This page evaluates CKD models generated by the PyKdis tool. Note that only present-day conditions are currently evaluted. The evaluates are performed against line-by-line calculations on the CKDMIP "Evaluation-2" dataset.


Two CKD models have been generated, one using the 13-band CKDMIP "narrow" band structure with 476 g-points, and the other using the 5-band CKDMIP "wide" band structure with 235 g-points. (These are reduced from original "narrow" and "wide" CKD models with 3211 and 1840 g-points, respectively.) The results for present-day conditions are as follows:

The evaluation of these two models in individual bands is shown next:

In summary:

  • Heating rates in the troposphere and lower-to-mid stratosphere (4-1100 hPa) are quite accurate.
  • Heating rates above this (lower pressures) are significantly underestimated, which is due to inadequate handling of the line centres of the 2.8 and 4.3 micron CO2 bands associated with the very steep rise of higher absorption in the cumulative distribution. Note that the error is significantly less for the "original" CKD models with 3211 and 1840 g-points.
  • O3 heating rates in the UV bands appear to be well captured.
  • Downwelling fluxes in the stratosphere are a little high because of the missing upper atmosphere absorption in the near IR.  There then seems to be too much absorption or scattering in the lower atmosphere because the surface downwelling fluxes are biased low (and there is a lot of scatter).

  • Upwelling fluxes tend to be biased high, most noticeably for the "wide" band structure, and is mostly associated with the visible and UV bands.  This is because Rayleigh scattering uses a fixed value per band rather than per g-point, leading to an overestimate of the scattering.

Write a comment...