| description abstract | This paper demonstrates how satellite observations of the cloudiness over a complex area such as the European Mediterranean area can be classified into distinct cloud regimes by application of a K-means clustering algorithm to pixel-level cloud properties. The study contrasts with previous approaches in the fact that the clustering is done on the cloud physical properties at the pixel level and not on statistics of these properties over a coarser grid. A method to choose the number of clusters is described. ?Shallow cumulus,? ?stratocumulus,? and ?frontal? clusters are robustly identified, and associated environmental properties are described. The approach helps to refine the diagnosis of errors in model simulations. In addition to isolated classical errors of climate models (lack of midlevel clouds, overestimation of the cloud optical thickness, and underestimation of the stratocumulus) and a dramatic underestimation of the shallow cumulus clouds over land, an underestimation of the boundary layer depth is detected for some regimes as well as an incorrect stratification for the shallow cumulus. The clustering on the individual cloud properties classifies them in a much more homogeneous way than the clustering on the statistics of the cloud properties; the use of individual cloud properties at the pixel level produced by climate models activating simulators combined with subgrid-scale sampling procedures may be considered as an alternative to the use of the statistical products for the evaluation of models. The approach can also be applied to a high-resolution regional climate model. | |
