IG pour
servateur
evre, M,
parcours
> Africa-
pastorale
ection, p
s effets
erritoires
lellites a
1ontagne
IRV de
eille III,
modeles
:RMES,
n de la
Sahel.
aris, pp.
], Paris.
LAND COVER CHANGE DETECTION USING THE CORINE METHODOLOGY
G. Büttner, G. Maucha and M. Birô
FOMI Remote Sensing Centre, Budapest, Hungary
Commission VII, Working Group No.2.
KEY WORDS: land cover, computer aided photointerpretation, change detection, statistical change indicators
ABSTRACT
A methodology has been developed to update the CORINE Land Cover - Hungary database using computer assisted photointerpretation
(CAPI). The procedure requires the simultaneous analysis of original land cover database, the associated satellite image and a new satellite
image in a Geographic Information System (GIS). The methodology has been tested in retrospective analysis ("downdating") of a map-sheet of
the CORINE Land Cover-Hungary database. Results showed that careful decision on significant changes was important to characterise
processes of evolution properly. Thresholds were found for area and perimeter of polygons of the change databasg to guide an automatic
elimination of slivers. Areas of change, as small as 4 hectares proved to be meaningful. Between 1984 and 1992, 229 significant land cover
changes have been found, affecting 4.41 % of the total area.
1. THE CORINE LAND COVER DATABASE
The CORINE Land Cover (CLC) database is produced oy
computer-assisted photo-interpretation of satellite images, with the
simultaneous consultation of ancillary data (European Commission,
1993). The fundamental characteristics of the database are presented
in Table 1.
Table 1
Fundamental characteristics of the CLC database
Primary satellite imagery: LANDSAT TM, SPOT XS
Ancillary data (topographic maps, aerial photos, etc)
Working scale: 1:100 000
Minimum mapping unit: 25, hectares
Minimum linear feature width: 100 m
Nomenclature: 3 levels; 5 classes for level-1, 15 classes for
level-2, 44 classes for level-3
The CLC nomenclature has been defined for a comprehensive
characterisation of European landscapes based on physical and
physiognomic characteristics of the Earth’s surface. Since the
information on biophysical land occupation is dynamic, permanently
affected by modifications caused mainly by anthropogenic activities,
there are two major requirements concerning the CLC database:
e possibility of the quick updating in desired time intervals (5-10
years),
e availability of advanced techniques and systems for acquiring,
compiling and managing new geographic information.
The technology of updating has to be used for retrospective analysis
(called *downdating") as well, to provide CLC database for a
previous date.
2. UPDATING/DOWNDATING THE DATABASE
Some of the basic principles of the CLC methodology have
important features, which should have to be taken into account for
the database updating:
interpretation is not based exclusively on satellite images,
majority of the CLC classes are spectrally non homogeneous,
subjective delineation of certain CLC classes,
precision constraints.
Considering the above facts, automatic change detection techniques
(using standard image processing procedures) can be applied only
for a limited number of categories (e.g. water) in the updating or
downdating of the database. Guidelines for the computer-assisted
photo-interpretation (CAPT) for updating the CORINE Land Cover
database have been developed by Joint Research Centre (Perdigao
and Annoni, 1997).
Based on several methodological studies and on our own
experiences, the updating/downdating procedure results in many
cases the recognition of the geometric and thematic errors of the
original database. Therefore, the clear differentiation between errors
and changes are essential (PTL/LC, 1998):
e An eror always belongs to the original database. When
detected, it must be always corrected in the original database
first, and it will be subsequently introduced into the updated
database.
e Change is caused by real (physical) change of land cover type
in a particular area, and is identified in the new (updated)
database.
3. DOWNDATING THE CLC-HUNGARY DATABASE
The CLC-Hungary database was prepared between 1993 and 1996
in the frames of the Phare programme. Landsat TM satellite images,
taken in the period of 1990-1992 was used as basic data support. In
addition to mapping at the standard 1:100.000 scale, experimental,
1:50.000 scale database was also elaborated for selected parts of the
country (Büttner et al., 1995).
In preparation for a downdating project covering the whole country,
land cover changes between 1984 and 1992 was studied for a
1:100.000 scale sheet (48 km * 32 km) of the Unified National
Projection system in Hungary. The study area in NE of Hungary
around the town of Miskolc (map-sheet number: 88) was selected,
because of the variety of landscape types and the availability of :
International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998 685
