ed in 
ynomial 
riables 
th, and 
91]. In 
into 5 
by the 
and DE 
candard 
ons is 
imation 
(e.g.) 
apacity 
y has 
bility 
1 (see 
'ainage 
learing 
Supply 
sp Set 
bility 
to in 
dicted 
; these 
verall 
dicted 
1(0.98, 
d Soil 
search 
ied by 
tables 
levant 
f the 
razing 
loured 
Then 
3 and 
S were 
y map 
ii) a 
d good 
ty map 
jJ. The 
QUALITY OVERLAY 
  
  
LEGEND 
Probability (x? 
SE Less than 10% 
S 10 - 30 
S 30 - 40 
&j 40 - 50 
   
Ej 50 - 60 
JV 
se 
Figure’ 2 - Probability overlay 
suitability 
The multicoloured grazing suitability map and the 
3-class probability map were then combined to give 
a map which showed, in colour (reds through yellow 
to green) the grazing suitability of the 
landparcels and the quality of these predictions 
(based on probabilities) as a grey stipple 
overlays (light-grey through mid-grey to 
dark-grey). 
(See the reference [VAN ELZAKKER, RAMLAL, 
DRUMMOND, 1992] in these Archives (Commission IV) 
for a further discussion of this project and the 
visualisation of the data and information 
quality.) 
5. CONCLUSIONS 
In section 2 some proposals for components of an 
uncertainty subsystem were presented. These were 
that: 
Positional and attribute quality parameters be 
directly linked to the database descriptions of 
individual real world entities, but that Lineage, 
Completeness, and Logical Consistency reports be 
linked to sets of the database descriptors of real 
world entities. Positional and attribute quality 
parameters were successfully stored in attribute 
database tables which contained at least one 
record (tuple) for each real world entity. 
Lineage, Completeness and Logical Consistency 
Reports were created, but software has not yet 
been developed to access these. Such reports could 
be used to update positional and attribute quality 
parameters in database tables, when necessary. 
As most GISs require the user to insert the 
processing model, the associated dialogue should 
ask the user about model quality. Alternatively, 
for wellknown or frequently used processing 
FOR GRASSLAND SUITABILITY 
for grazing 
361 
   
models, a Model Quality Report could be stored in 
a GIS. 
Error Propagation uses either variance propagation 
or set theory. Both should be supported by a GIS, 
although we only developed the latter so far. 
Considering the uncertainty surrounding the 
uncertainty (!) of many GIS variables and 
processing models, Fuzzy (Sub-) Set Thoery is 
probably more appropriate than Crisp Set Theory 
when processing Logical Models. 
Finally we propose that a user should be able to 
‘toggle’ between a visualisation of the 
information requested and a visualisation of the 
quality of that information. So far in this 
project we have implemented one approach to 
visualisation of information quality. More will 
follow! 
6. ACKNOWLEDGEMENTS 
A paper similar to this was presented at the EGIS 
/92 Third European Conference and Exhibition on 
Geographical Information Systems, Munich, Germany 
[DRUMMOND and RAMLAL, 1992]. 
We acknowledge the support of the Ministry of 
Development Cooperation of the Dutch Government 
which provided the finance permitting this 
research; Mr J. ten Cate and Mr. F Brouwer of the 
Dutch Soil Research Institute for their 
cooperation; the Landinrichtingsdienst of the 
Dutch Ministry of Agriculture, Land Management, 
and Fishery for permission to use their soils data 
and topographic data; and Mr C. van Elzakker of 
ITC for useful discussions.