Research On Cloud Image Interpretation Based On Fengyun Geostationary Meteorological Satellite

The meteorological satellite emerged in the 1960 s is a breakthrough in meteorology research and application which provided continuous high-temporal-spatial-resolution cloud images.Satellite cloud images can be used as auxiliary sources for daily weather analysis and forecasting.At present,visual interpretation of cloud image is still the main methods in the meteorology research and application.Subjective factors make it difficult to effectively extract useful information in satellite cloud images,which hinder the accuracy of the numerical weather forecasting.With the rapid development of domestic meteorological satellite technology,the interpretation of the satellite cloud images has become one of the popular research topics in recent years.In 2016,China launched her first three-axis geostationary satellite—FY-4A(Fengyun-4A)which is located at 104.7°E and equip with three advanced optical instruments:AGRI(Advanced Geosynchronous Radiation Imager),GIIRS(Geosynchronous Interferometric Infrared Sounder),and LMI(Lightning Mapping Imager).AGRI has 14 spectral channels with high spatial and temporal resolution(about 15 min interval),The infrared cloud image of highfrequency day and night monitoring of weather system can be obtained,which makes meteorological satellites play an increasingly important role in weather forecasting and monitoring applications.The resolution of infrared cloud image is lower than that of visible cloud image.Using FY-4A infrared satellite cloud image for automatic identification and segmentation of target cloud system is of great value to improve the quality of weather forecast and early warning ability of meteorological disasters in China.This research first describes FY-4A infrared image data and its processing flow.Then based on interpretation method of EUMETSAT(European Organisation for the Exploitation of Meteorological Satellites)satellite cloud image,a set of algorithm flow for automatic recognition and segmentation of typical large-scale cloud systems(cold front?warm front and occlusion front)on FY-4A cloud image is established by using FY-4A infrared image combined with mathematical morphology and cloud system geometric shape characteristics.The main four contents of the study are as follows:(1)Constructing the frontal area classification model.After analyzing the characteristics of the frontal areas of the typical large-scale cloud systems,we defined the features of the frontal areas in the FY-4A satellite thermal image and extracted them with mathematical morphology operations and multi-threshold segmentation algorithms.A classification model was constructed to discriminate the frontal areas and non-frontal areas with XGBoost(Exterme Gradient Boosting)decision tree algorithm.The results of the prediction was evaluated,and the overall accuracy reached to 93.43%.(2)Comparing the performance of four optical flow models.The extrapolation accuracy of four optical flow models was compared to choose a better extrapolation algorithm which is used to solve the problem of the "jumping" phenomenon in frontal area recognition.The result shows that the performance of the DIS(Dense Inverse Search)optical flow algorithm and the backward advection scheme adopted by Dense model group were better than those of the ShiTomasi corner detection and LK(Lucas-Kanada)scheme adopted by Sparse model group.(3)This paper introduces the principle,algorithm and processing process of frontal rear side detection,and uses the extrapolated frontal area image and the current frontal area to judge whether there is a rear side of frontal area,which can be used as an important rule for the subsequent interpretation of cold and warm front cloud system.(4)Using mathematical morphological algorithms,extract the frontal area skeleton and distance map,determine the end points and branch points of the frontal cloud system skeleton,and then calculate the geometric shape characteristics of the frontal cloud system on the basis of skeletonization,with the help of the frontal area rear side conditions And the geometric shape characteristics of the image realize the interpretation of the typical cloud system.The results were identified by experts,the recognition rate of cold front was 79.12%,the rate of warm front was 73.89%,and the rate of occlusion front was 75.61%.The images similarity was also evaluated between the manually extracted regions and the automatically segmented cloud areas,90% similarity shows that the recognition and segmentation algorithm in the study has the potential practice value for automatically detecting the typical large-scale cloud systems.