The Method Research Of Tropical Cyclone Intensity Estimation And Wind Fieldretrieval Based On Satellite Data

Tropical cyclone is one of the commonest disasters in China. Once a tropical cyclone lands coastal cities, it will cause very serious losses of lives and economics. So the accurate prediction of tropical cyclone is important to reduce the impact of disasters. The intensity of tropical cyclone is an important index for weather forecast. By comparing the methods of reconnaissance aircraft, radar and satellite monitoring, satellite data and machine learning technique are used to estimate the intensity of a tropical cyclone because of its high temporal resolution. Finally the result of tropical cyclone intensity estimation is used to retrieval wind field of eyed typhoon. This paper includes three parts as flows:(1) The structure identification of a tropical cyclone based on the deviation angle. Shape and structure of a tropical cyclone have different characteristics in different developing stages. Studies show that deviation angle can be used to describe the axial symmetry graph’s characteristics. If a tropical cyclone closes to axial symmetry graph, the deviation angle is smaller. The histogram of deviation angles based on infrared satellite cloud image is used to identify structure of a tropical cyclone. Typhoon and storm data are used to identify developing stage of a tropical cyclone cloud image by the histogram of deviation angles of the tropical cyclone.(2) Tropical cyclone intensity estimation based on machine learning. Generally, the maximum wind speed of the sea surface is used as intensity of a tropical cyclone. Deviation angle variance and gray gradient co-occurrence matrix parameters are respectively related to the maximum wind speed by Radial Basis Function Neural Network, Least Squares Support Vector Machine and Relevance Vector Machine. The tropical cyclones are divided into two categories:eyed and non-eyed tropical cyclones. Generally, eyed tropical cyclones are eyed typhoon, and non-eyed tropical cyclones include non-eyed typhoon and storm. The experiment results show that three kinds of intelligent algorithms have good estimation result compared to the traditional linear regression method. What’s more, Relevance Vector Machine’s modeling performance is more stable than Radial Basis Function Neural Network and Least Squares Support Vector Machine. So Relevance Vector Machine is considered as the modeling method of wind field retrieval. Gray gradient co-occurrence matrix is suitable to estimate the intensity for eyed typhoon and storm, and deviation angle variance is suitable to estimate the intensity for non-eyed typhoon.(3) Eyed tropical cyclone wind field retrieval based on machine learning. The retrieval of wind field surrounding the tropical cyclone is exciting. However the wind field retrieval of inner core is not satisfactory. In this paper, we try to retrieval both surrounding and inner core of a tropical cyclone. Partial differential equation technology is used to segment eye wall of the tropical cyclone, then the bright temperature data is obtained. The maximum wind speed of corresponding cloud image is obtained by tropical cyclone intensity estimation and tropical cyclone yearbook. Finally Relevance Vector Machine is used to build the model between bright temperature and maximum wind speed. For any test infrared cloud image, its eye wall is segmented and through the model getting wind speed of on the eye wall each point as the reference wind speed. Then the test point’s wind speed is computed by using a distance formula. The experimental results show that the wind field retrieval result by the proposed algorithm is satisfactory and the comprehensive performance is superior to the traditional linear regression algorithm.