Evaluating Forecasts of Central US Mesoscale Convective Systems in a GCM with Explicit Embedded Convection

Recent work has demonstrated that the Central US organized nocturnal eastward propagating mode of convection can be captured in a prototype global climate model (GCM) which uses embedded cloud resolving models (CRM) instead of statistical parameterizations to handle sub-grid convection (superparameterization - SPCAM). This is a surprising result since the CRM in SPCAM is idealized in two dimensions with periodic boundary conditions, which restricts fast manifold storm propagation mechanisms to the horizontal scale of a GCM grid box and constrains CRM shear organization to a fixed horizontal plane. Nonetheless, these simulated storms have been qualitatively characterized as having realistic propagation speeds, and have been argued to result from slow manifold pathways linking large-scale dynamics with a prognostic convective lifecycle (CRM memory). However, rigidities in the technique (fixed CRM orientation) and apparent discrepancies in the simulated storm relative flow structure are unresolved issues requiring closer analysis. The Cloud-Associated Parameterizations Testbed (CAPT) approach to evaluating GCM error statistics in forecast mode is applied here to test the sensitivity of this result to CRM orientation and to quantitively compare against observed storm systems that passed over high value ground based sensors at the ARM SGP site. A Newtonian relaxation "nudging" technique has been developed to initialize both resolved scales in SPCAM (GCM and 2D CRM) for forecast simulations, spinning up the CRM by forcing the outer model to follow analyzed observations. Propagating storms are captured in SPCAM with zonal, meridional, and southeast CRM orientations, but comparisons to observations reveal many features of the storm location, thermodynamic structure, and condensate fields to be overly simulated or mispositioned.