THE CASE STUDY 
MONITORING ALAM-PEDJA NATURE RESERVE 
(FROM 1988 TO 1995) 
Area 
Land cover map of the Alam Pedja area covers 1176.63 km?. The 
change map with aggregated cover classes (4) and their transition 
areas is presented in this paper for 46 848 ha, comprising the 
nature reserve area (26 000 ha) and the 2 km wide buffer zone (20 
848 ha). 
Material used for composing the maps: 
I. Digital satellite data in 7 channels from Landsat 5 Thematic 
Mapper (TM) of 08.06.1988 and 12.06.1995; 
II. Digital data bases stored in ARC/INFO format: GIS coverages 
at scale 1.50 000: rivers, roads, settlement, nature reserve and 
buffer zone border; 
III. Auxiliary data and field work. 
Methods 
Pre-processing. At first, satellite data were pre-processed. For 
geometric correction purposes 60 ground control points were 
found from topographic maps. These points were used for image 
transfor-mation (third order polynomial fit, nearest neighbour 
resampling) into Estonian base map projection, Baltic Map 
System 1993. After this step manipulating with ARC/INFO GIS 
coverage's is feasible, that means, masking out nature reserve area 
(Figure 1) and adding all additional vector layers (Figure 4). 
  
Figure 1. Rectified area with mask of Nature Reserve and border of buffer zone 
Unification of the data on multi-date images was performed. For 
all bands experimental lower and upper brightness values for all 
cover classes were determined. With a linear stretch, a new image 
was created by linearly scaling the values between the specified 
minimum and maximum limits (User's Guide, 1997). In this 
procedure, it was important to follow and fit the mean values on 
multitemporal channel pairs. The similar range of brigthness 
value units was important for comparability of two images. 
Principal component analysis (PCA) was used to bringing forth 
the most informative channels of satellite data. For both dates 
approximately 90 per cent of the total variance within the entire 
TM data set belongs to channels 3, 4, and 5. These channels were 
used for creating false colour composite image — TM453. The data 
in these TM channels has been shown to form optimal subset for 
classification of forest and emergent wetland habitats (Sader, 
1989, Sader et al., 1995). 
This composition of channels serves as input data for 
unsupervised classification procedure (ISODATA) and for 
visualisation and delineation of the training sites in supervised 
classification procedure (maximum likelihood algorithm). For 
thematic extraction of land cover data using either automatic 
(unsupervised) or supervised classification algorithms, prior 
knowledge about cover classes typical to the territory under study 
is needed. 
Classification. The classification scheme (Table 1, right column) 
developed for the satellite monitoring purposes in Estonia 
(Aaviksoo, 1995b) was used in this work. The 34 units of the 
classification scheme reflect main categories in Estonian nature, 
differentation of which is possible using Landsat TM data. 
Independantly of the classification method used, the spectral 
classes were labelled by the land cover type in the table in case 
when they occured in the test area. Actually, working with Alam- 
54 International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998 
Ped; 
(all 
pres 
cate, 
Tab 
  
* y 
Cla 
star 
we | 
ider