Summary and Outlooks

Abstract

I have discussed from the conceptual, theoretical and empirical points of view the basic issues of knowledge discovery in spatial and temporal data. The kinds of knowledge that geographers are interested in are essentially spatial structures, processes, and relationships in various domains. The data that we are dealing with are voluminous, multi-scaled, multi-sourced, imperfect, and dynamic. Hidden in these complex databases are conceptually and practically meaningful structures, processes and relationships that might be important to the understanding and sustainable development of the human-land system. Our objective is to develop effective and efficient means to discover the potentially useful spatial knowledge that might otherwise lay unnoticed.

From our discussion, a basic task of knowledge discovery in spatial and temporal data involves the unraveling of structures or processes that might appear as natural clusters in data. I have discussed in Chap. 2 the main objectives and difficulties of such task and methods by which natural clusters can be identified. Therefore, clusters form the spatial knowledge hidden in the database in this kind of spatial data mining task. Geographical research often involves classification of spatial phenomena. Given pre-specified spatial classes, we need to discover from data a surface that can separate the classes which are constituents of the data set. Structure thus discovered serves as class separation surface in the general case. Instead of a separation surface, particularly under complicated situations, classes might be separated by a set of classification rules which can be unraveled from data. In Chaps. 3 and 4, a number of statistical and non-statistical methods for the discovery of separation surfaces and classification rules have been, respectively, examined. In the search for spatial relationships hidden in data, the basic issue is to uncover and determine whether or not a spatial association or a spatial casual relationship is local or global. In Chap. 5, I have examined various methods by which local and global spatial statistics can be established from data and the issue of spatial non-stationarity can be resolved. To be able to study the dynamics of spatial phenomena in various time scales, I have discussed the mining of scaling behaviors of spatial processes from data in Chap. 6. The discovery of time varying behaviors from temporal data is investigated on the basis of their self-similarity and long range dependence along a wide spectrum of temporal scales.