My research lies in the realm of preparing global climate simulations for the era of petascale and exascale scientific computing.

Resolution Sensitivity

For global atmospheric models, gains in computational power often translate into explicitly resolving scales that were once parameterized. Assumptions inherent in these parameterizations tend to break-down at finer scales, requiring a reformulation of the grid-scale closure. I run idealized experiments to isolate underlying causes of “scale-incognizant” behavior that manifest in global simulations.

One approach is to analyze the convergence properties of global atmospheric models. In a perfect world, the model solution would converge as the grid spacing is reduced. In reality, this does not always occur. This is because the equations governing the resolved fluid flow have inherent scale-dependencies at resolutions typical of global atmospheric models. I hypothesize (Herrington and Reed 2017; Herrington and Reed 2018) that convergent solutions may then be recovered through regulation of the buoyancy forcing passed to the dynamical core by the physics package.

Perhaps the biggest problem with non-convergent AGCM solutions is that it hinders our ability to use variable resolution grids (i.e. global grids with regional grid refinement). An example is provided below, which shows the solutions of an aqua-planet experiment (an all ocean covered planet) in which the mesh containing the right equatorial panel has been refined by a factor of 8. One can clearly spot the higher precipitation rates (color bar not shown) in the refined region, which leads to asymmetric heating across the refinement region and dynamical consequences.

Reduced Complexity Configurations

I am also interested in reduced complexity test cases, such as deterministic tests and radiative-convective equilibrium tests. State-of-the-art climate models are extremely complex. Reduced complexity test cases have the potential to more easily isolate underlying causes of behavior in the models. My emphasis is on understanding the interaction between dynamical cores and their physical parameterizations.

I’ve developed an idealized test case (Herrington and Reed 2018) to help understand the resolution sensitivity that occurs in more complex configurations. It is simply a moist bubble initialized at the equator of a non-rotating planet, with the rest of the planet having the same vertical temperature profile resembling the Tropics. The test case is performed with moist physics routine of varying complexity. The experiment is repeated with the bubble radius and grid spacing reduced by the same factor, mimicking the effect of increasing the models resolution. What follows is that the vertical motion scales like one would expect from the dry equations of motion; inversely proportional to the grid spacing. However, the scaling is very sensitive to the choice of physics time-step.

Physics-Dynamics Coupling

High-order finite-element methods are becoming popular in the weather and climate modeling community, but their unique numerics challenges the conventional physics-dynamics coupling paradigm. How should these models couple with the physics, or other components of the model for that matter, when the dynamical core has no formal definition of a control volume at the grid-scale? Herrington et al 2018; 2019 explore these issues in detail, and develop a method in which the physics are evaluated on a separate, a finite-volume grid. Compared with prior attempts to define control volumes in a high-order finite-element model (left panel), the finite-volume physics grid is approximately isotropic (middle panel), and more similar to the conventional latitude-longitude grids (right panel) historically used in weather and climate modeling.