What is a Cloud: the Choices People Make and Why They Regret Them

What is a cloud? There are various correct answers to that question, but perhaps the best answer is another question - why do you need to know? To detect cloud, an algorithm looks for particular signatures in the measurement that differentiates a cloud-contaminated pixel from a clear-sky background. These signatures can be spectral, textural or temporal in nature. The magnitude of the difference between the cloud and the background must exceed a "threshold value" for the pixel to be classified having a cloud-contaminated FOV. All detection algorithms employ multiple tests ranging across some portion of the ultraviolet, visible and/or infrared spectrum. A cloud is not a single, uniform object, but rather a has a distribution f optical thickness and morphology. As a result, a problem arises when the distributions of cloud and clear-sky background characteristics both subsume the thresholds from tests, making some test results indeterminate—leading to some amount of misclassification. Further, a correctly identified cloud-contaminated FOV is not necessarily as good candidate for cloud retrievals that use relatively simplistic forward radiative models. This presentation will investigate the different 'choices' folks make in cloud detection algorithms and the impact on regional and global cloud amounts and fractional coverage, cloud type and property distributions. We will explore the reasons why these choices are made, discussing both active and passive satellite observations. We will make reference to the different scientific uses that the community has for cloud detection and resulting properties, and choices that modelers have in defining a cloud. We will also present data sets we have created to meet the needs of different communities.