Simulating the human cost of climate change

Annual temperatures in the UK average a daily low of 6 degrees and a high of 14 Celsius, which during the summer months ranges from 15 to 25 degrees Celsius. In 2021 and 2022, temperatures during the summer months recorded the highest UK-wide average maximum daily temperatures since 1960. Specifically, in 2022, the extreme heat experienced between June and August was far more intense and widespread than previous comparable heatwaves, with a maximum of 40°C recorded for the first time in the UK’s history (data from currentresults.com). Whilst temperature increases for a few days of the year may not seem significant, when assessed against historical data, it is evident that extreme heat is a result of climate breakdown and has prolonged impacts on wider populations. According to the Department of Health and Social Care, severe heat will affect everyone’s health and increase heat-related mortality, which will inordinately affect vulnerable communities with chronic and severe illness, disability and poor housing conditions.

Traditional means of estimating the risks associated with higher temperatures are based on aggregate measures obtained from epidemiological studies, which are both sparse and expensive to replicate. Sensor-based monitoring systems and crowdsourced reporting efforts often help fill gaps in the data, but their qualities remain varied with limited measures for error corrections. Furthermore, measures such as the average temperature within an urban area assume that all members of the population experience the same temperature. However, how heat is experienced by an individual differs depending on their activities; their health conditions and lifestyle, the amount of time they spend outdoors and indoors, their working and living conditions, their daily activities and movements, and their modes of transport. Exposure, therefore, differs with activity, which subsequently interacts with their underlying vulnerabilities, such as existing co-morbidities and lack of resilience to health stressors, to increase or decrease an individual’s relative risk of adverse health outcomes due to heat exposure.

To predict how climate change or extreme heat might impact a given population and support creating coherent response strategies, decision-makers in local and national governments need granular knowledge of the dynamics and demographics of a changing population, the various communities within it and the potential consequences of policy and mitigation plans proposed for them. This requires a comprehensive analysis of information drawn from multiple domains including urban infrastructure, urban planning, transportation networks, land use, air quality, regional energy systems, public health (such as population exposure to heat), existing mitigation plans and other sources. While it would be highly challenging, and potentially invasive or unethical, to gather all of this data for all individuals within an area, researchers can build synthetic populations using aggregate data representative of the population of interest, such as from censuses and traffic movements, and simulate population characteristics using an analytical technique known as microsimulation.

AI method in focus: Microsimulations

Simulations are a computational approach to mathematical modelling, designed to derive conclusions from data by applying a set of algorithms predicting the behaviour of real-world or physical systems. Monitoring and forecasting are data-based modelling techniques that represent real-life systems to help capture data reflecting a specific state, such as temperature in a specific location, or effects of changes such as measuring the impact of climate change on the environment. Microsimulation techniques, unlike other modelling approaches, operate at the level of individual units such as persons within a population, rather than aggregate population data. There are two widely known simulation approaches - agent-based models and microsimulations (DOI: 10.1146/annurev-statistics-010814-020218). Agent-based models are always probabilistic, whereas microsimulations_ _are always data-driven. Before implementing a proposed intervention in the real world, microsimulation is often used for evaluating the impact of a possible outcome of a suggested policy. Despite the differences in applications-level, both approaches are identical from a mathematical and computational perspective – they both apply a series of algorithms to data (reflecting a specific state) and address the problem of probabilistically reproducing real-life phenomena.

Microsimulation models use unit-level data to construct a collection such as a group of individuals to represent a target population and their behaviours. Applying a series of algorithms and predetermined probabilistic rules, the model simulates changes in an individual’s characteristics over time giving different versions of the outcomes. Microsimulation models can be classified into ‘static’ and ‘dynamic’. Static microsimulation is a snapshot at a single point in time that can apply a change (such as ageing) to a representative sample, and look at the impact based on the expected characteristics derived from known data. Dynamic microsimulation models apply specific processes (other factors related to ageing) to a representative sample and predict the distributional impact of some intervention of interest, such as the impact of a health risk factor on the future likelihood of disease, or a policy change to improve healthcare access for different populations.

Microsimulation was first applied to demographic problems in 1957 and has remained a popular approach in demographic research (Schonfield et al, 2018). This technique is increasingly favoured for studies that are generally expensive, unsafe or unethical (such as in clinical trials), impractical or time-consuming for their implementations in the real world. It flips the angle of policymaking by modelling the interacting behaviour of individuals, families, and organisations within a larger system, where algorithms represent behavioural processes changing through time for each individual in the entire population of decision-makers. Taking complex interactions across subpopulations and considering the behaviours of an individual in connection with their environment or society (measured at multiple points in time), microsimulations predict the effects of policy changes on individuals.