Extreme events and statistics in an AI emulator of a global storm-resolving model

Abstract

We present HiRO-ACEv2 (High Resolution Output Ai2 Climate Emulator), a two-stage framework for AI emulation of the 3~km resolution X-SHiELD global storm-resolving model (GSRM), and highlight its ability to generate large ensembles of data for studying the characteristics of tropical cyclones and extreme events in high-resolution simulations at an inference computational cost over 300 times lower than that of traditional numerical GSRMs. HiRO-ACEv2 expands upon the precipitation-only HiRO-ACEv1 model to generate coherent precipitation, 10~m winds, and sea level pressure outputs, enabling a broader range of events to be simulated at high-resolution. Fine-resolution predictions are generated by running an autoregressive emulator of coarsened X-SHiELD at 100 km, whose multivariate outputs are regionally downscaled to 3~km resolution via a separately trained diffusion model. Particular importance is placed on capturing the statistics of extreme events during the diffusion downscaling step. Our ability to match the tails of distributions and generate realistic-looking extreme events in both the `perfect prediction' case as well as in free-running emulator rollouts is improved via the use of multiple denoisers at inference time and several updates to the data sampling, noise sampling, and loss weighting during training. HiRO-ACEv2 closely matches the high-resolution reference distributions of precipitation and near-surface winds out to their 99.9999th percentile values, and broadly tracks the reference distribution of tropical cyclone intensities.

HClimRep-Ocean: A Global Ocean Emulator on an Unstructured Mesh
Kacper Nowak, Aleksei Koldunov, Nikolay Koldunov +7
arXiv28 Sep 2026PreprintarXiv licenceFull textAtmospheric science & meteorologyarXiv
Spatializing daytime street-level thermal exposure frequency profiles in urban areas during extreme heat events using the operative temperature index
Thomas Lagelouze, Sandra Rome, Julia Hidalgo +3
Urban Climate30 Oct 2026Peer-reviewedCC BY 4.0Full textClimate science Environmental science & pollutionHAL
Building adaptive capacity of US Gulf Coast city practitioners to extreme weather events.
Gabrielle Wong-Parodi, Natalie Herbert, Sarah J Fendrich +14
PNAS nexus28 Sep 2026Peer-reviewedCC BYFull textClimate scienceEurope PMC
RHITA: a web tool for real-time detection of extreme weather events
Greta Cazzaniga, Anastasia Akakpo-Numado, Patrick Brockmann +3
Natural Hazards and Earth System Sciences2 Oct 2026Peer-reviewedCC BY 4.0Full textEnvironmental science & pollutionHALCrossref
Impact of extreme hydrological events on soil microbiome dynamics and ecosystem resilience
Vikas, Rajiv Ranjan
Discover Soil25 Oct 2026Peer-reviewedCC BYFull textHydrology & soilsDOAJ

Cite

BibTeX RIS

BibTeX

@misc{kwa2026extreme,
  title = {{Extreme events and statistics in an AI emulator of a global storm-resolving model}},
  author = {Anna Kwa and W. Andre Perkins and Troy Arcomano and Noah D. Brenowitz and Christopher S. Bretherton and Spencer K. Clark and Lucas M. Harris and Jeremy McGibbon and Michael S. Pritchard and Oliver Watt-Meyer},
  howpublished = {arXiv},
  year = {2026},
  month = oct,
  doi = {10.48550/arxiv.2610.17693},
  eprint = {2610.17693},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.17693},
  note = {CC BY 4.0},
}

RIS

TY  - UNPB
TI  - Extreme events and statistics in an AI emulator of a global storm-resolving model
AU  - Anna Kwa
AU  - W. Andre Perkins
AU  - Troy Arcomano
AU  - Noah D. Brenowitz
AU  - Christopher S. Bretherton
AU  - Spencer K. Clark
AU  - Lucas M. Harris
AU  - Jeremy McGibbon
AU  - Michael S. Pritchard
AU  - Oliver Watt-Meyer
PY  - 2026
DA  - 2026/10/13
PB  - arXiv
DO  - 10.48550/arxiv.2610.17693
UR  - https://arxiv.org/abs/2610.17693
AB  - We present HiRO-ACEv2 (High Resolution Output Ai2 Climate Emulator), a two-stage framework for AI emulation of the 3~km resolution X-SHiELD global storm-resolving model (GSRM), and highlight its ability to generate large ensembles of data for studying the characteristics of tropical cyclones and extreme events in high-resolution simulations at an inference computational cost over 300 times lower than that of traditional numerical GSRMs. HiRO-ACEv2 expands upon the precipitation-only HiRO-ACEv1 model to generate coherent precipitation, 10~m winds, and sea level pressure outputs, enabling a broader range of events to be simulated at high-resolution. Fine-resolution predictions are generated by running an autoregressive emulator of coarsened X-SHiELD at 100 km, whose multivariate outputs are regionally downscaled to 3~km resolution via a separately trained diffusion model. Particular importance is placed on capturing the statistics of extreme events during the diffusion downscaling step. Our ability to match the tails of distributions and generate realistic-looking extreme events in both the `perfect prediction' case as well as in free-running emulator rollouts is improved via the use of multiple denoisers at inference time and several updates to the data sampling, noise sampling, and loss weighting during training. HiRO-ACEv2 closely matches the high-resolution reference distributions of precipitation and near-surface winds out to their 99.9999th percentile values, and broadly tracks the reference distribution of tropical cyclone intensities.
ER  -

Abstract

Related

Cite