Here in Exelis VIS Tech Support, we are often asked, “How do I remove the clouds from my imagery?” Unfortunately, when working with optical imagery, it is usually not possible to recover accurate information for the land surface under clouds. Opaque clouds block light both from reaching the surface (i.e., shadows), and also from being reflected back to the sensor. Consequently, there simply is insufficient signal at the sensor for cloud covered areas. Even the more translucent cirrus clouds interfere with the signal of reflected light from the surface in a way that makes it difficult to analyze or interpret optical imagery in these areas.
One way to approach this problem is to create a mask of the cloud covered areas, and leave those areas out of any image processing or interpretation performed on the optical data. In fact, official cloud mask data products are generated from some optical sensor data (e.g., MODIS).
One of the newer optical sensors for which cloud mask products are available is the Visible Infrared Imaging Radidometer Suite (VIIRS) sensor. VIIRS data products are distributed by NOAA’s Comprehensive Large Array-Data Stewardship System. The VIIRS Cloud Mask (VCM) Intermediate Products are generated using relatively sophisticated algorithms, which not only identify cloudy and clear areas, but also indicate a level of confidence in that assessment.
There are a few down sides to the VIIRS cloud mask products. The spatial resolution does not match the highest resolution available for VIIRS data. Moreover, it can be challenging to identify the correct cloud mask product that corresponds to a particular VIIRS dataset. Noting these challenges, Exelis VIS’s own Mark Piper has adapted an algorithm developed for Landsat 7 ETM+ data to provide a simpler, though less informative and robust, alternative to the VIIRS Cloud Mask (VCM) Intermediate Product. The advantage of this algorithm is that can be applied to a VIIRS Imagery EDR to conveniently quantify the fractional cloud cover in the scene or in a spatial subset.