Inequality is airborne
Unveiling disparities in exposure to air pollution using agent-based models
The UK government cite poor air quality as the largest environmental risk to public health in the nation. Transport is a major contributor to harmful pollutants such as NOx and PM2.5. Though studies often show that low-income groups experience higher levels of air pollution, understanding the underlying factors driving this vulnerability remains a challenge.
At the City Modelling Lab (CML) we are developing a methodology to use Agent Based Modelling (ABM) simulation outputs to track exposure to air pollutants emitted by simulated road traffic. In using vehicle emissions to inform the dispersion modelling of air pollutants we can measure the exposure of agents. This allows for a feedback loop to provide data-driven insights for decision-makers trialling policies to improve air quality and observe their potential impacts on different groups of the population.
The opportunity
Epidemiological studies researching exposure to air pollutants typically only consider the residential locations of agents. This approach overlooks daily space-time activity patterns, hindering a comprehensive understanding of exposure disparities and the underlying driving factors.
ABMs provide a powerful opportunity to use simulated day plans of individual agents to tally exposure throughout their day. Our behavioural analysis approach allows us to capture responses to policy interventions and critically understand the causes of unintended consequences of policies. In applying our approach to equity analysis (introduced here) we can understand if and which groups of people are most exposed to air pollution from vehicles.
Our approach
Our ABMs operate at a low spatial level with high temporal resolution, allowing us to simulate the individual, and their choices, throughout every second in the day. We previously introduced Pandia, our carbon emissions toolkit, which takes advantage of this foundation to calculate carbon emissions and interrogate who is emitting, where, and when these emissions are occurring.
We have expanded Pandia’s ability to calculate multiple emission types such as PM2.5 and NOx. From here, we can spatially aggregate vehicle emissions to link level and then up to geographical areas. In aggregating the hourly average rate of emissions (g/s) at an h3 grid we gain an understanding of both the spatial and temporal variation of emissions throughout the day.
Using the simulated mobility plans, we can track an agent’s journey and activities throughout their day and tally their exposure to emissions as they move through time and space (and the h3 emissions grid). Based on existing research, exposure to air pollution is affected by whether an agent is indoors (completing an activity in our simulations) or outdoors. When on an active travel leg of a journey, an agent will experience the full effect of outdoor exposure, however, when in a vehicle this will be reduced by varying amounts depending on the mode of transport.
Sheffield insights
We applied this methodology to our ABM for the Sheffield region. Here are a few insights we found.
Demographic attributes have a considerable influence on total exposure emissions. Statistically significant variations were observed across household income, age groups, and car availability. In contrast, gender emerged as a non-contributor to exposure differences, suggesting the absence of gender-based exposure disparities. Young, low-income individuals without car access are identified to be the groups exposed to the highest levels of emissions.
Compared to exposure during other transport modes or activities, the level of active travel exposure emerges as a key determinant of total daily exposure. This aligns with the understanding that active travel often entails traversing densely populated areas with elevated pollutant concentrations. Active travel trip legs are prevalent among individuals without car access and belonging to low-income households. This alignment with anticipated behaviour reinforces the pivotal role of economic considerations in mode selection and its resulting effect on exposure to pollutants.
However, the duration of active travel legs for an individual is a weak predictor of their active travel exposure. That is to say, there is little correlation between active travel duration in an agents plan and their exposure during active travel. This disconnect suggests that where agents walk and cycle (rather than their duration of active travel) could be of particular importance in determining their levels of exposure.
What next?
The current methodology aggregates emissions to geographical boundaries in identical ways. However, different pollutants emitted from vehicles will react differently to meteorological conditions such as wind speed, temperature and sunlight based on the weight and reactivity of the particles. The use of dispersion modelling to simulate the distribution of different emission types is a research area of focus going forward.
Summarising the insights from our Sheffield simulation;
- Low-income, young and those without access to a car are those most likely to be exposed to the highest levels of air pollution.
- Exposure during active travel is strongly correlated with total exposure, and low-income agents are more likely to complete active travel legs.
These insights raise questions about the inequity of why non-emitting agents, who are already at a disadvantage, are those most exposed to air pollution. This capability to model exposure to air pollution from vehicles will enable our clients to form policies that enable a more sustainable and equitable transport system.
If you have any questions then please get in touch — citymodelling@arup.com
