A Prototype Feature System for Feature Retrieval Using Relationships

Beginning of article

Introduction

Conventional geographic information systems (GIS) have managed spatial data with layers that represent a group of features. The layer-based model has emerged as a method for managing spatial geometry and thematic attributes. In the real world, however, geographic phenomena are characterized by the dimensional attributes of space, theme, and time as well as relationships for each of these attributes. Individual features, rather than layers, should thus be used to represent real world phenomena using GIS.

A feature concept to represent geographic phenomena in GIS was developed in the 1980s (Guptill et al. 1990; U.S. Geological Survey 1997). A feature is a defined geographical entity represented in the digital environment as an object and used as the unit of aggregation and analysis (Usery 1993). A feature is represented by multiple dimensional attributes and relationships using space, theme, and time. Recently, the Open Geospatial Consortium (OGC) released the specifications of a general feature model that provides a framework for building GIS (OGC 1999a). Therefore, a feature approach for a geographic data model may be an optimal solution for building GIS.

Current commercial GIS such as the ArcGIS package, 4Dlink, and STEMgis are based on the feature approach, however, they all treat features as a group instead of as individual instances when representing relationships. For example, a relationship class of a geodatabase in the ArcGIS package represents relationships between different feature classes. Each instance in a feature class, however, may have relationships with other instances in the same feature class, which is not supported by current GIS.

Furthermore, current commercial GIS provide a limited method to store temporal relationships, even though many temporal GIS models and prototypes have been provided in the literature (Worboys 1992 and 1994; Peuquet 1994; Claramunt and Theriault 1995; Yuan 1996 and 1999; Le 2005; Choi et al. 2008). Tracking Analyst in ArcGIS, for example, uses a temporal layer to animate the change of a feature. Temporal .information is stored in a column Date_stamp of a feature class table in a temporal layer. Because all information of a feature is stored in one row in a feature class table, a feature has only one geometry and Datestamp. The change of a feature is represented by one event_ID with multiple Feature_IDs through multiple Date_stamps. This is an event-oriented method which uses multiple feature_IDs for an event. In reality, a feature has multiple events through time. In each event, the geometry and themes of a feature may be changed. In a feature-oriented approach, a feature can have multiple geometry and themes through time so that the change of a feature can be stored and retrieved in a temporal relationship class using a unique feature_ID instead of multiple feature_IDs. The temporal relationships of a feature should represent not only history of a feature (Le 2005; Choi et al. 2008) but also the relationships between features (Peuquet 1994).

While feature representation using three dimensions (spatial, theme, and time) is implemented in commercial GIS such as ArcGIS, not much research has been done on building GIS that would effectively manage dimensional relationships of individual features. The objectives of this paper are to extend OGC's general feature model in order to incorporate feature relationships, and to develop a prototype feature-oriented GIS (FOGIS) based on the feature data model, which can manage dimensional attributes and relationships of each individual feature. To store those attributes and relationships, a feature database is also designed and used in the developed system.

The following discusses feature representation in GIS. Then, a feature data model is designed based on the OGC's feature data model. The subsequent section discusses the implementation of the proposed feature data model. Next, the system is applied for storing and querying features using temporal and non-spatial relationships. The paper concludes with a summary of key ideas.

Feature Representation in a GIS

With the emergence of the feature concept, various generic lists of features useful in GIS operations have been compiled. These include the Topologically Integrated Geographic Encoding and Referencing (TIGER) system of the U.S. Census Bureau (Trainor 1990), the Digital Line Graph-Enhanced (DLG-E) of the U.S. Geological Survey (U.S.GS) (Guptill et al. 1990), feature lists for military requirements by the U.S. Defense Mapping Agency (DMA 1987), and the Feature and Attribute Coding Catalogue (FACC) of the National Geospatial-Intelligence Agency (NGA) (DGIWG 2000). However, these feature lists tend to vary according to their applications.

To provide standard features for representing geographic phenomena, USGS (1997) designed the conceptual feature model for the Spatial Data Transfer Standard (SDTS). The SDTS conceptual model has three parts: a model of spatial phenomena, a model of the spatial objects used to represent phenomena, and a model of spatial features that explains how spatial phenomena and spatial objects are related. The SDTS conceptual model identifies all possible features using abstraction methods of object-orientation--classification, generalization, aggregation, and association. However, it does not provide any details about feature relationships.

To enable the representation of feature relationships within SDTS, a conceptual framework of features has been built (Usery 1996). This framework includes both attributes and relationships of the spatial, thematic, and temporal dimensions of features. With this conceptual framework in place, prototype feature-based GIS (FBGIS) have been previously developed for both vector (Tang et al. 1996; Usery et al. 2002) and raster data representation (Usery 1994). All these prototype systems focus on the three-dimensional attributes but they do not utilize dimensional relationships explicitly.

Abstract specifications for a general feature model including feature geometry, feature collection, feature, and relationship between features were released by the OGC late in the 1990s (OGC 1999a, 1999b, 1999c, and 2001). According to these specifications, feature geometry specifies all types of geometry and their structures. Feature collection specifies a feature that is a composite of other features (i.e., a network). Geographic phenomena in the feature specification are modeled based on four types of attributes: locational, spatial, non-spatial, and temporal.

The specifications for geographic phenomena are intentionally detailed so that they can be used for building a GIS. However, the specification for feature relationships only models the relationships between feature classes, such as a one-to-one relationship between a road class and a river class or a one-to-many relationship between an agent class and a house class. The relationship between feature instances in the same feature class is missing. For example, a road "Broad Street" is a part of a road "Highway 80" and they are instances of the same road class. In this paper, therefore, we designed a feature data model that can accommodate relationships between individual features.

Feature Data Model with Relationships

The OGC's general …