普洱季风常绿阔叶林乔木物种多样性高分辨率遥感估测  

Applying high resolution remote sensing to assess tree species diversity of monsoonal broad-leaved evergreen forest in Pu′er City

作　　者：刘鲁霞 庞勇[2] 桑国庆 李增元[2] 胡波 LIU Luxia;PANG Yong;SANG Guoqing;LI Zengyuan;HU Bo(School of Water Conservancy and Environment,University of Jinan,Jinan 250022,China;Key Laboratory of Forestry Remote Sensing and Information System,National Forestry and Grassland Administration,Institute of Forest Resource Information Techniques,Chinese Academy of Forestry,Beijing 100091,China)

机构地区：[1]济南大学水利与环境学院,济南250024 [2]中国林业科学研究院资源信息所国家林业和草原局林业遥感与信息技术重点实验室,北京100091

出　　处：《生态学报》2022年第20期8398-8413,共16页Acta Ecologica Sinica

基　　金：国家自然科学基金项目(41801330,31570546);高分辨率对地观测系统重大专项(30-Y30A02-9001-20/22-7);山东省自然科学基金(ZR201911160131)。

摘　　要：季风常绿阔叶林是我国南亚热带典型的地带性植被,也是云南省普洱地区重要森林类型。季风常绿阔叶林乔木物种多样性遥感估测对研究区域尺度生物多样性格局及其规律具有重要作用。根据光谱异质性假说和环境异质性假说,首先使用1m空间分辨率的机载高光谱数据和激光雷达数据提取了光谱多样性特征和垂直结构特征。然后利用基于随机森林算法的递归特征消除方法选择对研究区森林乔木物种多样性指数具有较好解释能力的遥感特征,并对Shannon-Winner物种多样性指数进行建模、制图。研究结果表明:(1)基于机载LiDAR数据提取的垂直结构特征和机载高光谱数据提取的光谱多样性特征均对研究区森林乔木物种多样性具有较好的解释能力,随机森林模型估测结果分别为R^(2)=0.48,RMSE=0.46和R^(2)=0.5,RMSE=0.45;两种数据源融合可以进一步提高遥感数据的森林乔木物种多样性估测精度,随机森林估测模型R^(2)和RMSE分别为0.69和0.37。(2)机载激光雷达数据对研究区针阔混交林乔木物种多样性的估测能力优于机载高光谱数据。(3)机器学习方法有助于从高维遥感数据特征中选择适合于森林乔木物种多样性建模的少量特征。该研究在云南普洱开展对季风常绿阔叶林的遥感估测研究,可为森林生物多样性调查提供补充手段,有助于森林生物多样性大尺度、长期动态监测。The monsoonal broad-leaved evergreen forest is an important vegetation type of the Pu′er area which is located in the southern Yunnan Province, China. Tree species diversity mapping of monsoon evergreen broad-leaved forest plays an important role in studying the patterns of biodiversity at the regional scale. Remote sensing is an efficient alternative to the traditional field work to map tree species diversity over large areas. We tested the utility of using spatial heterogeneity in the airborne hyperspectral reflectance spectrum and structure heterogeneity in airborne LiDAR point cloud to model tree species diversity of monsoonal broad-leaved evergreen forest in Pu′er area. Shannon-Winner Index of tree species was calculated from field measurement. According to the spectral heterogeneity hypothesis and environmental heterogeneity hypothesis, the spectral diversity and structural diversity features were firstly extracted using airborne hyperspectral data and Lidar data. Then, the Random Forest based Recursive Feature Elimination(RF-RFE) was used to select the valuable airborne remote sensing features for forest tree species diversity modeling. At last, Random Forest regression model was used for modeling and mapping. Results showed that the spectral and structural diversity variables extracted from airborne hyperspectral, and Lidar remote sensing data explained 48% and 50% variance in tree species diversity. However, combining both variables explained 69% variance in tree species diversity. The Lidar variables had a better performance for coniferous and broad-leaved mixed forest tree species estimation than the hyperspectral variables. According to the correlation analysis, the result showed that not all RF-RFE selected remote sensing variables had a significant correlation with field measured Shannon-Winner diversity index. Machine learning such as Random Forest model is helpful to select a few valuable features for forest tree species diversity modeling from massive remote sensing data. This study shows the

关 键 词：季风常绿阔叶林 乔木物种多样性 机载激光雷达 机载高光谱 特征选择 林业遥感 

分 类 号：S771.8[农业科学—森林工程] S718.5[农业科学—林学]