4
|
Correcting Further corrections in
CLC100/1992 using |—»| CLC100/1992 based on
CL C50/1992 satellite images of 1984
Producing CLC100/1984
by modification of 4—
CLC100/1992
|
of Si
changes /
A — X
By product: 2 N
improved
CLC100/1992
Final product:
CLC100/1984
Figure 1: Methodology of change detection and updating
archived Landsat TM satellite images. The practical work has been
carried out using ArcInfo softwzze running on a UNIX platform.
3.1 Data collection
Five Landsat TM satellite images (scene 187-27) were selected,
taken on different dates to perform database verification and change
detection. Three images taken in 1991-1992 were used in making the
original CLC100 database, while two images were chosen from
1984 to derive the database for a previous date (Table 2). Twelve
maps at scale 1:50 000 in Gauss-Kriiger projection covering the
study area and the experimental 1:50 000 scale CORINE Land
Cover database (CLCSO) helped the interpreter in identifying
interpretation errors and possible changes on land cover.
Table 2
Landsat TM images used in the study
year month, day | function remark
1984 | July 15 creating partly clouded
CLC100/1984
1984 | July31 creating partly clouded
CLC100/1984
1991 October 7 | creating partial coverage only
CLC100/1992
1992 | July21 creating excellent quality
CLC100/1992
1992 | March 8 verifying partial coverage only
CLC100/1992
3.2 Data preparation
Correct geometry is an important requirement of reliable change
detection. Satellite images used in making the original CLC100/92
database were registered to topographic maps with an accuracy of 25
meters (RMSE) (Büttner and Maucha, 1995). To register images
taken in 1984, an image-to-image rectification was used, with a
RMSE of 14.5 meters.
To support this case study, specific tools for image comparison and
database manipulation have been developed under UNIX ArcInfo
environment, which provides the photointerpreter with the following
facilities:
e spatially linked dual-window environment
e simultaneous display of vector and raster data in both windows
e simultaneous vector editing capability.
3.3 Database updating
The main stages of the updating methodology are presented in Fig. 1.
Using the unique occasion to have independent land cover data at
two different scales (CLC100 and CLC50), the two databases were
overlaid and the differences analysed in order to improve the
CLC100/1992 database. Typical modifications were:
e supplementing omitted pastures,
e better separation of vineyards [221] and complex cultivation
[242] categories,
e improved differentiation among categories of artificial surfaces.
Further systematic corrections were introduced into CLC100/1992
by simultaneous visualisation and analysis of the database on the top
of satellite images taken on 1984 and 1992. Typical modifications
made at this stage:
e improved differentiation of active and passive gravel extraction
sites along rivers (active site coded as mine [131], passive site
coded as lake [512])
® refining interpretation of forest clear-cut [324] areas.
The downdating procedure included a set of decisions for all the
polygons of CLC100/1992 to create CLC100/1984 database (Figure
2). Obviously, the CLC100/84 was obliged to fulfil the
fundamental characteristics of CLC100 database, including the
size of minimum mapping unit (see Table 1). After analysing
changes and filtering out impossible or improbable transitions
further modifications and corrections were performed in both
databases in an iterative way. As a result, CLC100 database for
the new date and the corrected version of the original database
was obtained (Figure 1).
4. ANALYSING CHANGES
Having the CLC vector database for two dates, change detection is
an automatic procedure. The change database is created by
intersecting CLC100/1984 and CLC100/1992. Each polygon has
two attributes: CLC codes related to date T, (1984) and T» (1992).
686 International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998
A
mt
—
