地球信息科学学报 ›› 2022, Vol. 24 ›› Issue (2): 391-404.doi: 10.12082/dqxxkx.2022.210138

• 遥感科学与应用技术 • 上一篇    下一篇

雅鲁藏布江中部流域土地沙化遥感识别

詹琪琪1,2(), 赵伟1,*(), 杨梦娇1,2, 付浩1,3, 李昕娟1,2, 熊东红1   

  1. 1.中国科学院、水利部成都山地灾害与环境研究所,成都 610041
    2.中国科学院大学,北京 100049
    3.成都理工大学地球科学院学院,成都 610059
  • 收稿日期:2021-03-17 修回日期:2021-04-22 出版日期:2022-02-25 发布日期:2022-04-25
  • 通讯作者: *赵 伟(1984— ),男,江西上高人,博士,研究员,主要从事山地地表水热过程遥感监测与应用研究。 E-mail: zhaow@imde.ac.cn
  • 作者简介:詹琪琪(1998— ),女,四川广安人,硕士生,主要从事土地沙化遥感监测研究。E-mail: zhanqq@imde.ac.cn
  • 基金资助:
    第二次青藏高原综合科学考察研究(2019QZKK0404);国家自然科学基金项目(42071349);四川省科技计划资助(2020JDJQ0003);中国科学院“西部之光”西部青年学者A类(2016333)

Identification of Sandy Land in the Midstream of the Yarlung Zangbo River

ZHAN Qiqi1,2(), ZHAO Wei1,*(), YANG Mengjiao1,2, FU Hao1,3, LI Xinjuan1,2, XIONG Donghong1   

  1. 1. Institute of Mountain Hazards and Environment, Chinese Academy of Sciences, Chengdu 610041, China
    2. University of Chinese Academy of Sciences, Beijing 100049, China
    3. College of Earth Sciences, Chengdu University of Technology, Chengdu 610059, China
  • Received:2021-03-17 Revised:2021-04-22 Online:2022-02-25 Published:2022-04-25
  • Supported by:
    The Second Tibetan Plateau Scientific Expedition and Research Program (STEP)(2019QZKK0404);The National Natural Science Foundation of China(42071349);Sichuan Science and Technology Program(2020JDJQ0003);The Chinese Academy of Sciences "Light of West China" Program(2016333)

摘要:

雅鲁藏布江中部流域长期遭受土地沙化侵蚀,采取有效手段进行沙化土地信息快速识别,跟踪土地沙化现状和动态发展,是土地沙化防治的基本前提。遥感数据因其快速、大范围、高精度监测等特点已被广泛应用于土地沙化监测。为降低该区域沙化土地破碎化分布特征以及广泛分布的稀疏植被地表对沙化土地遥感识别带来的不确定性,本文利用Google Earth Engine平台获取2019年秋季雅鲁藏布江中部流域Landsat无云遥感影像,基于面向对象的分类思想,充分提取沙化土地的光谱、几何和地形特征,根据不同的分类器构建4种分类方案,包括单一分类器(支持向量机、决策树、最近邻)分类以及组合分类法分类,提取雅江中游河谷地区沙化土地信息并验证不同方案的提取精度。结果表明:① 利用面向对象组合分类模型提取的沙化土地信息效果最佳,总体精度高达91.38 %,Kappa系数为0.82;② 相较于采用单一分类器(支持向量机、最近邻和决策树分类)的面向对象分类方法,组合分类模型能更有效地识别破碎化的小面积沙化土地,降低沙化土地与稀疏植被地表的混淆情况,提高分类可靠性;③ 基于面向对象组合分类模型反演得到雅鲁藏布江中部流域2019年沙化土地分布信息,土地沙化面积达299.61 km2,总体上呈现沿河谷的带状不连续分布,且集中分布于河流北岸以及靠近河道的阳坡、低海拔地区。本研究可为土地沙化遥感监测提供新思路,其应用可服务于雅鲁藏布江中部流域土地沙化预防和治理工作。

关键词: 土地沙化, Landsat, 遥感分类, 空间分布特征, 雅鲁藏布江中游, 面向对象分类, GEE, 方法比较

Abstract:

As the political, cultural, and economic core zone of Tibet, the middle part of the Yarlung Zangbo River basin has suffered from serious land desertification for a long time, posing obvious negative impacts for the local socio-economic development and natural environmental protection. It is the basic precondition for land desertification prevention and control to obtain spatial distribution, track the status quo, and analyze dynamic development of sandy land desertification. Remote sensing images have been widely used in the dynamical monitoring of sandy land information due to its characteristics of fast, large-scale, and high precision. In order to reduce the uncertainty caused by the fragmented distribution of sandy land and the large area of sparsely vegetation surfaces in this region, this study developed an object-oriented integrated classification method, combining decision tree classifier and nearest neighbor classifier. The method is based on the Landsat cloud-free images from Google Earth Engine platform. The spectral, geometrical, and topographic features of sandy land were extracted as the inputs of the method to differentiate sandy land from other land cover types, including the sparsely vegetated surfaces with similar spectral pattern as sandy land. The results indicated that, firstly, with the validation sample data collected from the Google Earth high-resolution images and field investigation, the integrated classification method has the highest overall accuracy of 92.38 % and the Kappa coefficient of 0.82. Secondly, compared with other single classifier classification methods, such as supported vector machine, nearest neighbor, and decision tree, the integrated classification method achieved the best classification results in identifying sandy land with small area. In addition, it also reduced the confusion between sandy land and sparsely vegetated surfaces, thus increased the reliability of the classification results. Thirdly, the sandy land in the middle part of the Yarlung Zangbo River basin in 2019 was mapped based on the proposed method with an area of 299.66 km2, displaying a zonal and fragmented pattern along river valleys and concentrating on the northern bank of rivers and the regions with southern aspect, low altitude, and close to riverways. This study provides a new direction for sandy land desertification monitoring with remote sensing data, and its application can also serve the prevention and management of sandy land desertification in the middle part of the Yarlung Zangbo River basin.

Key words: land desertification, Landsat, classification, spatial pattern, the midstream of the Yarlung Zangbo River, object-oriented classification, GEE, method comparison