Evaluating Cloud Properties Simulated by Largescale Models with Groundbased Data

Ground-based observations, such as those derived from Cloudnet (Illingworth et al. 2007), or ARM-instrumented sites (Mather et al. 1998) have proved useful in evaluating models. Although they lack the global coverage of satellites, they have the advantage of greater spatial and temporal resolution with a more powerful array of remote-sensing instruments. Provided that analysis is restricted to times when winds are high enough to ensure that sufficient amounts of clouds advect past the sensor and a reasonable cross section of the model grid box is sampled, valid comparisons with the model can be made every hour. The Cloudnet study of seven operational models over one year showed that the representation of clouds in a given grid box over the observing site was surprisingly good, but particular biases could be identified. The vertical profile of mean cloud fraction revealed that all models underestimated the occurrence of mid-level cloud. Mean ice water content profiles in the models showed good agreement with observations, and more recent versions of the models captured the observed mean liquid water content well. It is interesting to note that the performance of a mesoscale (12 km resolution) model was not notably better than the same model when run at a global scale with 60 km resolution, although a fairer test would be to carry out the comparisons at the same scale by aggregating the 12 km model data up to the 60 km resolution. It is important to note that the models carry the correct mean values, but this is not the whole story. The PDF of cloud fraction showed that models have fewer completely filled grid boxes than observed and more partially filled grid boxes. Model PDFs of liquid water content were more peaked than observations. The PDFs of ice water content revealed that the one model that had the worst mean value below 7 km had actually the best PDF below 0.1 g m-3; however, because any higher ice water content was considered to be falling snow rather than cloud, the result was a mean value of ice water content that was far too low. Most models have low-level water clouds that drizzle all the time, with the drizzle reaching the ground; observations show the same mean drizzle rate, but a completely different PDF, with occasional bursts of heavier drizzle reaching the ground but usually much lighter drizzle, which evaporates 100 m or so below cloud base. Ground-based studies have also shown (Hogan et al. 2000) that the common assumption of maximum random overlap of clouds is appropriate for clouds shallower than 2 km, but should be modified so that as the clouds become deeper the overlap tends towards maximum. These Cloudnet results, which show that the NWP models have considerable skill in producing clouds at the right time in the right place, suggest that the models are capturing the fundamental meteorological processes that produce the clouds and are correctly locating the regions of ascent. These encouraging results suggest that the assimilation of clouds within NWP models may be feasible, and that such studies could lead to improvements in parameterization schemes. In addition, the success of NWP models give us confidence that climate models should also be representing clouds reasonably well, since they are using essentially the same cloud parameterization schemes.

This example shows how analyzing observations, both in terms of mean values and PDFs, helps to elucidate the processes responsible and should greatly help to improve the physical basis of statistical cloud schemes and to reduce the degree of empiricism in them. Further analysis should be undertaken. This could, for example, investigate if the implicit ice particle sizes in the models are correct, examine if the lack of mid-level clouds in the models is important, and establish the scale and relevance of the errors in the representation of drizzle in low-level clouds. Given the critical uncertainties associated with trade-cumulus clouds (see above), the analysis of long-time series of ground-based data collected in regions covered by such clouds would be very beneficial. Aspects to be investigated would be the values of liquid water content and liquid water path, cloud base, cloud top as a function of the depth of the boundary layer, the formation of any precipitation including small drizzle droplets, and their subsequent fate as they fall below cloud to evaporate or on occasion to reach the ground. A sensitive high-resolution (e.g., 6 m/30 s) ground-based Raman lidar should provide detailed observations of the PDF of water vapor within the boundary layer, which can be combined with the liquid water content within cloud, to provide the observed PDF of total water content.

This PDF, which is the fundamental basis for statistical cloud schemes, can then be compared with that predicted by models. The deployment of the mobile ARM facility in the Azores, from April to December 2009, to observe the springtime overcast stratocumulus regime and the summertime broken trade cumulus, should be particularly fruitful. It may well be that the observations made in recent fi eld projects, such as RICO and BOMEX, are also able to furnish some of the data required to see if climate models are simulating such clouds realistically.

Recent observational campaigns with advanced multiple wavelength and depolarization lidars may provide more information on the ability of aerosol particles to influence cloud properties. In particular, it seems that the size of aerosol particles can be inferred from the lidar backscatter and/or extinction spectrum (Müller et al. 2000, 2001) and the shape from the depolarization ratio, and that large non-spherical particles may act as efficient ice nuclei and promote glaciation. There is some evidence that Saharan dust may be a source of ice nuclei and that when the dust is lofted to high altitudes, such ice nuclei can cross the Atlantic (e.g., DeMott et al. 2003; Ansmann et al. 2008). The degree to which such dust particles promote the glaciation of supercooled clouds and how often this occurs is still questionable, but this is an area of active research which should yield quantitative results. Incorporating such phenomena into climate models will be difficult; the degree to which Saharan dust is lofted is dependent upon the performance of the transport model and the gustiness of the surface winds—a local effect which will be difficult to capture reliably in large-scale models.

0 0

Post a comment