The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
1314
specifications. We would like to present two results of the 4. ASSESSMENT OF THE REFERENCE3D TILES
assessments. OVER HOKKAIDO PROVINCE IN JAPAN
An across evaluation of the SRTM and the Reference3D was
conducted by NGA and IGN (Bouillon et al., 2006). Table 1
shows the full compatibility of SRTM DTED level 2 and the
Reference3 D products over various landscapes, from very flat
desert areas up to very high mountain areas.
Cell#
Figures @ 90 % (m)
landscape
Min
Max
Mean
Std.
Dev
e035n29
high relief
-5
15
4.4
4.0
e035n33
high relief
-10
4
-2.7
3.0
e048n32
flat + relief
-21
1
-7.3
4.4
high relief
-27
3
-9.4
6.0
flat terrain
-9
-1
-5.3
2.0
e049n30
flat + relief
-6
5
-0.4
2.4
high relief
-5
4
-0.5
2.2
flat terrain
-10
7
-0.5
3.7
e033n30
medium
relief
-6
10
3.7
3.3
e032n30
medium
relief
-6
11
1.5
3.7
e047n30
flat terrain
-9
-1
-4.6
1.9
e048n29
flat terrain
-12
-1
-6.7
2.5
e048n30
flat terrain
-10
1
-5.5
2.2
4.1 Assessment region
The assessment region is located in the east of Hokkaido
province. Two Reference3D tiles (N43E144 and N43E145)
cover the region from 144-degree E to 146-degree E longitude
and from 43-degree N to 44-degree N latitude (Figure 2). The
region includes mountains with maximum elevation 1400m,
volcano, lakes, agricultural land use, and sparse islands. In most
part of the mountains, the slopes are between 20 degrees and 40
degrees, and the maximum slope of the mountains is more than
50 degrees.
Figure 2. DEM layer of the Reference3D of the assessment
region (two adjacent Reference3D geocells)
Table 1. Cross evaluation of SRTM and the Reference3D
Figure 1 is another evaluation of the Reference3D performed by
Joint Research Center Ispra (JRC) and FOMI (Kay, Winkler,
2004). One important feature of this evaluation is the detailed
evaluations of the elevation accuracy, through segmented
classes describing i) the local slope (10 degree, 20 degree and
40 degree thresholds) and ii) the land use (agriculture, forest,
urban). All of them confirmed that the Reference3D accuracy
specifications were met at each slope class.
<10 10-20 20 - 40 >40
slopeclass
Figure 1. Elevation accuracy of the Reference3D, displayed for
difference classes of slopes and land uses (from Kay, Winkler,
2004)
4.2 Assessment method of the DEM layer
Two reference DEM data sets are used for this assessment. One
is a set of 2207 points from Japan Triangulation Network
maintained by GSI. These points have very high cm-level
horizontal and vertical accuracies maintained by the following
procedures: each point of this triangulation network is checked
from time to time, and if its errors in X, Y and Z are over their
thresholds (10 cm in X, Y and 20 cm in Z), then the data of that
point will be corrected by on-site survey.
Another is a raster data set of 50m mesh DEM produced by GSI
from the contours of GSI topographic map (1:25,000), and the
standard deviation of the contours of the GSI topographic map
is 5m. Additional error might be introduced during the
generation of the DEM raster data from the topographic map.
So the vertical accuracy of the GSI 50m mesh DEM is no better
than STDEV 5m. The GSI 50m mesh DEM is provided in
Tokyo Datum geodetic system with pixel spacings of 2.25
seconds in longitude and 1.5 seconds in latitude, corresponding
to approximately 50m.
To assess the Reference3D, the Reference3D DEM was
superimposed on the reference DEM data sets. Since the
Reference3D and the reference DEM data sets have different
geodetic system and different raster cell size, bilinear
interpolation was used to derive elevations from the
Reference3D tiles for comparison. Using the differences
between the elevations derived from the Reference3D DEM and
the reference data sets, the standard deviations, mean values,
maximum and minimum values were calculated. The linear
errors at the 90% confidence level were calculated to assess the
accuracy of the Reference3D.
