洞庭湖洲滩速生杨树林变化信息提取方法

洞庭湖是我国第二大淡水湖,其湿地资源具有重要的生态功能和经济价值。近20年来,洞庭湖洲滩速生杨树林发展迅速,其中西洞庭湖杨树林的扩张最为明显,极大改变了湖区湿地植被分布格局,隐含极大的生态风险。以Landsat ETM+和HJ-1A/1B CCD影像为数据源,提出了洞庭湖速生杨树林变化信息提取的两种方法,并对这两种方法进行了比较研究。一种是分类的方法,即采用面向对象分层信息提取的方法先提取出树林滩地信息,再将距离大堤一定范围内的树林滩地归为防护林,速生杨树林变化的面积即为两个时相提取结果的差值。另一种是变化检测的方法,它是基于像元进行变化检测,先确定出总的变化区域,再从中筛选速生杨树林的变化信息。结果表明:(1)两种提取方法都是可行的,不同方法提取的速生林变化信息存在一定差异,但空间分布大体一致;(2)基于分类的方法总体精度和Kappa系数均略高于基于变化检测的方法:其中基于分类的方法总体精度达84.00%,Kappa系数为0.67,基于变化检测的方法总体精度达83.00%,Kappa系数为0.65;(3)基于分类的方法图斑较大、图斑数较少,基于变化检测的方法图斑较小且较破碎、图斑数多;(4)基于分类的方法漏分较少、错分较多,基于变化检测的方法漏分较多、错分较少。为洞庭湖洲滩杨树林的动态监测提供了研究方法,也为杨树林扩张原因及其生态效应分析提供研究基础。

Research on methods for extracting change information of the fast-growing poplar in Dongting Lake

Dongting Lake is the second largest freshwater lake of China, and has an important ecological function and economic value for its abundant wetland resources. The fast-growing poplar trees have grown fast on the beach land in Dongting Lake during the past 20 years, especially in West Dongting Lake, where a dramatic wetland vegetation distribution change is observed with a high ecological risk. In this study, two fast growing poplar tree change extraction methods, which are the classification method and change detection method, are proposed using remote sensed data (Landsat ETM+ images and HJ-1A/1B CCD images, both with a resolution of 30 m) in two periods in West Dongting Lake. And comparisons of the results obtained by these two methods were analyzed. In the classification method, an object-oriented hierarchical information extraction approach was applied to extract the woods beach land information. Then the regions within a certain distance from the dike were classified as windbreaks fields. The change area of fast-growing poplar trees was extracted to be the difference between the results of the two periods. In the change detection method, the change area was firstly calculated based on the changed pixels. Then the change poplar trees were detected as the expansion information for the poplar trees according to the sample information and priori knowledge. The two methods were examined using experimental data, and the result shows that both methods are feasible to extract poplar tree change information. The spatial distributions of changed poplar trees are similar in the two results, though some differences in the location are observed. The overall accuracy and Kappa coefficient of the classification method are slightly higher than that of the change detection method. The overall accuracy is 84.00% and the Kappa coefficient is 0.67 for the classification method. The overall accuracy is 83.00% and the Kappa coefficient is 0.65 for the change detection method. More large polygon patches are observed in the classification result with low degree of fragmentation. By contrast, more small and isolated patches are observed in the change detection result with high degree of fragmentation. The classification method tends to obtain lower omission errors but higher misclassification errors from the accuracy statistic. The change detection method gains lower misclassification errors but higher omission error. This study on extracting change areas of fast-growing poplar trees in Dongting Lake provides a way to dynamically monitor changing areas of fast-growing poplar trees, and offers basic research for other further studies such as the research on the expansion of fast-growing poplar trees and ecological effects that poplar expansion brought about.