Mike Pritchard
@SciPritchard
Followers
2K
Following
829
Media
13
Statuses
561
Director of Climate Simulation Research for NVIDIA and Professor at UC Irvine. Enthused about Earth system science, HPC, machine learning. Views are my own.
Encinitas, CA
Joined November 2009
Excited to share this emerging team research led by @JaideepPathak on autoregressive AI for km-scale atmospheric prediction. First glimmers of 1-5h convective forecast skill competitive with HRRR alongside plausible generated moist updrafts. Preprint:
š Hurricane help from #NVIDIAResearch: StormCast is a new #generativeAI model for emulating high-fidelity atmospheric dynamics. This model can enable reliable weather prediction at mesoscale which is critical for disaster planning and mitigation. ā”ļø āļø.
1
4
37
Special thanks to Prof. Greg Hakim @UWAtmosSci for lending a nice animation from recent work on dynamical tests of AI weather models, and to Ankur Mahesh & Prof. Bill Collins @UCBerkeley whose new work on huge ensemble AI forecasting is also mentioned.
0
0
2
Thanks to so many wonderful collaborators & colleagues who inform these views & especially from @Nvidia @UCIPhysSci @LeapStc @UCBerkeley @UWAtmosSci @AllenInstitute.
1
0
2
Pleased to share my views in a new @TEDx talk on how AI offers compelling new ways to simulate the Earth with unprecedented resolution and interactivity:
2
2
12
Thanks for shining a light on our latest hybrid simulation results. Iāve been working on this problem for > 7 years starting with @PierreGentine and @raspstephan & these are phase-changing results. @zeyuan_hu absolutely excelled during his @nvidia internship. Check out his talk!.
Zeyuan Hu (Harvard University) and his colleagues have been working on developing machine-learning (ML) subgrid parameterizations for convection and radiation using the ClimSim dataset. By integrating microphysics constraints into the ML emulator, their work achieves stable and.
1
1
16
I'm hiring for 1-2 new FTEs within 1 year of PhD to join our @Nvidia research team. Outstanding ML research & technical ability are a must-have, ideally alongside climate/weather/fluids/physics domain expertise or equivalent experience. Details:
1
23
77
Happy to share this latest work led by @nvidia research colleagues on stochastic downscaling AI modeling demonstrated for weather and climate applications. Escalating adventures in multivariate super resolution & new channel synthesis using generative AI!.
š Superresolution misalignment? Conditional diffusion/flow models falter when low-res and high-res data donāt align. Meet Stochastic Flow Matchingāa framework mapping low-res to latent space, bridging high-res targets effectively. š Paper:
0
3
48
Please see this new work on reformulating diffusion ML models to more naturally handle the heavy tailed statistics that are intrinsic to water cycle state variables in the earth system. Itās delightful to work on atmospheric science with generative AI experts @nvidia research!.
š¢š„ Check out our new work on modeling heavy-tailed data (e.g., extreme weather events) using diffusion models.
0
2
32
RT @MardaniMorteza: šŖļø Can Gaussian-based diffusion models handle heavy-tailed data like extreme scientific events? The answer is NO. Weāveā¦.
0
47
0
RT @PierreGentine: Excited to share our latest paper led by ā¦@XuLian_PKUā© discussing the key role of windthrow and water deficit on the Amā¦.
science.org
Satellite data reveal that Amazon rainforests are susceptive to water deficit in dry seasons and storm damages in wet seasons.
0
17
0
RT @LeapStc: So pleased to share another paper by an amazing interdisciplinary team, including LEAPers @PierreGentine @katiedagon @DJGagneDā¦.
0
10
0
Despite how much has changed in the AI / climate space since this excellent 2022 workshop, many of the challenges reviewed in its summary paper remain relevant research frontiers. Grateful to its leaders for including our viewpoint from work @nvidia @uciess @UCIPhysSci @LeapStc.
Excited to share our new review paper on the use of ML for climate science led by Veronika Eyring and Bill Collins and based on a great workshop in Colorado in 2022.
0
0
14
This paves the way for the broader ML community to work on hybrid climate modeling.
0
0
4
š āClimSim-Online: A Large Multi-scale Dataset and Framework for Hybrid ML-physics Climate Emulationā arXiv:2306.08754, led by @SungdukYu , @zeyuan_hu & Akshay Subramaniam.
1
0
4
2ļøā£ Released ClimSim-Online, an updated ClimSim paper (Neurips 2023 outstanding dataset benchmark) with a containerized, cross-platform workflow for coupling PyTorch models into Fortran-based E3SM for hybrid climate simulations.
1
0
7
š āStable Machine-Learning Parameterization of Subgrid Processes with Real Geography and Full-physics Emulationā
1
1
8
1ļøā£ Achieved stable hybrid climate simulation with favorable multi-year climatology, including complexity of real geography, seasons, explicit cloud condensate & wind prediction, and land coupling. Internship project led by @zeyuan_hu & Akshay Subramaniam.
1
0
5