The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008 
1318 
AX 
(longitude) 
AY 
(latitude) 
Circular 
Error 
Mean (m) 
3.3 
-2.1 
- 
STDEV (m) 
5.2 
4.8 
- 
Max (m) 
17.2 
12.2 
18.2 
Min (m) 
-10.7 
-11.2 
- 
90% (m) 
- 
- 
12.1 
98% (m) 
- 
14.9 
Table 4. Statistics on coordinate differences and circular errors 
of the 120 checking points 
The statistic of the 120 checking points (Table 4) shows that the 
circular absolute horizontal accuracy is less than 12.1m for 90% 
of the points, far within the CE90 16m specification of 
Reference3D. Actually the absolute circular error is less than 
15m for 118 out of the 120 points (98%). Table 4 and Figure 12 
show a horizontal bias of 3.3m to East and 2.1m to South. Once 
this bias is removed, the horizontal circular accuracy will be 
improved to better than CE90 10.2m (Gray circle in Figure 12). 
5. CONCLUSION 
The accuracy of the Reference3D product was assessed against 
the Japanese Triangulation Network, GSI DEM and 1:25,000 
scale vector maps. 
The comparison against the highly accurate DEM of the 
triangulation network shows that the assessed Reference3D tiles 
perform better than its specification, with a LE90 error of 6.7m. 
During the assessment against GSI DEM, Reference3D was 
transformed to Tokyo datum, and then the 1 arc-second (approx. 
20m by 30m) Reference3D DEM was compared to the 50m GSI 
DEM. The statistics of the vertical assessment show that the 
DEM of the Reference3D meets its specification. The 
assessment also shows that the Quality Masks is very useful 
when judging possible error sources for location with 
significant errors. With a smaller posting size than GSI DEM, 
Reference3D is a good alternative DEM solution over whole 
Japan or other areas for applications needing high consistency 
over large territories (e.g. for mapping needs). 
The horizontal accuracy of the Reference3D orthoimage, 
assessed over the Japanese territory against 120 checking points 
extracted from GSI 1:25,000 map, is absolute CE90 accuracy of 
12.1m. This shows that Reference3D is a very powerful tool to 
build highly consistent frame over areas that were once very 
difficult to map with a good accuracy, such as, by instance, the 
neighbouring Northern Islands. 
ACKNOWLEDGEMENTS 
The authors would like to acknowledge Mr. Fabrice Triffaut 
from Tokyo Spot Image for his assistance in providing the 
Reference3D tiles. 
REFERENCES 
Bouillon et al.,2006, SPOT 5 HRS geometric performances: 
Using block adjustment as a key issue to improve quality of 
DEM generation, ISPRS Journal of Photogrammetry & Remote 
Sensing 60 (2006) 134-136. 
Kay, Winkler,2004, Quality checking of DEM derived from 
satellite data (SPOT and SRTM), 10th Annual Conference on 
Control with Remote Sensing of Area-based Subsidies. 
Budapest, 24-27 November 2004.