Automated Cloud Detection and Classification of Ground-Based Hemispheric Sky Images

Shallow cumulus clouds have a significant impact on the planetary boundary layer through the distribution of latent and sensible heat fluxes; their impact on solar energy production is also significant. While the study of these clouds is undeniably important, the process of finding past cases is often laborious. The historical record of cloud coverage from visual observations at meteorological stations terminated in the U.S. in the mid-1990's due to the shift to automated observations by ceilometers. The historical visual observations typically included the fraction (octas) of the sky covered with clouds as well as the cloud species; the end of manual observations resulted in little record of cloud species in the ensuing decades. Since then, the primary method for identifying shallow cumulus cases is manual inspection of satellite or photographic imagery.

By using TSI images from the Atmospheric Radiation Measurement (ARM) Climate Research Facility Southern Great Plains (SGP) site, the authors have developed a statistical classification system optimized for the automatic determination of the presence of shallow cumulus clouds by analyzing motion, texture, and color statistics of the all-sky photographs. This classification algorithm shows promise for eliminating the time-intensive process of manually finding shallow cumulus clouds from TSI sky images.