The New Dynamical Kernel of the Global Environmental Multiscale Model with Height-Based Vertical Coordinate

The dynamical kernel of the Global Environmental Multiscale (GEM) model, used operationally by Environment and Climate Change Canada (ECCC) for numerical weather prediction (NWP), employs a log-hydrostratic pressure-type vertical coordinate. Such a coordinate system permits the use of a direct solver for the discretized elliptic problem to resolve the dynamical component of the flow. However, for very high spatial resolution, e.g., for sub-kilometer horizontal grid spacing, the model is found to exhibit strong numerical instability – particularly over complex terrain where the mountain slopes can become substantially steep with increasing model resolution. With the growing demand for very high-resolution operational NWP systems, improving model stability over steep mountains has become an imperative. A number of approaches have been investigated to improve numerical stability of GEM with limited success that include increased off-centering in the discretized vertical momentum equation, a vertically-variable basic state temperature profile and modifications to the non-hydrostatic contributions in the linear system arising from the discretized GEM formulation.

The previous experience with the Mesoscale Compressible Community model at ECCC suggests that a dynamical kernel based on height-type vertical coordinate does not suffer from similar severe orography-induced instability. Furthermore, the advent of the new high performance computing infrastructure shows that the existing direct solver approach in GEM loses its scalability substantially for a very large number of processor cores. This implies that in order to take advantage of very large number of processor cores to integrate each model dynamical step, a highly optimized iterative solver will be essential in future. Such a requirement negates one of the most attractive aspects of the existing mass-based vertical coordinate system. All of these issues have motivated the development of a new dynamical kernel for the GEM model that utilizes a height-based vertical coordinate. Initial version of this new kernel is currently being tested for some theoretical benchmark cases including the well known bubble convection and nonhydrostatic mountain wave simulations. Work is in progress to implement the new coordinate for full three-dimensional atmospheric flow simulations. The new height-based dynamical kernel is planned for operational implementation in the near-future after all the necessary tests and validations are carried out.

Results pertaining to the newly-developed GEM dynamical kernel with respect to the theoretical test cases as well as three-dimensional atmospheric flow will be presented at the conference. The roadmap for the transition from research to operation for the new model will also be discussed.