defining the
in soil an
_yalko et a
the problem
sultant vector
ISing vectors
regression is
two bands
ower;
provided by
sensing data
IS kinds of
ar moisture
moisture a
;oil moisture
active nadir
ed and used
\aracteristics
ted radiation
ined by the
of its vertical
otely sensed
eceived both
were applied
| system ofn
ie (Lyalko et
new method
rborne nadir
jmenting the
ling to the
yetween the
nded into à
the limits of
-haracterized
which are
Homogenous landscapes usually have certain features of
textures. Hence, the discrimination of these textures in
satellite images helps to segment the images to localize
exactly boundaries of homogeneous landscapes, upon
which the regression functions are modelled.
To define a texture, the Gibbs random field probability
model with double interaction of responses (Jain,
Gimel'farb, 1995) was used. The model defines a spatial
homogeneous texture as realisation of Markov's random
field sample which is given by the Gibbs probability
distribution.
If the structure is known, the model allows to generate
sampling realisations by means of element-by-element
stochastic relaxation. The model parameters for each
kind of texture can be estimated in accordance to training
patterns. First the analytical initial approximation for
maximum likelihood potential estimations are calculated
for a lot of interaction types. A map of these interactions
is generated. This gives the possibility to compare them
regarding to its strength and to select the most
representative structure of interactions for the given kind
of texture.
After this, the initial estimations of potentials for the
selected kinds of interactions will be completely
determined by stochastic approximation. The texture
analysis proposed above was carried out for an area near
the Chernobyl Nuclear Power Plant (NPP). Five types of
landscapes were discriminated on this site:
1- pine xerophyto-lichenous
2- oak-pine hazelous grassy
3 - oak-hornbeam-pine fern-sedgous
4 - alder marshy forest
5 - long-fellow non drained lands.
As a result of the experiment, the discriminated types of
landscapes were distinguished according to their texture
except for oak-pine hazelous grassy and oak-hornbeam-
pine fern-sedgeous forests.
The result of the automatic segmentation of a satellite
image subset according to the textures of the landscape
types showed pine xerophyte-lichenous and sparse
growth of trees to be discriminated most significantly.
The other types of landscapes were discriminated some-
thing weaker.
The volume of information and the spatial relations of
investigated phenomena require the use of GIS
technology.
For this purposes the GIS TRIAS has been developed by
the uve Remote Sensing Centre Potsdam to match
remote sensing data and environmental information
(Lyalko, Marek et al., 1995a). TRIAS is a MS-Windows
based user friendly and low cost GIS working on IBM-
compatible PC's. Similar to other GIS TRIAS allows to
model the natural phenomena as point, line or areal
objects. These objects can be connected with relational
data bases. The definition of object hierarchies, text
objects and thematic relations between the objects is
possible, too.
The following properties and possibilities of TRIAS have
been used to carry out the investigations:
- overlaying of vector and raster data
- Interactive vectorization of raster data
- flexible catalogue of object classes which is oriented on
cartographic standards and can be extended on
thematic characteristics
457
- free possibilities for combining topographic and
thematic map information and various presentation
models
- possibilities for generating map output with a great
number of presentation tools and a comfortable
connection to network plotters under MS-Windows
- connection to retrieving features in a relational data
base.
The handling of primary remote sensing and ground truth
information allows to compare great numbers of thematic
and temporal layers and to derive tendencies with regard
to spatial and temporal changes of various natural
components.
Finally the use of GIS technology allows to carry out the
classification of the test site for further studies on a
qualitatively high level.
3.2 Results
The programs developed have been used to analyze the
optical vegetation properties within the southern trail of
pollution induced by the Chernobyl accident. Primary
material for the work were multispectral space images
acquired by the KATE - 200 camera with a focal length of
200 mm and a ground resolution of about 30 m on the
,Cosmos" satellite, taken at the 10th of July 1980, as well
as by the MK - 4 camera with a focal length of 300 mm
and a resolution of about 20 m on the ,Resource -F-2"
satellite, taken at the 27th of July 1989. The three
spectral bands were recorded in the green (500 - 600
nm), red (600 - 700 nm), and near-infrared band (700 -
840 nm). Double-negatives of the space photographs
enlarged by eight times were used for the investigations.
The brightness values were provided for the various
landscape elements by image processing at a quite
dense and regular net of observation points. Thus, each
point is characterized by its coordinates, brightness in
three spectral bands, landscape element feature and
radioactive contamination level.
In order to get the radioactive contamination level of the
landscape elements, the map of Cs-137 contamination
density for soil (scale 1: 200 000) by V. A. Nagorsky
(firm ,Pripyat', 1993) and also the map of Cs-137
contamination of coniferous canopy (scale 1: 200 000)
by À. |. Ostravnenko (1990) were used. Currently Cs-137
provide the main radioactive contamination of the
Chernobyl region.
Nine homogeneous landscapes were identified within the
limits of the test site. The major elements of landscape
are pine forest growing on dry soils, grassland, leaf-
bearing and mixed forest, and agricenouses on dry and
moist soils. The remaining landscape elements play a
minor role because of their small area sizes.
Using regression model (1) described above and the
programs developed by ZAKIZ, the dependencies for the
major landscape elements were determined for the red
and near infrared bands of the space photographs. The
calculations were carried out for first order polynomials
(m = 1, m - maximum degree of polynomials).
For coniferous canopies before and after the Chernobyl
accident the relations between the spectral properties in
the red and near-infrared band are absolutely different
which can only be explained due to radioactive
contamination.
Even for vegetation growing on soils with low levels of
radioactive contamination (less than 2 Ci/km? ) the
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996
