2004
————
) are
Ive a
ent to
erally
s for
tified
mage
z the
mage
tions
s the
| map
ontal
4).
cally
and
aphic
bling
ional
aphic
'd on
thout
| and
)onse
1etric
mage
ages,
1ceds
1age)
vable
'oach
ution
' the
the
on -
jetric
n be
>nsor
netry
nap).
'ased
st to
ellite
10del
their
and
n) of
(the
1sing
d are
level
ends
es of
have
ssing
1odel
:
delivered with image) image
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
Model GCPs Remarks
Rigorous (physical) Usually 8 | IGM/RPC file is
satellite model for all ignored but
three information from
sensors metadata file are
required
Effectively repeats
work done by
Rational Functions (RPC) Many
calculated from user GCPs GCP's
required, image provider
usually
>20
Image geometry model No Limited geometric
(IGM / RPC delivered with | additional | accuracy
image) GCP's
Refined image geometry 2104
model (IGM / RPC GCPs per
Table 2. 3D geometric models and GCPs requirements.
The first and last options make up a reasonable and cost
effective choice where high accuracy of final product is
intended.
Several published papers (Kay et al., 2003, Cheng et al,
2001) confirm quite good results using RPCs refined by a few
GCPs. This refinement adjusts the RPC increasing
geopositional accuracy of the image, improving co-registration
between the image and DEM, thus ensuring more accurate 3D
geometric rectification.
Of course, the GCPs (and also check points) need to be
carefully selected and identified in the image before their
accurate surveying in the field. Attention should also be drawn
(during the ground point selection) to the fact that in
orthorectification distortions due to topography and varying
Earth surface heights are removed; however, man-made
features standing on the terrain surface like buildings, bridges
or channels will have significant displacement not modelled in
the DEM, and thus cannot be used reliably for GCPs.
3. THE VHR VALIDATION PROGRAMME
3.1 The aims of the programme
With regard to information content and potentially achievable
positional accuracy, VHR satellite images are considered to be
extremely useful for CwRS and Land Parcel Identification
System (LPIS) applications in the CAP. In this context, the
images (after orthorectification) are utilized to determine the
area and land use of declared agricultural parcels. The
importance of VHR satellite images, their potential usefulness
and advantages are also considered from the perspective of
CAP reform process and its control, monitoring, and
management requirements.
The study managed by the European Commission in
parallel to its operational work related to CwRS campaign, was
intended as a broad validation programme of VHR data use
with the main goals described as follows:
* establishing the realistic geometric target specification for the
different data types concerned, initially assumed to be 2.5m
RMSE; (EC, 2003)
* determining the performance of orthocorrection systems in
relation to the different site conditions experienced (ancillary
data quality), -
* confirming the potential of existing platforms to deliver
reliable and timely data to:
a) the CwRS programme,
b) other CAP applications (IACS) in general.
The detailed technical tasks include identification of factors
influencing the geometric quality of the final orthoimage
product, as well as testing different approaches and models used
for the orthorectification.
The VHR images were orthorectified bv the respective
contractors responsible for each site in the operational context
of the 2003 CwRS campaign purpose. A brief protocol with
recommendations for ancillary data quality was circulated, and
technical assistance offered on an ad hoc basis. The raw and
orthorectified images, together with ancillary data, were
delivered to JRC, where additional orthorectification tests and a
final evaluation were done.
3.2 The study area and data
The sites, located in a wide range of geographical conditions in
different European countries (Fig. 1), represent a quite diverse
set with respect to:
* terrain (relief) characteristics,
* operational, production conditions,
* ancillary data source and quality.
12 sites were covered by Ikonos images (Geo Ortho kit)
with view angles up to 20.15?, 17 sites by QuickBird (Standard
Orthoready) with view angles up to 14.2?, and 3 sites by EROS
images (Level 1A) with view angles up to 20.18°. The majority
of sites were covered by single scene images, but some sites
were covered by 2 or more overlapping images.
The ground control points and independent check points
were acquired in some countries through GPS-survey (accuracy
of 1m or better), while for others the points were obtained from
orthophotomaps (scale 1:5000) or digital vector maps (with
reference scale 1:2500). DEM cell sizes also varied, i.e.
between 5m and 50m, although according to reported metadata
all were equivalent to a RMSE;z of «5m, this being the
recommended maximum uncertainty. Although the number of
GCPs required was defined by recommendations, the number
delivered for some sites was fewer than expected, decreasing in
such cases the flexibility and possibilities of additional variants
of orthorectification planned to be done at JRC. In addition, due
to some incompleteness of ancillary data, not all sites are
included in the set of results described in this paper.
4. APPLIED METHODS AND PROCEDURES
4.1 Orthorectification of images
Raw and orthorectified images, delivered by contractors with
ancillary data (DEM, GCPs, check points, metadata), were
selected for further work done at JRC, which included:
creation of meta information concerning all the images,
ancillary data, sites, and also successively obtained results,
orthorectification with different options (other than chosen by
contractors),
comparison and evaluation of two types of orthorectification
results:
- delivered by contractors,
- done at the JRC (other variants).
Additional orthorectification done at JRC (cf. by
contractors) assumed different, pseudo-randomly chosen
options for processing, taking into account: number of GCPs,
type of 3D rectification model, other software. Consequently,
for sites with multiple images, more results were produced. The
final set of results can be considered as diverse and quite
representative, as well as useful for advanced factorial analysis.
.
1021
