Classification-based mapping of trees in commercial orchards and natural forests

Hyperspectral remote sensing (RS) and images of various spatial resolution open new vistas for classification and mapping trees. These approaches would improve plant classification in a complex population of forest trees of diverse species, genera, and families, as well as monitoring commercial orchards. In this work, we used new RS indices for cellulose, lignin, wax, chlorophyll, carotenoid, and anthocyanin for plant species classification in natural forests and commercial orchards. For proof of concept, the indices were applied to the classification and mapping of various horticultural crop orchards, where error due to the spatial mixing of different trees is minimal. The classification accuracy of the maps varied between 65 and 82%. This wide range was a result of the following factors: The RS index used, the season, and the spatial resolution of the hyperspectral images. The classification quality was highest when the full set of RS indices was used. The effect of the wax index on accuracy was significant. Furthermore, seasonality played an important role in the classification; the target species were better resolved in spring than in the summer. The higher spatial resolution of the images does not necessarily yield better classification and mapping results; it appeared to be case-specific and greatly depended on the species/crop and the unique environment.高光谱遥感（Hyperspectral Remote Sensing, RS）与不同空间分辨率的影像为林木分类与制图开辟了全新路径。此类手段可优化复杂林分中跨物种、属乃至科的林木植物分类任务，同时亦可用于商业果园的监测工作。本研究针对天然林与商业果园的植物物种分类需求，构建并应用了针对纤维素、木质素、蜡质、叶绿素、类胡萝卜素及花青素的新型遥感指数（Remote Sensing Indices）。为开展概念验证，本研究将上述指数应用于多种园艺作物果园的分类与制图任务，该场景下不同林木空间混交引发的误差极低。所得分类制图的精度区间为65%至82%，跨度较大，主要受以下三方面因素影响：所选用的遥感指数、成像季节以及高光谱影像的空间分辨率。当使用全部遥感指数集合时，分类质量可达最优；其中蜡质指数对分类精度的影响尤为显著。此外，季节因素对分类结果同样具有重要作用：相较于夏季，春季可更精准地分辨目标物种。需注意的是，更高空间分辨率的影像未必能带来更优的分类与制图效果，其表现需结合具体场景而定，且高度依赖目标物种/作物及其独特的生境条件。