Acta Scientiarum Naturalium Universitatis Pekinensis ›› 2018, Vol. 54 ›› Issue (4): 792-800.DOI: 10.13209/j.0479-8023.2018.010

Previous Articles     Next Articles

Bashang Forest Change Monitoring with Multi-Temporal MODIS Images and Random Forest Algorithm

ZHOU Jianing1, ZHANG Jie1, LI Tianhong1,2,†   

  1. 1. School of Environment and Energy, Peking University Shenzhen Graduate School, Shenzhen 518055
    2. College of Environmental Sciences and Engineering, Peking University, Beijing 100871
  • Received:2017-04-17 Revised:2018-01-09 Online:2018-07-20 Published:2018-07-20
  • Contact: LI Tianhong, E-mail: litianhong(at)


周佳宁1, 张洁1, 李天宏1, 2, †   

  1. 1. 北京大学深圳研究生院环境与能源学院, 深圳 518055
    2. 北京大学环境科学与工程学院, 北京 100871
  • 通讯作者: 李天宏, E-mail: litianhong(at)
  • 基金资助:


In order to reveal the dynamic characteristics of the forest in Bashang area of Hebei Province, MODIS reflectivity and NDVI data with a spatial resolution of 250 m were used for forest classification, and a Thematic Mapper (TM) image in 2005 was resorted to aid training sample selection. With Random Forest Algorithm and time series of MODIS images, the forest in Bashang area was monitored from 2000 to 2015 in every two years. Compared with widely used classifiers such as maximum likelihood classifier and BP artificial neural network algorithm, Random Forest classifier showed the best performance with its overall accuracy and Kappa coefficient being 91.89% and 0.88 respectively. Binary coding was applied to the eight phases of forest distribution images, which can easily and rapidly reflect the changing trajectory from phase to phase. It showed that the severe forest changes mainly occurred in counties of Fengning, Weichang, Zhangbei and Guyuan during the years of 2000, 2010, and 2013.

Key words: forest in Bashang area, MODIS, Random Forest Algorithm, dynamic monitoring


为了揭示河北省坝上地区林地的动态变化特征, 采用250 m空间分辨率的MODIS反射率和NDVI数据, 利用TM遥感影像辅助选择训练样本, 基于随机森林分类算法, 提取2000—2015年8个时相的林地信息, 并分析其空间变化情况。结果表明, 与常用的最大似然法和神经网络法相比, 随机森林法分类的精度更高, 总体精度和 Kappa系数分别为91.89%和0.88。通过二进制编码方法, 快捷地揭示了8个时相的林地信息在空间上的动态变化, 识别出变化幅度较大的年份和空间分布。结果显示, 林地退化严重的地区集中在丰宁、围场、张北和沽源四县, 时间集中在2002, 2010 和2013年。

关键词: 坝上林地, MODIS, 随机森林算法, 动态监测

CLC Number: