cks
rocks
rocks
ocks
ocks
'etation of
t over the
one.
in the GIS (see fig. 1). Haze and histogram corrections
were applied to the data in order to minimize atmospheric
influences.
The second data set comprises the available geologic
information available for that area. It is contained in a
geological map that was produced through a combination
of visual image interpretation of "optimized" Landsat TM
images (KAUFMANN & SCHWEINFURTH, 1989), field work,
and laboratory investigations. The available aeromagnetic
and gravity data were also transformed to the UTM grid
and input into the GIS.
The geological map of the study area, comprising about
40,000 squ. km, shows 61 geologic/lithologic units. For
digital classification as well as for representation on a
small-scale map, this number is far too high. Conse-
quently, similar lithologic units were lumped together until
27 major units were left. These units were arranged into 5
groups:
1. Quaternary (3 units);
2. Intrusive rocks (6 units);
3. Sedimentary and metasedimentary rocks (7 units);
4. Volcanic rocks (8 units);
5. Volcano-sedimentary rocks (3 units).
For each group, a mask was created and a maximum
likelihood classifier was applied to the ratio data within this
mask, using the respective lithologic units as classes. All
in all, 240 representative sites (ROls or AOls) about which
reliable field and petrographic evidence was available
were carefully selected within the entire map region and
used as "training areas". In this way, both the a priori
knowledge from the map and the spectral data could be
used in the classification (Orr, 1996). Subsequently, the 5
separate classification results obtained for each of the
masked areas were joined for the final classification
shown in fig. 7. The classification is presented in its "raw"
form, without any smoothing algorithm applied.
The result of a first supervised digital classification using
these additional data sets was a significantly improved
discrimination of the geologic units as compared to a clas-
sification based on spectral data only. Actually, a spectral
classification into 27 lithologic classes would be a rather
hopeless undertaking in the first place.
The result is also interesting when compared to the origi-
nal map information used for the masking: There are sev-
eral cases in which the interpreter was somewhat doubtful
about divisions he had made on the map, e.g. the separa-
tion of acidic vs. basic intrusives. In this combined classifi-
cation it can be seen that some of the doubtful decisions
were, in fact, not correct.
The resolution of the available geophysical data is, unfor-
tunately, too low to be used in the classification for dis-
cfiminating most of the individual geologic bodies; only
larger structural blocks within the different terranes can be
Separated. Therefore, the geophysical data sets were not
used in the classification. It was also tried to use neural
networks for this rather time-consuming process. At the
429
moment, however, the amount of data involved cannot be
handled satisfactorily by neural networks.
5 CONCLUSIONS
The combination of image processing and GIS greatly
enhances the potential of interpretation and classification
of remotely sensed data. The use of GIS technology is a
realistic way to include a priori knowledge, topographic,
mineralogic and geophysical data sets into the classifica-
tion process, leading to results superior to both purely
spectral classification and existing field/interpretation geo-
logical maps.
The combination of topographic and geophysical data with
remotely sensed imagery improves the visualization and
thus the interpretability of geologic data. GIS classification
of geophysical data sets provides a "second opinion" on
results of spectral analyses.
6 ACKNOWLEDGEMENTS
The research described here was conducted within the
Special Research Project on Arid Regions (Sonderfor-
schungsbereich 69). Funding by Deutsche Forschungs-
gemeinschaft and Freie Universitat Berlin is gratefully
acknowledged.
7 REFERENCES
ALMOND, D. C., AHMED, F. & DAwouUD, A. S. (1983): Tec-
tonic, metamorphic and magmatic styles in the north-
ern Red Sea Hills of Sudan. - Bull. Fac. Earth Sci.,
King Abdulaziz Univ., 6, p. 450 - 458, Jeddah.
ANONYMOUS (1938): Aerial photography used extensively
in New Guinea oil search. - World Petroleum, 8, 10,
p. 44 - 47, Houston, TX.
BOLSTAD, P.V. & LILLESAND, T.M. (1992): Rule-based clas-
sification models: flexible integration of satellite imag-
ery and thematic spatial data. - Photogramm. Eng.
remote Sens., 58, 7, p. 965 - 971, Bethesda, MD.
GILLESPIE, A. R., KAHLE, A. B. & WALKER, R. E. (1986):
Color enhancement of highly correlated images. |.
Decorrelation and HSI contrast stretches. - Remote
Sens. Envir., 20, p. 209 - 235, New York, NY.
GiLLESPIE, A. R., KaHLE, A. B. & WALKER, R. E. (1987):
Color enhancement of highly correlated images. ll.
Channel ratio and "chromaticity" transformation tech-
niques. - Remote Sens. Envir., 22, p. 343 - 365, New
York, NY.
GRANT, F. S. & WEST, G. F. (1965): Interpretation theory in
applied geophysics. - 584 p., New York (McGraw
Hill).
HELBLING, R. (1938): Die Anwendung der Photo-
grammetrie bei geologischen Kartierungen. - Beitr.
geol. Karte Schweiz, N.F., 76, p. 1 - 67, Bern.
HELBLING, R. (1948): Photogeologische Studien im An-
schluf$ an geologische Kartierungen in der Schweiz,
insbesondere der Tódikette. - 141 pp., Zürich (Orell
Füssli).
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996
