img

官方微信

高级检索

中国沙漠 ›› 2023, Vol. 43 ›› Issue (6): 60-70.DOI: 10.7522/j.issn.1000-694X.2023.00072

• • 上一篇    下一篇

结合GEE平台与机器学习算法的荒漠信息提取

芦瑞杰1,2(), 刘树林1(), 康文平1, 冯坤1, 郭紫晨1, 支莹1,2   

  1. 1.中国科学院西北生态环境资源研究院 沙漠与沙漠化重点实验室,甘肃 兰州 730000
    2.中国科学院大学,北京 100049
  • 收稿日期:2023-03-23 修回日期:2023-05-31 出版日期:2023-11-20 发布日期:2023-11-30
  • 通讯作者: 刘树林
  • 作者简介:刘树林(E-mail: liusl@lzb.ac.cn
    芦瑞杰(1999—),男,山西临汾人,硕士研究生,主要研究方向为荒漠生态遥感。E-mail: luruijie@nieer.ac.cn
  • 基金资助:
    第二次青藏高原综合科学考察研究项目(2019QZKK0305)

Combining the GEE platform and machine learning algorithm for desert information extraction

Ruijie Lu1,2(), Shulin Liu1(), Wenping Kang1, Kun Feng1, Zichen Guo1, Ying Zhi1,2   

  1. 1.Key Laboratory of Desert and Desertification,Northwest Institute of Eco-Environment and Resources,Chinese Academy of Sciences,Lanzhou 730000,China
    2.University of Chinese Academy of Sciences,Beijing 100049,China
  • Received:2023-03-23 Revised:2023-05-31 Online:2023-11-20 Published:2023-11-30
  • Contact: Shulin Liu

摘要:

快速、准确获取不同荒漠类型的分布信息,对于环境保护和生态修复有着重要意义。然而,受光谱与分辨率等因素影响,目前关于不同荒漠类型信息提取研究存在明显不足。选择青海省都兰县作为典型区,基于谷歌地球引擎(Google Earth Engine,GEE)并结合多源数据对研究区荒漠进行了分类,对比评估了不同分类特征组合应用于3种机器学习方法(RF、SVM、CART)的分类性能。结果表明:(1) RF的性能要优于CART和SVM,使用RF分类器并以光谱特征、雷达特征、地形特征、纹理特征为分类依据的总体分类精度最高,整体准确度为95.68%,Kappa系数为0.95,FM得分为94.28%,获得的都兰县荒漠面积为29 039 km2。(2) 在特征重要性得分评估中,海拔与VH对荒漠分类的贡献比较突出,其他特征则差异不大。(3) 在使用光谱数据的基础上,雷达特征对识别砾质荒漠与壤土荒漠的贡献突出,而地形特征则更适用于识别其他类型的荒漠。

关键词: 荒漠分类, 机器学习, GEE平台, 都兰县

Abstract:

The rapid and accurate mapping of desert type distribution is of great significance for environmental protection and ecological restoration. However, due to the influence of spectral and resolution factors, the current research on the extraction of information about different desert types is obviously insufficient. In this study, Dulan County of Qinghai Province was selected as a typical area. The desert was classified based on GEE platform and multi-source data, and the classification performance of different classification features combined with three machine learning methods (RF, SVM and CART) was compared and evaluated. The results show that (1) RF outperforms CART and SVM, and the overall classification accuracy using the RF classifier and spectral features, radar features, terrain features and texture features is the highest, with an overall accuracy of 95.68%, a Kappa coefficient of 0.95, an FM score of 94.28%, and an obtained desert area of Dulan County is 29 039 km2. (2) In the assessment of feature importance scores, elevation and VH contribute more to the desert classification, while other features do not contribute much. (3) Based on the use of spectral data, radar features are the most helpful for identifying gravel and loamy deserts, while terrain features are more suitable for identifying other types of deserts.

Key words: desert classification, machine learning, GEE platform, Dulan County

中图分类号: