关键词：优势树种识别；GEE；时序轨迹特征；归一化退化指数；CCDC算法；时间谐波分析中图分类号：S771.8 文献标识码：A DOI： 10.7525/j.issn.1006-8023.2025.03.00

Abstract：Theidentificationof dominant tree species isanimportant partofforestry resource surveys.Improving the accuracy of dominant tree species identification has significant practical implications for conducting forest resource surveys andrelated research.Using the Google Earth Engine（GEE）cloud platform，we obtained Sentinel-2 time series images forthe Huodong mining areafrom January to December 2O23.Theannual growth trajectory featuresof dominant tree species wereconstructed basedonthe CCDC algorithmand the NDFIindex.Adominant tree species hierarchical identification method combining \"trajectory features + spectral features + texture features\" of long-time series remote sensing images was proposed. A control group of \"spectral features + texture features\"was setup，and hierarchical classification andrandom forest clasificationalgorithms were used to identify7dominant treespecies （Pinus tabuliformis，Quercus wutaishansea，Betulaplayphylla，Lrixprincipis-rupprechtii，Platycladusorientalis，Populus davidiana，andpoplars spp.）inthe Huodong mining area.Theresultsshowed that：1）The NDFIindex can efectively distinguish between deciduous forests and evergreen forests；2）The dominant tree species identification based on \"trajectory features + spectral features + texture features\" performed well，with an overallclassification accuracy of 79.6%and a Kappa coeffcient of 0.742 in the study area，which was 7. 3 % higher than the control group.

Keywords：Dominant tree species identification；GEE；temporal trajectory features；normalized disturbance index; CCDC algorithm；time series harmonic analysis

0 引言

树种信息在森林资源动态监测、生物多样性评估以及森林生物量和碳储量估算中发挥着至关重要的作用[1]。但如何克服不同树种间的\"同物异谱\"\"异物同谱'现象，快速、准确地获取森林优势树种信息和空间分布是亟待解决的热点问题[2-3]。传统的树种调查以人工实地调查为主，辅以简单的测量和记录工具，这种方法不仅耗时耗力且无法及时有效地更新森林资源信息。而遥感技术能够在较短的时间内获取大范围、多种类型的地表信息，为准确而高效的植被分类和优势树种识别提供技术支持[4-7]。已有研究表明综合运用光谱特征、纹理特征和时间特征能有效提高优势树种识别精度[8-10]。

光谱特征主要指影像的波段光谱特征及植被指数。不同植被因结构特征、色素含量和含水量等不同造成光谱曲线差异[11-12]，梁爽等[13]研究了不同植被在不同季节的平均地表反射率，结果表明不同波段特征对植被提取的贡献度不同。植被指数反映植被的生长状况和生物量，目前在植被分类、优势树种识别中应用最为广泛的植被指数是：归一化植被指数（normalizeddifference vegetation index，NDVI）[14-16]，但NDVI在低植被密度区域，例如落叶时期可能受到阴影和土壤背景的影响[17]，而归一化退化指数（normalizeddifferencefractionindex，NDFI）[i8]充分顾及森林冠层阴影分量和王壤分量，恰好有效弥补NDVI对非植被区域敏感的问题。由于不同植被间的影像光谱特征存在一定程度的相似性，仅利用光谱特征往往无法很好地实现目标识别，因此在信息提取的过程中通过引人纹理特征增强影像识别的可靠性。……