The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008 
good data sources for DTM automatic generation and accuracy 
analysis. With the cooperation of several Institute of Surveying 
and Mapping in Shanxi, Heilongjiang and Sichuan, about 157 
well-distributed GCPs were collected with differential GPS in 
2006. The measurement accuracy was better than lm in 
planimetry and 1.2m in height. With aids of these GCPs, 
block-adjustment was made and the orientation accuracy is 
about 5m in planimetry and 2m in height. 
Fig. 4: Shaded DTM of the whole test-field in zone of 
headstream of Three Rivers, the DTM is automatically 
generated from 11 SPOT-5 HRS and nearly 20 HRG images. 
White areas (holes in DTM) are cloud-covered areas. White 
boxes show that coverage of 2 SPOT-5 HRS images, which 
used for DTM accuracy study. 
After the block-adjustment, the proposed DTM generation 
approach was applied to 2 SPOT HRS stereo image strips and 6 
HRG images simultaneously. As a result, 25m grid-spacing 
DTM of the accuracy study area has been generated 
automatically after about 26 hour. As results, about 90 millions 
feature points and 5 millions feature lines have been matched 
which is equivalent to match a 4 x 4 density grid on the original 
image, and it is good enough for generation of DTM with 25m 
interval. Since our DTM generation approach not only generates 
a large number of mass points but also produces line features, 
which are necessary for the modeling the rugged mountainous 
terrain. Parts of DTM results are shown in Fig 5. It can be seen 
that the resulted DTM reproduced quite well not only the general 
features of the terrain relief but also small géomorphologie and 
other features visible in the SPOT-5 images. 
Since the study area is within the WChTM project covered area, 
there are not good enough reference data. Therefore we apply 
the following three accuracy evaluation methods for DTM 
accuracy evaluation: 
(1) Overlay the automated generated DTM onto stereo image 
pairs to apply the manual visual checking under stereoscopic 
display device. Checking results show a good enough match 
between DTM and stereo images expect some small blunders 
within shadow areas, which have to apply necessary manual 
editing. 
(2) Accuracy checking by using GCPs and check points 
measured by differential GPS. Upside of Table 1 give the DTM 
accuracy evaluation results by using this method. We computed 
the differences as GCPs minus the interpolated heights from our 
generated DTM. The accuracy of the DTM is between 1.7 - 6.7 
m depending on the terrain relief and land cover. With this 
method, the errors probably include influence of all three error 
(a): Shaded DTM of arid/semi-arid mountainous terrain. 
(c): Shaded DTM of high-plateau mountain ranges 
Fig. 5: The shaded terrain models of 25m over 3 sub-areas 
in Zone of headstream of Three rivers, Tibet Plateau, 
China. The resulting DTM reproduced quite well not only 
the general features of the terrain relief but also small 
géomorphologie and other features visible in the images. 
Results show that it very necessary to use the multiply 
image matching and edge matching with SPOT-5 
HRS/HRG images over rough and steep mountainous areas. 
(3) Accuracy checking by using manually measured checking 
points which were acquired on digital photogrammetric