Application of a spatially explicit backcasting model: A case study of sustainable development in Salzburg, Austria

Application of a spatially explicit backcasting model: A case study of sustainable development in Salzburg, Austria

Applied Geography 58 (2015) 128e140 Contents lists available at ScienceDirect Applied Geography journal homepage: Ap...

3MB Sizes 2 Downloads 37 Views

Applied Geography 58 (2015) 128e140

Contents lists available at ScienceDirect

Applied Geography journal homepage:

Application of a spatially explicit backcasting model: A case study of sustainable development in Salzburg, Austria Eva Haslauer a, b, * a b

University of Salzburg, Department of Geoinformatics, Z_GIS, Schillerstraße 30, 5020 Salzburg, Austria Research Studios Austria, Studio iSPACE, Schillerstraße 25, 5020 Salzburg, Austria

a r t i c l e i n f o

a b s t r a c t

Article history: Available online

The high cost of land for housing within urban centres and the common desire to live within extensive residential areas in the green countryside have, in some cases, led to increasing residential development in the urban hinterland, often resulting in dispersed and sprawling development. In order to counteract such development this article seeks to provide a methodology for implementing strategies which aim is to achieve sustainable development in spatial planning. The proposed methodology, which is known as “backcasting”, aims to improve our ability to avoid undesirable future developments and to encourage those developments that are desirable. Backcasting has previously mostly been used within theoretical processes or frameworks. The backcasting exercise presented in this paper used a Python-based model to create often visionary future scenarios based on interviews with relevant experts, and then used these scenarios as input for a backwards running model. This model simulates a development that runs backwards in time, converging towards the present situation. The backcasting model presented herein has been applied to a case study in Salzburg, Austria. © 2015 Elsevier Ltd. All rights reserved.

Keywords: Decision support model Geographical information science Backcasting Spatial planning

Introduction Backcasting can be described as a decision support method that first designs a desired future scenario (or vision) and then defines how this desired future scenario could be achieved. Only then are specific strategies and follow-up activities identified (e.g. in the form of trajectories) that lead towards the desired future scenario (Quist & Vergragt, 2006). The first backcasting approach was introduced in the 1970s; it was then known as “backwards looking analysis” and dealt with energy consumption and optimization strategies. Since this approach has been developed in different directions, e.g. for sustainability, for transport, and for spatial planning. Backcasting has often been used as a theoretical tool, and especially as a decision support tool in spatial planning where it has been used in specialized workshops to define strategies for reaching a particular future scenario. In contrast, the model introduced herein is spatially explicit, automated, and model-based, none of which applied to previous versions of backcasting. The presented model is set up as a Python model and not exclusively linked to any

* University of Salzburg, Department of Geoinformatics, Z_GIS, Schillerstraße 30, 5020 Salzburg, Austria. Tel.: þ43 662 908585 224. E-mail address: [email protected] 0143-6228/© 2015 Elsevier Ltd. All rights reserved.

particular Geographical Information System (GIS) software but is independent of the type of software used. The model starts by creating desired future scenarios based on possibilities envisaged by a number of relevant experts. The desired future scenarios are compiled from interviews with relevant experts and serve as input for the backwards running model. Starting from these scenarios, the model simulates the development in reverse along the timeline until the present time and situation are reached. Milestones are created during the simulation at time intervals set by the user. These serve as interim goals and represent the steps that need to be taken and intermediate goals that need to be reached in order to achieve the desired future scenario. The backcasting model has been applied to a case study of a rural area in the hinterland of the city of Salzburg, Austria. It is a sprawling residential region with an extent of about 1000 km2. The model is used to encourage sustainable residential development and reduce urban sprawl within this region. The objective of this paper is therefore to determine whether interviews with relevant experts can be successfully used to develop normative scenarios, and whether the attainability of these scenarios can be simulated using a spatially explicit backcasting model. This paper starts with an overview of previous research and a description of the case study to which the proposed backcasting

E. Haslauer / Applied Geography 58 (2015) 128e140

model has been applied. It then describes the backcasting approach per se, starting with a summary of the interviews with revenant experts that aimed to delineate the main spatial problems within the case study area, followed by a description of the setup and functionality of the model. The implementation of the model is then explained followed by a description of the model's output and results. This is followed by “Discussion and conclusion” and “Outlook” sections to conclude the paper. Previous research According to Dreborg (1996) backcasting can generally be applied if the problems faced are predominantly complex, involve a major trend, and are greatly influenced by external factors, provided the timespan for implementing changes is sufficiently long. Backcasting is therefore clearly aimed at changes over long term perspectives. The first use of backcasting dates back to the 1970s when it was introduced by Amory Lovins as a planning method for electricity supply and demand. Since then backcasting has been regularly used in energy studies, often referred to as “energy analysis of the final consumption” (Dreborg, 1996; Quist, 2007). Following Lovins' implementation in energy studies backcasting has also been used in a much wider range of applications, and sustainability and planning have become important research fields. In the 1990s Robinson proposed a generic backcasting method to analyze environmental and development problems on national and regional levels by considering the interactions between natural systems and anthropogenic development (Robinson, 1990). Holmberg (1998) proposed the use of a backcasting analysis for strategic sustainable development in the “Natural Step” project. This project aimed at the “transition of existing structures and organisations in society towards sustainability and to help the new things that are always being created be sustainable from the start”. Backcasting was also used in the Netherlands, in a project on climate change known as COOL (Climate Options for the Long-term), where it was used in an interactive process with stakeholders to elaborate strategy options in response to climate change (Van de Kerkhof, Hisschemller, & ^t-Regamey and Brunner (2011) also preSpanjersberg, 2002). Gre sented a backcasting analysis aimed at adapting spatial planning tasks to climate change. For their methodology they suggested splitting backcasting into inverse modelling for spatially explicit quantitative problems (such as trade-offs in planning processes), and a strategic backcasting analysis for qualitative and complex problems (such as the elaboration of spatial development steps or coordinative measures). A model-based approach linking backcasting with inverse modelling for the analysis of uncertainties in models was also presented by Osidele and Beck in 2003. In England, backcasting analyses have been used in the VIBAT (Visioning and Backcasting for Transport Policy in London) project. The main task in VIBAT was to investigate how to reduce CO2 emissions in London until 2030 and 2050 through a backcasting and scenario approach (Hickman & Banister, 2005, 2007). In Sweden, backcasting has been applied to various other aspects of sustainability research such as, for instance, transportation systems. Other studies have used backcasting to investigate the development of sustainable water supplied, cities, mobility systems, and households (CarlssonKanyama, Dreborg, Moll, & Padovan, 2008; Miola, 2008; Quist, 2007). Carlsson-Kanyama et al. (2008) applied workshop-based backcasting with the involvement of local stakeholders that resulted in ideas, scenarios, and images for sustainable cities of the €chter, Ornetzeder, Rohracher, Scheurer, and Knoflacher future. Wa (2012) used a backcasting exercise in a workshop to develop strategies and milestones towards an efficient spatial organization for the Austrian energy system. The simulation of future scenarios


using Agent Based Models (ABMs) was proposed by Van Berkel and Verburg (2012), who used them in a backcasting exercise: they reported a case study that used ABMs to create region-specific scenarios for the Dutch region of Achterhoek. Stakeholders then participated in a backcasting exercise to define landscape objectives and propose ways to achieve these objectives. Other backcasting approaches linked to spatial planning issues have been developed by Haslauer, Blaschke, and Biberacher (in press) and Haslauer, Biberacher, and Blaschke (2012), and some are also summarized in Quist (2007). Justification for research Swart, Raskin, and Robinson (2004) carried out research into plausible and sustainable future social and environmental systems. They proposed the integration of a scenario analysis into a sustainability toolkit, clearly stating that a transition to sustainability is needed not only because of ongoing population growth but also because of growing economic output. This approach provides a valuable addition to our proposed backcasting approach since both deal with the simulation of population distribution and population development up to the year 2050. In their paper on sustainability Raskin, Chadwick, Jackson, and Leach (1996) summarized some preliminary results of the PoleStart Project, which aimed to develop long-term strategies and policies for sustainable development. Their report described a scenario relating to demographic and economic development, energy and land resources, and the environment, for the year 2050, and analyzed the policies required for sustainable development. They examined a range of intermediate future scenarios that could guide the transition to a sustainable, global future thereby identifying core-questions for the future. Their research supports the use of the proposed backcasting approach to foster sustainable development by including long-term goals and long-term strategies into the model, aiming to maintain desired developments until the year 2050. Brown, Hanson, Liverman, and Merideth (1987) described “sustainability” as a “desired goal of development and environmental management”. They stated that the meaning of this word is strongly related to the context in which it is used, and that its definition must therefore take into account the context, as well as the temporal and spatial scales. The proposed backcasting approach is relatively new and this paper describes the setting up of a spatially explicit backcasting model for use in spatial planning as a decision support model and as a means of estimating future developments. It maintains the spatial and temporal scales and takes into account the context through a Cellular Automata approach. Since urban sprawl can be explained as the uncontrolled spread of urban development (residential and non-residential) over time, it can definitely be considered as a suitable subject for a spatially explicit backcasting approach because of its relationship to both time and space (for further definitions and a more detailed discussion of the topic of urban sprawl see Section Problem definition). The presented model can be easily used by decision makers in the spatial planning sector to translate their individual visions into normative scenarios and propose alternative development pathways that would need to be followed to attain these scenarios. The advantages of this approach are (1) its automation and model-based implementation (since backcasting has previously only been applied as theoretical exercise), and (2) its spatial explicitness. The drawbacks of the proposed approach are (1) the resolution of the model (125 m  125 m) is rather low for planning issues relating to population development, and (2) the transferability of the model is as yet untested, since it has to date only been applied to the one case study presented in this manuscript.


E. Haslauer / Applied Geography 58 (2015) 128e140

Case study area The area selected for a case study using the spatially explicit backcasting model (Fig. 1) is located in the northern part of the Austrian province of Salzburg and covers the political district of Salzburg-Umgebung. This district is a rural area covering about 1000 km2; it includes 37 municipalities and in 2008 had 145,119 inhabitants (Statistik Austria, 2008b). The area was selected because of the excellent data availability and the extensive regional analyses that have previously been carried out by other researchers (c.f. Haslauer, Schnürch, & Prinz, 2013; Schnürch, 2011). For the backcasting exercise the case study area was analyzed with respect to land use patterns and changes in land use. The data were derived from Corine Landcover (CLC) data (EEA, 2006) acquired in 1990, 2001, and 2006 (the latter being the latest CLC data available: a new CLC dataset was expected to be available in 2011 but has not yet been published). The population distribution and the projections for population development between 2008 and 2050 by Hanika (2010) were also incorporated into the exercise. Information concerning the population numbers and the population distribution across the case study area in 2008 was obtained from the Austrian Statistical Agency (Statistik Austria, 2008a, 2008b) in the form of an equal-area polygon raster with 125 m edge-length. Physicalegeographical parameters for aspect, slope, and elevation derived from a freely available digital elevation model provided by the provincial government of Salzburg (OGD Land Salzburg, 2014), were used in the exercise as they were expected to have a significant influence on the population distribution within the case study area. Finally, public transport services (i.e. the rail network and bus stops) were integrated and all relevant datasets describing the area were rasterised into cells with an edge-length of 125 m.

Table 1 Interviewed experts and their working background. Interviewee


Interview 1 Univ. Doz. Dr. F. Dollinger

Spatial Planner working for the Spatial Planning Department of the Provincial Government of Salzburg. Interview date: 23 January 2013. Interview 2 Mag. J. Reithofer Urban Planner working for the City Planning and Transportation Department of the Salzburg City Government. Interview date: 25 January 2013. €ck Spatial Planner and GIS-expert working for Interview 3 Mag. P. Weissenbo the Spatial Planning Department of the Provincial Government of Salzburg. Interview date: 01 February 2013. Interview 4 Dipl.-Ing.P. Lovrek Regional Planner working in a planning association for Salzburg and its surrounding municipalities. Interview date: 02 February 2013.

thus still retaining the participatory backcasting approach. These experts were interviewed during the initial phase of model development; the questions posed and the experts' responses are summarized in Expert interviews section. Backcasting approach The proposed backcasting model was applied to a selected case study area in Salzburg, Austria. This was carried out in two stages, with the first stage involving the creation of the future scenario and the second stage involving the backwards modelling, from the future scenario back to the present situation.

Data and material Expert interviews The implementation of backcasting as an automated, spatially explicit model is a very recent development. To the author's best knowledge backcasting has to date only been utilized in theoretical exercises during workshops or discussions. Such workshops have tended to use participatory backcasting approaches (c.f. Quist & Vergragt, 2006) with a substantial stakeholder involvement. Since both the time and financial support available for the implementation of this model-based backcasting exercise in Salzburg were limited, the author decided to involve only four planning experts from various local institutions and departments as stakeholders,

Experts in spatial and urban planning were interviewed on their preferred sustainable developments for the case study area in order to develop the desired future scenario. Details of the expert interviews can be found in Table 1. All interviewees were chosen on the basis of their expertise and working experience in spatial, regional, and urban planning within the Salzburg case study area. The interviews were conducted in German since the interviewer and interviewees were all German speaking. The experts were asked the following questions:

Fig. 1. Case study area (framed by a red rectangle) in the province of Salzburg (light grey), Austria (white) covering the political district of Salzburg-Umgebung (dark grey). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

E. Haslauer / Applied Geography 58 (2015) 128e140

1. How would you describe urban sprawl in general, and within the Salzburg case study area in particular? 2. Do you see urban sprawl as a problem within the case study area? 3. What actions could be taken to counteract urban sprawl within the case study area and to foster sustainable development? 4. What is your future vision for spatial development within the case study area? The answers were all recorded; they have since been compiled and are summarized below. The summaries do not assign the statements to any particular interviewee but aim to provide an overview of the situation within the case study area. The responses to the question “How would you describe urban sprawl in general, and within the Salzburg case study area in particular?” are summarized below: The case study area, which covers part of the Alpine gateway, is characterized by urban sprawl in the form of scattered settlements. Typical compact Bavarian settlements developed historically within the case study area but the area is today characterized by a dense network of development, which is quite unstructured and illogical. This urban sprawl developed mainly as a result of increased private mobility. The case study area is characterized by remote, scattered housing with large distances between the houses and the town centre or core of the settlement. Low-density housing and detached houses are furthermore typical of the case study area. One reason for this is the traditional agricultural settlement development, in which a farmer's children were usually given a parcel of land when they left the parental farm. These parcels of land could be located on any part of the parental farm and the farmer's children would then build detached houses on their land. It is quite difficult for the planning sector in Salzburg to introduce binding guidelines for the municipalities since the responsibility for spatial planning generally rests with the municipalities themselves. Spatial development programmes serve as guidelines for municipalities but it is up to the individual municipalities to decide which way to implement the guidelines, if at all. The responses to the question “Do you see urban sprawl as a problem within the case study area?” are summarized as follows: Yes it is a problem, a massive problem, and it is also a widely recognized problem but it is rather late to start counteracting it since urban sprawl in the case study area is already an established fact. It has developed over the last 30e40 years. What is alarming is that this form of sprawling development is still continuing. Using only spatial planning instruments to counteract this problem is not enough; people's attitudes towards this topic also need to be changed. It is not helpful to divide opinions into professional, political, and social views. The answers to the question “What actions could be taken to counteract urban sprawl within the case study area and to foster sustainable development?” are summarized below: In the long-term it may not be possible to reduce urban sprawl, but only to counteract it. The only way that it could adjust itself is if mobility becomes more and more expensive (e.g. through rising fuel prices) and people can no longer afford to live in the green countryside, far away from an urban centre, with no efficient public transport connection. However, it would be a worst-case scenario if urban sprawl only decreased because people were forced to move by rising fuel prices. For sustainable development to be achieved through planning it may be enough to freeze any new


developments and to re-build some of the existing sprawling developments. New settlement locations should be checked in terms of sustainability and accessibility to public transport, but it may still take generations to make any noticeable reduction to the existing urban sprawl. The city of Salzburg needs to provide support for actions as well that will discourage urban sprawl, and city planners should encourage residential development within the existing city area. New residential areas need to be developed through urban densification rather than through a re-classification of open spaces. The low density of buildings within the city leaves a great deal of potential for the introduction of more compact building structures or for converting existing detached houses into multi-family homes. There are also high quality areas that are currently occupied by industrial and commercial units, which is an inefficient use of space. An adequate number of attractive and affordable new living areas need to be provided together with other measures to halt the trend towards suburbanization. Other strategies can be implemented concurrently: services such as schools, medical facilities, employment opportunities, public transport, etc. should be further improved in those municipalities within the case study area that already have good infrastructure. They can then serve as regional centres, providing alternatives to city living, with all necessary services and excellent connections to the city of Salzburg. Any further residential development should be coupled with improvements in public transport and polycentric developments should be enforced. There should be no new developments outside regional centres, and no new zoning of building land should be permitted within small villages. Existing efficient transport routes in all directions, including across the border to Bavaria, need to be extended. Cooperation with the adjacent German city of Freilassing needs to be strengthened with respect to residential facilities, work opportunities, and improving the efficiency of public transport. Finally, public awareness of the need for sustainable housing policy and public transportation needs to be improved. The answers provided to the last question “What is your future vision for spatial development within the case study area?” are summarized as follows: An ideal settlement design would involve well-developed regional centres that have urban characteristics and offer a wide range of products and services. Regional centres that are already welldeveloped therefore need to be identified. The inner parts of these centres then need to be defined as core areas within which densification should be supported by, for example, encouraging residential buildings and services at suitable locations. Residential development outside these areas should be restricted. Nevertheless, other smaller centres with good transport connections to the larger regional centre, or to the city of Salzburg, can be further developed but in this situation efficient transport services become very important in order to reduce car-dependency. This form of spatial development would perhaps encourage people to move to those regional locations that have excellent infrastructure with respect to local supplies, medical care, schools, and employment opportunities, together with public transport that offers frequent and reliable services. For a sustainable future it could also be important to consider the re-development of existing structures within the case study area, since these structures cannot suddenly be simply demolished or removed. With regard to the transport sector, sustainable concepts need to be implemented and realized such as the concept of the “Salzburg City Region Train”, or other concepts such as those following the example of Karlsruhe in Germany, which has implemented a robust linkage between local regional train networks and public transport


E. Haslauer / Applied Geography 58 (2015) 128e140

within the city itself. Such concepts could allow access to areas that cannot as yet be reached by public transport. None of these ideas can be implemented overnight, but if political decision makers would come to a common, definite decision, they could indeed be possible.

Problem definition The interviews carried out for this research identified urban sprawl as major problem in the case study area and all interviewees agreed that counteracting the negative aspects urban sprawl represents an important step towards achieving sustainable spatial development within the case study area. Urban sprawl is a complex topic and its causes are varied. The accesses to green space, low crime rates, and greater availability of land for residential building (and hence lower building costs) are just some of the reasons for dispersed development. Changes in economic structures, such as increases in working and operational area requirements, also provide additional incentives (Spitzer, 2007; Squires, 2002). Urban sprawl is very difficult to define in only a few words, or to parameterize (Siedentop, 2005; Spitzer, 2007). Squires (2002) defined urban sprawl as “a pattern of urban and metropolitan growth that reflects low-density, automobile-dependent, exclusionary new development on the fringe of settled areas often surrounding a deteriorating city”. Siedentop (2005) identified different aspects of urban sprawl, taking into account density, form, function, socially relevant effects, and planning perceptions, as itemized below:  low settlement density and the decay of inner cities  de-concentration of urban functions into peripheral rural areas  transformation of mono-centric, compact settlements into discontinuous, polycentric, and disperse settlements  negative impacts of traffic, evoked segregation processes (Spitzer, 2007), and the loss of fertile soils  un-planned urban development and sprawl-supportive urban planning (commuter subsidies, extension of road network in the countryside, etc.) The above-mentioned characteristics of urban sprawl lead, amongst other things, to the development of peripheral cities and fragmented land use (Squires, 2002), and to increased traffic due to a greater separation between housing and jobs. The prevention of environmental degradation in rural areas in general is a primary objective of sustainable planning (Botequilha Leitao & Ahern, 2002). The aim of the proposed backcasting methodology is thus to provide valuable support for sustainable planning and to help counteract the negative spatial effects of urban sprawl. Future scenario A future scenario was developed on the basis of the interviews and indicators or measurements derived that could be used to gauge the progress towards this scenario. The main objectives selected for consideration in the future scenario generated by the backcasting model are:  To support development in regional centres within the case study area, and thus encourage compact settlement.  To foster residential development around existing small regional centres, in order to support a compact distribution of buildings.  To encourage settlement development along public transport routes e predominantly railway lines e in order to improve accessibility within the case study area.

Backcasting is usually applied to regions without taking into account existing artificial structures (such as road networks, service supply facilities, etc.) or influences from any endogenous factors. The only exogenous factors usually taken into account in backcasting analyses are physicalegeographical factors (Haslauer et al., 2012). However, in order to take into account the views of the planning experts and to paint a more realistic picture of the future, the following endogenous and exogenous parameters were selected as influencing factors during the design of the future scenario: 1. Development of the following dominant land use classes: “natural areas”, “building areas”, “agricultural areas”, “water areas”, and “forest areas”. 2. Physicalegeographical parameters: elevation, slope, and aspect. 3. Population distribution and population projection. 4. Public rail-transport network.

Land use development A distribution of the desired dominant land use classes in 2050 was simulated for the future scenario. The land use classes included in this layer were: “natural areas”, “building areas”, “agricultural areas”, “water areas”, and “forest areas”. Since the area of “water areas” changed least of all in the years between 1990 and 2006, the author decided to retain the 2006 “water areas” for the 2050 future scenario. The model thus included only the four remaining land use classes of “natural areas”, “building areas”, “agricultural areas”, and “forest areas”, in the simulation. Physicalegeographical parameters “Building areas” in the future scenario were delineated on the basis of three physicalegeographical parameters: elevation, slope, and aspect. These parameters have been identified in previous local studies (e.g. Spitzer, Dollinger, & Prinz, 2008) as influencing settlement and building development. Spitzer et al. (2008) defined permanent settlement areas in the case study area as areas below 1600 m a.s.l. with a slope of less than 20 that were not zoned as forest, Alps, flowing water, water body, marsh, mountain meadow, harbour, glacier, wasteland, or protected area. In order to arrive at a statistical estimate of the influence that each of these parameters is likely to have on future zoning, the present population distribution per cell and the corresponding elevation, slope, and aspect values for each cell were analyzed using a Geographically Weighted Regression (GWR). The output, using the population in 2008 as the dependent variable and the slope as the explanatory variable, revealed a relationship, albeit a weak one (r2 ¼ 0.3), between slope and population. The same procedure was repeated using aspect as the explanatory variable and a weak relationship between those two variables (r2 ¼ 0.28) was again identified. An analysis of the correlation between population and elevation using a GWR did not indicate any significant relationship between these two parameters. In addition to the GWR, the intensity of the influence that these three parameters had on settlement development was analyzed using an adapted distance decay function. This function analyses relationships between static elements with respect to their proximity to each other (Jermann, 2002). In this analysis one element was the population per cell and the other was the average value of the slope, aspect, or elevation per cell. The analysis revealed that the intensity of influence that slope has on population distribution decreases as the slope increases. The slope values on which there is any population settlement range from 0 to 53 . Intensity values greater than 0.5 (within a range of 0e1.0) relate to areas with slopes

E. Haslauer / Applied Geography 58 (2015) 128e140

of less than 25 . With regard to elevation, there are no population values below 381 m a.s.l., this being the lowest altitude within the case study area to have any associated population values. The intensity of influence of elevation increases with elevation, with the population falling as the elevation increases until it reaches zero at 1296 m. This is the highest point in the case study area to have any associated population values. Finally, with regard to aspect, which is the geographical orientation of the landscape, intensity values greater than 0.5 relate to areas with aspects ranging between 100 and 300 (i.e. ranging between south-easterly and north-westerly directions, since 0 represents due north, 90 represents due east, 180 represents due south, and 270 represents due west). From the local study by Spitzer et al. (2008), the GWR analysis, and the intensity of influence analysis it was concluded that the population settles preferentially in areas below 1600 m a.s.l. that have a slope of less than 30 and southerly aspects between 100 and 300 . This information was used to generate physicalegeographical probability raster images representing settlement probabilities between 0 and 1. The results presented in Fig. 2 show probability values of less than or equal to 0.45, 0.6, or 0.75 (from left to right). In order to estimate the area of future agricultural zones a trend extrapolation was completed (based on CLC data from 1990, 2001, and 2006), which estimated the future number of agricultural cells. The same procedure was followed to delineate “forest areas”, again using a trend extrapolation to assign the future number of cells zoned as “forest areas”. The extrapolated numbers for “agricultural areas” and “forest areas” are listed in Table 2. The trend extrapolation estimates a reduction of 10 km2 in “agricultural areas” by 2050. The number of future agricultural cells in 2050 is thus estimated to be 26,692. Present-day “agricultural areas” within the case study area were found to be restricted to below 1700 m a.s.l. and to slopes of less than 45 . Spatial analyses of the area did not lead to any convincing conclusions concerning aspect and hence the limiting factors for the delineation of future “agricultural areas” were elevation and slope, with upper limits of 1700 m a.s.l. and 45 , respectively. The trend extrapolation estimated a reduction of 14 km2 in “forest areas”, which translates to 29,287 cells in 2050. The development of “forest areas” does not have any limitations in this model approach (e.g. they are not restricted by altitude or slope). All remaining cells not zoned as “building areas”, “agricultural areas”, “forest areas”, or “water areas” are zoned as “natural areas”.

Population distribution and development In the future scenario each cell zoned as “building areas” was assigned a population value. The population numbers for 2050 were based on figures from Hanika (2010), who projected a population of around 158,000 in 2050 for the case study area. This number was taken as the target for the model's future population, distributed across the case study area. The population distribution


Table 2 Absolute changes in area for “agricultural areas” and “forest areas” between 1990 and 2050 (extrapolated). Year

1990 2000 2006 Milestone year 5 Milestone year 4 Milestone year 3 Milestone year 2 Milestone year 1 e 2050 Change in area

Land use Agricultural areas (km2)

Forest areas (km2)

429 428 425 425 423 421 419 419 10

472 470 468 467 465 462 460 458 14

for 2006 (the last year of CLC data) was unfortunately not available in raster format and data from 2008 was therefore used. Each cell in “building areas” was assigned a threshold limiting the maximum number of inhabitants per cell. The settlement density (inhabitants per km2 of “building area”) within the case study area in 2011 was, according to Statistik Austria (2011), 4244 inhabitants per km2. This represents a maximum of 66 inhabitants per (125  125 m) cell, which was therefore used as a limiting factor for the model. The distribution of the future inhabitants was calculated sequentially over those cells zoned as “building areas”. As soon as a cell's threshold was reached it was marked as “closed” and no further additions could then be made to the population. The model then continued with the distribution by assigning additional population to the closest suitable neighbouring cell (in any direction: N, E, S, or W), provided it was zoned as a “building area” and had not yet reached the population threshold. The model completed the population distribution as soon as all the inhabitants (about 158,000) had been allocated to cells, and a map of this future scenario was then produced. Public transport network The public transport data for the case study area used in this model was derived from the OpenStreetMap (OSM) website. However, only the railway connections were taken into account within the case study area in order to ensure that the land use classes, especially the “building areas”, were developed close to public transport hubs, so that each individual cell would be serviced and ensured of good connections to the public transport system (Batty, Xie, & Sun, 1999). Additionally bus stops were included, which may be considered herein as “small” transport hubs. Special topic: sustainable areas The special topic of “sustainable areas” is included here since the model allows different themes and topics to be included when designing the future scenario. It is represented by an additional

Fig. 2. Physicalegeographical probability raster images for the case study area: less than or equal to 0.45, less than or equal to 0.6, and less than or equal to 0.75 (from left to right).


E. Haslauer / Applied Geography 58 (2015) 128e140

Fig. 3. Sustainability raster as additional probability raster in the scenario model: dark grey areas within the case study area are considered to have potential for sustainable development.

probability value (in addition to the physicalegeographical-based probability) that depends on sustainability criteria. These criteria are mainly based on the expert interviews and comprise:  Preferential establishment of new building areas in “service areas” within a 500 m radius of bus stops.  Preferential establishment of new building areas within 500 m “buffer zones” along the existing railway network. The service areas and buffer zones were defined according to a number of local spatial planning documents and guidelines (c.f. Schnürch, 2011). The Spatial Development Concept for the City of Salzburg, for instance, defined a 500 m buffer radius around bus stops as a walkable service area (Magistrat Salzburg, 2009). Jermann (2002) suggested a radius of 400 m for bus stops and 750 m for local railway stations, while Gerike (2005) suggested that there should be a stop of one kind or another within 300e500 m. These references were used to arrive at the figure of 500 m for service areas around bus stops and railway stations, and for the buffer zones along railway axes. These areas were taken to represent sustainable living areas with respect to public transport accessibility. Since there are no concrete plans for any extension of the railway network within the case study area, the present network was also used for the future scenario. All criteria were overlaid to create the sustainability raster, which is presented in Fig. 3. Scenario model In the scenario model the user is able to vary the following settings:  The target population for the future scenario: in the presented future scenario the population was set to 158,000, which approximates the population projected for 2050 by Hanika (2010).

 A population surplus added at each model step: these figures facilitate the process of population redistribution and land use changes.  The parameter defining the nature of the population increase between the model steps (e.g., linear increase, sub-linear exponential increase, or super-linear exponential increase).  The future number of agricultural cells.  The future number of forest cells.  The parameter defining the nature of the increase in the number of agricultural cells or forest cells between the model steps (e.g., linear increase, sub-linear exponential increase, or super-linear exponential increase).  The probability of converting non-building areas into “building areas”, and vice versa.  The maximum possible rate for converting vacant “building areas” into “natural areas”, “agricultural areas”, or “forest areas”. To generate a future scenario for 2050 the model performs the following process steps: (1) The model starts by adding a certain number of people to the case study area at each model step, according to specific rules. All cells are populated until a certain threshold is reached which, in the presented case study, was 66 inhabitants per cell (Statistik Austria, 2011). People are only added to cells where the physicalegeographical suitability is above a certain threshold and where the suitability criterion allows for such additions. If the cell selected for an increase in population is not classified as “building area”, a conversion probability is determined for the cell converting into a “building area”. This is achieved taking into account the neighbourhood effect. Increasing the population initiates a plus of “building areas” preferably in a neighbourhood

E. Haslauer / Applied Geography 58 (2015) 128e140


(3) (4)


dominated by “building areas”. This activity has a high probability of converting other land use classes to “building areas”. This emphasizes the fact that, in order to change the land use classification of a cell to “building areas” there must be a certain number of other cells with similar classifications within the neighbourhood (the neighbourhood effect). The population was increased by a fixed value, which simply facilitates population re-location within the modelled region; in this particular exercise the number added in each step was 5000. The population increase was removed again. Vacant “building areas” are converted into “natural areas”, “agricultural areas”, or “forest areas” according to the conversion probabilities defined in the model settings. Cells are converted into agricultural or forest cells in order to reach the desired number of cells for these land use classes. To change the land use classification of a cell to “agricultural areas” or “forest areas” there must again be a certain number of cells in the neighbourhood with similar land use classifications (the neighbourhood effect).

Backwards modelling The backwards-running model starts from the defined future scenario. The model, which runs on a reverse timeline, identifies milestones during the backwards-running simulation producing new land use patterns for each milestone, with the patterns of these milestones becoming increasingly similar to the present situation as they approach the present time. The final modelled milestone represents the present situation and the model is therefore a forced model. The user can choose the number of milestones to be generated by the model. In this exercise the model produced five milestones between 2006 (the latest date for which CLC data was available) and 2050, yielding a milestone approximately every nine years. A period of about ten years between milestones can be considered a suitable timespan since relevant (although not legally binding) development plans in Austria, such as the Austrian Development Concept (c.f. Austrian Conference on Spatial Planning, 2011) have a revision period of about 10 years. During the backwards modelling the probability of land use changes is evaluated by assigning a random change value (between 0 and 100%) to each raster cell at each time step. This random change value represents the probability of a change in land use classification in a cell from one milestone year to the next. Such a change takes place if the random change value is less than or equal to 5. The principal involved in such a change is illustrated in Fig. 4. In Fig. 4 a cell zoned “agricultural areas” in the future scenario is assigned a random change value greater than 5 in the first two milestones. In the third milestone a value less than or equal to 5 is assigned and the land use class of the cell changes. Since each change is influenced by the present situation (because the status of a cell always tends towards the present situation) this cell changes its zoning to “building areas”, which is the land use class at present. In the backcasting model the user can vary the following settings:  The number of model steps, representing the milestones.  The maximum population per cell (population density).  The probability of CLC change: in the backcasting model all cells are processed successively. A random number is assigned to each cell; if this value is below a certain threshold the cell is set to its status in 2006, and if not the status remains as it is. This is the stochastic principle of this model.  The development power, which defines the nature of the population development between model steps (e.g., linear increase,


sub-linear exponential increase, or super-linear exponential increase).  The population variability, which is a random variance: if, for example, this value is 0.05 the total population varies ±5%. The backwards running model starts by (1) assigning random change values to each cell that ranges between 0 and 100, such that a value of less than or equal to 5 indicates a change in the cell's status if it has not yet reached its state for 2006, (2) relocating any population that is not located in building areas, and (3) distributing the population at random during the re-location process, in cells zoned as “building areas” e taking into account the maximum population density per cell. The model is a stochastic model with a given starting point (the land use pattern of 2050) and a given end-point (the land use pattern of 2006), with interpolation between these points. Implementation A literature review identified the modelling approach used in Cellular Automata (CA) as being the most suitable. A cellular automaton can be described as a discrete processing mechanism that is characterized by internal states (Torrens & Benenson, 2005). CA are simple mathematical models used to model discrete dynamic systems. They evolve through a repeated application of simple, deterministic rules (Wolfram, 1982). The strength of these models is the simplicity of the implemented rules, which allows more flexible system behaviour than in other models. CA models can, for instance, be used to simulate the transition from non-urban to urban landscapes in a system such as cells within a grid space (Batty et al., 1999; Kim & Batty, 2011). In a CA model changes in land use for a particular cell can depend on the previous land use at this location, on multi-criteria factors, and on the land use in its neighbouring cells (Kim & Batty, 2011). In this approach a cellular automaton simulates changes in equal-area polygon cells with an edge length of 125 m, covering the entire case study area (c.f. Batty et al., 1999); the land use type of the cells changes through the repeated application of simple rules and is dependent on the land use in neighbouring cells. Results Having generated the future scenario (based on interviews with relevant experts and taking into account the land use pattern in 2006, physicalegeographical parameters, the population distribution, the railway network, and a suitability raster) the model then generates a possible future land use pattern for 2050. A possible future land use pattern for 2050 and the associated future population distribution are depicted in Fig. 5. The legend for Fig. 5 shows the 5 simulated land use classes of “natural areas” (pale green, in the web version), “building areas” (red, in the web version), “agricultural areas” (orange, in the web version), “water areas” (light blue, in the web version), and “forest areas” (green, in the web version), and is the same for all land use maps presented in this paper. Similarly, the legend indicating the number of people per raster cell is the same in all of the presented maps, with blue (in the web version) indicating zero population, light blue (in the web version) indicating a population of less than or equal to 5, yellow (in the web version) indicating between 6 and 25, pale red (in the web version) indicating between 26 and 50, and red (in the web version) indicating more than 50 persons per raster cell. The land use classes in 2006 and the population distribution in 2008 are depicted in Fig. 6, for the sake of comparison.


E. Haslauer / Applied Geography 58 (2015) 128e140

Fig. 4. Principle of the backwards modelling.

Fig. 5. Possible future scenario for 2050 showing the pattern of land use classes (left) and population distribution (right).

In order to indicate numerically the major changes in land use, absolute changes in area between 2006 until 2050 for the five land use classes are presented in Table 3. The backcasting model shows how the generated future scenario converges towards the pattern of the present situation (in 2006). The number of milestones in the model can be set by the user. Fig. 7 shows the input file for the model (i.e. the generated

land use pattern for 2050), four intermediate outputs (the milestones) over part of the area (enlarged for clarity), and the land use pattern in 2006. The milestones are separated by regular time periods of about 9 years. The associated population development and population distribution between the starting point in 2050 and the end point in 2006 are presented in Fig. 8. The milestones are also depicted

Fig. 6. Land use classes in 2006 (left) and population distribution in 2008 (right).

E. Haslauer / Applied Geography 58 (2015) 128e140 Table 3 Absolute changes in area for land use classes between 2006 and 2050 (simulated), within the case study area. Year

Land use Building Agricultural Natural Water Forest areas (km2) areas (km2) areas (km2) areas (km2) areas (km2)

2006 64 2050 70 þ6 Absolute changes in area

425 417 8

41 53 þ12

37 37 0

468 458 10

showing the population development in between the two end points (again shown over an enlarged part of the area for better visualization). The total population modelled for 2050 was 158,000 (based on the figure from Hanika, 2010), while in 2008 (the last year for which rasterized population data was available at the commencement of this study) the population was 145,119 (Statistik Austria, 2008a). This number was therefore used as the population for 2006 in the backwards running model. For this exercise the population development was set to superlinear, which meant that population decreased more rapidly during the early years of the backwards modelling (from 2050), with the rate of decrease flattening off in later years (towards 2006). The rate of change in the individual land use classes decreased from the initial state (in 2050), via the milestones, back to 2006, and are shown in Table 4.


In order to determine how accessible “building areas” would be in 2050 (according to the model) the areas of all land use classes within a 500 m “buffer zone” along railway lines and a 500 m radius “service areas” around bus stops were analyzed (Table 5). The total area classified as “building areas” within these “buffer zones” and “service areas” was shown to increase between 2006 and 2050 which is a favourable and realistic development, since none of the other land use classes have the same requirement for good access to public transport. These numbers show a reduction in “agricultural areas” and “forest areas” from 2006 to 2050, with increases in the other two land use classes (“building areas” and “natural areas”). The increase in “building areas” reflects the model locating additional “building areas” preferentially within 500 m of railway lines or bus stops, which is a favourable development. “Water areas” did not change at all, which is in accordance with the model's settings.

Discussion and conclusion The backcasting model has been used successfully for a case study area in Salzburg, Austria, to demonstrate its ability to support sustainable spatial development planning. The model presented herein was split into two parts: a model to generate a future scenario and a backwards running model. To the author's best knowledge, this is the first time that backcasting has been applied as an automated, spatially explicit model; backcasting has to date only been used a-spatially as backcasting exercises have mostly

Fig. 7. A future scenario, milestones created during the backwards modelling of land use classes, and the land use pattern in 2006.


E. Haslauer / Applied Geography 58 (2015) 128e140

Fig. 8. Milestones created during the backwards modelling of population distribution.

been performed in workshops, or in administrative units on the basis of statistical reports. The spatially explicit use of the backcasting model has been demonstrated in the case study described in this paper. The presented model is, however, a rather simple version (or model) of reality and does not truly reflect all spatial processes between cells but simply uses spatial relationships to invoke land use changes. Reality is of course more complex, which needs to be addressed in any future model developments based on the presented framework. To start the model a trend extrapolation (or forecast exercise) was carried out. This analysis took into account land use patterns, physicalegeographical parameters, population numbers, population distributions, and a suitability raster. Future patterns of land use and the associated population distribution were then modelled on the basis of these inputs. One possibility would be to generate a completely new

scenario, but it is also possible to start the backwards modelling from a scenario defined in a workshop by relevant experts. The backcasting model's outputs are time slices representing snapshots of land use patterns at particular times (with no time continuum). The model is implemented in Python open source software and acts as a cellular automaton in which equal-area cells cover the entire case study area. These cells change their land use classification according to a simple, repeatedly applied rule-set. The output from the model demonstrates that answers obtained from interviews with appropriate experts can be successfully turned into normative scenarios, and that the attainability of these scenarios can be simulated (or modelled) using a spatially explicit backcasting model. The author suggests that the proposed backcasting model can provide valuable support for sustainable spatial planning by

Table 4 Changes in area (km2) of land use classes and population estimated by the backcasting model. Building areas

Agricultural areas

Future scenario (2050) Milestone 1 Milestone 2 Milestone 3 Milestone 4 2006

70 68 66 65 65 64

417 420 422 423 424 425

Change in area (km2): 2050e2006



Natural areas

Water areas

Forest areas

Population (super linear decrease)

53 49 46 44 43 41

37 37 37 37 37 37

458 461 463 465 466 468

158,000 154,665 151,761 149,234 147,034 145,119




E. Haslauer / Applied Geography 58 (2015) 128e140 Table 5 Changes in area for the different land use classes within 500 m of railway lines and bus stops. Land use class

Area in 2006

Area in 2050

Change in area (km2)

Natural areas Building areas Agricultural areas Water areas Forest areas

2.5 55.4 182.7 9.5 72.9

5.6 63.2 175 9.5 69.7

þ3.1 þ7.8 7.7 0 3.2

Total (km2)




offering possible future development steps that can be taken in order to reach a certain desired future scenario e preferably a sustainable future state e and by identifying development pathways that can be followed to attaining the desired future scenario. The valuable support offered by the model, together with its usability, is underpinned by the fact that (1) the model can back-cast any future scenario that satisfies the model's requirements, and (2) the model considers long-term perspectives. The latter point is particularly important when defining sustainability science (Swart et al., 2004). The backcasting approach paves the way for decisionsupport systems to be integrated in decision makers' workflows. Decision makers in spatial planning have to decide what to alter and what to leave as it is; if “sustainability” and “development” are coupled together in decision making it implies that changes will need to be made to plans for future development. These changes need to be identified, analyzed, and delimited (Blaschke, 2006). Rather than presenting any new software, this paper presents an innovative use of existing open source software to set up a model that can, for the first time, be used to implement theoretical backcasting. The model is not limited to sustainability issues such as those selected for this particular case study; other objectives such as, for instance, to maximize the quality of life in a region, can also be supported by simply changing the sustainability raster. The author sees this approach as forming a bridge between spatial planning concepts and strategies (including sustainable development) and Geographic Information Science, as the latter can be used to support model-based innovative solutions, alternatives, and developments in spatial planning problems. Outlook Further work on the proposed model will involve investigating its transferability to other case study areas. Various other testing sites are currently being analyzed, one of which will soon be selected for further investigation; a sensitivity analysis is planned for the backcasting model in order to evaluate the effects that changes in the model settings have on the spatially explicit model outputs. Acknowledgement The research was funded by the Austrian Science Fund (FWF) through the Doctoral College GIScience (DK W 1237-N23). The author wants to thank Markus Biberacher and Thomas Blaschke for supporting the research, Roland Hufnagl for his programming support, and Ed Manning for his many valuable comments and for proof reading the article. Finally, thanks to the reviewers for their comments which improved the paper substantially. References Austrian Conference on Spatial Planning. (2011). Austrian spatial development concept. Vienna. Web


m_u._Region/1.OEREK/OEREK_2011/Dokumente_OEREK_2011/OEREK_2011_ EN_Downloadversion.pdf, 16.04.14. Batty, M., Xie, Y., & Sun, Z. (1999). Modeling urban dynamics through GIS-based cellular automata. Computers, Environment and Urban Systems, 23, 205e233. Blaschke, T. (2006). The role of the spatial dimension within the framework of sustainable landscapes and natural capital. Landscape and Urban Planning, 75, 1198e1226. Botequilha Leitao, A. B., & Ahern, J. (2002). Applying landscape ecological concepts and metrics to sustainable land planning. Landscape and Urban Planning, 59, 65e93. Brown, B. J., Hanson, M. E., Liverman, D. M., & Merideth, R. W. (1987). Global sustainability: toward definition. Environmental Management, 11, 713e719. Carlsson-Kanyama, A., Dreborg, K.-H., Moll, H. C., & Padovan, D. (2008). Participative backcasting: a tool for involving stakeholders in local sustainability planning. Futures, 40, 34e46. Dreborg, K.-H. (1996). Essence of backcasting. Futures, 28, 813e828. EEA e European Environmental Agency. (2006). CORINE (coordination of information on the environment) e Landcover data e A program of the European Commission. Gerike, R. (2005). Wie kann das Leitbild nachhaltiger Verkehrsentwicklung konkretisiert werden? e Ableitung grundlegender Aufgabenbereiche [How can the Leitbild of sustainable traffic development be substantiated?] (Dissertation). TU Dresden. ^t-Regamey, A., & Brunner, S. (2011). Methodischer Rahmen für den Einsatz von Gre Backcasting zur Anpassung an den Klimawandel [Methodological framework for the application of backcasting on climate change issues]. disP, 184, 43e51. € €umige Bevo €lkerungsprognose für Osterreich Hanika, A. (2010). Kleinra 2010e2030 mit Ausblick bis 2050 [Small-scale population projection for Austria 2010e2030 e And outlook until 2050]. Vienna. Haslauer, E., Biberacher, M., & Blaschke, T. (2012). GIS-based backcasting: an innovative method for parameterisation of sustainable spatial planning and resource management. Futures, 44, 292e302. Haslauer, E., Blaschke, T., & Biberacher, M. (2015). A spatially explicit backcasting approach for sustainable land use planning. Journal of Environmental Planning and Management (in press). Haslauer, E., Schnürch, D., & Prinz, T. (2013). A spotlight on mobility and interconnection in rural vs. urban areas. Conference paper. In A. Car, T. Jekel, & J. Strobl (Eds.), GI forum 2013, Salzburg (pp. 177e186). Heidelberg: Wichmann-Verlag. Hickman, R., & Banister, D. (2005). If, at first the idea is not absurd, then there is no hope for it: Towards 15 MtC in the UK transport sector. Conference paper at the 45th ERSA conference, Amsterdam, August 23e27, 2005. Hickman, R., & Banister, D. (2007). Looking over the horizon: transport and reduced CO2 emissions in the UK by 2030. Transport Policy, 14, 377e387. Holmberg, J. (1998). Backcasting: a natural step in operationalizing sustainable development. Greener Management International, 23, 30e51. €tze der GIS-gestützten PotenzialberJermann, J. (2002). Vergleich verschiedener Ansa echnung [Various approaches of a GIS-based potential estimation]. Conference paper at the 2nd Swiss Transport Research Conference, March 20e22, Monte , Switzerland. Verita Kim, D., & Batty, M. (2011). Calibrating cellular automata models for simulating urban growth: Comparative analysis of SLEUTH and METRONAMICA. UCL working papers series 176. Magistrat Salzburg. (2009). Die zukünftige Entwicklung der Stadt Salzburg. €umliches Entwicklungskonzept der Stadt Salzburg e REK 200 [The future Ra development of the city of Salzburg. Spatial development concept e REK 2007]. Salzburg: Schriftenreihe zur Salzburger Stadtplanung 35. Miola, A. (2008). Backcasting approach for sustainable mobility. JRC scientific and technical reports. European Commission, Joint Research Centre - Institute for Environment and Sustainability. OGD (Open Government Data) Land Salzburg. (2014). Digital elevation model of the province of Salzburg (10 m resolution). Web gdClient/, 19.09.14. Osidele, O. O., & Beck, M. B. (2003). An inverse approach to the analysis of uncertainty in models of environmental systems. Integrated Assessment, 4, 265e282. Quist, J. (2007). Backcasting for a sustainable future: The impact after 10 years. Delft: Eburon. Quist, J., & Vergragt, P. (2006). Past and future of backcasting: the shift from stakeholder participation and a proposal for a methodological framework. Futures, 38, 1027e1045. Raskin, P., Chadwick, M., Jackson, T., & Leach, G. (1996). The sustainability transition: Beyond conventional development. Stockholm: Stockholm Environment Institute. Robinson, J. (1990). Futures under glass: a recipe for people who hate to predict. Futures, 22, 820e843. Schnürch, D. (2011). Grenzüberschreitende Wohnstandortbewertung und -analyse in der Europaregion Salzburg [Trans-border analysis and assessment of residential locations in the “Europa-Region Salzburg”] (Master thesis). Salzburg: University of Salzburg. Siedentop, S. (2005). Urban sprawl e verstehen, messen, steuern [Urban sprawl e understanding, measuring, steering]. disP, 160, 23e35. Spitzer, W. (2007). Zersiedelung e Quantifizierung eines mehrdimensionalen Begriffs [Urban sprawl e Quantification of a multidimensional concept] (Master thesis). University of Salzburg. Spitzer, W., Dollinger, F., & Prinz, T. (2008). Neuabgrenzung des Dauersiedlungsraumes für das Land Salzburg [Re-delineation of the permanent settlement area in the province of Salzburg]. In J. Strobl, T. Blaschke, & G. Griesebner (Eds.), Angewandte Geoinformatik 2008 (pp. 655e660). Heidelberg: Wichmann Verlag. Squires, G. D. (2002). Urban sprawl: Causes, consequences & policy responses. Washington: The Urban Institute Press.


E. Haslauer / Applied Geography 58 (2015) 128e140

Statistik Austria. (2008a). Regionalstatistische Einwohnerraster 2008 [Regional statistical population raster 2008]. €lkerung der Statistik Austria. (2008b). STATcube e statistical database. Bevo € €nder 2008. Web Osterreichischen Bundesla cations_services/superstar_database/index.html, 17.04.14. €lkerung der Statistik Austria. (2011). STATcube e statistical database. Bevo € €nder 2011. Web Osterreichischen Bundesla cations_services/superstar_database/index.html, 17.04.14. Swart, R. J., Raskin, P., & Robinson, J. (2004). The problem of the future: sustainability science and scenario analysis. Global Environmental Change, 14, 137e146. Torrens, P. M., & Benenson, I. (2005). Geographic automata systems. International Journal of Geographical Information Science, 19, 385e412.

Van Berkel, D., & Verburg, P. (2012). Combining exploratory scenarios and participatory backcasting: using an agent-based model in participatory policy design for a multi-functional landscape. Landscape Ecology, 27, 641e658. Van de Kerkhof, M., Hisschemller, M., & Spanjersberg, M. (2002). Shaping diversity in participatory foresight studies. Greener Management International, 37, 85e99. €chter, P., Ornetzeder, M., Rohracher, H., Scheurer, A., & Knoflacher, M. (2012). Wa Towards a sustainable spatial organization of the energy system: backcasting experiences from Austria. Sustainability, 4, 193e209. Wolfram, S. (1982). Cellular automata as simple self-organizing systems. Caltech preprint CALT-68-938. Web demic/statistical-mechanics-cellular-automata.pdf, 22.08.12.