Knowledge Resource Center for Ecological Environment in Arid Area
| 基于遥感的湖泊面积提取及其时空变化研究-以亚洲中部地区为例 | |
| 其他题名 | Study on Extraction and Spatial and Temporal Change of Lakes Based on Remote Sensing Data – A Case Study of Asia Central Area |
| 车向红 | |
| 出版年 | 2018 |
| 学位类型 | 博士 |
| 导师 | 孙九林 |
| 学位授予单位 | 中国科学院大学 |
| 中文摘要 | 湖泊的形成与消失、扩张与收缩对生态环境演化和社会经济发展都有重要影响。由于受气候、生态环境和人类活动等因素的综合影响，湖泊水域范围的变化速度快、幅度大，对观测的频率和分布都有很高的要求。近几十年以来，卫星遥感技术以其快速、覆盖面广、成本低廉等优点，为较大区域的湖泊动态监测提供了重要数据基础。 本文以亚洲中部地区为例，针对大范围、高精度、长时间序列的湖泊变化分析对遥感数据时空分辨率的需求，重点开展了三项研究：1）基于MODIS观测数据，研究湖泊监测方法及在青藏高原的应用；2）研究基于Landsat等中高分辨率卫星数据的湖泊提取方法，提高对小水体的监测能力；3）利用2000年以来的Landsat多颗卫星的观测数据，结合高性能数据存储和处理能力，提取亚洲中部地区2000-2015年湖泊分布，分析其湖泊数量、面积和分布变化。本文主要结论如下： （1）MODIS的广域、快速观测数据为大区域较大的湖泊分析提供重要的数据源，但同时也存在局限。例如宽视场角遥感观测的BRDF效应导致在水体边界提取不稳定以及混合像元问题。论文利用BRDF校正方法和降尺度方法（IMAR）有效地提高了湖泊监测的空间精度和准确度，并在青藏高原开展了应用研究。 （2）构建了自适应的Landsat自动水体提取算法，通过结合遥感波谱指数、地形数据以及相关参考信息数据，构建分层加权采样空间，形成基于决策树分类算法的水体提取模型，能够对每一景Landsat进行自动水体提取。基于每个像素的遥感重复观测，判断每个观测的获取时间，通过时间序列内插分析，构建逐月的水体观测分布和变化过程。收集研究区2000年以来的云量小于80%的所有Landsat数据，获得共96278景影像（约25T数据量），结合高性能数据处理能力，生成中亚干旱区和青藏高原2000-2015年的湖泊时空变化数据集。 （3）对自动水体提取结果进行全面、定量的评价和精度验证。利用分层随机采样采集样点，通过人工解译，获取能够代表不同时空分布的验证样点。评价结果表明：研究区时间序列水体数据总体精度为99.45%(±0.59)，水体用户精度（错分）为85.37%（±3.74），制图精度(漏分)为98.17%(±1.05)。错分误差主要由预处理部分残留的云阴影、山体阴影以及冬季冰雪造成。夏季水体提取精度最高，冬季提取精度较低，对于冰雪的提取有待改进。时间序列内插法生成逐月水体数据交叉验证的总体精度为98.57%，稳定陆地上插值结果的精度最高（98.86%），稳定水体的插值精度为98.46%，波动水体上精度最小（87.47%），表明时间序列内插法的可行性。 （4）湖泊数量和湖泊面积的关系是反映水体分布特征的重要指标。论文发现，中亚干旱区和青藏高原的湖泊数量均和湖泊面积呈幂指数关系，与全球范围的类似研究结果相一致。但对比发现，该关系在中亚呈现相对两头大中间小的分布，而青藏高原则在中型湖泊的统计得到较高的密度，表明中亚干旱区湖泊小而多，且分布不均匀，而青藏高原地区的湖泊面积和分布比较均一。此外，2000-2015年中亚干旱区湖泊数量减少率约为青藏高原9倍，且湖泊数量波动约为青藏高原地区的6倍。 （5）在2000-2015年间，中亚干旱区湖泊面积变化要比青藏高原湖泊更为显著。在该16年内，中亚干旱区湖泊总面积整体呈减少趋势，减少速率约为1146 km2/年（年变化率1.2%），而青藏高原则整体呈增加趋势，增加速率约为356.86 km2/年（0.72%）。中亚干旱区约75.24%的湖泊面积处于减少趋势，主要分布在西北部平原区和新疆中北部沙漠区域，扩张湖泊主要分布在哈萨克斯坦大草原东部区域，中亚河谷湿地和昆仑山中部地区。青藏高原地区湖泊约46.54%的湖泊面积呈扩张趋势，主要集中在青藏高原中部的高山草原和灌木林地，相比中亚干旱区，青藏高原的中小型湖泊扩张比例明显增大。面积缩小的湖泊大多面积小于10km2，主要分布在西藏东南部林地和灌木地和草地区。 （6）研究逐月湖泊面积和变化，有助于理解区域湖泊季节性特征和变化驱动因素。论文分别以中亚干旱区和青藏高原的两个典型湖泊为例，研究发现两区域湖泊面积的月年际变化呈相反趋势。从2000至2015年，中亚干旱区典型湖泊在4-10月的年际面积均呈减少趋势，其中4-5月的年际面积减少最为显著，融雪减少可能是其减少的主导因素，且受人类活动影响明显。青藏高原地区典型湖泊在6-9月年际面积均有增加趋势，其中6月的年际面积增长最快，相关性分析表明，湖泊面积年变化和6-9月年际变化均受降雨影响比较显著。 |
| 英文摘要 | The distribution of lakes in space and its change over time are closely related to many agricultural, environmental and ecological issues, and are important factors that must be considered in human socio-economic development. Remote sensing has provided important data source for monitoring temporal lakes dynamics with its advantage of rapidness, wide coverage and lower cost. Taking Asia central area as an example, accurate dynamics of lakes from 2000 to 2015 is analyzed using high temporal-spatial resolution remote sensing images. First, lakes detection method is designed using MODIS observation data, and is applied to lakes on Tibetan Plateau. Second, in terms of issues from coarser images, we present an adaptive Landsat water extraction algorithm to improve the performance of small area water bodies. Finally, based on extracted water results, the distribution of lakes from 2000 to 2015 is produced with servers of high performance data storage and processing capability, and their change of the number, area and distribution are analyzed. The conclusions are as follows: (1) The wide swath and rapidness of MODIS observation provide important data source for the analysis of large lakes on large-scale region. However, there are still some limitations. For example, BRDF effect caused by wide swath scanning make the water boundary detection unstable and lead to mixed pixels. We used BRDF correction method and developed a downscaling method (IMAR), which facilitates improving the spatial accuracy of lakes monitoring. (2) Incorporated spectral indices, MODIS water data and topographic data, the C5.0 decision tree is built to extract water separately on each scene with hierarchical weighted sampling space. Subsequently, based on multiple observations and acquisition times on pixel scale, monthly water extents are derived by time series interpolation. We collected 96278 scenes images (25 terabytes) since 2000, and spatial-temporal dynamics dataset is produced for the arid region of central Asia (CA) and Tibetan Plateau (TP) considering high performance data storage and processing capability of servers. (3) Extracted water results are validated systematically and quantitatively. Specifically, a stratified random sampling strategy is used to produce validation samples representing different spatial-temporal distributions, which is interpreted manually. Comparisons of extracted and interpreted samples indicate that proposed method provide good performance for water detection (Overall accuracy: 99.45(±0.59); User accuracy: 85.37%±(3.74); Produce accuracy: 98.17(±1.05)). The commission error is primarily caused by cloud and hill shadow on pre-processing step and ice/snow in the winter. Therefore, water results from colder season should be used discriminately. In addition, Overall accuracy of cross validation for interpolated results is 98.57%, and the accuracy for three stratifications (permanent land, permanent water and variable water) is 98.86%, 98.46% and 87.47% respectively, which demonstrates the effectiveness of time series interpolation. (4) Lakes number and size can effectively reflect the distribution characteristics of water bodies. The results show there is a prominent power exponent relation between lake size and lake number for both CA and TP, which is similar to existing analysis of global scale. However, the majority of lakes on the CA is small size and distributed unevenly, while the lake size and distribution on the TP is more homogenous. The number density of the smaller and larger size of lakes is higher than that of medium size of lakes on CA, whereas the number density of medium size of lakes on TP is way higher than that of CA. The change rate of number of lakes on CA is 9 times compared to TP, and the number of lakes on CA is more volatile (nearly 6 times compared to TP). (5) From 2000 to 2015, lakes area change on the CA is more prominent in comparison to TP. In particular, from 2000 to 2015, the total area of lakes on CA experiences a decreasing trend, and decreasing rate is 1146km2/year (yearly change rate: 1.2%). However, lakes area on TP has an increasing trend, and corresponding rate is 356.86km2/year (0.72%). The percentage of shrinking lakes with the decreasing rate on TP accounts for 75.24%, which are mainly located at the northwest plain area and desert region of north-central Xinjiang. The expanded lakes on CA are primarily distributed in the eastern region of the Kazakh steppe, central Asia valley and wetland zones as well as the central area of Kunlun Mountain. In terms of TP, 46.54% of lakes area are increasing, and mostly distributed in Alpine grassland and shrub woodlands zone of the central Tibetan plateau. Additionally, compared to CA, TP has higher proportion of expansion for medium size of lakes. Shrinking lakes mostly with the area less than 10km2 is situated on the woodland and shrub and grass areas in southeastern Tibet. (6) The study of monthly lake area and change can help understand regional seasonality and driven factors of lakes. The study took four typical lakes on CA and TP as examples, analyzed their dynamics, which shows there is an opposite trend for monthly variation from 2000 to 2015. The monthly area change of lakes on CA experiences decreasing trend from April to October, and decreasing rates on April and May are largest, which probably results from the reduction of snow-melting ae well as increasingly human activities. However, monthly change rate is positive, and the largest increasing rate is also on June. By comparing with climate variables (precipitation, temperature and evaporation), the yearly and monthly lake area change on TP are significantly influenced by precipitation. |
| 中文关键词 | 水体提取 ; 中亚干旱区 ; 湖泊变化 ; MODIS ; Landsat |
| 英文关键词 | water detection arid region of central Asia lakes dynamics MODIS Landsat |
| 语种 | 中文 |
| 国家 | 中国 |
| 来源学科分类 | 地图学与地理信息系统 |
| 来源机构 | 中国科学院地理科学与资源研究所 |
| 资源类型 | 学位论文 |
| 条目标识符 | http://119.78.100.177/qdio/handle/2XILL650/288074 |
| 推荐引用方式 GB/T 7714 | 车向红. 基于遥感的湖泊面积提取及其时空变化研究-以亚洲中部地区为例[D]. 中国科学院大学,2018. |
| 条目包含的文件 | 条目无相关文件。 | |||||
| 个性服务 |
| 推荐该条目 |
| 保存到收藏夹 |
| 导出为Endnote文件 |
| 谷歌学术 |
| 谷歌学术中相似的文章 |
| [车向红]的文章 |
| 百度学术 |
| 百度学术中相似的文章 |
| [车向红]的文章 |
| 必应学术 |
| 必应学术中相似的文章 |
| [车向红]的文章 |
| 相关权益政策 |
| 暂无数据 |
| 收藏/分享 |
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。
