Pinned Tweet
🌎 We developed an ML-based atmospheric model suitable for climate prediction! Long-term stable, capable of assessing mass/moisture conservation, and predicts diagnostics for (eventually) coupling to ocean.
Very excited about this project! 🎉
We're excited to present ACE (AI2 Climate Emulator), a long-term (at least 10 yr) stable ML emulator of a comprehensive 100-km resolution global atmospheric model: arxiv.org/abs/2310.02074
Oliver Watt-Meyer retweeted
🌍☀️❄️ Can AI forecast year-to-year differences in the seasons?
New research with @metofficeUK shows our ACE2 ML model demonstrates seasonal forecasting skill—matching traditional physics-based methods while using dramatically less compute. 🧵
Oliver Watt-Meyer retweeted
We partnered with researchers @UCSanDiego and @ucsd_cse to supercharge climate modeling, speeding up predictions and enhancing accuracy.🌍 Check out the blog post below and come find us at #NeurIPS this week!
Researchers @UCSanDiego @ucsd_cse & @allen_ai combined #generativeAI and physics-based data to improve #climate models: today.ucsd.edu/story/acceler…
@yuqirose
Oliver Watt-Meyer retweeted
We are now seeking applicants for a full-time research scientist position on the climate modeling team at Ai2! Come work with an excellent team at the frontier of using AI for climate modeling. 🌎🌦
Hybrid in Seattle, app review begins Oct 15.
Details 👉job-boards.greenhouse.io/the…
Oliver Watt-Meyer retweeted
Excited to share our new review paper on the use of ML for climate science rdcu.be/dRMdh led by Veronika Eyring and Bill Collins and based on a great workshop in Colorado in 2022
Oliver Watt-Meyer retweeted
We are delighted to announce the #ICML2024 ML4ESM Best Paper Award winner:
[2406.14798] Probabilistic Emulation of a Global Climate Model with Spherical DYffusion (arxiv.org)
Congrats to @salvaRC7 and co-authors @ai2_climate @yuqirose for their amazing work!
Oliver Watt-Meyer retweeted
Hope everyone is enjoying #ICML2024! Don't forget to stop by our workshop on Machine Learning for Earth System Modeling tomorrow, Friday 26th in Stolz 1! Schedule here: icml.cc/virtual/2024/worksho…
Great work on probabilistic emulation of an atmospheric model from @salvaRC7 who interned on our @ai2_climate team!
🌍We've developed the first conditional generative model for accurate & efficient ensemble climate simulations!
📄Read the full paper: arxiv.org/abs/2406.14798
🎉Excited to share that a preliminary version was accepted as an oral presentation @ml4esmworkshop at #ICML2024!
🧵
Oliver Watt-Meyer retweeted
Don't miss this expert panel, incl. Dave Lawrence, LEAP's @NCAR_Science Model Dev. Liaison, discussing the responsible integration of #AI + #climate modeling.
📅 THIS THURS, 6/27
🕜 12:30-2p EDT
💻 Online: bit.ly/3RHUTW1
@NOAAClimate @CUSEAS @columbiaclimate @NSF
Oliver Watt-Meyer retweeted
How can AI help us prepare for a changing climate? That's one of the big questions our climate modeling senior director Chris Bretherton wants to answer. Chris spoke with @nvidia's @noahkravitz about how technology like localized models can help us prepare for the future — listen here: blogs.nvidia.com/ai-podcast/
Oliver Watt-Meyer retweeted
We have a new ACE preprint led by our former intern, James Duncan, where we’ve successfully emulated a second major atmospheric model, DOE’s EAMv2! 🌎💻
doi.org/10.22541/au.17086417…
Oliver Watt-Meyer retweeted
Was interesting to see how well a simple ML approach could learn skillful climate model cloud patterns that could improve coarse-grid model climate
See another newly published article, led by @brianmhenn1, where we machine-learn cloud properties (from a 3 km reference simulation) needed for unbiased radiative fluxes in coarse-resolution FV3GFS (200 km). ⛅
doi.org/10.1029/2023MS003949
Recently out in JAMES!
A new published article, led by @oliwm, investigates direct learning of heating/moistening rates from a coarsened 3 km (realistic topography!) storm-resolving simulation for use in a 200 km global climate model. 🌡🌧🌐
agupubs.onlinelibrary.wiley.…
5
296
Oliver Watt-Meyer retweeted
For those of you at @ametsoc's #AMS2024 this week, there will be two @ai2_climate in-person presentations on Tuesday. Andre Perkins (@frodre) is presenting on ML emulation of climate model microphysics ams.confex.com/ams/104ANNUAL…
Oliver Watt-Meyer retweeted
Did you like the IA NWP models? You'll love the AI climate models!
The authors present a climate model based on SFNO block, and with an impressive numerical stability of 10 years (!) at least.
arxiv.org/abs/2310.02074
Oliver Watt-Meyer retweeted
It was a breakthrough year for environmental AI, with many new foundation models and innovative solutions for tackling our biggest environmental challenges. Learn more about 2023's major advances and key themes in climate and sustainability research: blog.allenai.org/ai-for-eart…
Updated version of AI2 Climate Emulator (ACE) paper with:
-> 100 year run ⌚️
-> testing on unseen realistic SST dataset 🌍
-> code+data+checkpoint released github.com/ai2cm/ace
arxiv.org/abs/2310.02074
Replying to @ai2_climate
The updated ACE arxiv paper is available, now with code release arxiv.org/abs/2310.02074 . We're excited to get to share these results with the community!
6/
I'll be at @theAGU fall meeting presenting this next week—look forward to chatting with others there. Lots to catch up on.
1
181
Tagging team @gideoknite @jeremy_mcgibbon @spencerkclark @BrianMHenn1 @NoahBrenowitz @SciPritchard @mattthemathman and James Duncan, Karthik Kashinath, Boris Bonev, Chris Bretherton.
415
Oliver Watt-Meyer retweeted
Improved weather forecasting is known as the "quiet revolution" by meteorologists, & for good reason. AI is helping generate forecasts faster and could take models beyond weather prediction. @ScienceMagazine spoke w/AI2's Chris Bretherton & other experts: science.org/content/article/…
Some details: used Spherical Neural Operator architecture from @nvidia (arxiv.org/abs/2306.03838). Dataset is 100 years of output from a 1° atmospheric model. ML is trained to predict 6 hours into the future, then used autoregressively for 10+ year long inference.
1
195
See our paper here with more details on ML setup, physical conservation, climate fidelity and more:
arxiv.org/abs/2310.02074
1
2
208














