The Use of GIS as a Decision Support Tool for Evaluating Fragmentation by Roads

Carola Stauch
Institute for Landscape Planning and Ecology, Stuttgart University, Keplerstr. 11, 70174 Stuttgart, Germany

1. Introduction

Traffic has increased dramatically in the last decade, traffic induced by heavy goods vehicles by 54 percent since 1980 and traffic induced by personal motor vehicles by 46 percent since 1985 (EUA, 1998). Even though technology is able to reduce more and more pollution by air and noise, the ecological problems such as fragmentation by roads still remain.

Regional planning processes are aiming for environmentally beneficial mobility. Planning is a decision support process in which almost every step within a planning process contains an evaluative part, so that there has to be a comprehensible scale for evaluating these data. Regionalized ecological guidelines should be set up in order to serve as a scale.  ATKIS, the federal topographic-cartographic information system of Germany, is used to establish an evaluation scheme for the impact of road construction which takes the different priorities into account.  The study area, the Region of Stuttgart in Southern Germany, can be subdivided into categories of different types of urbanization, which can be automatically derived. It is shown at the example of two different guidelines how the evaluation of roads can differ depending on the urbanization category.

2. Spatial Decision Support and Planning

"Planning and management are based on a generic problem solving process which begins with problem definition and description, involves various forms of analysis which might include simulation and modelling, moves to prediction and thence to prescription or design which often involves the evaluation of alternative solutions to the problem" (Batty, 1996). One type of uncertainty which is typically involved in such planning problems is the formulation of goals and their relative weights (Arentze et al., 1996; Densham, 1991). The decision between one site or the other is not only a scientific question anymore but depends on values of the society, ecological, moral or political concepts, etc.

Taking the impact analysis of roads as a planning example, there are several types of analysis imaginable. This could be

Roads act as a disturbance source and produce fragmentation. Fragmentation is a detachment or separation of spatially connected landscape elements into spatially segmented tracts. Causes of fragmentation are disturbances defined as "an event that causes a significant change from the normal pattern in an ecological system" (Forman, 1986), which can be a single or a chronic event, a natural or anthropogenic disturbance. Fragmentation can be divisive cutting a landscape element in halves by a disturbance corridor. Roads can be regarded as chronic, anthropogenic disturbance source for divisive fragmentation (Bell, 1995).

There exist three dimensions which influence the result of an evaluation of divisive fragmentation. These are (1) the ecological goal or guidelines which determine the selection of the criteria and the evaluation, (2) the set of alternatives if a new site is planned/the evaluation only of the existing site/the set of measures if an existing site is evaluated and finally (3) the different multicriteria evaluation techniques.

The following shows how the impact analysis of a new road can be approached. After identifying several levels of spatial units in a hierarchy, the top level is further elucidated. It is proposed to distinguish between the evaluation of differently urbanized areas. It is shown how the urbanization category can be automatically derived. An example demonstrates how these categories influence the evaluation.

3. Structural Approach

"Management and planning for sustainability at an intermediate scale, the landscape or region, appears optimum" (Forman, 1997, p. 488).  Therefore, a scale of 1:25.000 to 1:50.000 was chosen. The evaluation of divisive fragmentation is done in a hierarchical process (Stauch, 1998). A top-down approach  distinguishes two main levels. The type level determines the threshold values for the ecological guidelines of the planning process and therefore serves as the evaluation scale (Fig. 1). The spatial unit is the landscape which in this case is defined as the Natural Region. There exist 12 Natural Regions within the study area, the Region of Stuttgart, which is an administrative unit of 3600 km². The Natural Regions were delimited by such features as geology, geomorphology, climate, biotopes (Meynen & Schmithüsen, 1953). The type level describes the landscape; threshold values are set up for ecological guidelines which have to be met. These guidelines can comprise ecological targets and standards of a region, which can range from legal pollution levels to the restoration of natural shores of the rivers. It is important, that they are expressed as explicit as possible, in order to elucidate the decision process.

At the object level the impacts of roads are measured. The object level can be subdivided into sublevels: at object level I the impact of the road on the catchment area is described, at object level II the impact of the road on the immediate surrounding is specified. At this level the spatial unit is the road which is an aggregation of road sections (Fig. 1). The catchment areas of the road in question (object level I) are investigated by determining for example the crossing of rivers by roads or the parallel running of roads within a critical distance. The road (object level II) can be evaluated for example by the area in which certain pollution levels are exceeded. The remaining paper focusses on the type level.


Figure 1: Hierarchical Process of Criteria Selection

4. Type Level: Establishment of Ecological Guidelines

Conservation of resources without goals cannot be successful in the long run. Goals are needed as a measure for every assessment or analysis.  They have an integral approach and consist of objectives of many sciences, and provide concrete ideas. A goal is a conglomerate of a historical analysis, the capture of the present state and aims of the society, discussion between experts, users and representatives of specific groups. Part of a goal should be ecological guidelines, such as the general demand of sustainable mobility, as well as concrete standards such as pollution levels. Environmental quality objectives, ecological goal functions, etc., are indispensible elements for every evaluation procedure (Wiegleb, 1997). The ecological guideline is dependent on the priorities for nature conservation for this region.

Strategies and aims of nature protection are often transferred from the open space onto urban area (Breuste, 1994). But nature in open space and urban nature have to be subdivided because they follow differing goals. Therefore a categorization of the Natural Regions is carried out into urban, suburban and rural. Later the evaluation priorities have to differ between these three categories. For example a parc area in a metropolitan area has a much higher priority than a parc area in a rural setting.

The remaining paper will show how this categorization can be derived by using GIS. Then rules are developed to satisfy two ecological guidelines, which are

     Figure 2: Type Level

It is suggested to apply these rules depending on the type of landscape, i.e. a more urbanized or more rural landscape. The division of regions into different categories with regard to their urbanization is necessary because the ecological guidelines of an urban area have to differ from the guidelines of a suburban or rural area. When taking the ecological guideline "preservation of the typical landscape pattern" in an urban area there are typically other functions to preserve (e.g. recovery zones) as in a rural area (e.g. biotope for specific species). Similarly for the ecological guideline "saving large unfragmented areas" there exist different threshold values for the three categories. In all three categories it is important to save large connected areas; in urban regions they serve as recovery zones, in rural areas they save the existence of species. Methods to derive urbanization categories will be explained in more detail in the next chapter.

4.1    Categorization of the Natural Regions

An area can be roughly subdivided into three categories, urban, suburban and rural. Suburbanization is defined as the transfer of uses and population from the city, the rural area and other metropolitan areas into the urban surrounding. At the same time the uses and population of the whole area of the metropolitan area will be re-organized (Friedrichs, 1995). The town in its administrative boundaries is called central city, the adjacent area is called suburban zone, followed by communities which area called exurbs. The terms agglomeration, conurbation, metropolitan area are used for the area of central city and suburban zone. The separation of the suburban zone is usually done by the number of commuters. The analysis of suburbanisation serves to force cities and communities to a common planning, for example to reject unregular development, uneven burden on central city and suburban zone as well as the spatial segregation of types of households and their consequences. An additional consequence of suburbanisation is the strongly increasing traffic (Friedrichs, 1995).

Two methodologies will be introduced on how these categories urban - suburban - rural can be attained.  For the first and easier possibility there has to exist a digital landscape model which in this case is represented by ATKIS, the federal topographic-cartographic information system of Germany (Kophstahl and Sellge, 1995).  The second possibility needs digital road network data as a database.

Database: A Digital Landscape Model

If a digital landscape model is available as a database, then the classification of a region into urban, suburban and rural can be simply calculated by capturing the distribution of land cover/use.  The percentual distribution of the built-up area gives information on the urbanization of a region.

A digital landscape model classifies the land use and function. In order to obtain a digital landscape model there exist several possibilities such as digitizing maps (topographic, cadastral, thematic maps) or deriving the classes by remote sensing. An example for a digital landscape model is ATKIS which consists of an object-structured object class catalogue.  The catalogue comprises six object domains (Fig. 3). Each object domain is composed of one or more object groups which in turn consist of individual oject classes. For example object domain "Settlement" possesses three object groups ("built-up areas", "open space" and "buildings and other constructions"), the object group built-up area consists of the object class 2101 city, 2111 living area, 2112 industrial area, etc.. Object classes can have attributes as shown in Fig. 3.

Figure 3: Construction of the ATKIS - Object Class Catalogue

In order to calculate the percentual distribution of the individual object classes from ATKIS, one has to make sure that one polygon represents only one object class. Especially when looking at object domain 2000 (settlement) this is often not the case. A polygon area can be within the city (class 2101), is a power plant (2126) in an area defined as industrial area (2112). In this case the polygon got the class 2101. For this application the structure of the city is not important, it is enough to know that it is a city. Several rules were put up, to deal with multiple naming of one polygon. These rules are of course dependent upon the objective of the analysis, which in this case is a rough categorization of the Natural Regions into three categories. Other examples for multiply-named polygons are

For the Region of Stuttgart (218,825 polygons) there were 25.1 % multiply-named polygons. For 97.9% of the multiply-named polygon applied the case that they were characterized as city area (class 2101) in addition to other classes. These polygons were classified as city area. 0.32% of the multiply-named polygons possessed one class of the object group 5100 (water area) which was considered dominant and 1.8% of the multiply-named polygons were other cases (sometimes also mistakes were found, e.g. an area was classified as a field and greenland at the same time).

For categorizing the Natural Regions the proportion of the sum of the area of all object classes within the object group 2100 from the total area of the Natural Region was calculated (Tab. 1). When looking at the distribution of object class 2100 in table 1,  one can see several breaks (Examples for the three categories are shown in figure 4):

  •     IF Object Class 2100 > 50 %, THEN URBAN
  •     IF Object Class 2100 < 10 %, THEN RURAL
  •     IF Object Class 2100 > 10% and < 50%, THEN SUBURBAN

    Object group 2100 * 2.57 3.76 4.32 5.31 7.21 8.57 8.53 15.52 16.47 18.17 30.51 54.71
    Table 1: The Distribution of Object Class 2101 in the Region of Stuttgart 
    * Percentage of the area of object group 2100 at the whole area of the Natural Region
    For the Natural Region 106 with a percentage of 30.51 one can say, that it has a tendency towards urbanity, whereas Natural Regions with a percentage between 10 % and 15 % would have a tendency toward a rural area (no case included in the study area).

    Figure 4: Examples for the three different categories

    Database: Digital Road Network Data

    Another possibility to obtain the categories urban - suburban - rural is to establish unfragmented areas of similar sizes with digital road network data. In this case ATKIS was used again, but there are other digital road network data like GDF (Geographic Data File) which could be used.  (see e.g. Heres and Wood, 1992)
    Since this application is done on a regional scale, all roads which were smaller than a community road were deleted. Then the road network was used to build a polygon topology. Afterwards the area sizes of the polygons were classified into eight classes with class 1 being the smallest ( < 1 km²) and class 8 being the biggest (> 8 km²). The classes were subdivided by steps of 1 km² .

    With the classes at the x-axis and the relative frequency at the y-axis, graphs were established for all Natural Regions within the Region of Stuttgart (Fig. 5). Three characteristic types of graph could be distinguished: the urban category has a peak at class one (ID 105 in Fig. 5), the rural category at class 8 (e.g. ID 104) and the suburban category is characterized by small peaks at class one and class 8 (e.g. ID 122).

    For assigning an urbanization category to the Natural Region, the following methods were tested:

    1. A regression line was applied to the graphs (Fig. 5)

    2. The slope of the regression line was used for categorization. A positive slope means a rural category, a negative slope shows an urban category and more or less no slope (slope = 0) means a suburban category (Tab. 2). The difference between Natural Region 106 and 105 is not distinguished very strongly. Without having more examples for urban areas, one cannot be certain that there is a break between these two regions.
    3. A sum of slopes was calculated for each Natural Region by adding up the slopes between the classes (Tab. 2).

    4. As in the first method a positive sum of slope represents a rural category, a sum of slope close to 0 suggests a suburban category, a negative slope is an urban category. In contrast to the slope of the regression line the difference between the sum of slopes for 106 and 105 is much higher.
    These two methods are very similar but the  sum of slopes produces a clearer result than the regression equation. The latter shows the central tendency whereas the sum of slopes shows the difference between class 1 and class 8.  Natural Region 102 was categorized differently for the two methods. This can be explained by the peak at class 3 (Fig. 5). This peak is taken into account by the regression line but not by the sum of slopes .

                         Figure 5: Distribution of unfragmented areas for the 12 Natural Regions of the Region of Stuttgart.
                                Class 1: < km², Class 2: 1 - 2 km², Class 3: 2 - 3 km², Class 4: 3 - 4 km²,
                                Class 5: 4 - 5 km², Class 6: 5 - 6 km², Class 7: 6 - 7 km², Class 8: > 8 km².
    ID Regression Equation Category Sum of Slopes Category
    124 y=0.0655 x - 0.1698 Rural 0.764 Rural
    96 y=0.0592x - 0.1414 Rural 0.633 Rural
    94 y=0.0582x - 0.1369 Rural 0.685 Rural
    104 y=0.0443x - 0.0744 Rural 0.546 Rural
    107 y=0.0407x - 0.058 Rural 0.488 Rural
    101 y=0.0239x + 0.0174 Suburban w. Tendency Rural 0.294 Suburban w. Tendency Rural
    102 y=0.0208x + 0.0315 Suburban w. Tendency Rural 0.409 Rural
    108 y=0.0194x + 0.0378 Suburban w. Tendency Rural 0.245 Suburban w. Tendency Rural
    122 y=0.0064x + 0.0963 Suburban 0.095 Suburban
    123 y=0.0003x + 0.1235 Suburban 0.052 Suburban
    106 y=-0.0211x + 0.2198 Suburban w. Tendency Urban -0.148 Suburban w. Tendency Urban
    105 y=-0.033x + 0.2734 Urban -0.41 Urban
    Table 2: Derivation of the Urbanization Category

    There exist three Natural Regions which have a transient function, i.e. the regions 101, 108 and 106. The first two have a tendency to a rural setting and the last has a tendency towards urbanity.  These spatial units should be treated as the next urbanisation category. That is, a suburban area with a rural tendency should be treated as a suburban area and not as a rural area and a suburban area with an urban tendency should be treated as a urban area. It needs more samples to verify the categorization of these units.


    When comparing the results of the two methods using different databases for establishing a categorization of urbanity there can be seen differences. The sequences within the categories, e.g. from most rural to least rural, do not coincide. The reason is that object group 2100 (first method) represents class 1 of the second method, but it doesn't give evidence about the biggest class (class 8). Using digital road network data, on the contrary, depends on the different frequency of the classes 1 to 8. This can be demonstrated very well for Natural Region 108 (Fig. 6): this region belongs clearly to the rural category at the first method, but looking at the second method it is rather found in the suburban category (even though this is somewhat ambiguous as was discussed in the previous section). It depends on the further use of the categorization which method should be chosen.  In addition, it depends on the availability of the database. It is easier to derive the categorization by using a digital landscape model. But this information will often not be available. Digital road network data, in the contrary, are gathered more frequently. GDF-data, for example, are captured for almost all Europe.

                              Figure 6: Comparison of the two methods for categorization for ID 108

    4.2 Establishment of a Ecological Guidelines

    Once the landscapes are categorized, an evaluation scheme can be constructed depending on the different categories. This evaluation scheme can be automatically derived by using the digital landscape model of ATKIS. For the region of Stuttgart rules for a goal could look like the following (the object classes and, where necessary, the attributes of ATKIS are set in parentheses):
    1. Ecological Guideline: "Saving large unfragmented areas"

    2. In general, the larger an area the more protected is this area.
    3. Ecological Guideline: "Saving the typical distribution of land use/cover"

    5. Summary

    Every evaluation either in planning or other sciences needs a scale, an evaluation scheme, against which the impact of a construction is measured. For the evaluation of streets on a regional scale a multi-level approach is suggested reaching from the local level, the road, to the regional level, the landscape. At this level, the type level, the evaluation scheme is defined. It is proposed that the ecological guidelines are dependent on the urbanization category of the landscape. For extracting this category automatically from digital data, two approaches are discussed. The urbanization category then determines the threshold values of the different ecological guidelines, which can also be derived automatically. This was shown at the example of two parts of an ecological goal, the protection of large unfragmented areas and the protection of the landscape pattern of a region.


    Arentze, T.A., A.W.J. Borgers and H.J.P. Timmermans: Design of a view-based DSS for location planning. Int. J. GIS 1996 Vol. 10 (2) (1996) 219-236

    Batty M. and P.J. Densham: Decision Support, GIS, and Urban Planning. (1996)

    Bell,  P.D.: Forest Fragmentation: Adverse Effects on old growth forest Biodiversity. In: G. Blouin (ed.), Old Growth Forests. Canadian Forestry Association (1995) 45-47.

    Breuste, J.: Urbanisierung des Naturschutzgedankens. Naturschutz und Landschaftsplanung 26 (6) (1994) 214-220.

    Densham , P.J.: Spatial Decision Support Systems. in: Maguire, D.J., M.F. Goodchild, D.W. Rhind: Geographical Information Systems - Principles and Applications. Longman Scientific & Technical: Essex (1991) 403-412.

    EUA: Die Umwelt in Europa: Der Zweite Lagebericht. Europäische Umweltagentur, Kopenhagen, 1998 (1998)

    Forman, R.T.T.: Land Mosaics - the ecology of landscapes and regions. University Press: Cambridge, UK.  (1997)

    Friedrichs, J.: Stadtsoziologie. Leske + Budrich: Opladen, Germany (1995)

    Heres, L. und T.F. Wood: GDF, a Lingua Franca for Geographic Information. In: EURONAV 92 - Digital Mapping and Navigation - The 1992 International Conference of The Royal Institute of Navigation and The German Institute of Navigation (DGON), London (1992)

    Kophstahl, E. und H. Sellge: Das Geoinformationssystem ATKIS und seine Nutzung in Wirtschaft und Verwaltung, Vorträge anläßlich des 2. AdV-Symposiums ATKIS am 27. und 28. Juni 1995 in Hannover, Niedersächsisches Landesvermessungsamt, Hannover (1995)

    Meynen, E. and Schmithüsen, J.: Handbuch der naturräumlichen Gliederung Deutschlands. Hrsg.: Bundesanstalt für Landeskunde und Zentralausschuß für deutsche Landeskunde (1953)

    Stauch, C.: Modeling the Ecological Effects of Traffic on a Regional Scale using GIS, In: J. Breuste, H. Feldmann, O. Uhlmann (eds): Urban Ecology, Proceedings of the International Conference "Urban Ecology" Leipzig 1997, Springer: Berlin, Heidelberg, New York, in print  (1998)

    Waterstraat, A., H. Baier, R. Holz, H.-J. Spieß und J. Ulbricht: Unzerschnittene, störungsarme Landschaftsräume - Versuch der Beschreibung eines Schutzgutes. Schriftenreihe des Landesamtes für Umwelt und Natur, Mecklenburg-Vorpommern (1996) 15-24.

    Wiegleb, G.: Leitbildmethode und naturschutzfachliche Bewertung, Z. Ökologie u. Naturschutz 6(1) (1997) 43-62