g has made 
rich claims 
h regard to 
with regard 
sion, some 
concerning 
ince of the 
rsion of the 
phics 4D35 
transferred 
raphs were 
xel size. 
summarized 
ossible nor 
ive projects 
'eo-overlap. 
in principle 
forest areas 
he accuracy 
the DEM 
suited for 
n, although 
| the system 
ree projects 
hotography. 
and concern 
terrain and 
  
  
  
  
  
  
specifications project 1 project 2 project 3 project 4 project 5 
hoto scale: 1:7000 1:14000 1:12000 1:22600 1:30000 
flying height : 1000 m 2100 m 1900 m 3200 m 4500 m 
size of DTM: 543 x 1247 m? 552 x 1075 m? 1450 x 2100 m? 2.3 x 3,8 km ? 3.3 x 3.5 km? 
grid width: 3m (15pm) Sm (15pm) Sm (15pm) 10m (15 pm) 14m (15 pm) 
6m (30pm) 10m (30pm) 10m (30pm) 20m (30 um) 28m (30 um) 
area description: plain slightly hilly hilly mountainous 
agricultural same as project 1 agricultural smooth rough, rocky 
dH - 5m dH = 45m dH = 180m dH = 600m 
residential area 
residential area 
several villages 
terrain breaks 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
interest points 2.9 105 (15pm) | 0.6 10° (15pm) | 3.6 10° (15pm) | 3.5 10% (15pm) | 6.0 10° (15pm) 
(left and right): 0.7 10° (30pm) | 0.15 10° (30pm) §| 0.9 10° (30pm) {| 0.9 105 (30pm) || 1.5 10° (30um) 
measured terrain 499005 (3m) 148771 (3m) 758658 (3m) 714997 (3m) 876425 (3m) 
points: 187120 (6m) 49275 (6m) 300346 (6m) 393356 (6m) 443159 (6m) 
grid points: 75712 (3m) 23976 (3m) 122511 (3m) 87400 (3m) 97821 (3m) 
18928 (6m) 5994 (6m) 30806 (6m) 21850 (6m) 24455 (6m) 
points per mesh 6 (15pm) 6 (15pm) 6 (15pm) 8 (15pm) 8 (15pm) 
6 (30pm) 8 (30pm) 9 (30um) 16 (30pm) 16 (300m) 
# check points: 2123 1847 4375 3292 1778 
(photogrammetric) 
# check points: 135 135 - - - 
(tacheometric) 
Table 1 Geometrical and statistical parameters of the DEM test projects (values in brackets 
indicate the pixel resolution and the grid width, respectively) 
Project 1 
image scale : 1 : 7000 
  
Project 3 
image scale : 1 : 12000 
  
edges of total MATCH-T DEMs 
Project 2 
image scale : 1 : 14000 
  
Project 4 
image scale: 1 : 23000 
  
excluded arcas 
Figure 1: Image and test areas of projects 1 - 4. 
987 
2.2 Accuracy tests 
Objective assessment of accuracy performance requires 
independent checks on the results. Here, the DEM results are 
compared with independent measurements as obtained by 
tacheometric field survey and, for most parts, by conventional 
photogrammetric measurements with an analytical plotter. The 
tacheometric checks refer only to projects 1 and 2. The field 
measurements of 135 check points were compared with the 
corresponding heights which were interpolated from the DEM. 
The photogrammetric check measurements referred directly to 
the grid points of the DEM. 
The photogrammetric accuracy checks do not cover the 
complete DEM areas. Instead, representative smaller areas 
were selected, and diagonal profiles were measured in addition. 
Nevertheless, the tests can be considered quite comprehensive 
as they comprise in each project several thousand check points. 
The photogrammetric check measurements were done with the 
same original photographs the digital data were scanned from 
(except for project 2 for which the check measurements of 
project 1 could be used, because of common overlap). This 
implies the difficulty that the photogrammetric check 
measurements are not more precise than the DEM heights to be 
checked. In fact, they have about the same accuracy, or may 
be even poorer. 
The emprical accuracy evaluation is based on the r.m.s. height 
differences between the DEM heights and check heights. From 
these differences an estimate © ,,, is derived for the height 
accuracy of the automatically generated DEM, by taking the 
variance estimates of both components into account. If the 
digitally derived DEM heights and the photogrammetric check 
measurements are equally accurate, the DEM estimate comes 
down to dividing the r.m.s. height difference by V2.