Forecasting the spatio-temporal evolution of natural hazards

Project Summary: You will develop a new Bayesian framework for forecasting the occurrence and evolution of natural hazards by
adapting a cutting-edge spatio-temporal modelling package, originally developed for ecological systems.

Project Background: Forecasting the spatial- temporal evolution of natural hazards is an important societal issue. Statistically, there are several interesting challenges including (i) how to model the spatio-temporal evolution efficiently, (ii) how to deal with self-exciting cluster processes such as the triggering of aftershocks or the spread of fire, (iii) how to incorporate heavy tail statistics. In this project you will adapt an open-source R package (http://inlabru.org) that has been developed by theoretical and applied statisticians primarily for modelling ecological systems. These cutting-edge methods are highly efficient using Nested Laplace Approximations for improved performance compared with Markov Chain Monte Carlo methods. This innovation has led to the application of more realistic and flexible spatio- temporal models than was previously possible. You will build on preliminary work using earthquakes, landslides and fire datasets with point process and raster based modelling.

The supervisory team brings a range of expertise:
- Dr Mark Naylor is a physicist and statistical seismologist with expertise in natural hazards and statistical seismology.
- Prof Finn Lindgren is the Chair in Statistics at the University of Edinburgh and one of the main theoretical and software developers behind inlabru.
- Prof Janine Illian is the Chair in Statistics at the University Glasgow (from July 2019), focusing on spatio-temporal point-process methodology that is relevant in practice.
- Dr David Milodowski is a postdoctoral ecologist with extensive experience of spatio- temporal remote sensing and is developing an inlabru framework to understand fire hazard in tropical forests.