地理科学进展 ›› 2021, Vol. 40 ›› Issue (4): 703-712.doi: 10.18306/dlkxjz.2021.04.014

• 研究论文 • 上一篇    下一篇

基于时序遥感的庐山自然保护区植被分类及其变化分析

张琍1(), 李斌1,2, 阳文静1, 李琼3   

  1. 1.江西师范大学鄱阳湖湿地与流域研究教育部重点实验室,南昌 330022
    2.中国石油大学(华东)海洋与空间信息学院,山东 青岛 266580
    3.武汉市测绘研究院,武汉 430022
  • 收稿日期:2020-03-04 修回日期:2020-10-16 出版日期:2021-04-28 发布日期:2021-06-28
  • 作者简介:张琍(1981— ),女,湖北武汉人,博士,副教授,主要从事雷达与光学遥感应用、植被遥感分类等研究。E-mail: zhanglinu@126.com
  • 基金资助:
    国家自然科学基金青年项目(41701514);国家自然科学基金青年项目(41761076);江西省自然科学基金项目(20161BAB213074);鄱阳湖湿地与流域研究教育部重点实验室开放基金(PK20170005)

Forest vegetation classification and its variation in Lushan Nature Reserve using Proba-V vegetation products

ZHANG Li1(), LI Bin1,2, YANG Wenjing1, LI Qiong3   

  1. 1. Key Laboratory of Poyang Lake Wetland and Watershed Research, Ministry of Education, Jiangxi Normal University, Nanchang 330022, China
    2. College of Oceanography and space informatics, China University of Petroleum (East China), Qingdao 266580, Shandong, China
    3. Wuhan Geomatics Institute, Wuhan 430022, China
  • Received:2020-03-04 Revised:2020-10-16 Online:2021-04-28 Published:2021-06-28
  • Supported by:
    National Natural Science Foundation of China(41701514);National Natural Science Foundation of China(41761076);Natural Science Foundation of Jiangxi Province(20161BAB213074);Open Fund of Key Laboratory of Poyang Lake Wetland and Watershed Research, Ministry of Education (Jiangxi Normal University)(PK20170005)

摘要:

自然保护区是生物多样性最丰富的区域,针对近年来庐山自然保护区森林生物多样性进行调查,了解保护区内植被的分布及其变化,对于保护庐山境内的林业资源具有重要的意义。论文利用2015—2018年Proba-V卫星100 m分辨率5 d植被指数产品,使用最大值合成法、Hants时间序列谐波分析方法对数据进行平滑降噪处理,并采用随机森林算法进行分类,同时对庐山林地变化进行分析。结果表明:① 2015—2018年庐山林地随机森林分类的总体精度分别为85.00%、84.25%、84.75%、85.25%,Kappa系数分别为0.80、0.79、0.81、0.80。分类结果与LC-CCI、Globcover、Globeland30、MODIS等植被覆盖分类产品的对比分析表明,基于时序NDVI产品的随机森林分类器在空间分辨率和制图精度上都取得了较好的分类效果和较高的分类精度。② 庐山自然保护区内针叶林、阔叶林分布较为集中,混交林与竹林零散分布,其中,针叶林的分布面积最大,占比39.36%;竹林的分布面积最小,占比为14.63%。③ 2015—2018年间,庐山林地类型分布变化不大,阔叶林、针叶林、混交林变化率均低于10%,阔叶林变化最小,竹林由于受到人类活动影响,4 a间变化幅度波动较大。

关键词: Proba-V, 随机森林, 森林植被制图与分类, 时空分布, 庐山

Abstract:

Nature reserves are the areas with the most abundant biodiversity. Conducting a survey of current forest biodiversity in Lushan Nature Reserve and understanding the distribution and changes of vegetation in the protected area is of great significance for the protection of forest resources in Lushan. In this study, we used the Proba-V satellite 100-meter resolution 5-day vegetation index product from 2015 to 2018 using the maximum synthesis and Hants time series harmonic analysis methods to smooth the data and reduce noise, and used a random forest algorithm to classify the forest vegetation. The changes of Lushan vegetation communities were also analyzed. The research results show that: 1) The overall accuracies of the random forest classification of Lushan forestland from 2015 to 2018 were 85%, 84.25%, 84.75%, and 85.25% while Kappa coefficients were 0.80, 0.79, 0.81, and 0.80, respectively. Compared with the Cover-CCI-2010, Globcover-2009, Globeland30, and MODIS-2010 land cover products, the classification results by random forest classifier based on time series NDVI products have achieved good classification effect and high classification accuracy in spatial resolution and mapping accuracy. 2) In Lushan Nature Reserve, coniferous forest and broad-leaved forest are concentrated while mixed forest and bamboo forest are scattered among them. Coniferous forest has the largest distribution area, accounting for 39.36% of the total forest area, and the bamboo forest has the smallest distribution area, accounting for 14.63%. 3) From 2015 to 2018, the spatial distribution of forest vegetation communities in Lushan did not change much. The change rates of broad-leaved forest, coniferous forest, and mixed forest were all less than 10%. Bamboo forest fluctuated during 2015-2018 because of the interference of human activities while the change of broad-leaved forest was the smallest.

Key words: Proba-V, random forest, forest vegetation mapping and classification, spatial and temporal distribution, Lushan