MOC Atmospheric Model Physics
A model’s physics options are presented as a list of independent choices, and they are not independent. This map is organised around the couplings, because the errors that cost me the most time have come from the interfaces between schemes rather than from any scheme being wrong.
The core dependency chain runs: the land-surface model decides how energy splits between heating and evaporation, surface-layer schemes convert that into fluxes using Monin-Obukhov similarity, PBL schemes mix them upward, convection schemes act on the resulting instability, and radiation responds to the clouds all of that produces — then feeds back into the surface energy budget where the chain started.
Because it is a loop, a change anywhere propagates everywhere. Changing the convection scheme without checking what the radiation scheme assumes about convective cloud fraction is the single most reliable way to acquire a temperature bias with no obvious source.
Convection
The choice here is between single-closure schemes and ensemble ones. Kain-Fritsch is the common default and fires too early, which shows up as rainfall peaking near local noon; KF-CuP was built to fix exactly that by tying the trigger to boundary-layer turbulence. The Grell family takes the other route, averaging over many closures: Grell-Devenyi established the approach, Grell 3D added subsidence spreading, and Grell-Freitas made the mass flux scale-aware for nests spanning very different resolutions. New Tiedtke is a further alternative, and stochastic parameter perturbation treats the uncertainty explicitly rather than averaging it away.
The comparison table is the place to start. Below about 4 km none of this applies — see convection-permitting modeling.
Boundary layer and surface
The WRF PBL options divide into local and non-local closures, with MYNN and YSU the usual representatives of each. Whichever is chosen constrains the surface-layer scheme that can accompany it — MM5 similarity is the standard pairing — and both sit on top of Noah or a relative. Soil moisture initialisation matters more here than the scheme choice does.
Radiation and microphysics
RRTMG handles both shortwave and longwave through the correlated-k method and is the usual choice. What it produces depends entirely on the cloud field it is handed, which comes from microphysics — WSM6 being a common option — and from the convection scheme. In convection-permitting runs the microphysics sets rainfall intensity directly, so its weight in the configuration increases substantially.
Putting a configuration together
The tropical WRF setup is the configuration I settled on for the Galapagos, and the DARWIN experiment suite is the set of runs that tested the choices rather than assuming them. That distinction is the point: a scheme that performs well over the continental United States has no obligation to perform well over a tropical archipelago, and the only way to find out is to vary it.