【作者】 李萍；

【导师】 陶夏新；

【作者基本信息】 哈尔滨工业大学 ， 防灾减灾工程及防护工程， 2009， 博士

【摘要】 地震是危及人民生命财产的突发式自然灾害。在大地震发生后,快速获取灾情信息、形成救灾决策、部署应急救援对减轻灾害有重要意义。地理信息系统(Geographic Information System,GIS)技术的发展为实现这一减灾目的提供了技术手段。我国近二十年来已经建立了数十个城市和大型企业的基于GIS的防震减灾及辅助决策系统,投入了大量的人力、物力。本文从系统维护和升级的角度,研究此类系统如何快速估计地震灾害,为减灾发挥切实有效的作用,从软件层面和技术层面做深入探讨。2008年的5.12汶川地震是对现有减灾技术手段的一次验证。本文从汶川地震的实际震例中,分析上述防震减灾系统功能的局限,急需增加新的处理模块、配合新的信息源,评估震害。提出、研究了增加多源信息修正等震线模块、基于减灾需求的中巴卫星图像分景索引、考虑云及其阴影的遮挡提取山地灾害的模块。本文结合多个实际震例,对系统评估震害做深入分析,指出了系统准确评估震害分布的关键在于提高对地震动影响场的估计精度,为了快速提高精度,需要引入新的信息源。遥感和互联网所包含的信息在空间上分布范围都比较广,而且能够在一定有效期限内实时更新;特别是互联网作为“信息高速公路”,信息的获取成本低,处理信息的工作量小、耗时少,完全可以开发用于减灾。本文以5.12汶川8.0级地震为例,创新性地提出一套基于互联网的描述性信息提取信息烈度、空间插值得到信息烈度预测面、生成信息等震线的方法。发展了人机交互修正等震线的方法,通过平移、放大经验性等震线解决宏观震中与微观震中不吻合、极震区与计算结果不吻合的问题,得到初步修正等震线,然后根据信息烈度和信息等震线的空间分布与修正等震线的空间关系,对其做二次修正;除了考虑随时间变化设置统计节点,还考虑了互联网能搜集到的遥感评估结果,将其转化成烈度,作为修正等震线的辅助信息。与实际调查结果比较看来,随着日期逐渐推移、信息量不断增加,相应的修正精度也越来越好,在震后第12天得到的信息累计结果,对等震线的评估结果已经和随后实际调查得到的烈度分布较为接近。在处理遥感信息源的方面,本文首先考虑了数据下载前的筛选过程的自动化。遥感图像文件较大,一景中巴资源卫星的多光谱图像就是百兆以上;按照现有的数据提供方给出的遥感图像的查询方式,搜索、下载图像后,仍然需要筛选、比较得到最需要处理的目标图像,这个过程也很耗费时间。本文以中巴资源卫星图像的筛选下载为例,研究突破网站现有搜索方式,建立面向减灾需求的中巴卫星分景索引模块。首先基于中巴资源卫星网站一定时间、空间范围的卫星景心搜索结果,提取图像信息,包括PATH、ROW、景心坐标、成像时间等,建立景心点群数据库,考虑卫星沿轨道规律成像的特性,从景心点群提取了标准景心点阵;以此为依据,对研究范围内的卫星图像的搜索结果做空间筛选,去除轨道异常点,得到规律成像的图像景心点;另外,考虑变化检测的需求,实现两个时间筛选准则,即震前震后图像成像时间间隔最短、季节相近准则,用Matlab编程实现空间、时间筛选准则;最后,做筛选景心数据和地震影响场的叠加分析,得到位于不同烈度区的卫星图像的下载优先级列表,实现考虑评估地震灾害需求的中巴卫星分景索引。针对汶川地震影响范围内时、空准则筛选后的结果,考虑云及其阴影的影响,做变化检测,提取地震引发的山地灾害分布,并给出灾害判定的准确度。首先,根据云及其阴影的波谱特征,用决策树法、支持矢量机(SVM)法及K均值聚类法,确定震前、震后图像中,云及其阴影的分布。用分层随机采样法给出随机点,验证分类精度。比较三者分类精度,决策树法精度好、简单易行。合并震前震后的遮挡区(云及其阴影区),将研究范围分成有一定准确概率的两部分,将震前、震后图像的变化检测结果与其做叠加分析,得到位于地物区的灾害区准确度一般比较大,而落入遮挡区的灾害区真的属于灾害的概率一般比较小。最后,对本文的研究工作进行了全面总结,展望了有待进一步深入研究的几个问题。更多还原

【Abstract】 Earthquake is a very fearful natural disaster that occurs suddenly and endangers people’s life and property. It is imperative for disaster relief agencies and civil protection bodies to evaluate the damage as soon as possible after the event for emergency response. The development of Geographic Information System technology provides a good measure to cope with this disaster mitigation demand. In China, dozens of cities and large enterprises spent a lot of manpower and material resources on building their own Earthquake Prevention and Disaster Reduction System (EPDRS) based on GIS since the beginning of 1990’s. Do these systems are useful or not, and how to improve the validity of those systems are discussed in knowledge and software aspects.Wenchuan Earthquake occurred on May 12, 2008 in Sichuan Province of China, which is a validation of our country’s disaster mitigation work. Considering this real earthquake lesson, this paper analyzed the functional limits of EPDRS and the imperative work to improve the capability to face the sudden earthquake damage. Those works included in this paper are as follows: add new information sources and the corresponding new processing modules to assist in modifying the experiential isoseismal calculated by EPDRS; build image index of CBERS satellite on basis of the demand of disaster mitigation; detect the mountain disasters from CBERS satellite image partially overlaid by the clouds and their shadows.This paper took several real earthquake experiences as validation cases of EPDRS. The validations with thorough analysis illumined that the precision of isoseismals generated by EPDRS or any other system must be improved for a better estimation of earthquake damage; that need to absorb the new information sources.The information included in Remote Sensing and Internet is distributing in a large space scope and that can be almost updated in real time at a certain time limit; especially, being“information highway”, Internet is a cheap information source, collecting information from Internet cost less workload and time consuming, it can be used to assist disaster mitigation. This paper took 5.12 Wenchuan earthquake as example to build a series methods that include: collect the describing information from Internet, get“information intensity”from the collection, interpolate in spatial range to build the prediction surface and get the contours which represent isoseismals. Human-computer interaction method was developed in this paper by move, enlarge the experiential isoseismals to deal with the deviation of the macro- from micro-epicenter, the meizoseismal area from the calculated zone by EPDRS, by which the 1st modification of the experiential isoseismals is completed; and then considering the distribution of the“information intensity”and“information isoseismals”, the 2nd modification can be made. Besides setting different time nodes of information collections, the building damage assessment results of remote sensing image published in the government Webpage was considered to be converted to RS-intensity to be used to modify the isoseismals. By comparing with the ground survey result, the precision of the modification isoseismals is better more and more along with the time changing and the information amount increasing. On the 12th day after earthquake the modification of isoseismals is already similar with that from the survey report.In the aspect of RS image processing, this paper considered to realize the automation of choosing image process before downloading. RS images are big files, one scene of CBERS-CCD multi-spectrum image can be more than 100 Mb. The searching methods on the CBERS webpage are limit, and the searching results still need more time to choose the image to study. This paper took the choosing process of CBERS image as example, and broke through the conventional searching methods in, and then built the index module of choosing CBERS images by considering the demand of disaster mitigation.Firstly, ordered a certain time and space range in the CBERS Webpage to search image,built a scene-center database from the information of the search results, such as PATH, ROW, longitude, latitude etc., from which the standard scene-center point’s matrix was calculated according to the satellite imaging character. Take the standard scene-center point’s matrix as normal points; pick up the scene-center points in a luminal distance. Secondly, in order to obtain the better result of change detection, two choosing rules about imaging time were founded and actualized by Matlab. In the end, the chosen points were overlaid with the isoseismals, the spatial analysis method“identity”is applied to get the priority download list of CBERS images.According to the download list, pick up several images pair (pre- and post-earthquake) to detect mountain disaster. The influence of the clouds and their shadows in the RS images on the change detection of mountain disaster. Three classification methods is carried out to get the distribution of the clouds and their shadows in this paper, the precision of classification is calculated. By comparison, the decision tree built in this paper is simple to apply and had a good precision of classification. Merge the regions of clouds and their shadows in pre- and post- image to get the shielding zone, the else region is ground object zone; the damage located in the shielding zone is false damage, that in the ground object zone is the real damage.Finally, all progresses in this dissertation are summarized; the assumptions and prospects for further study are mentioned.更多还原