基于定量反演与云模型的山区森林主要建群种多光谱遥感识别研究

Tree species recognition with remote sensing data is very important for forest efficient management, but it still has some difficulty with methods. Taking forests in Mountain Tai as a case, firstly, the optimum periods for extracting tree species information would be chosen by contrasting measured canopy reflectance among different constructive species. Secondly, with a overall consideration of effecting factors on canopy reflectance including topography, atomophere, pixel components and conopy construction, the ground surface reflectance would be retrieved from high-resolution multispectral remote sensing images through radiometric correction using DEM and atmospheric radiation transfer model, and the reflectance of tree canopy with different canopy constructions would be further retrieved by using pixel unmixing method and canopy reflection models. The retrieval accuracy of each species would be assessed by the comparison of retrieval reflectance with measured reflectance of samples. Thirdly, the most sensitive spectral indices would be constructed and chosen by mathematical transformation of band reflectance, then they would be used as independent variables to construct the recognition models for each species with the methods of cloud model, as well as the texture feature indices for tree species in confusion region. Finally, recognition methods of about 6-8 species would be proposed and a useful integrated recognition pattern would be constructed. This study will provide a feasible method reference for acquisition of tree canopy reflectance, forest type, and other base data for precison forest management.

树种遥感识别对森林高效管理具有重要意义，但仍是一个难点。本项目以泰山森林为例，通过分析主要建群种不同物候期的冠层实测光谱差异，确定多光谱遥感识别的最佳时相；以高分辨率多光谱影像为数据基础，充分考虑树种遥感识别中的影响因素（地形、大气、冠层结构、像元组分等），将改进的地形辐射校正算法、混合像元分解方法和植被冠层反射模型用于冠层反射率定量反演，提出针对不同冠层结构的反演模型；基于反演反射率与重采样的实测反射率构建并筛选各树种的敏感光谱指数，利用云模型方法建立基于光谱特征的树种识别模型；对于易混淆区，引入图像纹理特征参数，建立基于纹理特征的树种识别云模型。提出6-8个建群树种的遥感识别方法，并整合构建主要树种综合分析识别模式。该研究可为森林分类信息的及时获取提供可行方法，也为"精准林业"发展提供科学依据。

树种的多光谱遥感识别已成为遥感应用的重要领域之一，其方法仍在探索中，尤其对于山区来说识别难度更大。以泰山森林为例，从分析主要建群种不同物候期的冠层实测高光谱差异入手，确定多光谱遥感识别的最佳时相为10月份中下旬和11月上中旬；以定量遥感方法为手段，通过构建云指数、阴影指数采用方位搜索的方法区分云影与山地阴影，采用改进的地形辐射校正算法获取近似的冠层反演反射率，有效恢复了地形阴影处的植被信息，缓解了过校正现象；基于反演反射率与重采样的实测反射率构建并筛选出各树种的敏感光谱指数；利用云模型方法建立基于光谱特征的6个树种识别模型，平均识别精度为90.89%，对于易混淆树种，建立基于纹理特征参数的树种识别云模型，但平均精度低于光谱特征云模型；整合构建了6个建群树种的综合分析识别模式，获得研究区的主要树种分布图。该研究为我国东部暖温带森林主要树种的快速识别提供了可行方法，也为“精准林业”发展提供了科学参考。