incer on
specially
has also
vard and
els. The
Planetary
y the
? sensing
art of the
re by the
examples
e Internet
2 DATA ANALYSIS
The basic goal of the remote sensing data analysis is to
support flood protection and prediction systems. According to
the user requests known so far the following fields of work
have to be in the main focus:
e contributions to flood modelling and prediction,
e methods of data acquisition and data and information
management,
e derivation of information to identify the shortcomings in
the flood protection systems and the risk potential for
defined areas.
In relation to these main tasks the analysis and interpretation
of the remote sensing data acquired during the Oder flood in
1997 was focused in a first step on
|. derivation and presentation of the flooded areas in relation
to the surroundings and the river plain,
2. extraction of current land use information,
determination of precise land elevation data and
4. integration of these information into a flood risk
information system on the basis of a GIS.
22
2.1 Satellite image maps
To realize the first two points based on the acquired satellite data,
a number of satellite image maps of a scale of 1: 100.000 or
| : 50.000 covering the flooded areas at different times during the
flood event and with a land use classification have been created.
The geographical location and the size of the satellite image maps
have been aggreed on with the Regional Environmental Office of
Brandenburg and is shown in Fig. 4. This set of 5 overlapping map
sheets contain the lower Oder river from Gartz (northern part of
the common German-Polish river border) till Krosno and Nova
Sol in the western part of Poland.
An additional map sheet was prepared of the area near
Frankfurt/Oder including the flooded river plain ,Ziltendorfer
Niederung". The scale of this map was 1 : 50.000. The satellite
data of the Indian IRS-1C were acquired at the September 16,
1997, so the map shows this area about one month after the flood.
2.1.1 Radiometric ^ preprocessing. The radiometric
preprocessing and correction for the optical data (Landsat-TM und
IRS-1C) includes the atmospheric correction of the data sets and a
special histogram preparation and matching procedure to ensure a
coloration in a near natural color scheme. For the atmospheric
correction a model of a midlatitude summer atmosphere with
calibration parameters from Bolle was applied. Special problems
arised with strong differences of the acqisition date of the scenes,
so prior to mosaicking, a detailed investigation of the histograms
of all bands was done. To realize a nearly natural color scheme for
all images from April till August the combination of the Landsat-
TM - bands 7-5-3 has given the best results. After histogram
matching all single scenes were mosaicked and the final histogram
tables for the color composition were defined.
The main aim of preprocessing radar images is the speckle noise
reduction. Because of the significant differences of the statistical
features of the SAR data different combinations of local filters
were used. For most of the data a noise reduction on the basis of a
Sigma-Lee kernel with an additional smoothing of disturbing
image structures was made. But an optimal procedure for all data
sets to ensure a consistent and uniform data preprocessing
however could not be found.
= Blatt Frankfu
i YH £2 t i
Fig. 4: Geographical location and size of the satellite image
maps
Therefore the achieved results of the data preprocessing were
different too, especially in relation to the statistical features of the
images (variance, skew). That is of importance, because most of
the further image processing procedures for radar data statistical
features or local neighborhood relations take into account. A
detailed investigation showed that the main differences occured on
the land surface, while the water areas were comparatively equal.
That is why the SAR data sets were only used for the computer-
based extraction of the areas covered with water, but were not
included into the land cover classification carried out in the next
processing step.
2.1.2 Geocoding. The resampling of all images was done on the
basis of reference points to a geocoded digital map. The accuracy
of the resampling process was verified with conventional map
sheets TK 10 (scale 1 : 10.000) of Brandenburg. The RMS error of
the ground control points was of the order of several meters for the
merged IRS- and the fine mode Radarsat scenes and of 15 to 25
meters for the Landsat-TM data which means a sub-pixel
accuracy. The coordinates of the maps are based on the Gauss-
Krueger projection with the reference ellipsoid of Bessel. The
middle meridian of the projection strip is 15°.
2.1.3 Extraction of the flooded area. As shown in Table I,
satellite data for the extraction of the flooded areas were available
in the time period from 22.07.97 till 10.08.97. Due to the different
dynamic of changes during that time, the data sets included in the
cartographic presentation of the flooded areas were also different.
So, for example, in the area of the ,Ziltendorfer Niederung"
(mapsheet Frankfurt/Oder), where several dike breakes occured
after the 23.07.97, more data sets with a shorter time intervall were
included. On the other hand the differences in the northern part
International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998 185
