欢迎访问《草业学报》官方网站,今天是

草业学报 ›› 2026, Vol. 35 ›› Issue (2): 1-14.DOI: 10.11686/cyxb2025108

• 研究论文 • 上一篇    下一篇

基于无人机和Landsat数据的近30年三江源地区土地退化动态监测

赵琳兴1(), 王雁鹤1, 王子超2, 徐马强1(), 李泽宇1, 祁昌贤1, 崔宝祖1, 王宗保1   

  1. 1.中国地质调查局西宁自然资源综合调查中心,青海 西宁 810000
    2.北京林业大学草业与草原学院,北京 100083
  • 收稿日期:2025-03-27 修回日期:2025-04-21 出版日期:2026-02-20 发布日期:2025-12-24
  • 通讯作者: 徐马强
  • 作者简介:Corresponding author. E-mail: 651110732@qq.com
    赵琳兴(1992-),男,青海西宁人,本科。E-mail: zhaolx9208@163.com
  • 基金资助:
    中国地质调查局项目(DD20243409)

Dynamic monitoring of land degradation in the Three-River Headwaters Region over the past 30 years using unoccupied aerial vehicle imagery and Landsat data

Lin-xing ZHAO1(), Yan-he WANG1, Zi-chao WANG2, Ma-qiang XU1(), Ze-yu LI1, Chang-xian QI1, Bao-zu CUI1, Zong-bao WANG1   

  1. 1.Xining Center of Natural Resources Comprehensive Survey,China Geological Survey,Xining 810000,China
    2.School of Grassland Science,Beijing Forestry University,Beijing 100083,China
  • Received:2025-03-27 Revised:2025-04-21 Online:2026-02-20 Published:2025-12-24
  • Contact: Ma-qiang XU

摘要:

土地退化严重威胁我国生态系统稳定与粮食安全。三江源地区作为西部重要的生态屏障,面临突出的土地退化问题,影响区域生态安全与社会经济发展。本研究基于实地采样、无人机与Landsat数据,随机森林(RF)、支持向量机(SVM)、分类和回归树模型(CART),构建多源数据的土地退化监测框架,监测近30年(1993、2003、2013、2023年)三江源地区土地退化动态,并分析其时空演变特征。结果表明:1)无人机与卫星数据结合使用可以明显提高退化识别精度,基于“光谱-植被指数-地形”特征的随机森林模型精度最优,土地沙化识别精度达94.73%,F1分数为95.85%,“黑土滩”型退化识别精度达90.98%,F1分数为95.18%。2)1993-2023年,未退化与“黑土滩”型退化面积先增后减,盐渍化面积呈波动变化,先增加后减少再增加,沙化面积持续减少,各类型面积稳定不变的面积占比超1/2。3)总体上,黑土滩和沙化等级呈减轻趋势,重度黑土滩与中度沙化面积明显减少;轻中度盐渍化变化较小,重度盐渍化面积下降。本研究可为生态脆弱区土地退化监测提供新思路,并为区域生态保护与可持续发展提供科学依据。

关键词: 三江源地区, 土地退化, 无人机, Landsat, 遥感监测

Abstract:

Land degradation is a critical ecological and environmental issue that threatens ecosystem stability and food security in China. As a key ecological barrier in western China, the Three-River Headwaters Region is undergoing severe land degradation, which presents significant challenges for regional ecological security and socioeconomic development. Based on field sampling, unoccupied aerial vehicle (UAV) imagery, and Landsat data, this study established a multi-source data framework for monitoring land degradation by employing random forest (RF), support vector machine (SVM), and classification and regression tree (CART) models. The framework was applied to monitor land degradation dynamics in the Three-River Headwaters Region over the past three decades (1993-2003, 2003-2013, and 2013-2023), and to analyze its spatiotemporal evolution patterns. The results indicate that: 1) The integration of UAV and satellite data significantly improved the accuracy of degraded land detection. Among the models tested in this study, the RF model based on spectral and vegetation indexes and topographic features achieved the highest accuracy. Specifically, the accuracy of identifying desertified land reached 94.73% with an F1-score of 95.85%, while the accuracy of detecting degraded black soil beach land reached 90.98% with an F1-score of 95.18%. 2) From 1993 to 2023, the areas of non-degraded land and degraded black soil beach land initially increased and then decreased. The area of salinized land showed a fluctuating trend-increasing initially, then decreasing, and increasing again, while the area of desertified land continuously declined. For all degradation types, more than half of the affected areas remained in a stable state throughout the monitoring period. 3) Overall, both the severity of black soil beach degradation and desertification showed a decreasing trend, with substantial reductions in the area of severely degraded black soil land moderately desertified land. In contrast, there were only small changes in the area of mildly and moderately salinized land, but a notable decrease in the area of severely salinized land.

Key words: Three-River Headwaters Region, land degradation, unoccupied aerial vehicles, Landsat, remote sensing monitoring