草地学报 ›› 2026, Vol. 34 ›› Issue (2): 559-572.DOI: 10.11733/j.issn.1007-0435.2026.02.018

• 研究论文 • 上一篇    

基于无人机多源遥感数据的灌丛草原地上生物量估算

马磊超1,2, 王兴1, 杨秀春1, 杨东1, 王子超1, 邢晓语1   

  1. 1. 北京林业大学草业与草原学院, 北京 100083;
    2. 中国地质调查局自然资源综合调查指挥中心, 北京 100055
  • 收稿日期:2025-03-28 修回日期:2025-05-16 发布日期:2026-01-22
  • 通讯作者: 杨秀春,E-mail:yangxiuchun@bjfu.edu.cn
  • 作者简介:马磊超(1981-),男,汉族,河南沈丘人,硕士研究生,主要从事草原遥感研究,E-mail:leichaoma@126.com;
  • 基金资助:
    国家自然科学基金项目(41571105)资助

Aboveground Biomass Estimation in Shrub-Encroached Grasslands by Using Multi-Source UAV Remote Sensing

MA Lei-chao1,2, WANG Xing1, YANG Xiu-chun1, YANG Dong1, WANG Zi-chao1, XING Xiao-yu1   

  1. 1. School of Grassland Science, Beijing Forestry University, Beijing 100083, China;
    2. Command Center for Natural Resources Comprehensive Survey, China Geological Survey, Beijing 100055, China
  • Received:2025-03-28 Revised:2025-05-16 Published:2026-01-22

摘要: 气候变化与人类活动加剧了全球草原灌丛化,严重威胁草原生态系统健康,亟需高效精准的生物量监测技术。本研究以锡林郭勒盟灌丛草原为研究区,基于无人机激光雷达(LiDAR)和高光谱数据,结合地面实测数据,采用逐步多元线性回归、支持向量机和随机森林等方法探究了灌丛草原地上生物量(Above-ground biomass,AGB)的最佳反演模型。结果表明,基于LiDAR数据计算的草层平均高度、高光谱数据的红边区间反射率与AGB显著相关。高光谱在草本AGB反演中优于LiDAR精度(R2=0.87 vs. 0.24),而LiDAR较高光谱则更适用于灌木AGB反演(R2=0.58 vs. 0.47)。数据融合能显著提升模型精度,LiDAR和高光谱数据多源特征融合的随机森林模型使草本和灌木AGB的R2分别达到0.88和0.64。相关成果可为灌丛草原可持续利用和精细化管理提供重要科学支撑。

关键词: 草原地上生物量, 激光雷达, 高光谱, 机器学习, 灌丛草原

Abstract: Shrub encroachment in grasslands, intensified by climate change and anthropogenic activities, represents a substantial threat to the health of grassland ecosystems. This underscores the pressing need for effective and precise biomass monitoring methodologies. The present study was carried out in the shrub-encroached grasslands of the Xilingol League, to investigate optimal above ground biomass (AGB) estimation model by utilizing unmanned aerial vehicle-based LiDAR and hyperspectral data validated against field measurements. Stepwise multiple linear regression, support vector machine, and random forest algorithms were employed for analysis. The results demonstrated a robust correlation between AGB and both LiDAR-derived mean canopy height and hyperspectral reflectance within the red-edge spectral region. Hyperspectral data demonstrated superior accuracy to LiDAR for herbaceous AGB inversion (R2=0.87 vs. 0.24), whereas LiDAR performed better than hyperspectral data for shrub AGB inversion (R2=0.58 vs. 0.47). Data fusion significantly improved the model performance, with the random forest model integrating features from both LiDAR and hyperspectral data achieving R2 values of 0.88 and 0.64 for herbaceous and shrub AGB, respectively. The related findings offer critical scientific support for the sustainable utilization and precise management of shrub-encroached grasslands.

Key words: Grassland above-ground biomass, LiDAR, Hyperspectral, Machine learning, Shrub-encroached grassland