摘要
开展香格里拉市针叶林优势树种识别研究,为该区森林资源管理和研究提供参考。基于Google Earth Engine(GEE)平台和2020年Sentinel-1/2时间序列影像,构建植被的时序特征,并结合雷达特征、光谱特征、纹理特征和地形特征等共计43个特征,通过对特征的不同组合方案,采用分层分类和随机森林分类算法对香格里拉高山松、云南松、云冷杉和落叶松4种优势树种进行精细识别。结果表明,多源时间序列影像结合所有特征在3个层次上分类精度均较高;研究区森林与非森林分类的总体精度为98.73%,Kappa系数为0.97,用户精度和制图精度的调和平均值F_(1)为98.71%;针叶林和阔叶林分类的总体精度为92.80%,Kappa系数为0.85,F_(1)为92.58%;4种优势树种识别的总体精度为89.51%,Kappa系数为0.86,F_(1)为89.36%。不同树种在不同特征上均具有可分离性;多特征结合能在一定程度上提高树种识别的精度;基于GEE平台和Sentinel-1/2多源时间序列数据可以实现10 m空间分辨率下的森林优势树种精细识别。
The research on the identification of dominant tree species in the coniferous forest of Shangri-La was carried out to provide a reference for subsequent forest resource management and research in the area. Based on the Google Earth Engine(GEE) platform and Sentinel-1/2 time series images of 2020, the temporal characteristics of vegetation were constructed, and a total of 43 features were combined with radar features, spectral features, texture features, and terrain features. Through different combination schemes of features, hierarchical classification and random forest classification algorithms were used to finely identify the dominant tree species of four coniferous forests of Shangri-La: Pinus densata, Pinus yunnanensis, Picea asperata and Larix gmelinii. The results showed that the classification accuracy of multi-source time series data combined with all features was the highest at three levels. The overall accuracy of forest and non-forest types in the study area was 98.73%, and the Kappa coefficient was 0.97, harmonic average of user accuracy and mapping accuracy F_(1)was 98.71%. The overall accuracy of coniferous and broad-leaved forests was 92.80%, the Kappa coefficient was 0.85, F2was 92.58%. The overall accuracy of 4 dominant tree species was 89.51%, the Kappa coefficient was 0.86, F_(1)was 89.36%. Different tree species had separability on different features. The combination of multiple features can improve the accuracy of tree species identification to a certain extent. Based on GEE platform and Sentinel-1/2 multi-source time series data can perform fine identification of forest dominant tree species at a spatial resolution of 10 m.
作者
刘灵
张加龙
韩雪莲
许冬凡
王书贤
程滔
LIU Ling;ZHANG Jialong;HAN Xuelian;XU Dongfan;WANG Shuxian;CHENG Tao(Faculty of Forestry,Southwest Forestry University,Kunming 650244,China;National Geomatics Center of China,Beijing 100830,China)
出处
《森林工程》
北大核心
2023年第1期63-72,81,共11页
Forest Engineering
基金
云南省教育厅科学研究基金项目(2022Y583)
2020年云南省高层次人才培养支持计划“青年拔尖人才”专项(81210468)。
