Categorical databases are widely used in GIS for different kinds of application, analysis, planning, evaluation and management. Database generalization that derives different resolution databases from a single database with more detail is one of the key research problems and a hot research point in the GIS and Cartography field. This dissertation presents a framework for categorical database generalization in GIS. It includes defining conceptual aspects of current categorical database generalization transformation and constraints for generalization transformation, elaboration on supporting data structure and transformation units, development of auxiliary analysis methods, and demonstration of some application examples. Database generalization is considered as a transformation process. Three kinds of transformation are defined based on the characteristics of categorical database and categorical database generalization. They are geo-spatial model transformation, object transformation and relation transformation. Each transformation has a certain function and deals with some aspects of database. Geo-spatial transformation is mainly used to define the content framework of a new database and decide the theme of a new database. Object transformation and relation transformation deal with transformations of thematic and geometric aspects of objects and relationship between objects from an existing database to a new database. Database generalization (transformation) requires a data structure that strongly supports data organization, spatial analysis and decision-making in a database. The design of a data structure should take two functions into account. One provides the basis for describing and organizing spatial objects and the relationships between them. and the other is for analyzing and supporting operations on spatial objects. This thesis introduces the IEFDS, an integrated and extended version of FDS, as a data model to support automated database generalization transformation. The addition to FDS is triangles. The triangles and their classification are proposed based on constituent properties of triangles in IEFDS which plays an important role in the extended adjacent and inclusion relations and extracting the skeleton line. Some examples of spatial query operations that make use of the extended adjacent relation and semantic triangles are also provided in this thesis. In a categorical database, similarity between object types can be described by a similarity measure. The similarity is application-dependent. In a sense, the similarity will control and guide database transformation operations. The similarity evaluation model and similarity matrix are proposed for analyzing and representing similarity between objects and object types in this study which is based on Set-theory, classification and aggregation hierarchy. The constraints such as transformation conditions play a key role in the process of databasegeneralization. Constraints can be used to identify conflicting areas, guide choices of operationsand trigger operations as well as govern the database generalization. The processes of generalization should be performed by a series of operations under the control of constraints. Three types of constraints, data model. object and relationships based on an object-oriented database are proposed in landuse database generalization. These constraints can be specified interactively by users and varied to reflect different objectives or purposes. These types of constraints are applicationdependent. This will make the database generalization process very flexible/adaptive, and the decisionmaking can be based on geographic meaning and not simply on the geometry of an object. An important element proposed in this study is the transformation unit. It is an important process unit as many generalization problems need to be solved by considering a subset of related objects as a whole, rather than treating them individually. In a sense, the transformation unit is a basic analysis. processing, decision-making unit and a trigger to aggregation operation processes and it plays an important role in database transformation. The conflicted objects and its (their) related objects are organized into a transformation unit. A transformation unit that "brings together- a subset of objects can he created by conflict,; in thematic and /or geometric aspects of objects or spatial relation among objects or integrating them. The main purpose of creating a transformation unit is for the preparationofan aggregation operation. It limits the area and numberofa setofrelated objects in an aggregation operation. The different conflict types will create different types of transformation units. For this study, four types of transformation units are considered based on the constraints discussed. Each of which has a corresponding aggregation operation. The auxiliary analysis methods (algorithms) are needed to actually perform spatial analysis and transformations. The most fundamental tasks are to identify where to generalize, how to generalize, and when to generalize. The thesis introduces a number of auxiliary analysis methods that have been developed to solve a number of important geometric and thematic problems in database trans, form ati on. These auxiliary analysis methods include semantic similarity rnatrix, computing a model of similarity, detection and creation of transformation units, area object aggregation analysis and the process based on transformation units, multineighborhood, object cluster and creation of catchments hierarchy etc. Such examples of the application are included in the thesis as object cluster, land use aggregation and automated organization of hierarchical catchments. The application examples demonstrate the applicability and benefits of the IEFDS and similarity evaluation model. These supporting models play a key role in organizing thematic and geometric data, spatial analysis and spatial query in database generalization. It also proved that a lot of critical geometric and thematic problems in database generalization can be solved, or can be solved in a more efficient way, with the support of an adequate data model
