A Land-Use/Transport interaction model for Austria
Anna MAYERTHALER, Reinhard HALLER, Ginter EMBERGER
Vienna University of Technology - Institute for Transportation - Center of Traffic
Planning and Transport Engineering
Gusshausstrage 30/231, 1040 Vienna, Austria
anna.mayerthaler@ tuwien.ac.at, reinhard.haller@ tuwien.ac.at,
quenter.emberger@ tuwien.ac.at
ABSTRACT
This paper presents an attempt to set up the land-use transport interaction model MARS
(Metropolitan Activity Relocation Simulator) for a nation wide case study of Austria.
The purpose of the model is to capture the most important interactions and feedback
mechanisms between the land-use- and the transport system. To this end we adapted the
urban MARS model. Particular attention was paid to the structural changes of the
model and the estimation of the transport model parameters as well as the land-use
model parameters, which are modelled with a gravity model approach. For this purpose
we used the build-in optimizer of the modelling software Vensim by minimizing the sum
of squared deviations between observed and predicted data. We present the model fit,
estimated parameters and results of a first model run (30 years).
1 INTRODUCTION
Interaction between transport planning, land-use planning and the economy is highly
complex. The numerous feedback loops within and between transport, land-use and
economy are effective on different temporal and spatial levels. As a result even effects
caused by the change of a single policy instrument can be difficult to predict. Because
of the notion that transport and land-use are strongly interrelated, a series of land-use
and transport interaction (LUTI) models have been developed in the past decades
(Wegener 2004). One distinctive feature of this interrelation is that changes within these
two systems occur at significantly different speed. Whereas transport users respond
relatively fast to changes in the transport system, the land-use system is characterized by
a considerable degree of inertia, mainly due to the fact that land-use systems are
embodied in physical structures such as buildings and infrastructure. This makes system
dynamics modelling powerful in modelling land-use/transport interactions.
Most of these models though concern cities or urban agglomerations. This is
understandable as transport and land-use related problems, such as congestion, various
forms of pollution, scarcity of natural land, etc. are most apparent in densely populated
urban areas. However, neither a theoretical point of view nor empirical evidence suggest
that land-use/transport interactions are absent in rural areas. Quite the opposite seems to
be the case many of them have been subject both to significant transport infrastructure
construction and to considerable migration processes.
Another important notion is that it seems that most urban models are relatively custom-
tailored implementations for specific urban case studies, sometimes only loosely related
to more generic model environments. This raises the question of generality of LUTI
models.
To undertake the two research themes outlined above, namely the application of LUTI
modelling to rural areas and an investigation of the generality of the urban LUIT model
MARS, we set up a nation - wide case version of the existing urban LUTI model MARS
(Pfaffenbichler 2003a) for Austria.
The paper starts with a short description of the different model parts and their structure
in section 2. In this section the focus lies in the structural changes that were necessary
for the nation wide case study setup. Section 3 describes the application of the model to
Austria and depicts the case study area. In section 4 first results from model calibration
of the different model parts are presented. We describe the methodology as well as
model fit and estimated parameters. While in section 5 first forecasts conceming
population and workplace development are shown. The paper closes with conclusions
and an outlook in section 6.
2 THE MARS MODEL
2.1 Introduction
The MARS model is a dynamic land-use/transport interaction (LUTI) model, which is
based on the principles of synergetics (Haken 1983). To date the MARS model has been
applied to 10 European (Edinburgh, Gateshead, Leeds, Madrid, Trontheim, Oslo,
Stockholm, Helsinki, Vienna and Bari), 2 Asian (Hanoi, Ubon Ratchathani) and 1 South
American (Porto Alegre) city. Ongoing projects cover setting up the MARS model for
Hoh Chi Minh City in Vietnam and Washington D.C. in the US.
The model description in this paper will focus on the overall model structure and some
specific modules relevant for the issues addressed in this paper. For a more
comprehensive presentation, we refer the reader to Pfaffenbichler (2003a, 2008).
2.2 General model structure
The MARS model consists of sub models which simulate passenger transport, housing
development, household migration and workplace migration; additionally accounting
modules calculate assessment indicators and pollutant emissions. The overall structure
of the model is shown in Figure 1. The main link between the transport model and the
location choice model are accessibilities (formulated as potential to reach workplaces
and shopping opportunities), which are passed on from the transport model to the
location choice models and the spatial distribution of households and employment
which are input from the location choice models to the transport model. The land price
influences both the residential location- and the workplace sub model whereas these two
sum models change the availability of land.
Transport
sub model
Spatial distribution Spatial distribution
residents workplaces
Accessibility Accessibility
a workplaces consumers, workforce
Residential Work place
location sub |«t Land price | sub model
model
ol of a
Figure 1 Subsystem diagram with the three main sub models
2.3 Time dynamics
An important feature of MARS is that it considers rapid transportation dynamics and
slow migration dynamics.
The transport users can react immediately (from one time step to the next) to changes in
the transport system.
Residents who want to change their location need available housing space in the zone of
destination (the time to build new housing units is three years). This availability of
housing units is checked via three time steps, going sequentially from the first best
choice to the third best choice of destination zone. Residents who could not move into
their destination zone, due to the lack of available housing space are added to the
potential in-movers in the next time step.
A similar procedure is valid for the migration of workplaces, in order to move into
another zone MARS controls whether there is enough spatial capacity for workplaces in
the destination zone or not.
Because of this inertia embodied in the system, policy measures affect the different sub
systems of MARS over different timeframes. For example, a transport policy measure
like an increase of parking fees might have an immediate effect on transport users (a
shift in modal spilt to other transport modes), but might also in the long run change the
settlement structure of residents and firms (for example people choose to live closer to
their workplaces to reduce the need for commuting).
2.4 Model parts
2.4.1 | The Transport sub model
The transport model in MARS simulates passenger transport and comprises trip
generation, trip distribution and mode choice. Trip distribution and modal split are
calculated simultaneously by a gravity (spatial interaction) type model.
In the trip generation the number of trips originating or designated for a particular
model zone are calculated. The trip distribution, allocates the total number of trips to all
origin-destination (OD) pairs and the mode choice is the distribution of the trips to the
different modes of traffic, normally specified as percentage share.
The modes considered in the model are slow, car, public transport (bus) and public
transport (rail). The slow mode represents the non-motorized modes walking and
cycling. Due to the zone size in the MARS Austria model, this mode is almost
exclusively relevant for intrazonal trips (except for interzonal trips in Vienna where the
model zones represent municipal districts).
Trip distribution and mode choice in the MARS model are calculated per origin-
destination (OD) pair. Due to the heterogeneity of the case study area (see section 3.1)
we had to improve further the possibility of modelling commuting trips distribution for
intrazonal trips. The model setup was perfectly suitable for the urban case studies, but
the wider geographical scope made some changes in the structure of it necessary.
Therefore we extended the model zones for intrazonal distance classes. For each of the
121 model zones the intrazonal trips are now spilt up to 5 distance classes, where mode
split and number of trips are calculated separately for each zone and distance class.
2.4.2 The land-use sub model and its modules
The residential location model
In the urban MARS model, migration is modelled in a three step approach: first, out-
migration per model zone is estimated. The overall out-migration of the whole case
study is constrained to a given rate; this rate was usually assumed to be 0.06 per year,
equivalent to an average time between two residential moves of 15 years, which is
reasonable for European cities(ODPM 2002). Second, migrants were pooled over the
whole case study. In a third step, the migrants are distributed to destination zones.
Both out-migration, OMj, and in-migration, IMj, are modelled based on a LOGIT
function of the form:
OM. = @Zor@iPOP +aabR+axGL, +anACC
t=
Formula 1 Out-migration formula
IM. = @% ta POP) +a2bRj +a2GL, +asACC,
j
Formula 2 In-migration formula
Ay ly Parameters
POP, Number of residents in a zone J
LR; Land rents in model zone J
GL; Area of green land in model zone J
ACC; Access attractiveness in model zone J
The same set of variables is assumed to influence out-migration and in-migration as
described above. However, the direction and strength of the link between the
explanatory variables and migration, i.e. the parameters of the model, is different for
out- and in-migration.
The geographical location of origin and destination zones is not taken into
consideration. In other words, the model assumes that the destination choice of migrants
is not influenced by the location of their current domicile.
The choice of variables considered (accessibility by car and public transport, level of
housing costs and share of recreational green land) is based on several different lines of
argument: Firstly, they repeatedly rank among the most important determinants of
migration in empirical migration research (ODPM 2002). Secondly, own empirical
studies focusing in particular on Vienna confirmed this importance (Pfaffenbichler
2003b). Third, each of the variables is highly endogenous especially from a land-use/
transport perspective and in an urban context. As an example, in an urban context the
share of green land is both an important cause of migration - in that it constitutes a
major amenity perceived be potential migrants - and is simultaneously influenced by
migration - as new development can significantly reduce this amenity in urban areas.
An earlier attempt to implement the model for Austria without structural changes
revealed the inappropriateness of this structure for a larger spatial scale (Emberger et al.
2007). One major shortcoming was that the observed length distributions of migration
were not reflected in the model output: whereas domestic migration in Austria (and
elsewhere) is largely short-distance, the model predicted significant population shifts
from the West to the East of the country, i.e. over a couple of hundred kilometres.
The assumption of unconstrained destination choice was perfectly justifiable given the
urban applications of the model. In most cases, the case studies were small enough to
make it possible for migrants to maintain large parts of their “everyday life” (including
place of work, social networks, spare time activities, etc.) irrespective of their choice of
residence.
To improve on the model we reviewed literature on migration theory (e.g. Greenwood
1985, Muth 1971, Bode and Zwing 1998) and applied migration models (e.g. ODPM
2002, Flowerdew and Amrhein 1989, Roy 2004). Migration theory states that migrants
evaluate benefits and costs of migration. Migration related monetary costs include
actual costs of migration, the loss of social networks. In most applied work on
migration, due to the intangible nature of these effects, distance is taken as a surrogate
for the various types of migration costs. Moreover, distance also reflect an information
aspect of migration, as people are usually deterred from moving to more distant place
they know less about.
In order to account for the overwhelming importance of distance while changing model
structure as little as possible, we implement a two stage migration model: First, the
number of out-migrants per zone is estimated following the approach of the existing
urban MARS model. Second, a migration destination choice model distributes the out-
migrants (which it takes as an exogenous input from the out-migration model) over the
possible destinations based on characteristics of the destinations and the distance
between two zones.
The model takes the form of the well-know gravity/spatial interaction model. In general
terms, the number of migrants between origin i and destination j, Mi, is modelled as
expla 404X415 +02X2j+..+OnXnj+YnY ij Jai?
gyri
Yewla 404K) +02K 9) ++ OnXaj tray Ja
j
M
Formula 3 General form of the formula for calculating migration flows from zone i to J
where Oi represents the number of out-migrants of origin i (given exogenously to the
distribution model); X1j...Xnj a set of n attributes relating to destination j with the
associated parameters co... On; Yj an origin-destination pair specific (dummy) variable
with the associated parameter y; di the distance between origin i and destination j.
Workplace location sub model
The workplaces migration sub module has a structure, very similar to the residential
migration model. In the current version it consists of two parts: one for the production
sector and one for the service sector.
At the moment the relative attractiveness of a zone for potential workplace migration
considers:
e The zone’s potential for activity participation (accessibility),
e the abundance of building land,
e the cost for building in a zone and
e the average household income.
Access attractiveness, formulated as potential to reach workplaces and shopping
opportunities, presents the zones potential for activity participation. The possibility to
build in a zone is just restricted by the limits of land availability in a zone. The cost of
building in a zone is approximated by the land price. The average household income is a
signal for firms whether there is consumption potential and is a proxy for labour cost.
For the out-moving model an average time workplaces move has to be defined,
identified in empirical studies. The total number of workplaces in the study area
multiplied by the reciprocal of the average time workplaces move gives the total
number of out-movers in the study area.
In a next step the attractiveness to move out a certain zone is calculated with the above
mentioned influence factors, except for the land availability which of course is just
relevant for the in-moving sub model. This is modelled again as exponential function of
the form, separately for each sector:
Attr tt = qi" ACC tea" Land _ price att gems +a3*HHL,)
ti
Formula 4 Attractiveness to move out for workplaces
Oy. Ay Parameters
ACC; Access attractiveness in model zone i
Land_price_attr.i sector Land price attractiveness per zone i and_ sector
(production/service)
HHI; Household income in model zone i
The workplaces, which want to move in, are defined similar to the out-moving
workplaces, but an external growth rate is added, which can be negative or positive
depending on the sector. Then MARS calculates the amount of space available for
business use and allocates the total potential re-allocating and newly developed
workplaces to the different locations using a LOGIT model (see Formula 3)
Housing development model
In the MARS model developers decide whether how much and where to build new
housing units. Their decision is based on four factors:
e The rent they can achieve after the housing units are ready for occupation. It is
assumed that this is the rent paid in the year of the development decision.
e The land price in the decision year.
e The availability of land in the decision year.
e The demand from potential in-movers in the zones.
The potential for new domiciles is distributed to the zones according to the
attractiveness to build in a zone, which is dependent on the above mentioned factors.
These will be ready to occupy after an extemal defined time lag. MARS controls
whether there is enough land for the planned developments. If not, the number of
developments in the certain zones is constrained. There is currently no redistribution
process to other locations in the development sub model. Changes in the available land
influence land price and rent.
3 THE APPLICATION OF THE MODEL TO AUSTRIA
3.1 Study area and model zones
The study area comprises the whole territory of Austria, totally 121 model zones which
are based on the district subdivision (‘politische Bezirke’) of Austria plus the 23
municipal districts of Vienna. A first attractive feature of the district structure is that it
includes the so-called ‘independent cities’ (Statuarstadte) which are administratively
separated from their hinterland districts. Hence, it is possible to represent core-periphery
interactions (such as commuting flows and suburbanization) for these districts in the
model. Second, for many statistics, the district level is the most detailed level for which
data are available. Third, the number of districts (121) is a good comprise from a
technical point of view in that it keeps calculation time of the system within a
reasonable limit.
There are two important features of the case study worth mentioning. First, the model
zones are very heterogeneous amongst each other. It comprises highly urbanized,
service-sector oriented zones with highly positive commuting balance; sparsely
populated zones with significant agricultural production and high out-commuting rates;
mountainous regions influenced by tourism where settlement areas are concentrated or
constrained by alpine valleys to name just a few examples. All in all, diversity is much
greater than in usual urban agglomeration models.
Second, as the case study covers the entire Austrian territory, it is apparent the model
area is polycentric and, additionally, comprises several levels of central places.
We set up the model with data from 2001 for all available data. Due to some lack in
data expert guesses were necessary. This concerns first and foremost guesses in data for
the transport model, like parking place search time, parking fees, etc. For the average
rent data covers just the year 1991. Also for the calibration of the different model parts
some exceptions due to data availability were necessary, which will be described
separately in each section.
4 FIRST RESULTS FROM MODEL CALIBRATION
Model calibration and testing are currently in progress. This section therefore presents
some insights gained from first model runs.
4.1 Approach to model calibration and testing
We define model calibration and testing based on Ortuzar and Willumsen (1994).
Model calibration consists in finding parameter values that optimize the goodness of fit
between model outputs and observed data. Model validation is a related but not identical
concept. It consists in comparing in model predictions with observed data based on a
dataset not used in calibration. However, in line with Sterman (2000), we prefer the
term model testing instead of validation, as model “validation” in the strict sense of the
word is impossible as a matter of principle.
The transport sub model of MARS simulates transport flows within a time period.
Model calibration is thus carried out on a cross-sectional basis to improve model fit in a
base year. In contrast, the land-use sub models simulate changes in a time interval
which requires calibration of changes in observed land-use.
Due to data availability reasons we had to restrict the data analysis to the period from
1991 to 2001 for employment data and from 2002 to 2006 for residential migration data
in a first step. This implicates that migration trends in Austria which where happening
between 2002 and 2006 are assumed be valid also from the beginning of 2001 and that
the development of employment in Austria between the years 1991 and 2001 also
continued from the beginning of the year 2001 (the base year of the MARS Austria
model).
We used the optimization functionality implemented in Vensim (Ventana Systems Inc.
2007) for our calibration purposes. This automated calibration procedure was applied
both to the transport and the land-use sub models. The optimizer consists of an
algorithm (Powell) which numerically maximizes or minimizes an arbitrary objective
function; in Vensim terminology the objective function is called “payoff”. In the
calibration mode, the payoff is automatically specified by the software as the sum of the
squared deviations between the observed values and the model output for one or more
user-specified variables; a weight can be attached to each of the variables. Parameter
values are chosen in an iterative process to minimize payoff.
4.2 Transport model
Table 1 shows the modal split calculation data to which MARS Austria is calibrated.
The mode split for trips home - work is calculated using data from 1995 for commuting
trips (Herry 2007). For the trips home - other, the average mode split weighed by the
share of additional trip purposes (education, leisure, shopping, official business trips) in
other trips is taken.
Mode Home-W ork Home-Other Total
car 63.6% 47.6% 55.6%
pt bus 11.1% 10.2% 10.7%
pt rail 7.1% 6.9% 6.9%
slow 18.2% 35.4% 26.8%
Table 1 Mode split calibration data 1995. Source: (Henry 2007), own calculations
For the transport model the k factor (peak/opeak) for the trips HWH (home - work -
home; commuting trips) and HOH (home - other - home; all other purposes) is
estimated.
Ay Kacraaaasl f(ttyc#)
Tyr =/P* * ™ om
dA K soak opeak / f ti Ci
m) HWH /HOH
Formula 5 Simultaneous trip distribution and mode choice
ry Number of trips my mode m from source i to destination j
Pi Production of trips at source i
Aj Attraction of zonej as destination
ty Travel time my mode m from i to j (min)
fy Travel cost for a trip my mode m from i toj (€)
£(t™,c%y) Friction factor for a trip my mode m from i to j (min)
HWH Tour home - work - home
HOH Tour home - other activities - home
The friction factors are indicators to measure the subjectively perceived effort in terms
of time and money which is necessary to travel from origin i to destination j.
4.2.1 Calibration of the transport model parameters
Table 2 shows the mode and purpose specific parameters derived from the calibration of
MARS Austria of the base year result to the mode split data shown in Table 1. The
calibration was accomplished within the MARS Austria model using the Vesim built in
calibration function.
Trip purpose car pt bus prail slow
Home - Work | 1.96 0.28 0.24 0.21
Home - Other | 1.03 0.56 1.03 0.70
Table 2 Mode and trip purpose specific parameters k peak/opeak of the calibration
4.2.2 Model fit - transport model
The conformity between observed and calculated data was assessed by comparing home
- work trips by OD-pair and intrazonal home - work trips as calculated by the calibrated
version of MARS Austria and data from 2001. Unfortunately no similar data were
available for trips home - other.
The resulting regression coefficients are in general reasonably high. The R’ for the total
trips home - work is 0.91, the R? for the intrazonal trips is even higher with 0.97.
At the bottom of the left hand figure in Table 1 it can be seen that MARS Austria is
underestimating some commuting connections from the core cities (Linz, Salzburg,
Graz and Innsbruck) to surrounding suburban districts. The intrazonal trip distribution is
working very well, due to the fact that we implemented intrazonal distance classes (see
section 2.4.1).
‘Trips by OD pair Tours intrazonal
90000 100000
90000
0000
70000
60000
50000
0000
MARS predicted trips
30000
20000
10000
o
0 20000 ‘40000 ‘0000 ‘0000 if 20000 40000 60000 ‘0000
DATA 2001 DATA 2001
Table 3 Comparison data of commuting trips to MARS results - all commuting trips and intrazonal trips. Source:
(Statistik Austria 2002a)
4.3 Residential migration
In order to asses the model output we briefly summarize domestic migration trends in
Austria in the period from 1991-2001 (Statistik Austria 1995, 2002b) and 2002 to 2006,
which is the period for which detailed data on migration are available in Austria
(Statistik Austria 2005a, 2005b, 2006, 2007a, 2007b).
The most outstanding observation on domestic migration is that migration takes place at
fairly short distances. The median distance (air-line) between an old and a new domicile
is 12.6 kilometres; for 80% of the migrants the distance is shorter that 24.3 km, between
the years 2002 and 2006. As a result, migration linkages are less quantitatively
significant on higher levels of spatial aggregation; as an illustration may be noted that
migration between the districts of Vienna exceeds the flows between all other Austrian
provinces. This observation was a strong argument for explicitly considering the
distance between the zones for migration choices (see section 0).
Between the years 1991 to 2001 the most striking trend was a suburbanization progress.
Population shifts from the core to cities to surrounding suburban districts affected all
major agglomerations. This includes all cities exceeding a population of 100.000:
Vienna, Graz, Linz, Innsbruck and Salzburg.
In the following years urban agglomerations are the winners in domestic migration.
With the exception of Salzburg and Bregenz, all provincial capitals, several medium-
sized cities and the capital of Vienna experience population gains in the period. Within
the major agglomerations, the process of suburbanization is continuing, population
shifts from the core cities to surrounding suburban districts affect all major
agglomerations.
10
4.3.1 Calibration of the residents migration model parameters
As mentioned above (see section 4.1) we used the built in optimizer of Vensim
(Ventana Systems Inc. 2007) to calibrate the parameters of the moving-out and the
migration-flows model. In estimating parameter values, the optimizer draws on the
same criterion as (ordinary) least square estimation (OLS) in regression analysis. The
difference to OLS is that the model need not necessarily be linear in parameters when
estimated using the optimizer. The gravity model is not linear in parameters.
A point of criticism of this method might be that there are no analytically known
measures of model significance. However, through simulation some experimental
indications on model and parameter significance can be obtained. Another method we
will implement in the future for the parameter estimation process is
“bootstrapping” (Moore and McCabe 2006, Dogan 2007). Bootstrapping enables to get
the distribution of the parameter values and in further consequence allows conclusions
about statistical significance of the estimated parameters.
We build two stand-alone models for the calibration (see Figures below), to keep the
time consumed for parameter calibration little. For the out-migration model the
parameters for the out-migration rate per zone are estimated (see Formula 6).
OM —rate, = om—rate* e( 2's saACey saaPOP saul; )
Formula 6 Calculation of the out-migration rate in MARS Austria
om - rate Average out-migration rate throughout A ustria
Ay ry Parameters
HC; Housing cost in model zone j
ACC; Access attractiveness in model zone j
POP; Population in model zone j
LS; Living space in model zone j
For the migration flow model, parameters for the migration flows from model zone i to
zone J are estimated (see Formula 3).
11
Sum out-migrants
DATA
out-migrants Moving out rate per
—_—!_
DATA zone DATA
Residents J t=0 out-migrants Sum out-migrants
pred. ped.
Moving outrate__. out-migration rate pred,» average moving out
DATA rate pedicted
Residents J T-1 HOwtieatve sees Ait Living Space HU J t
relat a) ot b ¢) out © () out realtive sno
Figure 2 Structure of the out-migration calibration model
Oi
Sum Mij obs<¢——Mij DATA
Mij predicted ————————> Sum Mj
mousing. cost acces attr. distance car i. dum functional y ¢) in
WN LE e ¢) in
Sum
enumerator
Figure 3 Structure of the migration flow model
12
Variable
Description
Out-migration model
Residents J T-1 relative
Population of destination J per 100.000
inhabitants
HC J relative Ration of housing cost (rents) in a zone to
average housing costs (rents)
Access attr. Population accessibility potential with a
quadratic decay function based on
generalized cost between origin and
destination
Living space HU J relative
Ratio of living space per housing unit (m*) in
a zone to average living space per housing
unit (m’)
Migration-flow model
Residents J T-1 relative
Population of destination J per 100.000
inhabitants
Housing cost
Housing rents at destination J
Access attr.
Population accessibility potential with a
quadratic decay function based on
generalized cost between origin and
destination
Recreational green land
Share of green land as a percentage of total
zone area
Dummy functional (FUA)
Dummy for the “functional urban areas”
Distance car iJ
Car distance between districts i and J
Table 4 Overview of the migration variables
Table 4 above describes all variables considered in the migration models in MARS
Austria. The number of residents in the previous time step, the housing cost and the
access attractiveness are considered in both the out-migration and the migration-flow
model. Moreover the living space is a variable for the out-migration model.
As mentioned in section 2.4.2 in the migration-flow model the car distance, for
considering cost of migration is explicitly taken into account.
Furthermore it tured out that there where some core-hinterland relations, which result
in migration patterns that cannot be explained by the variables already considered.
The dummy variable “functional urban areas” (FUA) tries to capture these core-
hinterland relations. Functional urban areas based on patterns of major commuting
catchment areas are defined. Inter-district relations within the same FUA are singled out
in the estimation using this dummy variable.
The variable recreational green land, which in the urban MARS case studies has always
been a very strong explanatory variable, turns out to have no influence in the nation
wide set up any more. The aggregate level of observation leads to the situation that the
amount of green area in a model zone is sufficiently large, therefore its signal effect as a
variable for living quality is gone.
13
4.3.2 Model fit - migration model
The model fit both for the out-migration as well as the in-migration model is in general
satisfactory. The out-migration model including the explanatory variables listed in
Table 6 yields a R’ value of 0.93. The migration-flow model yields to an R? value of
0.96. Further variables were also tested but generally did not notably improve the model
fit; for reasons of compatibility with the existing urban MARS model, they are not
included in the migration model from the time being.
Out-migration Migration from ito}
30000 2000
° 000 19000-15000 20000-25000 «20000
a 000 10000» 15000 © 20000-25000
out-migratnts-data
Mij -data
Table 5 Goodness-of-fit of the estimated models: scatter plots of predicted vs. observed out-migrants and migration
flows
4.3.3 Parameter estimations for the residents location model
All parameter estimates are presented in Table 6. As expected the population parameter
(Residents J T-1 relative) is negative for the out-migration and positive for the in-
migration model.
The parameter on housing rents (HC J relative and Housing Cost) is also positive for the
out-migration and negative for the migration-flow model.
The parameter related to accessibility (Access attractiveness) is positive for the out-
migration and negative for the in-migration model. The negative sign for the migration
flow model is a frequent finding in empirical models of migration and has been
explained in terms of stronger competition between more accessible destinations (which
are closer to major population concentrations) and in terms of lack of information on
specific destinations in larger population “cluster” on the part of migrants (ODPM
2002).
The living space parameter (Living space HU J relative) has a strong negative influence
on out-migration which is in line with common sense.
The recreational green land, which was a very important explanatory variable,
influencing migration in the urban MARS models, has no influence on the national scale
(see section 4.3.1).
Distance between two districts exerts a strong negative influence on migration between
these two zones. And as in line with the argumentation of the dummy for the functional
urban area (cf. 4.3.1), migration flows are positively affected by this parameter.
14
Parameter Value
Out-migration model
a (r) out (Residents J T-1 relative) -0.276639
b (x) out (HC J relative) 0.200801
c (r) out (Access attractiveness) 0.180238
e (1) out (Living space HU J relative) -1.34024
c-out 1.08443
Migration-flow model
a (r) in (Residents J T-1 relative) 1.18259
b (r) in (Housing cost) -0.820271
c (r) in (Access attractiveness) -0.433847
d (x) in (Recreational green land) 0
e (r) in (Distance car iJ) -1.50692
g (x) in (Dummy functional) 1.35996
c -1.93192
Table 6 Parameter estimates for the residential location sub model
4.4 Workplace migration
As mentioned in section 2.4.2, the workplace migration sub model has a structure very
similar to the residential migration model. Therefore we do not explicitly describe the
calibration of the workplace sub model in this paper. Similar to the residential migration
calibration of parameters a stand-alone model was built and again the built in optimizer
of Vensim (Ventana Systems Inc. 2007) was used. We refer the reader to contact the
author s for further information.
5 RESULTS FIRS MODEL FORECAST
With the calibrated model a first long term model run (30 years) was completed. In the
following the results of the residents’ migration and the workplaces migration model are
presented.
For the transport model part, there are no substantial changes in the mode split and trip
distribution. The growth in the mode share of cars, which could be observed in the last
years, is continuing. The implemented fuel price rise is just approximately 1% per year.
For the population model a growth in population of 0.3% per year is assumed.
Assumptions for the workplace model are a slight decrease for the production sector
workplaces and a growth for the service sector workplaces.
5.1 Predicted population development
Figure 4 below shows the population development by district from the year 1991 to
2001 from data. The above (section 4.3) mentioned suburbanization processes can be
seen as well as significant population losses in the province of Styria, due to structural
economic changes. These districts were dominated by heavy industry.
15
Legend
cz
mmm -15,000 — -10,000
Im -10,000 — -5,000
-5,000 — -500
“500 — 500 ie meet
500 — 5,000] pe Ce oe -
lm 5,000 — 10,000 e os bw
fm 10,000 — 15,000 6 —.
satug ¢
=) AT
jo
mo De sia
CH e Ls a
rT bs Koger
ae
si HR
Figure 4 Change in population by district 1991-2001 (data). Source: (Statistik Austria 2002a)
Figure 5 shows the model output after a forecasting time of 15 years. In continuation
with the observed development the model predicts an ongoing suburbanization process
for some major cities, like Vienna, Graz, Linz and Salzburg; their hinterland districts
gain in population. Apart from that there is a significant loss of population in a district
at the western part of Austria. The styrian districts which had substantial population
losses between 1991 and 2001 can gain population in the model prediction.
EW +15years
<-15,000
-15,000 - -10,000
10,000 - -5,000
-5,000 - -1,000
1,000 - 1,000
1,000 - 5,000
[Ea 5.000 - 10,000
10,000 - 15,000
= >15,000
Figure 5 Predicted population changes after a forecasting time of 15 years (year 2016) to base year 2001. Source:
authors’ calculations
16
Figure 6 presents the predicted population development after a forecasting time of 30
years. All major agglomerations are now winner in absolute population values
compared to the years 2001. Bregenz, the districts on the western part of Austria, which
was loosing population in the first 15 years, is a winner of population compared to the
year 2001 and reaches a value (6000 people) higher that the population value of the year
2001. Also Graz which lost about 10.000 people in the first 15 years of simulation now
reaches a population of about 250.000 people which is approximately 15.000 people
more than in the year 2001.
EW 430years
<-15,000
[ll -15,000 - -10,000
10,000 - -5,000
= -5,000 - -1,000
-1,000 - 1,000
1,000 - 5,000
5,000 - 10,000
10,000 - 15,000
lg > 15,000
Figure 6 Predicted population changes after a forecasting time of 30 years (year 2031) to base year 2001. Source:
authors’ calculations
17
5.2 Predicted development of workplaces
In the following the predicted workplace development of the two sectors (production
and service) are presented. The first two graphs depict the development in the provincial
capitals, except for Vienna. The second two graphs depict 4 Viennese districts, two
small and dense districts and two big and less concentrated ones.
Figure 7 shows the development of the workplaces in the production sector after a
forecast of 30 years. All capital cities loose workplaces in the observed sector, except
for Eisenstadt which is a capital with just little population (about 10.000 people), where
the number of workplaces in the production sector slightly increases. This development
follows an overall observed Austrian trend of the last years. Growth rates over the
whole case study area in the production sector are assumed to be slightly negative.
Workplaces J
60,000
45,000
30,000
15,000 pr et tt
0
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
Time (Y ear)
Workplaces J[a101,production] : first full_run_30
Workplaces J[a201 production] : first full_ run 30
Workplaces J[a302,production] : first full” run 30
Workplaces J[a401,production] : first full_ run 30
Workplaces J[a501,production] : first_full_run_30
Workplaces J[a601,production] : first full_run_30
Workplaces J[a701, production] : first full” run 30
Workplaces J[a801 production] : first full_ run 30
Figure 7 Workplace development of the production sector of the provincial capitals
18
The development of the workplaces in the service sector looks totally different (Figure
8). All observed capital cities experience gains in the number of workplaces. Also in
this case Eisenstadt makes an exception, it wins relatively more workplaces than the
other capitals. Also this development is in line with observed Austrian trends, the
service sector gained workplaces in the last years.
Workplaces J
200,000
a ase ae
om Se
50,000 SS ee
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
Time (Y ear)
Workplaces J[a101 service] : first full_run_30
Workplaces J[a201 service] : first full_run_30
Workplaces J[a302,service] : first full_run_30
Workplaces J[a401 service] : first full_run_30
Workplaces J[a501 service] : first full_run_30
Workplaces J[a601 service] : first full_run_30
Workplaces J[a701 service] : first full_run_30
Workplaces J[a801 service] : first full_run_30
Figure 8 Workplace development of the service sector of the provincial capitals
19
The prediction of the development of the service sector workplaces for four Viennese
districts is shown in Figure 9. All depicted districts experience a growth in workplaces
except for the first district.
Workplaces J
200,000
150,000
100,000
50,000
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
Time (Y ear)
Workplaces J[a901,service] : first_full_ run_3¢
Workplaces J[a907,service] : first_full_run_3@
Workplaces J[a916,service] : first_full_run_3@
Workplaces J[a923,service] : first_full_run_3¢
Figure 9 Workplace development of the service sector of four Viennese municipal districts
6 CONCLUSIONS AND OUTLOOK
This paper presents the experiences made in the setup of a national land-use transport
interaction model for Austria. The objective of the exercise is to test how LUTI
modelling can be applied in rural contexts and to asses the generality of an urban LUTI
model at a higher spatial level.
Overall we showed that a nation wide setup of a land-use transport interaction model is
possible. The LUTI model appears to be relatively generic in structure, just few model
structure adaptations were necessary and it seems to be possible to model a very
heterogeneous case study area. Without external system intervention (for example a
transport policy) the model produces reasonable behaviour, where trends to date seem to
continue into the future.
A next step will be to gain more insights from simulation. Policy scenarios will be
designed to model whether and how the system takes other development paths.
There are of course some parts of the model structure that need revision and
improvements. Especially the workplace migration model seems to be the most
unsuitable for modelling a wide geographical scope like this. One major problem is the
lack of migration-flow data in this field. Furthermore we want to include a more
detailed sectoral disaggregation than the current two sectors or a differentiation
according to business size (number of workplaces).
20
Research will also be devoted to the way quality of life can be considered in the
residential migration model of MARS Austria. The urban variable “recreational green
land area” is not influencing migration flows any more.
The calibration procedure will be repeated for the residents- and the workplace model
part within the MARS structure, not just with the stand-alone models (see section 4.3.1).
Furthermore the bootstrapping methodology will be tested and results will be reported
Another next step will be extensive model testing of the MARS Austria model. For this
task, lacks in data base have to be filled.
21
7 REFERENCES
BODE, E. & ZWING, S. (1998) Interregionale Arbeitskraftewanderungen: Theoretische Erklarungsansatze und
empirischer Befund. Kieler Arbeitspapiere. Kiel, Universitat Kiel.
DOGAN, G. (2007) Bootstrapping for confidence interval estimation and hypothesis testing for parameters of system
dynamics models. System Dynamics Review, 23, 415-436.
EMBERGER, G., PFAFFENBICHLER, P. & HALLER, R. (2007) National scale land-use and transport modelling:
the mars Austria model. European Transport Conference (ETC) 2007. Leiden, The Netherlands.
FLOWERDEW, R. & AMRHEIN, C. (1989) Poisson regression models of Canadian census division migration
flows. Papers in Regional Science, 67, 89-102.
GREENWOOD, M. J. (1985) Human migration: Theory, models, and empirical studies. J ournal of Regional Science,
25, 521-544.
HAKEN, H. (1983) Advanced Synergetics. Instability Hierarchies of Self-Organizing Systems and Devices, Berlin,
Springer-Verlag.
HERRY, C. G. (2007) Verkehr in Zahlen. IN 5, A. V. I. (Ed.), Bundesministerium fiir Verkehr, Innovation und
Technologie.
MOORE, D. S. & MCCABE, G. P. (2006) Introduction to the Practice of Statistics, New Y ork, W.H. Freeman.
MUTH, R. F. (1971) Migration: Chicken or Egg? Southern Economic J ournal, 37, 295-206.
ODPM (2002) Development of a migration model. London, Office of the Deputy Prime Minister.
ORTUZAR, J. D. D. & WILLUMSEN, L. G. (1994) Modelling transport, Chichester, Wiley.
PFAFFENBICHLER, P. (2003a) The strategic, dynamic and integrated urban land use and transport model MARS
(Metropolitan Activity Relocation Simulator). Institute for Traffic Planning and Traffic Engineering.
Vienna, University of Technology.
PFAFFENBICHLER, P. (2008) MARS - Metropolitan Activity Relocation Simulator
A System Dynamics based Land Use and Transport Interaction Model, Saarbriicken, VDM Verlad Dr.
Miller.
PFAFFENBICHLER, P. C. (2003b) The strategic, dynamic and integrated urban land use and transport model MARS
(Metropolitan Activity Relocation Simulator). Development, testing and application. Institute for Transport
Planning and Traffic Engineering. Vienna, Vienna University of Technology.
ROY, J. R. (2004) Spatial Interaction Modelling, Berlin, Springer.
STATISTIK AUSTRIA (1995) Volkszahlung 1991. Binnenwanderung 1986-1996. Wien, Statistik Austria.
STATISTIK AUSTRIA (2002a) Volkszahlung 2001. Hauptergebnisse I - Osterreich. Wien, Statistik Austria.
STATISTIK AUSTRIA (2002b) Wanderungsstatistik 2001. Wien, Statistik Austria.
STATISTIK AUSTRIA (2005a) Wanderungsstatistik 2002. Wien, Statistik Austria.
STATISTIK AUSTRIA (2005b) Wanderungsstatistik 2003. Wien, Statistik Austria.
STATISTIK AUSTRIA (2006) Wanderungsstatistik 2004. Wien, Statistik Austria.
STATISTIK AUSTRIA (2007a) Wanderungsstatistik 2005. Wien, Statistik Austria.
STATISTIK AUSTRIA (2007b) Wanderungsstatistik 2006. Wien, Statistik Austria.
STERMAN, J. D. (2000) Business Dynamics - Systems Thinking and Modeling for a Complex World, McGraw-Hill
Higher Education.
VENTANA SY STEMS INC. (2007) Vensim Reference Manual, Harvard, Ventana Systems Inc.
WEGENER, M. (2004) Overview of Land-Use Transport Models. IN HENSCHER, D. A. & BUTTON, K. J. (Eds.)
Handbook of Transport Geography and Spatial Systems. Oxford, Elsevier.
22
POTENTIAL
OPEAK
POTENTIAL PEAK
.
Tie > ge owt ne work,
toaltie peak
ed
wave | Sayoore: seen
entageon | _Sota tours peak i-t———_ $
weit ili a ie en , cs,
( e om Rae
aire whens OER
PEAK
ax Ue epeak per
totaling opeak te oREaK Re
veiw coon cama oa en ee
\ — mode | ~t tours ir peak
eee
shri pr_p Wongaces Gr art
sas ~ attr. matrix peak:
etatee Ped *worolaces i te
— stactverest g—scomatn
NE wei ee ==
me betes es occupancy asimee ase sence <WorkpliCes | wins
- eri edistance cz
sre Gort / atwer\ aerd
‘WP ij service pc per ds i
ousted pe ooo
\ {mode ir opeak
woislees i
ee: oie Stowe navn) ‘eice asides} >
ee matiixae—* oF
sum ate manic — 970k
Toran at ma Sean
ms car_ ms ptbus ms ptrail ms slow bah ead
if We oPeak
— ‘Aotaltime: ‘tip time ir mode <tip time
ook Sambi eae
Firgure 1 MARS Austria - Transport model view
ceiiert) -accesst0Mn> hg space
SumResidents) Tr ewe ong aut ate
eat
Acces Ket
Residents J Tl. sf Residents esi
‘ca - veut potetaly
/ Access At
se Pobsi landed | actdens por, “secu
resent ty Wester
ba asides] 1 te faster)
Sum atraciveness a 7
Rema aa ‘Share owener "Tova Rullneceie ‘Control Supply)
demand resets | i
acipled using} —
Residents
__r meat
~etusmadeg Rests pr
et way
Resides 1) este
/ = Aesiers wa
| cl Ss fi —-
ry Readers oa sae mie PO
Lf ! a
seacl ans fowneatiog/ “testes sunk 01 is pea od
rmoungintay/ mows) 7 porrtal @) Te
sip f) | ot ey
No dt woe
sane, N Eg Ae, tassel —nterteniny saan
ie
‘moung nia) resents
om
oe ee
nai swoynD
dene pen
Sek
Figure 2 MARS Austria - Residents view
Va
Land availabe
air
k supply
Attractiveness
Gaia niente moving out
‘moving in "workplace aie ieee Potential moving
Household
loutworkplaces
income j relatve
Workplaces:
out]
<< w capacity i
\ Used WP are: i Free WP area
Spatial WP demand Specie
unaccupied J T- Spatial WP i
pied TL capacityper Zone ———*"capacityT-0
1
St and avaiable
‘demand
Workplaces: _eWorkplaces
sector ‘waren
=<
Aug tne workplaces move
smaccopledy -————peeirery Workplaces out demand total
demand J jin workplaces ;
u ss od —
demand wosatsted Workplaces Workplaces
carers demand workplace mi is
workplaces secorT= scenario
Pal \ i
|Stock unsatisfied| }
ra Workplaces
Ursatsted demand] demand [pono Shomces ospeces
workplaces workplaces satisfied —
woplces) 1-0 Shave daecarT-0
workplaces }>
Figure 3 MARS Austria - Workplaces view
