从信息熵的基本概念出发，认为地学空间数据子集划分产生的互信息或熵减源于子集划分，使得各个子集的不确定性或模型糊降低，并且子集之间的差异性增在 最大熵减的子集划分方案代表一定的地学模式和地不规律。以此为基础分别探讨了地学数据属性要素的子集划分产生多维属性关联规则，以间和时间的子集分割来进行了聚类的方法。

More and more interest has been paid on Geodata mining and knowledge discovery from large database with the rapid growth of Geodata volume and eagerness for the Geoknowledge. This paper presents a dataset partition model based on information entropy and mutual information. The author argued that the largest information entropy deduction is in accordance with the significant Geodata pattern. With this kernel theoretical base, information entropy based decision tree model and spatial temopral clustering by partition model were developed.