New Numerical Algorithm for the Multi-Layer Shallow Water Equations Based on the Hyperbolic Decomposition and the CABARET Scheme

Purpose. The present article is devoted to describing a new method of numerical solution for hydrostatic approximation of incompressible hydrodynamic problems with free surfaces and variable density. Methods and Results. The algorithm is based on the hyperbolic decomposition method, i. e. representation of a multilayer model as a sum of the one-layer models interacting by means of the reaction forces through the layers’ interfaces. The forces acting on the upper and lower interfaces of each layer are interpreted as the external ones which do not break hyperbolicity of the equations system for each layer. The explicit CABARET scheme is used to solve a system of hyperbolic equations with variable density in each layer. The scheme is of the second approximation order and the time reversibility. Its feature consists in the increased number of freedom degrees: along with the conservative-type variables referred to the centers of the calculated cells, applied are the flux-type variables related to the middle of the vertical edges of these cells. The system of the multilayer shallow water equations is not unconditionally hyperbolic, and in case hyperbolicity is lost, it becomes ill-posed. Hyperbolic decomposition does not remove incorrectness of the original system of the multilayer shallow water equations. To regularize the numerical solution, the following set of tools is propose: filtration of the flow variables at each time step; super-implicit approximation of the pressure gradient; linear artificial viscosity and transition to the Euler-Lagrangian (SEL) variables that leads to the mass and momentum exchange between the layers. Such transition to the SEL variables is the basic tool for stabilizing numerical solution at large times. The rest of the tricks are the auxiliary ones and used for fine tuning. Conclusions. It is shown that regularizing and guaranteeing the problems’ stability requires not only reconstruction of the computational grid at each time step, but also application of the flow-type variables’ filtering and the artificial viscosity simulating turbulent mixing.