基于盐城国家级珍禽自然保护区核心区2014年3个月份的Landsat 8遥感影像及其矢量数据,采用基于CART算法的决策树分类方法提取研究区芦苇、碱蓬、米草、鱼塘、浅滩、海域等湿地信息,并分析2014年植被变化情况。其中采用植被指数NDVI,RVI,DVI时间序列光谱分析曲线获得湿地植被类型窗口期,通过各植被指数、第一主成分分量、缨帽变换、原始波段(红外、近红外)、非监督分类影像等因子构建时序因子集。结果表明:1)3—12月份为植被分类窗口期,芦苇、碱蓬、米草区分度最大;2)CART算法的决策树分类方法对盐城湿地植被区分度较好,3个月份影像分类总体精度分别为99.88%,99.18%和97.61%,Kappa系数分别为0.99,0.99和0.97;3)2014年间,芦苇的面积从61.69km ²增长到63.08km ²,米草从38.01km ²增加到44.78km ²,碱蓬从26.37km ²锐减到19.63km ²。

Based on three months of Landsat8 remote sensing images and vector data of the core area of Yancheng national rare birds nature reserve and vector data in 2014,using CART-based decision tree classifier to extract phragmites,suaeda,spartina,fish ponds,mud flat,waters and wetlands information and then changes of vegetation are analyzed.Using vegetation index NDVI,RVI,DVI series spectrum curves to get a wetland vegetation type in the window period,through the vegetation index,the first principal component,tasseled Cap transformation,the original bands(red,near-infrared),unsupervised classification image,sequence subsets are built.Results showed that:1) The window period for vegetation classification is from March to December,phragmites,suaeda,Spartina discrimination of maximal order during this decision tree classification of remote sensing images for the data source can improve the accuracy of classification;2) CART-based decision tree classifier to Yancheng wetland vegetation and distinguish have good overall classification accuracy for 99.88%,99.18% and 97.61% and good Kappa coefficient for 0.99,0.99 and 0.97;3) In 2014,the area of phragmites grew from 61.69km ² to 63.08km ²and spartina from 38.01km ²to 44.78km ² while suaeda had fallen from 26.37km ²to 19.63km ².