Using Cloud Fraction and Condensate Decorrelation Lengths to Reproduce Cloud Field Statistics

A few years ago the exponential-random (ER) paradigm has been proposed as a representation of cloud fraction overlap that is superior to earlier paradigms such as maximum and maximum-random overlap. Within the ER framework the degree of overlap is often simply expressed in terms of a decorrelation length (scale). Similarly, the rank correlations of condensate (a measure of the degree of vertical alignment of relatively thin and thick parts of cloud layers) has also been modeled with decorrelation length scales assuming that the rank correlation drops as an inverse exponential with separation distance. The applicability of these approximations has been tested with two-dimensional condensate fields derived largely from a ground-based cloud radar. Specifically, we use the so-called MICROBASE evaluation product of the ARM Climate Research Facility which resolves highly (10 sec horizontally and 45 m vertically) the distribution of cloud condensate. Seven years of data from ARM's Southern Great Plains site were analyzed. The initial decorrelation climatology analysis shows a conspicuous seasonal cycle for both overlap and rank decorrelation lengths with peaks in the summer months and a significantly faster tendency towards randomness for condensate rank correlations. For domains equivalent to about 75 km, the range of monthly cloud fraction decorrelation length values is about 1.9 to 3.3 km while for rank correlations 0.8-1.2 km, with slightly larger values when the domain size is doubled. The purpose of the presentation is also to examine whether decorrelation length scales used in cloud generators are capable of reproducing the statistical properties of the condensate field (e.g., total cloud fraction, profiles of cumulative cloud fraction, variance of water path) with acceptable fidelity. We investigate the level of spatial and temporal detail needed in the specification of decorrelation lengths in order for this capability to be realized (if at all, while maintaining the simplicity needed for actual implementation in cloud modeling parameterizations. We attempt to address this question with a variety of tests that reveal decorrelation length dependencies on cloud type and sensitivities to specific choices in the analysis methodology.