Inventory of population synthesizers

Commissioned by SIVMO, a partnership of Dutch governments and infrastructure authorities, Panteia conducted research into methods and tools for population synthesis. The study focuses on the role of synthetic populations in transport models, with attention to Dutch data sources, privacy rules, geographical scale levels and the transition towards tour-based, activity-based and agent-based models.

The report “Population synthesis: an exploration of methods and tools for SIVMO” was prepared by Jan Kiel 

of Panteia. It provides an overview of existing population synthesis methods and tools, assesses their relevance for the Dutch modelling practice, and advises on possible routes for further implementation within SIVMO.

Background
Transport models increasingly require detailed information about persons and households. This applies to tour-based, activity-based and agent-based models, where travel behaviour is linked to household composition, age, work status, income, car ownership and other characteristics.

A synthetic population is a statistically plausible representation of a real population. It usually combines microdata, such as person or household surveys, with aggregated control totals, such as demographic statistics. The result is a population of synthetic households and persons that can be used as input for transport models.

For SIVMO, population synthesis is relevant because it can support a shared basis for national and regional model applications. It can also make it possible to analyse policy effects for dynamically defined target groups, for example by income, household type, age or access to a car. This requires a synthesis process that is transparent, reproducible, scalable and suitable for use within the Dutch data and privacy context.

Overview of the report
The report provides an inventory of population synthesis methods and tools. It covers both national and international examples and combines literature review, tool analysis and interviews with experts and organisations involved in population synthesis.

The main topics covered in the report are:

• Methodological families. The report distinguishes between iterative reweighting methods, optimisation-based methods, probabilistic reconstruction, simulation or generative methods, and data fusion or hybrid approaches.
• Software tools. Several tools are described and compared, including SigPopu, Quad, Octavius, PopulationSim, PopGen, PopSynWin, MATSim, SynthPop and other population synthesis systems.
• Dutch data context. The report discusses the role of CBS data, ODiN, geographical scale levels, data suppression, PRAM noise, Remote Access conditions and privacy-related constraints.
• Practical experience. Interviews were held with PTV, the UK Department for Transport, TNO, CBS, Goudappel and Significance. These interviews provided insight into implementation, validation, maintenance, governance and tool usability.
• Assessment framework. The tools are assessed on method, scalability, reproducibility, validation, maintenance, accessibility and suitability for the Dutch transport modelling practice.

Results
The study concludes that there is no single method or tool that is suitable for all applications. The usefulness of a population synthesizer depends on the model type, available data, geographical scale, privacy constraints and the need for reproducible results.

A key finding is that practical success depends less on the algorithm itself and more on data quality, consistent control totals, clear definitions, validation, documentation and governance. Population synthesis should therefore be treated as a managed process, not only as a software choice.

The report identifies three possible routes for SIVMO:

• Route A: standardising an existing tool already used in the Dutch modelling practice.
• Route B: adopting and adapting an existing open-source synthesizer for the Dutch context.
• Route C: developing a new generic population synthesizer specifically for SIVMO.

The recommended direction is Route B: using an existing open, optimisation-based synthesizer as a starting point and adapting it to Dutch data sources, zonings, privacy rules and model requirements. This route offers a balance between transparency, reproducibility, adaptability and manageable development effort.

The report also recommends that SIVMO first defines the required variables, scale levels, output format and validation steps. A pilot on a limited study area is proposed as a practical next step, before wider application or standardisation.

About SIVMO
SIVMO stands for Samenwerkingsverband en Innovatie Verkeersmodellen door Overheden. It is a partnership of Dutch governments and infrastructure authorities that focuses on the development and improvement of transport modelling methods. The partnership includes Rijkswaterstaat, ProRail, provinces, transport regions and major municipalities.

Through studies such as this inventory of population synthesizers, SIVMO supports the gradual renewal of transport modelling practice in the Netherlands.

More information and download
Interested parties, including policymakers, researchers, model developers and transport planning practitioners, are invited to download the report.

Download the full report here.