Hazardous weathers triggered by deep convective clouds such as thunderstorms, strong winds have always endangered people’s lives and social economy, so it is very important to conduct researches on convective clouds.Taking advantage of the multi-spectral and high temporal resolution features of Geosynchronous Meteorological Satellites, this paper concentrates on the study of detection of deep convective cloud and forecast of its evolution. Firstly deep convective cloud is analyzed and compared to clouds of other types in terms of its spectral characteristics and textural features. Based on the analysis an algorithm for deep convective clouds detection is put forward and the process is divided into two main parts. The first step is the recognition of high dense clouds into which the fitted slope of pixels’distribution in Water Vapor-Infrared spectral space is introduced on the basis that clouds of different types possess varied distribution pattern in the spectral space. The method shows its superiority over traditional brightness temperature threshold technique by comparing the boundary of the detected clouds to its corresponding TBB contour map. In the second part deep convective cloud regions are extracted from the results of the first step with respect to their rough texture and decreasing cloud temperature in developing stage. In addition, tropopause temperature data from FNL material is utilized as an adaptive criterion for the detection of deep convective clouds in mature phase.With the aid of a new image matching approach based on a pyramidal scheme, a complete field of cloud motion vectors is retrieved. Based on this displacement field short range forecast extrapolation of convective clouds can be obtained using backward trajectory technique. Evaluation of the extrapolation of convective clouds show that vectors calculated by the image matching algorithm can well reflect the local changes of convective clouds as well as clouds motion dominated by large scale flow and the forecast performs well in short time range. The results are more accurate for clouds with large spatial dimensions than that of small ones. |