4. GROUND TRUTHING
To yield acceptable classification results, training data is
needed. This means that the image analyst must develop
training statistics for all spectral classes constituting each
information class to be discriminated by the classifier. A
supervised classification requires a thorough knowledge of the
geographic area. Most importantly, the quality of the training
process determines the success of the classification stage and,
therefore, the value of the information generated from the
entire classification effort. The overall objective of the training
process is to assemble a set of statistics that describe the
spectral response pattern for each land cover type to be
classified in an image [Lillesand/Kiefer 1987].
An adjusted mapping key for ground truth measurements was
generated together with the Agency for Ecology of Lower
Saxony. This mapping key is adapted to the special mapping
key for biotopes in Lower Saxony with respect to § 28a NNatG
(Lower Saxonys Environmental Protection Statutes)
[Drachenfels, 1992]. For selected queries and overlay analysis,
these mapped test areas were transformed to separate GIS layer
in ArcInfo® format. This procedure was facilitated by an
adapted graphical user interface written in ArcInfo's macro
language AML (Figure 2) [Schelling, 1995].
+
+
nmmmnmnnmmnn PT
HTH HH HHH
+
5. GEOMETRIC PRE-PROCESSING
To use remotely-sensed imagery and their classification results
in geographic information systems, these image data have to be
transformed to an uniform reference system. Using the
polynomial correction techniques, an image can be registered
to a map coordinate system allowing its pixels to be addressed
in terms of map coordinates rather than pixel and line numbers
[Richards, 1994]. Many applications of remote sensing image
data require more than one scene of the same geographical
area, acquired at different dates, to be processed together. Such
a situation arises when, as in our monitoring project, changes
are of interest, in which case registered images allow a pixel
by pixel comparison to be made. There are two ways to register
two images to each other. You can register two images to each
other by registering each to a map coordinate base separately,
or alternatively, one image can be chosen as a master image to
which the other, known as the slave, is to be registered. We
chose the second method for our study. In our case the master
image was the Landsat-TM imagery database from 1990/9]
covering the complete area of Lower Saxony. This imagery was
georeferenced using lst order polynomial procedures with a
nearest neighbour resampling technique to a German
topographic map with a Transverse Mercator projection based
on a Bessel ellipsoid. As a result, the maximum residual error
is 1.5 Pixel (45 meters). The scrutiny of residual errors was
made on the one side by overlaying geometrically correct
ATKIS-datasets and on the other side by measuring the same
unequivocal radiometric pixels in both imagery (master to
.000 Scale
. 000
1 = Select 2 = Next 3 = Who 9 = Quit
Label 55 User-ID: 55 (3459938.500.5844350.500) Angle 0.000 Scale
4 = Select 2 = Next 3 = Who 9 = Quit
Label 53 User-ID: 53 (3459776.500,5844365.000) Angle 0.000 Scale
1.000
1 = Select 2 = Next 3 = Who 9 = Quit
Label 60 User-ID: 60 (3459344.000,5844205.500) Angle 0.000 Scale
1.000
1 = Select 2 = Next 3 = Who 9 = Quit
Label 48 User-ID: 48 (3453367.500,5844471.000) Angle 0.000 Scale
1.000
1 = Select 2 = Next 3 = Who 9 = Quit
Figure 2: ArcInfo Menu for Information Input
686
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996
^-^ o Fh PM üÜP-—5 5 ab mud
--
^» Fh MM NS
"^ o» ALI
