基于环境减灾卫星CCD数据与决策树技术的植被分类研究

doi:10.13249/j.cnki.sgs.2012.012.1488

摘要/Abstract

摘要：

以内蒙古呼伦贝尔地区为例,基于遥感数据获取区域7种典型植被的NDVI时间序列曲线。在此基础之上,分析曲线趋势及其特征值,研究基于曲线差异的植被分类信息提取方法。同时,以国产环境减灾卫星CCD数据作为主要遥感数据源,提取研究区5月上旬与8月上旬两期NDVI数据及其比值,采用决策树分类方法研究得到区域30 m空间分辨率植被分类结果。经实地验证,一级类型总体分类精度为83.64%,二级类型为70.91%,其中乔木林的分类精度最高,然后是农田与草地,灌丛的分类精度相对最低。结果表明该方法能够快速、准确据提取植被分类信息,为国产环境减灾卫星CCD数据的广泛深入应用提供理论与数据支持。

关键词: 植被分类, 环境减灾卫星CCD数据, NDVI时间序列曲线, 呼伦贝尔

Abstract:

Vegetation is critical for researches about global environmental change and regional sustainable development, and remote sensing is an important method for obtaining classification result. However, the Normalized Difference Vegetation Index (NDVI) time series method classification was limited by the coarse spatial resolution, and application of the medium high data such as Landsat TM was limited by the coverage and accessibility of remote sensing data. The Chinese environmental mitigation HJ satellite CCD sensors are capable of large area, all-time monitoring, and have a great advantage in coverage and frequency of repeated observations. A case study of Hulunbuir, Inner Mongolia was carried out in this paper. The NDVI time series curve of 7 vegetation types were extracted from both MODIS and HJ CCD data. Then, the curves and eigenvalue were analyzed. The result showed that between the 7 vegetation types, there was significant differences in the value range of early May NDVI, early August NDVI and the ratio of the two NDVI image. The vegetation classification rules were extracted based on these differences. The HJ-CCD was used as the main data sources in this paper. Three images including two NDVI and one ratio were extracted and the decision tree method was applied. Based on the result, 30 m spatial resolution vegetation classification result was carried out. By field verification, the result shows a 83.64% overall accuracy in the level one classification, and 70.91% in the level two classification. The cartographic accuracy of evergreen coniferous forest can achieve 100%, followed by cropland 82.61%, mixed forest 76.19% and desert steppe 75%. The accuracy of shrub is relatively low to 50%. This result proved a fast, simple and accurate method for vegetation classification, and provided the theory and data support for application of the Chinese HJ satellites.

Key words: vegetation classification, HJ-CCD data, NDVI time series curve, Hulunbuir

中图分类号:

TP79

刘睿, 冯敏, 孙九林, 廖顺宝, 王卷乐. 基于环境减灾卫星CCD数据与决策树技术的植被分类研究[J]. 地理科学, 2012, 32(12): 1488-1495.

Rui LIU, Min FENG, Jiu-lin SUN, Shun-bao LIAO, Juan-le WANG. The Vegetation Classification Based on HJ-CCD Data and Decision Tree[J]. SCIENTIA GEOGRAPHICA SINICA, 2012, 32(12): 1488-1495.

图/表 12

图1

表1

图2

图3

表2

图4

图5

表3

图6

图7

表4

表5

参考文献 21

[1]	Rees M, Condit R, Crawley M, et al.Long-term studies of vegetation dynamics[J]. Science, 2001, 293: 650-655. pmid: 328097111516348217162284200129222322221147410188561568643785470271
[2]	Hansen M C, DeFries R S. Detecting long-term global forest change using continuous fields of tree-cover maps from 8-km Advanced Very High Resolution Radiometer (AVHRR) data for the years 1982-99[J]. Ecosystems, 2004, 7(7): 695-716. doi: 10.1007/s10021-004-0243-3
[3]	Lunetta R S, Knight J F, Ediriwickrema J, et al.Land-cover change detection using multi-temporal MODIS NDVI data[J]. Remote Sensing of Environment, 2006, 105(2): 142-154. doi: 10.1016/j.rse.2006.06.018
[4]	南颖, 刘志锋, 董叶辉, 等. 2000~2008年长白山地区植被覆盖变化对气候的响应研究[J]. 地理科学, 2010, 30(6): 921~928.
[5]	孙艳玲, 杨小唤, 王新生, 等. 基于决策树和MODIS数据的土地利用分类[J]. 资源科学, 2007, 29(5): 169~174.
[6]	张景, 姚凤梅, 徐永明, 等. 基于MODIS的土地覆盖遥感分类特征的评价与比较[J]. 地理科学, 2010, 30(2): 248~253.
[7]	王长耀, 骆成凤, 齐述华, 等. NDVI-Ts空间全国土地覆盖分类方法研究[J]. 遥感学报, 2005, 9(1): 93~99. doi: 10.11834/jrs.20050114
[8]	林川, 宫兆宁, 赵文吉. 基于中分辨率TM数据的湿地水生植被提取[J]. 生态学报, 2010, 30(23): 6460~6469.
[9]	Xie Y, Sha Z, Yu M.Remote sensing imagery in vegetation mapping: a review[J]. Journal of Plant Ecology, 2008, 1(1): 9~23.
[10]	刘睿, 孙九林, 王卷乐, 等. 环境与灾害监测预报小卫星CCD数据质量评价[J]. 地球科学进展, 2011, 26(9): 971~979.
[11]	Gutman G, Byrnes R, Masek J, et al.Towards monitoring land-cover and land-use changes at a global scale: The global land survey 2005[J]. Photogrammetric Engineering and Remote Sensing, 2008, 74(1): 6-10.
[12]	Savtchenko A, Ouzounov D, Ahmad S, et al.Terra and aqua MODIS products available from NASA GES DAAC[J]. Trace Constituents in the Troposphere and Lower Stratosphere, 2004, 34(4): 710-714. doi: 10.1016/j.asr.2004.03.012
[13]	Levelt P F, van den Oord G H J, Dobber M R, et al. The ozone monitoring instrument[J]. Geoscience and Remote Sensing, IEEE Transactions on, 2006, 44(5): 1093-1101.
[14]	刘睿, 孙九林, 张金区, 等. 中国北方草地覆被的HJ星NDVI校正研究[J]. 草业学报, 2011, 20(1): 189~198. doi: 10.11686/cyxb20110127
[15]	诸云强, 孙九林, 宋佳, 等. 地学E-Science研究与实践——以东北亚联合科学考察与合作研究平台构建为例[J]. 地球科学进展, 2011, 26(1): 66~74.
[16]	张增祥, 汪潇, 王长耀, 等. 基于框架数据控制的全国土地覆盖遥感制图研究[J]. 地球信息科学学报, 2009, 11(2): 216~224. doi: 10.3969/j.issn.1560-8999.2009.02.013
[17]	廖顺宝, 刘睿, 尹芳. 全国1:25万土地覆被遥感制图精度在内蒙古东部的实地验证[J]. 地理研究, 2011, 30(3): 555~563.
[18]	中国科学院中国植被图编辑委员会.中国植被图集(1:100万)[M].北京:科学出版社,2001.
[19]	金良. 呼伦贝尔市土地利用动态变化研究[J]. 干旱区资源与环境, 2004, 18(2): 110~114. doi: 10.3969/j.issn.1003-7578.2004.02.022
[20]	宫攀. 基于MODIS数据关键物候特征参数的东北地区植被覆盖分类[J]. 资源科学, 2010, 32(6): 1154~1160.
[21]	陈彬彬, 王卫平, 黄文震, 等. 呼伦贝尔市降水时空分布特征研究[J]. 干旱区资源与环境, 2008, 22(9): 71~75. doi: 10.3969/j.issn.1003-7578.2008.09.014

遥感传感器	空间分辨率(m)	时间分辨率(d)	波段数
MODIS	250/500/1000	1/4	36
HJ/CCD	30	2	4

一级类型	二级类型	含义
森林	常绿针叶林	郁闭度>30%,高度>2 m的常绿针叶天然林和人工林
	落叶针叶林与混交林	郁闭度>30%,高度>2 m的落叶针叶林、阔叶天然林和人工林
	灌丛	郁密度>40%,高度>2 m的灌丛和矮林
草地	草甸草地	覆盖度>30%,以草本植物为主的各类草地
	典型草地	覆盖度在10%~30%,以旱生草本为主的草地
	荒漠草地	覆盖度在5%~10%,以强旱生植物为主的草地
农田		种植农作物的耕地

	常绿针叶林	落叶针叶林与混交林	灌丛	农田	草甸草地	典型草地	荒漠草地
5月上旬	0.50~0.70	0.35~0.50	0.26~0.38	0.19~0.25	0.25~0.30	0.20~0.30	0.15~0.22
8月上旬	0.85~0.9	0.84~0.91	0.78~0.88	0.65~0.90	0.55~0.75	0.40~0.55	0.30~0.42
8月上旬/5月上旬	1.3~1.7	1.6~2.4	2.1~3.2	2.7~4.7	1.8~3.0	1.3~2.8	1.4~2.8

	森林	草地	农田	非植被	总和	制图精度(%)
森林	24	1	3	0	28	85.71%
草地	4	47	5	0	56	83.93%
农田	1	3	19	0	23	82.61%
非植被	0	1	0	2	3	66.67%
用户精度(%)	82.76%	90.38%	70.37%	100.00%
总和	29	52	27	2	110
总分类精度：83.64%			Kappa系数：0.73