自然资源学报

自然资源学报 ›› 2002, Vol. 17 ›› Issue (6): 750-756.doi: 10.11849/zrzyxb.2002.06.014

• 研究方法 • 上一篇    下一篇

青藏高原生态资产地域划分中的SOFM网络技术

李双成1, 郑度2, 张镱锂2   

  1. 1. 北京大学城市与环境学系,北京100871;
    2. 中国科学院地理科学与资源研究所 土地科学中心,地表过程分析与模拟教育部重点实验室,北京100101
  • 收稿日期:2002-04-26 修回日期:2002-07-19 出版日期:2002-12-25 发布日期:2002-12-25
  • 作者简介:李双成（1961～)，男，河北平山人，副教授，博士后，研究方向为区域环境与生态系统评价与持 续化管理，地学非参数化建模等，近年发表论著40余篇（部～)。E-mail:scli@sohu.com
  • 基金资助:

    国家重点基础研究发展规划(G1998040816)经费资助。

Application of SOFM neural network to ecological assets regional ization in Qinghai-Tibet Plateau

LI Shuang-cheng1, ZHENG Du2, ZHANG Yi-li2   

  1. 1. Department of Urban and Environmental Sciences,Peking University,The Center for Land Studies,Lab for Earth Surface Processes,Beijing100871,China;
    2. Institute of Geographic Sciences and Natural Resources Research,CAS,Beijing100101,China
  • Received:2002-04-26 Revised:2002-07-19 Online:2002-12-25 Published:2002-12-25

PDF (PC)

赞

被引次数

76 | 20

摘要： 针对目前地域划分中存在的问题,论文尝试以人工神经网络技术作为区划工作的理论支撑,构建了自组织特征映射SOFM网络,以青藏高原环境与生态系统资产作为待分客体,探索了新技术和方法在生态资产地域划分中的应用。结果表明,对于自然界中广泛存在的非线性问题,SOFM网络具有比聚类分析等线性分类器更强的适应性。应用SOFM网络在对待分客体生态资产进行类型划分的基础上,使用策略性循环尺度转换(SCS)范式对其进行了区域转换,最终完成了青藏高原范围内生态资产的地域划分。

关键词: 地域划分, 人工神经网络, SOFM网络, SCS范式, 青藏高原

Abstract: Artificial neural networks(ANNs)whose elements are inspired by biological nervous systems are composed of simple elements operating in parallel.Commonly neural networks are adjusted,or trained,so that a particular input leads to a specific target output.Neural networks have been trained to perform complex functions in various fields of application including predic-tion,pattern recognition,system identification,classification and optimization.Conventional statistical models fail to deal with non-linear relations among the physical factors.However,as an alternative approach,ANNs can map complex temporal and spatial pat-terns by using non-linear transfer functions.In this paper,regionalization of ecological assets is conducted by unsupervised artificial neural network,namely Self-Organizing Feature Mapping(SOFM).The field data employed as input for training represent spatial ecological features such as longitude,latitude,annual mean temperature,annual mean precipitation,aridity,biological tempera ture,assets demand index,assets scarcity,NPP of unit area and ecological value of unit area collected at84sites on Qinghai-Tibet Plateau.After the iterative learning phase in the SOFM analysis,each of the84sites is associated with an output unit.Each output unit contains some of the sites and there is obvious discrete grouping of cases.The SOFM,therefore,appears to have organized the sites such that the various output units are associated with different eco-logical assets classes.In order to assess the performance of SOFM,the comparison with cluster analysis is carried out.The result indicates that the overall performance of the neural network algorithm was better than that of cluster analysis for ecological regionalization.Finally,using SCS paradigm,conver-sion from SOFM classification to ecological assets regionalization is conducted.ternsbyusingnon linearconductedbyunsupervisedSOFM.Thefielddata

Key words: regionalization, artificial neural network, SOFM model, SCS paradigm, Qinghai-Ti-bet Plateau