Augmenting a Conceptual Model with
Geospatiotemporal Annotations
Vijay Khatri, Member, IEEE, Sudha Ram, Member, IEEE, and
Richard T. Snodgrass, Senior Member, IEEE
Abstract—While many real-world applications need to organize data based on space (e.g., geology, geomarketing, environmental
modeling) and/or time (e.g., accounting, inventory management, personnel management), existing conventional conceptual models do
not provide a straightforward mechanism to explicitly capture the associated spatial and temporal semantics. As a result, it is left to
database designers to discover, design, and implement—on an ad hoc basis—the temporal and spatial concepts that they need. We
propose an annotation-based approach that allows a database designer to focus first on nontemporal and nongeospatial aspects (i.e.,
“what”) of the application and, subsequently, augment the conceptual schema with geospatiotemporal annotations (i.e., “when” and
“where”). Via annotations, we enable a supplementary level of abstraction that succinctly encapsulates the geospatiotemporal data
semantics and naturally extends the semantics of a conventional conceptual model. An overarching assumption in conceptual
modeling has always been that expressiveness and formality need to be balanced with simplicity. We posit that our formally defined
annotation-based approach is not only expressive, but also straightforward to understand and implement.
Index Terms—Data semantics, database design, semantic model, geospatial databases, temporal databases.
æ
1INTRODUCTION
M
ANY real-world georeferenced (e.g., land information
systems, environmental modeling, transportation
planning, geomarketing, geology, archaeology) and time-
varying (e.g., accounting, portfolio management, personnel
management, inventory management) applications need to
organize data based on space and/or time. Underlying
these applications are temporal and/or geospatial data,
collectively referred to as geospatiotemporal data. Conceptual
database design is widely recognized as an important step
in the development of database applications [1], [3], [7]
such as those listed above. During conceptual database
design, a conceptual model provides a notation and formal-
ism that can be used to construct a high-level description of
the real world—referred to as a conceptual schema—inde-
pendent of implementation details. The data semantics
provides a mapping from the conceptual schema to aspects
in the real world. However, conventional conceptual
models [1], [3], [7] do not provide a straightforward
mechanism to explicitly capture the semantics related to
space and time. As a result, it is left to the database
designers to discover, design and implement—on an ad hoc
basis—the temporal and spatial concepts that they need. In
this paper, we present a methodical approach that
augments a conventional conceptual model using geospa-
tiotemporal annotations.
Many prior studies [10], [23] attribute project failures to
lack of identifying real needs during conceptual design.
One of the problems with developing geospatiotemporal
applications is that there is “a gulf between the richness of
knowledge structures in the application domains and the
relative simplicity of the data model in which the structures
can be expressed” [33], which in turn impacts the ability to
elicit the application requirements. Considering that geo-
graphic data are finding their way into traditional applica-
tions (e.g., insurance, retail, distribution), there is a need for
an overall geospatiotemporal conceptual database design
methodology that can be integrated into conventional
conceptual design. Thus, it would be helpful to develop
an approach that is compatible with an existing general-
purpose methodology [1], [3], [7].
Our annotation-based approach divides geospatiotem-
poral conceptual design into two steps: 1) elicit the current
reality of an application using a conventional conceptual
model without considering the geospatial and temporal aspects
(“what”) and, only then, 2) annotate the schema with the
geospatiotemporal semantics of the application (“when”
and “where”). Rather than creating new constructs in a
conceptual model, we use annotations to elicit the geospa-
tiotemporal aspects of the application. Our annotation-
based approach is generic and can be applied to any
conventional conceptual model [1], [3], [7] to transform that
model into a geospatiotemporal conceptual model. In this
paper, we apply our annotation-based approach to the
Unifying Semantic Model (USM) [22]—an extended version
of the Entity-Relationship (ER) Model [3]—to propose the
geoSpatioTemporal Unifying Semantic Model (ST USM).
We mention here the assumptions in this paper to
delineate the scope of our work.
1. Based on perception, space may be differentiated as
large-scale and small-scale [17]. As with Mark and
Frank [19], we construe large-scale space as equiva-
lent to geographic space. In the following, we use the
1324 IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEER ING, VOL. 16, NO. 11, NOVEMBER 2004
. V. Khatri is with the Information Systems Department, Kelley School of
Business, Indiana University, 1309 East 10th Street, BU 572, Blooming-
ton, IN 47405-1701. E-mail: vkhatri@indiana.edu.
. S. Ram is with the Department of MIS, University of Arizona, 1130 E.
Helen St., Tucson, AZ 85721-0108. E-mail: ram@eller.arizona.edu.
. R.T. Snodgrass is with the Department of Computer Science, University of
Arizona, 711 Gould Simpson, PO Box 210077, Tucson, AZ 85721-0077.
E-mail: rts@cs.arizona.edu.
Manuscript received 10 Jan. 2002; revised 26 Aug. 2003; accepted 13 Oct.
2003.
For information on obtaining reprints of this article, please send e-mail to:
tkde@computer.org, and reference IEEECS Log Number 115685.
1041-4347/04/$20.00 ß 2004 IEEE Published by the IEEE Computer Society
term space interchangeably to mean large-scale space
or geographic space.
2. According to Peuquet [21], absolute space is objective
since it provides an immutable structure that is
purely geometric. On the other hand, relative space is
an ordering relation between objects that determines
their relative position. We concentrate on absolute
representations, which are typically employed in
databases.
3. A database schema can evolve with time. Schema
versioning [24] is an important area of research;
however, we do not focus on schema versioning.
In summary, this paper focuses on establishing a
foundation for capturing the geospatio temporal data
semantics during conceptual design and does not delve
into peripheral research areas.
The rest of the paper is organized as follows: In Section 2,
we outline requirements related to geospatiotemporal
conceptual modeling. According to Wand et al. [31], “the
power of a modeling language lies in the semantics of its
constructs” and “ontology can be used to define concepts
that should be represented by the modeling language.” The
basis for annotations is the time and space ontol ogy
summarized in Section 3. We describe our annotation-
based geospatiotemporal conceptual design methodology,
which first focuses on “what” is important for an applica-
tion in the real world and then associates “what” with
“when” and/or “where.” In Section 4, we summarize a
conventional conceptual model, USM, which provides
various abstractions to capture “what” is important for an
application. In Section 5, we apply our annotation-based
approach to USM to realize ST USM, which captures the
semantics related to “when” and/or “where.” We round
out the paper with an evaluation and contributions. In this
paper, we provide the essence of our approach; complete
details are available in a comprehensive report [15].
2DESIDERATA
A precursor to designing and developing a geospatiotem-
poral conceptual model is identifying the conceptual
modeling requirements that need to be met. Based on a
hydrogeologic study at the US Geological Survey (USGS),
we provide an example of an application that needs to
capture the geospatiotemporal data semantics. We then
outline the evaluation criteria for a conceptual model that
can capture t he data semantics for geospati ote mporal
applications like that at the USGS.
2.1 Hydrogeologic Application
We are working with a group of researchers who are
developing a ground-water flow model [5] for the Death
Valley region in the state of Nevada. Beneath the earth’s
surface, there is a zone where all interstices are filled with
water referred to as ground water. The objective of the
ground-water flow model is to characterize regional 3D
ground-water flow paths so that policy-makers can make
decisions related to radionuclide contaminant transport and
the impact of ground water pumping on national parks and
local communities in the region. However, the quality of the
model output and the predictions based on these models
are dependent on the data that forms an input to these
models.
A large part of the input data for this model is geospatial
and/or temporal in nature. For example, two key objects of
interest in the application are spring-water sites and borehole
sites. Both these objects need to be spatially referenced to the
earth. A spring-water site is represented as a point whose
location onthe surface of the earth is given by the geographic x
and y-coordinates, with a geospatial granularity of degree.
Spring-wat er sites are points where spring discharge is
measured. Similar to springs, boreholes are access points to
the ground water system. Information of their construction
and condition are important for monitoring ground water
supply and remediation. A borehole site refers to a part of the
borehole whose 3D location is given by the x and y-coordinates
on the earth’s surface with a geospatial granularity of degree,
along with the depth below land surface with a geospatial
granularity of foot; there can be different borehole sites at
different depths at the same surface location. Physically, a
borehole is composed of hole-intervals with different dia-
meters. A borehole can also be thought to be composed of a
sequence of casings and openings. A casing is a section of a
borehole with concrete, steel, or plastic installed on the
borehole. An opening is a section of a borehole that is open to
allow water flow. Additionally, a borehole site may have
associated access tubes that provide access to a section of the
borehole. Casings, openings, hole interval, and access tubes
define the characteristics of a borehole, and the water-level
measurements taken at the borehole site are influenced by
these aspects.
A primary input data for the ground-water flow model
includes discharge (in cubic feet per second) at a spring-
water site and water depth (in feet below land surface) at a
borehole site, both of which are collected by a source
agency. Discharge and water depth need to be associated
with the time of measurement (in minute). The researchers
evaluate the collected water level and discharge measure-
ments to decide which of them will be included as input for
the ground-water flow model; the measurements used as an
input to the model are referred to as io-water-level (that is,
input-output water level) and io-discharge (that is, input-
output discharge), respectively. There are various hydraulic
tests conducted at the borehole site and the results of these
tests need to be coupled with the time (in minute) when the
test was conducted. Additionally, a borehole site may have
a pumplift that removes water from the borehole site; the
existence of a pumplift can affect other data collected at the
borehole site.
Capturing the data semantics related to, e.g., spring-
water site, borehole site, borehole, casing, water level,
source agency, requires a proposed spatiotemporal con-
ceptual model to:
1. allow the data analys t to model nongeospatial,
nontemporal, geospatial, and temporal aspects of
the application in a straightforward manner,
2. provide a framework for expressing the structure of
spatiotemporal data that is easily understood and
communicated to the users,
3. support a mechanism for a methodical translation
into implementation-dependent logical models, and
4. include a mechanism to represent various spatial
and temporal granularities (e.g., minute, second,
degree) in a conceptual schema.
Having summarized some of the requirements for a
typical geospatial application, we next describe evaluation
criteria for a geospatiotemporal conceptual model designed
to capture the data semantics illustrated above.
KHATRI ET AL.: AUGMENTING A CONCEPTUAL MODEL WITH GEOSPATIOTEMPORAL ANNOTATIONS 1325
2.2 Evaluation Criteria
Batini et al. [1] posit that conceptual models should possess
the following qualities: expressiveness, simplicity, minim-
ality, and formali ty. Additionally, to augment extant
conventional conceptual models [1], [3], [7] with geospatio-
temporal concepts, we need to take into account upward
compatibility and snapshot reducibility [2].
Expressiveness refers to the availability of a large variety
of concepts for a more comprehensive representation of the
real world. Wand et al. [31] posit that “conceptual modeling
can be anchored in the models of human knowledge” and
that ontology be employed as the basis for a proposed
formalism. One of the conflicting goals related to expres-
siveness is simplicity, which requires that the schema
developed using a conceptual model be understandable to
both users and data analysts. Prior research [20] contends
that one of the deficiencies of the existing conceptual
models that can represent geographic phenomena is their
inability to “represent information in way that is more
natural to humans.” While minimality ensures that no
concept can be expressed through composition of other
concepts, formality specifies that the model must present a
unique, precise and well-defined interpretation. Similarly,
Wand et al. [32] posit that effective use of conceptual
modeling constructs requires that their meaning be defined
“rigorously.”
Upward compatibility [2] refers to the ability to render a
conventional conceptual schema geospatiotemporal with-
out impacting or negating that legacy schema, thus
protecting investment in t he existing schemas. I t also
implies that both the legacy schemas and the geospatio-
temporal schemas can coexist. Upward compatibility
requires that the syntax and semantics of the traditional
conceptual model [1], [3], [7] remain unaltered. Snapshot
reducibility [2] implies a “natural” generalization of the
syntax and semantics of extant conventional conceptual
models [1], [3], [7] for incorporating the geospatiotemporal
extension. Snapshot reducibility ensures that the semantics
of geospatiotemporal model are understandable in terms of
the semantics of the conventional conceptual model. Here,
the overall objective is to help ensure minimum additional
investment in a data analyst training.
Juhn and Naumann [13] posit that conceptual represen-
tations “drive discovery” and should be precisely and
rigorously defined; on the other hand, discovery needs to be
“validated,” and the schemas should be clear and compre-
hensible. In essence, the challenge of adding the space and
time dimension is balancing simplicity and understand-
ability with preciseness and completeness.
3ONTOLOGY-BASED GEOSPATIOTEMPORAL
SEMANTICS
Geographic applications require data referenced by geo-
graphic coordinates, with time sometimes referred to as the
fourth dimension. We summarize key temporal, geospatial,
and time-varying geospatial terminology in this section.
Next, we describe the annotation syntax and illustrate how
ontology is the basis for annotations.
3.1 Temporal Ontology
A time domain is denoted by the pair (T; ), where T is a
nonempty set of time instants and “ ” is a total order on T .
We ass ume that the tim e domain is discrete (as the
measurements modeled are captured at a time known to a
discrete value). For example, (Z; ) represents a discrete
time domain where instants are denoted by int egers,
implying that every instant has a unique successor. An
instant is a point on the time line. The time between two
instants is referred to as a time period. An unanchored
contiguous portion of the time line is called a time interval,
e.g., one day (or Gregorian day). Unlike time periods, a time
interval is a directed duration of time with “no specific
starting or ending instants” [11]. A nondecomposable time
interval of fixed minimal duration is referred to as a
chronon. A finite union of nonoverlapping time periods is
referred to as a temporal element [8].
A temporal granularity—an integral part of the temporal
data—is defined as a mapping TG from index i to subsets of the
time domain [6]. Although the index of a temporal granular-
ity is constrained to be contiguous, the granules are not
constrained to be contiguous on the time domain. Thus, a
temporal granularity defines a countable set of nondecom-
posable granules TGðiÞ. Additionally, a special granule
called the origin, TGð0Þ, is nonempty. Some examples of
temporal granularities are Gregorian-day (with each such
granule composed of a sequence of 24 contiguous Gregorian-
hour granules, or 86,400 contiguous Gregorian-secon d
granules), business-day and business-week (with each such
granule composed of five Gregorian-day granules). While
Gregorian-day is a temporal granularity with contiguous
granules of day, business-day contains some noncontiguous
granules (e.g., Friday is followed directly by Monday). Each
nonempty granule may have a textual representation referred
to as a label (e.g., “2001-10-5 EST”), which can be mappedto an
index integer with label mapping. A designated point of time is
referred to as an anchor with respect to the time domain. The
union of granules is called an image of a temporal granularity.
Facts can interact with time in two orthogonal ways
resulting in transaction time and valid time [28]. Valid time
denotes when the fact is true in the real world and implies
the storage of histories related to facts. Existence time, which
applies to an object, is the valid time when the object exists.
On the other hand, transaction time links an object to the
time it is current in the database and implies the storage of
versions of a database object. While the temporal granular-
ity can be specified for existence time and valid time, that
for transaction time is system-defined. Time-varying data
may be modeled as an event or a state. An event occurs at a
point of time, i.e., an event has no duration. A state has
duration, e.g., a storm occurred from 5:07 PM to 5:46 PM.
3.2 Geospatial Ontology
Any data that can be associated to a location on the earth is
referred to as geographic data. A space domain may be
represented as a set (e.g., R
3
; R
2
; N
3
; N
2
) with elements
referred to as points. For geographic applications, horizontal
space is segregated from vertical space; correspondingly, we
define horizontal and vertical geospatial gran ularities [16].
Intuitively, the horizontal space domain corresponds to the
earth’s surface while the vertical space domain corresponds
to the depth/height below/above the sea level. We define a
horizontal geospatial granularity as a mapping from integers
to any partition of the horizontal space; the partition may
arise from pixellation of space and may be a regular square or
any other shape like triangular irregular network (TIN) or
even irregular shapes (e.g., county). Examples of horizontal
geospatial granularities are dms-deg and dms-min.
A geospatial object is associated with position and
geometry. The position in space is based on the coordinates
1326 IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEER ING, VOL. 16, NO. 11, NOVEMBER 2004
in a reference system, e.g., latitude and longitude. Geometry
represents the shape of an object: a point,aline, and a region.
A point is “a zero-dimensional geospatial object with
coordinates,” a line is “a sequence of ordered points, where
the beginning of the line may have a special start node and
the end a special end node,” and a region consists of “one
outer and zero or more inner rings” [30]. The difference
between a line and a region is that the line itself is the
carrier of information, while the area is of primary
importance for a region.
3.3 Time-Varying Geospatial Ontology
In geography, space is indivisibly coupled with time. Three
types of interaction between an object and space-time are
possible [29]:
1. moving objects, i.e., objects whose position changes
continuously but the shape does not (e.g., a car
moving on a road network),
2. objects whose geospatial characteristics and position
change with time discretely, i.e., changing shape
(e.g., a change in the shape of land parcels for a
cadastral application), and
3. integration of the above two behaviors, i.e., contin-
uous moving and changing phenomena (e.g., mod-
eling a storm).
Having summarized the temporal, geospatial, and time-
varying geospatial semantics that need to be captured in a
geospatiotemporal conceptual model, we describe how
ontology manifests into annotations.
3.4 Geospatiotemporal Annotations
Annotations provide a mechanism to specify the context
(“when” and “where”) associated with “what” is important
in the real world. As shown in Fig. 1, the overall structure of
an annotation phrase is:
htemporal annotationi==hgeospatial annotationi==
htime-varying geospatial annotationi:
KHATRI ET AL.: AUGMENTING A CONCEPTUAL MODEL WITH GEOSPATIOTEMPORAL ANNOTATIONS 1327
Fig. 1. Annotation syntax in BNF.
The temporal annotations, geospatial annotations, and time-
varying geospatial annotations are each separated by a
double forward slash (//).
The temporal annotation first specifies existence time (or
valid time) followed by transaction time. The temporal
annotation for valid time and transaction time is segregated
by a forward slash (/). Any of these aspects can be specified
as not being relevant to the associated conceptual construct
by using “-.” Valid time or existence time can be modeled as
an event (E) or a state (S) and has an associated temporal
granularity. For example, “S(min)/T//” a ssociated with
DISCHARGE denotes that DISCHARGE exists in a bitem-
poral space and that the temporal granularity of the states
(S) is minute (min). Additionally, we also need to capture
transaction time (T) associated with DISCHARGE. In this
example, the granularity associated with transaction time is
not specified, as it is system-defined.
The geospatial annotation includes geometry and posi-
tion in x, y,andz-dimension, and each dimension is
segregated by a forward slash (/). For example, “// P(deg)
/ P(deg) / -” for SPRING_SITE describes a geometry of
points (P) in the x-y plane. The associated horizontal
geospatial granularity is degree .
The interaction between an object and space-time can
result in a change in the shape and/or a change in the position
of an object. A time-varying geospatial annotation can be
specified only if geospatial and temporal annotation have
already been specified. For example, a moving car tracked by
satellite may be represented by an annotation phrase “E(sec) /
- // P(deg) / P(deg) / - // Pos@xy” that denotes a time-varying
position and a time-invariant shape. The geometry is a point
(P) in the x-y plane with geospatial granularity of degree. The
position changes in the x-y plane (Pos@xy) over time and
each geometry is valid for time granules (E) measured in
second. Our annotation also includes a formalism to model
indeterminacy; details related to modeling indeterminacy
can be found elsewhere [16].
Having outlined geospatiotemporal annotations, we next
apply our annotation-based approach to a conventional
conceptual model, USM, to propose a geospatiotemporal
conceptual model called ST USM. However, our annota-
tion-based approach is not specific to USM and can be
applied to any conventional conceptual model [3], [7]. In the
next two sections, we exemplify our geospatiotemporal
conceptual modeling methodology via USM and ST USM.
4 USM: REPRESENTING “WHAT”
The abstractions supported by typical conventional con-
ceptual models [1], [3], [7], [22], [27] include classification,
association, aggregation, and generalization/specialization. The
underlying principle of these abstractions is selective
emphasis of detail. We summarize below the data semantics
that can be elicited using conventional conceptual model-
ing, specifically USM. Fig. 2 illustrates a USM schema that
represents “what” is important for the hyd rogeologic
application described in Section 2.
All real-world objects are referred to by the term entity.
Characteristics or properties of entities are called attributes
(A
i
, where i ¼ 1; ...;n). Each attribute has an attribute domain
1328 IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEER ING, VOL. 16, NO. 11, NOVEMBER 2004
Fig. 2. The USM schema for ground-water flow model.
