Head in a Cloud

troposphere and stratosphere meet blogosphere

Head in a Cloud random header image

Cloud Overlap Statistics

September 13th, 2006 by Sean Davis · 2 Comments

I’ve been back from vacation for a couple of days, and things are starting to settle back down a little. On my flight to Ecuador, I read a recent GRL paper entitled “Instantaneous cloud overlap statistics in the tropical area revealed by ICESat/GLAS data”, by Likun Wang and Andrew Dessler. …In the paper, Wang and Dessler use data taken by the GLAS lidar on the ICESat satellite during 2003 to identify cloud type/overlap statistics between 10°S – 20°N. This seemed like an appropriate paper to read, as it was precisely the region through which I was flying :)

So why are cloud overlap statistics important? In short, GCM representations of clouds, which are on a relatively coarse spatial scale (relative to GLAS observations), rely on assumptions about the overlap between clouds at different altitudes. Such assumptions have profound implications for atmospheric heating/cooling rates and the radiative balance within the lower atmosphere.

The most common cloud overlap scheme in GCM’s is called the “random-maximum” assumption. This assumes that if there are two layers of clouds at different heights (separated by clear air) within a GCM gridbox, the clouds overlap in a “random” fashion. If the clouds are continuous in the vertical, they are assumed to overlap maximally (think deep convection). Under the random overlap assumption, the cloud fraction is given by Crandom = C1+C2-C1C2, where C1 and C2 are the fractional cloud coverage from the two independent layers.

From the GLAS observations, the observed cloud fraction is C1+C2-Coverlap. Wang and Dessler test the random overlap assumption for different cloud types by comparing the product of C1 and C2 (measured from GLAS data) to Coverlap (also measured from GLAS data). They divide the GLAS cloud observations into five types: boundary layer, shallow convection, mid-level, deep convection, and cirrus. C1, C2, and Coverlap are then computed for the various combinations of two-layer cloud systems formed by these groups (e.g. cirrus + boundary, cirrus + deep convection, etc.). What they find is that in some cases (cirrus over boundary layer clouds or deep convection), there is less overlap (by 50 – 100%) than the random overlap assumption would predict, whereas in others (cirrus over mid-level or shallow clouds) the random overlap assumption underestimates the actual overlap (by ~20-30%).

So what are the implications of these disparities for GCM’s or future efforts to improve cloud overlap representations? The article of course does not approach this topic, being a short 4-pager GRL, but does note that the (recently launched) CALIPSO and CloudSat missions will be able to extend this work by extending the time period and geopraphical regions studied. It will be interesting to see the larger set of statistics provided by these new data sources.

I am curious to know, however, how the results from these type of studies could be
incorporated into GCM parameterizations of cloud overlap? I assume one could replace the C1XC2 term by a geographically, seasonally, and cloud-type-combination varying Coverlap term. That seems like it would get really messy really quickly, but perhaps this is the kind of improvements that modelers seek? I also wonder to what extent (if any) the “instantaneous” nature of the cloud overlap statistics from GLAS would be smeared out (and approach the “random” overlap assumption) if the data were averaged over horizontal gridboxes of the order of GCM grid sizes? Anyone?

Tags: climate · satellite

2 responses so far ↓

  • 1 Monica // Sep 28, 2006 at 2:29 pm

    Hi Sean,
    I can’t answer your question, but you may want to contact Bryan Baum. I’ve worked with him on nighttime cloud overlap and he probably would know the answer–or at least who does!

  • 2 seand // Sep 28, 2006 at 2:41 pm

    Hi Monica,
    Thanks for the useful link!

Leave a Comment