Pith. sign in

REVIEW 2 major objections 6 minor 93 references

At 10–15 day leads, AI weather emulators match or beat dynamical models on temperature skill but blur extremes and trail IFS on heat-extreme recall.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-31 14:15 UTC pith:HOWAFLLQ

load-bearing objection Solid multi-model benchmark at 10–15 d heat: AI wins on RMSE/ACC via blurring, IFS still leads extreme recall; reference asymmetry is real but secondary. the 2 major comments →

arxiv 2607.28220 v1 pith:HOWAFLLQ submitted 2026-07-30 physics.ao-ph cs.LG

Weather Emulators at the Frontier of Heat Extremes Predictability

Cas Decancq , Thomas Mortier , Jessica Keune , Diego G. Miralles This is my paper
classification physics.ao-ph cs.LG PACS 92.60.Wc92.60.Ry07.05.Mh
keywords weather emulatorsextreme heatsubseasonal predictionblurringspectral fidelitymedium-range forecastingheat early warningdeterministic AI weather models
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Beyond about ten days, weather forecasts lose most of their ability to pin down specific states and mainly capture broad trends. That horizon is exactly where early warnings of deadly heatwaves would be most valuable. This paper systematically tests six leading deep-learning weather emulators against three major physics-based forecast systems and simple statistical baselines on global land surface temperature and on heat extremes (temperatures two standard deviations above climatology) at 10–15 day lead times. Several emulators equal or beat the dynamical models on ordinary error and correlation scores, yet they achieve that skill by smoothing the fields—an effect called blurring—so they systematically under-represent the intensity and spatial detail of heat peaks. All models retain some skill above chance for identifying extremes, but the European IFS system still recalls more true extreme-heat area than any emulator. The work therefore shows both that AI is already competitive at the edge of the classical predictability window and that purpose-built training, spectral fidelity, and better Earth-system inputs will be needed before the models can deliver reliable, actionable heat warnings.

Core claim

Several state-of-the-art deterministic weather emulators rival or surpass leading dynamical systems on deterministic near-surface temperature skill at 10–15 day leads, but they do so at the cost of reduced spectral fidelity (blurring), most under-represent peak extreme-heat intensities, and none matches IFS recall of heat extremes defined as T ≥ μ + 2σ.

What carries the argument

Fixed-lead-time evaluation of six deterministic emulators versus three dynamical systems and climatology/persistence baselines, using continuous scores (RMSE, R², ACC, temporal correlation, spectral score) plus categorical heat-extreme metrics (precision, recall, ETS, reliability) on both global land and six recent large-scale heat events.

Load-bearing premise

That scoring emulators against ERA5 while scoring dynamical models against their own initial states, and applying one shared ERA5 climatology threshold to everyone, produces a fair ranking of extreme-heat skill.

What would settle it

Re-run the identical 10–15 day extreme-heat recall comparison after forcing every model onto a common reference and a common climatology derived from that reference; if IFS no longer leads every emulator, the central ranking claim fails.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Ordinary temperature skill at the medium-to-subseasonal boundary can already be improved by existing AI emulators without new physics.
  • Heat-risk applications cannot rely on those same emulators until spectral fidelity and peak-intensity bias are fixed.
  • IFS-style dynamical systems remain the stronger choice for recall of rare extremes at these leads.
  • Purpose-built subseasonal training, probabilistic ensembles, and richer land–ocean–ice inputs are the concrete next steps the paper identifies.
  • Even current deterministic emulators retain marginal but real skill above chance, so large cheap ensembles could still raise practical early-warning value.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Because blurring is largely an artifact of Lp loss and multi-stage or ensemble averaging, simply changing the training objective or adding a spectral penalty may close much of the intensity gap without new architecture.
  • The same computational cheapness that lets emulators scale to large ensembles also makes them natural candidates for Forecast-based Financing triggers once reliability is demonstrated.
  • If the shared ERA5 threshold systematically under-counts extremes in dynamical fc0 fields, the apparent IFS recall advantage may shrink once each system is scored against its own climatology.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 6 minor

Summary. The manuscript benchmarks six deterministic deep-learning weather emulators (Pangu-Weather, FuXi, ArchesWeather, AIFS, GraphCast, Aurora) against three dynamical systems (ECMWF-IFS, CMA, NCEP) and climatology/persistence baselines for global land near-surface temperature and extreme heat (T ≥ μ + 2σ from ERA5 1990–2019 climatology) at 10–15 day leads, using six recent large-scale heat events. Continuous skill is assessed via area-weighted RMSE, R², ACC, temporal correlation, and a proposed spectral score based on zonal Fourier power spectra; categorical skill uses precision, recall, ETS, and reliability conditioned on extreme forecasts. The central claims are that several emulators rival or surpass NWP on deterministic temperature metrics but at the cost of spectral blurring and under-representation of peak intensities, that all models retain some extreme-heat skill beyond climatology/persistence inside events, and that IFS recall of heat extremes exceeds that of every emulator, with AIFS the most balanced emulator overall.

Significance. This is a timely, multi-model, open-code benchmark at the medium-range/subseasonal interface for a societally critical hazard. Strengths include explicit fixed-lead concatenation, land masking, area weighting, out-of-sample post-2020 events, dual global vs event-level evaluation, and a reproducible spectral-score definition that makes the blurring trade-off quantitative rather than purely visual. The finding that deterministic emulators can beat or match NWP on Lp-type scores while losing extremes intensity and recall is actionable for both the AI-weather and early-warning communities, and the case-study and reliability analyses usefully separate “skill above climatology” from “actionable extreme warnings.” Code and data provenance are clearly stated.

major comments (2)
  1. [§2.1, Methods 4.2–4.6; Supp. Fig. 4, Tables 4–5] §2.1 and Methods 4.2–4.6: Emulators are verified against ERA5 while dynamical systems are verified against their own fc0, yet anomalies, binary extreme labels, and reliability thresholds for all systems are defined from a single ERA5 1990–2019 climatology. Supp. Fig. 4 and Tables 4–5 already show that CMA/NCEP fc0 hot-extreme areas differ substantially from ERA5; applying an ERA5 threshold therefore changes the recall denominator and chance hits inside ETS. Because IFS is climatologically closest to ERA5, this design can systematically favour IFS recall (and, after ETS correction, penalise hot-biased NCEP/CMA) relative to ERA5-trained emulators. The continuous scores are less exposed, but the load-bearing claim that “IFS recall is greater than that of any of the emulators” needs a sensitivity test: recompute categorical metrics for NWP using (i) each system’s own analysis climatology if
  2. [§2.2, Fig. 3] §2.2 and Fig. 3 (left): The multi-metric ranking that crowns FuXi (global) and AIFS (events) equally weights RMSE, R², ACC, temporal correlation, and spectral score. These metrics are not independent (blurring improves RMSE/R² while degrading the spectral score by construction), and the equal-weight average is not justified. A short sensitivity (e.g., rank by RMSE+ACC only, or by a Pareto front of skill vs spectral score) would show whether the headline ordering is robust or an artefact of the chosen weights.
minor comments (6)
  1. [Methods 4.5, Eq. (4)] Eq. (4) writes f′(t,i,j) = f(t,i,j) − o(t,i,j) and o′ = o − ō; the first line appears to subtract the reference field rather than the climatological mean. Clarify that anomalies are relative to ERA5 climatology ō for both forecast and reference (as stated in the surrounding text).
  2. [Methods 4.5; Supp. S8] The spectral score (Eq. 11) is a useful contribution; state explicitly the number and range of log-wavelength bins M and whether results are sensitive to that choice (Supp. S8).
  3. [§2.2; Table S3] ArchesWeather is evaluated as a 4-member micro-ensemble mean on a 1.5° grid while others are single deterministic 0.25° forecasts. The text notes the smoothing effect; a one-sentence caveat in the global ranking paragraph would help readers not over-interpret its RMSE advantage.
  4. [Fig. 4] Fig. 4 right panels: the −1.5%/day and −2.7%/day annotations are not defined in the caption or Methods; add a brief explanation.
  5. [Title] Title on the arXiv PDF header (“Weather Emulators Push the Frontier…”) differs slightly from the manuscript title (“…at the Frontier…”); align for citation consistency.
  6. [§2.3; Supp. S7] Several supplementary cross-references (e.g., “Figures S6–S11” vs S7–S12 in the main text) appear off-by-one; renumber carefully.

Circularity Check

0 steps flagged

No circularity: empirical multi-model benchmark with standard scores on held-out events; no claim reduces to its inputs by construction.

full rationale

This paper is an out-of-sample verification study, not a first-principles derivation. Six pre-trained emulators and three NWP systems are scored with conventional continuous metrics (RMSE, R², ACC, temporal correlation) and categorical metrics (precision, recall, ETS) plus a spectral score built from Fourier power spectra following WeatherBench2. Extreme-heat labels use a fixed ERA5 1990–2019 μ+2σ threshold applied after the fact; no free parameter is fit to the evaluation events and then re-presented as a prediction. Emulators were trained by external groups on pre-2020 ERA5; all six events are post-2020 and thus temporally out-of-sample. Self-citations (e.g. Miralles on land–atmosphere feedbacks) supply only scientific motivation, not a uniqueness theorem or load-bearing premise for the IFS-vs-emulator ranking. Reference asymmetry (ERA5 vs fc0) and the shared ERA5 threshold are methodological fairness issues, not circular reductions: the reported recall/ETS numbers are still genuine empirical comparisons, not identities forced by definition. No step matches self-definitional, fitted-input-as-prediction, self-citation-load-bearing, uniqueness-import, ansatz-smuggling, or renaming patterns. Score 0 is therefore appropriate.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 1 invented entities

Load-bearing content is empirical protocol choices and domain conventions, not free parameters fitted to force the headline ranking. The claim rests on standard forecast verification practice, ERA5-centric truth and climatology, a fixed extreme threshold, and a curated set of six heat events.

free parameters (4)
  • Extreme-heat threshold multiplier (2σ) = 2
    Heat extremes are defined as T ≥ μ + 2σ from ERA5 1990–2019 climatology; the factor 2 is a conventional choice that sets prevalence and all categorical scores.
  • Climatology smoothing window = 31 days
    Mean and σ use a 31-day linearly tapered rolling window over day-of-year; window length is a processing choice affecting μ, σ, and anomaly-based metrics.
  • Event-level mask (≥3 extreme days) = ≥3 days
    Event evaluation is restricted to locations with at least three extreme-heat days inside hand-drawn boxes; the day count and boxes shape the ‘events’ skill panel.
  • Equal-weight multi-metric ranking = equal weights
    Global model ordering averages RMSE, R², ACC, temporal correlation, and spectral score with equal weight; weighting is author-chosen and affects who is ‘top’.
axioms (5)
  • domain assumption ERA5 two-meter temperature is an adequate verification truth for emulators trained on ERA5.
    Stated in §2.1 and Methods 4.2/4.5; standard but couples training distribution to the scorecard.
  • domain assumption Dynamical models should be verified against their own fc0 at valid time rather than ERA5.
    Following cited practice [27,71,72]; enables fair NWP self-consistency but complicates cross-class comparison.
  • domain assumption Atmospheric predictability at 10–15 days is near the practical limit where anomaly skill is weak and baselines are strong.
    Frames the study via Lorenz/predictability literature in the Introduction; motivates the lead window.
  • ad hoc to paper A single ERA5-derived climatology can be used to form anomalies and extreme thresholds for all forecast systems.
    Authors note model climatologies are unavailable (§2.1); this shared baseline is necessary for the paper’s protocol but is not model-native.
  • ad hoc to paper Discrete Fourier zonal spectra averaged in log wavelength bins yield a meaningful scalar spectral fidelity score.
    Spectral score defined in Methods 4.5 / S8 following WeatherBench2-style spectra; useful but one of many possible blur metrics.
invented entities (1)
  • Spectral score (1 − mean |log10 power ratio| over wavelength bins) no independent evidence
    purpose: Scalar summary of whether forecast temperature-anomaly power matches reference across spatial scales.
    New composite metric proposed in this paper; built from standard DFT spectra, not a physical entity.

pith-pipeline@v1.2.0-daily-grok45 · 47762 in / 3357 out tokens · 65534 ms · 2026-07-31T14:15:09.955325+00:00 · methodology

0 comments
read the original abstract

Atmospheric predictability declines rapidly beyond the next ten days, such that forecasts at longer lead times primarily convey large-scale trends rather than specific states. Yet in a warming world, improving early warnings of extreme heat is an increasingly critical challenge. Here we evaluate six state-of-the-art deep learning weather emulators - Pangu-Weather, FuXi, ArchesWeather, AIFS, GraphCast and Aurora - alongside leading dynamical systems and statistical baselines in forecasting global near-surface temperature and extreme heat at lead times of 10-15 days. We find that several emulators rival or even surpass physics-based forecasts in deterministic temperature skill, but do so at the cost of reduced spectral fidelity, in a process widely known as blurring. While all models show some degree of predictive skill for extreme heat, most emulators under-represent peak intensities, and IFS recall is greater than that of any of the emulators. These results highlight both the emerging potential of AI to enhance extended range temperature prediction, and the remaining challenges in delivering reliable, actionable early warnings in a changing climate.

Figures

Figures reproduced from arXiv: 2607.28220 by Cas Decancq, Diego G. Miralles, Jessica Keune, Thomas Mortier.

Figure 1
Figure 1. Figure 1: Composite view of six extreme temperature events in the past five years. The main panel shows the total number of days with ERA5 surface temperature [90] exceeding two standard deviations above its 1990– 2019 climatological mean (Section 4.4) during each event. Insets specify the studied period and display the evolution of near-surface temperature in the affected regions by means of a latitude-weighted ave… view at source ↗
Figure 2
Figure 2. Figure 2: Example anomaly maps. Near-surface temperature anomaly at the peak extent of the 2021 heat dome over the Northwest Pacific (2021-06- 30 00:00 UTC) in references (IFS-fc0, CMA-fc0, NCEP-fc0 and ERA5) and corresponding forecasts initialized 14 days prior (2021-06-16 00:00 UTC). Similar visualization is provided for all events in Figures S7–S12. March through May [58]. More recently, record-breaking heat has … view at source ↗
Figure 3
Figure 3. Figure 3: Model benchmarking results. Left panels: fixed-lead-time model forecast scores across events and averaged over lead days 10 to 15, per metric (Section 4.5). A dark blue fill denotes a model performing best for that metric, whereas a white fill denotes the opposite. For each metric, the best score is reported in white font. Metrics considered are RMSE, R2 , ACC, temporal correlation and spectral score (Sect… view at source ↗
Figure 4
Figure 4. Figure 4: Categorical forecast skill and reliability of extreme heat forecasts. Near-surface temperature and forecasts are cast into binary format indicating extreme heat occurrence. Left panels: precision, recall, and the ET S (Section 4.6) as a function of lead time at global (top) and event-level (bottom) scales. Right panels: given an extreme heat forecast, probability of experiencing above-normal (T ≥ µ, top), … view at source ↗
Figure 5
Figure 5. Figure 5: AIFS case studies. AIFS lead day 14 forecasts for the major cities highlighted in [PITH_FULL_IMAGE:figures/full_fig_p014_5.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

93 extracted references · 3 canonical work pages

  1. [1]

    author Zhao, Q. et al. Global, regional, and national burden of mortality associated with non-optimal ambient temperatures from 2000 to 2019: a three-stage modelling study. journal The Lancet Planetary Health volume 5 , pages e415--e425 ( year 2021 )

    2000

  2. [2]

    author Ballester, J. et al. Heat-related mortality in E urope during the summer of 2022. journal Nature Medicine volume 29 , pages 1857--1866 ( year 2023 )

    2022

  3. [3]

    Climate change mitigation and adaptation ( year 2025 )

    author European Environment Agency . Climate change mitigation and adaptation ( year 2025 ). ://doi.org/10.2800/3817344. note In: Section 3 ``Europe's environment and climate: state and outlook''

    work page doi:10.2800/3817344 2025

  4. [4]

    State of the Climate Update for COP30

    author World Meteorological Organization (WMO) . State of the Climate Update for COP30. howpublished Digital report ( year 2025 ). ://library.wmo.int/idurl/4/69674. note Climate Statement

    2025

  5. [5]

    & author Tsirkunov, V

    author Rogers, D. & author Tsirkunov, V. Implementing hazard early warning systems. journal Global Facility for Disaster Reduction and Recovery volume 11 , pages 1--47 ( year 2011 )

    2011

  6. [6]

    title Heatwaves and Health: Guidance on Warning-System Development

    author World Meteorological Organization & author World Health Organization . title Heatwaves and Health: Guidance on Warning-System Development . No. number 1142 in series WMO-No. ( publisher World Meteorological Organization , address Geneva , year 2015 ). ://library.wmo.int/idurl/4/54600

    2015

  7. [7]

    author S aki \'c Trogrli \'c , R. et al. Early warning systems and their role in disaster risk reduction. In editor Taylor, A. , editor Kox, T. & editor Johnston, D. (eds.) booktitle Towards the ``perfect'' weather warning: Bridging disciplinary gaps through partnership and communication , pages 11--46 ( publisher Springer , address Cham , year 2022 )

    2022

  8. [8]

    author Miralles, D. G. , author Gentine, P. , author Seneviratne, S. I. & author Teuling, A. J. Land--atmospheric feedbacks during droughts and heatwaves: state of the science and current challenges. journal Annals of the New York Academy of Sciences volume 1436 , pages 19--35 ( year 2019 )

    2019

  9. [9]

    author Domeisen, D. I. et al. Prediction and projection of heatwaves. journal Nature Reviews Earth & Environment volume 4 , pages 36--50 ( year 2023 )

    2023

  10. [10]

    , author Garc \' a-Herrera, R

    author Barriopedro, D. , author Garc \' a-Herrera, R. , author Ord \'o \ n ez, C. , author Miralles, D. & author Salcedo-Sanz, S. Heat waves: Physical understanding and scientific challenges. journal Reviews of Geophysics pages e2022RG000780 ( year 2023 )

    2023

  11. [11]

    author Scholz, S. R. & author Lora, J. M. Atmospheric rivers cause warm winters and extreme heat events. journal Nature volume 636 , pages 640--646 ( year 2024 )

    2024

  12. [12]

    , author Mayr, G

    author Fix, F. , author Mayr, G. , author Zeileis, A. , author Stucke, I. & author Stauffer, R. Detection and consequences of atmospheric deserts: insights from a case study. journal Weather and Climate Dynamics volume 5 , pages 1545--1560 ( year 2024 )

    2024

  13. [13]

    author Luo, M. et al. Anthropogenic forcing has increased the risk of longer-traveling and slower-moving large contiguous heatwaves. journal Science Advances volume 10 , pages eadl1598 ( year 2024 )

    2024

  14. [14]

    , author Janousek, M

    author Haiden, T. , author Janousek, M. , author Vitart, F. , author Ben Bouall è gue, Z. & author Ferranti, F. Evaluation of ECMWF forecasts, including the 2021 upgrade. type ECMWF Technical Memorandum number 884 , institution European Centre for Medium-Range Weather Forecasts (ECMWF) ( year 2021 ). ://doi.org/10.21957/90pgicjk4

    work page doi:10.21957/90pgicjk4 2021

  15. [15]

    author Rasp, S. et al. WeatherBench 2: A benchmark for the next generation of data-driven global weather models. journal Journal of Advances in Modeling Earth Systems volume 16 , pages e2023MS004019 ( year 2024 )

    2024

  16. [16]

    Extended Range Output -- Extended ENS ( year 2020 )

    author European Centre for Medium-Range Weather Forecasts . Extended Range Output -- Extended ENS ( year 2020 ). ://confluence.ecmwf.int/pages/viewpage.action?pageId=222490862. note Forecast User Guide, Section 8.2

    2020

  17. [17]

    author Lorenz, E. N. The Predictability of a Flow Which Possesses Many Scales of Motion. journal Tellus volume 21 , pages 289--307 ( year 1969 )

    1969

  18. [18]

    Insights into atmospheric predictability through global convection-permitting model simulations

    author Judt, F. Insights into atmospheric predictability through global convection-permitting model simulations. journal Journal of the Atmospheric Sciences volume 75 , pages 1477--1497 ( year 2018 )

    2018

  19. [19]

    author Zhang, F. et al. What is the predictability limit of midlatitude weather? journal Journal of the Atmospheric Sciences volume 76 , pages 1077--1091 ( year 2019 )

    2019

  20. [20]

    , author Riemer, M

    author Selz, T. , author Riemer, M. & author Craig, G. C. The transition from practical to intrinsic predictability of midlatitude weather. journal Journal of the Atmospheric Sciences volume 79 , pages 2013--2030 ( year 2022 )

    2013

  21. [21]

    author White, C. J. et al. Potential applications of subseasonal-to-seasonal ( S2S ) predictions. journal Meteorological Applications volume 24 , pages 315--325 ( year 2017 )

    2017

  22. [22]

    author Vonich, P. T. & author Hakim, G. J. Testing the limit of atmospheric predictability with a machine learning weather model ( year 2025 ). note Preprint at https://arxiv.org/abs/2504.20238

    Pith/arXiv arXiv 2025

  23. [23]

    Implementation of IFS Cycle 49r1

    author European Centre for Medium-Range Weather Forecasts (ECMWF) . Implementation of IFS Cycle 49r1. howpublished https://confluence.ecmwf.int/display/FCST/Implementation+of+IFS+Cycle+49r1 ( year 2025 ). note Created by Milana Vuckovic; last modified on 24 Nov 2025

    2025

  24. [24]

    GRAPES Global Ensemble Prediction System (TIGGE contribution, BABJ)

    author China Meteorological Administration (CMA) . GRAPES Global Ensemble Prediction System (TIGGE contribution, BABJ). howpublished https://confluence.ecmwf.int/display/TIGGE/Models ( year 2018 ). note TIGGE metadata indicates implementation of GRAPES\_GFS-based ensemble (BABJ) from 26 Dec 2018; includes transition from TL639L60 to GRAPES\_GFS, updated p...

    2018

  25. [25]

    Global Ensemble Forecast System (GEFS), version 12

    author National Centers for Environmental Prediction (NCEP) . Global Ensemble Forecast System (GEFS), version 12. howpublished https://www.emc.ncep.noaa.gov/emc/pages/numerical_forecast_systems/gefs.php ( year 2020 ). note Implemented 23 Sep 2020; TIGGE contribution (KWBC); 31-member ensemble, 25 km resolution (C384L64), coupled atmosphere-wave system

    2020

  26. [26]

    author Bi, K. et al. Accurate medium-range global weather forecasting with 3D neural networks. journal Nature volume 619 , pages 533--538 ( year 2023 )

    2023

  27. [27]

    author Lam, R. et al. Learning skillful medium-range global weather forecasting. journal Science volume 382 , pages 1416--1421 ( year 2023 )

    2023

  28. [28]

    author Chen, L. et al. FuXi: a cascade machine learning forecasting system for 15-day global weather forecast. journal npj Climate and Atmospheric Science volume 6 , pages 190 ( year 2023 )

    2023

  29. [29]

    , author Singh, R

    author Couairon, G. , author Singh, R. , author Charantonis, A. , author Lessig, C. & author Monteleoni, C. Archesweather & ArchesWeatherGen: a deterministic and generative model for efficient ML weather forecasting ( year 2024 ). note Preprint at https://arxiv.org/abs/2412.12971

    Pith/arXiv arXiv 2024

  30. [30]

    author Lang, S. et al. AIFS -- ECMWF 's data-driven forecasting system ( year 2024 ). note Preprint at https://arxiv.org/abs/2406.01465

    Pith/arXiv arXiv 2024

  31. [31]

    author Bodnar, C. et al. A foundation model for the earth system. journal Nature pages 1--8 ( year 2025 )

    2025

  32. [32]

    author Pasche, O. C. , author Wider, J. , author Zhang, Z. , author Zscheischler, J. & author Engelke, S. Validating deep learning weather forecast models on recent High-Impact extreme events. journal Artificial Intelligence for the Earth Systems volume 4 , pages e240033 ( year 2025 )

    2025

  33. [33]

    & author Vitart, F

    author Robertson, A. & author Vitart, F. title Sub-seasonal to seasonal prediction: the gap between weather and climate forecasting ( publisher Elsevier , year 2018 )

    2018

  34. [34]

    author Robertson, A. W. , author Kumar, A. , author Pe \ n a, M. & author Vitart, F. Improving and promoting subseasonal to seasonal prediction. journal Bulletin of the American Meteorological Society volume 96 , pages ES49--ES53 ( year 2015 )

    2015

  35. [35]

    AI Weather Quest ( year 2025 )

    author ECMWF . AI Weather Quest ( year 2025 ). ://aiweatherquest.ecmwf.int/. note Accessed: 2025-08-06

    2025

  36. [36]

    What is Forecast-based Financing? howpublished https://www.forecast-based-financing.org/about/ ( year 2025 )

    author German Red Cross . What is Forecast-based Financing? howpublished https://www.forecast-based-financing.org/about/ ( year 2025 ). note Accessed: 2025-10-16

    2025

  37. [37]

    & author Robertson, A

    author Vitart, F. & author Robertson, A. W. The sub-seasonal to seasonal prediction project (S2S) and the prediction of extreme events. journal npj Climate and Atmospheric Science volume 1 , pages 3 ( year 2018 )

    2018

  38. [38]

    , author Orenstein, P

    author Hwang, J. , author Orenstein, P. , author Cohen, J. , author Pfeiffer, K. & author Mackey, L. Improving subseasonal forecasting in the western US with machine learning. journal Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining pages 2325--2335 ( year 2019 )

    2019

  39. [39]

    author Donaldson, G. C. , author Keatinge, W. R. & author Saunders, R. D. Cardiovascular responses to heat stress and their adverse consequences in healthy and vulnerable human populations. journal International Journal of Hyperthermia volume 19 , pages 225--235 ( year 2003 )

    2003

  40. [40]

    author McMichael, A. J. & author Lindgren, E. Climate change: present and future risks to health, and necessary responses. journal Journal of Internal Medicine volume 270 , pages 401--413 ( year 2011 )

    2011

  41. [41]

    & author Rahmstorf, S

    author Coumou, D. & author Rahmstorf, S. A decade of weather extremes. journal Nature Climate Change volume 2 , pages 491--496 ( year 2012 )

    2012

  42. [42]

    Heatwaves are silent killers

    author Loughnan, M. Heatwaves are silent killers. journal Geodate volume 27 , pages 7--10 ( year 2014 )

    2014

  43. [43]

    author Horton, R. M. , author Mankin, J. S. , author Lesk, C. , author Coffel, E. & author Raymond, C. A review of recent advances in research on extreme heat events. journal Current Climate Change Reports volume 2 , pages 242--259 ( year 2016 )

    2016

  44. [44]

    , author Remenyi, T

    author Campbell, S. , author Remenyi, T. A. , author White, C. J. & author Johnston, F. H. Heatwave and health impact research: A global review. journal Health & Place volume 53 , pages 210--218 ( year 2018 )

    2018

  45. [45]

    Exploring machine-learning forecasts of extreme weather

    author Magnusson, L. Exploring machine-learning forecasts of extreme weather. howpublished https://www.ecmwf.int/en/newsletter/176/news/exploring-machine-learning-forecasts-extreme-weather ( year 2023 ). note ECMWF Newsletter No. 176, Summer 2023; published July 2023

    2023

  46. [46]

    author Ben Bouall \`e gue, Z. et al. The rise of data-driven weather forecasting: A first statistical assessment of machine learning--based weather forecasts in an operational-like context. journal Bulletin of the American Meteorological Society volume 105 , pages E864--E883 ( year 2024 )

    2024

  47. [47]

    , author Mo, R

    author Lin, H. , author Mo, R. & author Vitart, F. The 2021 western North American heatwave and its subseasonal predictions. journal Geophysical Research Letters volume 49 , pages e2021GL097036 ( year 2022 )

    2021

  48. [48]

    author Emerton, R. et al. Predicting the unprecedented: forecasting the June 2021 Pacific Northwest heatwave. journal Weather volume 77 , pages 272--279 ( year 2022 )

    2021

  49. [49]

    , author Schmeits, M

    author Vijverberg, S. , author Schmeits, M. , author Van der Wiel, K. & author Coumou, D. Subseasonal statistical forecasts of eastern US hot temperature events. journal Monthly Weather Review volume 148 , pages 4799--4822 ( year 2020 )

    2020

  50. [50]

    author Benson, D. O. & author Dirmeyer, P. A. The soil moisture--surface flux relationship as a factor for extreme heat predictability in subseasonal to seasonal forecasts. journal Journal of Climate volume 36 , pages 6375--6392 ( year 2023 )

    2023

  51. [51]

    , author Whan, K

    author Van Straaten, C. , author Whan, K. , author Coumou, D. , author Van den Hurk, B. & author Schmeits, M. Using explainable machine learning forecasts to discover subseasonal drivers of high summer temperatures in western and central E urope. journal Monthly Weather Review volume 150 , pages 1115--1134 ( year 2022 )

    2022

  52. [52]

    author Weirich-Benet, E. et al. Subseasonal prediction of central European summer heatwaves with linear and random forest machine learning models. journal Artificial Intelligence for the Earth Systems volume 2 , pages e220038 ( year 2023 )

    2023

  53. [53]

    , author Li, X

    author He, S. , author Li, X. , author DelSole, T. , author Ravikumar, P. & author Banerjee, A. Sub-seasonal climate forecasting via machine learning: Challenges, analysis, and advances. journal Proceedings of the AAAI Conference on Artificial Intelligence volume 35 , pages 169--177 ( year 2021 )

    2021

  54. [54]

    author Weyn, J. A. , author Durran, D. R. , author Caruana, R. & author Cresswell-Clay, N. Sub-seasonal forecasting with a large ensemble of deep-learning weather prediction models. journal Journal of Advances in Modeling Earth Systems volume 13 , pages e2021MS002502 ( year 2021 )

    2021

  55. [55]

    author Nathaniel, J. et al. Chaosbench: A multi-channel, physics-based benchmark for subseasonal-to-seasonal climate prediction ( year 2024 ). note Preprint at https://arxiv.org/abs/2402.00712

    Pith/arXiv arXiv 2024

  56. [56]

    author Henderson, S. B. , author McLean, K. E. , author Lee, M. J. & author Kosatsky, T. Analysis of community deaths during the catastrophic 2021 heat dome: early evidence to inform the public health response during subsequent events in greater Vancouver, Canada. journal Environmental Epidemiology volume 6 , pages e189 ( year 2022 )

    2021

  57. [57]

    , author Pascale, S

    author Faranda, D. , author Pascale, S. & author Bulut, B. Persistent anticyclonic conditions and climate change exacerbated the exceptional 2022 European-Mediterranean drought. journal Environmental Research Letters ( year 2023 )

    2022

  58. [58]

    author Zachariah, M. et al. Attribution of 2022 early-spring heatwave in India and Pakistan to climate change: lessons in assessing vulnerability and preparedness in reducing impacts. journal Environmental Research: Climate volume 2 , pages 045005 ( year 2023 )

    2022

  59. [59]

    & author Mishra, V

    author Aadhar, S. & author Mishra, V. The 2022 mega heatwave in South Asia in the observed and projected future climate. journal Environmental Research Letters volume 18 , pages 104011 ( year 2023 )

    2022

  60. [60]

    Extreme Sahel heatwave that hit highly vulnerable population at the end of Ramadan would not have occurred without climate change

    author World Weather Attribution . Extreme Sahel heatwave that hit highly vulnerable population at the end of Ramadan would not have occurred without climate change. howpublished https://www.worldweatherattribution.org/extreme-sahel-heatwave-that-hit-highly-vulnerable-population-at-the-end-of-ramadan-would-not-have-occurred-without-climate-change/ ( year ...

    2024

  61. [61]

    Record heat index of 62.3C scorches Brazil's Rio de Janeiro

    author Al Jazeera . Record heat index of 62.3C scorches Brazil's Rio de Janeiro. howpublished https://www.aljazeera.com/gallery/2024/3/18/photos-record-heat-index-of-62-3c-scorches-rio-de-janeiro ( year 2024 ). note Accessed: 2025-08-07

    2024

  62. [62]

    Extreme weather and climate impacts bite Latin America and Caribbean

    author World Meteorological Organization . Extreme weather and climate impacts bite Latin America and Caribbean. howpublished https://www.wmo.int/news/media-centre/extreme-weather-and-climate-impacts-bite-latin-america-and-caribbean ( year 2025 ). note Accessed: 2025-08-07

    2025

  63. [63]

    Climate change fuels record February heat in Brazil ahead of Carnival

    author Climate Central . Climate change fuels record February heat in Brazil ahead of Carnival. howpublished https://www.climatecentral.org/climate-shift-index-alert/brazil-february-2025 ( year 2025 ). note Accessed: 2025-08-07

    2025

  64. [64]

    Intense heatwave in southern Brazil forces schools to suspend return

    author Rogero, T. Intense heatwave in southern Brazil forces schools to suspend return. journal The Guardian ( year 2025 ). ://www.theguardian.com/world/2025/feb/12/brazil-record-heat-rio-grande-do-sul. note Accessed: 2025-08-07

    2025

  65. [65]

    author Masson-Delmotte, V. et al. Climate change 2021: the physical science basis. journal Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change volume 2 , pages 2391 ( year 2021 )

    2021

  66. [66]

    & author Corti, S

    author Simolo, C. & author Corti, S. Quantifying the role of variability in future intensification of heat extremes. journal Nature Communications volume 13 , pages 7930 ( year 2022 )

    2022

  67. [67]

    author Perkins, S. E. , author Alexander, L. & author Nairn, J. Increasing frequency, intensity and duration of observed global heatwaves and warm spells. journal Geophysical Research Letters volume 39 ( year 2012 )

    2012

  68. [68]

    author Beck, T. et al. Increasing likelihood of heat-related mortality events with global warming: a continental epidemiological extreme event attribution study ( year 2024 ). note Preprint at https://doi.org/10.21203/rs.3.rs-4128140/v1

    work page doi:10.21203/rs.3.rs-4128140/v1 2024

  69. [69]

    , author Kornhuber, K

    author Rousi, E. , author Kornhuber, K. , author Beobide-Arsuaga, G. , author Luo, F. & author Coumou, D. Accelerated western European heatwave trends linked to more-persistent double jets over Eurasia. journal Nature Communications volume 13 , pages 3851 ( year 2022 )

    2022

  70. [70]

    , author Migliavacca, M

    author Dosio, A. , author Migliavacca, M. & author Maraun, D. How fast is climate changing? One generation is sufficient for unfamiliar heatwave characteristics to emerge in Europe. journal Climatic Change volume 178 , pages 1--17 ( year 2025 )

    2025

  71. [71]

    title Manual on the Global Data-Processing and Forecasting System

    author World Meteorological Organization . title Manual on the Global Data-Processing and Forecasting System . address Geneva, Switzerland ( year 2023 ). note Appendix 2.2.35, Section 7

    2023

  72. [72]

    author Price, I. et al. Probabilistic weather forecasting with machine learning. journal Nature volume 637 , pages 84--90 ( year 2025 )

    2025

  73. [73]

    , author Couprie, C

    author Mathieu, M. , author Couprie, C. & author LeCun, Y. Deep multi-scale video prediction beyond mean square error. In booktitle 4th International Conference on Learning Representations ( year 2016 )

    2016

  74. [74]

    , author Zhu, J.-Y

    author Isola, P. , author Zhu, J.-Y. , author Zhou, T. & author Efros, A. A. Image-to-image translation with conditional adversarial networks. In booktitle Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages 1125--1134 ( year 2017 )

    2017

  75. [75]

    author Rahaman, N. et al. On the spectral bias of neural networks. In booktitle Proceedings of the 36th International Conference on Machine Learning , vol. volume 97 of series Proceedings of Machine Learning Research , pages 5301--5310 ( publisher PMLR , year 2019 )

    2019

  76. [76]

    & author Geer, A

    author Bonavita, M. & author Geer, A. J. Forecast verification using information and noise. journal Quarterly Journal of the Royal Meteorological Society pages e70109 ( year 2026 )

    2026

  77. [77]

    Accuracy versus activity ( year 2024 )

    author Zied Ben Bouallègue and The AIFS Team . Accuracy versus activity ( year 2024 ). ://www.ecmwf.int/en/about/media-centre/aifs-blog/2024/accuracy-versus-activity. note ECMWF AIFS Blog

    2024

  78. [78]

    author Owens, R. G. & author Hewson, T. D. title ECMWF Forecast User Guide . address Reading, UK ( year 2018 ). ://confluence.ecmwf.int/display/FUG/Section+6.2.2+Anomaly+Correlation+Coefficient. note Section 6.2.2: Anomaly Correlation Coefficient

    2018

  79. [79]

    Bias adjusted precipitation threat scores

    author Mesinger, F. Bias adjusted precipitation threat scores. journal Advances in Geosciences volume 16 , pages 137--142 ( year 2008 )

    2008

  80. [80]

    author Schaefer, J. T. The critical success index as an indicator of warning skill. journal Weather and Forecasting volume 5 , pages 570--575 ( year 1990 )

    1990

Showing first 80 references.