DyME Case Study

Shaping Climate Resilience Policy through Dynamic Microsimulation

Summary

This case study explores the use of dynamic microsimulations—a computational technique that simulates the behavior and interaction of individual entities to predict complex system outcomes (e.g. effect of local weather patterns on the health of individuals). The case study explores how such techniques could be used to analyse exposure to extreme heat and predict its impact on health outcomes for vulnerable people (e.g. elderly, outside workers). The goal is to use such methods to support public policy and enhance climate resilience.

Project Description

Climate change is escalating public health risks, such as exposure to extreme heat for vulnerable populations. Current epidemiological studies and health impact analyses often rely on aggregate measures of heat across a local area, assuming uniform exposure across populations. However, in reality, exposure to heat varies significantly among individuals depending on their demographic, behavior, location and environment.

A research team based in the UK have developed an AI-based microsimulation that can be used to model and predict how exposure to extreme heat (i.e. the hazard) would affect the health outcomes of vulnerable people. They are working with public health officials and policy-makers to simulate these varying individual-level effects in changing climate scenarios.

Their current approach combines a range of individual-level data, such as daily movement and activity patterns, socio-economic status, and health attributes from synthetic population datasets that are representative of the UK population. These data are integrated with future climate projections, including temperature changes and rainfall from regional climate models adapted to the specificities of local areas. Included in their approach is a risk assessment framework that is used to estimate key variables, such as vulnerability and risk of exposure at the individual level.

Based on their initial success, the team now wish to scale their system for global adoption and application. They wish to ensure that the system achieves similar performance with novel demographic, and activity data, making it a versatile tool for assessing diverse environmental stressors, aiding decision-making across various sectors and locations worldwide.

Technology Description

This tool is based on dynamic microsimulation modelling, designed for scenario projections at a high spatial resolution. These models require as input climate projections data and population health data which are aligned to identify vulnerable population groups in a local area across space and time. The population health data can draw from synthetic population data, which is enriched with socio-economic and health attributes using iterative proportional fitting techniques. Regional climate models deliver the temperature data and need to be bias-corrected for reliable predictions at a spatial resolution matching the population data.

add here technical description on the risk & exposure assessment for health metrics

Key Issues

• Generalisability of the system to different countries (e.g. bias mitigation)
• Data privacy and protection
• Trust and social license. Attitudes and acceptability of system's use across countries.
• Interpretability and explainability of system behaviour to support decison-making

Deliberative Prompts

1. AI-enabled tools and techniques (e.g. microsimulations, digital twins) are claimed to "support local-level decision-making". What needs or capabilities would need to be present at the local level to ensure this opportunity can be realised? How will variations across geographical (e.g. sociopolitical differences) be addressed?
2. What assurance would need to be provided to different stakeholders or end users to ensure they are confident in deploying a tool in a new environment (e.g. evidence of performance, means for contesting or reviewing outputs?
3. What technical and ethical tradeoffs are important to consider when using real vs. synthetic population health data for the model?

Stakeholders and Affected People

• Climate researchers
• Policy makers (specifically in the domain of Health & Healthcare)
• Governments
• Healthcare professionals and providers (e.g. local service providers, emergency response workers, doctors)
• Affected individuals (eg sub-groups of the population especially vulnerable to heat-related conditions) can we give more explicit examples here?

Datasheet

Notes

Discussion with Ruth

Need to incorporate these into the case study.

• Preliminary data work:
  • Population survey that supports synthetic data generation.
• Research question(s) (assuming initial data wrangling and preprocessing):
  • What percentage of {group X}
  • What interventions
• Define "micro-environments" in which interventions could be considered:
• Interventions:
  • Nature-based (e.g. green/blue space, increased shade)
  • Built environment (e.g. biophillic design)
  • Behavioural change (e.g. home environment management)
  • Short-term versus long-term
• Hazard: heat
  • High-risk group(s):
    • Comprised thermoregulation (e.g. elderly people, chronically dehydratyed, taking certain medications, pregnant people (and their foetuses))
    • High exposure (e.g. outside workers, athletes, military personnel, pedestrians)