KMS XINJIANG INSTITUTE OF ECOLOGY AND GEOGRAPHY，CAS
| 中亚阿姆河干流地表水体面积时空变化分析 | |
| Alternative Title | Analysis of Spatio-Temporal Variation of Surface Water of the Main Stream of Amu River in Central Asia |
| 李冬冬 | |
| Subtype | 硕士 |
| Thesis Advisor | 杨辽 |
| 2020-06-30 | |
| Degree Grantor | 中国科学院大学 |
| Place of Conferral | 北京 |
| Degree Discipline | 工程硕士 |
| Keyword | 陆表水体 水体提取 遥感 时间序列 机器学习 Land surface water Water extraction Remote sensing Time series Machine learning |
| Abstract | 本研究针对面对水体提取对河流，尤其是中亚干旱与半干旱区域河流提取分类研究较少，阿姆河历史影像冗余利用率不高，利用时间序列的中高分辨率Landsat 卫星遥感影像， 通过对比多种水体指数与构建简单随机森林分类器的机器学习分类对阿姆河流域陆表水河流的提取精度，对其进行水体提取，构建时间序列阿姆河主干河流水体数据集。 通过机器学习的方法，借助 Python 编程的方法实现近 2900 景数据的处理以及水体提取，最终统计出 1986-2018 年这 33 年的阿姆河主干河流的水体面积，进行变化分析探究。结果表明：(1)通过对比主要 NDVI、 MNDWI、 WI2015、 Wetness 遥感指数在不同时间节点分类阈值的不一致，已经复杂多变，虽然 MNDVI 在对于选中的验证数据的分类精度对于水体达到 93.5%， 但是对于非水体的分类精度确存在一定偏差， 低于 90%， 而且同一阈值运用到另一验证样本是精度均不及当前精度。(2)基于像元的特征学习构建的简单随机森林分类器，通过对不同时序的样本点进行学习，预测分类来区别水体与非水体。 避开繁琐的经验阈值的尝试与选择，在总体分类上看， 对于水体与非水体的正负类别的预测识别精度相对较稳定均高于 92%， 与此同时由于学习样本时间跨度覆盖月份范围交广，能够避免由于季节性影像差异带来的水体分类的错分的影响。 通过综合精度较高的随机森林RF 分类器对研究区的 Landsat 多种数据源的历史影像进行分类预测， 对分类结果通过算法优化进一步消除错分类别进一步提高提取结果，最终通过影像拼接裁剪，构建阿姆河主干河流的水体提取数据集。(3)通过Landsat数据集对阿姆河干流进行水体提取，整理拼接裁剪得到1986-2018 年这 33 年的阿姆河主干河流水体 4 月、 7 月和 9 月的水体提取结果。从历年提取结果中看：阿姆河主干河流年内面积： 7 月>4 月>9 月。 1986-2018 年年平均水面积变化具有具有较高的线性相关， R²=0.7837。水体面积整体呈现下降趋势。 1986 年主干河流水体面积为 965.102Km²，截止 2018 年水体面积为 720.814Km²，面积减少约为 25.31%，平均每 10 年减少 7.67%。（4）通过面积变化对比分析，阿姆河主干河流水体面积变化可以分为三个阶段： 1986 年-2001 年、 2002-2008 年、 2009—2018 年。其中 1986 年-2001 年为第一阶段，水体面积变化巨大，波动较为明显，水体面积小范围增加，随后减少，整体呈现下降趋势。 2002-2008 年间与前一阶段相近，出现小范围极值，但是面积变化波动较小。 2009-2018 年间为第三阶段，水体面积平稳中下降，逐年下降幅度约为-1.6%。 |
| Other Abstract | In this study, there are few studies on river extraction and classification in the faceof water extraction, especially in the arid and semi-arid areas of Central Asia. Thehistorical imagery of the Amu Darya River is not highly redundant. By comparingvarious water body indexes with the machine learning classification of a simple randomforest classifier to extract the accuracy of land surface water rivers in the Amu DaryaBasin, the water bodies are extracted to construct a time series Amur River main riverwater body data set. Through the method of machine learning, the processing of nearly2900 scenes data and the extraction of water are realized by Python programming, andfinally the water body area of the main river of the Amu Darya River in the 33 yearsfrom 1986 to 2018 is counted and analyzed for changes. The results are listed in thefollowing.(1) By comparing the inconsistency of the classification thresholds of the mainNDVI, MNDWI, WI2015, and Wetness remote sensing indexes at different time nodes,it is already complex and changeable. Although the classification accuracy of MNDVIfor the selected verification data reaches 93.5% for water bodies, for non-water bodiesThere is a certain deviation in the accuracy of the classification, which is lower than90%, and the accuracy of the same threshold applied to another verification sample isless than the current accuracy.(2) A simple random forest classifier constructed based on the feature learning ofthe pixel, by learning the sample points of different time series and predicting theclassification to distinguish the water body from the non-water body. In order to avoidthe tedious experience threshold and selection, in terms of overall classification, theprediction and recognition accuracy of positive and negative categories of water andnon-water bodies are relatively stable and are higher than 92%. It can avoid theinfluence of misclassification of water body classification caused by the difference ofseasonal images. The random forest RF classifier with high comprehensive accuracy isused to classify and predict the historical images of Landsat multiple data sources inthe study area. The classification results are further optimized by algorithm to eliminate misclassification and further improve the extraction results. Finally, the image isstitched and cut to construct A dataset of water extraction from the main river of theAmu Darya.(3) The Landsat data set was used to extract the water body of the main stream ofthe Amu Darya River, and the water body extraction results of the main water bodiesof the main river of the Amu Darya River in April, July and September from 1986 to2018 were sorted and stitched. From the extraction results of the past years, the area ofthe main river of the Amu Darya River is: July> April> September. The annual averagewater area change from 1986 to 2018 has a high linear correlation, R² = 0.7837. Theoverall water area showed a downward trend. In 1986, the water body area of the mainriver was 965.102 Km². As of 2018, the water body area was 720.814 Km². The areareduction was about 25.31%, with an average decrease of 7.67% every 10 years.(4) Through comparative analysis of area changes, changes in the area of the main riverof the Amu Darya can be divided into three stages: 1986-2001, 2002-2008, and 2009-2018. Among them, 1986-2001 is the first stage. The area of the water body changesgreatly, the fluctuation is more obvious, the area of the water body increases in a smallrange, then decreases, and the overall trend is decreasing. From 2002 to 2008, it wassimilar to the previous stage, with a small range of extreme values, but the area changewas less volatile. The third phase from 2009 to 2018, the area of the water body declinedsteadily, with a year-on-year decline of about -1.6%. |
| Subject Area | 测绘工程 |
| Language | 中文 |
| Document Type | 学位论文 |
| Identifier | http://ir.xjlas.org/handle/365004/15431 |
| Collection | 中国科学院新疆生态与地理研究所 研究系统 |
| Affiliation | 中国科学院新疆生态与地理研究所 |
| First Author Affilication | 中国科学院新疆生态与地理研究所 |
| Recommended Citation GB/T 7714 | 李冬冬. 中亚阿姆河干流地表水体面积时空变化分析[D]. 北京. 中国科学院大学,2020. |
| Files in This Item: | There are no files associated with this item. | |||||
Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.
Edit Comment