【作者】 朱晓荣；

【导师】 鞠洪波；

【作者基本信息】 中国林业科学研究院 ， 森林经理学， 2012， 博士

【摘要】 洞庭湖湿地地处长江中下游，对维持生态平衡和保持本区域的经济可持续发展具有重要意义。然而，由于自然环境的变迁、长期的泥沙淤积和人类活动干扰的日益加剧，洞庭湖湿地受到了严重的破坏，湿地面积急剧下降。人类活动与河流水沙情势制约着湖泊的演化，围湖垦殖和入湖泥沙淤积加速着洞庭湖的衰亡，洞庭湖敞水区的面积急剧缩小，洲滩面积逐渐增多，洞庭湖湿地的结构和功能正在发生着巨大的变化。本研究以洞庭湖区为研究对象，利用不同分辨率遥感影像数据的光谱与纹理特征，结合其他辅助数据，探索洞庭湖湿地信息的高效提取方法。同时根据多期影像数据的分类结果，揭示湿地的演变规律。主要研究成果与结论如下：（1）利用人类活动中洞庭湖防洪大堤对洞庭湖区域进行分区，结合GIS知识的空间分析功能，研究洞庭湖湿地的分布特征，考虑冬夏季相的不同，充分挖掘数据，构建决策树实现对研究区湿地类型的精确获取。将分类结果与传统最大似然法监督分类所得结果进行对比可知：利用知识的决策树分类方法对湿地类型进行分类，较传统的最大似然法监督分类总体精度提高12.05%；总体kappa系数提高0.1407；特别是外湖区的林地，芦苇滩地，泥滩地、水体等覆盖类型其生产者精度和用户精度有大幅提高。（2）利用SPOT-5高分辨率影像进行洞庭湖湿地土地覆盖分类，选择全色波段作为纹理特征计算的数据源；通过选定样本的J-M距离确定各湿地类型相对应的最佳纹理尺度；选用QUEST算法对遥感影像光谱、纹理信息构成的数据集进行数据挖掘，构建决策树模型，对高分辨率影像进行分类。结果表明结合多尺度最佳纹理信息的高分辨率影像分类，分类精度达到78.57%，而单一光谱数据分类和结合单尺度纹理数据的分类精度分别为71.98%和76.76%。可见，纹理信息能够有效地提高地物的识别程度，多尺度纹理能够更好地描述地物的纹理特征，更有效解决分类结果中的同谱异物现象，有助于提高高分辨率影像分类精度与效率。（3）水体、泥滩地、苔草滩地、芦苇滩地、水田是本研究最主要的土地覆盖类型。水体、苔草滩地面积在所选时间跨度上先减少，后趋于稳定；泥滩地呈现持续下降的趋势。作为本区鱼类与水鸟主要栖息、觅食地的水体、泥滩地、苔草滩地的减少，表明奔去湿地退化严重威胁本区的湿地多样性保护。从1987年至1996年，研究区湿地类型发生剧烈变化，主要转化类型为泥滩地、水域、苔草滩地与林地的转化。1987年至2004年洞庭湖外湖区的林地面积仍在持续增加，同时泥滩地持续减少；在2004年-2009年区间，外湖区的林地面积一定程度出现下降，但不明显，其已经成为洞庭湖区特别是西洞庭湖区一种主要的地面覆被类型。（4）研究区景观总体的多样性和异质性变化不大；景观中各优势地类所占比重呈现先减小后增大的趋势；景观类型团聚程度增大，分散程度降低。研究区地类斑块在1987年至1996年总体呈现斑块趋于破碎化和小型化的趋势；在1996年至2009年2个时间段内，则表现为斑块趋于聚合的趋势。研究区地类斑块密度变化总体呈现先增大后减小的趋势，斑块破碎化程度在1996年达到最高，到2004年时又有很大程度好转，景观趋于完整；斑块边界密度在1996年至2009年之间总体下降，研究区地类形状趋于规则，受到较为剧烈的人为干扰。研究区地类呈现出聚集度增加的趋势。更多还原

【Abstract】 The Dongting wetland is located in the Yangtze River. It’s a great significance of theecological balance and the regional economic sustainable development in this area. However，because of the change of natural environment，long-term sediment siltation and humanactivities increasing，the Dongting Lake wetland has been destroyed seriously. The wetlandarea decreased sharply in Dongting Lake area. Human activities and the river water sandregime is restricting the lake evolution，reclamation from lakes and lake sediment depositionaccelerated decline of Dongting Lake. In open water area reduces，bottomland area graduallyincreased，the Dongting Lake wetland structure and function is undergoing tremendouschanges.In this study based on different resolution remote sensing image，using the remote sensingdata of spectrum and texture features by combination with other auxiliary data we extracteDongting Lake wetland information. At the same time according to the multiple image dataclassification results，it reveales the evolution of wetland.The result showes that:(i)By using flood control dam on Dongting Lake wetland zoning，combined with theknowledge of GIS spatial analysis function，studying of Dongting Lake wetland distributioncharacteristics in winter and summer，considering the phase different，sufficient mining data，a decision tree is constructed to obtained achieve the study area wetland types. Than wecompared classification results to the traditional maximum likelihood supervised classification.Which shows: The method using of knowledge of decision tree classification to classify typesof wetlands, increased by12.05%to the traditional maximum likelihood supervisedclassification overall accuracy; overall kappa coefficient was increased by0.1407; Such aswoodland, reed beaches, mudflats, water coverage types’ producer accuracy and user accuracywere greatly improved outside the Lake area. (ii) SPOT-5high resolution images is used to classification land coverage type.inDongting lake wetland. We selected panchromatic as texture features to calculate data source.Through the various J-M distance for the selected sample of wetland types we determined thebest texture scale.,than we used QUEST algorithm for remote sensing image spectral，textureinformation form data sets ofr data mining to constructe the decision tree model for highresolution image classification.The results show that by selecting the optimal texture scale combination，using a decisiontree on the spectral data and multi scale texture data for high resolution remote sensing imagehaving a high classification accuracy of78.57%. The spectroscopic data classification andcombining single scale texture data classification accuracy were71.98%and76.76%. It showsthat texture information can effectively improved the wetland object recognition level，at thesame time multiscale texture can better described the features of texture features and moreeffective to solved the classification results of the foreign body in the same spectrumphenomenon. It helped to improve the accuracy and efficiency of high resolution remotesensing image classification.(iii) Water，mudflat，carex，reed，forest land and paddy field are the most important typesof land coverage in the study area. Water，carex area at the selected time span decreases first，then tends to be stable; Forest land to explosive increase，then tends to be stable.From1987to1996，in the study area the land coverage types have a acute change. Themain transformation is happened between mudflat，carex，reed and water to forest land. From1987to2004forest land area continues to increase，at the same time the mudflat continued todecline. In the years2004to2009interval， the forest land outside the Dongting lake dam areais decline，but not obvious. It has become a major ground cover types in the Dongting Lakearea especially in Western Dongting Lake area.(iv) The study area’s overall landscape diversity and heterogeneity changed little. Thesuperior class proportion appears first decrease and then increase. Landscape types ofagglomeration degree increases，the degree of dispersion is reduced. Land coverage types in the study area, plaques in overall plaque tended to thefragmentation and miniaturization trend from1987to1996. Form1996to2009it shows asplaque tends to polymerization tendency. The study area to class plaque density changesoverall appears first increased and then decreased. Patch fragmentation peaked in1996to2004，which showed it has a great degree improvement and landscape tends to be integrity.Patch boundary density in1996to2009has a overall decline. The study area shape tends to beregular which has a more intense human disturbance. The class of the study area presents aaggregation degree increasing trend.更多还原