TM10 | Applying Machine Learning to better reconstruct and understand paleoclimates
Wed, 19:00
Applying Machine Learning to better reconstruct and understand paleoclimates
Convener: Dan Lunt | Co-conveners: Junxuan Fan, Xin Ren, Tianyi Chu
Wed, 30 Apr, 19:00–20:00 (CEST)
 
Room M2
Wed, 19:00
Although Machine Learning (ML) is becoming relatively common in many areas of climate science, it is still relatively under-used in paleoclimate studies. This is despite the fact that ML offers exciting novel techniques for addressing long-standing problems in the field, through improving computational efficiency and accuracy, and enabling the management of large, complex datasets. Whereas traditional methods often rely on a limited set of statistical techniques that may not fully capture the complexity inherent in the data, ML provides a suite of algorithms capable of handling non-linear relationships and high-dimensional data, allowing for a more sophisticated analysis. When integrated with proxy data and output from Earth system models (ESMs), ML provides innovative approaches to efficiently reconstruct, interpret, and analyze paleoclimate conditions in ways that were previously unattainable. There are also applications of ML to improve proxy calibrations, and to improve ESMs themselves, for example, through developing more accurate model parameterisations. Furthermore, ML techniques can be employed to develop surrogate models that approximate the behaviour of more complex ESMs. These surrogates can be used to efficiently conduct sensitivity analyses or long simulations.

We aim to bring together those from the modelling and proxy communities, along with experts in machine learning, to showcase existing work applying ML to paleoclimate studies, and to discuss opportunities for future work. Through this meeting, we hope to foster collaborations and discussions that leverage ML for new insights into paleoclimates. The Town Hall meeting will cover, but not be limited to, the following topics:

Modelling
*) Emulators as surrogate models for ESMs to enable efficient paleoclimate simulations;
*) Efficient model tuning to enhance the performance in simulating paleoclimate;
*) New model parameterisations developed through ML;

Proxies
*) Improvements in management and stratigraphic calibration of large proxy dataset;
*) Advances in proxy calibration and quantifying uncertainties;

Model + Proxy Integration
*) Data assimilation and field reconstruction;
*) Proxy system modelling;
*) Downscaling model results for model-data comparisons

The session will be led and facilitated by scientists from the University of Bristol, UK, and Nanjing University, China.
The oral presentations are given in a hybrid format supported by a Zoom meeting featuring on-site and virtual presentations. The button to access the Zoom meeting appears just before the time block starts.