Fuzzy Sets for Cloud Detection in Atsr–2 Images over Land

This paper presents a cloud detection scheme for the second Along–Track Scanning Radiometer (ATSR– 2) instrument data, that has been developed using a fuzzy set methodology. Fuzzy sets allow pixel classes to be defined with soft “real–world” decision boundaries, rather than the traditional “yes/no” characterization. Described by a membership function, the grade of membership of a given fuzzy set is assigned for all image pixels. For this scheme the cloud and surface data sets produced by the International Satellite Cloud Climatology Programme (ISCCP) are used as a knowledge base of a priori data. A simple atmospheric model allows the ISCCP data to be translated to an “as seen by ATSR–2” form, from which bi–variate fuzzy sets for a variety of Earth surfaces and cloud types are calculated, based on their probability distributions. Applied to cloud detection in ATSR–2 images, this varying degree of membership for an images pixel in the “cloudy/clear” fuzzy sets results in a superior image classification.