COMMUNICATIONS OF THE ACM April 1997/Vol. 40, No. 4 103
Geographic information systems are
used to collect, analyze, and present
information describing the physical
and logical properties of the geo-
graphic world. Geographically refer-
enced data is the spatial data that
pertain to a location on the earth’s surface.
Shashi Shekhar, Mark Coyle, Brajesh Goyal,
Duen-Ren Liu, and Shyamsundar Sarkar
Using object-oriented database technology to model the real world.
Data Models in
Geographic
Information
Systems
There are four major functional units in a typical
geographic information systems (GIS):
• Data Input Unit. Measurements in GIS are taken
by sensors such as cameras and global positioning
systems. A manual process is then used for
inputting data that cannot easily be processed
automatically. The measurements are discretized,
for example, by imposing a regular, multidimen-
sional discrete grid over the surface to be mea-
sured, allowing points of interest to lie only at
the intersection of the grid lines. In addition to
the error imposed by this discretization process,
measurement errors also reduce the accuracy of
attribute values. The data input therefore needs
validation. Various integrity constraints including
topological constraints also need to be checked.
An example of a topological constraint is “Min-
neapolis should be inside Minnesota.”
• Data Model. A conceptual data model is a type of

data abstraction that hides the details of data
storage [9]. It uses logical concepts, which may
be easier for most users to understand. It supports
data input, manipulation and result presentation.
Many GIS are organized as a collection of themes.
Each theme represents the values of a unique
attribute of the geographic space. A theme may
independently partition, decompose, and frag-
ment the continuous space for a particular value
(or value range) of the attribute. The partitions
and fragments of space within each theme are
often stored within the database and can be
treated as enti-
ties or objects.
• Data Manipu-
lation Capabil-
ities.
Geographic
data is queried
and analyzed
for various
operations,
including spa-
tial searches and overlays. Operations on primi-
tive vector data-types include geometric
operations (e.g., area or boundary, intersection),
topological operations (e.g., connectedness) and
metric operations (e.g., distance).
• Result Presentation Facilities. A GIS presents
results visually (e.g., cartographically) in the form
of maps, consisting of graphic images with vector
data displayed over raster data; 3D display; ani-
mation; and cartographic production. Carto-
graphic maps sometimes highlight semantically
interesting information at the expense of loca-
tional accuracy. This phenomenon is called map
generalization.
Related Work and Contribution
The GIS data models can be categorized into field-
based models and object-based models. Field-based
models see the world as a continuous surface (layer)
over which features (e.g., elevation) vary. Layer alge-
bra [2] provides a field-based view. It defines a set of
operations that can manipulate different layers to
produce new layers. The object-based model treats
the world as a surface littered with recognizable
objects (e.g., cities, mountains, rivers), which exist
independent of their locations. GraphDB [7],
GODOT [5], Worboy [12], OGIS [3] and GeoOOA
[8] are some attempts to model GIS using the
object-based approach. OGIS provides a library of
spatial types (e.g., point, line, chain) and operations
on these types (e.g., intersect, overlap) to facilitate
data exchange across different GIS. GeoSAL, Wor-
boy and GODOT propose extensive class hierarchies
to model the geometry and topology of spatial
objects. GraphDB supports the explicit modeling
and querying of graphs. GeoOOA adds a geographic
dimension to each object modeling a spatial entity,
and it supports a fixed set of geometric types and
topological relationships. The GERM model [11]
attempts to unify the two approaches and provides a
set of concepts as an add-on to the ER model for
modeling GIS.
These models do not explicitly support the dis-
cretization process and interpolation to invert the
discretization. This omission leads to several com-
plexities in the existing models, including the
dichotomy between field- and object-based
approaches.
T
HIS ARTICLE PROPOSES A NEW MODEL
called the Geographic Information Sys-
tem Entity Relational model (GISER).
This model explicitly represents the
discretization aspects so as to unify the
two approaches. GISER extends the Enhanced
Entity Relationship (EER) model [9] to include con-
tinuous fields. The continuous fields are associated
with discretizations and interpolation models.
104 April 1997/Vol. 40, No. 4 COMMUNICATIONS OF THE ACM
Data Input
Data Modeling
Data Manipulation
Result Presentation
Data Consistency and Quality
Continuous Space
Geometry
Topology
Set Operations, Spatial Relationships,
Network Analysis
Visual Representation
Constraints, Interpretation
Discretization, Interpolation models
Points, curves, polygons, etc.
Networks, Partitions
Topological, Direction, Metric
Visualization constraints, Maps
Functional Unit Data Modeling Requirement Examples / Issues
Table 1. Summary of requirements
GISER attempts to
support the entire GIS
process, from
the input of data mea-
sured and discretized
to the display of this entity
and all the data processing that must be
performed.