Machine Learning for Climate and Weather Working Group Announce

Hi everyone,

This month’s meeting of the Machine Learning for Climate and Weather Working Group will take place Friday, 3 October at 2pm.

Zoom details: Launch Meeting - Zoom
Zoom Meeting ID: 83187307759 (password: 123456)

Here is the agenda:

Facilitator: Micael Oliveira
Co-chairs: Sanaa Hobeichi, Ryan Holmes, Tennessee Leeuwenburg, Micael Oliveira (ACCESS-NRI liaison), Yue Sun (NCI liaison)

  • Acknowledgement of country
  • Updates from ACCESS-NRI and NCI
    • ACCESS-NRI 2025-26 Workplan for Machine Learning
    • New DL models ready to run at NCI
  • Updates from the Chairs
    • Highlights from the First Year of the Community.
    • ML session at AMOS2026
  • Updates from the community & new community member introductions
  • Presentation by David Fuchs (Details provided below).
  • Discussion and follow-up from NRI community workshop (ongoing):
    • Proposals for interacting/supporting other working groups
    • Plans for using project nm47 (100TB gdata storage, 700 kSU/quarter) on NCI

See you on Friday!

Details of this month’s presentation:

Presenter: David Fuchs | Senior Scientist, DCCEEW
Title: The benefits of lateral connections: towards a stable neural network surrogate for climate model parametrization
Abstract: This study compares four neural network (NN) architectures to serve as surrogates for moist convection in a global atmospheric model. The four architectures share a common backbone but differ in implementation, and are tested offline and online in the CAM atmosphere model. A network architecture that enforces a mesh of short and long pathways proves superior to an increase of depth alone, or replacing ResNet shortcuts with DenseNet. An online hybrid climate model run based on this architecture runs stably for 25 years to completion of an atmospheric-only, and 12 years to completion of a coupled ocean-atmosphere scenario. This shows that the learning process loses online numerical stability while reducing online error. Earlier offline learning epochs mastered cloud liquid water while later epochs reduced the error in cloud ice, potentially due to a naive sampling approach. In the coupled ocean-atmosphere scenario, these translated to steep surface temperature increases when earlier training epochs were used as surrogates, which flattened when later training epochs were used. These results emphasize the need to explore further NN architectures as surrogates to existing parameterizations.

Hi everyone,

This month’s meeting of the Machine Learning for Climate and Weather Working Group will take place this Friday, 7th November, at 2pm (AEDT), and includes a talk from Ajitha Cyriac (CSIRO) - A machine learning approach to downscale sea surface temperature extremes and thermal stress on the Great Barrier Reef

NEW ZOOM DETAILS (for this month’s meeting only): https://anu.zoom.us/j/81754231561?pwd=b7cQMUUdqvTcXeYZ8acztclzavn8Rc.1

Meeting ID: 817 5423 1561, Password: 292009

Agenda:

Facilitator: Ryan Holmes
Co-chairs: Tennessee Leeuwenburg, Sanaa Hobeichi, Vassili Kitsios, Micael Oliveira (ACCESS-NRI liaison), and Yue Sun (NCI liaison)

—-----------------------------------------------------------------------------------------------------

Presenter: Ajitha Cyraic | Postdoctoral Research Fellow at CSIRO
Title: A machine learning approach to downscale sea surface temperature extremes and thermal stress on the Great Barrier Reef
Abstract: In this study, we have successfully developed a machine learning (ML) model based on a super resolution deconvolutional neural network to rapidly downscale SST on the Great Barrier Reef (GBR). When downscaling 80 km data to 10 km resolution, the ML model captures the spatial variability of SST and extreme thermal events significantly better than a conventional interpolation method. We have applied this model to independent datasets from both current and future climates to demonstrate its robustness. We further use this model to downscale thermal stress from five CMIP6 models and analyse coral bleaching risk under different emission scenarios on the GBR.
Biography: Ajitha Cyriac is a Postdoctoral Research Fellow at CSIRO based in Perth. Her research focuses on downscaling climate data using ML methods and analyzing coral bleaching risks at Australia’s Marine World Heritage Areas.

Hi everyone,

I would like to draw your attention the our guidelines for applying for Machine Learning for Climate and Weather WG Resources. This document has been recently revised to include guidelines and instructions to request technical support from ACCESS-NRI.

A project selected for this kind of support will have a Research Software Engineer from the Software Transformation Team assigned to it for a certain period of time. This is a new type of support provided by ACCESS-NRI and is currently only available for the Machine Learning for Climate and Weather WG.

1 Like

Hi everyone,

This month’s meeting of the Machine Learning for Climate and Weather Working Group will take place this Friday, 5th December, at 2pm (AEDT), including a talk from Christian Stassen (Bureau of Meteorology), Machine learning based climate downscaling for the Australia Climate Service

Zoom details: Zoom link
Meeting ID: 831 8730 7759
Password: 123456

Agenda:

Facilitator: Sanaa Hobeichi

Co-chairs: Vassili Kitsios, Tennessee Leeuwenburg, Ryan Holmes, Micael Oliveira (ACCESS-NRI liaison), and Yue Sun (NCI liaison)

  • Acknowledgement of country

  • Updates from ACCESS-NRI

  • Updates from NCI

    • AIFS training benchmark
    • Experimental software environment modules used in Momentum Partnership ML Hackathon, not finalized yet, can be further improved before officially released
      • sda/25.10
      • pet/0.4.0
      • anemoi/25.10
    • Module updates
      • Rapids
  • Updates from the Chairs

    • Feedback on Momentum Partnership ML Hackathon - 10-14 November
    • AMOS 2026 - Session: 10. Advancing weather, climate and ocean research with AI and machine learning
  • Updates from the Community & new community member introductions

  • Presentation by Christian Stassen (BoM) (Details provided below).

  • Discussion (ongoing, dependent on time available in meeting):

—--------------------------------------------------------------------------------------------------------

Details of this month’s presentation:

Presenter: Christian Stassen | Modelling Research Scientist | Atmospheric Modelling, Research, Science and Innovation - BoM

Title: Machine learning based climate downscaling for the Australia Climate Service

Abstract: In the first phase of the Australian Climate Service (ACS), two physical models, BARPA and CCAM, were used to dynamically downscale CMIP6 projections over Australia to regional (11km - 17km) and convection-permitting (4km) resolutions. However, the high computational cost of high-resolution physical simulations makes them impractical for large ensembles or extended historical and future periods. Machine learning offers a faster, more cost-effective alternative by leveraging the existing physical model outputs as training data. State-of-the-art generative machine learning models produce realistic, high-fidelity outputs and these approaches can complement and significantly extend the reach of physical downscaling by enabling more global CMIP models, emission pathways, and ensemble members, to be downscaled at convection-permitting resolutions. This will help reducing uncertainty in regional climate projections. This talk will cover plans for the use of ML methods the next phase of ACS climate downscaling.

Biography: Christian is a research scientist at the Bureau of Meteorology working in the Atmospheric Modelling team. He co-leads the climate-ML activity in ACS Work Package 1 - Foundational Climate Modelling.

Upcoming EGU session on ML…

Abstract submissions are now open for the EGU session, ‘OS1.5 Machine Learning and Ocean Modelling for the Earth System’. This is a session proposed by Thomas Wilder and colleagues, Anna Sommer, Said Ouala, and Adam Blaker. We are interested in the following, but not limited to, physics emulation, development and availability of training data, benchmarking framework, and calibration techniques. I’d be very grateful if you could share this across your network with those who might be interested.

Full session details and how to submit an abstract are found here → https://meetingorganizer.copernicus.org/EGU26/session/55899

Deadline for submission is 15 January 2026, 13:00 CET

Information about financial support can be found here → EGU26 - Financial support and waivers

Hello,

ACCESS-NRI is going to have significantly more compute resources on Gadi for the first half of 2026. As a consequence, all working group projects will receive an allocation of 1.5 MSU per quarter for that period. We will still look at re-allocations during the quarter to optimise usage.

If you have any experiment idea that fits within the guidelines, January is the time to go ahead and use the resources.

Self-nominations for the ACCESS-NRI Scientific Advisory Committee (SAC)

The ACCESS-NRI Scientific Advisory Committee (SAC) is changing the way it operates, and we are now inviting self-nominations for new members. We encourage applications from across the ACCESS Community. We are seeking a diverse committee that includes representatives from each of the ACCESS Community Working Groups (CWGs), as well as members from different career stages and areas of research, technical or infrastructure expertise.

To apply and for more information go to: Self-nominations for the ACCESS-NRI Scientific Advisory Committee (SAC)

We are also seeking your recommendations for potential international members to join our SAC. Please send their details (name, organisation and email and a brief rationale for why you’re recommending them) to access.nri@anu.edu.au

Hi,

This Friday we will resume the Machine-Learning Working Group monthly meetings. Meeting will be at the usual time, 2pm (AEDT), and we will have a talk by Skye Williams-Kelly (UNSW), Interpolating daily in situ precipitation observations with Bayesian Neural Fields.

Zoom details: Zoom link
Meeting ID: 831 8730 7759
Password: 123456

Agenda

Facilitator: Yue Sun (NCI liaison)

Co-chairs: Tennessee Leeuwenburg, Sanaa Hobeichi, Vassili Kitsios, Ryan Holmes, and Micael Oliveira (ACCESS-NRI liaison)

  • Acknowledgement of country
  • Updates from ACCESS-NRI
  • Updates from NCI
    • FCN3 inference notebook available
    • HPC for large scale weather prediction and climate modeling hosted by CCRS, RIKEN R-CCS, and NIES Japan. updates from SCA26
  • Updates from the Chairs
    • AMOS 2026 - Session: 10. Advancing weather, climate and ocean research with AI and machine learning - 17th February 2026, after the afternoon tea.
  • Updates from the Community & new community member introductions
  • Presentation by Skye Williams-Kelly (UNSW) (Details provided below).
  • Discussion (ongoing, dependent on time available in meeting):

Details of this month’s presentation:

Presenter: Skye Williams-Kelly | PhD Candidate | UNSW Sydney - ARC CoE for the Weather of the 21st Century

Title: Interpolating daily in situ precipitation observations with Bayesian Neural Fields

Abstract: Accurate precipitation predictions are vital for water resource management and risk mitigation. Interpolated precipitation estimates derived from in situ observations are frequently used to evaluate climate models and analyse trends. However, these inadequately represent its spatio-temporal characteristics and significantly smooth out extremes, inhibiting effective evaluation of dynamical models and analysis of trends. Machine learning methods may be suited to addressing these limitations due to their ability to identify patterns in large datasets and use of GPU acceleration. Here I discuss results for one such method, the Bayesian neural field, which was inspired by the frequently used Gaussian Process and kriging methods. Its performance is evaluated using traditional and distributional metrics, including on out-of-sample prediction. Results are further compared against existing precipitation products to identify the relative strengths and limitations of each method.

Biography: Skye is a 2nd year PhD candidate at the Climate Change Research Centre, UNSW Sydney. Following a bachelor’s degree in data science, including an Honours project on multivariate Bayesian fitting methods, she briefly worked in data engineering before returning to complete her PhD. Her current research is looking at machine learning and Bayesian methods for interpolating precipitation, with a focus on better characterising extremes.

Hi,

This Friday we will resume the Machine-Learning Working Group monthly meetings. Meeting will be at the usual time, 2pm (AEDT), and we will have a talk by Vassili Kitsios (CSIRO), Data-driven approaches to subgrid turbulence parameterisation and climate emulation.

Zoom details: Zoom link
Meeting ID: 831 8730 7759
Password: 123456

Agenda

Facilitator: Yue Sun (NCI liaison)

Co-chairs: Tennessee Leeuwenburg, Sanaa Hobeichi, Vassili Kitsios, Ryan Holmes, and Micael Oliveira (ACCESS-NRI liaison)

  • Acknowledgement of country
  • Updates from ACCESS-NRI
    • Community input invited: ACCESS-NRI Annual Work Plan (FY2026-27)
  • Updates from NCI
    • Aurora model update
    • Performance Improvement Opportunities
  • Updates from the Chairs
    • Tennessee - short discussion on experiment management
  • Updates from the Community & new community member introductions
  • Presentation by Vassili Kitsios (CSIRO) (Details provided below).
  • Discussion (ongoing, dependent on time available in meeting):
    • Plans for using project nm47 (100TB gdata storage, 875kSU/quarter) on NCI.
    • Experiment management - if not mlflow,what? And who needs it?

Details of this month’s presentation:

Presenter: Vassili Kitsios | Senior Research Scientist | CSIRO

Title: Data-driven approaches to subgrid turbulence parameterisation and climate emulation

Abstract: There is increasing interest in the application of machine learning methods to emulate the entire climate, and also to parameterising the contribution of the unresolved subgrid-scales within general circulation models. These two separate but related problems will be presented in terms of: 1) the mathematical formulation; 2) the data requirements; and 3) our previous solutions using statistical and machine learning. The intention of this seminar is to educate the community on the importance and typical approaches to climate emulation and subgrid parameterisation, and inform discussions on the potential future initiatives of our working group.

Biography: Dr Kitsios completed a PhD with the University of Melbourne and the Université de Poitiers (France) on fluid dynamical stability and model reduction of aerospace flows. He then undertook post-doctoral research with the CSIRO and then Monash University, on the numerical simulation and stochastic subgrid-scale parameterisation of atmospheric, oceanic and boundary layer turbulence. He then held an industrial research position at a hedge fund developing trading algorithms based on macroeconomic themes and market conditions. Since re-joining CSIRO, he has been undertaking research on data assimilation methods for improved climate state / parameter estimation and forecasting. His most recent research involves the application of machine learning for climate emulation, and quantifying the influence of climate on agriculture, financial markets, health indicators and social unrest. He is currently a co-chair of the Machine Learning for Climate and Weather Working Group of the Australian Climate Community Earth System Simulator National Research Infrastructure consortium, an associate editor for the Theoretical and Computational Fluid Dynamics journal, and an elected committee member of the Australasian Fluid Mechanics Society.

Hi,

A reminder that the first period for community input on the ACCESS‑NRI Work Plan for FY26–27 is now open and will close on Monday, 16 March . If you haven’t already, please take a moment to share your feedback, through this form—your insights are essential to shaping the upcoming work plan.

We welcome all input, big or small, new ideas or projects already discussed, ideas on model development, usability, release, accessibility, evaluation, tools, etc.

You can find more information on the ACCESS-Hive forum.

Claire

Applications now open for the ACCESS-NRI PhD Internship Program!

This 3-month internship provides an opportunity to work alongside ACCESS-NRI staff on activities related to the development, testing, and release of climate models as well as the supporting software and data that underpin them.

This is the first of two rounds this year and is open to PhD students from the Australian National University, UNSW, and the University of Tasmania. We expect to open a second internship round later in 2026 that will also include Monash University and the University of Melbourne.

Application deadline: 10 April 2026

Full information and details for this round can be found on the ACCESS-NRI internship webpage.