nly 
ine 
Jl. 
nly 
‘he 
to 
nly 
ers 
ith 
ure 
ind 
her 
rol 
ers 
ing 
‚it 
should be permitted to correct mistakes in the datasets. 
This is a rather delicate issue, because no clear guidelines 
can be given. Only experience and regularly monitoring 
can provide tools for efficient update management. 
6. EXPERIENCES 
One TOP10vector-dataset, based on aerial photography 
from 1995, has been compared with a previous edition, 
based on aerial photography from 1991. All files have 
been compared with a tolerance value of 40 cm. The 
number of changed features is given in figure 7. Changel 
contains changes in the GEOMETRY-files, change2 
contains changes in the ATTRIBUTE-files. The total 
number of changes of the dataset is also given. 
FILTERED-changes are not displayed in figure 7. The 
number of changed features for the FILTERED-deletions 
is reduced by 20% with reference to GEOMETRY- 
deletions and for FILTERED-additions, the reduction 
amounts to 30% with reference to GEOMETRY- 
additions. The reductions mean a significant acceleration 
of the checking process. 
Figure 7 shows that the total number of features has 
slightly decreased. Most updated TOP10vector-files have 
been increased in size, only the number of symbols has 
dropped enormously. This is caused by a change in the 
rules: it was decided to omit individual trees. Almost 
6,400 individual trees were deleted from the SYM-file. 
Rows of trees are still captured, they do appear in the 
PAT-file. 
A tolerance value of 40 cm was used, a value which may 
be reviewed. In case of change-only delivery, a tolerance 
can not be used as all changes have to be delivered. In 
this case, when no tolerance value was applied, the 
numbers of changed features in the BAS-files were 
increased by 100, which is only a fraction of the total 
number. Figure 7 then gives an impression of the size of 
change-only datasets for delivery to users. 
Looking at the number of changes, one might question 
the value of delivering change-only data-sets to users. It 
is expected that the average number of changes will drop. 
Three considerations support this expectation: 
1. The dataset was one of the first TOP10vector-datasets 
and contained therefore more inconsistencies than 
average. 
2. During the development of TOPlOvector the 
specifications were altered significantly. 
3. The number of field changes in the area is more than 
average. 
Generally spoken, about 45% of the dataset has been 
changed. The digitizing time for updating is estimated at 
50% with reference to the digitizing time of the initial 
TOP10vector-production. It is expected that this 
percentage will drop when more experience is gained and 
when the update process is fine-tuned. The reduced 
workload due to updating is compensated by the higher 
update frequency (from 4, 6 or 8 years, depending on the 
area to 4 years overall). 
  
  
80 000 
70 0007 
  
  
60 000 
  
50 000 
  
40 000 
  
30 000 
20 000 
  
  
0 
  
del add del add del add del add 
BAS AREA BLD SYM 
  
del add del add 
PAT TOT 
  
  
Figure 7 Number of feature changes 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996