基于地物光谱季节曲线特征的毛竹林分布提取

Extraction of Moso Bamboo Forest Distribution based on Characteristics of Vegetation Spectral Seasonal Curves

毛竹是我国南方广泛分布的重要竹种,具有良好的生态效益和经济价值。毛竹林与其他森林区分难度大,现有提取方法多直接采用已有的晴空观测,未充分考虑分类时相的影响,限制了提取精度。以浙江省庆元县为例,从地物光谱的季节曲线特征入手,利用MODIS高时间分辨率观测充分挖掘各植被类型光谱季节曲线特征和差异,结合多时相Landsat OLI影像进行分类实验,优选毛竹林与其他植被区分度最大的季相,并采用随机森林方法实现了毛竹林分布的有效提取。结果表明:①初、中秋是区分研究区毛竹林与其他植被的最优时相,夏季次之,春季与冬季较差;②当初、中秋无晴空影像时,结合夏冬季影像的毛竹林提取精度最佳,用户和制图精度分别达到85.57%和78.06%;③10月影像提取毛竹林分布精度最高,用户和制图精度分别达到89.00%和86.91%,与当地森林资源调查数据相比精度优于89.23%。实验表明:在类似亚热带地区毛竹林提取中,应优先选择秋季初、中期影像;若此时期无晴空观测,应优先采用夏季与冬季影像共同分类。

As an important bamboo species,moso bamboo forests are widely distributed in southern China and has great ecological and economic benefits.However,it is difficult to distinguish moso bamboo forests from oth⁃er forests.Most of existing extraction methods directly use available clear sky observation,which do not fully consider the influence of classification time phase,limiting the extraction accuracy.Taking Qingyuan county,Zhejiang Province as an example,a method of moso bamboo forest extraction was established in this paper.The characteristics and differences of seasonal spectral curves were evaluated for typical local vegetation types using MODIS high resolution images,and 16 classification experiments were carried out on single and multi-temporal Landsat OLI images.Based on these analysis and experiments,the best seasonal phase to distinguish moso bamboo forest from other vegetation types was selected,and the distribution of moso bamboo forest was extract⁃ed effectively by using random forest classifier.The results showed that:(1)Early or middle autumn is the best period to distinguish moso bamboo forest from other vegetation in the study area,followed by summer and worst in winter and spring.(2)When there is no clear-sky observation in early and middle autumn,the extrac⁃tion accuracy of moso bamboo forest is the best for combination of summer and winter images,with user and producer accuracy of 85.57%and 78.06%,respectively.(3)The extraction accuracy is the highest based on Landsat image in October,with user accuracy and producer accuracy up to 89.00%and 86.91%,and the extrac⁃tion accuracy is better than 89.23%when compared with the local forestry resources census data.Experiments show that in extraction of moso bamboo forest in similar subtropical areas,the early or middle autumn image should be selected first;if there is no clear-sky observation in this period,the combination of summer and win⁃ter images should be chosen priority.