North American Extreme Weather in CESM2: Regionally-refined Simulations with Next Generation Dynamical Cores

Accurately predicting extremes, such as tropical cyclones and severe convective storms, in earth system models requires high resolution to resolve mesoscale dynamics at horizontal scales of O(10km). Simulating such resolutions globally for climate timescales (years to decades) remains infeasible. However, regional refinement can curtail required computational costs by applying resources in areas of interest while maintaining a globally-continuous solution. This work describes multi-decadal global simulations using variable-resolution meshes with both the Model for Prediction Across Scales (MPAS) and Spectral Element (SE) dynamical cores in a developmental version of the Community Earth System Model version 2 (CESM2). Here, a variable-resolution nest with a grid spacing of 14km is centered over the continental United States (CONUS).

When compared to baseline unrefined simulations (111km), the mean global climatologies produced by regionally-refined simulations for both CESM-MPAS and CESM-SE are similar, indicating that the technique possesses utility in improving resolvable scales locally while doing no harm to the global solution. We focus on simulated climatology of three CONUS extreme weather phenomena: tropical cyclones, northeastern U.S. snowstorms, and severe convective outbreaks, which are quantified using recently-developed objective detection techniques. Tropical cyclone structure and frequency improves with increased resolution and results imply that realistic patterns of CONUS landfalls can be captured with regional refinement. Individual northeastern U.S. snowstorms are also well-matched to historical observations, although subtle biases in near-surface temperature and precipitation phase highlight under-appreciated dynamical core sensitivities. Operational proxies used for severe weather forecasting (such as the Significant Tornado Parameter) show promise in detecting severe convective storm outbreaks in CESM2, although persistent large-scale biases in predicted column instability demonstrate that some large-scale errors in low-resolution CESM are not necessarily mitigated with finer grid spacing.