We propose to develop, test, and evaluate two methods for correcting severe anomalies in simulated convective activity and precipitation in numerical models with spatially-varying grid spacing. These anomalies have been demonstrated in previous studies and arise from multiple causes. Among these is that the ability of convective motions to develop on the model grid is strongly dependent on resolution, such that convection occurs most readily at the finest resolution and is effectively suppressed at grid scales much larger than convective scale. Another principal cause occurs in cases where parameterized convection in the model is selectively applied only to regions of the grid where convective circulations cannot be resolved, while resolved convective motions and bulk microphysics parameterization represent convection and its effects in regions of the grid with sufficiently fine spacing. The transition from parameterized to unparameterized convection can result in artificial horizontal gradients and solenoidal circulations that favor convection on one side of the transition and suppress it on the other side. The enormous potential benefit of variable resolution grids and their growing popularity in global climate models make it imperative that these problems be corrected as soon as possible.

Our first proposed method is aimed primarily at controlling the response of the convective parameterization so that it more closely matches that of resolved convection with bulk microphysics parameterization in timing and intensity. Achieving a balance between these two disparate representations of convection in models can eliminate most of the lopsided competition for moisture and convective energy that can occur between the two spatial regions. The second proposed method is a unique form of "superparameterization" where a cluster of grid cell columns called a "convection grid" is inserted in coarse model grid cells where convection cannot be resolved but which have atmospheric environments that may be sufficiently unstable to support convection. Our proposed form of convection grid is horizontally unstructured and provides a mixture of high and medium resolution grid cells that permit resolution of principal convective updrafts and downdrafts and areal coverage of a broader region of compensating mesoscale subsidence with a reasonably modest total number of grid cells. This configuration provides 3D representation of convection and its impacts at a cost well below that of a uniform convection-resolving grid. Both proposed approaches have high potential for mitigating the largest convective anomalies that can result from spatially variable grid spacing and thus offer significant improvement to climate models that employ such grids.

These methods will be developed and tested as a component of the Ocean-Land-Atmosphere Model (OLAM) atmospheric model, whose global grid is configured with local mesh refinement to a high degree. The associated Fortran source code will be written in portable format that can be transferred to other modeling systems with relative ease. The new code, like OLAM itself, will be publically available, and we will offer assistance to other model developers who wish to apply the methods and/or source code.