Generative machine learning emulation of km-scale regional climate simulations

Abstract

High-resolution climate simulations are very valuable for understanding climate change impacts and planning adaptation measures. This has motivated use of convection-permitting models (CPMs). These are regional climate models run at sufficiently fine resolution to capture important small-scale atmospheric processes, such as convective storms, rather than relying on parameterizations. However, these CPMs have very high computational costs, limiting their applicability.

This thesis presents the development and evaluation of CPMGEM, a novel application of a diffusion model (a form of deep generative machine learning), to skilfully emulate CPM simulations from the Met Office's United Kingdom Climate Projections Local product over England and Wales between 1981 and 2080 at much lower cost. This emulator enables stochastic generation of high-resolution (8.8km), daily-mean samples conditioned on coarse-resolution (60km) weather states from a global climate model. The emulator is first designed to predict the precipitation output of the CPM, based on requirements gathered from interviews with potential users. The output is fine enough for use in applications such as flood inundation modelling. The emulator is stochastic, which allows it to produce precipitation samples with more realistic small-scale structure than a similar deterministic model and to be better skilled in reproducing extreme events. The emulator captures most of the 21st century climate change signal. An extension of the emulator to output more variables (near-surface relative humidity and air temperature) is demonstrated. Potential applications include producing high-resolution predictions for large-ensemble climate simulations and downscaling different climate models and climate change scenarios to better sample uncertainty in climate changes at local-scale.

Date of Award	13 May 2025
Original language	English
Awarding Institution	University of Bristol
Supervisor	Peter A G Watson (Supervisor) & Laurence Aitchison (Supervisor)