making the neccessary corrections and-updating 
it according to the date of the satellite imagery 
by interviewing local field workers. 
.3. STUDY SITE DESCRIPTION 
The site of investigation is a part of the a 
forested area near Nour city in Mazandaran 
province in northern Iran (Nour Forest Park). 
The relief is mostly flat and altitudes range 
between —25 and 40 m above sea level. The 
analysis of mountaineous regions has its own. . ... 
problems (Itten et al. 1992) which are pointed 
out in later studies. In this study, a plain and 
flat forested area is chosen to avoid negative 
effects of topography on geometry and 
radiometry. 
The study site covers 366 x 316 pixels (after 
the geometric correction) equalling to 10'409 
hectars. In addition to the forest, there are other 
land use types present in this region such as 
sea, farmland, forest sluice and urban areas. 
The forested area is integrated but the ratio of 
its preimeter to the perimeter of a circle with the 
same surface is relatively high. 
4. Processing and Analysis Methods 
Although the TM data used has been corrected 
to system level (level 5), it is neccessary to 
investigate the radiometric and geometric 
quality. Therefore the bands are visually 
controlled after contrast-enhancement. Image 
banding is quantitatively assessed in a 250 x 
400 pixel part of the full scene in a 
homogenous area within the Caspian sea. The 
banding is calculated as the average digital 
number (DN) in each scan line (each scan line 
in the sixteen-fold period is recorded by a 
detector). The images are also investigated on 
other errors such as dead pixels, duplicated 
rows and columns, and undesired pixels. 
4.1 Geometric Correction 
Because the satellite data is accompaigned by 
geocoded data such as ground truth maps. For 
being able to use the forest maps in a GIS- 
database, it is necessary that the satellite data 
are corrected geometrically. 
For this purpose, a ground control method with 
10 points, which are apparent on the colour 
composite image of the region, as well as on 
the 1:25'000 topographic (base-map) sheet, are 
used. The geometric correction is done using a 
first order polynomial transformation. During 
this process, the original 28.5 m pixels are 
resampled to a 30 m image using the nearest 
neighbour resampling method. 
42 Image Enhancement | 
Reliable information extraction is performed 
using principal component analysis and 
relevant rationing is based on studies by Koch 
et al. (1993): 
R1: TMA/TM2 
R2 : (TM4* TM3y(TM4-TM3) 
R3 : TM4-TM3 
R4 : TM4/(TM3-TM2) 
4.3 Forest/Non-forest Classification 
To separate the forest from other classes, the 
obtained geometrically corrected satellite data is 
classified using a supervised method with . 
Maximum Likelihood (ML) and Parallelepiped 
(PPD) classifiers. By using the reference data 
and on an TM RGB (bands 3, 4, and 5) colour 
composite image, required sample areas for 
forest/non-forest classification are chosen. The 
non-forest class includes a set of water, 
farmland, and urban areas. Several small places 
of thick forested regions are used by villagers 
as sluice (an forest-area to store water for 
agricultural irrigation purposes). The special 
conditions of these areas and the lack of a 
comprehensive definition of forest for these 
specific areas leads to the question if these 
areas belong to the forest or non-forest class. 
Finally, classification is performed in two 
cases: sluice defined as forest and as well as 
non-forest. 
Best band-sets for classification are chosen 
through the calculation of separability 
employing the Jeffrey Matuzita method,adopted 
by leiss (1995). | 
4.4 Classification Accuracy 
Assessment 
In order to determine the accuracy of the 
forest/non-forest map obtained from satellite 
224 International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998 
e 24 ft Wn "0C. 
€) He mt 
e 9 € 
op rs Peg (^ tt Pd bpd bep o (7 pa à.