Based on the table of threshold values for NDVI and TRRI the authors have developed an algorithm for automated classification using an external table of legend. This algorithm allows classification of different data sets by the same table of legend which is defined by a pair of NDVI and TRRI values . The computation flowchart of this algorithm is shown on Figure 5.
Table 1: Threshold values of TRRI and NDVI for different land cover categories

TRRI

100	11	11	11	11	11	11	11	11	11	11
44	8	8	8	8	8	8	8	8	8	8
31	8	8	8	8	6	5	3	4	4	4
28	9	9	8	7	6	5	3	4	4	4
26	10	9	9	7	6	5	3	4	4	4
22	10	9	9	7	6	5	3	4	4	4
20	10	9	9	7	6	5	3	2	4	4
17	10	9	9	7	6	5	3	2	1	1
15	10	9	9	7	6	5	3	2	1	1
11	10	10	10	7	6	5	3	2	1	1
2	12	12	12	12	12	12	12	12	12	12
0	115	127	141	152	167	193	207	217	226	255

NDVI

1 – Closed forest	7 – Land with little vegetation cover
2 – Forest of medium leaf coverage	8 – Dry agricultural land
3 – Sparse forest	9 – Wet agricultural land,
4 – Forest plantation	10 – Water
5 – Orchard or shrub.	11 – Cloud
6 – Other agricultural cultivation	12 – dummy

Figure 5 Flowchart of classification programme
After reading the table of legend, the programme will determine how many classes are to be classified. For each pixel vector the programme computes pairs of NDVI and TRRI values and compares it with those in the table of legend. The pixel which has a pair of NDVI and TRRI values meeting the threshold range of a category in the legend will be classified to that category. With the same table of legend, the programme can classify different input images so we can talk about automated classification. Figure 5 shows a flow chart of the proposed classification scheme.

Change Analysis and Prediction
After classification of the series of image data from 1989 to 1998 we can make change analysis of land cover as well as predictions of the development trends. The natural forest will be a key point of our analysis. The forest cover of 1965 has been computed by using 1965 topographical maps. The data on forest cover of other time is derived from classification result of the TM data series. If we take the forest cover of 1965 as a reference then about 27.7% and 52% of forest cover had been cut by 1989 and 1998 respectively. The rate of forest lost varies over the time. In the period from 1965 to 1989 on average about 1516 ha of forest (4.78%) per year was cleared. During the period from 1989 to 1998 the forest cutting reached a higher rate. Annually about 3.74% forest cover was lost in this period. The analysis of forest cover change is shown in detail in Table 2. When we compare the previous result with official statistical data from the National Statistical Yearbook 1996 we discover very large discrepancy. According to the official data for the period 1989 – 1996 only 11,922 ha of natural forest has been cut in the area of provinces Dong Nai and Binh Thuan, but for the study area that spreads over only two districts the forest loss, estimated from remote sensing data, is 22,262 ha - almost two times greater than that from the Yearbook. This is a fact that reflects unreliability of the existing forest inventory mechanism which needs to be improved urgently. If the forest lost will continue as in the period 1989 – 1998 then by 2016 no natural forest will exist in the study area.
Table 2. Summary of forest cover change in the study area for the period 1965 – 1998

Natural forest area (ha)
1965	1989	1992	1996	1998
131321	94929	81323	69786	63012

Natural forest cover lost rate
1965-1989		1989-1992		1992-1996		1996-1998
Area (ha)	Percentage	Area (ha)	Percentage	Area (ha)	Percentage	Area (ha)	Percentage
1516	1.15%	4535	4.78%	2884	3.55%	3386	4.85%

Conclusion
Based on change analysis of forest cover in the study area from 1965 to 1998 by topographical maps and multitemporal remote sensing data we can make the following conclusions:

• It is necessary to carry out forest cover monitoring and to warn local government on time on their exceed change.
• There is an urgent need to develop automated classification method of land cover to formulate Remote Sensing Based Operational Forest Monitoring System (RSBOFM)
• The classification method based on a pair values of NDVI and TRRI gives acceptable results and it can be used for a data set with at least two spectral channels in red and infrared regions.
• The classification data on land cover and forest cover derived from remote sensing data should be used as a reliable complementary information source to the official annual statistical data.

Acknowledgement
The author would like to acknowledge the Fundamental Research Program of Vietnam for financial support to this research.

Reference

• Nguyen Dinh Duong. Total Reflected Radiance Index – An Index to Support Land Cover Classification. Proceeding of the 19 th Asian Conference on Remote Sensing. 16 –20 November, 1998, Manila, Philippines.
• Shunji Murai. Special Course on Advanced Remote Sensing / GIS Technologies. S-3 Land Cover Change Monitoring. Asian Institute of Technology 1994.
• The National Statistical Yearbook of Vietnam 1996. Statistics Burro Publishing House 1997