Surface temperature extremes produced by huge machine learning hindcasts of summer 2023
2026
Read Surface temperature extremes produced by huge machine learning hindcasts of summer 2023 (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-09754.txt.
Introduction
Risser and eight co-authors ask how much worse summer heat peaks might have become under the same setup. Weather forecast groups model other heatwave paths, but small samples limit worst-case picture. Using the Spherical Fourier Neural Operator model, the team makes 7,424 path forecasts of summer heat peaks. Despite shared boundary states, the machine-learning set produces hotter runs than weather forecast groups. For 30 percent of land areas, machine-learning peaks fall outside the range drawn from smaller forecast groups, showing large groups improve worst-case pictures.
Catalog record
| YEAR | 2026 |
|---|---|
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 33652 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.09754
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Surface temperature extremes produced by huge machine learning hindcasts of summer 2023 Abstract The summer of 2023 was the second hottest on record, with numerous extreme heatwaves across the globe. Using the Spherical Fourier Neural Operator machine learning (ML) weather model, we generated a massive ensemble of 7,424 weather scenarios simulating summer temperature extremes. The ML ensemble produced extreme heatwave scenarios exceeding temperatures from reanalysis and numerical weather prediction ensembles. Our results show that the ML model’s extreme surface temperatures were not unusual for approximately two-thirds of the global land area. However, for the other one-third, ML-generated extreme events were well outside the prediction envelope from extrapolating smaller ensembles with extreme value theory. Furthermore, the ML ensemble readily generates storyline simulations of humid heat extremes, which yield more dangerous categories of public safety alerts than can be simulated from smaller ensembles. This research highlights the potential of huge ensemble simulations to improve understanding and prediction of both humid and dry temperature extremes. † † journal: Geophysical Research Letters † † authors: Mark Risser, Ankur Mahesh, Joshua North, William D. Collins, Boris Bonev, Karthik Kashinath, Thorsten Kurth, Shashank Subramanian, and Michael S. Pritchard † † corresponding: Mark D. Risser, [email protected] Earth and Environmental Sciences, Lawrence Berkeley National Laboratory (LBNL), California, USA Department of Earth and Planetary Science, University of California, Berkeley, California, USA NVIDIA, Santa Clara, USA Department of Earth System Science, University of California, Irvine, California, USA National Energy Research Scientific Computing Center (NERSC), LBNL, Berkeley, California, USA These authors contributed equally to this work. keypoints Surface temperature extremes generated by a huge ML-based ensemble exceed both observations and those from a traditional forecast ensemble. For 30% of the globe, the ML ensemble predicts more intense heatwaves than those derived from extreme value theory applied to NWP ensembles. Going beyond extreme value theory, the larger ML ensemble generates global storylines of humid heat extremes relevant to public safety. Plain Language Summary Devastating heatwaves now threaten vast areas of the Earth’s surface, presenting significant challenges for adaptation and impact planning. The extreme, record-breaking heatwaves experienced during the summer of 2023 highlight the necessity of understanding the full plausible range of how hot temperatures can become. This study uses machine learning (ML) to simulate alternate scenarios of heatwave risk, with the ability to explore more extreme outcomes than current simulation capabilities. We find that the ML model was able to generate extreme heatwaves that exceeded observed temperatures for summer 2023 and those predicted by traditional forecasting methods. For nearly one-third of the global land area, including parts of Greenland, Russia, Alaska, and China, the ML-generated heatwaves were well outside the prediction range of traditional forecasts. We furthermore show how the ML model yields important information about the risks associated with humid heat extremes. Our study highlights the capacity of ML to simulate realistic and otherwise unanticipated extreme weather events at a fraction of computational cost of state-of-the-art weather forecasting tools. 1 Introduction Extreme heatwaves are among the most damaging hazards associated with a changing climate, and understanding how intense such events can plausibly become is an important scientific challenge. The 2021 heatwave in the North American Pacific Northwest and British Columbia is a compelling example of a recent extreme heatwave ( 52 ) . It was generated by interactions among atmospheric circulation ( 50 ) , sea-surface temperature patterns, and soil moisture ( 45 ) . These collectively amplified the event beyond five standard-deviations away from the historical median surface air temperatures ( 3 ) with anomalies relative to climatology as high as 25 ∘ C ( 14 ) . The daily maximum temperatures are well above the upper bounds set by Generalized Extreme Value (GEV) theory applied to the climate records for that region ( 4 ) . Events of this severity can be reproduced using traditional Earth System and numerical weather prediction models using large ensembles ( 38 ) , plausible counterfactual “storylines” identified using ensemble boosting techniques ( 17 ) , and ensemble weather and seasonal forecasts ( 30 ; 16 ; 26 ) . Machine learning (ML) may help understand changes and enhance predictability in heatwave occurrence at regional scales ( 2 ) up to months in advance ( 6 ; 8 ; 22 ; 37 ; 39 ) . The applications of ML to heatwaves are expanding to include causal inference of the precursors and drivers for heatwave formation ( 7 ; 25 ; 27 ) , time of emergence relative to natural variability ( 28 ) , and attribution to anthropogenic climate change ( 24 ) . We focus this study on the behavior of heatwaves during the boreal summer of 2023, which was the hottest summer in the observational record at the time ( 53 ) and probably during the last 2000 years ( 15 ; 20 ) . It has since been surpassed by summer 2024 ( 41 ) and was nearly passed by summer 2025 ( 54 ) . The temperature anomalies during summer 2023 significantly exceeded those predicted by observationally-calibrated statistical climate models ( 44 ) . The annually averaged temperature of 2023 was higher than a climatological baseline during 1850-1900 by 1.43 ∘ C ( 18 , central estimate) and warmer by 1.49 ∘ C relative to a pre-1700 baseline ( 23 , central estimate) . We elected to construct our huge ensemble of hindcasts for this summer to understand how several types of extremes, including heatwaves but also tropical cyclones and atmospheric rivers, would respond to these warmer conditions. In this paper, we examine whether ML emulators can generate heatwave storylines that are more intense than those found in conventional numerical weather prediction (NWP) ensembles, yet remain physically plausible. Because ML emulators can generate ensembles orders of magnitude larger than those from traditional NWP models, they sample far into the tails of the forecast distribution, capturing rare and extreme events that conventional ensembles would likely miss ( 35 ) . We discuss the time period for this study, the datasets used, and our use of GEV theory in Section 2 . Our results on the information regarding particularly intensive but plausible heatwaves generated by ML and missing from ERA5 and NWP are detailed in Section 3 . Section 4 concludes the paper. 2 Data and Methods 2.1 Overview of HENS We created a huge ensemble (HENS) of weather extremes using an ML emulator of global numerical weather reanalyses and simulations of the recent climate record. Our goal is to conduct the first studies of whether massive ensembles from ML emulators improve the ability to simulate “low likelihood, high-impact” (LLHI) weather events, using terminology from the sixth Intergovernmental Panel on Climate Change Report, which states that we currently have low confidence in current and future projections of LLHIs ( 46 ) . Studies of LLHI extreme weather and climate events require massive ensembles to capture long tails of multivariate distributions ( 19 ) . Until recently, generating 1,000- or 10,000-member ensembles of high-resolution hindcasts was exceptionally challenging because of prohibitive compute and data storage costs. The climate community has recognized, however, that such massive ensembles are needed to characterize LLHIs ( 49 ) . HENS is an ensemble hindcast consisting of