In the case of Finland (Jaakkola, 1994) and Great Britain
(Wyatt, Fuller, 1992), the mapped results will serve as a
basis to produce the national CORINE land cover map
by applying suitable map generalisation procedures. The
maps should be completed in 1997 and will constitute
the challenging task of combining a maximum amount
of automatic processing with the CORINE legend
requirements. In the framework of Cooperation in
Science and Technology with Central and Eastern
European Countries a cooperative project: the
"Application of integrated methods for the monitoring
and evaluation of natural and cultivated landscape
vegetation for status, stress and drought, using remote
sensing (with participation of Telespazio/Italy, EU JRC,
FÓMI RSC/Hungary, OPOLIS/Poland, Czech Technical
University/Czech Republic) is being implemented. The
first results are connected with crop monitoring while the
next phase will focus more on the natural vegetation
monitoring.
Our own experience at the Joint Research Centre - on
land cover inventory and mapping at a scale of 1/50.000
and for complex and variable physiographic conditions
(Megier et al, 1991; Hill, 1993) suggests that at the
present stage and especially for demanding conditions
which are often encountered in Europe, a compromise
has to be found between reproducibility and large scale
spatial consistency obtained with automatized computer
processing and a more detailed legend affordable by
visual interpretation. The necessity for consistent
radiometric ^ preprocessing and calibration of
multitemporal imagery (Hill et al.,1995) - to accumulate
information throughout the vegetation growing season -
has also to be counted in. Future developments in
sensor technology and computer-based image
understanding might of course alter the present
situation.
3. Low resolution mapping at continental level
Mapping by remote sensing over the whole of Europe,
from Portugal to the Urals, was first performed in 1992
(Pseiner et al.,1992) for producing a digital forest map of
Europe with only two classes ("forest', "non-forest’) at a
maximum scale of 1/1.000.000 by using NOAA-AVHRR
data (1,1 km ground resolution).
A number of exercises have started meanwhile, aimed
at producing similar scale, large extension land cover
maps over Europe using AVHRR data (Veldkamp et
al.,1995), but the results are not yet available. The
anticipated objectives range from global land cover
monitoring to agro-meteorological modelling and global
climatological assessments.
The use of these data bases must naturally be
compatible with the reduced number of broad land cover
classes technically affordable in this context (four to six
maximum, excluding inland surface water). However,
even such a modest legend requires the implementation
of a systematic use of multitemporal satellite data over
the whole growing season (typically, one coverage per
86
month from March to September or October) in order to
reach the required potential of class discrimination.
Preliminary geometric and radiometric calibration of the
data are thus mandatory although not at all a trivial
problem, especially for the latter.
The European AVHRR land cover map at scale
1/1.000.000 undertaken at the JRC Ispra well illustrates
the above mentioned requirements and problems. The
first rather extended results on Belgium, France,
Germany and The Netherlands are presented in more
detail in this congress (Hoffmann,C.: The fusion of GIS
information and remotely sensed data for mapping
European scale land cover, Com.IV. W.G.1.) They are
obtained with 4 land cover classes (built-up, sparse
vegetation, vegetation, cropland) but forest will be
separated from vegetation in a second step.
The AVHRR data have been previously stratified into 13
ecosystem regions and are then processed
independently on a regional basis (European
Commission, 1995). 68 relatively cloud-free AVHRR
mosaics have been used and reduced to eight monthly
maximum value composites from March to November
1995. Each month NDVI and surface temperature
values (Ts) are used together in an attempt to more
effectively discriminate the regional land cover classes
(Hoffmann C., Roy D., Stein A., 1995; Roy D., Kennedy
P., Folving S., 1996). The first accuracy assessments
indicate a high degree of consistency with the high
resolution mapped results on eight test areas of 40x40
km extension and a reasonable correlation with
EUROSTAT regional statistics of soil occupation
together with a constant spatial consistency of the class
labels across the boundaries between the various
ecosystem regions considered.
The trade-off on using this type of low resolution data
obviously lies between the limitation of the broad class
legend achievable and the possibility of easily mapping
and updating extensive areas up to continental level,
although the near availability of the "Vegetation"
instrument on SPOT 4 (1998) will improve the class
discrimination potential, due to the presence of the
middle IR band around 1,7 mm.
References
Applications, Environmental Mapping and Modelling
Unit, E.C. Contract No 5609-93-11 ED ISP F, pp.12-
69.
Bittner et al. The CORINE Land Cover - Hungary
Project. EN&IN Conference, Budapest, 1995.
Cornaert, M., Maes,J., 1992: Land cover, an essential
component of the CORINE information system on
the environment. GIS implications, European
"International Space Year" Conference 1992, Munich,
Germany, pp.473-481.
EC: CORINE Land Cover, Guide Technique, EUR
12585, 1993.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B6. Vienna 1996
Mr 2 b
— (f) n
Tin ———0OU0 0n (€)
f rtf
