浙江省森林信息提取及其变化的空间分布

如何利用遥感技术提取森林信息是遥感应用的重要领域之一。以不同时相的Landsat TM/ETM+为数据源,采用面向对象和基于多级决策树的分类方法得到浙江省2000年、2005年以及2010年的森林植被覆被图。经实地采样点验证,2010年分类精度达到92.76%,精度满足要求。介绍了浙江森林信息的快速提取方法,即统计不同森林类型的Landsat TM影像原始波段和LBV变换值以及各种植被指数在各时相上的差异,经过C5决策树训练,选取合适的规则和阈值实现森林信息的提取。结果表明,面向对象分割与决策树算法结合可以作为森林信息提取的有效方法。最后,通过对3期森林专题图进行空间叠加分析,得到了森林资源动态变化的空间分布,并以此为基础对林地变化的类型及原因进行分析,结果显示浙江省森林资源变化主要分布在浙西北山区、浙中南山区以及沿海地带,这一结果可以为有关部门的决策提供依据。

The extraction of forest information and the spatial distribution of its change in Zhejiang Province

The accurate forest inforrnation extraction through remote sensing technology is an important content of remote sensing applications. The land cover maps of Zhejiang province in 2000, 2005 and 2010 are generated based on object-oriented segmentation and multi-level decision tree classification technologies. Landsat TM/ETM+ images of variance times are used in this process. The forest in this area mainly contains evergreen coniferous forest, evergreen broad-leaf forest, deciduous coniferous forest, deciduous broad-leaf forest, mixed broadleaf-conifer forest and shrubbery. The forest information extraction for the studied area is carried out through a multiple level scheme, which is the main innovation of this work. The multi-scale segmentation technology is used to construct a 3-level segmentation system to get different scale objects in different classification stages. The multi-level decision tree is emploied as the classification tool. The first level objects will be classified as vegetation or non-vegetation. The second level objects within vegetation will be classified as evergreen forest or deciduoud forest. The third level objects within the evergreen forest and the deciduoud forest will be respectively classified as the relavent sub-type forests. Particularly, some features are computed and used as the input of the decision tree for the sub-type forest classification, including LBV(Level Balance Variance) transform from the 7 TM bands, NDVI(Normalized Difference Vegetation Index) from infrared and red bands of TM, and Tessal transform. Through the decision tree training, we find that in different stage there is a particular feature which plays the key role. For example, NDVI is a typical index to distinguish the vegetation and non-vegetation. NDVI of winter image is also a key index to differentiate evergreen forest and deciduous forest.V derived from LBV transform of the summer TM data is proved to be the best index for classifying the evergreen broad-leaf forest and evergreen coniferous forest. It is also used for deciduous coniferous forest and deciduous broad-leaved forest. In a nother hand, the field data is used to evaluate the mapping result. 1216 samples validate that the classification result of 2010 is highly precise with the overall accuracy of 92.76% and the Kappa of 0.893. The evergreen forest has higher classification accuracy than the deciduous forest. The evergreen coniferous forest and evergreen broad-leaf forest are usually confused with the mixed broadleaf-conifer which leads to the main errors.The results indicate that the decision tree classification combined with object-oriented segmentation provides an effective method for the extraction of forest information. Finally, the dynamic spatial distribution of forest resources is implemented through the spatial overlay analysis for the woodland thematic maps of the three periods. The results show that the forest increasing mostly accur in the central and the southern mountainous areas of zhejiang province, and the deforestation mainly accur in the coastal zones.The categories and reasons for the forest cover change are analyzed as well which can be helpful for the decision-making for the relevant departments. It should be noticed that there are dramatic land cover changes of forest in Zhejiang province. A large number of cultivated lands have been changed into evergreen coniferous forest or evergreen broad-leaf forest which is believed to be a response of the relevant policy.