地理科学进展 ›› 2011, Vol. 30 ›› Issue (7): 912-919.doi: 10.11820/dlkxjz.2011.07.018

• 土地利用与生态环境 • 上一篇    下一篇

土地利用变化预测CBR方法的适应性分析

孙晔然1,2, 杜云艳1, 苏奋振1, 周成虎1   

  1. 1. 中国科学院地理科学与资源研究所资源与环境信息系统国家重点实验室,北京100101;
    2. 中国科学院研究生院,北京100049
  • 收稿日期:2011-01-01 修回日期:2011-04-01 出版日期:2011-07-25 发布日期:2011-07-25
  • 通讯作者: 杜云艳,E-mail: duyyr@lreis.ac.cn
  • 作者简介:孙晔然(1987-),男,硕士研究生,主要研究方向为GIS和空间数据挖掘。E-mail: sunyr@lreis.ac.cn
  • 基金资助:

    国家863 计划重点项目(2009AA12Z148);国家自然科学重点基金项目(088RA400SA)。

Study on the Suitability of CBR Method in the Estimation of Land Use Change

SUN Yeran1,2, DU Yunyan1, SU Fenzhen1, ZHOU Chenghu1   

  1. 1. State Key Laboratory of Resources and Environmental Information System, Institute of Geographic Sciences and Natural Resources Research, CAS, Beijing 100101, China;
    2. Graduate University of Chinese Academy of Sciences, Beijing 100049, China
  • Received:2011-01-01 Revised:2011-04-01 Online:2011-07-25 Published:2011-07-25

摘要: 研究土地利用变化的方法很多,常见的有统计回归方法、概率统计方法、系统动力学方法、元胞自动机方法以及贝叶斯网络方法等。这些方法都各具特点,并得到了实际应用。虽然有些研究指出了一些方法的优点和不足,但是对其在土地利用变化问题上的适应性研究还比较有限。案例推理(Case Based Reasoning ,简称CBR)作为研究土地利用变化的一种新方法,目前同样缺乏适应性方面的研究。因此,本文在土地利用的CBR方法的研究基础上,具体探讨该方法在土地利用变化预测上的适应性问题。研究分别从案例的选取方式、模型指标的选取以及指标权重的设定3 个方面开展,通过对比试验以深入讨论3 个方面的因素对预测精度的影响。研究结果显示,在历史案例丰富的情况下,通过选择重要的指标,并对其赋予较高权重,可以保证CBR方法在预测土地利用变化时的稳定性。此外,CBR中“地理环境”组分的引入,有效地提高了土地利用变化预测的精度。研究表明,CBR在解决土地利用变化问题上具有简单灵活、适用范围广、预测精度高以及保持形态稳定的特点,是一种解决地学问题的新方法。

关键词: 案例推理(CBR), 三元组模式, 土地利用变化

Abstract: There are various approaches used to study land use change (LUC), such as regression analysis, probability statistics, system dynamics, cellular automata and Bayesian network. These approaches have their specific characteristics and practical applications in the LUC. Although there were some researches revealing the advantages and disadvantages of some approaches, there were relatively few studies on the suitability of these approaches. This study focused on the suitability of CBR approach for LUC estimation, on the basis of the CBR model for the LUC estimation. The comparison experiments were conducted from three aspects, selection approach of the test cases, selection of variables and weights of the variables, to explore the influences of these factors on the estimation accuracy of LUC. The land use changes in Zhuhai region, China during 1995-2000 were used as a case study to conduct the comparative experiments. The concrete comparison strategies include: (1) To choose the test cases by selective approach and stochastic approach to explore the effects of the selection approach on the LUC estimation accuracy; (2) to neglect different variables in turn representing three categories of impacts respectively to explore the effects of the neglect of the variables on the estimation accuracy; (3) to change the weights of variables in turn to explore the effects of the weights of specific variables on the estimation accuracy. The experimental results are shown as follows. Firstly, the selection approach of test cases has insignificant effects on the LUC estimation accuracy under the circumstance that the historical cases are abundant. Secondly, the neglect of the ordinary variables has insignificant influences on the estimation accuracy on the condition that vital variables are selected. Thirdly, the weights of the ordinary variables have insignificant effects on the estimation accuracy in the event that greater weights are assigned to the vital variables. These results demonstrate that CBR is an effective method for solving LUC problems with the advantages of simple construction, wide application, high accuracy and stable pattern. The stability of the LUC estimation accuracy based on CBR approach can be kept on the condition of plentiful historical cases when vital variables are selected and higher weights are assigned to them. In this case, CBR method shows a good suitability for LUC estimation. In addition, the incorporation of the new component“geographic environment”into the CBR model efficiently improves the estimation accuracy of LUC.

Key words: cased-based reasoning (CBR), land use change (LUC), three-component model