基于增强型燃烧指数的中小型火烧迹地自动提取研究  

Automatic extraction of small and medium-sized burned area based on enhanced normalized burning ratio

作　　者：贺忠华 李少虹 毛燕军[2] 王宁杭 张育慧 何月[2] 方贺 HE Zhonghua;LI Shaohong;MAO Yanjun;WANG Ninghang;ZHANG Yuhui;HE Yue;FANG He(Zhejiang Academy of Safety Science and Technology,Key Laboratory of Safety Engineering and Technology Research of Zhejiang Province,Hangzhou 310012,China;Zhejiang Climate Center,Hangzhou 310056,China;Zhejiang Aerial Forest Protection Administration,Hangzhou 310020,China)

机构地区：[1]浙江省应急管理科学研究院,浙江省安全工程与技术研究重点实验室,浙江杭州310012 [2]浙江省气候中心,浙江杭州310056 [3]浙江省航空护林管理站,浙江杭州310020

出　　处：《自然灾害学报》2023年第3期192-204,共13页Journal of Natural Disasters

基　　金：浙江省自然科学基金项目(LQ21d050001,LQ21d060001);浙江省安全工程与技术研究重点实验室开放基金项目(202106);浙江省气象局科技项目(2021ZD09,2021YB07)。

摘　　要：火烧迹地提取是林火灾后评估和重建的重要内容。中小型森林火灾分布范围广,发生频率高,人工判识耗时耗力。卫星遥感能大范围、高时效地提取火烧迹地,但不同区域地表环境和气候条件差异使得其自动提取难度大。以浙江省2019—2021年7个中小型林火为研究对象,基于Sentinel-2和Landsat-8卫星影像开展火烧迹地自动提取研究。首先,利用实测信息进行火场模糊定位,结合谷歌地球引擎完成火灾定位和卫星数据获取;其次,基于卫星影像多波段反射率,利用云过滤、大津法(Ostu)、指数分割、图像膨胀等预处理识别和剔除云、水体和非林地等干扰信号。最后,通过燃烧面积指数和归一化燃烧指数计算,图像增强和波段合成等,自主构建增强型归一化燃烧指数(ENBR),并结合高斯滤波和Ostu完成中小型火烧迹地的自动提取。结果表明:ENBR能显著凸出火烧迹地和周边环境特征差异;2021年实验样本自动提取经目视判读和实测数据验证,面积偏差分别为3.69±3.09和2.51±3.33 hm^(2),提取精度达(94.20±3.03)%和(90.54±8.21)%,显著优于dNBR和BAI的人工阈值判识结果;2019—2021年12次提取应用整体偏差为5.00±4.70 hm^(2),提取精度达(92.50±5.06)%。研究表明,ENBR和Ostu自动提取方法精度高,且区域自适应能力强。因此,研发的ENBR和火烧迹地自动提取方法对广泛发生中小型林火的中国南方地区具有重要应用价值。Burned area extraction is important for the post-assessment and reconstruction of the fire hazard.Small and medium forest fires are widely distributed with high frequency,and the interpretation manually can be a very time-consuming and labor-intensive process.Observations of satellites have advantages in extracting the burned area in a large area efficiently.However,differences of surface environment and climatic conditions in different regions make it difficult for the automatic extraction of burned area.In this study,we selected 7 small and medium forest fire samples in Zhejiang Province during 2019—2021 for researches on automatic extraction of burned area based on medium-high resolution optical satellite images.Firstly,fuzzy positioning of fires based on measured information and satellite data acquisition through Google Earth engine were conducted.Secondly,processes of cloud filtering,Ostu,index segmentation,and image expansion,based on multi-bands of satellites observed surface reflectance,were used to identify and eliminate the interference information,such as cloud,water and non-forest signals.Finally,the enhanced normalized burned ratio(ENBR)was constructed by processes of burned area index(BAI)and normalized burned ratio(NBR)calculations,image enhancement and band synthesis,and the automatic extraction of burned area could be accomplished by combining with Gaussian filtering and Ostu methods.Results showed as follows.ENBR can significantly highlight the difference between the burned and surrounding areas.After validation by the visual interpretation and measured data,area biases of automatic extraction using experimental samples in 2021 were 3.69±3.09 and 2.51±3.33 hectares,and their accuracy were(94.20±3.03)%and(90.54±8.21)%,which is significantly better than the results from dNBR and BAI artificial threshold identification.Overall area bias of 12 extraction applications from 2019 to 2021 was 5.00±4.70 hectares,and the accuracy was(92.50±5.06)%compared with visual interpretation results.Results

关 键 词：增强型燃烧指数 火烧迹地 自动提取方法 中高分辨率卫星 中小型林火 

分 类 号：S762[农业科学—森林保护学] P237[农业科学—林学]