Historically, two established classes of numerical weather prediction models are coarse global circulation models (GCMs) and higher resolution limited area models (LAMs). GCMs are inhibited by their inability to resolve important smaller-scale features, while LAM forecasts are sensitive to the larger-scale, downscaled initial and boundary conditions. The non-hydrostatic atmospheric core of the Model for Prediction Across Scales (MPAS-A) falls into a new class of models bridging this gap. Local horizontal refinement allows for desired resolution in regions of interest, while a smooth transition to a relatively coarse surrounding mesh increases efficiency and reduces the numerical complications that typically arise from nesting approaches. Understanding the benefits and limitations of this new class of models is fundamental.
Towards this understanding, we examine the predictability and sensitivities for forecasts of the longest lived Arctic TPV known to date, lasting 86 days in the summer of 2006. An ensemble of simulations leverages varying regions of refinement, vertical spacing, and physics parameterizations. Using a novel tracking algorithm to define TPVs objectively, simulations achieve an extended range of TPV predictability. Degradation of skill in both vortex track and intensity is clearly associated with mesh coarsening in the Arctic rather than lower latitudes. Coarser lower latitudes lead to distinct differences in wave amplification. An analysis of physics tendencies and potential vorticity budgets within these features reveals that the physics parameterizations are the main sources of sensitivities between the simulations, highlighting the limitations of schemes that do not scale consistently with the underlying mesh. The implications of these sensitivities for both atmospheric and coupled (atmosphere-land-ocean-sea ice) simulations are discussed.