土地利用变化预测的案例推理方法

当前,基于案例的推理(Case-Based Reasoning,CBR)在解决复杂的地学问题时.对地学案例的表达和历史案例的相似性计算与推理存在明显缺陷,需要在CBR的表达模型和空间相似性计算与推理算法进行拓展.本文针对土地利用变化问题,首先在分析土地利用变化各种定量方法基础上,提出利用CBR进行土地利用变化分析的研究思路:其次,针对土地利用变化的空间特性及隐含的空间关系特性.给出土地利用变化案例的表达模型,案例间内蕴空间关系抽取算法,以及考虑案例间空间关系的CBR相似性推理模型:最后,进行珠江口区域土地利用变化的CBR方法试验.预测精度达到80%.为了进一步评价CBR方法对土地利用变化预测的有效性.在实例部分采用同样的实验数据进行贝叶斯网络的预测方法实验.由两种方法对比可知.CBR是从复杂到简单进行地学问题求解的一种有效方法.

A CBR Approach for Land Use Change Prediction

A variety of methods, including Markov chains, multivariate statistics, optimization, system dynamics, and CLUE/CA, have been widely used to study land use change in different areas. Previous studies indicate that these methods obviously have their own pros and cons when they are applied to the studies on land use change. New approaches will probably provide a better alternative if it can assimilate some of the advantages of current available methods.Case-based reasoning (CBR) is an effective method which was widely used to study geographical problems. However, the CBR approach is far from perfect in presenting complicated geographical phenomena, particularly in computing and reasoning the similarity between current study cases to those ones that have been studied. Research is in great need to improve CBR-based geographic information portrayal modeling and reasoning algorithm. This paper reports a CBR-based method, including a spatial relationship extracting algorithm and a model describing the similar reasoning between spatially related cases. These methods were tested by examining the land use change in Zhuhai City, which is located on the western Pearl River Mouth of Guangdong, China. In order to evaluate the prediction accuracy derived from CBR-based method, we also use Bayesian network method to study land use change in our study area.As the results indicate, both CBR and Bayesian network approaches yielded similar prediction accuracy. However, the advantages in CBR approach are obvious, particularly in dealing with complicated geographic phenomena. When using the CBR method, it is unnecessary to define those complicated conversion regulations. Instead, the method predicts land use change simply based on knowledge retrieved from old cases, hence significantly improving the efficiency in building the case library, as well as case querying in the library. By contrast, Bayesian networks require extensive computation and more unrealistic assumptions, i.e., complete dataset, no preferred selection, and non-continuous variables.