Cloud Classification Based On Geostationary Meteorological Satellite Imagery

Cloud radiation properties and distribution significantly affect the forecasting accuracy and global climate's change. Studies about satellite images have been carried on since the first meteorological satellite was launched.Various algorithms of cloud feature extraction and selection, as well as cloud classification, are reviewed first. Then clouds radiation properties and characteristics in three satellite channels are introduced. Taking FY-2C satellite data for example, methods of the earth location and data calibration are discussed particularly.This study identifies immature cumulus clouds using a structural threshold approach. These immature cumulus clouds maybe develop into storm or other strong convective weather. Quick and accurate identification of these cumulus clouds is helpful to reduce the number of computation in strong convective weather forecasting. The study also analyzes the results of this threshold cloud classification machine with three cases.For classifying more types clouds, this study also classifies clouds with Support Vector Machines (SVM) method. First of all, one-against-one classified structure is used, and the kernel function of SVM uses Radial Basis Function. Secondly, the research data, which use the cloud classification productions of National Satellite Meteorological Center as types'labels, are divided into Land, Mixed cloud, Altostratus or Nimbostratus, Cirrostratus, Cirrus Spissatus, Cumulonimbus and Stratocumulus or Altocumulus. There are 9 features about radiation and texture information composed the feature vectors of training and predicting data. Thirdly, the SVM parameters C andγare set by dichotomy method. Then SVM models are got by training samples. At last, by predicting data with these models, the classification results are got. And the results from models with two best parameter sets are analyzed.