Speed Dating Topics
- Peter Caldwell
- Bryce Harrop
- Mark Taylor
Thursday, 2/27/2025
8:40 - 10:30 Breakouts:
The first 4 rounds of speed dates are 20 min long with 10 min transition time
Room 1: Infrastructure meets with (in order):
8:40 - 9:00 Water Cycle + HES:
GCAM coupling next steps.
LE data management
9:10 - 9:30 EAMxx + Omega:
What helpful tools/tests/etc has each group created or planned that would be helpful to others?
Can you come up with project-wide C++ coding standards?
OMEGA and cpl7?
comparison of CPU vectorization approaches between our implementations
9:40 - 10:00 Land + v3 Atm:
Making more diagnostics possible: hub-height winds.
How do we move to an MPAS-land grid?
When is a PR for high-res land going to happen?
Getting land diagnostics as routine output for coupled runs.
10:10 - 10:30 Polar + Ice:
What are the near-term coupling needs?
ATM grid imprinting on ELM surface mass balance - possibility of higher-order conservative remapping for precip/snowfall?
To what extent are diagnostics the domain of component versus infrastructure. Can we collaborate:
Addition of support for LIVVkit to standard diagnostics / analysis when running w/ dyn ice sheets
Development of diagnostics across groups E3SM-diagnostics/MPAS-Analysis/zppy
Ice sheet coupler support
Albany support for MALI
Room 2: Performance meets with
8:40 - 9:00 EAMxx + Omega
What can we do to eek out more performance from our models?
Avoiding resource contention if any when sharing nodes
…
9:10 - 9:30 Land + v3 Atm
Land: GPU performance, development roadmap
Land: I/O and new partitioning approaches for km-scale simulation
v3 atm HR AMIP(?), other sims performance
…
9:40 - 10:00 Polar + Ice
Sea ice coupling requirements - within 30˚ of the equator
Wavewatch deprecation plans
Greenland ice sheet performance - GPU utilization questions for V4, Frontier usage.
10:10 - 10:30 Water Cycle + HES
HR performance, PE layouts, target machines
…
Room 3: Coupled
8:40 - 9:00 Land + v3 Atm
Transition of land to the MPAS grid, and a unified land+MOSART+Ocean grid
Getting standard land diagnostics for future coupled runs
Lake model updates from Hongyi Li to address negative runoff
Land surface T biases - discussion with polar group
…
9:10 - 9:30 Polar + Ice
From ice:
Data S. Ocean freshwater fluxes from sub-ice shelf melt (more physically realistic implementation of “snowcapping”):
this capability has been thoroughly tested and now appears ready for (regular?) use
is there support from coupled team for exploring the impacts of this in, e.g. a comparative PI control run?
if so, discuss / flesh out details a bit more
Status of new snowpack model testing in 100 PI control run
New capabilities in Greenland ice sheet and sea ice for V4
From Polar:
BlueTip simulation campaign
Significant polar biases: ice sheet climate
Do we have a place for a list/branch of incremental coupling improvements for V4? e.g.
Roughness length over ice sheets by 2-4 orders of magnitude reduction
PMSL extrapolation to use local not global lapse rate
9:40 - 10:00 Water Cycle + HES
How are BGC early results looking, any major feedbacks?
Next steps with v3.1 and v4 coupled runs with BGC/HES
10:10 - 10:30 EAMxx + Omega
What is the roadmap for coupled runs with EAMxx and Omega? Can we come up with tasks/expectations for each group in this work?
Room 4: AI/ML - @Peter Caldwell thinks this group (consisting solely of @Andy Salinger ) should just take notes on all the AI applications being done within E3SM and discuss how they fit into E3SM’s AI strategic plan.
8:40 - 9:00 Polar + Ice
Model emulation: How can we contribute to an E3SM coupled emulator?
Process expansion:
Wave emulator with sea ice feedbacks (Erin)
Emulation of high-fidelity calving model in MALI, POC: Trevor Hillebrand (LANL), Funding: LDRD (LANL)
Parameter refinement:
Vertical sea ice deformation
shelf melt flux refinement
Autotuning of MALI solver settings for performance, POC: Jerry Watkins (SNL), Funding: SciDAC
Uncertainty quantification
Emulation of MALI velocity solver, POC: Mauro Perego (SNL), Funding: SciDAC/MMICCS
Emulation of regional sea-level model for sea-level UQ, POC: Matt Hoffman (LANL), Funding: ECR
Computational infrastructure: AI for converting code between languages, e.g. Icepaxx
Signal detection: Pattern detection, low-order behavior, causality for tipping points.
9:10 - 9:30 Water Cycle + HES
AI/ML to help with BGC spin-up or auto-tuning??
How to incorporate more coupling?
How to leverage emulators for future ensembles?
Will emulators work at both ne1024 and ne128?
9:40 - 10:00 EAMxx + Omega
emulation of aerosol mixing state and cloud droplet collision processes (POC Jiwen Fan; funding LDRD)
10:10 - 10:30 Land + v3 Atm
AI/ML help with Land spin-up.
AI/ML calibration of land parameters. Multiple sites.
10:30 - 10:40 Break
10:40 - 11:10 Breakouts:
Then for the final 30 min speed dating session, the 4 core teams mix it up:
Room 1: Infrastructure + Coupled + Performance + AI/ML
What do we want in a model experiment tracker?
Any progress on identifying AI/ML runtime best practice?
Data formats for AI/ML ?
Room 2: Water Cycle+HES+Polar+Ice
From Polar + Ice:
Sharing between groups analysis of BlueTip, Large Ensemble.
Frazil development and stratification
Update on progress towards replacing snowcapping in the water cycle with active ice sheets
…
Room 3: EAMxx+Omega+Land+v3 Atm
How do we integrate high-res land with EAMxx and Omega?
What are Land’s and River’s plans in terms of their own C++ port?
How do v3 Atm members get integrated into C++ tasks?
What advice (say, todo and not todo) would the EAMxx and Omega teams have for the land group as it undertakes the C++ port?