First CASSINI International Workshop
"Data Quality in Geographic Information: from Error to Uncertainty"
21-22-23 APRIL 1997 - PARIS - FRANCE
SCOPE
The scope of this workshop is to address the quality problem in geographic
information (GI) from several points of view, in order to help to modelize the
quality information and help to use it when combining different data for
different purposes.
Quality has to be quantified as well as quantities have to be qualified.
The "propagation" of the quality within a spatial analysis process starts as an
"error propagation" due to "location accuracy", but becomes rapidly much more
complex as soon as we are concerned by consequences on derived information:
topological relations, surface computations, interpolation
processes ....
A standardisation process is already ongoing to help GI providers to include
quality information together with geometric-geographic information. Is this
worth while ?
Will users require it ? use it ? accept its extra cost ?
What models and software tools to develop ?
How to evaluate the gain of using quality information ?
How to choose the right quality level of the information for the right
application ?
The range of approaches to answer these questions is still widely open from
statistics to artificial intelligence. The goal of this meeting is to help us
to better organise this landscape.
Workshop DELIVERY:
a book has been published:
"Data Quality in GI",
Goodchild & Jeansoulin eds., by Hermes Publisher, Paris ;
a selection of the best papers
forms a special issue of GeoInformatica (Oct.1998), and
the abstracts are printed below.
About ``CASSINI'':
a "groupe de recherche" of the CNRS since January 1993. It is
sponsored mainly by CNRS and IGN, and by several other French public
organisation interested in GI. It is the main French research consortium on
GIS, and its projects cover most of these aspects: "Time, Quality and Dynamics
in GIS", "Spatial Analysis and Generalisation", "Communication and Co-operation
in the use of GI".
This international workshop is the first one of several
- Cassini's - devoted to study emerging problems in the field of
representation and use of GI.
The next one (next year) will focus on Groupware for Urban Planning.
The present workshop is sponsored by:
AFIGeO:
Association Française pour l'Information GÉOgraphique ;
CNIG
Conseil National de l'Information Géographique
CNRS LIM:
Laboratoire d'Informatique de Marseille ;
CNRS GdR CASSINI: Quality group (Line C of 93-95) ;
CNRS GdR "Bases de Données": "Spatial DataBases group" ;
IGN:
Direction de la Recherche.
Monday, 21 April - (IGN - Saint Mandé)
14h00- Welcome address: J-C. Lummaux CNIG, F. Salge, IGN
14h30-15h30: Session 1
- N. Chrisman:
Error in social context: or why content standards efforts are doomed to failure
- S. Guptill:
Building a geospatial data framework - Finding the 'best available' data
- B. David:
Quality equals conformity to specifications or fitness to needs
16h00-17h00: Session 2
- M. Goodchild:
Communicating the results of accuracy assessment: metadata, digital libraries, assessing fitness for use
- R. Laurini, S. Servigne, B. Tellez:
Structural maps for matching aerial photos and cadastre
- P. Burrough:
Using information on uncertainty to improve the modelling of runoff, erosion and deposition with Digital Elevation Models
Tuesday, 22 April (Hotel IBIS, La Villette, Paris)
09h00-10h20: Session 3: Reporting Quality
- S. McMaster:
Using GIS-based sensitivity analysis to undertake data quality and model assessment
- S. Faiz, K. Abbassi, P. Boursier:
Applying data mining techniques to generate quality information within geographical data bases.
- F. Vauglin:
Statistical modelling of positional accuracy for linear features
- F. Harvey: Quality needs more than Standards
10h40-12h00: Session 4: AI Approaches
- M. Worboys:
An algebraic approach to spatial uncertainty
- R. Jeansoulin:
Using spatial constraints as redundancy information to improve quality
- P. Fisher:
Economic improvement of quality information in digital elevation data
- N. Chrisman:
Transformations of Geographic Information as a Guide to Error
14h00-16h00: Session 5: Uncertainty Management
- J. Wood:
Activating Scale-based Elevation Uncertainty from Digital Elevation Models
- D. Cazemier, P. Lagacherie, R. Martin Clouaire:
Use of fuzzy and uncertain data extracted from a regional pedological map
- M. Molenaar:
Identification of fuzzy objects through remote sensing and photo interpretation
- M. Azouzi:
Data Quality and Decision Support System
- A. Frank:
Hierarchical spatial reasoning and data quality estimates
Wednesday, 23 April (Hotel IBIS, La Villette, Paris)
9h00-9h40: Session 6: User Session
- A. Arnaud: Is there still a need to improve Address quality?
- A. Hoekstra: Quality Issues in Cadastral Map Renovation
10h00-11h00: Session 7: Quality Control
- G. Edwards: The propagation of boundary errors from maps to models
- D. Lanter: Priority-based Quality Assurance for Consumer GI products.
- R. Laurini, S. Servigne, T. Ubeda: A Methodology for Spatial Inconsistency Checking and Correcting
G.I. Data Quality - ABSTRACTS - Paris, 21-23 April 1997
Nicholas R. Chrisman <chrisman@u.washington.edu>
Department of Geography, Box 353550 -
University of Washington - Seattle WA 98195-3550 USA
Research about error in cartography and GIS often makes the implicit
assumption that some 'terrain nominale' exists as an objective, external
truth. This assumption is critical to making progress on mathematical tools,
but it is important to examine the degree to which people and organizations
can agree about the underlying models. Instead of a universal and unitary
'reality', a geographic information system is a social construction based on
agreement and negotiation. Error research should be aware of the radical
differences in perspective (and hence in the conceptualization of a 'terrain
nominale').
This presentation will demonstrate how an approach to the
study of scientific knowledge (a theory of standardized packages and boundary
objects) explains the divergences between different groups in defining the
content of a geographic database. As a case study, this presentation will
focus on the definition of 'wetlands' and related classes of land cover/land
use. Different disciplines establish different systems of meaning (cultures)
through which the world is measured and represented. These ideas control the
content of a GIS far more than the ideas behind the software developers.
The social processes of creating a GIS require substantial negotiation, and
the topic of this negotiation can be predicted by a theory of geographic
measurement, better than the classical theory of cartographic representation
or a social theory of cognitive communication.
Standards efforts for the
content of a GIS have been mostly motivated by the illusion that one data
source can be applied to all potential users. The evidence from the FGDC
Wetlands Subcommittee report regarding Wicomico County shows something quite
different. Geographic categories, such as wetlands, are constructed in
context. When treated at face value, comparisons show utter incoherence.
Another understanding of the social construction of geographic information
seems to be the logical result. With it comes a humility that standards may
serve a much more limited role in the GIS community.
Stephen C. Guptill <sguptill@usgs.gov>
U.S. Geological Survey - 519 National Center - Reston, VA 20192 USA
In traditional cartography, map makers decide if the quality of the available
data is sufficient to produce a multi-purpose map product. At best, the
quality information available to the user of an analog produced map includes:
(i) a quality statement furnished by the producing organization,
(ii) the reputation of the producer, and
(iii) any experiences resulting from using the product.
Cartographers have rarely dealt with the data quality issues of nonpositional
data. By the time positional accuracy has been obtained, the other attributes
of the features have been collected with accuracies that also satisfy the
cartographer.
In the digital environment users select the digital data sets that they
determine are necessary for the production of a desired product. The
producer makes the basic decisions concerning data quality and provides
metadata to the user. Both then decide whether or not the data are useful
for their specific purpose. The concept of a geospatial data framework is
based on the premise that digital spatial data collected by various
producers can be integrated into a larger (national) coverage.
When there are multiple data producers, each can add attributes or
relationships to features already in the database or add new features.
Thus, any data file may be the product of a number of producers.
In an environment of collaborative data production, there is the potential
to satisfy today's requirements for current, accurate geospatial data.
However the key to building a geospatial data framework is to use the
"best available" data. On what basis is this selection made?
Finding the "best available" data means having a standard set of data
quality elements and metrics.
There is general agreement on the categories of information that a spatial
data quality standard should require. These categories include:
Lineage: a description of the source material from which the data
were derived and the methods of derivation, including all transformations
involved in producing the final digital files.
Positional Accuracy: measures of the horizontal and vertical accuracy
of the features in the data. Descriptions of positional accuracy shall
consider the quality of the final product after all transformations mentioned
in the lineage portion.
Attribute Accuracy: the accuracy of scalar or nominal values associated
with features or relationships contained in the data set. Evaluation of scalar
attributes shall be performed using procedures similar to those used for
positional accuracy. Effective measures describing the accuracy of nominal
attributes need to be developed.
Completeness: the degree to which all applicable features,
relationships, and attributes have been encoded in accordance with data
capture rules.
Logical Consistency: the number of features, relationships, or
attributes that have been correctly encoded in accordance with the integrity
constraints of the feature data specification.
Semantic Accuracy: the number of features, relationships, or attributes
that have been correctly encoded in accordance with the feature representation
rules.
Currentness: the temporal aspect of the data. It defines the date of
observation, type of update (creation, modification, deletion, unchanged),
and validity periods.
The research to develop measurement, rating and testing methodologies for
all of these elements remains to be undertaken. In addition, there are
complex interrelationships between these data quality elements that make
the determination of the 'best' data a very difficult task.
Dr. Benoit David <Benoit.David@ign.fr>
Institut Geographique National,
2/4 Avenue Pasteur - F-94165 SAINT-MANDE CEDEX - FRANCE
Qualite = conformite a la specification ou adequation aux besoins
La qualite est definie dans la norme ISO 8402 comme l'aptitude a satisfaire
des besoins exprimes ou implicites. Cette definition s'applique ˆ
l'information geographique.
Cependant les normes internationales en cours
d'Žlaboration (CEN, ISO) limitent la portŽe de la definition en se
restreignant a la "conformite a la specification" car c'est la seule approche
que l'on sache actuellement traiter. Cette limitation est clairement orientee
producteur et ne repond pas aux besoins des utilisateurs. L'objet de mon
expose est de proposer une approche plus globale de la qualite en information geographique et de commencer a poser les problemes pour aborder ce sujet.
Michael Goodchild <good@goodrs.geog.ucsb.edu>
University of California at Santa Barbara, CA, USA
The concept is proposed of a data life cycle extending from the initial
collection of data in the field or by remote sensing to eventual archiving.
Although GIS has been perceived to date largely as an analytic technology
confined to a small part of the data life cycle, its influence is
increasingly seen as extending throughout the cycle, in the form of newer
technologies such as field GIS and digital spatial data libraries. The
concept of a data model is introduced and examples are given of the role of
data modeling along the life cycle. Accuracy assessment and concern for
broader issues of data quality are critical if data are to be assessed as
to fitness for use by the various actors in the life cycle. GIS data models
must be extended to accommodate essential information on accuracy. The role
of accuracy assessment in metadata is discussed, and a new view of metadata
is proposed with extended functions and a hierarchical structure.
Bruno Tellez, Sylvie Servigne, Robert Laurini <laurini@if.insa-lyon.fr>
LISI - INSA Lyon
Maintaining quality in geographical database is a crucial problem.
Urban planning could not be done if quality about objects is not ensured.
Especially, cadastral maps emphasise this situation because after detecting
changes, updates are generally time-consuming and expensive. So, this paper
presents how aerial photos can be used to process updates in cadastral data
by using structural maps.
If raster-based data (photo) seems to be unusable as they stand, vector data
(cadastre) are semantically richer and easier to compare in order to update
objects. If a description by vector data could be easy for cadastre, it leads
to huge difficulties to transform raster data into vectors. In the past, most
of numerical cadastre only stored entities with their geometry and
alphanumerical values (owner, area...). Locally, the information is
accessible to describe the object but nothing concerns relative positions
between objects.
A methodology was designed to manipulate the two types
of data in order to represent and store them in a unique topological model.
The methodology is based on Delaunay triangulation in order to define plane
subdivision and to deal with empty space. Empty space between two objects
suggests a kind of proximity.
Typically, in aerial photos, objects are extracted with topological
objectives. After image processing, features are extracted to propose a base
for triangulation. Finally, triangles are combined to obtain high level
objects (with topological and colour considerations). The pictorial object
representation model integrates both geometry and connections to the exterior
which define a hierarchical structure. For photo and cadastre, objects and
extended relations define a structural map (based on hierarchical structure)
that will be used to process comparison and detect modifications.
Directions (at Frank's sense), objects dispositions so as temporal intervals
(Allen) and topological relations (Egenhofer) are used to increase semantics
and information on structural map.
Peter A. Burrough <P.Burrough@frw.ruu.nl>
Utrecht University, NL.
The derivation of flow networks from altitude matrices (raster digital
elevation models) is now standard practice for surface water modelling
and catchment analyses carried out with GIS. Different algorithms for
computing slope and flow linkages may lead to substantially different
nets but as most procedures are deterministic the variety of results is
due to method and not data; further, most GIS procedures assume data to
be exact. This presentation demonstrates how uncertainties in DEMs
propagate through to the derived drainage nets. It describes how errors
can be modelled and added to the DEM to obtain an enhanced and enriched
understanding of surface transport that explains some of the problems
encountered by persons attempting to validate runoff models. Finally a
presentation will be given of how simulation of certain geomorphological
processes is only made possible by the addition of spatial errors to the
model algorithms.
Susanna A. McMaster <mcmaster@macalester.edu>
Department of Geography, Macalester College, St. Paul, Minnesota,
55105-1899, USA.
GIS-based sensitivity analysis has been used as a methodology for assessing
the sensitivity of model results in response to variations in specific inputs.
For example, variations can be imposed to study the effects of measurement
error, resolution changes or model assumptions, e.g., different weighting
schemes associated with suitability analysis. This approach can be useful
in gauging the reliability of data or model assumptions in terms of the
impact on model results that support different types of decision-making.
It can help to identify, for example, the appropriate level of data quality
or type of model specification necessary for a particular application.
This paper discusses the use of sensitivity analysis to examine variations
in attribute error and resolution on a forest management model designed to
assess the suitability of an area for pulpwood production. It assesses the
impact of these variations in terms of the impact on management decisions
such as which areas to cut for timber harvest. Opportunities for employing
GIS-based sensitivity analysis to explore the effects of other error sources
are also discussed. Several recommendations for improving GIS reliability and
decision-making capabilities are proposed including the conceptual design of
a visual error assessment interface.
Keywords: sensitivity analysis, data quality, model specification
Sami Faiz, Karim Abbassi, Patrice Boursier <Sami.Faiz@lri.fr>
University of Paris-Sud / Computer Science Research Lab. LRI -
Bat. 490 / 91405 Orsay - FRANCE
The work that we describe aims at reducing the cost of generating quality
information associated to geographical data. We do not propose another way
to generate quality information, but a means to locate areas that need,
more than the others to be certified or "qualified" with terrain reference.
We present a set of knowledge extraction methods, and their application to
geographical databases.
We also propose an approach for applying these
methods for the generation of quality information in geographical information
systems.
Francois Vauglin <Francois.Vauglin@ign.fr>
Institut Geographique National, Saint Mande, France
This will be an excerpt of the PhD dissertation that Francois presented on
Monday 21, morning.
Francis Harvey <Francis.Harvey@dgr.epfl.ch>
EPFL-IGEO-SIRS, Lausanne, Suisse
Standards are good. Quality is better.
The myriad number of standardization efforts belie the importance of quality
for the "suitable" use of geographic information. Standards help make
producers and users of geographic information aware of quality concerns, but
without enforcement mechanisms or incentives, quality remains in many cases a moot topic. Furthermore, the complexity and number of standards in circulation dilute issues of quality. In the sense of the "fitness for use" paradigm, standards should be seen in light of their institutional and disciplinary setting. What can be summarized as the standard's culture has strong impacts on its reception and adoption. Standards can readily become contested vehicles for the advancement of certain institutional views. In awareness of these issues, I present work on algorithms that follow a pragmatic approach to including quality in geographic information processing. Instead of separating quality measures and indicators from the data as metadata, such data should be incorporated in the basic geographic information data structures. In contrast to often cumbersome metadata, this information could be readily exchanged and utilized. Incorporating data regarding quality at the lowest level of data structures can directly aid geographic information processing.
Prof Michael F Worboys <michael@cs.keele.ac.uk>
Department of Computer Science
Keele University, Keele, Staffordshire ST5 5BG, UK
The usual current approach to reasoning about geographic spaces with
uncertainty is fuzzy logic. This talk proposes an approach based on
generalizations of Boolean algebra. We will consider the problem of
reasoning about regions in the plane with indeterminate boundaries.
Our approach makes connections between mereology and category theory.
The relevant algebraic structures are Heyting and co-Heyting algebras.
The talk will provide the formal foundations, and explain the relevance
of the formalism to reasoning about regions with indeterminate boundaries.
Robert Jeansoulin <Robert.Jeansoulin@lim.univ-mrs.fr>
CNRS LIM and GdR CASSINI
Let's consider merely the quality as the compliance of data with user needs.
In numeric approaches (statistics, RMSE ...) it's a matter of more or less.
In symbolic approaches (constraints ...) it's a matter of satisfiability:
is there a model for such a theory ? Often, inconsistency (who yealds
insatisfiability) arises when giving a new fact (axiom) contradicts previous
ones (facts or derived facts).
We will focus on a particular aspect of GI quality: the compliance between
coordinates and topology. Ideally, given contour coordinates, topologic
relations are deductively computed. The general "topologic model" based
on node-arc-face relations is in the core of most of present GIS. Hence,
giving coordinates (translated as axioms) implicitely gives topology
(as theorems).
Topologic constraints are useful to determine spatial object when their
coordinates are unknown, but we can also choose to explicitely give
topologic relation between objects, even with known coordinates. This
redundancy is useless as far as such knowledge (given as a new axiom)
was already a theorem. But if coordinates positional accuracy is
unsufficient, bad theorems may have been derived.
The inconsistency will be partially removed whenever invoking specific
subset of axioms. For example, give priority to coordinates to display
data, and give priority to topology to compute constraints.
Through this example, non-monotony, logical revision and para-consistency
will be illustrated.
Pr. Peter Fisher <pff1@leicester.ac.uk>
Department of Geography, University of Leicester, LE1 7RH, UK
Traditionally the quality of elevation data is reported by a single
parameter, the Root Mean Squared Error (RMSE). It has been pointed out
that this parameter of quality fails to incorporate a number of aspects
of the actual quality. The most important failures are that the mean error
may not be zero but may be either positive or negative, and that the spatial
distribution must be structured (high values must neighbour other high
values).
Unfortunately, any attempt to propagate the error in a Digital Elevation
Model (DEM) is dependent on an estimate of the amount of spatial structure.
This will be demonstrated with respect to the determination of the visible
area where use of RMSE without spatial structure yields very large
divergences in the area visible from a point and very small areas of high
probability, but if assuptions are made about the spatial structure, the
divergence reduces and the area of high probability increases. A simple step
is advocated as a means to address the short comings. Using spot heights
from plublished maps, the spatial structure in the error in the elevation
model can be estimated, as can local measures of bias and RMSE. This is
combined with stochastic simulation to provide an improved error modelling
stategy. The approach to modelling error in elevation models is achieved,
by abandoning the passive reporting of the RMSE as a measure of quality,
and adopting an analysis of spot heights wich were surveyed at a greater
resolution and accuracy in independent surveys. To be convenient this
approach requires the data provider to supply information, beyond that
currently provided. On the other hand, the information exploited here exists,
and oes not require further survey.
From this empirical approach, there are general lessons to be learnt, wich are valid for most if not all measures of quality used in spatial information :
¡ like all spatial information, quality has spatial structure,
¡ traditional measures of quality are passive, and not intended for direct use,
¡ activating these measures will usually require recourse to more detailed information,
¡ that information may already exist, and not require extra cost,
¡ the products of analyses using revised error measures may provide more robust and believable error estimates wich are not possible with traditional quality parameters.
Nicholas R. Chrisman
Department of Geography, Box 353550 -
University of Washington - Seattle WA 98195-3550 USA
Analytical cartography developed in the era of Chomsky's transformational
grammars. Tobler (1976) set out a systems of transformations based largely
upon the geometric component of geographic information. This approach
informed the 3X3 (point, line,area) or 4X4 (with the addition of volume)
matrix in Clarke (1995, Figure 11.1, page 184) and Unwin, among others.
This schema for transformations hinges on the geometric form of the
representation, perhaps quite appropriate for analytical cartography
purposes.
The GIS literature has a series of alternative schemes used
to present the different kinds of operations. Perhaps the most developed
is Tomlin's (1990) Map Algebra. Goodchild (1987) and Burrough (1992) have
contributed extensions to Tomlin's basic scheme. This paper will describe
some limitations in the formulations presented in the literature.
One limitation of the current literature involves conversions between
different forms of information. There is too much inherited from the
cartographic representation, not based on the measurement framework inherent
in a data source. This paper presents a scheme for transformations with a
number of novel elements. First, it separates the representation
(symbolization) from the underlying measurement. Then it presents a scheme
of measurement frameworks that cover the bulk of geographic databases. In
place of the raster/vector divide, this scheme includes a dozen data models
that should not be confused with each other. These measurement frameworks
should be the basis for understanding transformations in GIS.
While a 3X3 or 4X4 matrix can be quickly comprehended, a 12X12 (or larger)
is difficult to describe or communicate. This paper reviews a unifying scheme
for transformations based on two elements: a geometric neighborhood plus a
rule to combine or process attributes (Chrisman, 1997). The rules fall into
three classes (dominance, contributory, and interaction) based on the
treatment of multiple attribute values. Viewed in this way, the operations
of a GIS (including map overlay analysis, neighborhood operations, plus the
items now treated as transformations) can all be relocated as various kinds
of transformations.
This new scheme for transformations has ramifications the study of error in
GIS. The error to be expected in an element of 'control' should be different
from the error in a component that is free to vary. Many products are
overconstrained due to the cumulative effects of a series of indirect
measurements. This scheme helps explain differences that have been known,
but not well accomodated in the previous frameworks.
Burrough, P.A. 1992: Development of intelligent geographical information systems. IJGIS 6(1): 1-11
Chrisman, N.R. 1997: Exploring Geographic Information Systems. New York: John Wiley.
Goodchild, M.F. 1987: A spatial analytical perspective on geographical information systems. IJGIS 1(4): 327-334
Tobler, W. 1976: Analytical cartography. The American Cartographer 3: 21-31.
Tobler, W. 1979a: Cellular geography. In Philosophy in Geography, eds. S. Gale and G. Olsson, p. 379-386. Dordrecht NL: Reidel.
Tobler, W. 1979b: A transformational view of cartography. The American Cartographer 6: 101-106.
Tomlin, C.D. 1983: Digital Cartographic Modeling Techniques in Environmental Planning. unpublished Ph.D., Yale University.
Tomlin, C.D. 1990: Geographic Information Systems and Cartographic Modeling. Englewood Cliffs NJ: Prentice Hall.
Jo D. Wood <jwo@leicester.ac.uk>
Department of Geography, University of Leicester, LE1 7RH, UK
Surface properties commonly extracted from DEMs within GIS may well vary
with scale. This introduces a degree of uncertainty into the extraction of
such properties, that is normally ignored. Selecting the 'correct' scale at
which to perform analysis is in part a function of the application involved,
and part a property of the surface itself. As far as possible, the scale of
analysis should not be determined by the fixed scale implied by raster
surface models. Using software written in Java, several methods are introduced
that allow surface properties to be measured over a range of scales. This
allows scale dependency, and hence an element of uncertainty, to be visualised
and quantified. Results indicate that it is possible to identify 'critical
areas' whose surface properties vary greatly with scale of analysis.
D.R. Cazemier*, P. Lagacherie* & R. Martin-Clouaire** <lagache@ensam.inra.fr>
*INRA Laboratoire de Science du Sol,
2 Place Viala, 34060 Montpellier Cedex 1
**INRA Station de Biometrie et Intelligence Artificielle,
B.P. 27 31326 Castanet-Tolosan Cedex
Information about soil is required in different fields of environmental
research and land management concerning regions or countries as a whole.
In most situations, small scale soil maps (e.g. 1:250.000), are the main
repository of soil information. Such sources usually provide only incomplete
information with respect to the desired soil parameters at any particular
point in the area covered. The reasons are threefold. Firstly, the soil
properties are specified as ranges of values so as to convey the spatial
variability. Secondly, the mapping units of small scale soil maps most
often correspond to soil associations which implies that the soil unit
boundaries are not known. Thirdly, the parameters of interest are only
loosely linked to the variables described in the map and data base.
Therefore it is essential to represent faithfully (as precisely as possible
but without introducing arbitrariness) the available pieces of knowledge so
as to be able to exploit at best all kinds of overlapping of information
that can be used to reduce uncertainty.
This study handles the use of soil information in crop yield prediction
models designed for agricultural studies. These models require information
on specific soil parameters, in particular, hydraulic conductivity and water
retention. We describe an approach based on a fuzzy constraint representation.
The addressed problem amounts then to specify the available information as a
set of mathematical relations (constraints) between the variables involved
and to search for these variables the set(s) of values that are compatible
with all these relations. In order to take into account the high imprecision
that pervades the available knowledge we use fuzzy relations and,
consequently, each set of values returned as a possible solution is qualified
by a degree expressing to what extend any considered set of value is
consistent with the knowledge described by the constraints.
Martien Molenaar <molenaar@itc.nl>
Waggeningen Agricultural University and ITC, NL.
Spatial objects have a geometric and a thematic description. Both can be
uncertain in the sense that there is no full confidence that they are
correct. Traditionally the geometric uncertainty has ben expressed by
variances of the coordinates and measures like epsilon-bands expressing
the possible discrepancies between the real object boundary and its
digitised representation. The thematic uncertainty can be expressed by
variance of the attribute values of the object or by fuzzy measures. The
thematic uncertainty of objects will often have a spatial effect in the
sense that it is difficult to decide where the object boundary is. This
effect will be called the extensional uncertainty of the object. The
geometric uncertainty mentioned before indicates then how accurate the
shape of a boundary and the position of boundary points can be measured,
once a decision has been made where the boundary is. The concept of the
extensional uncertainty of spatial objects will be explained in detail,
some examples related to remote sensing and photo interpretation will be
given to illustrate it.
Keywords: fuzzy objects, extensional uncertainty, topology, data model, syntax.
Mounir Azouzi <Mounir.Azouzi@dgr.epfl.ch>
EPFL Institute of Geomatics/Geodetic Engineering Lab.
IGEO/TOPO - Lausanne Switzerland
The (Swiss) noise propagation model is used in a GIS to produce the trafic
noise map. And the noise is one of the creteria used in a muticriteria
method combined with a GIS for land suitability assessment. The purpose of
this paper is to study the data error propagation in the model above, and
to evaluate the influence of data quality in a decision support system.
Andrew U. Frank <frank@geoinfo.tuwien.ac.at>
GeoInformation, Technical University of Wien, Austria.
Data quality descriptions are crucial, but methods to produce and use them
have not significantly improved during the past 10 years.
Current quality descriptions are from the perspective of the producer of the
data, not the user. Actual quality descriptions are mostly verbal and not
suitable for rapid comparison with a required standard to make a decision
about Ôfitness for useÕ of a certain dataset for a task.
This limits business with geographic data over the net.
The paper introduces the concept of a metamodel as a framework to compare
data quality from a producer and a user perspective in a single model.
It is based on category theory and morphisms, which link the model of reality
with the model of the GIS data, and their collection and use.
The achieved quality of a decision based on using the data can be derived.
It is shown that data quality descriptions are dependent on the intended use
of the data. A Ôuse independentÕ, generic data quality description is
not possible. Fortunately, a large set of GIS functions demand the same data
quality description, therefore not every potential use requires a different
data quality description of a data set.
Ant—nio Morais Arnaud <ama@di.fct.unl.pt>
FCT / New University of Lisbon, Portugal
The current trends are towards the generalized use of coordinate based referencing, to locate and identify fixed objects, of smart card plus public key code to identify individuals and of GPS to locate moving objects. Automatic point location and path determination with GPS is expected to spread rapidly but the traditional locator "Postal Address" must prevail and still needs to be improved in its quality, as of recent evaluations and arguments.
A number of current GI applications, using the "address" as locator, including emergency dispatch, epidemics studies, transportation planning, ..., bring global savings (in human lifes, petrol, time,..); however the benefits from these applications depend strongly on address quality. These expected savings justify the investment in urgent political measures to improve address quality, suported in current technology and hopefully partially with public funding.
By other side, address with quality is still needed for traditional business and shopping as it is also becoming a Geomarketing infrastructure to prospect, Direct Mailing, to deliver goods by Post or by private fleets. Large corporate Databases are now becoming spatial, allowing spatial analysis for Market Analysis and Decision Making; it is through the address elements in these databases that the G is added to them using middleware.
This new pressure over the address, as it is, in so many different organizations, risk to create even more mess if address normalisation and official street gazeteers are not promoted. The Portuguese Post Office, CTT, is promoting a new Post Code and contributing to the diffusion of a normalized address through a series of initiatives.
A set of measures to improve the quality of address is expected to be promoted by the Government and CTT, with short term results, supported in an Internet site, allowing spatial queries, to validate and pin-point addresses.
Martin Salzmann, Auke Hoekstra, Ted Schut <K.VanDerHoek@ap.kadaster.nl.net>
Cadastre and Public Registers Agency of the Netherlands, Apeldoorn, NL
Map renovation is at the moment an important isue for the Cadastre and Public Registers Agency (Cadastre) of the Netherlands. The cadastral map is available in digital form and the revision and use of the map require that its quality is guaranteed. Revision of the map is basically an internal process.
It goes without saying that the map revision process benefits of maps that meet specifications. In many cases our customers conbine the information of the cadastral map with other geographic information (e.g. topographic base maps).
Inconsistencies between maps are directly noted. Therefore the Cadastre of the Netherlands is undertaking a major effort to bring its maps up to national standards. In this contribution we will discuss the approach followed in the Netherlands. We will focus on the aspect of quality and in particular on geometric quality (positional accuracy).
The cadastral map of the Netherlands is improved by reconciling it with the large-scale base map of the Netherlands.
We will discuss the characteristics of both maps, before we discuss the reconciliation process itself. Quality improvement of the cadastral map is a major objective of these activities and therefore we will pay attention to the description and specification of quality. For this we use the concepts of the recently completed manual of the technical activities of the Cadastre ((Polman and Salzmann, 1996); in Dutch known as the Ç HTW È (Handleiding voor de Technische Werkzaamheden)). This manual provides a unified approach to quality assurance, including a model for the description of geometric quality. Besides it contains a separate chapter on map renovation. We will discuss the quality model as it is used in the Netherlands and we will consider more closely the procedures as sketched in the HTW. Finally we describe the requirements of a procedure for a successful map renovation process.
Geoffrey Edwards <Geoffrey.Edwards@scg.ulaval.ca>
Universite Laval, Quebec, Canada
Having developed a model for managing boundary errors from multiple
interpretations of the same scene, and another model for characterising
boundary errors in classified remotely sensed scences, we have begun to
address the issues of combining these models with fuzzy classes and other
representations of uncertainty information within the context of inputting
maps to models. We have developped an integrated framework for combining
these maps in order to generate a single uncertainty/error surface to
accompany the mean surface produced as a result of a chain of operations.
We have performed extensive sensitivity analyses on both the mean surface
and the range of uncertain surfaces which can be associated with the mean,
within the context of a hydrogeological model used to characterise the
vulnerability of the water table to pollution.
David Lanter <dlanter@microsoft.com>
Microsoft, Redmond, WA, USA
Microsoft integrates multi-scale, heterogeneous, data sources to create
consumer geographic information products. Prior to inclusion within our
products, these data are prioritized for examination. Thematic and spatial
prioritization assures that quality assurance efforts are focused efficiently.
To this end, research is focused on creating abstract views that represent
the "Terrain Nominal" for our markets.
Alain Puricelli, Sylvie Servigne, Thierry Ubeda, Robert Laurini <laurini@if.insa-lyon.fr>
LISI - INSA Lyon, France
Quality is essential regarding data processing. The aim of quality control
is to ensure result reliability of reasoning, queries, and so on. Geographic
data sets bring specific quality aspects related to the spatial attributes of
data. Regardless of the data source (map digitization, aerial photos, GPS
data, etc.), resulting geographic data sets must be consistent in order to
be used in spatial analysis, such as in urban planning applications.
This paper addresses the problem of spatial consistency checking in geographic
databases in vector format. A general methodology of consistency checking is
proposed. It is based on error definition, detection and correction processes.
Two kinds of errors are distinguished, geometric errors which can occur in
the shape representation of objects and semantic errors which are defined
according to the meaning of objects (a polygon representing a lake).
The main issue of the geometric errors detection is a shape admissibility
process. It allows to detect inconsistency in shape representation of objects
(for example an unclosed polygon). In our case, semantic errors detection
relies on topological integrity constraints. They permit to find inconsistent
topological scene in the data sets (for example a road inside a lake). Those
error detection processes lead to determine a set of errors to be corrected.
The correction can be manual by means of manipulation tools or semi-automatic
by means of correcting scenarios suggestion. Scenarios are computed by
applying elementary transformations to an object (such as object displacement).
A visual interface has been designed for each kind of correction processes.
Proposed corrections ensure not to create a new error.
Application of this methodology to the Lyon Urban Community geographical
database is currently conducted. First results are presented.