Asking for community input for the new FY plan - what could/should ACCESS-NRI work on that would most benefit the community
Draft will be shared April
Final draft May-June (expect minimal changes)
Please add ideas to form by 16 March (submit a new form for each idea): this online form
Scope of work could be large or small (they have had a submission for idealised modelling)
CSET IntroductionJorge Bornemann
Community Seamless Evaluation Toolkit
Support assessment, evaluation, verification, across space and time scales (nowcasting to weather forecasting to climate) including ensembles, physics-based and ML models
A process-oriented focus for model assessment (compare models against each other and observation)
Modern software, open access, comprehensive documentation, easy to use and contribute to
Design concepts discussed including fields and operators to develop recipes and produce a diagnostic
Designed to make it easy for researchers without strong computing skills to use, as well as for researchers developing and coding new diagnostics to contribute to.
Uses Iris, GitHub, recipes defined in YAML, plots add metadata to JSON file that is used to display on a webpage, cylc workflow, verification uses MET and METplus, VERPy used to produce plots
Walked through some example recipes, such as spatial plots at a specific time, ensemble mean spatial plot
Showed how the output plots are organised on the webpage, and example plots: spatial maps, domain mean timeseries, difference plots, histograms, transects, profiles, convective cell statistics (52 recipes so far added including elevated convection diagnostics, age of air)
Next steps: climate diagnostics, overlay observations with diagnostics
Prerequisites: xp65, a project that hosts a web service, hr22, desirable GitHub account
Stepped through installation, including how to install cylc workflow
Old installation by Scott works and can open the suite (needs to be updated to latest CSET version)
Bureau developing an nci site file, but not required (contact Vinod for nci site file)
Walked through how to use the cylc suite to set up to read in model experiment data, select which fields to visualise, which diagnostics to run and plot
Provided commands to run the cylc workflow and showed the website displaying the outputs
Compute/storage resources:Heidi Nettelbeck
Get your experiment proposals in for Q1/2026. We have 1.5 MSU allocation this quarter. See here for how to apply for Atmosphere WG Resources.
heidi
(Heidi Nettelbeck, ACCESS-NRI Atmosphere Modelling Team Lead)
2
Atmosphere WG meeting minutes 25/03/2026
11 participants on Zoom
Brief update from WG co-chairs and ACCESS-NRI:Yi Huang, Charmaine Franklin, Matt Woodhouse, Heidi Nettelbeck
Charmaine (BoM): reported that the Access-A and -AE ensemble systems are now running in near real-time on the Bureau’s new HPC, with research shifting focus to LFRic and RAL4, including new ancillary file development and testing of an interactive aerosol scheme.
Yi provided an update on computing resources, noting strong usage by university members for cloud-interaction simulations and cloud feedback studies led by various team members.
Matt mentioned that CSIRO is undergoing restructuring, which may impact their activities, including potential reduced capacity for CMIP7 preparations.
Machine Learning/Atmosphere Workshop (Melbourne, Aug 2026):Heidi Nettelbeck
Discussed aim of workshop to focus on how Atmosphere WG could use machine learning to complement and advance existing physical atmospheric models rather than replacing them with emulators. Matt and Yi expressed interest in parameterisation aspects and hybrid physics-machine learning approaches, while Charmaine emphasized the need to bridge differences with the ML working group’s focus. The group agreed their aim should be to develop machine learning techniques that enhance current models through parameterisation, bias reduction, and post-processing of model outputs.
Atmosphere WG co-chair shake-upMatt Woodhouse, Heidi Nettelbeck
Two Atmosphere WG co-chairs (Matt and Charmaine) are ready to hand over the baton as co-chairs after 2.5 years in the position. They announced their intention to step down and open up the positions to the community, with a focus on finding two new co-chairs to work with Yi and Heidi. The role involves approximately two hours of work per month, including monthly planning meetings, and would be beneficial for someone’s CV or career development. Self-nominations are open! The group aims to have new co-chairs in place by their April meeting.
Science talk
Implementation of aerosol-aware ice nucleation particle (INP) parameterisations in the high-resolution regional UM model over the Southern Ocean, Zhangcheng Pei
Zhangcheng from the University of Tasmania presented on investigating cloud ice processes in the UM regression model over the Southern Ocean, highlighting improvements in liquid water path and radiation simulations when using aerosol-microphysics coupling.
Inferring long-term cloud responses to warming from their short-term variability, Qinggang Gao
Qinggang from the University of Melbourne discussed cloud responses to warming from natural variability, examining cloud feedback in the UK Earth System Model and finding significant biases in outgoing radiation over the maritime continent.
Compute/storage resources:Heidi Nettelbeck
Get your experiment proposals in for Q1/2026. We have 1.5 MSU allocation this quarter. See here for how to apply for Atmosphere WG Resources.
Yi Huang, Matt Woodhouse, Heidi Nettelbeck (Charmaine Franklin was an apology)
Yi (21st Century Weather/University of Melbourne):
ModSci team have build a ‘minimum working example’ for the ACCESS rAM3 suite. Hard coded a single domain driven by ECMWF with simplified cylc logic and configuration files. Primary use was to help develop the centre’s regional coupled model, but it could also be used to support tutorials and training materials for NRI.
ModSci team is also engaging NRI and BoM to work on netCDF conversion of AM3/rAM3.
AM3 team has been testing N512 (~25 km) runs which will support AM3-beta release. Testing is underway using N96 ancillaries reconfigured to N512.
Seeking feedback for ENSO experiments using this model
rCM3 work is going well.
Contact Julie Arblaster (lead CI for ModSci team) and Shixue Li for AM3
Matt (CSIRO) :
Still working through the CSIRO restructure which is unsettling for everyone.
Anyone using an NRI-supported atmospheric model at a university, MUST sign the updatedUM partnership licensing agreement. Click on Accessing UKMO licensed models
Let us know if you want to run, or participate in, any training or tutorial session
Show of hands showed the majority are keen to hold our annual WG meeting during the ML workshop in Melb on 21 Aug.
Co-chair shake-up!
Matt and Charmaine are standing down. Welcome to Sonya Fiddes and Andrew Brown in their new roles!
Yi will still co-chair the WG to manage transition to new co-chairs.
Steve Sherwood proposal:
Steve Sherwood wants to run high-resolution simulations to support machine learning into small scale atmospheric interactions. We need high temporal resolution to train a hybrid ML models. We are hoping to run a suite to run and do model training ‘on-the-fly’ on data with very high temporal and spatial resolutions. Can we start talking about this before the August meeting. If you’re interested, have comments, or share experiences on what is currently underway, please share?(Steve took a shapshot of people raising hands showing interest)
Claire Courage and Chermelle Engel - plans for AM3/rAM3
FYI Chermelle is now in land team as she’s working on coupling rAM3 with CABLE
Claire : AM3 plans
Beta release. Single atm and land source code for ACCESS3
Upgrade to cylc8, Using UMv13.8
Use CCI ancillaries (hopefully!)
Support for N96 and N512 resolutions (with help from 21st Century Weather)
Publish documentation
Training videos and possible a training event
Science features in the future
Update ancillaries to CMIP7 dataset to support recent time periods
Support Australian tree PFTs (?) land cover and parameters for these types
Activate biochemistry
Working on biogenic emissions and some depositions
Urban model (hopefully) dependent on extra NRI funding
Extra configurations?
Evaluation
Gathering atmospheric recipes. At the moment we only evaluate atmospheric / water balance
AM3 - ILAMB (land-based evaluation) workflow
Rapid Evaluation Framework
CABLE coupling in CM3 and rAM3
River routing
Evaluation
Optimisation
Some profiling
Full optimisation not in this Financial Year
Claire showed a schematic workflow. The ultimate aim is to make AM3 and rAM3 (and ultimately CM3) a common code base using same ancillary suites
Q : Julie Arblaster re: consistency b/w AM3 and CM3. If you did an AMIP run with AM3, that will be different to atmosphere CMIP?
A : Claire - will depend on the release schedule. When we get to UMv13.8 and when we incorporate biochemistry.
Chermelle : rAM3plans
Science features
UM v13.8
CABLE land surface model
Generalised GRIB paths for BARRA and CMIP driving models
Support for longer-running regional climate runs (including monthly-varying aerosol ancils)
Adopting GLOMAP/CASIM to support chemistry community (stretch goal)
Technical support
Working with BoM/21st Century Weather on netCDF post-processing options
Q : Sonya re: chemistry?
A : Chermelle - the existing UM partnership regional climate suite has its own workflows for dealing with aerosols. NRI have decided not to support another suite, rather will add this functionality (after discussions with UKMO) to the existing (and better supported) RNS suite.
A : Matt - running these configs requires adding aerosol fields to start dumps. Myself and Sonya have a lot of experience with thus
Q: Julie re: urban ancillary funding?
A : Claire. Waiting on NCRIS decision. This will determine what we support.
Q : Paul Leopardi : Longer term CMIP7 campaign depends on CSIRO, not NRI AM planning?
A : Claire : NRI is also waiting for funding announcements.
Ashley Barnes - rCM3 updates
First run of a ‘Frankenmodel’. Created by
Using rAM3 to define a domain, ancils, namelists and ICs and LBCs
Using rOM3 to do the same. Currently a small test domain centred around Lismore
Latest CM3 config. Update UM version from 13.0 → 13.5. Remove ICE and other problems associated with global configurations (e.g. global river runoffs)
Have to activate c_tiles in rAM3 namelists to allow fractional land.
Mash them all together
Currently have two frankenmodels, one for GAL9 (global model) and one for RAL3 (hi-res convection permitting resolutions)
Currently running outside of rose-cylc, which removes some of the obsfucation that rose can perform in generating namelists
We can start thinking about preliminary science runs. But workflow is still fully manual.
Next steps
Some bug fixes and issues with RAL3 precipitation
Set up some science runs and validate
Create an automated workflow
Reconcile rCM3 and CM3 source. We think we can share a common executable for both models in the future
Source code. Need to handle ice and run-off smoothing with if statements. Mainly related to NUOPCY cap. Kieran has already started doing this for OM3 work.
Changing to read mesh from disk, rather than creating the mesh in memory
Using total precip rather than resolved + convective in the coupling mediator
Do we wait for Spack build for CM3 and move to UMv13.8 or do a big push now and have two branches that use different UM versions?
Necessary fixes
Surface stresses are currently zero. May be related to removing ICE
Have to figure out restarts
Should we use rAM3 patches rather than using vanilla UMv13.5?
Ideally - use netCDF outputs
Check land/sea mask behaviour
Turn off river run-of
Early science and validation runs : Andrew Brown - sea breezes, Madi Rosevear - ocean fluxes
Automation pipeline
We can use rAM3 rose/cylc minimum working example for a single-nest run
Keen to build a python notebook which can test and modify ancillaries (e.g. vegetation, forcing and bathymetry) and view them in a single package
Hope to have single-nested science runs completed by late 2026.
Heidi : resource update
We have 2MSU available this quarter. No applications yet! Please apply in the usual way.
