森林动态变化分析对揭示生态系统环境变化及植被恢复和布局重建等具有重要意义,时间序列的遥感数据为森林监测提供了基础数据.本文根据森林植被的统计学特性,在暗目标法的基础上,利用归一化植被指数NDVI实现森林样本自动选择;并融合NDVI构建了新的综合森林特征指数(Integrated Forest Z-Score,IFZ);以时间序列的IFZ分析森林动态信息,实现森林变化动态监测.以三峡大坝及周边区域森林为研究区,利用2001年至2012年每年生长季节(5月—10月)的Landsat TM影像检验本文算法.基于2002年、2006年和2010年三期7月—9月的QuickBird影像的精度分析结果发现:研究区森林变化检测的总体精度可达96.53%,Kappa系数为0.9512.在添加NDVI指数后构建的IFZ提高了总体监测精度.其中,毁林类别的检测精度提高显著,漏检率和误检率分别为2.74% 和3.64%;干扰后重建的森林类别的检测精度有一定提高,其漏检率和误检率分别为10.79%和10.51%.研究结果表明,改进暗目标法能提高森林样本的选样效率,添加NDVI的IFZ能提高森林动态变化的识别度.此外,本算法不仅能定性识别森林变化,而且能定量提供森林干扰发生时间和干扰强度.

Research on forest dynamic changes has a significant meaning for revealing changes in ecosystem, overall arrangement, and vegetation recovery. Time series remote sensing image provides abundant data for forest monitoring. This paper presents a novel method for forest cover change detection with time series images. (1) On the basis of the statistical properties of a forest area, a modified dark object identification method, which added NDVI to filter other dark objects, was used to acquire forest samples. (2) A new Integrated Forest Z-score (IFZ) that added NDVI was constructed on the basis of the forest samples to indicate forest characteristic. (3) A time series IFZ value was used to identify forest cover changes. The research area was in Three Gorges Dam. Peripheral area and Thematic Mapper images in every growing season (May-October) from 2001 to 2012 in this area were acquired. The accuracy of the change detection results was evaluated by t-test using QuickBird images in the growing season (July-September) in 2002, 2006, and 2012. The overall precision was 96.53%, and the Kappa coefficient was 0.9512. The commission and omission errors of this class were 2.74% and 3.64% respectively. The accuracy of the disturbance-restoration class was also improved but not as significant as that of the deforestation class; its commission and omission errors were 10.79% and 10.51% respectively. The modified dark object method could improve the efficiency of sampling, and the new IFZ that added NDVI could identify forest effectively. In addition, the novel method not only identified forest quality changes, but also detected the time and degree of disturbance quantitatively.