地球信息科学学报 ›› 2021, Vol. 23 ›› Issue (4): 680-691.doi: 10.12082/dqxxkx.2021.200161

• 遥感科学与应用技术 • 上一篇    下一篇

基于沙地指数模型的沙地监测方法

李宇君1,2(), 张磊1,*()   

  1. 1.中国科学院空天信息创新研究院 数字地球重点实验室,北京 100094
    2.中国科学院大学 电子电气与通信工程学院,北京 100049
  • 收稿日期:2020-04-03 修回日期:2020-06-16 出版日期:2021-04-25 发布日期:2021-06-25
  • 通讯作者: 张磊
  • 作者简介:李宇君(1996— ),女,山西太原人,硕士生,主要从事沙化土地分类研究。E-mail: liyj@radi.ac.cn
  • 基金资助:
    国家重点研发计划项目(2016YFC0500806)

Sandy Land Monitoring Method based on Classification Index Model

LI Yujun1,2(), ZHANG Lei1,*()   

  1. 1. Key Laboratory of Digital Earth, Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China
    2. School of Electronic, Electrical and Communication Engineering, University of Chinese Academy of Sciences, Beijing 100049, China
  • Received:2020-04-03 Revised:2020-06-16 Online:2021-04-25 Published:2021-06-25
  • Contact: ZHANG Lei
  • Supported by:
    National Key Research and Development Programof China(2016YFC0500806)

摘要:

沙漠化是干旱、半干旱地区的重要生态环境问题,我国西北地区沙漠化土地分布广泛,加剧的沙漠化问题影响着区域经济和社会的发展,遥感技术的进步为沙漠化评估与制图提供了重要手段。本文以内蒙古自治区浑善达克沙地为研究区,基于面向对象方法,对研究区Landsat8 OLI影像进行沙地最优尺度分割。以分割对象为基础,实验在冬夏季影像上分层分阶段提取沙地。在冬季影像上,本文提出新比值型指数RSBI(Ratio Soil Brightness Index)对沙地进行提取,精度较SBI指数提高4.11%。后基于改进型植被覆盖度指数(FMSAVI)与反照率(Albedo)构建二维特征空间,建立沙地分类指数模型(DCI),对夏季影像沙区分类。该方法总体精度为83.24%,较NDVI-Albedo二维特征空间模型精度提高5.59%,较FMSAVI模型提高16.20%。本文结合RSBI指数与FMSAVI-Albedo特征空间反演的DCI指数模型来提取沙地信息并对沙地分类,减少了沙地提取误差,提高了分类精度,为沙地信息的研究提供了新思路。

关键词: 面向对象, 沙化土地, 沙化指数, 特征空间, 分类模型, NDVI, MSAVI, Albedo

Abstract:

Desertification has become one of the most serious environmental problems facing the world today, which is also one of serious environmental threats in southwest China. Remote sensing technology offers substantial information for assessment of desertification. Based on remote sensing technology, our study aimed to extract the sandy land information and map the land cover classification in Otindag sandy land. We proposed a new sandy land extraction index and a new sandy land classification model to extract and classify sandy land at different times and levels on the basis of object-oriented classification. First, we segmented the image of study area using optimal segmentation scales evaluated by ESP2 in eCognition. The results show that the optimal segmentation scales for sandy land, shrub, and herb were 200, 145, and 185, respectively. Then in order to extract sandy land, we proposed the Ratio Soil Brightness Index (RSBI). Compared with SBI, RSBI can avoid the misclassification between water body and non-sandy land and hence had higher precision. In order to avoid the misclassification between the fixed sandy land and areas with relatively high vegetation coverage, we extracted sandy land at different times and levels. We first used RSBI to extract sandy land in the winter image. The index of MSAVI was introduced to derive the coverage of vegetation, named as FMSAVI. FMSAVI and Albedo were selected to construct two-dimensional feature space in sandy land area extracted by RSBI in the summer image. After normalization of FMSAVI and Albedo, a liner regression analysis was performed. Based on this, a Desertification Classification Index (DCI) was developed. The DCI was a collection of lines perpendicular to the fit line. Different positions of the lines mean different sandy land classifications. A map of sandy land classification was generated and we grouped sandy land into three classes, namely shifting sandy land, semi-fixed sandy land, and fixed sandy land. Our result show that the overall accuracy of DCI model was 83.24%, higher than traditional methods using modified vegetation coverage (67.04%) and NDVI-Albedo feature space (77.65%) to classify sandy land. We proposed the RSBI to extract sandy land in the winter image and constructed FMSAVI-Albedo feature space to classify sandy land in summer image. These two indexes improved classification accuracy. Our methods are simple, robust, powerful, and easy to use for the extraction and classification of sandy land.

Key words: object-oriented, sandy land, sandy land index, feature space, classification model, NDVI, MSAVI, Albedo