Workshop 2 (Full Day – English)
Bayesian spatiotemporal modelling for environmental data with NIMBLE
Course Faculty: Garyfallos Konstantinoudis (United Kingdom), Robbie Parks (USA)
Time: 09.00 – 13.00 & 14.00-18:00
Target Audience: PhD students and researchers interested in Bayesian models and R
About This Workshop:
In environmental epidemiology, spatial and spatiotemporal methods are increasingly used with high-resolution environmental and health data, which allows detailed insights into associations between exposures and outcomes. In this context, Bayesian methods provide a natural setting to incorporate uncertainty and prior knowledge, borrow information between neighbouring units, and create hierarchical structures. However, appropriate usage of these methods requires careful consideration and knowledge of building technical models, which can be intimidating to the uninitiated user. During this full-day course, our aim is to introduce the ideas of Bayesian inference and modelling for spatiotemporal environmental data. The workshop is designed to be as approachable and friendly as possible while still providing technical and practical know-how. The objectives of the workshop are:
- An Introduction to Bayesian inference
- An Introduction to hierarchical model structures
- An Introduction to spatiotemporal priors
- Applications in real world environmental data and health outcomes
We will focus on three illustrative case studies. In the first case study, we will evaluate the effect of long-term air-pollution exposure on COVID-19 mortality in England and examine the crucial role of spatial autocorrelation. In the second case study, we will estimate the non-linear effect of temperature in all-cause mortality in Italy and examine how this effect has changed over time and varies in space using spatiotemporal Gaussian processes. In the third case study, we will show how spatiotemporal models can be used to evaluate excess deaths due to events such as heatwaves in Switzerland and ways to propagate the different sources of uncertainty. All analysis will be performed in R with the NIMBLE software and data and code will be made available for future studies. By the end of this workshop, attendants shall develop an understanding of Bayesian spatiotemporal models and be able to write their first spatiotemporal model in NIMBLE.