bul 2004
ereo-model
; Clear of
> compared
ble 3. The
ints are in
61m from
| data was
rth surface
of these
DTM was
IM minus
difference
tography
DSM and
a sets. The
les 4 and 5.
er an open
a mean
graphy is
The mean
extmap is
orithm has
rea. This
hich also
ean. This
nt in both
SM has a
the Lidar
).38m and
[m]
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004
Comparison
Area p
Sub Terrain Land
Type cover
Rmse,
n Vmin Vmax VMean o[m] [m]
Lidar DTM (5m) vs. Nextmap | hilly
DTM (46-61m)
crops
11484 |-1.55 |0.32 -0.38 10.30 0.48
Lidar DTM (5m) vs. Nextmap | Flat
DTM (49-53m)
Do
crops
2166 -1.14 10.04 |-041 |0.1 0.46
Table 4. Lidar compared to Nextmap DTM (selected areas)
: Terrain Land
Sub Area Comparison Type rana n Ymin Vois: Vues lem] E.
Aerial DSM vs. Nextmap | hilly ; 11484 10.01 t A Z
DSM (46-61m) crops -10. 1.094 |-0.31 |0.44 | 0.54
Aerial DSM vs. Nextmap | hilly Tm
l DTMS (46-611) crops 11484 |-0.97 0.54 -0.07 10.20 10.21
Aerial DSM vs. Nextmap | Flat
2 3 2 ^7
2 DSM (49-53m) crops 2166 1.38 0.99 0.23 |0.37 [043
Aerial DSM vs. Nextmap | Flat i
2 DTM (49-53m) Crops 2166 -0.54 0.37 -0.00 {0.17 (0.17
Table 5. Aerial photography compared to Nextmap DTM (selected areas)
The areas chosen for validation in Shrewsbury and Worcester
cover a flood plain of the River Severn with some relief of
about 60m above the river. It also contains a variety of land
cover types, including bare ground, crops, woodland and built
up suburban areas.
The main conclusions are as follows:
e The vertical accuracy of the Nextmap data varies according
to the type of the terrain where the comparison is made,
and in particular the land cover. For example, it is known
that forest and dense urban areas significantly decrease
vertical accuracies of Digital Elevation Models. In general
it can be stated that the mean surface of the Nextmap data
is higher than the reference data. This is expected because
of the size of the Nextmap footprint, and the general effect
of vegetation in IfSAR measurements. The elevation
measured for any IfSAR DSM sample (square footprint
somewhat larger than the 5m DSM sample distance) result
from a combined signal of scattering objects. Thus, raised
objects such as trees and hedges located in the sample area,
contribute to the elevation value measured.
e When comparing the Nextmap DSM with
photogrammetric checkpoints, which were measured in
open terrain, a mean difference in elevation of -0.61m and
an rmse of £0.83m were observed.
* The best accuracy of the Nextmap DTM is obtained over
an open field, which is interpreted as bare earth, where a
mean difference between the Nextmap and aerial
photography was —0.001m and the rmse was +0.17m
(Nextmap higher).
* The Nextmap and Lidar surfaces are in good agreement,
overall. Over a cropped area the Nextmap DSM had a
mean difference of —0.61m and a rmse of £0.77m, from the
Lidar DSM. Over the whole area the Nextmap DTM was -
0.22m difference with a rmse of £1.01m, from the Lidar
DTM.
e The GPS profiles along the roads show good agreement
with the Nextmap DTM data when the effects of trees and
hedges have been removed.
The evaluation of the Nextmap Great Britain data reveals a
number of characteristics of the IfSAR and LiDAR data, and of
the filtering techniques, which are general to this type of data.
[t also reveals unexplained differences which need further study.
7. VALIDATION
The validation of any DEM is clearly very important. This
however can be quite difficult and expensive when the precision
of the product is so high, and interpolation is necessary in the
process. Some operators have high confidence in their product
and do not consider validation necessary. The quality of the
positioning can be checked from the GPS record. The normal
procedures for validation include the use of reference data such
as check points located as targets or points on open surfaces,
reference DEMs or profiles. Checks can also be made for
consistency and for outliers, and correlations can be
investigated between the data and vegetation or slope. Where
the point density is high enough, targets provide an very good
validation surface for LIDAR. The Highways Agency (HA) in
the UK specifies that boards, 1.2m x 1.2m, centred over a co-
ordinated point and accurately levelled must be used. The HA
requires a point density of 7 — 10 points per m°, and thus about
10 points are expected to fall on the board, allowing significant
statistics to be generated. An illustration is shown in figure 6.
For IfSAR this is not usually appropriate but traditional corner
reflectors can be used. Because of the footprint size of IfSAR
comparisons over large areas or profiles are better suited.
Kinematic GPS profiles along roads have proved to be very
useful for checking both LiDAR and IfSAR. Examples are
given in Morley et al (2000) for the Landmap project and
Dowman and Fischer (2003) for Nextmap UK.
