5 DISCUSSION 
In fuzzy mode guide assigns the membership of a single data 
point to all spatial output classes. This result can be analysed 
and displayed, and thus used for further improvements in 
classification. Transparency and easiness to use has made guide 
a successful tool for learning cognisance in image classification 
and expert system use (Gumbricht, 1996; Gumbricht and 
McCarthy, 1996). However, a good classification accuracy 
demands many iterations, and is rather tedious. 
A major problem in this application was that the geometrical 
registration was to poor. By manual inspection of training and 
ground truth data it was clear that position errors between 
images were two to three pixels, and not less than one pixel (as 
indicated by the RMSE of the geometric transformation). The 
Mazurian landscape has a very small scale topography, and 
finding points for geometric transformations is hence difficult. 
The position problems made the use of expert rules very 
uncertain in the fragmented terrain of the studied area. 
6 CONCLUSION 
Knowledge acquisition is the bottle neck of expert system 
applications (cf. Robinson and Frank, 1987). However 
compared to advanced classification methods expert 
classification can be intelligible, and used for hypothesis 
testing. Important relations between processes and patterns can 
be inferred and evaluated. A problem is that when using 
multisource and/or multitemporal images geometrical 
registration must be very accurate. 
Methods to improve knowledge acquisition include co- 
occurence matrices, discriminant analysis and Bayesian 
approaches (cf. Argialas and Harlow, 1990; Franklin and 
Peddle, 1989; Lauver and Whistler, 1993; Dymond and 
Luckman, 1994), and Fuzzy set and Dempster-Shafer theory of 
evidence (Srinisavan and Richards, 1990 and 1993). Further 
improvements in image classification, we feel, also need to 
consider contextual relationships, and we are presently 
developing and testing an expert system for such a 
classification (cf. Gumbricht et al., 1995). 
7 REFERENCES 
Argialas, D.P. and C.A. Harlow, 1990. Computational image 
interpretation models: an overview and perspective. 
Photogrammetric Engineering & Remote Sensing, 56, pp. 871- 
886. 
Civco, D.L., 1993. Artificial neural networks for land-cover 
classification and mapping. Int. J. Geographical Information 
Systems, 7, pp. 173-186. 
Congalton, R.G., K. Green and J. Teply, 1993. Mapping old 
growth forests on national forest and park lands in the Pacific 
Northwest from remotely sensed data. Photogrammetric 
Engineering & Remote Sensing, 59, pp. 529-535. 
Desmet, P.J.J. and G. Govers, 1994. Potentials of a GIS-based, 
three-dimensional USLE-approach for the identification of 
186 
critical areas at a catchment scale. Working Paper USLE.| 
Laboratory of Experimental Geomorphology, Katholieke 
Universitet Leuven, 23 pp. 
Dreyer, P., 1993. Classification of land use cover using 
optimized neural nets on SPOT data. Photogrammetric 
Engineering & Remote Sensing 59, pp. 617-621. 
Dymond, J.R. and P.G. Luckman, 1994. Direct induction of 
compact rule-based classifiers for resource mapping. Int, J 
Geographical Information Systems, 8, pp. 357-367. 
Eastman, J.R., 1993. IDRISI Version 4.1. Update manual 
Clark University, Graduate School of Geography, 209 pp. 
Fiorella, M., and W.J. Ripple, 1993. Determining succesional 
stages of temperate coniferous forests with Landsat satellite 
data, Photogrammetric Engineering & Remote Sensing, 59, py, 
239-246. 
Franklin, S.E. and D.R. Peddle, 1989. Spatial texture for 
improved class discrimination in complex terrain. Int, J. 
Remote Sensing, 10, pp. 1437-1443. 
Gumbricht, T., 1995. Watershed structure and symmetry with 
runoff and water quality. In: B. Wiezik (ed), Hydrological 
processes in the catchment. Cracow University of Technology, 
Institute of Water Engineering and Water Management, pp. 37- 
48. 
Gumbricht, T., 1996. Application of GIS in training for 
environmental management. Journal of Environmental 
management, 46, pp. 17-30. 
Gumbricht, T. and J. McCarthy, 1996. Transparent land surface 
modeling in GIS. In: Proceedings Geoinformatics "96, West 
Palm Beach, Florida, USA, April 26-28, in press. 
Gumbricht, T., C. Mahlander and J. McCarthy, 1995. Rule 
based and contextual classification of landscape patches and 
boundaries. In: J.T. Bjorke (ed), ScanGIS '95 Proceedings, 
Trondheim June 12-14, pp. 245-255. 
Hepner, G.F., T. Logan, N. Rittner and N. Bryant, 1990. 
Artificial neural network classification using a minimal training 
set: Comparison to conventional supervised classification. 
Photogrammetric Engineering & Remote Sensing, 56, pp. 469- 
473. 
Lauver, C.L. and J.L. Whistler, 1993. A hierarchical 
classification of Landsat TM imagery to identify natural 
grassland areas and rare species habitat. Photogrammetric 
Engineering & Remote Sensing 59, pp. 627-634. 
Leung, Y., and K.S. Leung, 1993. An intelligent expert system 
shell for knowledge-based geographical information systems: 
1. The tools. Int. J. Geographic Information System 7, pp. 189- 
199. 
McCarthy, J, 1996. Leaf area estimation for 
hydroclimatological models. Manuscript accepted for Nordic 
Hydrological Conference 1996. 
Middelkoop, H. and L.L.F. Janssen, 1991. Implementation of 
temporal relationships in knowledge based classification of 
satellite images. Photogrammetric Engineering & Remote 
Sensing 57, pp. 937-945. 
Ripl, W. and T. Gumbricht, 1996. Integrating landscape 
structure and clean water production. Presented at Stockholm 
Water Symposium "95, submitted to Ambio. 
Robinson, V.B., and A.U. Frank, 1987. Expert systems for 
geographical information systems. Photogrammetric 
Engineering & Remote Sensing, 53, pp. 1435-1441. 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996