Figure 12.3. Mapping of some ecological structures in the Kwilu region, Zaire (SPOT XS 1988, band 1–
2-3, half resolution, minimum distance to mean classification). A, stream corridor—gallery forest; B,
environmental resource patch—open forest; C, stream corridor—palm trees; D, environmental resource patch—woodland savannah; E, matrix—pseudo-steppe.
The use of remote sensing (SPOT) 163
Landscape change
A multi-temporal analysis of processed satellite images permits the construction of variation curves which describe variation of landscape characteristics (e.g. patch shape, corridor width, matrix porosity, etc.) as a function of time, by three independent parameters: general tendency, amplitude of oscillation, and rhythm of oscillation.
Considering two images of the ‘same’ landscape taken within a certain time interval, and by looking at how each landscape element is maintained or replaced by another type of element, one can construct a transition matrix and calculate the replacement rate: the percent change for each of the conversions, based on the total number of points observed.
Those rates are important because changes in the structure of a landscape are related to changes in its functioning (Forman and Godron, 1988:428–445). A multi-temporal image classification makes it possible to follow the rate of regeneration of the vegetation affected by fire raising.
As a case study the focus is here on an open forest. On the images of 1987, the forest is clearly burned and a vast part of the forest is classified as fire patch. Using the same classification method on the data of 1988, only a part of the forest is classified correctly.
The part, that most affected by fire, is classified as steppe. This can be explained by the low degree of soil coverage because in this area the vegetation is not yet recovered. A follow-up for several years enables an estimation of the damage caused by the forest and steppe fires.
Belgium (Kempen)
Environmental settings
The second test site is the Kempenland in the north-eastern part of Belgium (Figure 12.4).
It is a rather flat area characterized by sandy soils where dune formation has taken place during the glacial periods. Originally this region was completely covered with oaks. The oak wood was cleared in a very early stage, and it was slowly altered in a moor landscape. In the late eighteenth century a vast part of this moor was planted with pines due to a lack of fuel and mine wood. During the nineteenth century many parts were (again) cultivated due to the appearance of artificial fertilizers.
As a result of the above changes, five landscape units can be distinguished:
1. an old agricultural land characterized by open arable fields;
2. a small-scale complex of arable land and pasture, characterized by an irregular shape of the parcels—many of these parcels are bordered by hedges and trees;
Landscape ecology and geographic information systems 164
Figure 12.4. Localization of the test site in Belgium. (Reproduced with permission from Monkhouse (1974:
Figure 6.1).)
3. a large-scale complex of arable land and pasture, with regular parcels bordered by hedges and trees;
4. units with small, long parcels mixed with small poplar planting; and 5. moor and woods.
The use of remote sensing (SPOT) 165
Detection and mapping
Introduction
The detectability of two different features, ‘land blocks’ and linear elements, is investigated. From a bird’s-eye view, one seems to recognize parcels in the landscape on the satellite images. A closer look at the SPOT images, however, shows that some parcels cannot be distinguished from each other, e.g. parcels with the same land use separated by barbed wire, furrows, etc.
The ability to recognize parcels is a matter of contrast between them. Therefore, the term parcel detection in the context of satellite image interpretation should not be used, and the new term land blocks is introduced: a surface surrounded by at least three linear elements. As linear elements we consider rows of trees, hedges, drainage ditches, roads, etc., or a combination of these ecological corridors.
All these patterns and structures can be detected because of clear differences in contrast, which can be enhanced by filters. The contrast can be caused by many different factors.
1. The occurrence of hedgerows—these linear elements are, in general, too small to be detected on the SPOT multi-spectral images as pure pixels. Differences in their reflection and their shadows, however, can generate mixed pixels.
2. Differences in land use and/or the phenological stages can cause differences in the spectral reflection.
3. Drainage ditches can sometimes be detected due to the effect of the groundwater table.
4. Roads and paths can cause the occurrence of some mixed pixels by their strong contrast with the vegetation.
In general, the corridors or the linear landscape elements can be detected due to their effects on spectral signature. In most cases, however, they occur as mixed pixels.
Results and discussion
Topographical maps at a scale of 1:10000 and updated during field surveys are compared with the edge-enhanced SPOT multispectral image in order to evaluate the possibilities and restrictions of satellite data for the detection of corridors and patches (land blocks), in relation to their size and shape.
The results can be summarized as follows. Land blocks with an area of more than 3ha are unmistakably detectable on SPOT imagery; those with an area smaller than 1.2ha cannot be detected. A transition zone occurs between 1.2 and 3ha (Figure 12.5). As far as the length is concerned one can say that the smallest size must be 300m. When the length is smaller than 130m it is not possible to detect the land blocks. A transition zone occurs between 130 and 300m (Figure 12.6). Concerning the width, the smallest size must be 120m. Land blocks cannot be
Landscape ecology and geographic information systems 166
Figure 12.5. Detectability of land blocks in relation to their area. 0, Not detectable; 1, detectable.
detected when the width is smaller than 80m. A transition zone occurs between 80 and 120m (Figure 12.7).
In terms of SPOT XS images, one can say that a block must be built up by 28 pixels.
The edges of the blocks will normally consist of some mixed pixels. This means that only 10 of the 28 pixels are pure pixels.
Figure 12.8 gives an overview of the detectability of land blocks in relation to their area and their length/width ratio. Two major groups can be distinguished: on the one hand, land blocks larger than 3ha (group 1), and on the other hand those with an area smaller than 3ha (group 2). All land blocks of more than 3ha can be detected (class 1B) except those with a length/width ratio exceeding 7 (class 1A). The
The use of remote sensing (SPOT) 167
Figure 12.6. Detectability of land blocks in relation to their length. 0, Not detectable; 1, detectable.
Figure 12.7. Detectability of land blocks in relation to their width. 0, Not detectable; 1, detectable.
lack of contrast due to an uniform land use plays also an important role for the undetectability of the land blocks in the class last mentioned.
Two trends can be seen in group 2 (<3ha). (1) Land blocks with a length/ width ratio larger than 4 are generally not detectable (class 2B). (2) Land blocks with a length/width
Landscape ecology and geographic information systems 168
ratio smaller than 4 form a transition zone (class 2A): two-thirds are visible, and one-third cannot be detected. It must be mentioned that classes 1A and 2B do not count enough elements to be statistically relevant, they only give an idea about possible relationships and correlations. Further investigation is necessary to accept or to reject the stated hypotheses.