Open Thread

In the time-step-iterated earth system models, the radiative transfer code is executed separately from the physics operations for mass and energy flow. I realized this fact some time ago as I was reading about the NCAR CAM5 atmospheric model. I followed up recently to better understand why this is so.

My prompt to Grok 4.6 on 9-27-2026:
“About the earth system models that are used for CMIP scenarios please describe how the mass and energy flow computations at each time step are necessarily completed separate from the execution of the radiative transfer component of the model. I am interested in understanding the nature of the sequential processing which must occur so as to compute new shortwave and longwave radiative fluxes for continuing the iteration.”

Here is the first paragraph from the response from Grok 4.6. Please read the full conversation captured in the pdf linked below it:
“Earth-system models used for CMIP simulations advance the atmosphere by operator splitting: the dynamical core and most physical parameterizations update mass and energy on a relatively short time step, while the full radiative-transfer calculation is performed far less often and then held (or only weakly updated) until the next radiation call. The two calculations cannot be performed simultaneously on the same instantaneous state because radiation is both computationally expensive and depends on a consistent thermodynamic and cloud field that itself results from the preceding mass- and energy-flow steps.”

https://drive.google.com/file/d/1zjB9oZfPiEuQvByzkplbg3hlLksiKusQ/view?usp=sharing

So what? The execution of the radiative transfer component of the model is therefore best understood as a “freeze-frame” computation from the skin surface up to space through the modeled layers of the atmospheric column, using the simulated physical state conditions from the most recent iteration.

This is the same, in concept, as running the static-atmosphere MODTRAN model manually at each radiative transfer execution, using a set of cloud, surface temperature, and atmospheric selections, including CO2 and CH4 concentrations. The resulting “looking up” IR flux in W/m^2 is applied as one of the surface energy inputs. A higher CO2 concentration will always give a higher number.

The CMIP/IPCC models therefore “run hot” – some more than others – because they return a built-in “warming” trend with rising concentrations of CO2, CH4, N2O. They operate in these scenarios – by construction – as a stepped-static energy accumulator as the iteration proceeds, so as to make long-term sensible heat gain the only possible result on land, in the oceans, and in the lower atmosphere . This applies to hindcasts of the past and to projections into the future. Or in the case of a pulse exercise, a more rapid warming response occurs, modified by thermal storage.
The premise of the modeling exercise is that the computed static radiative “warming” tendency, from rising concentrations of the non-condensing IR-active trace gases, operates as a climate system “forcing” to which amplifying “feedbacks” then arise.
  
But that framing is based on assumptions. NO ONE KNOWS those starting assumptions to be valid in the proper context of the observed circulation. Cloud formation and dissipation within that circulation are the dominant influence on the independently observed longwave emission to space. Meanwhile, the power intensity of dynamic energy conversion, as computed within the ERA5 reanalysis, is the overwhelming influence on the vertical disposition of the energy involved in the incremental radiative “warming” tendency. And let’s not forget how the latent energy of water vapor makes horizontal and vertical energy transport so powerful within the circulation.  

For these reasons, among others, the entire time-step-iterated CMIP/IPCC modeling exercise is inherently circular. The simulations animate the premise but cannot justify it.

Thank you for your patience in this matter.

P.S. Here’s something more to think about. There could well be a set of parameter-tuning values for cloud properties, for radiative transfer itself, and for representation of sub-grid-scale processes, etc., that would make one of these time-step-iterated models completely insensitive to rising CO2, even with the execution of the radiative transfer code separated from the physics steps. That would be an interesting design-of-experiments exercise.