Introducing BERTIE — an Agent-Based Model for Transport East
Using ABMs to improve transport in the Transport East region
It’s an absolute pleasure to pull back the cover and reveal a project we’ve been working on for the last year with Transport East (TE), one of England’s sub-national transport bodies (SNBs).
Transport East
TE provide a regional perspective to support their partners in Norfolk, Suffolk, Essex, Thurrock, and Southend-on-Sea. They have a strong focus on decarbonisation and improved outcomes for everyone in the region, so they were open to using people-focused modelling to help deliver on their objectives. Their role is:
“To provide a single voice for our councils, business leaders and partners on our region’s transport strategy and strategic transport investment priorities, working in close collaboration with the government and the rest of the UK”.
Before getting into the details of the model, I wanted to say how much we’ve enjoyed working with Transport East, who are genuinely committed to using innovation to better support their local authorities in improving transport services. They recognised that this kind of model gives them something new that’s highly relevant for their regional focus, becoming the first regional transport body in the UK to build an ABM.
BERTIE
In this post, we introduce the BEhaviouR & Transport: Impact & Equity Model, more affectionately known as BERTIE. We will explore the origins of the model and some of the exciting stuff we’ve been doing with it to date.
“The East is facing ever-more complex challenges as it adapts to emerging mobility trends, changing behaviours, and unexpected shocks. Transport East is committed to thinking creatively to solve the transport challenges we face to improve outcomes for our communities, now and in the future.
But how do we achieve this in a sustainable and equitable way? This is the power of BERTIE — our Agent-Based Model, developed by Arup.”
Dan Johnson, Senior Transport Planner, Transport East
We built two versions of the model, one to represent pre-pandemic 2019 (which has the most available data) and a future forecast for 2040 (for long-term scenario testing). We undertook a wide range of analyses on the outputs and ran several different scenarios to test the model’s responses (more on these later).
The Network: Bigger is Better
The network covers a huge geographic area, as shown in the map above, including full coverage of public transport services and the entire road network in the core study area bounded by the red border. There are varying levels of detail across the road network outside the core study area to cover trips that begin or end outside the region, e.g. long-distance freight.
The road and public transport networks are joined to allow agents to make fully multimodal trips. A trip joins two locations and can comprise multiple legs, where each leg might use a different mode. For example, your daily trip from home to work could involve walking to a station, catching a train, and then a bus trip to your final destination.
The Population: Two profiles of transport demand
We created two versions of the population: a baseline to represent the present day, and a 2040 population to reflect expected population changes in the future.
We used our usual population and activity synthesis process to generate a population of agents with activity plans for 2019. We then used the Office of National Statistic’s NTEM forecasts to incorporate future growth assumptions into the 2040 population.
We used PAM to modify this future population to represent future working-from-home patterns. We picked specific job categories and allocated a subset of agents with these jobs and changed their work locations to become home-based.
Decarbonisation of transport is one of Transport East’s strategic priorities, so BERTIE needs to understand future vehicle mixes in order to analyse carbon emissions. Bearing this in mind, the final ingredient for our 2040 population was the level of Electric Vehicle (EV) uptake in 2040. We used the Department for Transport’s assumptions about future levels of EV uptake, coupled with some additional modelling to distribute EVs sensibly across the population.
We also built a road freight population based on data from National Highways to cover non-household demand.
What did we test?
Our goal for BERTIE is to answer questions, test assumptions, and generate insights to make better decisions. We tested several different scenarios to show a range of possible outcomes on top of the 2040 base model:
- EV uptake: We ran two scenarios, looking at the most likely and best-case uptake levels for the population. How much operational carbon is left once most of the population has switched to an electric vehicle?
- Active mode improvement: We ran a scenario where we boosted the attractiveness of walking and cycling. What does this do to people’s transport choices?
- Increasing the cost of driving: We tested the response to an increase in the cost of driving. We scaled up the costs of car use to reflect this to give us a straightforward scenario. EVs were assumed to be cheaper to run than combustion engine vehicles, but otherwise, we made driving more expensive. We wanted to understand who would be impacted by these changes.
- Combination scenario: The structure of our simulations means that we can combine all the above and see what happens. Are benefits cancelled out, or do we see an overall effect greater than the sum of its parts?
Each scenario required large amounts of compute power (BERTIE runs on AWS) and some nontrivial orchestration and processing. We generated comparable outputs for each scenario and the baselines, which we then compared to build valuable insights.
What analysis did we do?
Delving into the detail of so many simulations requires a tremendous amount of effort, so we focused on a few critical analysis themes that ran across all of the model runs:
- Carbon emissions: What are total carbon emissions? How are they distributed?
- Mode shift: How are people choosing to travel? How does it change when we run a scenario?
- Short journeys: What behaviour do we see for all trips under 5km?
- Equity: How do different groups of people’s experiences differ? How do changes impact these groups?
- Utility analysis: How much are people able to adapt to changes? Who could have used public transport but didn’t?
Each of these analyses is involved enough to become a blog post in its own right, so we’ll describe some of them in more detail in future posts.
What did we find?
Running a large number of highly detailed simulations leads to a wealth of insights — more than we could usefully discuss in a single blog post. But as a quick teaser, I wanted to share an interesting comparison of equity between each scenario.
We can analyse simulation outputs using basic demographic attributes such as agent age, gender, or income level, but we also capture transport-related personal attributes at the agent level, including possession of a driving licence and access to a private car.
The chart below shows how people with and without private car access responded to each scenario. We can see how agent utility (how efficiently they completed their daily activities) changes compared to the 2040 baseline.
The chart shows us that people without access to a car were disbenefitted (had lower utility) more than people with access to a car in the scenarios where we increased the cost of driving.
At first glance, this seems counter-intuitive — why should people who don’t drive be the most negatively affected when we increase the cost of driving? We needed to dig a little more to understand what was going on.
Further analysis uncovered the answer. Increasing the cost of driving prompts drivers to switch to public transport, which is a good thing from a decarbonisation perspective. However, this shift produced an important side-effect — more crowding on the public transport network. Thus, many agents without cars were crowded out of their usual buses and could not reach their activities on time.
This finding leads us to the conclusion that interventions aimed at moving people out of cars and onto public transport must also include improvements to public transport services. While this may not be a controversial statement, being able to quantify the impacts via the simulation is hugely novel.
Happily, we can use our agent-based models to test combined changes to both the public transport network and the cost of driving to discover the best overall strategy. This kind of complex response across multiple modes is exactly why Transport East are so excited about the possibilities that BERTIE opens up.
“BERTIE presents a huge opportunity to rethink how we plan transport in the East — to understand how the region can meet its goals through changes within an integrated transport system that really shifts the way people and goods move around.
As we focus on the solutions the region will need over the next 30 years, BERTIE will give us strong evidence for new interventions.”
Esme Yuill, Communications Lead, Transport East
What’s next?
We are currently working on the next iteration of BERTIE, refining the model’s assumptions and performance, which will help inform and prioritise transport investment across the region, focusing on delivering decarbonisation in an equitable way and supporting growth.
It’s exciting for us to see another example of how our modelling aligns so well with a client’s objectives and the outcomes they’re trying to achieve. We are confident that this people-focused approach to thinking about transport will go from strength to strength as organisations like Transport East turn these simulations into real live transport policy.