can 
1g à 
t be 
with 
lata. 
the 
are 
lity 
data 
rele 
teil, 
r für 
zT 
'aher 
lung 
Die 
iden 
ung, 
von 
ben, 
innel 
nsed 
bing 
]uery 
the 
their 
rs to 
vhich 
ality. 
wing 
, and 
that 
even 
se to 
ional 
and 
these 
y. To 
loped 
r, aD 
' 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004 
  
representations appear like a DEM, but instead of terrain 
height the corresponding data quality attribute is displayed. 
The remainder of this paper is structured as follows: section 
two explores previous work on spatial data quality, the 
theory behind visualization methods and their combination, 
the use of visualization to convey data quality, and it 
describes existing projects on data quality visualizations. 
Section three discusses the selection of the quality 
attributes we chose in our approach. In section four we 
discuss effective visualization methods and introduce the 
visualizations that we have developed. Section six provides 
conclusions and future work. 
2. RELATED WORK 
In the last two decades data quality has become an important 
research topic. Scientists argued that users of spatial data 
should have access to data quality information 
(McGranaghan, 1993; Buttenfield and Beard, 1994; Beard, 
1997). Soon it became obvious that the nature of spatial data 
lends itself perfectly to the communication of quality 
parameters by visualization in the form of images and 
graphics. As a result, the call for visualization of data 
quality surfaced (Beard and Mackaness, 1993; van der Wel et 
al, 1994). 
Since the early nineties researchers took formal approaches 
to the visualization of spatial data quality (Clapham, 1992). 
The National Center for Geographic Information and 
Analysis devoted a lot of energy in exploring this area and 
spearheaded a research initiative on "Visualization of the 
Quality of Spatial Information" (Beard et al., 1991). Results 
from this initiative are introduced in (Buttenfield and Beard, 
1991). 
In the remaining part of our literature review we present 
various terms used to describe data quality aspects, and we 
discuss related visualization approaches and past project 
efforts. 
2.1 Discussion of Terminology 
In the literature a substantial number of expressions are used 
to describe data quality, namely quality, error, reliability, 
uncertainty, validity, accuracy, vagueness precision and 
fitness for use. 
The term quality is used as an umbrella-term that covers all 
aspects of the issue. It is used by practically everybody in 
the field (Beard, 1997; Veregin, 1999). The use of the term 
error is also widely used, and there is broad consent on what 
the word describes when used for image data, namely the 
difference between true value and the value stored in the 
database (Hunter and Goodchild, 1995; Buttenfield, 1993). 
Reliability can be defined as the level of confidence a data 
provider has that the data are correct (Evans, 1997). 
The term uncertainty is used in various ways, one being that 
the resolution of the data does not allow a user to make an 
assured decision about the content of the data. For example, 
pixels in remotely sensed images might contain uncertain 
information because of sub-pixel mixing or sensor sampling 
bias (Bastin et al., 2002). Worboys and Duckham (2004) use 
the term uncertainty to describe the doubt that users have 
about the right use of data. In this sense it is a measure that 
describes the state of mind of the user. 
Other terms that are used to describe different outlooks on 
data quality are validity (Goodchild et al., 1994), and 
accuracy (Veregin, 1999). Vagueness describes the 
impossibility to determine the exact location or boundary of 
an object in space (Duckham et al., 2001). For example ‘the 
East of Maine’ is a vague area in that its boundaries are not 
exactly determinable. Precision denotes the exactness with 
which the measurement is made that led to the entry in the 
database (MacEachren, 1992). An overall phrase that is used 
frequently is fitness for use. It indicates whether the data has 
the specifications that the users need to solve their task 
(Paradis and Beard, 1994). 
2.2 Visualizations 
Beard and Buttenfield (1999) listed the following 
challenges in the visualization of data: graphic design, 
metadata, error analysis, and user satisfaction. In this 
research we concentrate on the graphic design issues. For the 
combined display of data and data quality three possible 
forms are mentioned in the literature (MacEachren, 1994; 
Beard and Buttenfield, 1999). First, there are side-by-side 
images, where one picture shows the data and the other one 
the quality of the data. The second approach is to generate 
composite images that display data quality superimposed 
on the visualization of the data. Thirdly, sequenced images 
of data and data quality can be presented, either affording 
the user to toggle between the displays or providing an 
animation (Evans, 1997). 
The following two visualization approaches have also been 
discussed: variation in color hue and saturation to convey 
the quality of data (Schweizer and Goodchild, 1992; Howard 
and MacEachren, 1996), and, showing quality attributes as 
the z-axis in a 3D elevation model, which was mentioned as a 
worthwhile endeavour by van der Wel et al. (1994) without 
any follow-up projects implementing the idea. We take up 
the concept of the latter approach and incorporate it in our 
3D visualizations. 
2.3 Previous Projects 
The following works concentrate on the communication of 
quality of geospatial data. The R-VIS project introduces a 
model which shows the reliability of water quality data 
(Howard and MacEachren, 1996). A visualization of 
uncertainty in meteorological forecast models was also 
developed showing the discrepancy of multiple weather 
forecasts over time (Fauerbach et al., 1996). Various graphs, 
bivariate images and animations are used in the FLIERS 
project to visualize uncertainty in multi-spectral remotely 
sensed imagery (Bastin et al., 2002). Davis and Keller (1997) 
offer quality information for risk management decisions. 
Spatial data uncertainty was also communicated using 
animation (Ehlschlaeger et al., 1997). 
3. DATA QUALITY ATTRIBUTES 
Metadata contain a wealth of information about the data at 
hand. From the attributes that are typically described by 
metadata information we selected the ones that convey data 
quality, and more specifically, those which pertain to 
geospatial image quality. Our goal has been to display the 
optimum number of essential data attributes, avoiding 
redundancies which could confuse the user. We based our 
selection on the US Spatial Data Transfer Standard's (SDTS) 
section on data quality (NIST, 1992), which is quite 
1003