zijing 2008
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Voi XXXVII. Part B4. Beijing 2008
363
ised by the
Iso by the
ted that the
:r than 1 m.
ìolds value
ients which
in table 4,
th an error
•>3
0,07%
0,01%
scover the
significant
iposed on a
in a legend
owed us to
the largest
are located
)undations,
¡flection of
RASS as a
>n instead
tie manual
aspect to a
rdinia and
itive value
ìypothesis.
intro 1 it is
ication
Tigure 4).
a logical
e a height
hat others
the same
Figure 4. Problems with wharfs or undeep water: many points
are classified as terrain
The control of the GRASS DSM was performed with a
comparison with Sardinia DSM. The areas in which the control
was performed are always the same already used in the previous
analysis.
To create a DSM with GRASS it is sufficient to interpolate the
points classified as first pulses. In this case we used bicubic
splines with resolution of 4 m and we choose a resolution of the
DSM always equal to 2 m.
The differences were calculated with the procedure already used
for the DTM and are shown in table 5.
area
mean
a
•< 1
<N
v
VI
CO
V
VI
<N
•>3
B
0,03
0,84
87,70%
8,17%
2,66%
1,47%
C
0,04
0,52
94,91%
3,63%
0,96%
0,50%
Table 5. Distribution of the DSM error (units in m)
The mean of the difference is almost zero but the standard
deviation has a larger value than in the DTM analysis. Anyway
the error is at least at 87 % lesser than 1 m and is due mainly to
the usage of two different interpolation methods.
4. DTM/DSM ABSOLUTE CONTROL
4.1 Dataset and procedure
The controls previously shown were performed with the raw
LiDAR data, starting from the filtering up to the interpolation
on a regular grid. They were based on the comparison with
another DTM/DSM. This analysis demonstrated a good
correspondence between GRASS and Sardinia results, but it is
not sufficient to check the absolute precision of the
implemented algorithm.
The goal of this paragraph is to show the results related to the
difference between gridded GRASS products and 218 points
measured with a GPS. The precision of the RTK GPS survey
was ±0.03 m, while raw LiDAR data have a precision larger
than ±0.15 m. This implicates that GPS coordinates become
useful to test the absolute accuracy of GRASS DTM and DSM.
The procedure to check GPS and GRASS data was based on
raster differences. Also in this case a GPS raster file was created.
After an analysis on the distances between each GPS point we
choose a resolution of the GPS raster map equal to 0.5 m, which
avoided overlapping between the GPS raster points. The map is
composed of only 218 full cells, and the remaining part is
completely empty.
The next step was to calculate the differences between the GPS
maps and the DTMs. The control was performed in 3 areas (2
areas have been already used during the relative control, in
addition we choose another area). To complete the tests also the
original Sardinia’s DTM was used in this comparison. The
results are shown in table 6.
num
mean (m)
a(m)
max (m)
min (m)
GPS-GRASS
area 16
88
0,21
0,17
0,47
-0,48
area 15
72
-0,01
0,14
0,26
-0,45
area 14
58
0,36
0,16
0,73
-0,19
GPS-Sardinia
area 16
88
0,24
0,08
0,50
0,11
area 15
72
0,21
0,07
0,39
0,08
area 14
58
0,27
0,14
0,57
-0,38
Table 6. Difference between GPS measures and grid products
The results using GRASS and Sardinia DTMs are close enough
but Sardinia DTM always presents a lower discrepancy. In any
case, considering the spatial resolution of 2 meter and the
complicated morphology of the chosen areas both results can be
accepted (we remember that Sardinia’s procedure is semi
automatic).
The last control regards the DSM precision. Also in this case the
control was based on the differences, but the points were
measured with a Total Station to obtain information also about
the point that lie on roofs (figure 5).
The main occurred problem was the position of the point used
to perform the control, because they are generally located next
to the building edge, where the DSM has a significant variation.
The roofs in the areas have typically pitched faces, and it is
obvious to forecast larger discrepancies than in the DTM
control, because an optimal and representative control must be
carried out in flat areas, where the spatial resolution of 2 m
becomes irrelevant.