Local Galerkin methods make use of quadrature grids to exactly integrate basis functions over an element. Typically parameterization tendencies are based on the atmospheric state at these quadrature points. This leads to an non-equidistant 'sampling' of the atmospheric state and parameterizations are not given grid cell average atmospheric states (as, e.g., convection schemes are expecting). The functionality to run physics on an equal-area finite-volume grid inside each element in CAM-SE has been implemented. The motivation for this functionality was fourfold:

• The purpose of the quadrature grid is to exactly integrate polynomials over an element inside the spectral element fluid flow solver. It may be more physically consistent to integrate these polynomial representations of the atmospheric state over equal-area control volumes rather than sampling them an-isotropically in space at the quadrature points when passing the atmospheric state to the physics package.

• Noise in, e.g., vertical velocity, typically appears near element boundaries, where the globally continuous polynomials are only C^0. By using an equal-area finite-volume-type physics grid within each element, the vertical velocity is averaged over a control volume and not evaluated at the element edge.

• A physics grid supports the finite-volume multi-tracer scheme CSLAM (Conservative Semi-LAgrangian Multi-tracer scheme in CAM-SE) in that tracer transport can be computed on the same grid as the physics grid.

• Running physics on a finer or coarser grid will enable new science in CAM-SE. Using linear numerical analysis theory it is well known that the dynamical core has not converged at the grid-scale.

This advocates computing sub-grid scale tendencies on a grid coarser than the dynamical core grid so that only the 'converged' model state is forwarded to the sub-grid scale parameterizations (Williamson, 1999). Perhaps counter-intuitively running physics on a grid finer than the dynamical core grid resulted in improved forecast scores in the ECMWF IFS (Niels Wedi; personal communication). Clearly this paradigm needs further investigation. Preliminary results from Held-Suarez simulations using the physics grid in CAM-SE will be presented.