View PDF

5 th AGILE Conference on Geographic Information Science, Palma (Balearic Islands, Spain) April 25th-27 th 2002

Growth and Densification Processes in Suburban Landcapes - a Spatial Agent - Simulation Wolfgang Loibl*), Rudolf Giffinger°), Tanja Toetzer*) *)Austrian Research Centers - Seibersdorf research, A-2444- Seibersdorf, Austria °)Technical University Vienna -Wien, Karlsgasse 13, A-1040 Wien, Austria, [email protected] Abstract: The paper discusses a model that performs the simulation of polycentric development of suburban systems. It introduces different settlement pattern growth speeds in different suburban regions considering housing area densification and land use change from open space to built up area. Accordingly, the approach concentrates on a Spatial Agent Model (SAM) to perform the movement patterns and the settlement densification processes as well as land use change. In particular, the approach takes into account the suburban population migration as driving force for polycentric development patterns. Therefore, this model is better adapted to real world suburban processes than Cellular Automata (CA) applications that are based only on general land use transition rules applied to neighbouring land use patterns.

Introduction Suburbanisation is an ongoing process that pushes landscape transition. New nuclei within the agglomerations emerge through the process of suburbanisation. Accordingly, a polycentric growth pattern dominates suburban development with different intra-regional growth velocities and directions. In this context landscape transition based on the increasing land demand for suburban housing and commercial activities – leads to new patterns of (sub-)urban cultural landscapes (Breuste, 2001). Due to the decrease of open space and increase of commutertraffic as result of suburban migration this is further an essential environmental issue (ÖROK, 1996) . In regard to landscape transition the process of densification within the suburban settlements is a crucial element in the growth of suburban settlements because they are strongly related with the velocities of land use change. Because of preferences for single family houses densification is not the predominant type of settlement development. Anyhow, if densification would be a more dominant process in suburban growth, the dynamics of land use change might slow down remarkably. Therefore, facing the increasing lack of open space and the shortage of available building lots in many suburban regions of European cities, densification has to be considered as a decisive condition in suburban growth. Thus densification related to suburban migration pattern is necessary to be modelled, in order to provide alternative development scenarios. The principal goal in this paper is to model of polycentric growth patterns with different velocities over time. The model approach presented here is a micro simulation approach - single action simulation (based on a two step household migration decision model) that leads to regional collective land use and housing density pattern changes.

Model approach City planners needs: accuracy vs. similarity of landscape micro-simulation Settlement landscapes are a result of complex spatial processes related to various economic, social, demographic and geographic influences. Therefore modelling of (sub-)urban growth is a well known task in urban-regional research (for an overview see Richardson, et al., 1996); different classes of models have been developed over the last decades. The - obviously chaotic – spatial development patterns within large scales could not be handled by macro simulation models (e.g. Lowry, 1964, Forrester, 1969) based on population-employees - equilibrium assumptions and bid-curve-gradients. Cellular Automata models (CA), disseminated since the 1980-ies (e.g. by Couclelis 1985, 1997, Clarke et al.,1996, White et al.,1997, Batty et al., 1997), are a step forward. Cellular settlement growth simulation, based on neighbourhood land use pattern and local transition rules, appears as fractal land use change that looks similar to real world processes. For scientists and even for regional planners the results were satisfying as long as they show the overall trend. CA-simulation applied to an identical land use pattern in two sub-areas will in - the long run - lead to the same land use change process due to the general transition rules. Therefore - when these models are used for simulating polycentric suburban landscape transition the results, when examined closely, do not match with

1


reality – settlement growth is simulated in the “wrong” municipality, or it takes place faster or slower. The reasons for different landscape transition velocity cannot be taken into account. Reaching only similarity of pattern does not meet the needs for today’s city planners, particularly in usually polycentric landscapes. As city planners expect accurate high resolution future scenarios on a large scale, valid for every municipality, one has to rethink the CA-model approach. As cells are not “responsible” for land use change, a different approach has to be considered to simulate landscape transition processes based on real behaviour of local actors. Spatial agent models: Landscape transition driven by local actors decisions To introduce spatial agents in a simulation context of landscape transition we will use the Chinese game “Go” (originally “Wei Ch’i”), invented 4000 years ago. (McAdams M., 2002). Here two “populations” compete for territory by many single actions while at the same time keeping track of a large scale strategy. Fig. 1: Go – territorial pattern change by local actions and large scale strategy. When playing Go territorial patterns change because of single stones that are placed on the board. In real world these stones can be compared to migrating inhabitants or households and the driving forces to the decisions of the moving people. In other words (cultural) landscape transition is the result of many single human decisions and activities. Therefore, not neighbourhoods or cell states of a land use grid map but interactions between local actors and their environment must be considered in order to “forecast” landscape transition with higher accuracy. These actors are integrated by spatial simulation models as “spatial agents”. But who are these Agents? Franklin and Graesser (1996) defined an “agent as a system situated within and a part of an environment that senses that environment and acts on it, over time .... ”. Regarding suburban spatial dynamics, the behaviour and movement of agents depend on its knowledge about the region, its perception of the surrounding area, its desires and (financial) constraints, its decisions and finally, on its actions to overcome discrepancies between desires and housing conditions in their (former) residential environment. Agents react on environmental disadvantages by responsive behaviour. Although each agent acts individually, the sum of all actions lead to collective behaviour patterns (with some stochasticity!) and thus effect changes in the spatial patterns. As agent-based models describe decisions and behaviour of single autonomous objects leading to multiple and even contradicting spatial effects, the model results come close to real world behaviour. Referring to urban systems, spatial agent models (SAM’s) are applied in order to simulate the movement of households, enterprises etc., which is taken as basic reason for landscape transition. Models that consider this paradigm of more or less autonomous agents were developed in the late 1990s (cf. Wegener & Spiekerman, 1997, Portugali, 1999, Torrens, 2001). An agent-based simulation of polycentric suburban settlement growth Our contribution demonstrates a simple but effective agent based model with some extensions referring to the development of polycentric settlements in order to gain higher accuracy regarding the location of transition and the speed of settlement development. The basic assumptions are : Agents are households with different socio-economic characteristics (and thus different desires, decisions and actions), that want to migrate into the respective suburban region. The landscape of the suburban region is modelled as cellular space with n-tuple cells - several grid cell layers contain different local information. Agents move virtually within the cellular landscape. The following figure 2 depicts the concept as overview. The model steps are: (a) Initialisation of the reservoir of migrating households: The initial model task is to define the total number of migrants moving (mainly from the core city) to suburban settlements within the urban system. The overall driving force for regional growth is regional economic welfare which attracts migrants (and start-up of enterprises creating employment opportunities). Besides economic pull factors the quality of residential environment is a strong influence on migration. Unsatisfied with living conditions in the core city due to environmental disadvantages a specific fraction of households decides to move out of the city. (b) Municipality choice model: In a first task each household decides, to which municipality it want to move, based on several selection criteria. The selection of the municipality is performed randomly taking into account a municipality choice probability 2


distribition. The probability for each suburban municipality to be selected by the agents was estimated by former migration numbers and by attractiveness data relating to the main selection criteria. The probability controls the single agents choice and in general the frequency of how often each municipality will be selected as potential migration target. (c) Agents allocation and landscape transition In a second task, after the selection of the municipality, the search will be continued within the targetmunicipality for an ideal or at least appropriate residential area. The local search depends on the desires of the different agent classes regarding housing type and population density and on supply criteria: on the amount of available open space - zoned as built up area or on the possibilities of densification in the housing areas of the respective municipality. If the agents search was successful, they will settle and the population density in the respective cell has to be increased by the household size (and if necessary/possible, the land use class has to be changed – performed by cellular automata). If not, a different municipality has to be selected according to the choice probability distribution. Thus the demand of the agents for new housing opportunities – when satisfied by local supply pushes the polycentric growth leading to different local velocities and directions of suburban landscape transition. Task 1 and task 2 are carried out by all agents, one after the other, to simulate the migration of single households. The decision of each agent where to settle is influenced by the actions of the former migrating agents as they cause new population density and/or housing area patterns. In addition each agent action might also influence the migration decision of the “remaining” moving agents. Model tasks, elements:

Model action:

Input data:

Initialisation

Population forecast: immigration total

Agents: perception desire/limitation decision action

Task 2 - search location Spatial Agents: agent / perception Cellular desire automata Landscape model

migration data pop. forecast data regional housing demand - by agents movement

Task 1 - select municipality Spatial agent model

socio-demographic data

Effects of national population dynamics regional migration balance

Municipality choice - model: target selection probability

regional attractiveness: - landscape - accessibility - services supply - land prices / rents

Agents housing demand - satisfied by local supply

loop back if search is not satisfying

Agents allocation Landscape transition: new built up area/ densification

local attractiveness: - population density - zoning - land use

Fig.2: General model structure

Model details The model objective is to simulate suburban development of built up area for 180 municipalities in the Greater Vienna Region in Austria during the past and for the near future. The simulation time frame to validate the model ranges from 1968 to 1999. The comparison of the observed land use patterns for 1999 with the model results for 1999 will allow a validation of the simulation. In a later stage a simulation for future developments of built up area until 2011 could be performed. Study area and basic data sets The study region, covering an area of approx. 90 x 120 km, consists of some 1000 census wards, 184 municipalities and 9 districts. Demographic data including migration allows to model landscape transition based on housing demand. Data sets regarding demography, employment and housing were derived from Austrian census data for 1971, 1981, 1991 and 2001. Economic and demo-graphic information is provided for census wards, while data on migration relations are provided only for municipalities. Spatial data were compiled as grid cell maps with a resolution of 100x100 m. A 12-classes land use map for 1968 was derived from U.S. satellite photographs, a congruent land use map for 1999 came from high resolution satellite scanner data (Steinnocher et 3


al., 2000). Additional spatial data sets are landscape attractiveness, accessibility maps, road network, services supply locations, land prices and land use zoning. (Loibl, 2001) Very important are population density maps calculated on grid cell level using population numbers per ward and the housing area covering each ward.

Fig. 3: Land use maps 1968 and 1999 for the Greater Vienna Region Agents characteristics regarding suburban migration Let’s first describe the formal aspects. We have introduced several agent classes. The agents generated here are discrete, reactive autonomous agents deciding about the location of their future residence by referring to spatial properties they prefer. The model provides a blackboard for each agent class to share information about successful searches of housing cells. The decision process of individual agents on the regional scale takes into account municipality choice probabilities with additional thresholds to reduce the number of the appropriate municipalities for each agent class. Within each municipality a very individual search process will be performed by single agents based on a simple multi-criteria decision making approach. Our agents have different characteristics and, accordingly, different desires/constraints and will decide on municipalities as appropriate future residence. The spatial knowledge base for migration decisions is provided as attractiveness layers. Four agent classes were defined through different household types related to socioeconomic characteristics and different behaviour patterns. The behaviour patterns were derived from prior migration data related to socio-economic characteristics (regarding attractiveness layers see 3.3): 1.

2. 3. 4.

High income households: they prefer single family houses in municipalities with high levels of accessibility, high standards of supply of social services, and high levels of landscape attractiveness and they don’t mind high levels of land prices Moderate income households: they can afford single family homes in municipalities with low levels of land prices, moderate standards in accessibility and in the supply of social services. Moderate income highly educated households: they prefer municipalities with high levels of accessibility and of social services , but accept flats in multi-storey buildings as they cannot afford single family houses. Lower income households: they accept flats in multi-storey buildings in municipalities even with low levels of accessibility and land prices or rents.

Two more agent classes were introduced but are not further discussed here: household looking for weekend houses and enterprise founders looking for appropriate commercial lots. Knowledge base for migration and target selection probability: attractiveness of municipalities The initial decision of the households, concerning the selection of a target municipality depends on the desires and financial limitations of the moving households with respect to their socio-economic status and thus their financial power The knowledge base of the agents to decide where to move is regional attractiveness, that further influences migration patterns. Attractiveness surface layers were generated as follows (Loibl & Kramar, 2001): •

Landscape attractiveness is derived from the land use map: the fraction of attractive land use classes (e.g. forest areas) was calculated using a focal neighbourhood window. Elevation range was selected as second landscape attractiveness variable, derived from a digital elevation model.

4


•

• • •

Indicators for local services supply were derived from the numbers of service facilities for each municipality. Density kernels estimate service supply quality for municipalities with and without facilities considering not only the affiliation to a municipality but the accessibility to those facilities. In this context attorneys, pharmacies, medical services etc. were considered as appropriate indicators. Accessibility of the Vienna city centre as commuting focus was calculated applying a shortest path model in order to find the shortest travel time. Accessibility was calculated for time steps: 1968, 1999 and 2015 to estimate the future accessibility, considering future road network. Availability of lots is represented by the total built up area as proxy data set. Average land prices per municipality were provided from sale statistics per municipality and further information sources.

Suburban immigration numbers per municipality have been explained by the regional attractiveness with the help of a regression analysis. (Loibl & Kramar, 2001). Separate equations for all agent classes were then calculated and applied to estimate municipality selection probabilities P. Pi k = f ( Dci,Li ,S i,Ai , Xk)

(1)

where Dci, = distance between center and target municipality i (accessibility of i) Li = landscape attractiveness at target i (forest quota, elevation-difference) Si = services supply at target i, A i = availability of lots, houses etc. all at targets i. Xk = weight – vector of agent class k with elements xD , xL, xS , xA The formal definition is that properties Dci, Li , S i, Ai of a municipality i builds a tensor E that is combined with the weight vector Xk of the respective agent class k. The result is the choice probability Pi k of municipality i referring to agent class k : P k = E Xk

(2)

where

Pi k =

( Dci

Li

Si

 xD    Ai )  xL  x   S x   A

(3)

which means Pi k = Dci xD k + Li xLk + S i xS k + Ai xS k

(4)

The decision, which municipality will be selected, is performed randomly referring to the probability distribution Pi k for each agent related to agent class k considering all municipalities i. (c.f. Fig.4) A further selection restriction that overrules the probability distribution was introduced via thresholds for 2 selection criteria for every agent class in order to exclude municipalities because of unacceptable properties (e.g. “travel time to Vienna > 40 min. ” for the high income households or “land price > 200 €” for low income households) . Fig. 4: The estimated general probability distribution of municipality choice as future residence (for 180 municipalities in the surrounding area of Vienna) Migration patterns probability

based

on

municipality

choice

Between 1971 and 2001 120.000 “migrants” have to be moved leading to an (observed) housing area increase of 6000 ha. Here we simplify and assume an average household size of 3 persons that leads to 40.000 migrating household-agents : Strictly speaking we do not simulate all migration-movements but those that lead to migration balance surplus that drives new housing demand. An in-/out-migration balance of zero is not considered to be relevant for new housing demand. This again is a simplification, as we ignore the decrease of household size and the increase of average flat size that also pushes the growth of housing areas. But as the change of household size and flat size is responsible for a small part of suburban growth when compared to the overall migration surplus, it is acceptable to integrate these influences by reduced densification rates instead of including them in the model by an extra task.

5


The probability distribution governs the agents’ choice regarding the selection frequency for their respective municipalities as migration target. The figure 5 depicts the model results for a part of the study area: The bar chart shows the growth of built up area cells in the district of “Moedling” south of Vienna with some 20 municipalities. The built up area of each municipality is represented by two bars. The upper bar (red/grey) show the housing area in 1999, the lower bar (magenta /light-gray) show the housing area in 1968. The stacked (purple/black) bar extending the 1968 bar show the growth of housing area per municipality. If the (lower) stacked bar reach a length similar to that of the upper bar, the simulation can be examined as valid. District Moedling: Simulation of built up area growth per municipality 1968-1999

Wienerwald Wiener.Neudorf Vösendorf Perchtoldsdorf Münchendorf Mödling Maria.Enzersdorf Laxenburg

housing 68

Laab im Walde

growth-simulation 69-99 housing 99

Kaltenleutgeben Hinterbrühl Hennersdorf Guntramsdorf Gumpoldskirchen Giesshübl Gaaden Brunn.am.Gebirge Breitenfurt.bei.Wien Biedermannsdorf

bebaute 100x100 mRasterzellen

Achau

0

200

400

600

800

1000

Fig. 5: Observed and modelled housing area growth 1968-1999 south of Vienna Modelling the agents’ spatial decision about the future residence location The final decision of the households for the selection of the appropriate residential area depends on the available lots, houses or flats and the local advantages or disadvantages (population density, land use zoning conditions and neighbouring land use pattern.) In contrast to the CA models local neighbourhood is just one of several border conditions that influence landscape transition. The local search is solved spatially by multi-criteria decision making regarding attractiveness within a cellular space where each cells contains several criteria, which were considered during the search procedure. The assumptions regarding local search are as follows: • •

•

Agents that prefer single family houses start their search on a random location in open space within the respective municipality moving to the settlement border and looking for a built up area cell that shows a potential population density that is small enough to allows single family houses. Agents that accept multi-storey buildings will start their search in the municipality centre and move in random direction in search of a cell with the lowest potential population density above a threshold that is big enough to expect multi-storey buildings. If there is a difference between the actual and the potential population density settling could take place. The agents will search for the cell with the smallest potential population density. If there is a difference between the actual and the potential population density settling is possible. Then additional cells in the vicinity that might be a better choice will be investigated regarding additional criteria of attractiveness or repulsiveness. The cell that shows the “best value”: e.g. low population density, long distance from highways, a lot of open space but adjacency to already built up area, etc will be selected. The sequence in which the criteria will be considered in case of similarity of characteristics is subject to “expert judgement”.

6


•

•

Th agents will only then chose a new open space cell if there is housing area zoning and all other already built up area cells in the local neighbourhood show only a small remaining population density difference between former and future density. Land use change transition will be performed through cellular automata application considering the land use pattern in the local neighbourhood. The future population density of open space cells to be newly covered by residential buildings will be defined by the density of the adjacent built up area cells. An increase in density is only possible for cells adjacent to the already densely built up areas. The communication between agents runs via a blackboard: each successful agent writes down the location of the last successful search within each municipality on the blackboard of the respective agent class. Before a search an agent of the same agent class will follow the information on the blackboard and search for a residential area in the surroundings of the previously occupied location in this municipality. If after several trials is not successful in the respective municipality the search will start again in a different municipality according to the municipality choice probability distribution.

Fig. 6: Residential location search procedure of single family home agents The flowchart in above figure 6 shows the structure of this local search task for the agents preferring single family houses. The last “diamond” contains the steps shown in fig.7. This Figure highlights the final choice of the most favourable cell within the local neighbourhood regarding further criteria related to residential attractiveness as described on the next page.

7


Fig. 7: Local search steps and final choice of the future residential location The details of how this final spatial choice decisions are performed are described below: 1. Search of local minimum of the population density surface in the respective municipality starting from a random location. The way to reach the settlement boundaries are described in fig.6. and shown in the “bubble” of fig.7. 2. Selection of all appropriate cells within the neighbourhood of the minimum population density cell, and investigation of additional properties of these cells. 3. Ranking these property characteristics, weighting of the properties and final selection of that cell with the maximum weight /attractiveness. The scheme how this last two steps are performed can be seen in fig. 7 on the right hand side. Step 1 takes into account the local population density minimum d at time t, but with several constraints that can be expressed as:

 (d t > 0) ∧ (d (t + 1) − d t > 0) : 1  (d t ≤ 0) ∧ (d ( t + 1) − dt ≤ 0 ) : 0

f(d) = 

(5)

d is the cell population density and d (t + 1) − dt is the difference of the potential population between time step t and (t+1) – the vacancies available to be “occupied” by settlers within a local set of cells S l : d ∈ Sl In step 2 the agents check other appropriate cells that might be more attractive in the vicinity of the local minimum population density cell because of other criteria. After reaching the (local) minimum population density cell min(S l ) a set of cells in a neighbourhood S will be selected within an defined extent of +/- n cells:

Sl = {( x, y) ( xi − extent < xi < xi + extent ) ∧ ( yi − extent < yi < yi + extent )} (6)

For those cells referring to equation (6) that show an appropriate population density referring to equation (5) a set of additional criteria will be investigated. The criteria are provided as 10-tuple space organised in 7 cell layers with a 100 x 100 m grid cell resolution (The last 3 criteria will be calculated out of the adjacent cells): -

actual population density related to built up area cells potential (future) population density related to built up area cells actual land use land use zoning regulations distance to next settlement(centre) distance to highway (weighted by traffic load) distance to major junctions number of neighbouring housing cells (3x3 Moore neighbourhood) 8


number of neighbouring industry cells (3x3 Moore neighbourhood) number of neighbouring open space cells (3x3 Moore neighbourhood)

-

In step 3 multi-criteria decision making is performed, which allows a rational spatial choice within all appropriate alternatives ( following Jankowski, 1995). Therefore all appropriate cells that might serve as future residence alternatives will be ranked by referring to each single criteria. The final rank sequence is provided after all criteria are weighted and summed for each cell. The set of local cells S l that meet equation (5) is characterised by a set C (j =1 to n) criteria:

C = {c1,....cn}

(7) In order to provide ranks in a fast way, all criteria c j of set C will be normalised to C’ : C

' C=→ 1 +C9'

j

cj max(C)

with

(8)

which leads to a set of normalised criteria C’ j (j=1 to n) of each cell i - now with values ranging from 1 and 10:

{

}

C ' = c'1,....c' n (9) The “ranking” by normalisation is more advantageous because we do not refer to rank differences but to relative differences that are related to the real values of the criteria. Starting with score 1 instead of 0 preserves, that some cells criteria will not be considered during the weighting process.

The set C is now weighted regarding the importance of the properties for residential attractiveness using a set of factors W .

{

}

W = w1,.... w n (10) The weights provide an order of attractiveness criteria that allows an easier choice between the appropriate alternatives. The normalised characteristics c’ weighted with the factors w deliver the attractiveness for each cell i : n

ai = ∑ c ' i j .wj

(11)

j =1

The step summed up as equation (11) is similar to equation (3). Here not the municipalities are weighted but the appropriate cells within a respective neighbourhood. All a i make up the set A – the attractiveness of the cells i (1 to x) for settling down. (12) A = {a1,....an} The final decision where to settle will be made after checking the characteristics of additional properties of all appropriate cells in the neighbourhood The optimal choice - the “ultimate attractive cell within the set of alternatives is the one with the maximum weighted attractiveness-total a i : afinal = max( A) where a final is member of the set A preferred against a x2....against a xn: (13)

afinal ⊃ ax 2 ⊃ .... ⊃ axn

Model results and empirical findings Figure 8 shows a subset of the results of a simulation run and the comparison with the observed land use patterns 1968 (top-left) and 1999 (top-right) in the South of the Greater Vienna region. The resulting map shows a high spatial coincidence with the 1999 observation although the parameterisation has not yet beet finished. The agent based landscape transition model allows a simulation a polycentric settlement development within a large and complex suburban settlement system with high accuracy. In contrast to the cellular automata model the agent based approach allows a simulation of very different settlement development speeds within the region according to different large scale attractiveness patterns. The results show that the spatial agent model approach is powerful enough to simulate virtual city dynamics in a complex and multi-centred landscape, where local strategies on a municipality level allow to trigger regional effects on landscape transition. Generally it has been proven that the attractiveness layers reflect the real influences on migration very well and control the municipality choice probability distribution sufficiently, so that estimations regarding future developments will become more realistic. Model-results and empirical findings will be used for discussions with local actors and city planners. Alternative planning regulations regarding densification or zoning of new housing

9


areas, simulated for various municipalities, will lead to very different migration as well as residential area patterns. These can be modelled by considering different scenarios of zoning and population density regulations as alternative model input layers.

Fig. 8: Model details: top: observation 1968(left), 1999(right), below: simulation 1999

References Breuste J., 2001. Kulturlandschaften in urbanen und suburbanen Räumen. In ARL & ÖGR (eds.) Die Zukunft der Kulturlandschaft zwischen Verlust, Bewahrung und Gestaltung. pp. 79-83, Eigenverlag der ARL. Hannover. Clarke K.C., S. Hoppen & Perez S., 1996. Urban Growth Model http://www.geo.arc.nasa.gov/usgs/clarke /hilt.html, (rev.03/2000). Couclelis H., 1985. Cellular Worlds: A Framework for Modeling Micro-Macro Dynamics. In Environment and Planning A. Vol. 17, pp. 585-596. Franklin S. & Graesser A., 1996. Is it an Agent, or just a Program?: A Taxonomy for Autonomous Agents. Proceedings of the 3rd Int. Workshop on Agent Theories, Architectures and Languages. pp.21-35, Springer-Verlag. http://www.msci.memphis.edu/~franklin/AgentProg.html, (rev. 08/2001). Forrester J.W., 1969. Urban Dynamics. MIT-Press Cambridge. Mass. Jankowski P., 1995. Integrating Geographical Information Systems and Multiple Criteria Decision Making Methods. IJGIS. 9(3):pp.251-273 Lowry I. S., 1964. A Model of Metropolis. Rand Corporation. Santa Monica. Cal, Memorandum RM.4035 – RC. Loibl W., 2000. Modellierung der Siedlungsdynamik mit einem GIS-basierten Zellularen Automaten - Konzeption, GISIntegration und erste Ergebnisse. In Strobl J. et al. (eds.) Angewandte Geographische Informationsverarbeitung XII, pp 297-306, Wichmann Verlag, Heidelberg. Loibl W., 2001. Simulation of sub-urban growth for the Greater Vienna Region with a GIS-based Cellular Automaton. In: Proceedings of the ISESS 2001, May 2001, Banff/Canada: International Society of Environmental Software Systems. Loibl W. & Kramar H., 2001. Standortattraktivität und deren Einfluss auf Wanderung und Siedlungsentwicklung In Strobl J. et al.(eds.) Angewandte Geographische Informationsverarbeitung XIII, pp 309-315, Wichmann Verlag, Heidelberg. McAdams M., 2002. What Is Go? http://www.well.com/user/mmcadams/gointro.html (rev.03/2002) ÖROK – Österreichische Raumordnungskonferenz (eds.) 1996. Siedlungsentwicklung in Österreich. Szenarien 1991 - 2011. Band 127. Eigenverlag der ÖROK. Wien. Portugali J., 1999. Self Organization and the City. Springer, New York.

10


Richardson H.W., Kenneth J.B., Nijkamp P. and Park H. (eds.) 1996. Analytical Urban Economics. Edward Elgar Publishing Ltd. Cheltenham and Brookfield. Steinnocher K., Kressler F. & Köstl M., 2000. Erstellung einer Siedlungsmaske aus Fernerkundungsdaten und Integration zusätzlicher Information aus Zensusdaten. In J. Strobl, T. Blaschke (eds.) Angewandte Geographische Informationsverarbeitung XII, pp 481-488, Wichmann Verlag, Heidelberg. Torrens P.M., 2001. Can geocomputation save urban simulation? Throw some agents into the mixture, simmer, and wait... http://www.casa.ucl.ac.uk/paper32.pdf, (rev. 06/2001). Wegener M. & Spiekermann K., 1997. The Potential of Microsimulation for Urban Modelling. In: Proceedings of the International Workshop on Application of Computers in Urban Planning, pp.129-143, Kobe University, Kobe, Japan. White R., Engelen G. & Uljee I., 1997. The use of constraint cellular automata for high resolution modelling of urban landuse dynamics. In Environment and Planning B, Planning and Design 24(2), pp. 323-343.

11