uncertainty in both Schemes for each of the stages
in Figure 5. Note how there is an assessment for
all polygons regardless of the Boolean outcome in
Figure 5. Thus in Scheme A, doubt can be cast on
the acceptance of unit 6 and in Scheme B doubt can
be cast on both the acceptance of units 1-7 and the
rejection of units 8 and 9 in the traditional
binary form of the analysis.
The software used in this small example is Genamap
which provides lineage tables through the use and
recording of active IDs. Table 4 shows the analysis
lineage. This type of record would permit
sensitivity analysis to quickly recompute the «E
Scores for changing levels of initial uncertainty
and hence identify what decrease in uncertainty is
required to improve the end result. Thus for unit 6
in Scheme A, the observer should collect additional
data on category B in Layer 1, only in the area of
unit 6, so that it can be improved to "moderately
certain' or better from the observer's point of
view. For Scheme B, category W in Layer 2 must also
be resurveyed so that it is mapped in as
"reasonably certain? or better.
IMPLICATIONS FOR GEO-INFORMATION THEORY
The model has a number of implications for geo-
information theory. There is a distinct need to
improve the ability of GISs to record natural
variation and uncertainty in data, not at the
generalised level of the coverage but at the level
of the individual object or entity. Precise, crisp
boundaries and totally homogeneous areal units are
in fact a special case in our mosaic of natural
environment. Why are we using a tool more suited to
these special cases for all our spatial analysis?
The point, line and polygon may be inappropraite
primitives to record imprecisely bounded objects.
To have proactive data, observers will need to
exercise greater responsibility in recording levels
of uncertainty and will have to develop strategies
in order to do so. Yet the user must take
responsibility for assessing fitness-for-use since
it is the user who is most familiar with the
prevailing context. Sensitivity analysis of data
should become a standard routine. Data structures
and database design will need to be modified in
order to faciltate this. There is the potential for
involving expert systems. Current relationships
between user and data will undoubtedly change.
Probably the greatest impact of the model is likely
to be in the area of research. Ways of combining
uncertainty measures will have to be sought. New
measures may be developed that can be better
embedded within the data. There are a number of
candidate metrics that could be used for
propagating uncertainty. These will need to be
investigated, as will the mapping from source
uncertainty measures. The propagation will need to
be modelled for a wide range of data
transformations. The mapping out process and
visualization of uncertainty distributions is
already a major concern (Beard et al. 1991). Most
of all, the model provides a framework around which
whole or partitial solutions to the problem of
uncertainty in GIS can be built and tested.
ACKNOWLEDGEMENTS
This research has been supported by the Hong Kong
Polytechnic.
Genamap is a registered trademark of GENASYS TI
Inc. Fort Collins, Colorado.
REFERENCES
American Society of Civil Engineers, 1983. Map
Uses, Scales and Accuracies for Engineering and
Associated Purposes. ASCE, New York, USA.
American Society of Photogrammetry and Remote
Sensing, 1985. Accuracy specifications for large
Scale maps. Photogrammetric Engineering and
Remote Sensing 51:195-199
Beard, K., 1989. Use error: the neglected error
component. Proceedings AutoCarto 9, Maryland
:808-817
Beard, K.; Buttenfield, B.P. & Clapham, S.B., 1991.
Visualization of spatial data quality. NCGIA
Technical Paper 91-26, NCGIA, Santa Barbara, USA.
Bedard, Y., 1987. Uncertainties in land information
systems. Proceedings AutoCarto 8, Maryland :175-
184
Bethel, J.S. & Mikhail, E.M., 1983. On-line quality
assessment in DTM. ASPRS/ACSM Annual Convention
:576-584
Blakemore, M., 1983. Generalisation and error in
spatial data bases. Cartographica 21 :181-139
Bureau of Budget, 1947. National Map Accuracy
Standards, US Government Printing Office,
Washington D.C., USA.
Burrough, P.A., 1986. Principles of Geographical
ID Entries
Selection/Criteria
Map/ID Selection Table Criteria
1 12 Liz ATTRIBUTE | TAGV1 | BOUNDED A
5 12 LiL2 ATTRIBUTE TAGV2 BOUNDED W
10 25 LiL2 ATTRIBUTE | TAGV1 | BOUNDED $C
11 25 L1L2 ATTRIBUTE | TAGV2 | BOUNDED #Y
12 19 1, 65 |" BETS - OR
13 18 10, 12 ] SETS - AND
14 17 11, 13 | SETS - AND
Table 4: Simplified extract of the Genamap ACTIVE ID list showing
lineage of analysis depicted in Figure 5.
766
L