Structural Uncertainty of Tropical Cyclone Simulations in General Circulation Models

High-resolution General Circulation Models (GCMs) with grid spacings around 30 km begin to represent strong tropical cyclones of categories 4 and 5. Although such GCM grid spacings still miss many small-scale processes, these simulations exhibit rather realistic features like a warm core, a distinct eye or slanted eye wall. This raises the question how robust these characteristics are with respect to the many sources of uncertainty in GCM experiments. These are the initial data, parameter and structural uncertainty. In particular, structural uncertainty arises due to the choice of the physical parameterizations or the dynamical core. The latter describes the fluid flow component of GCMs and determines the equation set, the computational grid, numerical methods and diffusion mechanisms.

The paper takes an in depth-look at the structural uncertainty of tropical cyclone simulations in NCAR's Community Atmosphere Model CAM 5.1 that arises from its dynamical core. CAM 5.1 offers five dynamical cores which include the current default Finite-Volume (FV) algorithm on a latitude-longitude grid, the future default Spectral Element (SE) model, as well as the optional Eulerian (EUL) and semi-Lagrangian (SLD) spectral transform methods and GFDL/NASA's FV model on a cubed-sphere grid. We analyze the structural uncertainty of deterministic ten-day simulations via idealized tropical cyclone simulations in aqua-planet mode. Using analytically prescribed initial conditions and identical physics packages, we spin up a single tropical-cyclone-like vortex and reveal the spread in the strength and structure of the storm. The choice of the dynamical core and its diffusion mechanism severely impacts the shape and strength of the cyclone. In particular, models with local numerical discretizations, like FV and SE, develop storms with much greater intensities in comparison to the rather weak EUL and SLD simulations which utilize a global numerical method. This sensitivity might be underappreciated in the GCM modeling community. The results are even robust if simplified physical parameterizations are used in place of the CAM 5.1 physics suite. The structural uncertainty is quantified and the role of the numerical schemes discussed.