Probabilistic machine learning methods for blending weather forecasts towards optimal health risk prediction

Background

Temperature records have been broken repeatedly in recent decades and heat-stress is increasingly a serious health risk (e.g. 2007 London marathon fatalities). This project aims to quantify the spatially and temporally varying heat-risk in the UK using statistical models and to understand how weather forecasts and climate projections can be leveraged for informing adaptation policy and mitigation action such as early warnings. Interpretable probabilistic machine learning (ML) methods will be used to model weather and health data available at the Met Office so the sits at the interface between ML, environmental science, meteorology and epidemiology. Skills such as data analysis and modelling, environmental and health data manipulation, risk mapping and decision making under uncertainty are expected to be gained by the student, who will have the chance to be hosted at the Met Office as a visiting scientist.

PhD Opportunity

The first challenge is the question of how to best use statistical machine learning methods for understanding the relationship between heat-stress and health outcomes. Heat-stress is a combination of unfavourable temperature, humidity and wind-speed over a generally unknown number of hours/days etc. Appropriate data modelling tools will need to be utilised to understand health-risk as a function of heat-stress, allowing for socio-economic factors of the population-at-risk, in addition to the inherent spatio-temporal variability.

Next, is the question of how to use the estimated risk in mitigation and adaptation strategies. For mitigation, prescriptive (decision making) approaches for issuing health warnings will be investigated. For adaptation, future climate projections of heat-risk will be computed and contrasted for various climate change scenarios and climate models. Statistical machine learning methods will be used to probabilistically quantify the uncertainty from the various sources of climate projections, enabling their use in decision making.

A key challenge is the question of how to optimally use the available weather and health data for quantifying the health risks and whether blending various data sources is beneficial for reliable forecasting.

Expertise and data from the Met Office (MO) will be available towards tackling these challenges, noting that the current ‘heat-health alert service’ is co-managed by the MO.

Applicant Profile

Students with a strong background in statistics and/or machine learning who want to apply these skills to environmental epidemiology and more generally the interface between climate change and health

Other information

https://experts.exeter.ac.uk/19235-theo-economou
https://experts.exeter.ac.uk/43970-christophe-sarran
https://www.metoffice.gov.uk/weather/warnings-and-advice/seasonal-advice/heat-health-alert-service
https://www.metoffice.gov.uk/research/climate/climate-impacts/health