LANDSAT TM data from 1989, 1992, 1996 and 1998 of the study area enable us to monitor the change of land cover. False color composites of the years 1989 and 1998 are shown in Figure 2. By only visual comparison one could recognize a big change in forest cover between the two dates. The destroyed forest is not only evident by clear cutting but also by forest quality degradation. The latter is possible to detect by remote sensing techniques, however, it requires more complicated analysis methods.

Figure 2. Forest cover of 1989 (left) and 1998 (right) based on LANDSAT TM. Color composite band 5 = red, 4 = green, 3 = blue



Figure 3. Forest cover classification result of 1989 (left) and 1998 (right) by LANDSAT TM
To carry out fast analysis of multitemporal TM data the authors have applied the classification method proposed in [1]. This is a simple automated classification method suitable for natural resource inventory in a large area. The classification method is fast and provides results with acceptable accuracy. This method is based on the normalized vegetation index NDVI and the total reflected radiance index TRRI. A scatter plot of NDVI and TRRI is shown on Figure 4. While the NDVI provides information about vegetation distribution in the study area, the TRRI shows levels of radiance reflectance of land cover objects. Apparently, there are many objects with the same NDVI values and by taking TRRI in to consideration they can be classified further into different classes.

The NDVI is computed by the following formula:

Where Infrared and Red are digital counts of infrared and red channels.

The TRRI index is computed by the formula:

Where
I_i ... Digital count of a pixel in channel i
n ... Number of spectral channels
Imax ... Maximum digital count
D ... Integral step


Figure 4. Scatter plot of NDVI and TRRI indices
Data Normalization and Classification
Multi-temporal data needs to be normalized before classification. There are several methods to do this but the simplest one is based on using mean values and standard deviation as shown in [2].

Suppose in time t the data set has a mean value m_t and standard deviation s_t and at time t+1 the corresponding values are m_t+1 and s_t+1 then values of time t normalized for time t+1 are given the following formula:

The time point 1998 has been chosen as a reference for normalization of data from another time. After normalization by the given formula the data set is ready for analysis by NDVI and TRRI indices . Thanks to data normalization threshold values defined for the data set 1998 and given in Table 1 can be applied to all other time points.