The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
4.3 Assessment result of the DEM layer
Assessment against Japan Triangulation Network: There is a
triangulation network in Japan providing a set of geographically
accurate points for survey, mapping and so on. Various
attributes such as longitude, latitude and elevation of these
points are freely available through a web service from GSI. The
latitudes, longitudes and elevations of the 2207 points of the
triangulation network in the assessment area were manually
extracted from the web service and used as one reference data
set to assess the two Reference3D tiles (Figure 3). The black
dots in higher elevation area and white dots in lower elevation
area (Figure 3) indicate the positions of the points used.
Figure 3. Points of Japan Triangulation Network over DEM
layer of the two Reference3D tiles (N43E144, N43E145)
Because the points of the triangulation network are not aligned
with the grid vertices of the Reference3D, an interpolation is
necessary to superimpose the two data sets. The elevation of
each point of the triangulation network was kept unchanged,
and the correspondent value of the Reference3D at the point
location was calculated by bilinear interpolation of four vertices
of the Reference3D grid enclosing the point. The elevation
difference was calculated for each point, then from those
differences, standard deviation and mean value were calculated
(Table 2a).
Whole area
Flat area
Mean (m)
-2.2
-1.7
STDEV (m)
5.3
1.9
Max (m)
11.7
3.3
Min (m)
-91.7
-8.3
LE90(m)
6.7
4.1
Table 2a. the Reference3D compared against Japanese
Triangulation Network: general statistics
-92 -84 -76 -68 -60 -52 -44 -36 -28 -20 -12 -4
Figure 4. The histogram of the elevation differences between
Reference3D and the Japan Triangulation Network
To confirm the specification of the Reference3D, the linear
error at 90% confidence level was also calculated (Table 2a).
An agricultural area of low slope Was separately assessed (area
within the white rectangle in the Figure 3) and the result is also
shown in table 2a. The histogram of the whole area is shown in
Figure 4. Difference intervals were summarized in table 2b.
Difference
intervals
Whole area
2,207 points
Flat area
367 points
-5 / +5m
85.5%
98.4%
-10 / +10m
96.6%
100%
-15 / +15m
98.3%
100%
-20 / +20m
99.0%
100%
Table 2b. the Reference3D compared against Japan
Triangulation Network: error intervals
The difference intervals (Table 2b) show that the difference for
99% of the points over the whole assessed area is less than 20m.
In the flat area, the difference is less than 5m for more than
98% of the points.
Unfortunately there are very few points in the mountain area
(white part of Figure 3) so we did not assess mountain area
alone. There is no point of the triangulation network over the
two islands within the assessment area, so no assessment was
done for the two islands.
Cross evaluation between Reference3D DEM layer and GSI
DEM: To superimpose the two raster data sets of the
Reference3D (30m mesh, WGS-84) and GSI DEM (50m mesh,
Tokyo Datum), the common raster cell size, geodetic coordinate
system and matrix size (number of rows and columns) are
prerequisites. Therefore, at the beginning of this assessment, the
Reference3D DEM raster data was transformed to the geodetic
coordinate system of GSI DEM (Tokyo Datum) through the
sophisticated software (ENVI), and a bilinear interpolation
approach was used to resample the Reference3D DEM to the
same raster cell size and matrix size. Then by superimposition
of the corresponding samples, the difference between the two
DEMs was calculated as “Reference3D DEM - GSI DEM” (see
histogram in Figure 5).
Figure 5. Histogram of the elevation differences between the
Reference3D and GSI (whole area)
From those differences, standard deviation and mean values
were calculated for tiles N43E144 and N43E145 as shown in
Table 3 a. The Table also includes LE 90% as well as maximum
and minimum differences for reference. The cell with elevation
1315