Abstract

Abstract: Ensembles of simulations from multiple climate models (‘simulators’) underpin much of our understanding of the climate system, and in particular the potential evolution of future climate in response to different scenarios of socioeconomic development and the associated greenhouse gas emissions. No simulator is perfect, however; and ensemble outputs contain structured variation reflecting simulator inter-relationships, as well as shared discrepancies between the simulators and the real climate system. This structure must be accounted for when using ensembles to learn about aspects of the real climate, especially when defensible assessments of uncertainty are needed to support decision-making. This talk will discuss the issues involved, and describe a statistical framework for addressing the problem. A theoretical analysis leads to a mathematical result with major implications for the design and analysis of multimodel ensembles; whilst the practical application of the framework will be demonstrated using future climate projections for the United Kingdom from two contrasting ensembles (UKCP18 and EuroCORDEX). These ensembles have different structures and properties: the approach is shown to reconcile the substantial differences between the original ensemble outputs, in terms of both the real-world climate of the future and the associated uncertainties.

Biography: Richard is a Professor in the Department of Statistical Science at University College London, where he has worked since completing his PhD at UMIST in 1994. He has extensive experience of developing and applying statistical methods for the environmental sciences. Particular interests include the analysis of time series and space-time data, with application areas including hydrology and the impacts of climate change. Other areas of interest include the assessment of uncertainty when interpreting model outputs; the use of mis-specified models; and the use of nonprobability samples to draw population inferences in ecology.