基于Sentinel-1和Sentinel-2影像的洪泽湖国家湿地公园水生植被信息提取

Extraction of aquatic vegetation in Hongze Lake National Wetland Park based on Sentinel-1 and Sentinel-2 images

【目的】探索利用光学遥感和雷达遥感数据进行湖泊湿地水生植被信息提取方法。【方法】以洪泽湖国家湿地公园为研究区,基于Sentinel-1的SAR影像和Sentinel-2的MSI影像,利用面向对象影像分析技术,结合EVSI、NDVI、SR特征指数和对象之间的上下文特征,以及挺水植被高度的差异所对应SAR影像上的后向散射系数,在对象级的基础上建立决策树模型对湿地水生植被进行分类,分析洪泽湖国家湿地公园水生植被以及挺水植被的分布状况。【结果】研究区水生植被类群分类精度为89%,Kappa系数为0.85;挺水植被种群分类精度为85.2%,Kappa系数为0.76。与基于像元分析方法的结果相比,面向对象的影像分析方法具有更高的精度;湿地水生植被以沉水植被和挺水植被为主,其中挺水植被中以荷叶和芦苇为主。【结论】本研究提出的湖泊湿地水生植被信息提取方法具可行性,可为湿地管理与决策提供科学依据。

[Objective]The objective of this study was to explore the extraction of spatio-temporal distribution of aquatic vegetation in lake wetlands using Sentinel-1 and Sentinel-2 data.[Method]Hongze Lake National Wetland Park was chosen as the research area.Based on the combination of Sentinel-2 MSI images and Sentinel-1 SAR images,the object-oriented image analysis was used.The feature set was constructed by using EVSI,NDVI,SR feature index and contextual features between objects,as well as differences in the backscatter coefficients of the SAR images corresponding to differences in the height of the emergent vegetation types.A decision-tree model was established at the object level to classify the wetland,and the spatio-temporal distribution of the aquatic vegetation and the emergent vegetation in the Hongze Lake National Wetland Park was acquired.[Result]The classification accuracy and the Kappa coefficient of aquatic vegetation were observed to be 89%and 0.85,respectively,and that of the emergent vegetation was 85.2%and 0.76,respectively.The results showed that,compared with the results of the pixel-based analysis method,the accuracy of object-based image analysis was higher.The wetland aquatic vegetation was dominated by submerged and emergent vegetation;among the emergent vegetation,lotus leaves and reeds were dominant.[Conclusion]The methods proposed in this study were feasible,and the results could provide a scientific basis for managers and planners of wetlands.