面向自然场景土地覆被分类的遥感物候模式分区

Remotely-sensed phenology pattern regionalization for land cover classification of natural scenes:A case study in China

选取适当地理边界对研究区或影像进行分区,降低区域内土地覆被复杂程度及其在影像中的特征变异性,能够有效提升土地覆被遥感分类的精度和效率。现有土地覆被制图中开展分区分类策略所借助的生态分区等区划数据缺乏目标针对性,其对遥感分类的适用性范围及精度提升效果仍存在限制。植被物候变化是造成自然场景土地覆被类内光谱异质的主要原因,本文利用遥感观测的反映地表植被绿度状态的植被指数和反映地表植被生长发育节律的关键物候期构建分区指标体系,以反映微地貌形态、坡面属性及地表物质组成的地貌小区为分区单元,采用数据驱动的空间约束层次聚类算法,提出了面向自然场景土地覆被分类的遥感物候模式分区。基于统计检验和多源土地覆被产品的分区评价结果表明,本文的遥感物候模式分区可有效降低区域内土地覆被复杂程度和植被物候变化引起的类内特征异质,在土地覆被代表性样本库构建以及分区分类策略实施等方面具有较高应用潜力。

Selection of appropriate geographic boundaries for zoning the study area or images can effectively improve the accuracy and efficiency of land cover classification by reducing the complexity of land cover within the regions and the variability of its features in the images.At present,the regionalization data used in land cover mapping based on stratified classification strategies,such as ecological regionalization,lack targeted objectives,which limits its applicability and effectiveness in remote sensing classification.Vegetation phenology is the main cause of spectral heterogeneity within the land cover of natural scenes.To address this issue,this study proposed a remotely-sensed phenology pattern regionalization scheme for land cover classification of natural scenes.The regionalization scheme is implemented by constructing a zoning index system using vegetation indices,which reflect the greenness status of vegetation,and key phenological periods,which reflect the growth and development rhythm of vegetation.Small geomorphic regions are used as the zoning units,and a data-driven spatially constrained hierarchical clustering algorithm is employed in the regionalization.The evaluation results based on statistical tests and multi-source land cover products indicate that the remotely-sensed phenology pattern regionalization in this study effectively reduces the complexity of land cover within the region and the intra-class feature heterogeneity caused by vegetation phenology,and shows high potential in constructing representative land cover sample libraries and implementing stratified classification strategies.