对比多源遥感数据在滨海湿地提取中的差异

Comparison of the Differences of Multi-source Remote Sensing Data in Coastal Wetland Extraction

Landsat 8与Sentinel-2数据是当前遥感分类最典型、常用的光学数据源,但这两种数据源在滨海湿地分类中的差异性仍有待进一步研究。基于Landsat 8、Sentinel-2和Sentinel-1数据,采用随机森林(random forest,RF)算法对南沙区湿地进行分类,并探讨光学数据源、主成分变换融合方法以及各特征变量对南沙区滨海湿地RF分类精度的影响。结果表明:(1)Sentinel-2数据参与的方案分类精度均优于Landsat 8数据参与的方案,其分类精度分别为90.24%(单一光学数据参与分类)、85.89%(光学数据与雷达数据融合参与分类)。(2)主成分变换融合方法对分类精度的提高不大,融合数据使滨海湿地总体分类精度下降(p<0.05),但融合数据在一些滨海湿地类型提取中具有一定优势。融合Sentinel-2与Sentinel-1数据减小了沿海滩涂与浅海水域之间的错分现象;融合Landsat 8与Sentinel-1数据更有利于库塘、红树林湿地信息提取以及湿地与非湿地信息的区分。

Landsat 8 and Sentinel-2 data are the most typical and commonly used optical data sources for remote sensing classification,but the differences between them in coastal wet⁃land classification remain to be further studied.Based on Landsat 8,Sentinel-2,and Sentinel-1 data,we use random forest(RF)algorithm to classify wetlands in Nansha District,and discuss the influences of optical data sources,the fusion method of principal component transformation,and various features variables on the accuracy of classification results of coastal wetlands in Nansha District based on RF algorithm.The results show that,first,the classification accuracy of the schemes involving Sentinel-2 data is better than those involv⁃ing Landsat 8 data,with the classification accuracy of 90.24%(single optical data involved in classification)and 85.89%(optical data and radar data fusion involved in classification).Second,the fusion method of principal component transforma⁃tion does not greatly improve the classification accuracy.Even though the fusion data reduce the overall accuracy of coastal wetlands(p<0.05),they have certain advantages in the extraction of some types of coastal wetlands.The fusion data of Sentinel-2 and Sentinel-1 reduce the misdivision be⁃tween coastal beaches and shallow waters.The fusion data of Landsat 8 and Sentinel-1 are more conducive to the extraction of information from ponds and mangroves,and the distinction between wetland and non-wetland information.