Use of a Pattern Recognition Technique for Determining Cloud Motions from Sequences of Satellite Photographs

Abstract

Recent geosynchronous satellites (ATS-I and ATS-III) have obtained photographs of clouds over broad regions of the earth at intervals of ∼20 min. Sequences of such photographs show the direction and speed of cloud motions. To extract the motion information in real time it is desirable to process the raw data by computer methods. One approach to this problem is to locate a limited number of centers of brightness that represent the cloud patterns in a photograph. Such centers, which am determined on the basis of area and quantified brightness, are analogous to centre of gravity in mechanics. Measurements of displacements of these centers in successive photographs provide an indication of cloud motion. Brightness centers are found by an objective computer technique called “ISODATA” that was developed in an earlier program of pattern recognition. The average distance between brightness centers is determined in part by specifying values of certain parameters in the ISODATA program. This average distance is a measure of the scale of the analysis, and can be adjusted according to the intended application. Cloud motions are derived by a separate program that matches brightness centers found on two pictures, and computes the distances between matched pairs. In cases selected for initial study, the computed cloud motions were reasonable and agreed qualitatively with motions perceived visually from time-lapse presentations of the same data. Further study is required to compare this technique with others, and to extend it to include infrared data (closely related to cloud heights) that will be obtained by future synchronous satellites.