干旱区研究 ›› 2024, Vol. 41 ›› Issue (4): 581-589.doi: 10.13866/j.azr.2024.04.05

• 水土资源 • 上一篇    下一篇

基于支持向量机的蓄水工程土地利用分类与动态变化

王军1,2(), 柴志福3, 马浩艳4, 赵志锰4, 邬佳宾1,2, 付卫平3   

  1. 1.中国水利水电科学研究院牧区水利科学研究所,内蒙古 呼和浩特 010020
    2.内蒙古阴山北麓草原生态水文国家野外科学观测研究站,北京 100038
    3.内蒙古自治区水利科学研究院,内蒙古 呼和浩特 010052
    4.华中师范大学,湖北 武汉 430079
  • 收稿日期:2023-08-15 修回日期:2024-01-19 出版日期:2024-04-15 发布日期:2024-04-26
  • 作者简介:王军(1987-),男,博士,高级工程师,主要从事水土资源利用与生态环境效应研究. E-mail: slwj1988@163.com
  • 基金资助:
    内蒙古自治区水利科技项目(NSK202104);内蒙古自然科学基金项目(2020MS05057);科技基础资源调查专项(2022FY101600);中国水利水电科学研究院基本科研业务费项目(MK0145B022021)

Land use type interpretation and dynamic changes due to water storage projects using support vector machine

WANG Jun1,2(), CHAI Zhifu3, MA Haoyan4, ZHAO Zhimeng4, WU Jiabin1,2, FU Weiping3   

  1. 1. China Institute of Water Resources and Hydropower Research, Institute of Water Resources for Pastoral Area, Hohhot 010020, Inner Mongolia, China
    2. Yinshanbeilu Grassland Eco-hydrology National Observation and Research Station, Beijing 100038, China
    3. Inner Mongolian Autonomous Region Research Institute on Hydraulic Sciences, Hohhot 010052, Inner Mongolia, China
    4. Central China Normal University, Wuhan 430079, Hubei, China
  • Received:2023-08-15 Revised:2024-01-19 Online:2024-04-15 Published:2024-04-26

摘要:

为进一步恢复和重建蓄水工程建成前后土地利用变化的历史过程,更好掌握和预报土地利用转移方向,本文利用支持向量机理论开展了土地利用类型解译的适应性研究,通过梳理土地利用动态变化,剖析了蓄水工程建成前后土地利用结构的自适应调节能力和演变方向。结果表明:(1) 依靠自学习和自适应等优势能力,支持向量机解译土地利用分类的总体精度高达91.7%、Kappa系数为0.90;除耕地生产者精度相对较低外,水体、林地等其他土地类型具有较高的分类识别能力。(2) 利用谷歌地球引擎(GEE)平台梳理土地利用类型演变过程发现,受”三北防护林”工程二阶段(2001—2020年)等项目实施影响,建设用地、林地面积出现较大增幅,其中,林地面积较2000年实施初期增加了近5倍。(3) 工程建设运行后林地和建设用地近2/3面积保持了原貌,水体和未利用土地受水利和城建工程影响,原貌类型超过65%以上面积变成了其他类型;“三北防护林”工程加快了林地面积的增加和草地植被覆盖度的提高,低覆盖度草地转移到中、高覆盖度草地的面积净增幅达48.0%、50.2%。

关键词: 土地利用, 支持向量机, 状态转移, 蓄水工程

Abstract:

In this study, to further restore and reconstruct the historical process of land use change before and after the construction of a water storage project and better grasp and forecast the direction of land use transfer, adaptive research on land use type interpretation was performed using the support vector machine theory. The adaptive adjustment ability and evolution direction of the land use structure before and after the construction of a water storage project were analyzed by examining the dynamic change in land use. The main conclusions were as follows: (1) The overall classification accuracy of the support vector machine for land use type interpretation is as high as 91.7%, and the Kappa coefficient is 0.90, depending on the advantages of self-learning and self-adaptation. In contrast with the relatively low accuracy observed for cultivated land producers, higher classification and recognition ability was observed other land types such as water bodies and forest land. (2) The Google Earth Engine (GEE) platform was used to examine the evolution process of land use types; it was found that the implementation of the second stage of the “Three-North Shelterbelt” project (2001-2020) significantly increased the area of construction land and forest land and increased the area of forest land by nearly five times compared with the initial stage of implementation in 2000. (3) Since the construction and operation of the project, nearly two-thirds of the area of forest land and construction land have maintained their original appearance, water bodies and unused land have been affected by water conservancy and urban construction projects, and more than 65% of the area has transformed from the original appearance type to other types. The “Three-North Shelterbelt” project accelerated the increase in forest area and improvement in grassland vegetation cover, and the net increase in the transformation of low-cover grassland to medium and high-cover grassland was 48.0% and 50.2%, respectively.

Key words: landuse, support vector machine, state transfer, water storage project