necessity of achieving mean geometric accuracies below
the size of half.a pixel, if pixelwise comparison is
performed. This accuracy .can be obtained by automated
image matching and co-registration of the geocoded
image data sets with acceptable efficiency under.certain
conditions. Investigations on co-registration methóds are
currently carried out at the Institute of Digital. Image
Processing of Joanneum Research in order to improve
the performance of the monitoring system.
4.2 Topographic Normalisation
Topography does not only affect the geometric properties
of an image but also has a significant impact on the
illumination and the reflection of the scanned area
(McCormick et al, 1995; Schardt, 1987 and 1990;
Kenneweg et al., 1996). This effect is caused by the local
variations of view and illumination angles due ‘to the
terrain and thus is particularly critical for applications in
alpine regions. Therefore, identical forest-cover might be
represented by totally different intensity values depending
on its orientation and on the position of the sun at the
time of data acquisition.
To overcome this problem, several radiometric correction
procedures are described in literature. Besides empirical
approaches which do not take into account the physical
behaviour of image elements, earlier correction methods
were based on the lambertian assumption, i.e. the
satellite images are normalised according to the cosine of
the effective illumination angle (Smith et al. 1980).
However, most objects on the earth's surface show non-
lambertian reflectance characteristics (Meyer et al., 1993;
Schardt, 1990; Schardt and Schmitt, 1996). As it could be
shown by investigations of Civco (1991) and Colby (1991)
and by the present study the cosine correction has to be
extended by introducing parameters, which simulate the
non-lambertian behaviour of the surface. The estimation
of these parameters is generally based on a linear
regression between the radiometrically distorted bands
and an incidence angle map, derived from the digital
elevation model. From the experiences of this study the
Minnaert correction model is best suited for topographical
corrections.
4.3 Image Calibration - Atmospheric Correction
Variations in atmospheric conditions have a serious
influence on the imagery from earth observing satellites,
especially in the 400 to 2500 nm region of the spectrum
(Haefner, 1992). The atmosphere influences the radiation
reaching a sensor in the following ways (Richter, 1990):
e the atmosphere contributes its own part in reflected
radiation
e it absorbs parts of the incident as well as the reflected
radiation
e it scatters reflected radiation in the vicinity around
object which leads to a reduction of the image
contrasts (adjacency effects). This is to be expected
especially with high resolution satellite sensors
The reflectance changes to be detected in the alpine
forests are small compared with the differences between
absolute calibrated images. For instance, an investigation
of Schardt et al., 1995 demonstrates that drastic changes
in canopy density of 20% only leads to insignificant
signature differences of 3 to 4 digital numbers in the
infrared bands TM 4 and 5 and about 1 to 2 digital
numbers in the green and red spectral range represented
by band TM 2 and 3. From these results it can be
deduced that image calibration is a critical issue in
monitoring applications and has to be carried out very
precisely.
In order to correct these errors image calibration was
performed relatively by using statistical regression of the
image data sets. For estimating the regression
parameters only'areas have to be used which are covered
by forest at all image acquisition dates. This procedure
ensures that only forest areas (pixels) will be considered
which have not drastically changed. Thus, no clear cuts
and strongly damaged areas which are characterised by
very low canopy densities were included. The forested
areas of the multitemporal images (forest masks) to be
used for monitoring were generated by applying threshold
values for each acquisition date (see chapter 4.4). Areas
which are forested at all acquisition dates are resulting
from the overlay of the multitemporal forest masks (AND-
Operator).
Olsson (1993, 1995) states, for example that errors in the
geometry between the multitemporal data sets will result
in calibration functions that underestimate light areas. AS
it could be shown by Olsson (1993, 1995) and this study,
too, (see chapter 4.5) this effect is far more critical in
heterogeneous areas such as forests than it is the case
for intensively used agricultural areas. In order to reduce
geometrical superposition errors the calculation of the
regression parameters is based on image data sets which
are aggregated by a 10 by 10 pixel window. The
aggregation is carried out by calculating the mean values
for each of the windows. This regression parameters are
then applied to the original topographical normalised
images. This procedure is resulting in relatively calibrated
images which are characterised by a similar signature
behaviour.
Besides of relative calibration also absolute calibration
methods exist and will be tested within the objective
studies. For the intended tests the 6S code which has
been shown to be powerful in the past will be applied to
the multitemporal Landsat and SPOT data (Sandmeier,
1997; Vermote et al., 1994). This calibration method takes
into account Rayleigh and aerosol scattering, as well as
gas absorption due to water, carbon dioxide, ozone,
oxygen, methane, nitrous oxide and carbon monoxide
between 0.25 and 4.0 um in a spectral resolution of 2.5
nm. The input parameters for 6S can be chosen from
proposed standard conditions, or specified by the user.
The monitoring system to be established is designed in
such a way that it is open to use both calibration methods,
alternatively. This is due to the fact that the subsequent
268 International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998
proces:
be ba:
images
image
system
4.4 Cla
Collect
For th
informe
indispe
necess
image !
informe
photog
sufficie
be inve
well sui
the acti
lassifi
As a fir
are cr
acquisi!
classific
mask)
masks.
areas
conside
one of
include
procedi
areas v
e COV
e are:
at a
e and
ope
Basical
using :
signatu
severe
vegetat
Therefc
is nece
and nor
As the
classific
pixel-wi
For the
SPOT-t
categor
e fore:
e fore:
* age
