Preprints
https://doi.org/10.5194/egusphere-2022-912
https://doi.org/10.5194/egusphere-2022-912
 
04 Oct 2022
04 Oct 2022
Status: this preprint is open for discussion.

Using Probabilistic Machine Learning to Better Model Temporal Patterns in Parameterizations: a case study with the Lorenz 96 model

Raghul Parthipan1,2, Hannah M. Christensen3, J. Scott Hosking2,4, and Damon J. Wischik1 Raghul Parthipan et al.
  • 1Department of Computer Science, University of Cambridge, Cambridge, UK
  • 2British Antarctic Survey, Cambridge, UK
  • 3Department of Physics, University of Oxford, Oxford, UK
  • 4The Alan Turing Institute, London, UK

Abstract. The modelling of small-scale processes is a major source of error in climate models, hindering the accuracy of low-cost models which must approximate such processes through parameterization. Red noise is essential to many operational parameterization schemes, helping model temporal correlations. We show how to build on the successes of red noise by combining the known benefits of stochasticity with machine learning. This is done using a physically-informed recurrent neural network within a probabilistic framework. Our model is competitive and often superior to both a bespoke baseline and an existing probabilistic machine learning approach (GAN) when applied to the Lorenz 96 atmospheric simulation. This is due to its superior ability to model temporal patterns compared to standard first-order autoregressive schemes. It also generalises to unseen scenarios. We evaluate across a number of metrics from the literature, and also discuss the benefits of using the probabilistic metric of hold-out likelihood.

Raghul Parthipan et al.

Status: open (until 29 Nov 2022)

Comment types: AC – author | RC – referee | CC – community | EC – editor | CEC – chief editor | : Report abuse

Raghul Parthipan et al.

Model code and software

Model code Raghul Parthipan https://zenodo.org/record/7118668

Raghul Parthipan et al.

Viewed

Total article views: 309 (including HTML, PDF, and XML)
HTML PDF XML Total BibTeX EndNote
246 55 8 309 1 2
  • HTML: 246
  • PDF: 55
  • XML: 8
  • Total: 309
  • BibTeX: 1
  • EndNote: 2
Views and downloads (calculated since 04 Oct 2022)
Cumulative views and downloads (calculated since 04 Oct 2022)

Viewed (geographical distribution)

Total article views: 306 (including HTML, PDF, and XML) Thereof 306 with geography defined and 0 with unknown origin.
Country # Views %
  • 1
1
 
 
 
 
Latest update: 29 Nov 2022
Download
Short summary
How can we create better climate models? We tackle this by proposing a data-driven successor to the existing approach for capturing key temporal trends in climate models. We combine probability, allowing us to represent uncertainty, with machine learning, a technique to learn relationships from data which are undiscoverable to humans. Our model is often superior to existing baselines when tested in a simple atmospheric simulation.