The Research On FY-2C Cloud Detection Technology Based On Block Partition

Satellite remote sensing has been widely used in various fields, like weather analysis, environment monitoring, energy investigation, global research and regional planning, etc. The experts are still looking for all sorts of ways, in order to further use of the massive remote sensing data resources. Among them, the cloud is the most intuitive one of satellite remote sensing data. The cloud is an important factor of climate changing, which has a directly effect on the radiation budget of the earth system. Therefore, it has always been the research target of meteorologists. There is great uncertainty in the time and space distribution of cloud, and it has always been a big problem because the change of height, thickness, constitution of cloud and solar altitude angle, azimuth will cause a great change of satellite cloud image features. The exploration of cloud detection methods has never stopped, and a variety of methods and ways have been tried by various scholars and experts, and the effect is remarkable.In this article, a new idea has been proposed that the research target should be changed from pixels to blocks. The size of a block is different from each other, there may be only a few pixels, or there may be a large amount of pixels. However, no matter the size of a block, the uniformity is necessary, which means that all pixels of one block belong to the same class. This is premise of conducting the research.Taking advantage of the multi-spectral and high temporal resolution features of FY-2C geostationary satellite, this article conducts the research of cloud detection. First of all, use boundary division algorithm, which combines the scan line seed fill algorithm with 8-connected boundary tracking algorithm to guarantee that the boundary is complete and closed, to conduct the boundary recognition. Then, use Zonal toolsets in ArcGIS and cluster analysis for blocks to extract characteristic values and conduct preliminary classification, with visual interpretation to compare the features of no cloud areas and cloud areas, to gain the characteristic, then define a new index I=RVIS·(TIR4-TIR1). And gain the feature that the values of I are lower in no cloud area after doing statistics. At last, get the threshold of I by conducting iteration of I, as the index to divide the cloud areas and no cloud areas, which is the result of cloud detection.This article uses the above method to detect clouds on the FY-2C satellite cloud images on 1 January and 1 July 2006 from 02:00 to 06:00 (UTC), and compare with the cloud classification products provided by National Satellite Meteorological Center, China Meteorological Administration. The results shows that over 70% of cloud detection results are the same as cloud classification products, and the most one is over 85%. After verifying the parts of different detecting results between two methods by visual interpreting, it shows that the cloud detection results have more correct areas. And there are two major reasons why the results are incorrect, one is that the data of VIS channel is unavailable, the other is the accuracy of the threshold of I. In summary, this new cloud detection technology has a better effect, and is worthy of further research.