RRTMG

RRTMG is the Rapid Radiative Transfer Model for GCMs, a radiation scheme covering both shortwave and longwave and the usual choice in WRF. It is the accuracy-oriented option in a field where the alternative is always a cheaper approximation.

Its method is the correlated-k approach. Exact radiative transfer means integrating over millions of spectral absorption lines, which is impossibly expensive. Correlated-k reorders wavelengths by absorption strength rather than by wavelength, exploiting the fact that what matters for the integral is the distribution of absorption coefficients, not which wavelength each belongs to. The reordered function is smooth and can be integrated with a few hundred points rather than millions. It is a genuinely clever piece of numerics and it is why RRTMG achieves near line-by-line accuracy at a tractable cost.

What using it commits you to

Using the same scheme family for both shortwave and longwave is worth doing deliberately. Mixing schemes means the two halves of the energy budget make different assumptions about gaseous absorption and cloud optical properties, and the resulting inconsistency is not visible in any single diagnostic — it shows up as a top-of-atmosphere imbalance that is hard to attribute.

RRTMG is expensive enough that it is called every 10-30 model minutes rather than every step. That is normally fine and occasionally not: too long an interval aliases the diurnal cycle, and it interacts badly with fast-evolving cloud. It is worth setting from the run’s timescale rather than leaving at a default chosen for a different application.

The dependency that actually determines the result

RRTMG’s accuracy in clear sky is excellent and largely irrelevant, because real skies have clouds and clouds dominate. The scheme does not compute cloud — it receives cloud fraction and condensate from microphysics and from the convection scheme via convective cloud fraction.

So the practical accuracy of radiation in a run is set by the cloud fields, not by the radiation scheme. Two failure modes recur: condensate counted twice when the cumulus scheme and the microphysics both report it, and the cloud overlap assumption — maximum, random, or maximum-random — which lives in the radiation code, is rarely discussed, and substantially changes total cloud cover for deep convective columns.

Choosing RRTMG and not checking those is a common way to have a well-validated scheme produce a poor answer.

See also: radiation schemes generally, microphysics, the tropical WRF setup, and the model physics map.