Versions Compared


  • This line was added.
  • This line was removed.
  • Formatting was changed.


Most radiation schemes in weather and climate models use the "correlated k-distribution" (CKD) method to treat gas absorption, which approximates the integration over hundreds of thousands of spectral lines by N pseudo-monochromatic radiative transfer calculations, where N is in the range tens to hundreds. In principle, there is a trade-off to make: larger N means more accuracy and a wider range of atmospheric conditions can be simulated, but at greater computational cost. Unfortunately most CKD schemes are a black box: there is no way for the user to adjust N according to their application. The purpose of CKDMIP is to address these issues, and specifically:

  • To use benchmark line-by-line calculations to evaluate the accuracy of existing CKD models for applications spanning short-range weather forecasting to climate modelling, and to explore how accuracy varies with number of g-points in individual CKD schemes.

  • To understand how different choices in way that CKD models are generated affects their accuracy for the same number of g-points.

  • To provide freely available datasets and software to facilitate the development of new gas-optics models, with the ultimate aim of producing a community tool to allow users to generate their own gas-optics models targeted at specific applications.

Potential participants are welcome to email Robin Hogan outlining what submissions they envisage being able to provide. See the results so far from the participating CKD tools and models.

Recent updates

The absorption spectrum of the nine gases considered in CKDMIP for concentrations at 100 hPa in the CKDMIP median atmospheric profile. Note that the wavenumber scale changes from linear to logarithmic at 2500 cm-1.


More information on the design of the experimental protocol is here:


  • ckdmip-1.0.tar.gz: Software package for performing longwave radiative transfer calculations both on line-by-line absorption spectra and CKD files produced by CKDMIP participants, as well as converting LBLRTM output files to NetCDF (June 2021)
  • ckdmip-0.9.tar.gz: Older version (March 2020)


The data are freely available here:

  • (multiple files can be downloaded with wget)
  • Via read-only FTP to (username: ckdmip, password: any non-empty string) NEW ADDRESS PENDING

Any publications using the data for any purpose should cite Hogan and Matricardi (2020).


Table 3 of Hogan and Matricardi defines four datasets consisting of a number of gas-concentration profiles plotted here:

Absorption spectra

The absorption spectra used to generate reference radiation calculations were produced at ECMWF by Marco Matricardi using the LBLRTM line-by-line model, and are available in NetCDF4/HDF5 format.  The volumes are as follows:

  • Evaluation-1: longwave 272 GB, shortwave 140 GB
  • Evaluation-2: longwave 228 GB, shortwave 117 GB (not publicly released until the end of the CKDMIP project)
  • MMM: longwave 31 GB, shortwave 16 GB
  • Idealized: longwave 143 GB, longwave 76 GB

The total volume is 1019 GB (949 GiB). Water vapour spectra are available both with and without the continuum contribution for all datasets except Evaluation-2.

Intercomparison files

As described in section 4 of the technical guide, participants are requested to provide files containing the output from their CKD model. In the longwave this contains the optical depth and Planck function at g-point of their model, for each concentration scenario. Here is an example of this file, and the fluxes that the CKDMIP software package will produce from it:

In the shortwave, the file should contain the absorption optical depth, Rayleigh optical depth and solar irradiance at each g-point of the model, for example:

Participants are also requested to provide information on any CKD model they generate in terms of which points in the spectrum contribute to each g-point. The format should match the following:

Note that these particular files have been built without any knowledge of where individual g-points are weighted within a band, but this sub-band information will be needed to most accurately assess errors in cloudy profiles in the final phase of the project (section 4.4 of Hogan and Matricardi).

Contact and social media

Any queries not answered in the documentation on this page should be addressed to Robin Hogan (

Search Twitter for the hashtag #CKDMIP.