基于C5.0决策树和时序HJ-1A/B CCD数据的神农架林区植被分类

我国神农架林区海拔高、气候复杂,森林类型多样,结构破碎,森林遥感分类难度较大。将2013年时间序列HJ-1A/B CCD遥感影像作为数据源,计算出植被指数(NDVI、DVI、RVI)和主成分第一分量(PC1),使用DEM数据生成地形因子(高程、坡度、坡向),构建植被分类时序因子集。运用C5.0决策树分类法将神农架林区植被细分为七类:针叶林;针阔混交林;落叶阔叶林;常绿和落叶阔叶混交林;常绿阔叶林;灌丛和草甸。结果表明:该方法的总体精度为72.7%,Kappa系数为0.67;在6~8月,针叶林、草甸和灌丛的植被指数明显低于常绿阔叶林、常绿和落叶阔叶混交林、落叶阔叶林和针阔混交林,对分类的贡献较大,称为植被分类的"窗口期"。PC1、NDVI和高程因子对神农架林地的区分度较高,而坡度、坡向和RVI因子对分类帮助不大。作为一种智能分类方法,C5.0决策树分类方法应用于30m分辨率的时间序列HJ-1A/B CCD数据,能够将地貌复杂的神农架林区植被分为七类,提高了类别精度,具有更高的应用价值。

STUDY ON VEGETATION CLASSIFICATION IN SHENNONGJIA FOREST DISTRICT BASED ON C5.0 DECISION TREE AND HJ-1 A/B DATA

Forest classification plays an important role in understanding the structure and function of forest ecosystem. Shennongjia Forest district belongs to the northern subtropical monsoon climate zone with vegetations distributed in different vertical belts. Because of its high altitude, complicated climate, rich forest types and fragmented landscape, it is difficult to classify the vegetation types in forest region of Shennongjia district. This paper constructed a time series analyses for vegetation classification. The time series of factors collection consisted of the vegetation indices, the first principal component (PC1) and topographic factors, which were calculated from 24 HJ-1A/B CCD images of 2013 and DEM. The vegetation indices collection included NDVI (Normalized Difference Vegetation Index), DVI (Difference Vegetation Index) and RVI (Ratio Vegetation Index). The topographic factors collection included altitude, slope and aspect. By means of the C5.0 Intelligent decision tree algorithm, seven categories were classified in the Shennongjia forest district. They were:cold-temperate coniferous forest, temperate coniferous and broad-leaved mixed forest, temperate deciduous broad-leaved forest, broad-leaved evergreen and deciduous mixed forest, north-subtropical evergreen broad-leaved forest, shrub and meadow. The results showed that:(1) Using the intelligent decision tree algorithm, the overall accuracy of the classification reached 72.7%, with a Kappa coefficient of 0.67; (2) the Vegetation Index values of north-subtropical evergreen broad-leaved forest, the broad-leafed evergreen and deciduous mixed forest, temperate deciduous broad-leafed forest and the temperate coniferous and broad-leafed mixed forest were higher than the cold temperate coniferous forest, scrub and meadow in June to August; the window period of forest vegetation classification was identified as June to August. Using the time series of factors collection can show up "the window" and provide a theoretical basis for the vegetation classification; (3) the time series of PC1, NDVI and altitude as well had a significant contribution to the forest classification; the time series of DVI can help to distinguish between north-subtropical evergreen broad-leaved forest, broad-leafed evergreen and deciduous mixed forest, temperate coniferous and broad-leaved mixed forest, temperate deciduous broad-leaved forest; while slope, aspect and the time series of RVI had much less help in the automatic generation of classification rules, thus, they were of little value to vegetation classification. Overall, the time series of data derived from 30m-resolution HJ-1A/B CCD imagery can accurately categorize the Shennongjia forest vegetation into seven classes based on the C5.0 decision tree method. As an intelligent decision tree classification algorithm, the method has the advantages of identifying more categories and is of high application value. There is still space to improve the classification accuracy, i.e., a more considerate classification system of Shennongjia Forest, the increasing of field samples at mixed vegetation belt, and the use of higher spatial and temporal resolution images.