1

Microsimulationof Land Use andTransport inCities

Michael Wegener

Advanced Modelling in Integrated Land-Use and Transport Systems(AMOLT) M.Sc. Transportation Systems TU München, 14 July 2009 2

Model levels

Dortmund

Regions

Dortmund

Zones

Dortmund City

Raster cells

Multi-level

Multi-scale

3

Model levels

Dortmund

Regions

Dortmund

Zones

Dortmund City

Raster cells4

Microsimulation

5

Microsimulation

Theory

Microsimulation is the reproduction of a macro process bymany micro events.

Events:

The basic building block of microsimulation is the event.

No deterministic assertions (that are valid with certainty)can be made about events, only probabilistic assertions(that are valid with probability) are possible.

6

Microsimulation

There are two types of events:

- Transitions are changes of an individual or object fromone state to another subject to transition probabilities(e.g. ageing of persons or buildings) .

- Choices are selections between alternatives by anindividual as a function of their perceived utility (e.g.modal choice, residential location).

If the causal chain behind choices is of no interest, alsochoices are modelled as transitions (e.g. birth, marriage,divorce).

7

Microsimulation

The method of microsimulation consists of generating asequence of events as a function of their probability ofoccurrence.

To generate an event, a lottery is drawn the result ofwhich is one of the possible outcomes of the event andsubject to their probability. This is why microsimulation isalso called Monte-Carlo simulation.

Another term for microsimulation models is agent-basedmodels (ABM). Cellular automata (CA) models are raster-based models of transitions of raster cells based on neigh-bourhood properties.

8

Microsimulation

The lottery consists of the mapping of a random numberbetween 0 and 1 to the vector of cumulated probabilities ofpossible outcomes of the event.

Example:An event has three possible outcomes with probabilities 0.3, 0.5 and0.2. If the random number 0.74 is drawn, Outcome 2 is the event.

Random number 0.74

Outcome1

Outcome2

Outcome3

0 0.3 0.8 1.0

9

Random number generators

Random number generator Pseudo random number generator 10

Micro database

Microsimulation models require micro data.

In reality micro data are rarely available or where availablecannot be used because of privacy concerns.

Synthetic micro data are micro data generated fromavailable spatially aggregate data with which they areconsistent in all known attribute distributions.

Microsimulation models can be operated with syntheticmicro data instead of real micro data.

11

Micro database

BuildingsResidential buildings (micro location)- Dwellings (type, size, quality)

Nonresidential buildings (micro location)- floorspace (industrial, retail, offices)

Households/BusinessesHouseholds (micro location)- Households (size, income,cars)- Persons (age, sex, education, job)

Businesses (micro location)- Firms (industry, size, vehicles)- Employment (skill)

12

Spatial disaggregation

For the synthetic micro database zone data are allocated toraster cells. Two steps are performed:

(1) Conversion of polygons to raster cellsThe polygons of a land-use map are converted to rastercells and each raster cell is assigned a land-use category.

Land-use categories

Residential high-density

Residential low densityIndustrial

Open Space

13

Spatial disaggregation

(2) Allocation of zone data to raster cellsThe data elements are allocated to raster cells by Monte-Carlo simulation according to their density.

Probabilities Number ranges n Weight of all zones

10050 data elements10

Weights

1,487 dataelements

10

5

1

1

14

Spatial disaggregation

100 x 100 m raster cells

1 km2

Spatial disaggregation

100 x 100 m raster cells

1 km2

14

15 16

Synthetic population for Dortmund

Population density in raster cells of 100 x 100 m

0 - 2020 - 4040 - 6060 - 8080 - 100

100 - 120120 - 140140 - 160160 - 180180 - 200200 -

Pop/ha

0

km

2

17

Dwellings in DortmundWorkplaces in Dortmund

17

Suburb-to-suburb commuter flows in Dortmund

17

Commuter flows in Dortmund

17 18

The ILUMASS Model

19

The ILUMASS Project (2001-2006)

The project ILUMASS (Integrated Land-Use Modelling andTransport Systems Simulation) embedded a microscopicdynamic simulation model of urban traffic flows into acomprehensive model system incorporating both changesof land use and the resulting changes in transport demandas well as their environmental impacts.

For testing the land use submodels, the transport andenvironmental submodels were replaced by the aggregatetransport model of the IRPUD model and simpler envir-onmental impact models (= reduced ILUMASS model).

20

Internal zones

External zones

Dortmund

Hamm

Hagen

Bochum

Study area

21

Study area

Hamm

0 10 km5

Dortmund

22

Syntheticpopulation- housing- persons- cars

Synthetic firms- floorspace- firms- jobs- vehicles

Base year data

Transportnetworks- roads- public transport

Firms- floorspace- firms- jobs

Person traveldemand- activities- week plans- trips

Goodstransportdemand- activities- trips

Environmental impacts

Emissions- air pollution,- traffic noiseat sources

Population- housing- households- persons- cars

Dynamic trafficassignment- network flows

Impacts- air quality,- traffic noiseat work

Impacts- air quality- traffic noise

at housing

Accessibility- of jobs- of shops- of population- of facilities

Land use

Transport

ILU

MA

SSm

odel

ILUMASS model

23

Dortmundmodel

Forecast ofSASI model

Microsimulation

ReducedILUMASSmodel

Forecast ofSASI model

Microsimulation

23 24

Modules

Travel

Environmentalimpacts

Firms

Households

Dwellings

Moves

Firms

Households

Dwellings

Moves

Firms

Households

Dwellings

Moves

Travel

Environmentalimpacts

Firms

Households

Dwellings

Moves

t=0 t=1 t=2 t=3

Start

meso

micro

25

Firms and householdsStart

End

Anotherhousehold?

Yes

No

Select a household

Demographic events- Ageing- Death- Birth- Marriage/cohabitation- Divorce/separation- Persons leave household- Persons move together

Economic events- Education- Place of work- Employment status- Income- Mobility budget of household

Update lists- Persons- Households- Dwellings- Firms

Start

End

Anotherfirm?

Yes

No

Select a firm

Select alternative location

Accept?No No

1-10 11

Yes

Update lists- Firms- Nonresidential floorspace- Employed persons

Check satisfaction with presentlocation

Satisfied?

No

Yes

26

Dwellings

Anotherproject?

EndNo

Update lists- Persons- Households- Dwellings- Construction

Yes

Select zone andmicrolocation

Select type and num-ber of dwellings

Upgrading

Select zone andmicrolocation

Select type and num-ber of dwellings

Demolition and result-ing moves

Select zone andmicrolocation

Select type and num-ber of dwellings

Construction

Start

Expected demand- Demolition- Upgrading- Construction

Select project- Demolition- Upgrading- Construction

27

MovesStart

End

Anotherdwelling?

Yes

No

Select dwelling- Area- Type

Select household- Immigration- New household- Forced move- Move

Accept?No No

Update lists- Households:

- with dwelling- without dwelling

- Dwellings- occupied- vacant

1-5

Housing supply

Yes

Start

End

Anotherhousehold?

Yes

No

Select household- Outmigration- Immigration- New household- Move

Select dwelling- Area- Type

Accept?No No1-5 6

Yes

Housing demand

Update lists- Households:

- with dwelling- without dwelling

- Dwellings- occupied- vacant

6

28

Environmental Impacts

29

ReducedILUMASSmodel

Forecast ofSASI model

Microsimulation

ReducedILUMASSmodelEnvironmentalimpacts

Forecast ofSASI model

Microsimulation

Environmental impacts

29 30Air quality

31Traffic noise

32Change in traffic noise 2001-2021

33Change in traffic noise relative to reference scenario in 2021

34Quality of open space

35Access to open space

36

Typical Model Run

37

Model dimensions

1.2 million households2.6 million persons1.2 million dwellings

80,000 firms92,000 industrial sites

8,400 public transport links848 public transport lines

13,000 road links

246/54 internal/external zones209,000 raster cells

30 simulation periods (years)

90 minutes computing time

Con

clus

ions

38

Model parametersMicro dataMicro data

39

Travel flowsPublic transportflowsTravel flowsTravel flowsTravel flows

40

Link loadsLink loadsLink loads

41

Link loadsLink loadsLink loadsPublic transportspeed

42

Link loadsLink loadsLink loadsPublic transportspeed

43

Air qualityNO2

44

Traffic noisedB(A)

45

FirmsFirmsFirmsFirmsFirmsFirms

46

HouseholdsHouseholdsHouseholdsHouseholdsHouseholdsHouseholdsHouseholdsHouseholdsHouseholdsHouseholdsHouseholdsHouseholds

47

HouseholdsDwellingsHouseholdsDwellingsHouseholdsDwellingsHouseholdsDwellings

48

Vacant dwellingsHouseholdsMovesMovesMovesMovesMoves

49

Aggregationto zonesMicro dataMicro dataSimulationcompleted

50

Conclusions

51

Microsimulation

New activity-based microsimulation models improveurban simulation models:

- Lifestyles can be represented, i.e. households andindividuals can be disaggregated to the agent level.

- Environmental impacts and feedback can be modelledwith the required spatial resolution.

- Population and employment can be represented by theirdecision making units, i.e. households and firms.

- Microlocations can be represented. Households affectedby environmental impacts can be localised.

PR

OP

OLI

S&

ILU

MA

SS

52

However ...

To date, no full-scale microsimulation model of urban landuse, transport and environment has become operational.There are still unresolved problems regarding the inter-faces between the submodels.The feedback between transport and environmental qualityand location has not yet been implemented.Serious problems of calibration, instability and randomfluctuations have not yet been solved.The computing time for existing models is calculated interms of weeks or days, not hours.

Con

clus

ions

53

Limits of microsimulation

There are ultimate limits to increasing the substantive,spatial and temporal resolution of behavioural models:

- There are theoretical limits when the number of pro-cesses simulated is too small to yield reliable results.

- There are empirical limits when the marginal costs ofobtaining micro data are larger than their added value.

- There are practical limits when the computing time of themodels exceeds the duration of the modelled processes.

- There are ethical limits to the collection of data aboutprivate lives for purposes of research.

Con

clus

ions

54

How much micro is enough?

There seems to be little consideration of the benefits andcosts of microsimulation:- Where is microsimulation really needed?- What is the price for microsimulation?- Would a more aggregate model do?

For spatial planning models, the answer to these questionsdepends on the planning task at hand.

Con

clus

ions

55

Conclusion

These considerations lead to a reassessment of the hypo-thesis that in the future all spatial modelling will be micro-scopic and agent-based.

Under constraints of data collection and of computingtime, there is for each planning problem an optimum levelof conceptual, spatial and temporal resolution.

This suggests to work towards a theory of balanced multi-level models which are as complex as necessary for theplanning task at hand and – to quote Albert Einstein – "assimple as possible but no simpler".

Con

clus

ions

56

More informationMoeckel, R., Schwarze, B., Spiekermann, K., Wegener, M. (2007):Simulating interactions between land use, transport and environment.Proceedings of the 11th World Conference on Transport Research.Berkeley, CA: University of California at Berkeley.

Wagner, P., Wegener, M. (2007): Urban land use, transport andenvironmental models: Experiences with an integrated microscopicapproach. disP 43(170):45–56.

More information on the development of the ILUMASS model can befound at http://www.spiekermann-wegener.de/mod/ilumassmod_e.htm.

Author:Prof. Dr.-Ing. Michael Wegener, Spiekermann & Wegener, Urban andRegional Research, Lindemannstrasse 10, 44137 Dortmund, Germany.mw@spiekermann-wegener.de. http://www.spiekermann-wegener.de.