The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008 
185 
For length is the most important factor when selecting 
mainstream, we can set the weight value of the length and 
calculate the other two values as follows: We suppose the 
weight of the length is wl then the weight vector is W = {wi, 
w 2 *, w 3 *}. 
Di = r u *a + r 2i *w 2 . + r 3i *w 3 . (1) 
b = 1-Wj (2) 
We calculate the w2 and w3 that will let di get maximum value. 
n n 
b lL r 2j b lL r v 
7=1 7=1 
i'=2 j-1 ;'=2 7=1 
Using expressions (1) and (2) we obtain the weight vector, then 
we can calculate the decision vector D. the maximum d of the D 
is the best choice, and the corresponding route is the 
mainstream, the headwater of the mainstream is the source of 
the river, and the length of the mainstream is the length of the 
river. 
5. EXPERIMENT AND CONCLUSION 
We use the 1:1000000 scale topographic data to check the 
validity of the method. The experiment area involves lake and 
double-line river. As shown by Figure 9, a tree-like river 
network is obtained through extracting the centerlines of the 
double-line river and lake. In this experiment, we assign the 
length weight with 0.5, for the length is the most important 
factor when identifying mainstream and headwater. As 
illustrated in Figure 10, the derived mainstream is reasonable 
and according to the three factors very well. 
For large rivers often have loops, In order to obtain a tree-like 
single-line river network we must cut the loops. This requires a 
manual decision or arbitrary decision. Further work should be 
done to develop automatic method to cut loops. Multiple criteria 
approach used in selecting mainstream and headwater brings the 
three important factors together into consideration, and it 
considers their importance as well. A great advantage lies in that 
it can be extended for further usage, for example, if the data 
have information of altitude, we can let the altitude of the 
headwater to be the fourth factor to help us make more 
appropriate decision. 
RERFERENCES 
GUO Qingsheng, et al, 2003. Intelligentize Processing of 
Geo-graphic Information, Wuhan: Publishing House of Wuhan 
University. 
LIU Shaochuang, 1999. Interpretation of Headwaters of Lan- 
cangjiang (Mekong) River by 3S Intergration Technique. 
Geo-Information Science. 2, pp. 28-30. 
Michael McAllister, Jack Snoeyink, 2000. Medial Axis Genera 
lization of River Networks. Cartography and Geographic 
Information Science. 27, pp. 129-138. 
QU Yonghe, 2004. Discussion of Determining the River Source 
of Xiuhe River. Jiangxi hydraulic Science and Technology, 30, 
pp.153-156. 
TAN Xiao, WU Fang, 2005. HUNG Qi, DENG Hong-yan. A 
Multi-criteria Decion Model for Identifying Master River and 
Its Application in River System Construction ACTA Geodaetica 
et Cartographica Sinica, 34, pp. 154-160. 
WU Hehai, 1995. The Automated Construction of Tree 
Structure of River Network. Journal of Wuhan Technical 
University of Survey and Mapping, 20 (Supplement), pp. 7-14. 
ACKNOWLEGMENT 
This research is supported by National science and technology 
support project (2006BAJ09B02): Key technologies study on 
Villages and small towns space data conformity and renewal.