3种时空融合算法在洪水监测中的适用性研究  

Analysis of the applicability of three remote sensing spatiotemporal fusion algorithms in flood monitoring

作　　者：石晨烈 王旭红[1,2] 张萌 刘状 祝新明 SHI Chenlie;WANG Xuhong;ZHANG Meng;LIU Zhuang;ZHU Xinming(College of Urban and Environmental Sciences, Northwest University, Xi’an 710127,China;Shaanxi Key Laboratory of Earth Surface System and Environmental Carrying Capacity, Northwest University, Xi’an 710127, China;College of Resources and Environment, University of Chinese Academy of Sciences, Beijing,100049,China)

机构地区：[1]西北大学城市与环境学院,西安710127 [2]陕西省地表系统与环境承载力重点实验室,西安710127 [3]中国科学院大学资源与环境学院,北京100049

出　　处：《国土资源遥感》2020年第2期111-119,共9页Remote Sensing for Land & Resources

基　　金：中国科学院战略性先导科技专项资助项目“泛第三极环境变化与绿色丝绸之路建设子课题”(编号:XDA 2004030201);国家自然科学基金面上项目“不同地貌类型区的遥感图像信息容量的差异性研究”(编号:41071271);陕西省自然基金面上项目“基于遥感图像信息容量的城市热岛效应研究”(编号:2015JM4132)共同资助。

摘　　要：洪水灾害的遥感监测依赖于高时空分辨率影像,但目前中高空间分辨率的遥感影像受卫星回访周期及天气的影响,限制了在洪水监测中的应用。为此,提出融合MODIS和Landsat影像生成高时空分辨率影像来监测洪水灾害。以Gwydir和New Orleans 2地区为研究区,利用时空自适应反射率融合模型(spatial and temporal adaptive reflectance fusion model,STARFM)、时空反射率解混模型(spatial and temporal reflectance unmixing model,STRUM)和灵活的时空融合模型(flexible spatiotemporal data fusion,FSDAF)3种流行算法融合MODIS和Landsat影像,获得Landsat融合影像,采用支持向量机(support vector machine,SVM)对融合影像分类来提取洪水信息,并对其结果进行精度评估。实验结果表明,3种时空融合算法能够有效应用到洪水监测中,且FSDAF算法融合结果在2个研究区都优于STARFM和STRUM。在Gwydir研究区,STARFM,STRUM和FSDAF 3种算法洪水分类总体精度分别为0.89,0.90和0.91,Kappa系数分别为0.63,0.64和0.67;在New Orleans研究区,3种融合算法洪水分类精度为0.90,0.89和0.91,Kappa系数分别为0.77,0.76和0.81。此研究表明时空融合算法能够有效应用到洪水监测中。Remote sensing images with high spatiotemporal resolution offer a reliable way to the monitoring of flood disasters.However,the application of high spatial resolution images is restricted by satellite revisit period and extreme weather.Therefore,this paper proposes a method that can blend Landsat and MODIS images to generate high spatiotemporal images for monitoring flood disaster.Selecting Gwydir and the New Orleans as study areas,the authors performed fusion of MODIS and Landsat TM based on three major spatiotemporal fusion algorithms,i.e.,the spatial and temporal adaptive reflectance fusion model(STARFM),the spatial and temporal reflectance unmixing model(STRUM)and the flexible spatiotemporal data fusion(FSDAF),which led to the formation of a new TM image.Meanwhile,classified flood information was extracted by applying support vector machine(SVM)to the new TM image.The results show that three spatiotemporal fusion algorithms can monitor flood disasters effectively,with FSDAF playing a more superior role in fusion accuracy and flood information extraction.Evaluation of flood classification shows that,in Gwydir,the overall accuracy of STARFM,STRUM and FSDAF is 0.89,0.90,0.91,and the Kappa coefficients are 0.63,0.64,0.67,respectively.In the New Orleans,the overall accuracy of three fusion algorithms is 0.90,0.89,0.91,and the Kappa coefficients are 0.77,0.76,0.81,respectively.This study shows that spatiotemporal fusion algorithms can be effectively applied to flood monitoring.

关 键 词：时空融合 洪水监测 高时空分辨率 STARFM模型 STRUM模型 FSDAF模型 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]