Challenges and a Solution in Coupling Dissimilar Models for Complex Planning Policy Analysis

Harutyun Shahumyan and Rolf Moeckel

National Center for Smart Growth Research and Education, University of Maryland

Methodology

Python wrappers were developed to loosely couple modelsdeveloped in different environments. ArcGIS Model Builderwas used to provide a graphical user interface and topresent the models’ links and workflow. With the use ofPython wrappers, the implementation of the coupler isseparated from the models’ source codes. This gives aflexibility, which can help in terms of portability,performance and maintenance of the codes.

Benefits

• No need to change the source codes of the models.• Runs models developed in different environments. • Can be extended with additional models over time.• General user interface showing process flow.• Rich visualisation & mapping capabilities with ArcGIS.• Easy to implement.

Limitations

• Parallel model runs and dynamic data exchange during simulation time steps are not supported.

• Model processes run independently from one another.• Data exchanged between modules are written to and

read from a hard drive. No in-memory data exchange.

Conclusion

This approach is especially efficient when the models aredeveloped in different programming languages, theirsource codes are not available or the licensing restrictionsmake other coupling approaches infeasible. A key findingof this research is that model integration should depend ondirection of information exchange and frequency of dataflows, as shown below. While this simple but robust looseintegration has satisfied the project’s initial goals, furthertighter integration within the CSDMS is currently exploredto enhance model performance and data exchange as wellas to widen the scope of applications.

Summary

Close model integration has become the mantra amongmodel developers. New tools under development, such asCSDMS or OpenMI, promote tight integration of verydifferent models and ease information transfer betweenthe same. Continuously increasing computationalcapacities enable ever more comprehensive modelintegrations. From a technical perspective, the prospects oftight model integration are excellent. However, theresearch presented also exemplified limitations anddifficulties of model integration.

Models Used for Integration

The modeling system needs to integrate models of thefollowing domains

- Transport: MSTM, Maryland Statewide Transport Model- Land Use: SILO, Simple Integrated Land Use Orchestrator- Transport Emission: MEM, Mobile Emissions Model- Immobile Emissions: BEM, Building Energy Consumption- Land Cover: CBLCM, Chesapeake Bay Land Change Model

Several models require frequent data exchange, as shownbetween the transportation and land use model below.

harut@umd.edu (301) 405 1112

rolf.moeckel@udo.edu (301) 405 9424

http://smartgrowth.umd.edu

ModelEnviron

ment

Operation

SystemLicensing

Simulation

Period

Sim.

yearsRuntime

MSTM CUBE Windows

Scripts: Open

source

CUBE: CitiLabs

2007 or 2030 115-16

houra

SILO Java Multi-platform Open source 2007-2030 234-5

houra

MEM CUBE Windows USGS, CitiLabs 2007 or 2030 1< 30

mina

BEM Excel Multi-platform n/a 2007 or 2030 1 < 1 mina

CBLCM C / C++ CentOS USGS 2007-2030 4 3 hourb

Main characteristics of the used models

a 20 x AMD Opteron Processor 6328 @ 3.20GHz, 42GB RAM, Windows 7 Virtual Machineb 2 x 2.56 GHz CPU’s, 24GB RAM, Centos 6 Virtual Machine 2000-

2007

• SILO

2007

• MSTM

• MEM

• CBLCM

2007-2030

• SILO

2030

• MSTM

• MEM

• BEM

• CBLCM

2030-2040

• SILO

To

From

MSTM SILO MEM BEM CBLCM

MSTM

Auto travel time between zones

Transit travel time between zones

Auto-operating costs

All trips within the region

Average speed distribution

SILO

Population

Employment

Auto availability

Building data: type, age, area, rooms, occupation, heating fuel, location, etc.

Population

Employment

Zonal accessibility to population by auto & transit

Data exchange between models

The models’ processing flow order and simulation periods.

ArcGIS geo-processing model organizing the models simulation workflow

SILOMSTM Auto-

ownership

model

Skimming

Trip

Generation

Trip

Distributions

Assignment

Household

relocation

Developer

location

choice

Mode choice

Auto availability

Transit travel time

Households

by zone

Auto-

operating

costs

Auto travel times