discusses the Connection Machine implementation of the model and its performance, as well as major changes that were made in the algorithms and numerical methods. In order to increase the parallel efficiency, we reformulated the barotropic equations to solve for the surface-pressure field rather than the volume-transport streamfunction. In addition to being more efficient, this method has several other advantages over the streamfunction formulation: (1) any number of islands can be included in the computational grid at no extra cost; (2) the model can handle steep gradients in the bottom topography; and (3) the free-surface height is a prognostic variable, which simplifies assimilation of altimetric data into the model. We have also developed a new parallelizable preconditioning method for the solution of the elliptic equation for the surface-pressure field. In order to use a standard conjugate gradient algorithm, an approximate-factorization technique is used to split off the Coriolis terms and obtain a symmetric operator. We then apply a symmetric local approximate-inverse operator as a preconditioning matrix, which is very effective in accelerating convergence to a solution,