IMAGe Brown Bag - Multivariate spatio-temporal modelling for geophysical inversions

10/26/2016 - 12:00pm to 1:00pm
ML Chapman Room

Andrew Zammit Mangion
University of Wollongong, Australia

Multivariate spatio-temporal systems are ubiquitous in the geophysical sciences. When analysing such systems, all covariances between all possible combinations of variables at any sets of locations and time points need to be modelled. These covariances need to be built with care, since any covariance matrix that is derived from such a model has to be nonnegative-definite. In this talk I show how a class of valid, flexible models can be constructed using a conditional approach. More generally I show how the approach can be used to construct multivariate models defined by networks of spatio-temporal variables. As case study I consider the problem of atmospheric trace-gas inversion, where the aim is to infer the unobserved sources and sinks of a gas (first variable) from noisy measurements of gas mole-fraction (second variable) from ground stations or remote sensors. These two spatio-temporal variables are related through meteorology and chemistry, simulated using deterministic numerical models. The talk concludes with an outline of other environmental applications that will benefit from this modelling approach. This is joint work with Noel Cressie from the University of Wollongong and Anita Ganesan from the University of Bristol.

Wednesday, October 26, 2016
12:00pm – 1:00pm
Mesa Lab, Chapman Room
(Bring your lunch)