Radar group home page Who's who in the radar group About Chilbolton Research projects Field experiments Quicklooks of radar data Real-time cloud radar and lidar images Radar Group publications Analysis software and documentation Radar links Back to Meteorology Department Back to University of Reading

Last update
5 Apr 2001

This page is maintained by
Mark Fielding

All content Copyright © The University of Reading unless otherwise stated.

Comparison of ECMWF cloud fraction with radar derived valued

Robin Hogan, Christian Jakob and Anthony Illingworth


  • In addition to the usual prognostic variable of cloud water content, global circulation models carry a value for cloud fraction which can be either diagnostic (e.g. UK Met Office Unified Model) or prognostic (e.g. ECMWF model).
  • This parameter is important for the model radiation budget but previous validation attempts using satellite data have suffered from poor vertical resolution, so only total cloudiness (i.e. vertically integrated) could be validated, not the cloud fraction at every model level.


  • The high vertical resolution of radar makes it the ideal tool with which to validate the model cloud fraction. The principle is simple - radar time height sections are divided up into boxes centred on the model levels and 1 hour in duration. Cloud fraction is then simply the fraction of pixels within the box that are cloudy.
  • The lidar ceilometer is very sensitive to liquid water clouds and in particular, is able to locate cloud base in the presence of rain or drizzle, so is used to modify the cloud fraction derived from the radar.


  • We use the 3 months of data taken by the 35 GHz Rabelais and 94 GHz Galileo during the Cloud Characteristics campaign. Quicklooks of the radar and lidar data, together with the retrieved and modelled cloud fractions for each day, can be found here.
  • An example of the comparison for a week of data is shown above.
  • This is then used to compile a climatology for comparison. The first panel below shows mean cloud fraction versus height for the model and the observations - one can see that the model tends to underestimate cloud fraction at mid-levels and overestimate it at high levels. However when this is divided into a `frequency of occurrence' (when cloud fraction is greater than 0.05) and an `amount when present' we see that the model is good at saying when there will be some cloud in a gridbox, but it tends to get the amount wrong.


  • Comparison of ECMWF winter-season cloud fraction with radar-derived values
    Hogan, R. J., C. Jakob and A. J. Illingworth, 2001, J. Appl. Meteorol., 40(3), 513-525. (Dowload from the Publications page)

See also