International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004 
  
already existing orthophoto product at the NGI. In order to 
combine the advantages of a grid approach (volume, easy 
transferability to the users of rasterdata) and a TIN-approach 
(preserving the linear structures) a hybrid form is envisaged. It 
should however always be possible to derive differently 
structured DTMs directly from the DEDSs, for example in case 
we want to work directly on digitized contour lines as described 
by Mizukoshi and Aniya (2002). It is therefore of paramount 
importance that the DEDSs alone are considered as the core 
data. 
On a third level we can envisage all kinds of derivative products 
such as contour lines, viewshed analysis results, ...). 
2.7. Available data sources 
The following data sources that contain information on terrain 
height are available for the production of a better DEM: 
e Points at ground surface and structure lines that were 
stereoplotted on aerial b/w photographs on 1:20.500 specifically 
for DTM production (DTM10.000 typel), available for 17% of 
the country but more will become available through co- 
operation with the regions. 
e Points from airborne laser (about 1 
point/25m”)These are at the moment only available for 160 km? 
but in due time much more may become available through co- 
operation with the regions. (The derived DTM is known as 
DTM 10.000 type 2) 
e Points at ground surface measured by terrain survey. These 
are available for a small number of very flat areas. The accuracy 
is very good (RMS 0.4m) but the data do not include points on 
linear terrain features. Addition of linear elements (e.g. dikes) 
could remedy this. (The derived DTM is known as DTM10.000 
type 3) 
e Contour lines that were stereoplotted using aerial b/w 
photographs on 1:20.500 and 1:10.000 for flat areas. Both were 
extensively checked in field surveys. These are available for a 
very large part of the country in vector format. Their general 
quality is quite good but locally gaps and blunders are present. 
(The derived DTM is known as DTM 10.000 type4). 
e Linear elements (embankments, hydrographical elements, 
.) in 3D that were stereoplotted for the production of a 
topographical map. The elements that are at the ground surface 
are a very interesting additional source of data for the build-up 
of a countrywide DTM (Ruiz,2000). Before integrating them, a 
selection and thorough control of the z-values is necessary as 
the emphasis in the past has been more on their planimetrical 
position. 
* Points from image correlation on aerial b/w photographs 
on 1:20.500 and 1:29.000. These are only available for a small 
part of the country but can be quickly derived for almost the 
whole country, as it is completely covered with the necessary 
aerial photographs. 
scanning 
Depending on the type of area a different kind of data source 
provides the optimal (within practical constraints) solution. 
Factors that are decisive in the choice of data source are the 
availability of the data source, the type of terrain, the amount of 
processing that has to be performed on the data source, the 
urgency of demand for data in an area... Sometimes a 
suboptimal solution needs to be chosen as a stopgap which will 
be replaced with a better solution afterwards. 
uA 
2.8.Quality issues 
2.8.1. Accuracy. There are two reasons for a thorough 
statistical quality analysis. On the one hand it is needed as a 
guide for the upgrading of the DEDS. As such it is really a tool 
and not so much an end in itself. On the other hand it is 
necessary to give the users of the data (or derived DTMs) 
metadata which they can actually use to estimate the usefulness 
of the data for their application. Statistical results on the 
accuracy of discrete elements contained in the DEDS should be 
clearly distinguished from statistics on the accuracies obtained 
at random locations in the DTM. 
We are setting up a separate DEDS which will only be used as a 
source of control points. The height accuracy is in the cm-range. 
Here again a policy of recovering past efforts is pursued by 
filtering suitable points at the ground surface from the 
thousands of control points used for aerotriangulation purposes 
in the past 15 years. 
Preliminary results show that all four DTM10.000 types yield 
significantly better results than the existing DTED level 2 
model. E.g. table 1. 
  
Flat terrain (H:3-10m) Hilly terrain (H:60- 
230m) 
DTM10.000 DTED DTM10.000 DTED 
type3 v2 typel lv2 
Mean -0.17 -1.30 0.39 -0.32 
Variance 0.09 0.67 0.63 4.28 
Deviation 0.29 0.82 0.80 2.07 
RMS 0.41 1.59 1.06 2.30 
  
Table 1: statistics on the height differences between check points 
measured in the field and values derived from a DTM. (H = terrain 
height) 
A visual inspection of the data by an operator is a step that is 
labor intensive but that cannot be neglected. The real bottleneck 
in a production environment is always this phase for which 
there is however no valid alternative. Even for very automated 
and homogeneous techniques such as laser scanning, the 
expenditure needed for a good data verification is still 
significant (Artuso et al, 2003). Visualizing the DTM as a 
shaded image is a very effective way to quickly spot possible 
problems (see also Dupéret,1999), although other visualization 
techniques are also useful. A mixture of different techniques, 
preferably by different operators is still the best quality 
assurance policy. The visualization also gives a subjective, quite 
intuitive impression of the quality. A comparison of figure 1 
and figure 2, both showing the same area gives a good 
impression of the improved height resolution of the model and 
hints at the better modeling of the terrain. 
  
Int 
as 
Du 
wi 
tor 
ou!