[ES_JOBS_NET] Graduate Teaching Assistant PhD Studentship, Harnessing Causally-Informed Deep Learning for Use in Climate Modelling, University of Reading, UK

[Todd Jones]

Harnessing Causally-Informed Deep Learning for Use in Climate Modelling
https://jobs.reading.ac.uk/Job/JobDetail?JobId=13080
Supervisor: Dr Todd Jones, t.r.jones at reading.ac.uk

4-Year PhD: 25% time as Teaching Assistant, Department of Computer Science, University of Reading, UK
Salary: £6,994-8,245 per annum + maintenance stipend £14,428 per annum
Requires: degree in Computer Science, Mathematical Science, Meteorology, Physics or closely related subjects
Closing date: 9 April 2024
Interviews expected the week commencing 22nd April 2024.

This project aims to develop a novel deep learning framework that leverages the power of causality to emulate complex atmospheric processes. This involves:
- Generating and utilizing high-fidelity training data from Large Eddy Simulation (LES) models.
- Enhancing model performance and computational efficiency, with a focus on enhancing GPU functionality in an HPC environment.
- Applying your deep learning model within a superparameterization framework to improve largescale climate projections.
By integrating causally-informed deep learning with LES model training data, we seek to emulate high-resolution atmospheric processes at a fraction of the computational cost. This approach not only enhances model performance via the potential to generate accurate high-resolution simulation data quickly, but it also offers a scalable solution to improve reliable climate projections, particularly in models that superparameterize subgrid with LES. That is, climate models must represent small scale phenomena which their coarse spatial grids cannot explicitly resolve with necessary approximations. Historically, this is done via formulaic mathematical functions as parameterizations. Some newer models instead represent these processes by embedding small area high-resolution LES models within the coarse grid to gain physical accuracy at exceptional cost.  By emulating the LES, we intend to capture the physical gains of high-resolution models at a fraction of the computational cost, a potentially pioneering step towards more accurate and computationally feasible climate modelling.


_________________________________
Dr Todd R Jones, BSc MSc PhD AFHEA
Lecturer in Computer Science
University of Reading
Polly Vacher Building, Room 149
+44 (0)118 378 3187
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.ucar.edu/pipermail/es_jobs_net/attachments/20240314/ba1c46df/attachment.htm>