Comparison Plexrt Longwave Radiation in ICON

Tue 12 November 2019

Felix found a weird discrepancy in fluxes between the RRTMG solver and the Tenstream solvers which he used in his WRF experiments. For his purposes, it was not really necessary to have the solver to be perfectly tuned so, after we couldnt immediately find the reason behind it, we just let it be. Now, I recently had a discussion about the implementation of another radiative transfer package in ICON and they found (maybe) similarly shaped differences when they compared their results against the RRTMG solver.

This is an excellent chance to delve into the issues again and maybe find an error in our use of the RRTMG optical properties.

Anyway, as a first step, here we have the output of mean(over all columns) longwave up and downwelling fluxes from the first timestep of a limited area setup with ICON.

In the top panels we have the complete profile from the surface(at index 0) up to the top of the model domain at index 151. The lower panels zoom in on the first couple of levels near the surface.

Lets first compare the orig rrtmg version against the plexrt rrtmg:

  • There are large discrepancies at the top of the dynamics model. This is probably because the original RRTMG neglects the upper atmosphere above the ICON mesh, whereas the TenStream solvers glue a background profile(US standard atmosphere in this case) on top of the ICON grid. One particular indicator is the fact that Edn at the top is zero in the case of orig rrtmg. This suggests there is no atmosphere above

  • Given above point, it is no surprise that we have an offset in downwelling irradiance that penetrates as far as the surface. While this could also come from wrongly implemented aerosol, ice clouds, traces gases or all of it together, we will have a look at this another time.

  • On the other hand, if we look at upwelling thermal radiation at the top of the domain, the discrepancies suggest that there is indeed something not yet fully implemented. This could again be trace gases(e.g. co2, ch4 n2o and o2 are not yet fully coupled), but this is not the real issue that we want to keep an eye on at this moment.

  • We should note that the skin temperature and the optical properties in the lower atmosphere are probably implemented correctly because the upwelling radiation above the surface agrees very well between the the rrtmg versions

Now, if we compare the rrtmg solvers against other 1D solvers in the Tenstream lib:

  • The schwarzschild solver(with 3 angles) compares very well against the disort version(16 streams) which is expected because the optical properties from RRTMG do not support scattering in the thermal spectral range
  • The TwoStream shows differences compared to disort computations but what to do about it? We usually use schwarzschild anyways because it is faster.
  • There is a different behavior for all the TenStream 1D radiation solvers for upward irradiance near the surface. This is where I am particularly curious if this also happens with other radiation packages.
  • This could either be a off by one error on my part or there is something weird/fancy going on in RRTMG

For now, this is it. I hope I can get back and disentangle the issues with the help of new radiation implementations from within ICON.