Re 
ts and 
Italy, 
round 
aerial 
(that 
de in 
"risks 
earch 
5 um 
jealth 
| and 
nsing 
f the 
g the 
993). 
lina 
may 
used 
up to 
1 the 
| the 
on is 
en if 
12/c 
ical, 
land 
| the 
onal 
ixel 
‘the 
the 
the 
Cavazzini, Armando 
  
geometrical distortion not insignificant. For the same reason, the pixel size on the ground is also quite different at the nadir 
and at the border of the same scan line. For example, for a flight height of 1500 meters the nadiral pixel size is of 3 meters 
while the last pixel of the scan line is 5.1 meters. The scan rate of the scanner can be varied according to the flight height 
from 25 scan/sec (1500 meters height), down to 6.25 scan/sec, making the scanner similar to a "push broom" sensor because 
of the high scan speed compared to the speed of the aircraft. The main geometrical distortions that affect the MIVIS images 
are due to roll, pitch yaw, variations in altitude and direction of the aircraft. The roll distortion is automatically 
compensated by the system itself, making the correction of the starting and ending point of registration. 
The scanner has a GPS system and a gyro. It is able to record every second the data coming from these instruments together 
with the spectral data; the CASA 212/C aircraft on which the scanner is installed, has a complete GPS system for accuracy 
navigation and determination of the trajectory of the aircraft. 
Both these systems are able to record data every second, that means, for low altitude flights, every 25 scan lines, and it is 
possible to record the position of the aircraft data set; unfortunately, it is not always possible to have a differential correction 
using a ground station; this aspect causes an error of an order of several meters (i.e. many pixels) in determining the right 
position. 
Finally, terrain elevation plays the most important role in the geometric distortion of the data, both in the position of the 
pixels and in theirsize; a difference of 100 meters in elevation corresponds to a linear difference of 0.2 meters in pixel size, 
te. 7% for a 1500 meters flight. Different routines have been written and applied to the MIVIS data to resample the images 
in order to minimize all these effects. 
Meanwhile, other geographic data sets have been made available to the community to be used for the adjustment of the 
MIVIS images. Colour orthophotos covering the whole italian country at the scale of 1:10.000 have been produced by CGR 
and are available for geographic co-registration of spectral data. Thanks to the use of the orthophotos, many examples of 
georeferenced data have been produced for different areas, with comparable results. For the purposes of this study, a second 
order polinomial warping algorithm has been used to coregister the MIVIS images to the orthophotos. 
The advantage of the integrated use of geographical and remote sensed data is becoming apparent (Catlow et al., 1984; Van 
der Laan, 1988; Kenk et al., 1988). The aim is to enable feed back of remote sensing derived information to a GIS. This 
article briefly describes an example of the use of MIVIS instrument to define different land use classes for risk assessment 
in an industrial area of the Pescara river basin (Sulmona, Italy) resulting in a GIS obtained by the integration of the MIVIS 
data, the classification layer and the orthophotos at the scale of 1:10.000. 
The study is being conducted by CISIG Consortium (Parma, Italy) for ISPESL, a National Health Research Agency that 
evaluates risks in industrial areas, in collaboration with the University of Chieti. 
2. METHODOLOGY 
A linear discriminat analysis method (Dillon et al., 1984) has been used to classify the MIVIS hyperspectral images. In 
previous studies and attemps (Ferrarini et al., in preparation), linear discriminant analysis has proven its suitability for 
classifying MIVIS images with high accuracy. 
Discriminant analysis is characterized as follows: there are two types of multivariate observations - the first, called training 
samples, are those whose group identity (i.e., membership in a specific group) is known a priori, and the second type, 
referred to as test samples, consists of observations for which a priori information is not available and wish to be assigned 
one within the a priori known groups. An observation is classified into a group if the squared distance (also called the 
Mahalanobis distance) of the observation from the group center (mean) is the minimum. An assumption is made that 
covariance matrices are equal for all groups (Dillon et al., 1984; Johnson et al., 1992). There is a unique part of the squared 
distance formula for each group (the linear discriminant function for that group). For any observation, the group with the 
smallest squared distance has the largest linear discriminant function and the observation is then classified into this group. 
Linear discriminant analysis has the property of symmetric squared distance: the linear discriminant function of group i 
evaluated with the mean of group j is equal to the linear discriminant function of group j evaluated with the mean of group i. 
Statistical considerations in discriminant analysis have to do with distributional assumptions concerning the observations 
and measures of separation among the groups. These assumptions have been tested before applying the discriminant 
analysis function. 
Figure 1 exhibits a portion of the area that has been classified. It is an intensively developed region with residential and 
commercial zones, industries, cropland, herbaceous and shrub-brushland, deciduous forest and barren land. All classes of 
interest have been carefully selected and defined to succesfully classify MIVIS data into land-cover information. This 
requires the use of a classification scheme containing taxonomically correct definitions (Jensen, 1996) of classes of 
information. There is a fundamental difference between information classes and spectral classes (Campbell, 1987): 
information classes are those that human beings define; spectral classes are inherent in the remote sensed data and must be 
labelled by the analyst. 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 237