Numerical Analysis (2020; please now see SCSC)

Dan Simpson


Title: The fine art of putting a bunch of straight lines through things

Ambient air pollution is a global public health problem, but if we want to know how big a problem it is, we need to do some statistical modelling.  The problem is that we do not have good measurements of air pollution everywhere that people live and so we cannot link exposure to adverse health outcomes. What we do have is a monitoring network, satellites, and a great deal of optimism. Now all we need is to turn that into a coherent estimate of air pollution. In this talk I will talk through the challenge of calibrating plentiful but inaccurate satellite measurements to rarer high-quality data. In the course of doing this I will touch on the principles of Bayesian modelling, techniques for approximating Gaussian random fields, and model assessment.