Abstract
The arbitrarily structured C grid, Thuburn–Ringler–Skamarock–Klemp (TRiSK), is being used in the Model for Prediction Across Scales (MPAS) and is being considered by the Met Office for their next dynamical core. However, the hexagonal C grid supports a branch of spurious Rossby modes, which lead to erroneous grid-scale oscillations of potential vorticity (PV). It is shown how these modes can be harmlessly controlled by using upwind-biased interpolation schemes for PV. A number of existing advection schemes for PV are tested, including that used in MPAS, and none are found to give adequate results for all grids and all cases. Therefore a new scheme is proposed; continuous, linear-upwind stabilized transport (CLUST), a blend between centered and linear-upwind with the blend dependent on the flow direction with respect to the cell edge.
A diagnostic of grid-scale oscillations is proposed that gives further discrimination between schemes than using potential enstrophy alone. Indeed, some schemes are found to destroy potential enstrophy while grid-scale oscillations grow.
CLUST performs well on hexagonal-icosahedral grids and unrotated skipped latitude–longitude grids of the sphere for various shallow-water test cases. Despite the computational modes, the hexagonal icosahedral grid performs well since these modes are easy and harmless to filter. As a result, TRiSK appears to perform better than a spectral shallow-water model.