Agent-based measurement

A synthetic human layer that turns static OOH metrics into living behavior.

Sweden ABM builds a full-scale synthetic population from SCB distributions and uses it to power mobility-aware exposure measurement. We are doing this to move beyond traffic counts and reach estimates, and instead measure how real people with real demographic and household context move through the OOH landscape across the day and the year.

Open Human Layer Open ABM Report

Population size

10,593,249 persons

Household base

5,117,752 households

Validation snapshot

3 core test suites passing

Car totals aligned

4,377,337 vehicles

Why ABM

OOH measurement needs behavior, not just counts.

The ABM approach lets us model where people actually are, when they are there, and who they are. That makes impressions time-weighted, demographically grounded, and ready for scenario testing.

Behavioral realism Day/night population shifts, commute flows, and household context are all captured.

Audience precision Agents carry education, occupation, income deciles, and social class markers.

Measurement continuity ABM outputs connect directly to the Human Layer and ABM report surfaces.

Foundation

Grounded in SCB reality.

The synthetic population is built from 37 SCB tables across demographics, households, labour, income, vehicles, and education. The pipeline preserves DeSO-level structure while providing municipal and national fallbacks to keep every region complete.

DeSO distributions Household synthesis Telework + health OD commuting

ABM surfaces

Explore the active population synthesis views

The landing page is the front door. Deep detail lives in the Human Layer and the report, with more specialist views queued for release.

Core

Human Layer

Interactive exploration of the synthetic population: household structure, demographics, and spatial distributions at DeSO resolution.

Agents, households, and demographic filters
Day/night area types and special populations

Open view

Core

QA/QC population synthesis report

Narrative readout of the model, validation checks, and what the population build implies for exposure measurement in Sweden.

Validation summary and statistics
Methodology and pipeline notes

Open report

Planned

Validation report

A focused QA surface that tracks sanity checks, margin comparisons, and data alignment over time.

Household integrity and demographic checks
Car ownership reconciliation vs SCB totals

Coming soon

How it is built

From SCB to agents in four stages

The pipeline is engineered for repeatable regeneration, with distributions curated from raw SCB tables and output validation baked into the build.

SCB ingestion

37 SCB datasets pulled from PxAPI 2.0 plus PxWeb 1 (RAMS) for occupation and industry data. Metadata is stored alongside the raw tables.

PxAPI 2.0 PxWeb 1 Metadata manifest

Distribution curation

Distributions cover age/sex, households, employment, education, income, tenure, vehicles, health, telework, and day/night population.

Household compositions Education priors Car ownership

Population synthesis

Agents and households are generated at DeSO resolution and written to Parquet for downstream mobility and exposure models.

Persons.parquet Households.parquet DeSO indexing

Validation + alignment

Sanity checks and aggregate comparisons verify demographic consistency, household logic, and car ownership alignment with SCB totals.

Sanity checks Margin comparisons Vehicle rebalance

Validation snapshot

Key results from the latest full build

The pre-grid validation confirms population integrity, household consistency, and alignment with SCB totals before mobility modeling begins.

Population scale

10,593,249 persons across 5,117,752 households.

Mean household size 2.07, median 2 persons.

Demographic balance

51.1% women, 48.9% men. Median age 38, mean age 40.29.

Under 18: 27.07%, 65+: 22.85%.

Vehicle alignment

Car totals rebalanced to match SCB: 4,377,337 vehicles.

62.25% of households have at least one car.

Household distribution

Most common sizes: 1-person (2,154,234) and 2-person (1,614,972).

3-person: 547,844; 4-person: 552,670.

Next releases

Expanding the ABM stack.

Upcoming work focuses on replacing heuristic special-population flags with official datasets, deepening workplace layers, and extending the ABM reporting suite.

Official homeless + tourism data

Workplace layer integration

Extended validation reporting