1-1] 
2. 4 
3. . 
4. | 
5.59 
A GIS UNCERTAINTY SUBSYSTEM 
The 
rea: 
SET: 
Bheshem Ramlal "li 
University of the West Indies, St. Augustine, Trinidad and Tobago gat. 
ow 
Jane E. Drummond have 
International Institute for Aerospace Survey and Earth Sciences (ITC), The Netherlands det: 
rea 
but 
PURPOSE: Cony 
e.g. 
Although variance propagation is well established in photogrammetry, this and other error propagation 
theory has not been transferred to GIS to exist as a standard analytical tool alongside such as Overlay 
Analysis, Buffer Analysis, and Network Analysis. This paper describes a prototype Uncertainty Subsystem 
implemented in ILWIS - a PC based GIS, and designed to provide an error propagation facility. The 
subsystem has been tested on a Dutch Land Reallocation problem which combines soils and topographic 
information. The procedures used to determine and record the quality of the processed data; the error 
propagation techniques which process the quality data through the models which generate the new Land 
Reallocation information; and the applied visualisation techniques - all used in the Uncertainty 
Subsystem, are described in the paper. 
KEYWORDS: Gis processing, Land reallocation, Land consolidation, Data quality, Information quality, 
Error propagation 
1. INTRODUCTION 2. PROPOSALS FOR SOME UNCERTAINTY 
SUBSYSTEM COMPONENTS 
For at least two hundred years, since surveyors 
began to exploit Error Theory while establishing A simple definition of GIS which contributes to 
survey control, map makers have been actively this discussion on data and information quality 
concerned with the quality of their data. But only is: 
recently has data quality within GIS become a "hot 
topic" - as demonstrated by the 1989 publication A Geographic Information System processes spatial 
of Goodchild and Gopal’s "Accuracy of Spatial data through models to provide information. in a 
Databases", NCGIA support for comprehensive computer managed environment. 
reviews of data quality  [VEREGIN, 1989] and its 
visualisation [BEARD, BUTTENFIELD and Spatial data are facts about real world entities 
CLAPHAM, 1991], re-evaluations of Openshaw’s falling into two categories: 
Monte-Carlo simulation work of the 1970's 
[OPENSHAW, CHARLTON and CARVER, 1991], etc. This primary data: identifiers; positional data; 
recent GIS-centred activity seems to have been attribute data; and, 
initiated by Chrisman [CHRISMAN, 1982] and 
Blakemore  [BLAKEMORE, 1984] in the early 1980's, secondary data: temporal data, quality parameters, 
but related concerns over the quality of gridded etc. i.e. facts-about-facts (or, 
digital data when derived from satellite remote sometimes, meta-data) 
sensing sources (e.g. [HORD and BROONER, 1976] and 
[VAN GENDEREN and LOCK, 1977]) and Digital Terrain With identifiers unique recognition of a real 
Models (e.g.  [MAKAROVIC, 1978]) were being world entity is enabled - if explicitly stated. In 
expressed in the 1970's, and have generated a some GISs identifiers are merely implied (e.g. by in 
literature which remains applicable when position). Positional data are represented by of 
considering data quality in today’s GISs. continuous variables. Attribute data may also be "log 
represented by continuous variables or tas} 
Classifications of error now exist [VEREGIN, alternatively discontinuous variables. Temporal owne 
1989], and could form the foundation for a data represent the date at which primary data were the 
standard GIS tool dealing with information originally observed or measured. meet 
uncertainty, but as this standard tool does not unic 
yet appear to exist, uncertainty is frequently Models embody the manipulative and analytical GIS 
ignored by GIS users. It is our intention, at ITC procedures which use data stored in a GIS to met; 
under the auspices of the XGIS project, to develop generate information, and can be considered to be 
and implement such a standard tool vithin ILVIS. either: 1. logical; or, 2. mathematical. Logical 2.1 
(The XGIS Project will provide Expert System based models (e.g. crop suitability rules) manipulate 
interfaces for ILWIS. ILWIS or The Integrated Land discontinuous variables. Mathematical models (e.g. 
and Watershed- management Information System is an projection change equations) manipulate continuous We 
MS-DOS based GIS developed at ITC.) The GIS tool variables (e.g. geodetic latitude and longitude) and 
dealing with information quality will be termed and constants (e.g. a,b the semi-major and [RAP 
the ‘Uncertainty Subsystem’ of ILWIS. It will semi-minor axes). Comp 
process quality information in (near) parallel data 
with the information generated for the users’ Secondary data include quality parameters. It is Qual 
applications, and provide quality information at now well established [CHRISMAN & MCGRANAGHAN, indi 
the user’s request. 1990] that in GIS there are five aspects of uses 
spatial data which have quality implications: 
356