| of
on.
ent
on.
| of
ent
ted
ded
lue
our
stic
and
Ily,
ole
tric
the
the
ain
re-
lly
der
ion
the
the
1ap
[he
of
ere
ich
om
ted
in the north of the Netherlands, bordered by geographic
latitudes 52°45'-53°30'N and longitudes 5°00'-6°15'E. It
consists mainly of agricultural areas, some urban areas,
water bodies and a coastal zone including the island of
Ameland as shown in Figure 2.
Friesland/Ameland
Figure 2: Location sketch of calibration test site in the
Netherlands
The actual research test site is located on the south-west
coast of Sumatra in the Province of Bengkulu, Indonesia.
It extends from 3?15' to 4?45'S and 102°00° to 102°
30'E. Sumatra is an equatorial tropical island with a
climate that is characterised by constant high
temperatures and extremely abundant rainfall, well
distributed over the year. With relief ranging from sea
level to 2,400 metres, and complex land forms, tropical
forests, and mixed agricultural systems, the research site
was selected as a representative, typical problem area
which experiences extensive cloud cover. Figure 3
indicates the location of the research test site in the
western part of the country.
dg
Li 1 e
AT
Figure 3: Location of research test site, Bengkulu -
Indonesia
Among the selected sensors that provided the remotely
sensed data were ERS-1 SAR, JERS-1 SAR, SPOT HRV
and Landsat TM. The sensors, supplying data of
different spatial and spectral resolutions, cover areas
ranging from 3,600 km? (SPOT) to 34,225 km? (TM).
For both test sites a full set of images from each sensor
was collected. In the case of ERS-1, a multi-temporal
data set was requested in order to be able to fulfil the
research objectives and test the influence of multi-
temporal and multi-orbit ERS-1 SAR data on image
fusion. The data of the calibration test site in the
Netherlands consisted of multiple ERS-1 SAR scenes
acquired during the first half year of 1993. The optical
data set contained SPOT XS data from 1986, SPOT PAN
dated 1989 and a Landsat TM scene from 1992.
Concerning the research site in Indonesia, it was much
more difficult to obtain the remote sensing data due to a
number of constraints. It was problematic to find optical
remote sensing data from SPOT and Landsat with less
than 30% cloud cover. In the case of Landsat TM one
scene could be identified in the archive that was not fully
657
covered by clouds collected in 1990. In terms of SPOT
XS the only useful scene dated back to 1987. Based on a
request, further scenes could be acquired in the summer
of 1994. Together with a three dates coverage of ERS-1
SAR and two JERS-1 SAR coverages with a time
difference of one year the data set was completed. It was
necessary to acquire two JERS-1 SAR scenes per
coverage which were used to build a mosaic in order to
cover the entire research test site.
4. IMAGE PROCESSING
First, the Dutch data was processed in order to calibrate
the image fusion approach and to identify suitable
techniques for the cloud removal without having to deal
with terrain induced distortions. In addition, up-to-date
topographic maps of 1:50,000 and larger scale were
available to verify the findings. The experience gained
from these experiments was later adapted and applied to
the Indonesian area.
4. 1 Radiometric Pre-Processing
An atmospheric correction was applied to the optical
remote sensing data in order to remove haze effects from
the imagery. Because no ancillary information, necessary
for correct atmospheric modelling, were available, only
an approximation (histogram modification) for this
correction was applied. The optical remote sensing data
of the calibration site (Netherlands) was not affected by
striping. The Landsat TM scene of the research site in
Indonesia contained 16-lines striping. One of the SPOT
XS scenes had a striping appearance too. For destriping,
each pixel was adjusted based on a comparison of the
local mean of the line to an unweighed local mean of
those closest neighbouring lines. But it has to be taken
into account that this destriping is a cosmetic act that
should be performed after spectral interpretation or
processing of the images (Crippen, 1989). However, the
destriping was done prior to image fusion to take into
account the radiometric characteristics of the VIR data
alone which cannot be separated from SAR in fused
imagery.
The SAR data were reduced from 16- to 8-bit applying
linear scaling in consideration of the mean and standard
deviation of the particular images in order to reduce the
data volumes. In addition, a speckle reducing filter (3x3
Gamma MAP) was applied for the spatial enhancement
of the data (Shi, 1994).
4. 2 Geocoding
The Dutch data set was geocoded using well-distributed
GCP's and polynomial rectification because the terrain
height influence on the geometry was negligible. The
largest height difference reaches + 10 m in that region.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996
