and 36 to
trending major structure, the Wadi Ashat zone (REISCH-
MANN & KRÔNER, 1994), cuts across the Haya terrane in
the middle of the mosaic. To the south a large belt of
metavolcanics stands out as a dark bow-shaped promi-
nent feature.
1.2 OBJECTIVES
Since the launch of Landsat-1 in 1972 multispectral data
have been widely used for the extraction of thematic infor-
mation in the fields of geology, mineral exploration, land
use and many others. This process of information extrac-
tion or classification involves several steps. Starting point
is our "real world", the object space, from which we pro-
ceed into image space and feature space. The classes we
want to obtain are defined in a thematic space and have
to correspond, in the end, with actual classes in the real
world (LIST, 1993). In this rather complex process two
divergent techniques are employed: visual interpretation
by a trained interpreter or "automated" classification by
means of a digital computer.
Visual interpretation is the time-honored approach used in
the earth sciences since the appearence of aerial photo-
graphs. As of 1938, "photogeology" can be considered as
an established operational method both in the US and in
Europe (e.g. ANONYMOUS, 1938; HELBLING, 1938, 1948).
Digital classification evolved rapidly after the launch of
Landsat-1. Today, a variety of "supervised" and "unsuper-
vised" classification algorithms is available to the re-
searcher, the maximum likelihood classifier being proba-
bly the most widely used one. Still, all these classifiers
utilize only the spectral properties of the individual pixels
in an image. As compared to visual interpretation, this
approach sacrifices a lot of information that is available
but not used in the classification process. Such informa-
tion consists of at least some existing information on the
geology of the area studied, or geophysical, lithological,
stratigraphic and geochemical data, just to name a few.
This is one of the reasons why digital classification of
multi-spectral data sets produces, in most cases, inferior
results in comparison to a visual interpretation by a
trained geologist. The other reason lies, of course, in the
fact that what is of interest to the geologist is often cov-
ered by weathering products, soil, and vegetation, ob-
structing the identification of lithology.
One possible solution for this predicament is the use of
data from imaging spectrometers like AVIRIS, GERIS,
HYDICE oder AMSS. These hyperspectral data sets allow
precise definition of individual minerals and mineral mix-
tures that cannot be achieved with the current broad-band
multispectral data from systems like Landsat-TM, IRS or
SPOT. Unfortunately, hyperspectral data are still scarce
and rarely available of the area of interest. Still, the prob-
lem of having to work with data that are diluted or altered
by interference from surficial material is there.
A different answer to classification problems is provided
y the integration of ancillary data in a GIS environment,
making use of the "complimentarity between GIS and
427
remote sensing" (WILKINSON, 1996: 85). In this way, impor-
tant information can be gleaned from the various sources
that are also taken into account in the actual thinking and
interpretation process, and be used in addition to spectral
information from remotely sensed data. In the following,
several examples are presented to illustrate these points:
Q Improved visualization of data by integrating remote
sensing, topographic and geophysical data;
Q A complementary method for the detection of iron
anomalies from band ratioing;
Q Classification of lithologic units in an image process-
ing / GIS environment using ancillary data.
2 IMPROVED VISUALIZATION
Satellite remote sensing data present a more generalized
and thus more abstract image of geologic features than
larger-scale aerial photographs. They also mostly lack the
stereo capability of aerial photography that leads to a
model-like rendition of objects that makes it easier for the
human interpreter to correlate information contained in the
image to the familiar three-dimensional view of natural
objects. The difficulties in interpreting satellite data can be
alleviated in several ways:
Q The spectral content of multispectral imagery can be
enhanced in order to provide additional information
to the interpreter. This can be achieved by
decorrelation of the spectral bands and transforma-
tion resp. re-transformation of the data in color space
(KAUFMANN & SCHWEINFURTH, 1986, 1988; GILLESPIE
ETAL., 1986, 1987).
Q The remotely sensed image can be draped over a
DEM and viewed from several positions in space,
thus providing an impression that comes close to the
human way of perceiving geologically related fea-
tures in the field. In combination with height exagger-
ation and superimposing geophysical or other data,
the interpretability of the image data is greatly en-
hanced. Of course, both methods can be combined
by draping color-stretched imagery over a DEM (e.g.
LIST ET AL., 1992; LIST & SQUYRES, 1996).
A subscene of the image mosaic of fig. 1 is presented in
fig. 2. It shows the region of Jibál Mugrar, situated at the
northern edge of the mosaic. An area of about 30 by 27
km at a scale of about 1 : 100,000 is draped over a DEM
derived from toposheets, with a vertical exaggeration fac-
tor of 4. The prominent range in the northern part of the
scene, Jibál Mugrar, consists of a syn- to late-tectonic
calc-alkaline granitic intrusion, surrounded by metamor-
phic volcano-sedimentary rocks with low relief. The poly-
gons resulting from an integrated image processing / GIS
analysis are superimposed on the model. To the east of
Jibál Mugrar the "iron anomaly" described in the follow-
ing chapter is shown. In this way the precise location of
the anomaly is clearly defined in relation to the topo-
graphic features.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996
