有云Landsat TM/OLI影像结合DEM提取青藏高原湖泊边界的自动算法研究  

Automatic Algorithm for Extracting Lake Boundaries in QinghaiTibet Plateau based on Cloudy Landsat TM/OLI Image and DEM

作　　者：王鑫蕊 晋锐 林剑[4] 曾祥飞[4] 赵泽斌[1,2] Wang Xinrui;Jin Rui;Lin Jian;Zeng Xiangfei;Zhao Zebin(Northwest Institute of Eco-Environment and Resources,Chinese Academy of Sciences,Lanzhou 730000,China;University of Chinese Academy of Sciences,Beijing 100049,China;Center for Excellence in Tibetan Plateau Earth Sciences,Chinese Academy of Sciences,Beijing 100101,China;Key Laboratory Knowledge Processing and Networked Manufacturing,Hunan University of Science and Technology,Xiangtan 411201,China)

机构地区：[1]中国科学院西北生态环境资源研究院,甘肃兰州730000 [2]中国科学院大学,北京100049 [3]中国科学院青藏高原地球科学卓越创新中心,北京100101 [4]湖南科技大学知识处理与网络化制造实验室,湖南湘潭411201

出　　处：《遥感技术与应用》2020年第4期882-892,共11页Remote Sensing Technology and Application

基　　金：中国科学院战略性先导科技专项(A类)“时空三极环境”项目(XDA19070104);NSFC项目(41531174、41471357)。

摘　　要：青藏高原的湖泊数量众多且分布广泛,约占全国湖泊总数量和总面积的41%和57%,对于全国甚至全球的湖泊研究十分重要。遥感监测湖泊分布历来已久,但光学遥感影像时常被云遮蔽,难以自动化提取得到完整的湖泊边界。提出了一种有云的Landsat TM/OLI影像结合航天飞机雷达地形测绘数据(Shuttle Radar Topography Mission,SRTM)30 m分辨率的数字高程模型(DEM)的湖泊完整边界的自动插值生成算法。首先,在Google Earth Engine平台上,利用Landsat TM/OLI影像的地表反射率Tier1数据,根据其中的像元质量评价(Pixel Quality Assessment,pixel_qa)属性,结合SRTM 30 m DEM数据,先剔除云、云阴影、积雪和山体区域的影响;计算改进的归一化差异水体指数(Modified Normalized Difference Water Index,MNDWI),采用Canny边缘检测算法,得到无云覆盖区域的已知部分湖泊边界(L)。在本地对DEM进行极差滤波,得到可能的湖泊区域;同时,利用DEM生成等高距间隔为1 m的等高线,将包围着可能湖泊区域的一系列等高线自动筛选出来,根据等高线间的包含关系建立树结构。叶子节点为最内部等高线,记为内等高线(C1)。由于Landsat和DEM的获取时间不同,随着湖泊扩张或萎缩,湖泊水位会相对于内等高线上升或下降,对此采用不同的外等高线(C2)确定方法;随后,建立内等高线C1、外等高线C2与已知部分湖泊边界L之间对应点的坡度坡向关系,插值得到未知的湖泊边界点;最后利用最近邻法连接已知的湖泊边界点与插值得到的湖泊边界点形成完整的湖泊边界。利用相近日期的资源三号影像或无云Landsat影像的手工数字化湖泊边界对提取的湖泊边界进行验证,发现两者基本重合,且长度差百分比为-6.81%~9.4%,面积差百分比为-2.11%~2.7%。表明该方法对于有云Landsat TM/OLI影像的湖泊边界自动化提取十分有效,并为在GEE等大数据平台中自动化提取长时间序列、高�Lakes in the Qinghai-Tibet Plateau are numerous and widely distributed,accounting for 41%and 57%of the total number and area of lakes in China,which are very important for the study of lakes in the whole country and even in the whole world.Remote sensing has been used to monitor the lake distribution for a long time,but optical remote sensing images are often obscured by clouds,from which it’s impossible to automatically extract complete lake boundaries.An automatic interpolation algorithm for lake boundary generation based on cloudy Landsat TM/OLI image and Shuttle Radar Topography Mission(SRTM)30 m resolution Digital Elevation Model(DEM)is proposed.Firstly,supported by the platform of Google Earth Engine,the tier1 data of Landsat TM/OLI images are used to eliminate the effects of cloud,cloud shadow,snow and mountain area,based on the Pixel Quality Assessment(pixel_qa)attribute and SRTM 30 m DEM.Then,the Modified Normalized Difference Water Index(MNDWI)is calculated,and the Canny edge detection algorithm are used to obtain the known part of the lake boundary(L)in cloud-free areas.The possible lake areas are obtained by range filtering of DEM locally.At the same time,DEM is used to generate contours with an isometric interval of 1 m,and a series of contours surrounding the possible lake area are automatically screened out.The tree structure is established according to the inclusion relationship between contours.The leaf nodes are the innermost contours,which are recorded as inner contours(C1).Because the acquisition time of Landsat and DEM is different,with the lake expanding or shrinking,the lake water surface will rise or fall relative to the inner contour.Different methods of determining the outer contour(C2)are adopted.Subsequently,the slope-aspect relationship between the inner contour C1 and the outer contour C2 and the known part of the lake boundary L is established,and the unknown lake boundary points are interpolated.Finally,the nearest neighbor method is used to connect the known lake boundary points

关 键 词：青藏高原 湖泊边界 Landsat影像 有云 DEM 

分 类 号：P941.78[天文地球—自然地理学] TP79[自动化与计算机技术—检测技术与自动化装置]