Comparison of ECMWF winter-season cloud fraction with radar-derived values

Of great importance for the simulation of climate using general circulation models is their ability to represent accurately the vertical distribution of fractional cloud amount. In this paper, a technique to derive cloud frac tion as a function of height using ground-based radar and lidar is describe d. The relatively unattenuated radar detects clouds and precipitation throu ghout the whole depth of the troposphere, whereas the lidar is able to loca te cloud base accurately in the presence of rain or drizzle. From a direct comparison of 3 months of cloud fraction observed at Chilbolton, England, w ith the values held at the nearest grid box of the European Centre for Medi um-Range Forecasts (ECMWF) model it is found that, on average, the model te nds to underpredict cloud fraction below 7 km and considerably overpredict it above. The difference below 7 km can in large part be explained by the f act that the model treats snow and ice cloud separately, with snow not cont ributing to cloud fraction. Modifying the model cloud fraction to include t he contribution from snow (already present in the form of fluxes between le vels) results in much better agreement in mean cloud fraction, frequency of occurrence, and amount when present between 1 and 7 km. This, together wit h the fact that both the lidar and the radar echoes tend to be stronger in the regions of ice clouds that the model regards as snow, indicates that sn ow should not be treated as radiatively inert by the model radiation scheme . Above 7 km, the difference between the model and the observations is part ly due to some of the high clouds in the model being associated with very l ow values of ice water content that one would not expect the radar to detec t. However, removal of these from the model still leaves an apparent overes timate of cloud fraction by up to a factor of 2. A tendency in the lowest k ilometer for the model to simulate cloud features up to 3 h before they are observed is also found. Overall, this study demonstrates the considerable potential of active instruments for validating the representation of clouds in models.