How are the cloud maps created in the Atmospheric Correction Module - FLAASH?

This Help Article discusses how the cloud classification maps are generated in ENVI's Atmospheric Correction Module - FLAASH and what they are used for in the atmospheric correction.
 
FLAASH tries to compute a mask of pixels that might contain clouds because it is important to eliminate these pixels from the regular processing in certain places (such as computing the adjacency correction). Several criteria are used to improve accuracy and allow multiple cloud types to be identified.

• The first finds pixels that have a (green:swir) ratio between 0.8 and 1.2 but also have unusually high albedo (greater than 0.3).
  
• The second test looks for the amount of water in the column compared to clear pixels.
  
• The third uses the cirrus channel to look for reflectance greater than a threshold (that is determined by the histogram of the cirrus channel pixels).

Cirrus clouds are a special case because they are not opaque in the visible and near infrared (VNIR). If the image doesn't have all of the bands required for a particular test, then that test is skipped. So, you might end up with cloud map that only reports opaque clouds if you didn't have cirrus channels available. In this sense, you might say that the cloud map is less reliable than if you had the cirrus channels. But another way to think about it is that FLAASH is doing the best it can with the data provided to it to find clouds and treat them accurately in the processing.

If the input image has some artifacts or bad data (for example, where images are mosaicked together or where fill values have been used due to a bad detector) then this can potentially lead to a strong response in one of these cloud criteria. Remember, the cloud map is an ancillary output from FLAASH (it's already computed so the map is output in case it might be useful for someone) but FLAASH isn't a cloud-mapping program. It is just used to attempt an incremental improvement to the reflectance results, mostly by improving the visibility estimate and adjacency correction. A good use for the cloud mask is to verify that its not reporting water retrieval errors across a large part of the image. If it does, then you may want to start looking into the journal.txt file to see if there's anything else wrong.

Review on 12/31/2013 MM