Estimating errors in fractional cloud cover obtained with infrared threshold methods

Abstract

Differences between spatial coherence and infrared threshold derived cloud cover were used to estimate errors in cloud cover that arise through the application of thresholds to imagery data. The differences indicated that threshold errors could be predicted by assuming that the cloud cover at the scale of a typical satellite imagery pixel (4–8 km) is uniformly distributed between some small value δ (δ < 0.2) and 1 − δ. In this study, results were obtained for single‐layered cloud systems from four oceanic regions. Cloud cover was analyzed for both (60 km)2 and (250 km)2 regions. Three distinct infrared thresholds were used: (1) a “cloud‐free threshold” for which the threshold was set slightly below the radiance observed for cloud‐free pixels in the region; (2) an “overcast threshold” for which the threshold was set just above the radiance observed for pixels that were overcast in the region, and (3) a “midpoint threshold” for which the threshold was set midway between the radiances observed for cloud‐free and overcast pixels. For the cloud‐free threshold, cloud cover was overestimated; for the overcast threshold, cloud cover was underestimated, and for the midpoint threshold there was relatively little difference between the threshold and spatial coherence derived cloud cover. Mean differences in cloud cover were as large as 0.25 for (60 km)2 regions and 0.18 for (250 km)2 regions. Mean and rms differences reached maximum values when the regional cloud cover was 0.5. The fraction of 4‐ to 8‐km imager pixels that were neither cloud free nor overcast was approximately 0.5. Furthermore, when the regional cloud cover (60–250 km) was small, the cloud cover in partially covered imager pixels was also small; likewise, when the regional cloud cover was large, the cloud cover in partially covered imager pixels was also large. This connection between regional scale cloud cover and 4‐ to 8‐km pixel scale cloud cover suggests a link between large‐scale fields which govern the regional cloud cover and the small‐scale disturbances to the large‐scale fields, which must govern the pixel scale cloud fraction.