导　　师： 黎夏

学科专业： 083001

授予学位： 博士

作　　者： ;

机构地区： 中国科学院研究生院

摘　　要： 湿地是一种介于水生生态系统/(深水湖、海洋/)和陆地生态系统/(森林、草地/)之间的一种重要而又特殊的生态系统，湿地是地球上最具生产力的生态系统之一，也是最富生物多样性的生态系统之一，它不仅能为人类的生产、生活提供多种资源，同时还具有很高的经济价值、环境效益和多种生态功能。由于近几十年来人类活动日益频繁，使得我国湿地资源急剧减少，以有效监测与合理保护湿地为目的的湿地遥感监测及其相关研究已成为湿地研究领域的一个重要研究方向。 遥感技术与GIS技术在湿地资源现状调查、动态变化监测和湿地制图等领域已经得到了广泛的应用。遥感技术所具有的观测范围广、信息量大、信息定量化、数据更新快、多时相、多平台、历史资料丰富、可对比性强等优点，使得遥感技术在湿地领域的相关研究中显得十分突出和重要。且随着近些年来计算机软件、硬件的不断完善，GPS和GIS技术的飞速发展和普及，“3S”技术的结合使得遥感技术在湿地研究中的应用范围及利用效率大大提高。 本文以珠江三角洲的核心地区/(珠江口/)作为研究区域，利用知识发现与数据挖掘方法从多时相、多分辨率、多种成像方式的遥感图像/(光学遥感、微波遥感/)中获取不同类型的湿地信息，并且监测了近20年来珠江口湿地的动态变化，分析其变化特征，同时利用时空关联规则挖掘方法来获取湿地演变与人类活动之间的关联规则。研究结果将为珠江口地区湿地资源的可持续利用提供了基础数据和决策依据，具有重要的理论意义、实践意义和社会意义。本文通过全面分析、讨论得到如下的主要结论： /(1/)在湿地遥感识别、动态监测研究中，知识发现与数据挖掘方法是一种十分有效的方法，能够取得较好的分类精度。本文研究中共使用到了决策树算法、神经网络算法、粗糙集算法、支持向量机算法和朴素贝叶斯算法。 /(2/)决策树算法识别湿地信息的精度最高。本文的研究表明，在使用的多种知识发现方法中，决策树算法的分类精度最高，且对遥感数据的要求较少。而其他的分类方法均需要在满足一定条件下，才能够得到较高的分类精度。 /(3/)1988～2004年珠江口湿地资源持续减少。本文利用4个时相的Landsat TM数据监测了珠江口湿地资源的动态变化，并分析了湿地资源的变化速率、双向动态度以及湿地资源的空间分布重心位移。分析结果表明，从1988年开始，珠江口的湿地资源在持续减少，但到了2002年，部分湿地资源受到了保护和恢复，面积略有增加。 /(4/)关联规则方法可以获取湿地演变规律。关联规则也是一种知识发现方法，它能够获取不同的项/(湿地与其他影响因子/)之间的关联性。本文以东莞市城市湿地为例，在像元尺度、镇区尺度和格网尺度三个尺度级别上挖掘湿地资源演变与社会经济统计属性因子、环境属性因子、土地利用属性因子之间的关联规则。研究表明，格网尺度是挖掘湿地演变关联规则的最佳空间尺度；东莞市城市湿地演变/(减少/)与环境属性因子的增加、外来人口数目的增加、农业人口与农业总产值的降低、城市用地的扩张之间具有强关联性。 /(5/)不同分辨率、不同成像方式的遥感数据在湿地遥感研究中得到了成功的应用。本文在利用遥感技术与知识发现方法研究湿地资源时，使用到了被动成像的光学遥感数据SPOT/(融合后其空间分辨率2.5m/)与TM图像/(空间分辨率30m/)，还有主动成像的雷达数据Envisat ASAR/(空间分辨率30m/)与Radarsat SAR图像/(空间分辨率6m/)，均取得了较好的应用。研究表明，多源遥感数据的结合使用能够将各自的优势综合起来，弥补单一遥感数据信息量的不足，扩大了各自信息的应用范围，同时还能够提高湿地信息提取的精度。 Wetland is an important and special ecosystem, which exists in hydrophilic ecosystem /(deep lake, ocean/) and land ecosystem /(forest, lawn/). Wetland is one of the most productive ecosystem and one of the most Biodiversity ecosystems. It provides many kinds of resources about production and life, and it also has high value of economics, environmental benefit, and much zoology function. In the recent years, wetland recourse in China has decreased because of the frequently action of human being. Wetland detection using Remote Sensing, which in the purpose of effective inspection and reasonable protection about wetland, and the research in concern have become the most important research direction in the area of wetland research. Remote sensing technique and GIS technique have been wildly used in the area of wetland recourse actuality survey, dynamic change detection and wetland mapping etc. Remote sensing technique has many advantages, such as wide observation, huge information, ration information, fast renewed information, multi-temporal, many plat form, rich history data, and can be contrasted feature etc, which makes the practice of remote sensing stand out and important in the area of wetland research. With the achievement of computer software and hardware these years, the GPS and GIS technology developed fast and popularization. The combination of 3S technology made the efficiency of improved in the practice of wetland research with the technology of remote sensing. This paper make the core area /(Pearl River Estuary/) of Pearl River Delta as the research area, gaining different types of wetland information using Knowledge Discovery and Data Mining method with multi-temporal, multi-resolution, multi-imaging mode remote images /(Optical remote sensing, Radar remote sensing/), and detection the dynamic changing of wetlands in the Pearl River Estuary in the recent 20 years. Analyzing the changing feature, and use the space-time association rule mining method to obtain the Association Rules about wetland evolvement and human actions. The results will support basic dada and decision gist, and have important theoretical meanings, practical meanings, and social meanings. Through overall analysis and discussion, the main conclusions were drawn in this article as follows: /(1/) In the research of wetland identify, and dynamic changes detection using remote sensing, the methods of knowledge discovery and data mining are effective, and can acquire good classify precision. Decision tree arithmetic, neural networks arithmetic, Rough Set arithmetic, Support Vector Machine, and NaiveBayes arithmetic were used in this article. /(2/) The identifying of the wetland information with Decision tree arithmetic has the highest precision. The research of this article make clear that during the several knowledge discovery methods, Decision tree arithmetic has the highest classify precision and require less remote sensing data. /(3/) The wetland resource of Pearl River Estuary decreased from 1988 to 2004. This article used 4 temporal Landsat TM data to detection the dynamic changes about the Pearl River Estuary Wetland, and analysis the changing speed, bilateral change dynamic degree, spatial center of gravity distribution displacement of wetlands. The analyzed result means that the wetland resource of Pear River Estuary decreased from 1988, but it had been protected and recovered, the area increased from2002. /(4/) The Association Rules method can obtain the wetland evolution law. Association Rules is one of the knowledge discovery method, and it can capture the association of different items /(wetlands and other influence attributes/). This article gives the sample of Dongguang City, on the level of three scales of pixel, town and grid to mining the Association Rules about the wetland evolution and the social economic statistical attributes, environmental attributes, and land use attributes. The research indicates that grid scale is the best space scale for mining the wetland evolution association rules. There are strong association among The decrease of wetland evolution of Dongguan, the increase of environment attributions, the increase of new comers and the decline of agricultural population and the agriculture outcome, the expand of city land. /(5/) The remote sensing data of multi-resolution, multi-imaging mode has been used successfully in the research of wetland remote sensing. When we using the remote sensing technology and knowledge discovery method to do the research about wetland resource, passive imaging optical remote sensing, such as SPOT /(2.5m spatial resolution/) and Landsat TM /(30m spatial resolution/) and active imaging Radar remote sensing, such Envisat ASAR /(30m spatial resolution/) and Radarsat SAR /(6m spatial resolution/) were used in this article. The research indicates that the combined use of multi-source remote sensing data can colligate all the advantages to offset the lack of information with single remote sensing data. It can enlarges the application of each information, and improve the precision of wetland information acquiring.

关 键 词： 珠江口 湿地 知识发现 数据挖掘 遥感 雷达遥感 演变规律 关联规则 决策树 神经网络 支持向量机 粗糙集 贝叶斯分类

分 类 号： [X37 X87]

领　　域： [环境科学与工程] [环境科学与工程]