Spatial and temporal varying thresholds for cloud detection in satellite imagery

A new cloud detection technique has been developed and applied to both geostationary and polar orbiting satellite imagery having channels in the thermal infrared and short wave infrared spectral regions. The bispectral composite threshold (BCT) technique uses only the 11  m and 3.9  m channels, and composite imagery generated from these channels, in a four-step cloud detection procedure to produce a binary cloud mask at single pixel resolution. A unique aspect of this algorithm is the use of 20-day composites of the 11  m and the 11 - 3.9  m channel difference imagery to represent spatially and temporally varying clear-sky thresholds for the bispectral cloud tests. The BCT cloud detection algorithm has been applied to GOES and MODIS data over the continental United States over the last three years with good success. The resulting products have been validated against “truth” datasets (generated by the manual determination of the sky conditions from available satellite imagery) for various seasons from the 2003’2005 periods. The day and night algorithm has been shown to determine the correct sky conditions 80’90% of the time (on average) over land and ocean areas. Only a small variation in algorithm performance occurs between day-night, land-ocean, and between seasons. The algorithm performs least well during the winter season with only 80% of the sky conditions determined correctly. The algorithm was found to under-determine clouds at night and during times of low sun angle (in geostationary satellite data) and tends to over-determine the presence of clouds during the day, particularly in the summertime. Since the spectral tests use only the short- and long-wave channels common to most multispectral scanners, the application of the BCT technique to a variety of satellite sensors including SEVERI should be straightforward and produce similar performance results.