165 
Figure 6. Five profiles showing elevation plots obtained from SIR-B derived DEM (solid line) 
and from topographic maps (dashed lines). 
- Failures at the correlation level severely 
impacted the DEM homogeneity such that some of the 
major terrain features have simply disappeared 
(e.g. profile portions IB and 2B). 
Table 2. Measured elevation accuracy values 
(average and RMS differences) for the 5 profiles 
shown in Figure 5. 
Average difference RMS 
Profile (m) dif f erence(m) 
1 
-75 
90 
1A 
-14 
56 
IB 
-126 
82 
2 
40 
124 
2A 
26 
52 
2B 
56 
168 
3 
79 
117 
3A 
-29 
46 
3B 
110 
112 
4 
20 
110 
5 
-20 
86 
5A 
-17 
38 
5B 
- 8 
97 
Average for pro 
files (1,2,3,4 & 5) 
45 
105 
4 CONCLUSIONS 
Comparisons with other results obtained with same 
side stereo radar images are not easy to carry out 
as stereo image parameters may differ 
considerably. However, our results seem to 
compare well with those published earlier such as 
by Leberl et al. in 1985. One can also suggest 
that even though the actual proposed method is not 
yet fully developed for reliable and precise 
digital elevation measurements, there are strong 
indications that under certain conditions (still 
to be identified) height extraction is feasible 
(e.g. profile portion 5A) and could be refined to 
extend to other parts of the terrain. 
An automated digital approach has been 
investigated for the production of digital 
elevation models on mountainous terrain using 
stereo SIR-B image data with 29° and 53° incidence 
angles. The results indicate that DEM production 
is achievable through only digital techniques, but 
tends to be more suitable for some part of the 
image than others. This non-homogeneity in the 
quality of results illustrates the constraints 
which probably originate from variations in 
detailed surface features used for spatial 
correlation.