Four-dimensional variational data assimilation based on the Global Environmental Multiscale (GEM) Model (see Cote et al, 1998) is scheduled to replace the three-dimensional variational data assimilation system at the Canadian Meterological Centre in Autumn 2004 (see Gauthier et al., 2004). The GEM model and its adjoint were recently adapted to a hybrid vertical coordinate and redesigned for a parallel distributed-memory implementation with explicit message passing.

Work is currently underway to extend the four-dimensional variational assimilation to the new non-hydrostatic limited-area capability of the GEM model. As a first step, the adjoint was developped. As a second step, tests will be performed using identical twin experiments so as to understand the impact of the boundary conditions on different assimilation problems. We will consider 2D idealized experiments in X-Z such as bubble convection and stratified flow over mountain, as well as 3D experiments. The results of these experiments will be presented.