An Interpretation of Cloud Overlap Statistics

Observational studies have shown that the vertical overlap of cloudy layers separated by clear sky can exceed that of the random overlap assumption, suggesting a tendency toward minimum overlap. In addition, the rate of decorrelation of vertically continuous clouds with increasing layer separation is sensitive to the horizontal scale of the cloud scenes used. The authors give a heuristic argument that these phenomena result from data truncation, where overcast or single cloud layers are removed from the analysis. This occurs more frequently as the cloud sampling scale falls progressively below the typical cloud system scale. The postulate is supported by sampling artificial cyclic and subsequently more realistic fractal cloud scenes at various length scales. The fractal clouds indicate that the degree of minimal overlap diagnosed in previous studies for discontinuous clouds could result from sampling randomly overlapped clouds at spatial scales that are 30%–80% of the cloud system scale. Removing scenes with cloud cover exceeding 50% from the analysis reduces the impact of data truncation, with discontinuous clouds not minimally overlapped and the decorrelation of continuous clouds less sensitive to the sampling scale. Using CloudSat–CALIPSO data, a decorrelation length scale of approximately 4.0 km is found. In light of these results, the previously documented dependence of overlap decorrelation length scale on latitude is not entirely a physical phenomenon but can be reinterpreted as resulting from sampling cloud systems that increase significantly in size from the tropics to midlatitudes using a fixed sampling scale.