Machine Learning emulation of high fidelity processes, with application to climate modelling
Project ID: 2228cd1429 (You will need this ID for your application)
Under Offer
Research Theme: Mathematical Sciences
UCL Lead department: Statistical Science
Lead Supervisor: Serge Guillas
Project Summary:
Emulators are statistical surrogates of complex computer models. These must be used to replace components of high resolution global climate simulations, which are computationally intractable due to the integration over long time scales. The project is about fusing Machine Learning techniques such as novel Gaussian Process regression with a climate model, so as to embed high resolution variability of cloud formation (or other processes) into a coarse resolution climate model. The challenges of creating such machine learning approaches are multiple: large dimensions, highly nonlinear behaviour, fast predictions.
Why this research is important This research will lead to more precise risk assessments of future extreme events such as droughts, floods, hurricanes, etc. Solutions and policies require this level of precision.
Who you will be working with This project spans Statistical Science (Serge Guillas, Met Office Chair in Data Sciences), Computer Science (Daniel Giles, Senior Research Fellow in Machine Learning for Climate) and the Met Office (Cyril Morcrette, Head of Clouds and Radiation group, Atmospheric Processes & Parametrizations). You will access supercomputing, as well as training and support, thanks to the strategic UCL Met Office Academic Partnership.
What you will be doing You will focus on creating new machine learning methods and algorithms to accelerate, and improve the quality and efficiency of the emulation with Gaussian Processes, and extensions (e.g. Deep or physics-informed Gaussian Processes). You will also be able to tailor these new methods to the computational set-up of hybrid machine learning and numerical solvers.
Who we are looking for Someone with a background in statistics/machine learning, with an interest in learning how to harness high performance computing, and a curiosity towards climate modelling.