Acta Agrestia Sinica ›› 2025, Vol. 33 ›› Issue (10): 3362-3371.DOI: 10.11733/j.issn.1007-0435.2025.10.024

Previous Articles    

Assessing Vegetation Coverage of Alpine Grassland Based on Unmanned Aerial Vehicle and Sentinel-2 Data

XU Gan-jun1,2, WU Sheng-yi2, NIU Yue-chuan3, YAN Wen-de1, KANG Xiao-ming4,5, ZHANG Xiao-dong4,5   

  1. 1. Central South University of Forestry Technology, Changsha, Hunan Province 410004, China;
    2. Key Laboratory of National Forestry and Grassland Administration on Ecological Hydrology and Disaster Prevention in Arid Regions, Northwest Survey and Planning Institute, National Forestry and Grassland Administration, Xi'an, Shanxi Province 710048, China;
    3. College of Life Sciences, University of Chinese Academy of Sciences, Beijing 100049, China;
    4. State Key Laboratory of Wetland Conservation and Restoration, Beijing Key Laboratory of Wetland Services and Restoration, Institute of Ecological Conservation and Restoration, Chinese Academy of Forestry, Beijing 100091, China;
    5. Sichuan Ruoergai Alpine Wetland Ecosystem Observation and Research Station, Aba, Sichuan Province 624500, China
  • Received:2024-12-04 Revised:2025-01-17 Published:2025-10-17

基于无人机和Sentinel-2数据的高寒草地植被覆盖度评估

徐干君1,2, 吴胜义2, 牛阅川3, 闫文德1, 康晓明4,5, 张骁栋4,5   

  1. 1. 中南林业科技大学, 湖南 长沙 410004;
    2. 国家林业和草原局西北调查规划院/旱区生态水文与灾害防治国家林业局重点实验室, 陕西 西安 710048;
    3. 中国科学院大学生命科学学院, 北京 100049;
    4. 中国林业科学研究院生态保护与修复研究所/湿地环境保护与生态修复全国重点实验室/湿地生态功能与恢复北京市重点实验室, 北京 100091;
    5. 四川若尔盖高寒湿地生态系统定位观测研究站, 四川 阿坝 624500
  • 通讯作者: 张骁栋,E-mail:zhangxiaod@caf.ac.cn
  • 作者简介:张骁栋,E-mail:zhangxiaod@caf.ac.cn
  • 基金资助:
    国家林业和草原局西北调查规划院科技创新项目(XBY-KJCX-2023-02);中国林业科学研究院生态保护与修复研究所科技项目(STSTC2023005)资助

Abstract: Multispectral remote sensing indices have been widely used to estimate grassland fractional vegetation cover (FVC) on the Qinghai-Tibet Plateau. However, evaluating the accuracy of different remote sensing models is challenging due to the lack of ground verification of pixel matching corresponding to the remote sensing pixels. In this study, based on Sentinel-2 multispectral data and near-ground RGB imagery captured by an unmanned aerial vehicle (UAV), a pixel dichotomy model (PD model) and a random forest model (RF model) were developed to estimate the FVC of Maqin County in the Yellow River source region in 2023. This study compared the inversion accuracy and spatial discrepancies in FVC predictions between the two models. The results showed that the goodness of fitting degree for the PD model and RF model were 0.68 and 0.78, respectively, with the RF model overall providing higher accuracy in FVC estimation. Both models displayed similar spatial patterns in FVC distribution across Maqin County, with higher values in the eastern and southern regions, and lower values in the northern and western regions. Notably, substantial differences were found in environmentally extreme regions, especially in areas with elevations exceeding 4800 m or FVC values below 0.2, where the RF model tended to predict higher values than the PD model. This study provides a methodological support for alpine grassland ecological restoration projects and regional fine-scale assessments.

Key words: Qinghai-Tibet Plateau, Vegetation indices, Pixel dichotomy model, Random forest, Binarization

摘要: 遥感多光谱指数常用于反演青藏高原草地植被覆盖度(Fractional vegetation cover,FVC),然而由于缺乏像元匹配的地面验证,不同遥感模型反演效果难以评估。本研究基于Sentinel-2多光谱数据和无人机近地面可见光影像,构建像元二分模型(Pixel dichotomy model, PD模型)和随机森林模型(Random forest model,RF模型)并反演黄河源地区玛沁县2023年区域FVC,比较两种模型的精度和区域差异。结果表明:PD模型和RF模型的拟合度分别为0.68和0.78,总体上RF模型精度优于PD模型。两种模型反演玛沁县FVC的整体空间格局一致,表现为东部和南部较高,而北部和西部较低。两种模型在极值区域反演差异较大,尤其在海拔高于4800 m或FVC低于0.2区域RF模型的估值高于PD模型。本研究能为高寒草地生态恢复项目及区域精细化评估提供方法学支撑。

关键词: 青藏高原, 植被指数, 像元二分模型, 随机森林, 二值化

CLC Number: