Daytime cloud classification over South American region using multispectral GOES-8 imagery

Abstract

Information about a minimal set of characteristic variables and types of scene was searched for Geostationary Operational Environmental Satellite (GOES)-8 Imager multispectral imagery over an extended area of South America, September 2002, 1600 UTC (near local noon over most parts of the area). Thirteen variables were considered for each pixel: five of them described reflectance and brightness temperature in four channels, three variables assessed temperature difference related to channel 4; finally, five variables assessed local homogeneity (texture) in each channel. Thirty-two clusters were determined by a classification scheme (‘dynamic cluster’) based on minimal Euclidean distance. Factor analysis in principal components (PCs) applied to cluster centroids shows that only five variables might be taken as non-redundant, namely reflectance in channel 1 and brightness temperature in channel 4 as well as their textures, together with difference between channels 5 and 4. Although factor analysis suggests defining about seven clusters (which in turn are consistent with image nephanalysis), PC analysis makes it evident that an objective minimal number of scenes is actually loosely defined but provides some useful criteria for definition of proper centroids, depending on the user’s convenience.