本文为美国亚利桑那大学(作者:KEITH STUART KRAUSE)的硕士论文,共292页。
物种的变化、减少或灭绝是当前地球面临的一个重大问题。目前正在努力测量、监测和保护具有高度物种多样性的栖息地。遥感技术通过绘制生态系统图和利用土地覆盖图或其他衍生数据作为物种数量和分布的代表,显示了监测物种多样性的重要价值。国家生态观测网(NEON)机载观测平台(AOP)由成像光谱仪、全波形激光雷达和高分辨率彩色摄像机等遥感仪器组成。
2014年5月,AOP通过OrdwaySwisher生物站(OSBS)收集数据。大部分OSBS场地被沙山生态系统覆盖,该生态系统包含非常高的植被物种多样性,是一些受威胁动物物种的原生栖息地。本研究探讨如何分析AOP资料,以绘制OSBS地点的生态系统图。本研究试图利用高空间分辨率的数据,研究在一个地标尺度内数据的变异性,并整合来自不同传感器的数据。从数据中提取数学特征,并将其纳入决策树分类算法(rpart)中,以便为站点创建生态系统地图。高光谱和激光雷达特征可作为每个生态系统植被类型的化学、功能和结构差异的代表。K-folds交叉验证的训练准确率为91%,验证准确率为78%,使用独立地面验证的准确率为66%。本文的研究结果对利用高光谱和激光雷达遥感综合数据进行生态系统制图作出了重要贡献,将一个地块尺度内数据的空间变异性与构成一个给定生态系统的植被类型集合联系起来。
The change, reduction, or extinction ofspecies is a major issue currently facing the Earth. Efforts are underway tomeasure, monitor, and protect habitats that contain high species diversity.Remote sensing technology shows extreme value for monitoring species diversityby mapping ecosystems and using those land cover maps or other derived data asproxies to species number and distribution. The National Ecological ObservatoryNetwork (NEON) Airborne Observation Platform (AOP) consists of remote sensinginstruments such as an imaging spectrometer, a full-waveform lidar, and ahigh-resolution color camera. AOP collected data over the Ordway-SwisherBiological Station (OSBS) in May 2014. A majority of the OSBS site is coveredby the Sandhill ecosystem, which contains a very high diversity of vegetationspecies and is a native habitat for several threatened fauna species. Theresearch presented here investigates ways to analyze the AOP data to mapecosystems at the OSBS site. The research attempts to leverage the high spatialresolution data and study the variability of the data within a ground plotscale along with integrating data from the different sensors. Mathematicalfeatures are derived from the data and brought into a decision treeclassification algorithm (rpart), in order to create an ecosystem map for thesite. The hyperspectral and lidar features serve as proxies for chemical,functional, and structural differences in the vegetation types for each of theecosystems. K-folds cross validation shows a training accuracy of 91%, a validationaccuracy of 78%, and a 66% accuracy using independent ground validation. Theresults presented here represent an important contribution to utilizingintegrated hyperspectral and lidar remote sensing data for ecosystem mapping,by relating the spatial variability of the data within a ground plot scale to acollection of vegetation types that make up a given ecosystem.
- 科学背景
- 研究目标
- ORDWAY-SWISHER生物站的植被群落
- 遥感与地面数据
- 利用高光谱遥感测量的光谱变异性进行植被分类
- 利用离散回波激光雷达测量的结构变异性进行植被分类
- 改进植被制图的综合遥感观测
- 结论
更多精彩文章请关注公众号: