Objective Methods for Registering Landmarks and Determining Cloud Motions from Satellite Data

Abstract

Cloud photographs are available from geosynchronous weather satellites at intervals of approximately 20 min. Sequences of such photographs show cloud motions over vast regions of the planet, providing a potential source of useful weather data, particularly over the oceans, since there are few weather-monitoring stations at sea. Our computer programs represent objective methods that require a minimum of human assistance to compute cloud motions. Computer algorithms are described for matching the earth's disk, giving a global registration; a more precise local registration is then achieved by matching landmarks. Accurate registration is needed because the satellite cannot be given an ideal synchronous orbit and perfect attitude stability. We have experimented with landmark templates of sizes 10×10 and 20×20 picture elements, representing 25-mi and 50-mi squares. An initial coarse scan of the template suggests where to try a succession of fine scans. The disk matching and the template matching are combined in the complete subsystem for registration. Properly registered cloud data are processed by means of the ISODATA clustering algorithm, which represents a cloud pattern by a number of cluster centers. Cloud motions are found by tracking these centers on successive pictures.