Forest Fire Burned Area Extraction From Polarimetric SAR Images

Forest fire occurs frequently in global forest, it is an important confounding factor that affects global forest ecosystem. So accurately grasping the forest fire information, including occurrence time, location and scope, is the basic for understanding the complex relationship between forest fires and the atmosphere, biosphere, and human activities. Satellite remote sensing is a feasible mean to obtain information on a wide range of forest fire.Early forest fire extraction based on satellite remote sensing data focuses on optical/thermal infrared remote sensing. However, it is influenced by cloud and weather, sometimes it can't detect burn area. Synthetic Aperture Radar data is few affected by weather, and at the same time SAR data includes unique information that can't be contained in visible/near-infrared and thermal infrared data. Since the 1990s, studies are carried out at abroad, research has not yet appeared at home, meanwhile, most studies are based on single polarized channel data, few are based on post-fire quad-polarization data. With radar sensor and polarization becoming diversity, extraction of forest fire burn area is developing forward using fully polarization data before and after fire. Therefore, this paper carries out the following tasks:(1) This article focuses on the comparative analysis of polarimetric parameters before and after forest fire for the first time, in order to clarify and reveal the scattering mechanism change before and after fire. On this basis, one of the most sensitive forest fire polarization parameter is selected from each type of polarization parameters for the subsequent extraction of forest fire burned area. Additionally, scattering characteristics of different vegetation, different burn severity and different aspect of the forest burned area is revealed.(2) In this paper, change detection method is firstly used to extract forest burned area. And foret burned area is achieved through binary segmentation to the scattering power difference image by a robust and reliable method proposed in this paper. The result shows that forest fire burned area extraction using change detection method is feasible and effective.(3) Extraction of forest burn area under different data condition is carried out using classification method. This paper firstly uses the full-polarization data before and after forest fire to achieve the forest fire scar extraction, experimental result shows that, compared with previous purely single-channel data before and after forest fire, full-polarization data before and after forest fire not only expresses the changes of backscatter intensity,but also reflects the change in vegetation structure, therefore combination of full-polarization data before and after fire are more conducive to accurately extracting forest burned area; In additional, compared with previous only using full-polarization data after fire,full-polarization data before and after fire is not only able to express the difference between burned area and unburned area after fire,but also takes advantage of burned area change information, thereby combination of full-polarization data before and after fire can effectively improve forest burned area extraction accuracy.