Statistics and Data Science Seminar: Dr Jonathan Owen (University of Sheffield)

Title: Bayesian Uncertainty Analysis of Prior Beliefs in History Matching Climate Models
Speaker: Dr Jonathan Owen (University of Sheffield)
Location: Harrison 170

Abstract: Computer models are vital in climate science to study drivers of climate change;
quantify uncertainties in future climate predictions; and to guide policy decisions.
However, their direct use is inhibited by: their complex structure; high-dimensional
input and output, including spatial-temporal fields; the numerous sources of
uncertainty present in linking models to the real-world; further compounded by their
long evaluation times, often taking weeks to months on High-Performance
Computers (HPCs).
Earth System Models (ESMs) are highly complex models integrating atmosphere,
ocean, land, ice, and biosphere. An important use of ESMs is to investigate natural
and anthropogenic aerosol emission interactions with clouds yielding large Aerosol
Radiative Forcing (ARF; the temporal change in Earth’s energy balance due to
aerosols) induced uncertainty in historical climate change. ARF is unobservable and
key to predicting future climate, yet research has resulted in little uncertainty
reduction in 30-years of IPCC reports.
Model-observation comparison is performed to calibrate ESMs enabling tasks such
as the robust constraint of ARF uncertainty. We employ history matching,
incorporating Bayesian emulators as fast statistical approximations, to perform an
efficient global parameter search. These are embedded within an uncertainty
quantification framework which includes structural model discrepancy linking ESM
output and the real-world, as well as representation and observation errors, to
meaningfully compare observations with the model. In this research we also address
challenges pertaining to subjective choices. These include: methods of carefully
selecting a set of observable model outputs over which to history match; an
interrogation of the prior belief specification; and analysing the implicit prior beliefs of
past studies. This methodology is applied to the UK Met Office UKESM1 model to
resolve parametric uncertainty as well as to identify sources of structural deficiencies
to aid further ESM development.