Development of a cloud detection method from whole-sky color images

A method is proposed for detecting clouds from whole-sky color images obtained with an all-sky camera (ASC) system. In polar regions, cloud detection using whole-sky images usually suffers from large uncertainties in fractional cloud cover retrievals because of large solar zenith angles (SZAs) and high surface albedo, which cause whitening in the images. These problems are addressed by using differences between real images and virtual clear-sky images for a particular observation time with the same SZA. The method is applied to ASC images obtained at Ny-Alesund, Svalbard in May of 2005-2007, and the results are compared with Micro-Pulse Lidar (MPL) measurements. When no clouds were detected by MPL, the false cloud detection rate from ASC classification was 2.1% in total hours. Conversely, when clouds were detected by MPL, the ASC classification underestimated the clouds by 11.6%. In most cases, this occurred when MPL detected very optically thin clouds. Furthermore, the variability of cloud fractions estimated by MPL and ASC was roughly constant regardless of the SZA. Thus, it is confirmed that the method developed in this study is valid for cloud detection from whole-sky color images. (C) 2014 Elsevier B.V. and NIPR. All rights reserved.