img

官方微信

高级检索

中国沙漠 ›› 2026, Vol. 46 ›› Issue (4): 204-216.DOI: 10.7522/j.issn.1000-694X.2026.00055

• • 上一篇    下一篇

基于kNDVI-Albedo-SWCI三维特征空间与XGBoost-SHAP的科尔沁沙地沙漠化时空格局与驱动因素分析

曹婉颖1,2(), 段翰晨1, 贾晓鹏1()   

  1. 1.中国科学院西北生态环境资源研究院 干旱区生态安全与可持续发展全国重点实验室,甘肃 兰州 730000
    2.中国科学院大学,北京 100049
  • 收稿日期:2026-01-07 修回日期:2026-04-05 出版日期:2026-07-20 发布日期:2026-08-27
  • 通讯作者: 贾晓鹏
  • 作者简介:曹婉颖(2002—),女,河南许昌人,硕士研究生,主要从事干旱半干旱区沙漠化遥感监测研究。E-mail: caowy2023@163.com
  • 基金资助:
    国家自然科学基金项目(42271316);中国科学院“西部之光”计划项目(E5290202);干旱区生态安全与可持续发展全国重点实验室青年交叉创新团队项目(E4500125)

Spatiotemporal patterns and driving factors of aeolian desertification in the Horqin Sandy Land based on kNDVI-Albedo-SWCI three-dimensional feature space and XGBoost-SHAP analysis

Wanying Cao1,2(), Hanchen Duan1, Xiaopeng Jia1()   

  1. 1.State Key Laboratory of Ecological Safety and Sustainable Development in Arid Lands,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:2026-01-07 Revised:2026-04-05 Online:2026-07-20 Published:2026-08-27
  • Contact: Xiaopeng Jia

摘要:

科尔沁沙地是中国北方典型的农牧交错区,也是“三北”防护林工程的重要实施区域,其沙漠化演变过程复杂,是开展沙漠化研究的典型区域。为精准揭示科尔沁沙地沙漠化时空演变规律及其驱动机制,本研究创新性地构建了kNDVI-Albedo-SWCI三维特征空间模型,实现对科尔沁沙地沙漠化的时空动态监测。在此基础上,引入XGBoost模型对沙漠化驱动因子进行解析,并结合SHAP方法定量分析各自然与人为因子对沙漠化演变的相对贡献。结果表明:(1)引入湿度分量后,kNDVI-Albedo-SWCI三维特征空间模型能够综合表征植被、反照率与地表水分信息,有助于刻画复杂地表环境下的沙漠化特征。(2)基于kNDVI-Albedo-SWCI的三维特征空间模型能够实现对科尔沁沙地沙漠化程度的有效识别,沙漠化分类总体精度为0.80,Kappa系数为0.75,表明该模型在沙漠化信息提取中具有较好的稳定性与可靠性。(3)从时序变化特征来看,2000—2025年科尔沁沙地沙漠化面积总体呈减少趋势,沙漠化逆转特征明显,反映出区域生态修复措施取得了积极成效。(4)基于XGBoost-SHAP的驱动力分析结果表明,土壤质地和实际蒸散发在各驱动因子中重要性占比较高,是影响科尔沁沙地沙漠化演变的关键因素。本研究构建的三维特征空间模型与XGBoost-SHAP可解释机器学习方法相结合,可为干旱半干旱区沙漠化动态监测、驱动机制分析及防沙治沙工程成效评估提供可靠的技术支撑。

关键词: 三维特征空间, XGBoost-SHAP, 沙漠化监测, 驱动因素, 科尔沁沙地

Abstract:

The Horqin Sandy Land, located in northern China, represents a quintessential agro-pastoral ecotone and serves as a critical area for the implementation of the Three-North Shelter Forest Program. Due to its intricate desertification evolution processes, it is an exemplary region for conducting desertification research. In order to precisely elucidate the spatiotemporal evolution patterns and driving mechanisms of desertification within the Horqin Sandy Land, this study introduces an innovative three-dimensional feature space model, denoted as kNDVI-Albedo-SWCI, to facilitate the dynamic monitoring of desertification across spatial and temporal dimensions. Building upon this framework, the XGBoost model is employed to examine the driving factors of desertification, and the SHAP method is utilized to quantitatively assess the relative contributions of both natural and anthropogenic factors to the evolution of desertification. The findings indicate that: (1) The incorporation of the moisture component within the kNDVI-Albedo-SWCI three-dimensional feature space model enables a comprehensive characterization of vegetation, albedo, and surface water information, thereby enhancing the depiction of desertification characteristics in complex surface environments. (2) The three-dimensional feature space model, utilizing kNDVI-Albedo-SWCI, demonstrates a robust capability in accurately identifying the degree of desertification in the Horqin Sandy Land, achieving an overall classification accuracy of 0.80 and a Kappa coefficient of 0.75. This indicates the model's substantial stability and reliability in extracting desertification information. (3) From a temporal perspective, the desertified area within the Horqin Sandy Land has generally exhibited a declining trend from 2000 to 2025, with marked signs of desertification reversal, suggesting that regional ecological restoration initiatives have yielded positive outcomes. (4) The driving force analysis, conducted using XGBoost-SHAP, highlights that soil texture and actual evapotranspiration are among the most significant driving factors, playing a crucial role in influencing the desertification dynamics of the Horqin Sandy Land. The integration of the three-dimensional feature space model developed in this study with the interpretable machine learning approach of XGBoost-SHAP offers reliable technical support for dynamic monitoring of desertification, analysis of driving mechanisms, and evaluation of the effectiveness of sand prevention and control projects in arid and semi-arid regions.

Key words: three-dimensional feature space, XGBoost-SHAP, aeolian desertification monitoring, driving factors, Horqin Sandy Land

中图分类号: