GIS-based Backcasting

A Method for Parameterisation of Sustainable Spatial Planning and Resource

Management

Eva Haslauer

, Thomas Blaschke

and Markus Biberacher

Interfaculty Department of Geoinformatics, University of Salzburg, DK GIScience, Schillerstraße 30, 5020 Salzburg,

Austria

Research Studios Austria, Studio iSPACE, Schillerstraße 25, 5020 Salzburg, Austria

1 ABSTRACT

Backcasting, if used as a decision support and

planning method, starts from a desired future state or

vision and simulates backwards until present time.

During the model run, which goes backwards in

time, development paths are created. They expose

which steps have to be taken in future to reach the

desired state. Besides, milestone scenarios are

created as outputs that represent interim goals.

The paper at hand proposes an automated, GIS-

based backcasting model, since backcasting so far

has only been applied in workshops or as theoretical

framework. Until now no spatially explicit

backcasting model has been set up. The proposed

backcasting model first creates a future scenario

utilizing an Agent Based Model approach.

Afterwards the model simulates backwards

implemented as a Cellular Automaton. This is

realized in a Python script and linked to the Open

Source GIS software Quantum GIS.

The general model is applied to a case study in

Salzburg, Austria. The topic concerns sustainable

spatial planning. The results of the model run show

in time steps a successful backcasting of land-use

classes from a future state back until present time.

2 STAGE OF THE RESEARCH

The preliminary title of my PhD work is ‘A GIS-

based Backcasting: An innovative method for

parameterisation of sustainable spatial planning and

resource management’. I started my PhD in

February 2011 and I am now in the final year. In

September 2011 I joined the Doctoral College

GIScience at the University of Salzburg which

provides a professional and excellent network of

students and faculty for potential collaborations and

knowledge exchange.

The initially defined research questions of my

PhD work are:

1. Can backcasting exercises be reasonably

translated into models?

a. Which options could be used?

2. Are Geosimulation tools applicable to a

backwards working approach like backcasting?

b. Can Cellular Automata and/or Agent Based

Models be used?

I meanwhile published one PhD-relevant article

in the ISI-referenced journal Futures. Its title is

similar to my PhD topic, namely ‘GIS-based

Backcasting: An innovative method for

parameterisation of sustainable spatial planning and

resource management’ and was published online in

November 2011. This paper covers the first ideas

and a theoretical framework of the backcasting

model. Since then I was working on the

implementation of the backcasting model itself,

which is now in the final stage.

Currently I am working on my second PhD

relevant article which is supposed to be submitted in

July to the journal ‘Environmental modelling &

software’. It presents the implementation of the

model and development steps towards the spatially

explicit model.

3 OUTLINE OF OBJECTIVES

The main objective of my PhD is to develop a

spatially explicit backcasting model. To explore the

principle of backcasting, which is not necessarily

immediately understood correctly by non-experts, I

may start with the often better known principle of

forecasting. In a forecasting exercise an analyst

usually takes a present situation and applies

forecasting methods as for instance trend

extrapolations or time series to explore how a future

Haslauer E., Blaschke T. and Biberacher M..

GIS-based Backcasting - A Method for Parameterisation of Sustainable Spatial Planning and Resource Management.

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

situation may look like or what the directions of

future trends are. It is a conditional method saying

‘If policy A is adopted, state B will happen.’

Forecasting is for instance applied in meteorology or

economics (c.f. Murphy & Winkler, 1974; Clements

& Hendry, 1998; Clements et al., 2004). Contrarily,

a general backcasting model starts from a future

vision, resp. a desired future state decoupled from

currently given structures. From this future state the

backcasting model simulates backwards a

development until the present state is reached (see

Fig. 1).

Figure 1: Backcasting principle.

The simulation is based on rules and creates

milestone scenarios as interim goals. Next to the

interim goals the simulation identifies trajectories.

They serve as development paths or guidelines and

express which steps have to be taken in order to get

from the present to the desired future state.

Backcasting is not a conditional method, as it is

forecasting, but it tells the modeller what needs to be

done and what needs to be achieved during the next

years in order to reach the (desired) future state.

The proposed backcasting model first creates a

future scenario. This part of the model utilizes an

Agent Based Model. The second part of the model

simulates backwards starting from the future state.

This part is implemented with a Python script and

linked to the Open Source GIS software Quantum

GIS.

The model is applied to a case study of

sustainable land-use planning in Salzburg, Austria.

The model shall support to reach a vision in land-use

planning in an appropriate time horizon. Thus the

results of the model run show in 10 years’ time steps

a successful back-casting of land-use classes from a

desired future state back until present time.

4 RESEARCH PROBLEM

Although the basic ideas of backcasting are

relatively old, general backcasting is until now

usually applied in non-spatial and non-explicit

context. Today, backcasting is often applied to back-

cast scenarios for whole cities or regions, for

instance, and often performed in workshops. The

unique characteristic of the presented backcasting

approach is its spatial explicity and automation: The

simulations in the proposed model are based on

polygonal equi-areal raster cells covering a selected

pilot region. So far no tangible method exists for the

application of the backcasting concept with spatial

reference and to my best knowledge also no model-

based, spatially explicit backcasting analysis has

been implemented yet.

An automated, GIS-based backcasting model will

reveal new perspectives and opportunities applicable

to research topics like spatial planning, safety,

construction, and health care. The basic research

results of this study will provide new possibilities

for the integration of decision support systems into

the workflow processes of spatial planning

specialists and decision makers.

The backcasting model will be applied in a pilot

region located in Salzburg, Austria. The model shall

help to counteract negative consequences of urban

development in this region by proposing a desired

future vision and related development paths how to

reach this future state.

This research forms a bridge between spatial

planning concepts and Geographic Information

Science and adds a new dimension to GIS. At the

same time it contributes to the provision of

innovative solutions.

5 STATE OF THE ART

The origin of the Backcasting concept goes back to

the 1970s, when it was introduced by Amory Lovins

as a planning method for electricity supply and

demand. At that time he referred to it as ‘backwards-

looking-analysis’. Since then, this method has been

regularly used in energy studies, often being referred

to as ‘energy analysis of the final consumption’

(Dreborg, 1996; Quist, 2000). According to Dreborg

(1996), general backcasting can be applied if:

 Complex problems are dominating

 Problems include a major trend

 External factors greatly influence the problems

 The time span for implementing changes is

SIMULTECH2014-DoctoralConsortium

long enough

Following the first implementation of

backcasting it took some time before it was realised

that the backcasting approach can be used in a much

wider range of applications. Sustainability became

an important field of application: In Sweden it has

been applied to various topics of sustainability

research, like transportation systems and air

transport. Further studies have been conducted on

the sustainability of water, mobility, or households

(Carlsson-Kanyama et al., 2008; Miola, 2008,

Banister et al. 2000 cited in Quist 2007). In Oxford,

backcasting has been applied e.g. within the VIBAT

project. There it was implemented to lower transport

carbon emissions (Hickman et al., 2009). A

Backcasting analysis aiming at an efficient energy

system with regard to the development of a

sustainable spatial organisation in Austria was

proposed by Wächter et al. (2012). Grêt-Regamey

and Brunner proposed a new approach in 2011,

namely to split the backcasting methodology into

inverse modelling (Grêt-Regamey & Brunner, 2011)

for explicit quantitative problems on the one hand

and a qualitative, strategic backcasting analysis for

complex problems in spatial planning on the other

hand. The latest research in this field dealt with the

implementation of a backcasting analysis with Agent

Based Models (ABMs). This was proposed by Van

Berkel and Verburg (2012) and was applied to

design policies for multifunctional landscapes to

simulate landscape changes.

6 METHODOLOGY

The proposed steps in my PhD towards a spatially

explicit and model-based backcasting approach

include (1) interviews with experts in the field of

spatial and urban planning. They have been selected

based on responsibilities for planning issues in the

pilot region. (2) Visions derived from the experts’

views. (3) Extensive analysis of the current state

including the land-use pattern and influencing

physio-geographical parameters, the population

distribution, and the future population development

in the pilot region. After having characterized the

pilot region (4) normative future scenarios are

created. In backcasting, future scenarios are often

visionary and decoupled from given structures or

circumstances, assuming that there are no constraints

for development. In the proposed approach the

developed future scenario represents a randomly

generated possible future pattern of land-use classes

in the pilot region. The development of the future

scenario is only influenced by exogenous parameters

which have been identified before to influence land-

use in the pilot region. The future scenario serves as

starting point for the next step (5), the backwards

modelling. This is performed by repeatedly applying

pre-defined rule-sets describing the backwards

development. The backwards modelling simulates

changes of equi-areal polygonal cells. Each of these

cells changes its state individually, resp. explicitly

from one time step to the next. These time steps are

represented by milestone scenarios that are created

as outputs during the whole backwards run,

representing interim goals. Further outputs of the

backwards modelling are development paths or

trajectories, revealing which changes have to take

place to reach the future scenario. Finally the

backwards modelling results in a modelled present

land-use pattern. This is compared with the actual

present land-use pattern in the pilot region which

should equal each other.

Based on a profound literature review two

principle modelling techniques have been identified

to be potentially suitable for the backcasting

approach. On the one hand these are Agent Based

Models (ABMs) which in general simulate actions

of autonomous agents, and display the behaviour

and dependencies between those agents and to other

agents. The aim of an ABM is usually not to reach

equilibrium, but to find out how a system will adapt

to changed conditions (Macal et al., 2010). On the

other hand these are Cellular Automata (CA). CA

are considered to be simple mathematical models

used for modelling discrete dynamic systems. These

models evolve due to a repeated application of

simple, deterministic rules (Wolfram, 1982). The

strength of those models is the simplicity of their

implemented transition rules which provide more

flexible system behaviour than other models. CA

can be applied to simulate the transition, for instance

from non-urban to urban landscape, of a system like

cells in a grid space depending on the

neighbourhood: Land-use changes at a certain cell

can thus depend on a previous land-use at this

location, on multi-criteria factors, or on the land use

of its neighbouring cells. These approaches allow a

numerical analysis of non-numerical geographic

systems. (Kim & Batty, 2011)

Based on an internal evaluation of strengths and

weaknesses of both approaches and discussions with

other experts, ABMs were chosen to be used for the

setup of the backcasting model. Nevertheless, the

ABM does not simulate individual agents moving

over a rasterized pilot area, but it simulates the

GIS-basedBackcasting-AMethodforParameterisationofSustainableSpatialPlanningandResourceManagement

changes of equi-areal polygon cells covering the

whole pilot area. It can be said that the ABM is

utilized as a CA.

The proposed model is applied to a selected pilot

region in Austria and is divided into two parts: The

first part creates future scenarios and is implemented

in ESRI’s Agent Analyst. This is an Agent Based

Modelling (ABM) extension for ArcGIS and builds

on Not Quite Python (NQPy) language. NQPy

satisfies nearly all requirements of GIS users, who

can also export the models and can add more

sophisticated functions.

Due to performance criteria the backwards

running process, consequently the second part,

which starts from a given future scenario, was

implemented as a Python script that is linked to the

open-source GIS software Quantum GIS (QGIS).

7 OUTCOME

The first results of the backcasting model present a

randomly generated future land-use scenario for

2050. This is input to a backwards running model

starting from 2050 and ending in 2006, as the final

stage. The ABM-based generated random future

scenario for 2050 is shown in Fig. 2.

Figure 2: Randomly generated future scenario of 2050.

This random future land-use pattern consists of the

four dominant land-use classes in the pilot region:

nature, settlement areas, agricultural land, and water

bodies. These data were extracted from Corine

Landcover data of 2006. The land-use class of nature

includes for instance forest areas, pastures,

grasslands, and other vegetated areas. Agricultural

areas include cultivated areas and arable land, and

settlement areas comprise urban areas, industrial

units, road and rail network, and other artificial

facilities (related to leisure, sports, etc.). All

simulated milestone scenarios, as well as the final

output, are made up of these four land-use classes.

To give an impression, two selected milestone

scenarios for the years 2040 and 2020 are presented

in Fig. 3a and b.

Figure 3a: Milestone scenario of 2040.

Figure 3b: Milestone scenario of 2020.

Finally, the output of the backwards running

model, which is similar to the land-use pattern of the

current time (here it is 2006), is presented in Fig. 4.

Figure 4: Final model output: the modelled land-use

pattern of 2006.

This modelled land-use pattern is similar to the

Corine Landcover data of 2006, which represents the

SIMULTECH2014-DoctoralConsortium

initial land-use pattern which was also input for the

model. This will always be the final output since the

backwards running model converges in any case

towards a fixed land-use pattern (in this case it is the

land-use pattern of 2006).

8 DISCUSSION

The aforementioned research questions that are also

addressed in my PhD and can be answered through

the presented outputs are:

1. Can backcasting exercises be reasonably

translated into models?

a. Which options could be used?

2. Are Geosimulation tools applicable to a

backwards working approach like backcasting?

c. Can Cellular Automata and/or Agent Based

Models be used?

With the gained results of the model, presented

in Fig. 2 to 4, it is possible to answer all research

questions: Yes, backcasting exercises can be

reasonably translated into models and yes,

Geosimulation tools are applicable to a backwards

working approach like backcasting. For the

generation of a random future scenario of 2050 an

Agent Based Model extension (Agent Analyst) for

ArcGIS is used. Cellular Automata (CA) models

might be used more often but it is made clear that

the ABM extension was not used in a sense of

studying independently acting agents. This

forecasting exercise resulted in a scenario

considering pre-defined rule-sets for the

development of future land-use patterns. The

backwards modelling is implemented as a Python

script linked to QGIS. This was preferred against the

Agent Analyst due to performance criteria and the

free availability of Python and QGIS. Both model

approaches, in Agent Analyst and Python/QGIS, act

like CA meaning that the equi-areal cells covering

the pilot area change their states according to

implemented rules. This statement also answers

question two, if Geosimulation tools are applicable

to a backwards working approach. Yes they are,

which has been proven by the herewith presented

approach of implementing backcasting with ABM

resp. a CA realized via Python.

9 OUTLOOK

Next developments will include the integration of a

baseline scenario for validation of the model. The

baseline scenario is created with Metronamica, land-

use simulation software, developed by RIKS at the

VU Amsterdam. “Metronamica is a unique generic

forecasting tool for planners to simulate and assess

the integrated effects of their planning measures on

urban and regional development. As an integrated

spatial decision support system, Metronamica

models socio-economic and physical planning

aspects. It incorporates a mature land use change

model that helps to make these aspects spatially

explicit” (Metronamica Website, 2014).

Furthermore a whole pool of future scenarios

will be created each having a different emphasis

representing for instance maximum compactness,

maximum Quality of Life or maximum autarky. The

rules for these scenarios are currently set-up.

An on-going update of the backcasting rule-set

includes the integration of neighbourhood influences

on settlement development and population

development in the pilot region. Preferred areas are

then not only influenced by physio-geographic

parameters but also by the surrounding environment.

Besides, the road network of the pilot region will

be included through an explicit raster input.

Since the two parts of the model, the creation of

a future scenario on the one hand, and the backwards

modelling on the other hand are for now

implemented in different software environments, it

is planned to implement both in the same

environment. Thus, the creation of future scenarios

is transferred to Python scripting as well.

Finally, a sensitivity analysis with respect to the

initial condition, which is obtained by the

backcasting model, is conducted in order to see

whether the same results can be achieved.

ACKNOWLEDGEMENTS

The research was funded by the Austrian Science

Fund (FWF) through the Doctoral College

GIScience (DK W 1237-N23).

REFERENCES

Carlsson-Kanyama, A., Dreborg, K.-H., Moll, H. C.,

Padovan, D., 2008. Participative backcasting: A tool

for involving stakeholders in local sustainability

planning. In: Futures 40, 34-46.

Clements, M. P., Franses, P. H., Swanson N. R., 2004.

Forecasting economic and financial time-series with

non-linear models. In: International Journal of

Forecasting, 20, 169-183.

GIS-basedBackcasting-AMethodforParameterisationofSustainableSpatialPlanningandResourceManagement

Clements, M. P., Hendry, D. F., 1998. Forecasting

economic time series. Cambridge University Press,

Cambridge, New York.

Dreborg, K.-H., 1996. Essence of Backcasting. In: Futures

28, 813-828.

Grêt-Regamey, A., Brunner, S., 2011. Methodischer

Rahmen für den Einsatz von Backcasting zur

Anpassung an den Klimawandel (Methodological

framework for the application of backcasting on

climate change issues). In: disP 184, 43-51.

Hickman, R., Ashiru, O., Saxena, S., Banister, D., 2009.

Visioning and Backcasting for Transport in London.

Stages 3/4, Background Report. Halcrow, London.

Web: www.vibat.org (May 26, 2014).

Kim, D., Batty, M., 2011. Calibrating Cellular Automata

Models for Simulating Urban Growth: Comparative

Analysis of SLEUTH and METRONAMICA. In: UCL

Working Papers Series No. 176.

Macal, C. M., North, M. J., 2010. Tutorial on agent based

modelling and simulation. In: Journal of Simulation 4,

151-162.

Metronamica, 2014. Metronamica Homepage. Web:

www.metronamica.nl (April 16, 2014).

Miola, A., 2008. Backcasting approach for sustainable

mobility. In: JRC Scientific and Technical Reports.

European Commission, Joint Reseqarch Centre -

Institute for Environment and Sustainability.-Ispra.

Murphy, A. H., Winkler, R. L., 1974. Probability

Forecasts: a Survey of National Weather Service

Forecaster. In: Bulletin of the American

Meteorological Society, 55, 1449–1452.

Quist, J., 2007. Backcasting, for a sustainable future: the

impact after 10 years. Eburon, Delft, 2007.

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. In: Landscape Ecology 27,

641-658.

Wächter, P., Ornetzeder, M., Rohracher, H., Scheurer, A.,

Knoflacher, M., 2012. Towards a Sustainable Spatial

Organization of the Energy System: Backcasting

Experiences from Austria. In: Sustainability 4, 193-

209.

Wolfram, S., 1982. Cellular Automata as Simple Self-

Organizing Systems. In: Caltech preprint CALT-68-

938.

SIMULTECH2014-DoctoralConsortium