In this paper a physically based method is used to retrieve cirrus cloud optical thickness at 0.87 microns, ice crystal effective diameter using the 1.6 and 3.7 micron channels simultaneously, and cloud-top pressure (cloud-top temperature) primarily using the 10.8 and 11.9 micron channels. The method is based on Optimal Estimation (OE). Rodgers has previously described the OE method applied to the inverse problem in the context of retrieving atmospheric temperature and humidity from infrared radiometers for assimilation into NWP models. The OE method although routinely used in retrieving atmospheric clear sky profiles is rarely applied to cloudy retrievals. The OE method attempts to combine information (this includes measurements, a priori cloud state, error characteristics of model and instrument) at ALL wavelengths using pre-calculated look-up-tables generated from the single-scattering properties and auxiliary information in the form of NWP profiles, and surface properties such as reflection and emission. The advantage of this retrieval method is that there is also information on the error characteristics of the retrieval such as the measurement residuals, that is differences between actual measurements and calculated measurements at the solution. If the model were a perfect representation of the cloud then the measurement residuals would be identically equal to zero at each of the ATSR-2 wavelengths. This then gives a powerful diagnostic analysis of the assumptions used in the forward model such as single-scattering properties and cloud geometry, as well as a useful quality control parameter for the retrievals.
Case examples of semi-transparent cirrus and convective cloud will be presented from the mid-latitudes and tropics to test forward model assumptions using the measurement residual analysis. After application of rigorous quality control the behaviour of retrieved parameters such as ice crystal effective diameter, cloud-top temperature, optical thickness, and estimates of ice water path will also be presented for particular cases. The information obtained from the ATSR-2 analysis should find application in comparing our retrieved parameters with similar output from Cloud Resolving Models.