时序Sentinel-1和Sentinel-2数据支持下的鄱阳湖湿地草本植物群落制图分类  

Classification scheme for mapping wetland herbaceous plant communities using time series Sentinel-1 and Sentinel-2 data

作　　者：张琍[1,2,5] 罗文庭 张皓寰 殷秀琬 李斌 ZHANG Li;LUO Wenting;ZHANG Haohuan;YIN Xiuwan;LI Bin(Key Laboratory of Poyang Lake Wetland and Watershed Research,Ministry of Education,Jiangxi Normal University,Nanchang330022,China;School of Geography and Environment,Jiangxi Normal University,Nanchang 330022,China;College of Urban and Environmental Sciences,Central China Normal University,Wuhan 430079,China;State Key Laboratory of Remote Sensing Science,Beijing Normal University,Beijing 100875,China;Nanchang Base of International Centre on Space Technologies for Natural and Cultural Heritage under the Auspices of UNESCO,Nanchang 330022,China;College of Oceanography and Space Informatics,China University of Petroleum(East China),Qingdao 266580,China)

机构地区：[1]江西师范大学鄱阳湖湿地与流域研究教育部重点实验室,南昌330022 [2]江西师范大学地理与环境学院,南昌330022 [3]华中师范大学城市与环境科学学院,武汉430079 [4]北京师范大学地理科学学部,北京100875 [5]联合国教科文组织国际自然与文化遗产空间技术中心南昌分中心,南昌330022 [6]中国石油大学(华东)海洋与空间信息学院,青岛266580

出　　处：《遥感学报》2023年第6期1362-1375,共14页NATIONAL REMOTE SENSING BULLETIN

基　　金：国家自然科学基金(编号:41701514,41961004,41761076);江西省自然科学基金(编号:20224BAB202037,20224BAB203034);鄱阳湖湿地与流域研究教育部重点实验室开放基金(编号:PK2020003,PK2021007);江西师范大学地理与环境学院经天纬地开放基金。

摘　　要：植被是湿地的核心,易受人类活动和气候变化的影响,湿地植物群落分类与制图可以为湿地生态监测与评估提供科学数据支撑。本研究以鄱阳湖国家级湿地自然保护区为研究区,基于2019年月度Sentinel-1和Sentinel-2时序数据,通过提取影像的水体指数和植被指数、红边指数、纹理特征、光谱特征、雷达极化数据5类,共计240个特征指标,使用随机森林、支持向量机和深度神经网络算法进行分类,探寻一套湿地植被分类最优的特征组合和分类方案。(1)光学数据在湿地分类与制图提取中明显优于雷达数据,雷达数据可以在光学数据不足时,作为光学数据的补充。(2)对时序Sentinel-2的各特征变量进行重要性筛选,有助于提高分类精度,优选时间段主要分布在1月、5月、8月、9月、10月和12月份;(2)当对5组特征变量单独分类时,分类精度排序为红边指数组>水体—植被指数组>光谱特征组>雷达极化数据组>纹理特征组;(3)对比组合变量和单独特征变量,组合变量不一定有助于提高分类效果,分类精度排序为:红边指数分类组>水体—植被指数分类组>组合分类组,其中,红边指数组随机森林分类总体精度达0.81,Kappa系数为0.76;(4)对比3种分类方法,分类精度排序为:深度神经网络>随机森林>支持向量机,其中,深度学习方法并没有太大幅度的提高分类精度,相对随机森林算法仅仅提高了2%。故深度神经网络和随机森林算法都可以作为优选算法。本研究给出的分类方案是,使用Sentinel-2和Sentinel-1多时序数据对湿地植被进行精细化分类,时段选择建议1月、5月、8月、9月、10月和12月份的卫星数据更优,特征变量可选红边指数组或者水体—植被指数组产品,分类方法可根据需求选择深度神经网络或随机森林对湿地植物群落进行分类,可得出较优的分类结果。这个分类方案可以有效的提升鄱阳湖湿地植被制�Plant communities play an important role in wetland elements and are vulnerable to human activities and climate change.Wetland plant community classification and mapping provide scientific important data support for wetland ecological monitoring and evaluation.This study aims to develop a classification scheme suitable for the wetland plant communities in the Poyang Lake wetland.Taking Poyang Lake National Nature Reserve as the research area and on the basis of the monthly Sentinel-1 and Sentinel-2 time-series data in 2019,this study extracts five types of image feature parameters,including water and vegetation index group,red edge index group,texture feature group,spectral feature group,and polarization radar backscatter group,with a total of 240 feature indexes,and uses Random Forest(RF),Support Vector Machine(SVM),and Deep Neural Network(DNN)algorithms for classification to explore a set of optimal feature combinations and a suitable classification scheme for wetland vegetation mapping in Poyang Lake.Results show the following:(1)Compared with radar data,the extraction accuracy of optical data is remarkably better than that of radar data in wetland plant community classification and mapping.Radar data can be used as a supplement to optical data when optical data are insufficient.(2)Screening the importance of each image feature of Sentinel-2 helps improve the classification accuracy.The preferred time periods are mainly distributed in January,May,August,September,October,and December.(3)Five groups of unitary image features are selected to classify separately,and the classification accuracy is as follows:red edge index group>water and vegetation index group>spectral feature group>radar polarization data group>texture feature group.(4)Comparing the combined image feature groups with the unitary image feature groups reveals that the combined image feature group is not necessarily helpful to improve the classification accuracy.The classification accuracy is as follows:red edge index group>water vegetation index g

关 键 词：遥感 鄱阳湖 湿地植被制图 特征变量 随机森林 深度神经网络 多时相光学与雷达数据 

分 类 号：P2[天文地球—测绘科学与技术]