Ali Farrag, Farrag 
  
window of the image was used for visual analyzing of image data and used for adjusting the histograms of the green 
bands of the images (details are given later). For further analysis a smaller portion of each image (corresponding to 12 
km x 17 km on the ground), that cover the cement factory and the surrounding area (which is possibly infected by air 
pollution) was processed. These image portions are shown as classified images in plates 3 and 4 for 1995 image and 
1998 image respectively. 
STEP 2: In this step major emphasis was made on the accurate identification and delineation of the contaminated area 
based on the information contents of the composite images. The unsupervised classification was carried out to illustrate 
the possibility of automatic discrimination of the polluted areas. 
The composite images were constructed and displayed in pseudo natural color, where band 2 was assigned to red, band 
3 was assigned to green and band 1 was assigned to blue. Visual interpretation was carried out to give an indication of 
number of classes that can be discriminated on the image and possibility of distinguishing the source of pollution and the 
contaminated places. Several tests were carried out in order to choose the optimum number of classes in the classified 
image. It found that 10 classes give reasonable results. The unsupervised classification was carried out for the purpose 
of automatic discrimination of the polluted area in the images and to identify different intensities of pollution. The 
classes that result from unsupervised classification are spectral classes. Accordingly, the identification of such spectral 
classes will not be initially known, because they based only on the natural groupings in the image values. In this study, 
the classified images were compared with maps and information from the field visits to determine the information 
classes. The resulting classified images by this technique are given in plates 1 and 2 for 1995 image and 1998 image 
respectively. 
The results of visual interpretation of the composite and classified images can be summarized as follows: 
a- The pollution source is clearly identified and the contaminated area is clearly distinguished in the processed images. 
b- There is clear difference between the identified polluted area as they identified from the two images, where the size of 
the identified polluted area from 1995 image is larger than that identified from 1998 image. 
c- Taking into account the wind direction at the time of imaging (which is the average wind direction in the study area, 
almost from North to South) the dispersion and the aerial extent of pollution as identified from the two images take, in 
its general shape, the Gaussian Plume model [Durucan (1995)]. 
d- The ten classes of the unsupervised classified images can be interpreted as follows: 
i- Desert mixed with scattered contaminated areas vi- Reclaimed areas 
ii- Desert vii- Contaminated areas (T) 
iii- Cultivated land (T) viii- Contaminated areas (II) 
iv- Cultivated land (IT) ix- Water bodies 
v- Built-up areas x- Irrigated fields 
e- As given above, different intensities of pollution can be identified and delineated. Accordingly, a smaller window of 
the images was considered for further analysis (12 km by 17 km as defined in step 1 above). Unsupervised classification 
was carried out in order to classify this window into ten classes as shown in plates 3 and 4 for 1995 image and 1998 
image respectively. These classes can be identified as, three classes of contaminated areas with different degrees of 
contamination intensities, four classes as different types of desert areas, one class as built-up areas and one class as 
cultivated (reclaimed) land. 
STEP 3: In this step major emphasis was made on the accurate identification and delineation of the contaminated area 
based on the information contents of the green band of SPOT images, where maximum information about air pollution 
can be extracted from the green band [Farrag (1997) and Nicolas Sifakis (1992)]. The green band of each of the two 
images was processed separately and the obtained results were compared. The processing includes histogram adjustment 
of band images and then applying stretching and slicing to the images of these bands. 
a- Histogram adjustment of image bands 
In order to be able to compare the reflectancy of temporal images, the DNs corresponding to common features were 
examined. For the purpose of linking the results of this study with previous research work which was carried out by the 
author, the reflectancy of the green band in SPOT mage of the same area that captured by SPOT-1 on 23 July 1992 was 
used as a reference for this purpose [Farrag (1997)]. As mentioned above, the window of the image which represent a 
square area on the ground of 458 km” contains several types of land cover, such as urban areas (cities and villages), 
cultivated land, water bodies (river Nile and canals), reclaimed desert areas, desert areas, paved roads, etc. Some of 
these land covers does not suffer from major changes during the period between 1992 to 1998, such as water bodies, 
  
48 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 
pex