基于Himawari-8和GF-1卫星的林火遥感监测

Fire Forest Remote Sensing Monitoring Based on Himawari-8 and GF-1 Satellites

以2017年4月11日11-16时(北京时间)辽宁省丹东市森林火灾为研究对象,基于Himawari-8(H8)和高分一号(GF-1)卫星开展林火密集监测、火烧迹地和森林燃烧受害程度监测。基于H8数据采用前后关联的火点识别算法进行火点判识,提取中心燃点温度及燃烧范围;基于GF-1数据采用近红外光谱(B4)、归一化植被指数(NDVI)和全球环境监测植被指数(GEMI)法进行火烧迹地识别;采用高斯函数拟合模型模拟健康林地生长曲线,基于当前影像数据重构火灾前的林地光谱值,以B4衰减量为评价指标对森林燃烧受害程度进行分等定级。结果表明:H8的密集监测显示中心燃点的位置由燃点1变化到燃点2、影响范围由2像元扩大至4像元、燃点温度由321 K降低至314 K;对比火烧迹地提取方法,B4波段的衰减变化最明显、GEMI模型次之、NDVI的衰减不明显,因此小尺度的火烧迹地提取采用B4衰减法最好;基于368个离散点林地光谱指数模拟健康林地生长曲线,拟合方程相关系数为0.89;以4月3日相距火灾发生仅8d的数据进行燃烧受害程度监测,重度受害程度等级漏分误差为90%;经过光谱重构后,重度受害程度分类精度提高24%、轻度受害程度提高10%,总体分类精度为69%、kappa系数为0.44,因此有必要进行火灾前遥感影像的林地光谱重构。

Forest as a vital natural resource,is closely related to human production and life.Fire is the most serious hazard to forests.Moreover,the problems of frequent fires caused by massive accumulation of forest combustible materials and global warming,etc.are prominent.Most of remote sensing monitoring on fire are conducted on the basis of polar orbit satellites;yet the monitoring frequency and precision are to be improved.In this paper,focusing on the a forest fire occurred in Dandong,Liaoning from 11am to 6pm(Beijing time)on April 11,2017,intensive fire monitoring,burned areas and the damage of forest burning monitoring are conducted based on Himawari-8(H8)and Gaofen No.1(GF-1)satellites.A fire point recognition algorithm that is contextual fire detection algorithm is adopted to recognize fire points of the 7-track H8 data.Meanwhile,the central ignition temperature and the burning range are also extracted.The near-infrared spectra(B4)and the normalized vegetation index were used before and after the fire.B4,The NDVI and GEMI methods are utilized to identify burning areas with the use of 2-scene GF-1 before and after the fire.A Gaussian function fitting model is proposed in this paper to simulate the growth of healthy forests due to the long time resolution of GF satellites and the close relationship between the burning damage and the growth of forest.On account of spectral values before the fire reconstructed by current imaging data,a standard deviation standardization method is employed to score the damaging condition with the B4 decrement as the evaluation index.Results show that:it s clear that the position of the central ignition point is changed from the ignition point 1 to the ignition point 2,the range changed from 2 pixels to 4 pixels,and the temperature decreased from 321K to 314K according to the intensive monitoring of H8;Comparing the three methods for extracting burned areas,it can be concluded that the decay is changed the most obvious in the B4 band;the GEMI model is second;while the decay of NDVI is not obvious.Therefore,the B4 decay method is the best way to extract burned areas.The growth curve of healthy woodland is simulated based on the spectral index of 368 discrete forest lands,and the equation is fitted.The growth curve of the healthy forest is simulated based on forest spectral indexes of 368 discrete points with a correlation coefficient of the fitting equation is 0.89.Also,burning damage is monitored with data of a fire occurred 8 days ago in 3rd,April,whose omission error of the severity of damage is 90%.However,the precision of data severity,the precision of mild damage increased and the overall classification precision after simulated are increased by 24%,10%and 69%,respectively,with the kappa coefficient of 0.44.Therefore,it s necessary to reconstruct the spectral spectrum of remote sensing images for forests before fire.