tion. The radiometric correction of the CASI data is done
by the empirical line correction method basing on spectral
reflectance data measured in the field.
The classifications were initially done following the usual
procedures of Maximum-Likelihood classification. At the
same time the spectral endmember for the spectral un-
mixing are determined and verified against the spectral
measurements in the field and laboratory.
The following chapters show first results of Landsat-TM
and CASI data to classify special sediments and vegeta-
tion types.
4. RESULTS
4.1 Classification of sediments
The investigations in the open-cast mining are concen-
trated on the distinction of sediments and water classes.
At the beginning the differentiation of sediment classes
was carried out with spectral clearly distinguishing Terti-
ary sediments due to the very heterogeneous conditions of
the area. The individual Tertiary sediment classes were
classified in the Maximum-Likelihood classification way
with added parallelepiped limits of CASI- and Landsat-
TM data based on field worked mapped training sites.
The mapping was done in a way that only areas with
homogenous sediments were shown. These areas are also
possible endmember for the spectral-unmixing for CASI
and Landsat scanner data. Figure 1 presents first results of
the classification of CASI and Landsat-TM data. Three
types of different sediments are shown. They could be
responded as dark up to dark brown quartzitic Tertiary
Bitterfeldian micaceous sands, as light brown Tertiary
Bitterfeldian micaceous sands as well as Upper Oligocene
up to Lower Miocene lignite on land-derived maps. Areas
with only one homogen sediment could be classified with
an accuracy of more than 95 %. However areas with
varying proportions of sediments within one pixel are
only determined with a limited accuracy. The last men-
tioned ones are more common due to the less spatial re-
solved data of Landsat-TM classification. The differences
one can recognise in the Northern part of the image.
A set of spectral-endmember forms the basis of the use of
spectral-unmixing procedures. It is assumed that the com-
ponents, captured within one pixel, are spatially sepa-
rated. That means the reflected light is only contacting
one component before getting mixed with the reflected
light from other components in the sensor. In that case the
mixing is linear according to the corresponding linear
area portions of the components inside one pixel.
Because of the mining technology there are only a few
areas with spatially separated components. Mixtures of
different sediments and lignite and the formation of so
called mixed sediments are much more common. That
leads to probable interactions between the mixed compo-
nents. The photons of the downwelling radiation can
contact, e.g. through multiple reflection, more than one
component before reaching the sensor.
The presented investigation is based on six homogeneous
initial materials, sampled form the former open-cast min-
ing area Goitsche. The homogeneous initial material con-
sists of the earlier mentioned:
- dark up to dark brown quartzitic Tertiary Bitter-
feldian micaceous sands (T1)
- light brown Tertiary Bitterfeldian micaceous sand
(T2)
- black-brown, upper Oligocene up to lower Miocene
lignite (K1)
and
- A yellow up to yellow brown sands and gravel deposits
of the Weichsel Glacial Periodical lower terrace (Q1)
- brown ground moraine of the Ice Age Saale (glacial
till) (Q2)
- grey ground moraine of the Ice Age Elster (glacial
till) (Q3).
The reflectance of various mixtures was checked to
find out what is the best suited mixing model to describe
the behaviour of the samples as there are
- A linear mixing (also areal mixture)
- intimate mixing
- A coatings and
- molecular mixing.
For this reason the spectral reflectance of various
mixtures of selected endmember and of lignite was tested
in the wavelength range from 400 to 2500 nm in the lab.
Mixtures with percentages between 0 and 100 % in 10
and 20 % steps respectively were modelled from the ho-
mogeneous initial material. The mixtures with lignite
were produced with a lignite percentage of 10, 20 and
30 %. A total of 40 mixtures and 6 homogeneous samples
were investigated and the percentage values were realised
as percentage by volume. The fraction of more than 2 mm
grain size was eliminated to minimise scattering because
of the relatively small measuring area.
Figure 2 shows the spectra of the six initial components.
All spectra were determined by air dry samples in the lab
and they were measured at least three times each. The
shown curves are the average mean of at least three of the
best suited curves of a sample. Besides smoothing
through forming the average mean of some curves a slid-
ing average mean with a window size of three spectral
neighbouring values was used.
One can clearly recognise the concave graphs of the Ter-
tiary samples T1 and T2 in the visible range in contrast to
the convex ones of the Quaternary samples Q1, Q2 and
Q3 and for the lignite K1. The water absorbing bands
occur at 1400 nm and 1900 nm in the VNIR NIR and
SWIR range and the mineral absorbing bands at around
72 International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998
- AN)
€) € DO Oo 959 77 OO = 6 - e
—.
"— Qo U NTI €
p
