International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004 
the total accuracy of the laser scanning is not as simple and as 
straightforward as it was thought. ..... It was observed that there 
is a flight line-dependent systematic and random error affecting 
on the total accuracy obtained. It was observed that the higher 
the flight altitude, the higher is the random error of terrain 
models. 800m flying altitude gives poorer results than 100m 
flying altitude. Laser measured heights are in general above the 
real ground surface. For asphalt surfaces a standard deviation of 
10cm is obtainable from H=550m and from lower altitudes the 
results are even better. A systematic error of typically 10 cm 
was observed due to observation angle changes.’ 
Abdullatif et al (2003) have also investigated the accuracy of 
LiDAR and report systematic errors, but an overall accuracy of 
about 12cm. Overall accuracy of LiDAR can be as good as 
10cm, but in practice varies according to the quality of the 
calibration and the terrain surface. 
Airborne IfSAR can achieve accuracies of 0.5m, but also varies 
according to calibration, altitude and terrain surface. Mercer 
(2003a) discusses the trade offs between accuracy and swath 
width and states that the theoretical accuracy from 30,000ft is 
0.45m and 0.30 m at 10,000ft. 
UCL has carried out an analysis of the Nextmap Great Britain 
dat which used two test areas (Dowman et al, 2003) and made 
use of LiDAR, GPS and aerial photography as reference data. 
The initial comparison of Kinematic GPS with the Nextmap 
DSM showed unexpectedly large errors, which turned out to be 
due to the effects of hedges and trees on the Nextmap due to 
the footprint size. These were removed by filtering in order to 
climinate outliers due to vegetation that bias the accuracy 
measures. The 3c threshold was used as starting point for 
filtering the difference data (KGPS minus Nextmap DSM). It is 
clear that points on the DSM are measured to be higher than 
their true value because of the size of the footprint of the 
Nextmap data. If the bare earth algorithm is effective, these 
errors should be corrected in the DTM, results are shown in 
table 3. It can be seen that a shift of between 0.3m and 0.8m has 
occurred and that this has therefore significantly improved the 
root mean square error. 
Photogrammetric check points collected from the stereo-model 
of aerial photography in open bare earth areas, clear of 
surrounding surface features within a 5m radius were compared 
with the DSM and the results can be found in Table 3. The 
Nextmap DTM and the photogrammetric checkpoints are in 
good agreement. A mean difference in elevation of -0.61m from 
the check points and a rmse of 0.83m was observed. 
Furthermore, the vertical accuracy of the Nextmap data was 
evaluated by comparing the Nextmap DTM bald earth surface 
with Lidar derived reference DTMs. Results of these 
comparisons are also listed in Table 3. The Nextmap DTM was 
subtracted from the reference DTMs (reference DTM minus 
Nextmap DTM. 
Two sub areas of open terrain type were selected and difference 
statistics produced. The Lidar DSM and the aerial photography 
DSM were chosen as a reference. Both, the Nextmap DSM and 
the DTM product were compared to the reference data sets. The 
results of these different comparisons are given in tables 4 and 5. 
The best accuracy of the Nextmap data is obtained over an open 
field, which is interpreted as bare earth, where a mean 
difference between the Nextmap and aerial photography is 
0.23m (Nextmap higher) and the rmse is 0.43m. The mean 
difference between the Aerial DSM and the Nextmap is 
effectively zero. This suggests that the bare earth algorithm has 
removed a mean difference of 0.23m in bare earth area. This 
corresponds to the finding discussed earlier, which also 
indicates that the bare earth algorithm affects the mean. This 
result needs further investigation. 
The Nextmap and Lidar surfaces are in good agreement in both 
the sub areas. Over a cropped area the Nextmap DSM has a 
mean difference of —0.61m and rmse of 0.77m, from the Lidar 
DSM. The Nextmap DTM has a mean difference of -0.38m and 
rmse of 0.48m from the Lidar reference DTM. 
  
Comparison Terrain Type | Land cover 
n Vmin Vmax VMean o[m] Rmse, [m] 
  
KGPS 3 vs. Nextmap 
DSM points > +1.5m 
Co flat) roads (bare earth) 
Mixed (hilly, | KGPS located along 
1994 -1.50 0.05 -0.95 [0.34 1.00 
  
  
KGPS 3 DIM vs | Mixed (hilly, 
Nextmap DTM flat) earth) 
KGPS located along 
road network (bare 
2647 -1.52 0.48 -0.66 |0.32 0.73 
  
KGPS6 DSM vs. 
Nextmap DSM points 
> +1.9m removed 
Mixed (hilly, 
flat) road network (ba 
earth) 
KGPS located along 
re 
1475 -1.85 1.00 -0.96 | 0.49 1.08 
  
KGPS 6 DTM vs.|Mixed (hilly, 
Nextmap DTM flat) cod UE (ba 
earth) 
KGPS located along 
re 
1568 -1.73 8. 
Un 
oo 
1 
e 
I 
0.45 0.47 
  
Air photo check points 
vs. Nextmap DSM Mixed 
Bald earth 
66 -1.66 0.43 -0.61 10.57 0.83 
  
Lidar DTM vs. | Mixed (hilly, 
Nextmap DTM (5) flat) Bald earth 
  
  
  
  
85362 |-9.20 12.04 |-0.22 | 0.10 1.01 
  
  
  
  
  
  
  
Table 3. Summary of results from NUI Nextmap DTM evaluation of Shrewsbury area 
Notes: All DEMs have 5m grid. KGPSi refers to ith profile recorded along roads. 
94 
Internatio 
NS 
[Sub 
Area _ 
to 
  
Table 4. | 
Sub Ar 
  
Table 3. / 
The areas c 
cover a flc 
about 60m ; 
cover types, 
up suburban 
The main co 
e “The ve 
to the 
and in 
that fo 
vertical 
it can t 
is high: 
of the s 
of veg 
measur 
somew 
from a 
objects 
contrib 
e When 
photog 
open te 
an rms: 
«The. be 
an ope 
mean 
photog 
(Nextn 
® The N: 
overall 
mean d 
Lidar [