Banzhaf, Ellen
three bands ia a first approximation to the atmospheric path radiance, and these minimum values are subtracted from the
respective images (Mather 1987). In both images band 6 is eliminated; a synthetic NDVI band calculated and attached.
Both classifications are calculated using the maximum likelihood classifyer with the non-parametric rule of the
parallelepiped optimization put first. A hierarchical classification needs to be generated as different settlement densities
and open pit mining are spectrally very similar, as well as fields without crops and unsealed ground (e.g. airport) are
difficult to be separated.
Tab. 2 Classified Land Use for the Region between Halle and Leipzig
Image Date and Sensor | 07.07.1989 TM-5 13.09.1999 TM-7 Change Detection
ATs ke [%] [%] [%]
Disperse settlement 6.4 12.9 +65
Dense settlement 3.5 4.2 10.7
Sealed area (e.g. roads) 2.0 72 +4.3
Area without green vegetation 12.5 6.5 -6.0
Fields with crop 34.4 29.8 -4.6
Green top and bush vegetation 10.1 16.7 + 6.6
Pasture and meadow land 20.6 9.9 - 10.7
Forest 77 10.2 +2.5
Water 1.9 2.6 +0.7
The change detection for this region is shown in the table above. It is obvious that especially disperse settlements and
sealed areas have increased at agricultural land's expense. As rather natural wetlands have remained under conservation
their share could augment. The quantified analysis is a first step to investigate land use changes but it does not show
structural modifications at this scale. Therefore a detailed classification is made for green spaces using IRS-1C data.
5.2 Binary Classification Concentrating on Green Spaces by Means of IRS-1C LISS Data
In this first classification phase a conventional, multispectral classification is applied to the IRS data. The produced
intermediate result provides a set of spectrally rather homogeneous landcover classes, and thus it is reliable to identify
landcover classes, like water or forest. A multi-step, hierarchical procedure is then undertaken, which was developed in
earlier projects, to classify both, satellite-based and airborne, multispectral scanner data (Netzband, 1998). In a first
step, an unsupervised classification (i.e. without signature analysis by the analyst) is executed which supplies 15
classes. These classes have to be assigned to land-use types by interactive, visual check and postprocessing or, if
necessary, aggregated. Furthermore, it is important to separate individual classes that are spectrally unique. The class
separation is performed by a multispectral, supervised classification in which each identified class is "extracted" by
masking it in the intermediate result, in order to exclude it from the following classification steps. For the classification,
a parallelepiped classifier is used. In this procedure pixels are not classified which do not belong to clusters of the
spectral signatures, and pixels in the overlap area of two clusters are classified according to the Maximum Likelihood
method. The resulting classes can be overlaid as masks on the finally resulting image and can be stored as independent
layers.
For the following calculation process especially two classes could be separated:
* Forest, stand of woods (larger trees),
* Allotments as well as grassland and meadow surfaces in the inner and peri-urban areas.
5.3 Calculating the Green Spaces according to the Ring-Sector-Model
To evaluate the green area distribution by classified satellite image data in the peri-urban area the so-called 'ring-sector-
model' is suggested. This space reference model was developed by Simon (1990) to analyse intra-regional occupation
commuter relations in Switzerland. It is based on the dimensional grid (same distances) of conurbations, by superposing
any number of concentric sets and sectors over a region.
The model guarantees that a uniform external limitation of different test areas is given. Additionally, it serves to
describe intra-regional characterisations of features. Gradients between the town centre and outskirts can be analysed
and quantified with the ring-sector-model in a differentiated manner.
122 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000.
