Def: City Modelling
An overview of our frameworks/tools and the terminology we use.
The world is a complex place — cities especially are complex systems. We think modelling them better will likely require the adoption and integration of many domains of knowledge and practice.
In Arup’s City Modelling Lab, we have been tinkering for a while now. Bringing together ideas and methodologies from a diverse range of theoretical and applied fields. From computing and statistics, via economics and transport, to human behaviour and ethics.
The breadth of questions we can answer through a more interdisciplinary practice are exciting. Arup has long been at the frontier — integrating theory and practice across domains in new ways. But how do we articulate our work in this new space? What language best reflects our work, and perhaps more importantly, what language makes sense to everyone else?
So read on for our definitions of words we like to use, including:
- Model vs. Simulation
- Strategic Transport Model
- Agent-based Model (ABM)
- Activity-based Model (AcBM)
- Population Synthesis
- Traffic Modelling vs Microscopic Traffic Modelling
- City Model
- Extends/Boundaries
These are our own definitions, they are not axiomatic. Think we got something wrong? We’d love your feedback!
Model vs Simulation
We use the word model a lot. We usually mean mathematical model, as in a representation of some real system like a city. Sometimes we mean physical mathematical model, like transit vehicles moving around a transport network. Sometimes we mean statistical mathematical model, like predicting where someone will choose to get their hair cut based on some past observations. We extend statistical models to include machine learning models.
When we use our models, we give them some data and/or some initial conditions and let them run. This gives us a simulation — a sort of prediction of the real or future world. Our simulations are our outputs, we are designing/training them to look like the real-world as much as possible.
Strategic Transport Modelling
In the field of transport, a transport model predicts the demand for travel. Transport models simplify and aggregate demand into trips or tours. Where a trip is a simple origin and destination journey and a tour is a more complex but still finite combination of trips.
We care a lot about agent decisions in our models, including decisions about travel. But these trip and tour simplifications fail to represent the majority of people’s real daily lives. Additionally, aggregation is contrary to the ABM philosophy. Travel, and more broadly behaviour, is not an aggregated decision made in isolation. It is an individual’s decision made in consideration of past events, future expectations and present concerns.
More recently the transport modelling field has started to adopt activity plans for modelling travel demand. We like activity plans combined with ABMs — together they capture and account for the variability of the real world and the needs of the individual. They also allow us to model some really complex processes.
Agent-based Model (ABM)
Agent-based models (ABMs) or modelling is a computational approach for simulating complex systems. It has been kicking around (very usefully) since the 1970s. One of the first ABMs, was Thomas Schelling’s Segregation Model, explored wonderfully here.
ABMs simulate the actions and interactions of autonomous agents within an environment (such as people within a city). They use agent level rules for behaviour with each other and their environment, from which complex systems are modelled and emergent global outcomes investigated.
We like ABMs because they offer a framework for better considering the complexity of cities. But we also like them because they provide direct observation of simulated real-world interactions. This makes engagement and communication with our work easier. These interactions also make feasible the use of new and more precise data for model validation. Giving us more confidence in the the quality of our models and allowing us to add more features. Stay tuned for a more detailed post on this concept.
For us, agents are individuals. Maintaining the concept of the individual is key in designing cities for actual people. People are unique in their behaviours, values, restrictions and needs and ABMs can allow for this.
Activity Based Models (AcBM)
Some of the latest strategic transport models are called activity-based models (AcBMs) or activity-based demand models. Their primary purpose is to model people’s demand for travel as chains of activities/plans. This is achieved through chains of statistical models for predicting specific choices. Choices range from the long-term, such as where to live and work, to the medium term, such as if to stop by the shop on the way home, to short-term, such as what time to leave the house. Confusingly, in addition to sharing an acronym, these activity-based models are sometimes called ‘pseudo’ agent-based models.
Much of the challenge of modelling cities is modelling agent choices. Where choices can be responding to any number of past, present and expected interactions with the environment or other agents. The ‘pseudo’ ABMs discussed above only very loosely model the autonomy of agents within their simulated environment. This is somewhat antithetical to the ABM approach. In addition, they typically use simplistic simulated environments and interactions. We build on their merits, by using more complex simulations and using a framework that models choice within these simulations.
We use a software framework called Multi Agent Transport Simulation (MATSim). MATSim is agent based and models shorter-term activity choices, such as modes, times and routes. For longer term choices we rely either on an AcBM as described above, or increasingly, on existing data, using population synthesis techniques.
Population Synthesis
A term to describe the creation of agents used in an ABM simulation. Population synthesis allocates attributes to agents. The attributes impact agents’ behaviours and therefore the outcome of the simulation.
For our work, we are focussed on making representative populations; we want our agents to reflect the real world population. Representation includes both overall population patterns, such as wealth distribution, and individual characteristics, such as living in a certain area and going to a certain school. We want our agents to not only have the same characteristics as real people, but also realistic and unique behaviours.
We are implementing existing techniques for agent synthesis and experimenting with some new ones. Our aim is to produce more detailed and representative populations from traditional sources (such as census surveys). Or use new sources when traditional data isn’t available (such as GPS data). Stay Tuned.
Traffic Modelling vs Microscopic Traffic Modelling
In addition to the population of agents in an ABM, we need to model the transport network. The transport network defines where and how agents in our model can travel.
Not all traffic models are created equal. A key feature of our work is the scale and complexity of the transport model that we use for simulation. We describe this as a microscopic model to differentiate it from less detailed models. We model a complex physical world that includes detailed multi-modal networks (incorporating walking, cycling, public transit, different vehicle types), scheduled transit operations, realistic building/activity locations and many of the corresponding interactions.
City Model
Now that we have introduced some of the component parts of a city model, we can define what we mean when we talk about building a City Model.
We consider the scope of City Modelling to be broad. It involves capabilities and techniques from domains like land-use modelling, economic modelling, air-quality modelling, behavioural modelling, and more. We use transport modelling (using ABMs) as a key thread to bring together these different fields. Broadly, we can do this because our transport models don’t just model agents’ movements and trips but also agent activities. By using ABMs and traffic models as our foundation, and integrating system modelling from other domains— we can start to build truly detailed and powerful models that reflect some of the interconnected systems within cities.
Extends/Boundaries
We keep saying city modelling — but actually we mean bigger. Cities have huge catchments and we have to allow for people to get access into the city or between cities. This becomes a massive computational challenge — so far we haven’t gone much bigger than a country. But please stay tuned.
