神农架川金丝猴栖息地优势乔木树种遥感识别及其分布特征

Determination of dominant tree species and effects of tree distribution on the habitat of Rhinopithecus roxellana using Remote Sensing imagery in Shennongjia

针对神农架川金丝猴生境基础研究中乔木树种大范围分布数据难以获取问题,尝试利用多源多时相遥感数据结合专家知识分层次实现树种识别。首先采用冬季Landsat8/OLI数据根据物侯特性分层提取常绿、落叶林的地域范围;进而依据夏季Worldview-2高分遥感影像的实地乔木样本的光谱特征分层次完成常绿树种(巴山冷杉、华山松、青$、刺叶栎)和落叶树种(红桦、日本落叶松、米心水青冈、漆树、锐齿槲栎、椅杨)的识别;并通过实地植被样方及专家知识通过高程数据完成分类结果的修正;最后结合GIS对主要优势树种的地形及地域分布特征进行了空间分析。实验精度表明常绿林中巴山冷杉、华山松、刺叶栎、虫害华山松整体精度较高,落叶林中红桦、漆树等识别精度相对较高,部分树种如椅杨、锐齿槲栎识别精度较低;总体上常绿树种的精度要优于落叶树种。从植物地理学、遥感、GIS三者相结合的角度,将多源、多时相遥感数据与物种物候特性、专家知识进行有效整合,提出了一种乔木树种识别的方法(1)提供了复杂山地环境的主要乔木优势种识别途径,且具有通用性;(2)完成了物种物候特性与遥感数据特性的整合利用,有效降低数据成本费用;(3)配合地面样方及专家知识修正结果,避免了过分依赖光谱特征引起的误判。这将为神农架川金丝猴栖息地保护与恢复提供更精确的数据依据。

Because of the difficulty in obtaining large-scale distribution data on tree species in Rhinopithecus roxellana habitat in Shennongjia, we attempted to use multi-source and multi-temporal remote sensing data combined with expert knowledge to identify species at different levels. Firstly, after analyzing the discrimination of sample trees, we used winter Landsat8/OLI image data to extract evergreen and deciduous forest, respectively. Secondly, we used summer Worldview-2 high resolution image data to for the recognition of tree species, which included the evergreen species （Abies fargesii, Pinus armandii, Picea wilsonii, Quercus spinosa） and deciduous tree species （Betula albo- sinensis, Larix kaempferi, Fagus engleriana, Toxicodendron vernicifluum, Quercus aliena, Populus wilsonii）, respectively. Thirdly, combining the vegetation quadrats and expert knowledge on elevation, we corrected the classification results based on the second step. Finally, making use of GIS spatial analysis, we analyzed the terrain and geographical distribution on the dominant species. The experiment revealed that accuracy was higher in evergreen forests, such as Abies fargesii, Pinus armandii, Quercus spinosa, and Pinus armandii affected by pests, whereas relatively higher in deciduous forest, such as Betula albo-sinensis and Toxicodendron vernicifluum. Some species, such as Populus wilsonii and Quercus aliena, showed poor accuracy. In general, evergreen species had higher accuracy than deciduous trees. By combining plant geography, remote sensing, and GIS, we integrated the multi-source, multi-temporal remote sensing data, phenological characteristics of the tree species, and expert knowledge to propose a method for identifying tree species. This method （1） provides an effective way to identify dominant tree species in complex mountainous environments, and it has the versatility for a variety of geographical environments; （2） makes full use of the integration of species phenological features and characteristics of remote sensing data to reduce data costs; （3） uses ground sampling and expert knowledge, ensuring the classification results are correct, which can avoid excessive reliance on spectral characteristics, and reduce the possibility of misclassification. This method will provide more accurate data for the protection and restoration of the habitat of Rhinopithecus roxellana in Shennongjia.