Examples of these approaches are given by Kontoes et
al., (1993), Janssen and Middelkoop (1990).
In this study IF THEN rules are defined using expert
knowledge which relate land and soil information, in
combination with topographic information, with land
cover. These rules are implemented in Arcinfo and
applied to an already classified image of approach 1.
Again an error matrix and classification accuracy report is
made for comparison.
Alternatively, rules derived from the geo-data sets are
calculated using discriminant analysis in SPSS.
Discriminant analysis was selected since the variables
were not truly independent (Liyana Arachchi 1997). In
this way relations could be found between topographic
variables from the DEM and agronomic data from the
geo-database. Relations between actual group
membership and predicted group membership were
shown in a table. Since the accuracy was comparable to
the others no rules were derived for further consideration
in this study.
3. STUDY AREA
The study area is situated near the village of Alora, 20 km
North of the town of Malaga in the south of Spain. The
climate of the area is typically Mediterranean with a warm
dry summer and a cool humid winter. The area receives
an average annual precipitation between 400 - 700
mm/year and the annual average temperature is around 17
°C. There is a distinct dry period from June until
September.
À description of the Major Landforms of the area is given
in section 4.
A wide range of cover types, representative for
Mediterranean areas is present in the study area (Fresco
and Guiking, 1998). To reduce the number of cover types
and also aggregate them, we used a .thematic
generalization procedure (Molenaar, 1998). This resulted
in the following classes: (1) irrigated tree crops, (2) rain-
fed tree crops, (3) annual crops, and (4) semi-natural
vegetation.
The irrigated tree crops consist of citrus species like
lemon, orange and mandarins. The rain-fed tree crops are
olive and almond. The annual crops in the area are mainly
barley, wheat and chick peas, while also fallow land and
horticulture is part of this class. The semi-natural
vegetation consists of land cover types like oak trees, wild
olive, pine trees, elm along rivers and rock outcrops.
From a spectral point of view this last class, semi-natural
vegetation, is most diverse. Because of their small size,
roads and houses are not considered in the classification
process. Villages are also excluded due to difficulty in
discrimination. Therefore, to verify and improve the land
cover classification in this study, only irrigated tree crops,
rain-fed tree crops, annual crops and natural vegetation
are considered in detail.
344
4. INFERENCE RULES AND THE SOIL-LAND
SYSTEM HIERARCHY
Because soils are important determinants for land cover,
we can use inference rules based on expert knowledge to
improve land cover classification. To understand the
application of these rules it is best to conceive soils as
part of a multilevel nested system (figure 2). The higher
levels of that system represent terrain objects defined in
geological/ morphological terms. Going down in the
hierarchy, the terrain objects are subdivided and more
precisely characterized. At each step downward more and
more specific information is added as to the determinants
for the soil types. This in turn means that soil properties
are specified (predicted) with increasing detail
(Wielemaker et al., 1996).
The hierarchy as applied in the study area recognizes
three semantic levels:
e The major landform level: a repetition of form units. It
has a particular relief and position and finds its
meaning in a (main) origin and a common morpho-
structure reflecting lithologic nature and composition.
Landscape understood in geomorphic/ geologic terms
has a similar connotation.
e The landform element level: The most detailed geo-
morphic unit of which the major landform is
composed. It is a landform classified on the basis of a
particular genesis (process) and parent material
e The soil-scape level (facets): The land facet is a small
area of the land surface (minimum size is set at about
200 square meter) which is uniform in terms of
topographic parameters and surficial materials. It
follows then that variation in soil forming factors and
hence also in soil characteristics is so minimal that the
pedons or profiles occurring within its boundary
conform to the definition of the most detailed unit of
soil classification which is the series.
ATTRIBUTES ATTRIBUTES
MAJOR
LANDFORM . |MAINLY DENSE RIVER | SANDY PERMEABLE
CLASS HILLS jo AvEY FLATS | GRAVELLY [DEEP SOILS
MAJOR LAND FLUVIAL MARI. +
FORM CAPPED SANDSTONE
SUBCLASS HILLS HILLS
AGGREGATION
DISAGGREGATION
LAND FORM . : TIGE
ELEMENT EROSION] TERRACE MARI SANDSTONE
ab dus GLACIS REMNANT SLOPES OUTCROPS
FACET UPPER SLOPE LOWER SLOPE SCARP RIDGES VALES:
SOIL- IRAVELLY DENSE CLAY VERY MOD. DEEP DEEP LOAM
SCAPE WER CLAY SHALLOW | | SANDY LOAM
Figure. 2 Illustration of the three-level nested terrain unit hierarchy :
major landforms, landform elements and facets. Map and legend are
presented in figure 3.
Information from a lower level can be used to
characterize a higher level: fluvial -capped hills (HsC in
Figure 3) is a property derived from the study of the
landform element composition of the Hills (figure 2). In
fact the third level of the legend of the major landforms
Intemational Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998
