Skip to content

Model design

MILES is a chain of five components. Each component's output is the next one's input; each has its own page: Synthetic population, Land use, School location, Activities and Traffic assignment.

Stages and what passes between them

The MILES chain. Blue: models; grey: input data; dashed: validation data. The skims are fixed: the assignment does not feed travel times back to the land-use and activity models.
Step Model Input from the previous step Output to the next step
1 Synthetic population (CVAE) Census microdata Households, persons, dwellings, jobs and workplaces of 2016
2 Land use (SILO) The 2016 base Persons, households, dwellings and jobs of every year to 2024
3 School location Students of each year The school of every student
4 Activities (MAST) Population, jobs and schools of 2024 One weekday plan per person: activities, locations, times, tour modes
5 Traffic assignment (MATSim) Weekday plans Routes, departure times, link volumes, travel times

Time line

The base year is 2016. SILO advances the population in annual steps from 2016 to 2024; the school step runs every simulated year. The activity model and the assignment run once, for a weekday of 2024, which is the validation year.

Study area and zone system

The study area is the six counties of the Baltimore region: Baltimore City and the counties of Anne Arundel, Baltimore, Carroll, Harford and Howard. It is embedded in a zone system of 1,588 zones covering Maryland, the District of Columbia and parts of Virginia, Pennsylvania, Delaware and West Virginia, so that commuting and migration across the regional boundary are represented. Land use runs with open migration among 44 counties.

Study area: the six counties of the Baltimore region and the major roads. Inset: location in Maryland.

Level of service

The land-use and activity models read fixed skims:

  • Car: base-year AM-peak times of the Maryland Statewide Transportation Model (MSTM).
  • Transit: door-to-door times computed on the OpenStreetMap network with the timetables (GTFS) of the 18 operators of the network area.

The road network of the assignment (links, lanes, road classes, posted speeds) is also taken from OpenStreetMap; free speeds and capacities are derived from it by fixed rules (assignment).

What is estimated, imposed, calibrated or assumed

Every component is in one of four classes.

Estimated
Parameters are maximum-likelihood or empirical estimates from microdata. This covers the synthetic population, residential location, car ownership, workplace and school choice, activity generation and scheduling, destinations, stops, tour formation and tour mode choice.
Imposed
A quantity is set to observed data: county population and employment totals, construction, deaths, job changes, incomes and school seats.
Calibrated
A parameter is adjusted until a model output matches a target: the moving decision, the workplaces of residents of Pennsylvania, Delaware and West Virginia, and the rates of births, marriages, divorces and leaving home (calibration).
Assumed
A value is set without estimation or calibration: the car skims, the rules for free speeds and capacities, the partner-choice distance decay and the value of time of the assignment.