a) Radiometric pre-processing. In a multitemporal
investigation, the first step concerns the different image
radiometric calibration. This calibration has been
performed by equalising the June illuminating conditions
to those of April, subtracting a linearly increasing digital
number (Dn) variable value to the June image. The
variation curve
(Ax — XJun — X Apr = Jon = OX jun + b) (1)
has been determined by means of a linear regression
that was calculated by taking into account non time-
variable features radiometric values (in this example,
these features are large industrial buildings roofs in Alba
surroundings). The features’ DN differences is only
related to different illuminating conditions, thus, the
difference can be used to equalise the whole image set.
The Dns recorded in correspondence to different
features, show a linear dependence between the Dns
variation and the Dns recorded in the reference image
(for example the June image). The calculated regression
curve and the correlation coefficient is shown in Tab. 3.
Calculated function Yz0.8659X - 15.923
Correlation index
R2 - 0.9989
Table 3 - Calculated regression curve and correlation
coefficient (R)
b) Geometric pre-processing. This step only concerns
the absolute georeferencing of one image and the relative
image to image geometric fitting: An accuracy of less
than half the geometric ground resolution (accuracy « 5
m in the case of Spot 2 PAN image) has been obtained,
allowing a good multitemporal overlapping.
c) Image classification. This step derives from the
assumption (verified by field evidence) that flooded areas
appear in brighter grayscale tones on the image, because
of the presence of mud and sand carried by the rivers.
Thus, extracting a radiometric range (a pseudo-spectral
signature in Spot-2 PAN images), it is possible to classify
the whole data set, subtracting the April image from the
classified June image in order to eliminate features that
are not directly linked to the flood event.
d) Generation of a clivometric model. A clivometric
model has been generated from the DEM acquired for the
Piedmont regional technical map (scale 1:10.000),
resampled to a 10x10 m grid. This model has been used
to define the hydrographic basins, and thus to eliminate
any noise from the features (landslide, mud areas. etc.)
not directly related to the flooded areas.
e) Photogrammetric coverage comparison.
Stereoscopic models derived from the panchromatic
photogrammetric coverage acquired in the days following
the flood event has been analysed in order to plot the
flooded areas; this vector plot has permitted one to
compare the classification results and to calibrate the
radiometric range extracted.
f) Final image generation. By forcing all the features
classified as not-flooded areas to a null value, a new
synthetic image has been generated, and it is shown in
fig. 4.
6. LANDSLIDE ASSESSMENT
Following the November 1994 flood in the southern part
of Piedmont (in particular in the Langhe hilly region),
several landslides (debris flows and slides) occurred. The
aim of the study is thus, to locate and classify the above
mentioned phenomena, using remotely sensed data.
Figure 4 - The flooded areas extracted by Spot-2
panchromatic data sets
Images acquired from satellite platforms present several
advantages, taking both the landslides regional scale and
the efforts that should be accomplished in the case of an
in-situ analysis into account.
As mentioned and shown in fig. 1, the new methodology
has been implemented in a training area, using the data
described in § 4.
A flow chart showing the different steps of the analysis is
presented in Fig. 5.
Radiometric
pre-processing
i
Geometric
pre-processing
First method Second method
em
Classification Featu re
extraction
Slope Slope
range range
Dipping Dipping
range range
Final
i
| :
| image
Figure 5 - Landslide assessment flow chart
68
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996
OQ OO QQ —
WN =
