植物生态学报 ›› 2020, Vol. 44 ›› Issue (6): 598-615.DOI: 10.17521/cjpe.2019.0347

所属专题: 生态遥感及应用 生物多样性

• 研究论文 • 上一篇    下一篇

利用聚类算法监测森林乔木物种多样性

衣海燕1,2, 曾源1,2,*(), 赵玉金3, 郑朝菊1, 熊杰1,2, 赵旦1   

  1. 1中国科学院空天信息创新研究院遥感科学国家重点实验室, 北京 100101
    2中国科学院大学, 北京 100049
    3中国科学院植物研究所植被与环境变化国家重点实验室, 北京 100093
  • 收稿日期:2019-12-12 接受日期:2020-02-07 出版日期:2020-06-20 发布日期:2020-03-26
  • 通讯作者: 曾源
  • 基金资助:
    国家重点研发计划项目(2016YFC0500201);国家自然科学基金(41671365)

Forest species diversity mapping based on clustering algorithm

YI Hai-Yan1,2, ZENG Yuan1,2,*(), ZHAO Yu-Jin3, ZHENG Zhao-Ju1, XIONG Jie1,2, ZHAO Dan1   

  1. 1State Key Laboratory of Remote Sensing Science, Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100101, China
    2University of Chinese Academy of Sciences, Beijing 100049, China
    3State Key Laboratory of Vegetation and Environmental Change, Institute of Botany, Chinese Academy of Sciences, Beijing 100093, China
  • Received:2019-12-12 Accepted:2020-02-07 Online:2020-06-20 Published:2020-03-26
  • Contact: ZENG Yuan
  • Supported by:
    National Key R&D Program of China(2016YFC0500201);National Natural Science Foundation of China(41671365)

摘要:

该研究基于机载激光雷达(LiDAR)和高光谱数据, 从森林物种叶片的生理化学源头探寻生化特征与光谱特征的内在关联, 探讨生化多样性、光谱多样性与物种多样性之间的响应机制, 选择最优植被指数并结合最优结构参数, 通过聚类方法构建森林物种多样性遥感估算模型, 在古田山自然保护区开展森林乔木物种多样性监测。研究结果表明: (1)从16种叶片生化组分中, 筛选出叶绿素a、叶绿素b、类胡萝卜素、叶片含水量、比叶面积、纤维素、木质素、氮、磷和碳可通过偏最小二乘法用叶片光谱有效模拟(R2 = 0.60-0.79, p < 0.01), 并选择有效的植被指数: 转换型吸收反射指数/优化型土壤调整指数(TCARI/OSAVI)、类胡萝卜素反射指数(CRI)、水波段指数(WBI)、比值植被指数(RVI)、生理反射指数(PRI)和冠层叶绿素浓度指数(CCCI)表征相应的最优生化组分; (2)基于机载LiDAR数据利用结合形态学冠层控制的分水岭算法获得高精度单木分离结果(R 2 = 0.77, RMSE = 16.48), 同时采用逐步回归方法从常用的森林结构参数中选取树高和偏度作为最优结构参数(R 2 = 0.32, p < 0.01); (3)基于6个最优植被指数和2个最优结构参数, 以20 m × 20 m为窗口通过自适应模糊C均值方法进行聚类, 实现了研究区森林乔木物种丰富度(Richness, R 2= 0.56, RMSE = 1.81)和多样性指数Shannon-Wiener (R 2 = 0.83, RMSE = 0.22)与Simpson (R 2 = 0.85, RMSE = 0.09)的成图。该研究在冠层尺度上获取了与物种多样性相关的生化、光谱和结构参数, 将单木个体作为最小单元, 利用聚类算法直接估算物种类别差异, 无需判定具体的树种属性, 是利用遥感数据进行区域尺度森林物种多样性监测与成图的实践, 可为亚热带地区常绿阔叶林的物种多样性监测提供借鉴。

关键词: 森林物种多样性, 自适应模糊C均值聚类, 高光谱, 激光雷达, 单木分离

Abstract:

Aims Monitoring forest species diversity continuously and efficiently is important to maintain ecosystem services and achieve sustainability and conservation goals. In this paper, we explored the relationship between leaf biochemical and spectral properties and their inner linkage with species diversity, then estimated the forest species diversity based on a clustering algorithm using airborne imaging spectroscopy and Light Detection and Ranging (LiDAR) data in the Gutianshan National Nature Reserve of China.
Methods Firstly, we isolated individual tree crowns (ITCs) with the watershed algorithm from the LiDAR data. Then we calculated the optimal vegetation indices (VIs) representing the key biochemical properties from the hyperspectral data and selected optimal structural parameters from commonly used LiDAR-derived structural parameters based on correlation and stepwise regression analysis with the field samples. Finally, a self-adaptive Fuzzy C-Means (FCM) clustering algorithm was applied to map the species diversity (i.e. Richness, Shannon-Wiener index and Simpson index) in the study area for each 20 m × 20 m moving window.
Important findings The results indicated that biochemical components (chlorophyll a & b, total carotenoids, equivalent water thickness, specific leaf area, cellulose, lignin, nitrogen, phosphorus and carbon) could be well quantified by leaf spectrum using partial least squares regression (R2 = 0.60-0.79, p < 0.01), and represented by hyperspectral VIs, namely, Transformed Chlorophyll Ratio Index/Optimization of Soil-adjusted Vegetation Index (TCARI/OSAVI), Carotenoid Reflectance Index (CRI), Water Band Index (WBI), Ratio Vegetation Index (RVI), Photochemical Reflectance Index (PRI) and Canopy Chlorophyll Concentration Index (CCCI). The individual tree isolation showed high accuracy (R 2 = 0.77, RMSE = 16.48). The correlation and stepwise regression analysis showed tree height and skewness were the optimal structural parameters among seven commonly used forest structural parameters (R 2 = 0.32, p < 0.01). The species diversity indices calculated from the self-adaptive FCM clustering algorithm based on the six VIs and two optimal structural parameters correlated well with the field measurements (species richness, R 2 = 0.56, RMSE = 1.81; Shannon-Wiener index, R 2 = 0.83, RMSE = 0.22; Simpson index, R 2 = 0.85, RMSE = 0.09). With the clustering method combined with crown-by-crown variations in hyperspectral biochemical VIs and LiDAR-derived structural parameters, we created continuous maps of forest species diversity in the examined subtropical forest without the need to identify specific tree species. Our case study in Gutianshan showed the potential of airborne hyperspectral and LiDAR data in mapping species diversity of the subtropical evergreen broad-leaved forest. It could also provide a pathway for monitoring the state and changes of forest biodiversity at regional scales.

Key words: forest species diversity, self-adaptive Fuzzy C-Means clustering algorithm, imaging spectroscopy, Light Detection and Ranging, individual tree isolation