pith. machine review for the scientific record. sign in

arxiv: 2604.18906 · v1 · submitted 2026-04-20 · ⚛️ physics.ao-ph

Recognition: unknown

Instability-Aware Steering of an Extreme Atmospheric River in an AI Weather Foundation Model

Moyan Liu, Qin Huang, Upmanu Lall

Pith reviewed 2026-05-10 02:31 UTC · model grok-4.3

classification ⚛️ physics.ao-ph
keywords atmospheric riverAI weather modelLyapunov exponentsweather controlextreme eventscloud seedinginstability
0
0 comments X

The pith

Small instability-guided perturbations can steer an extreme atmospheric river to reduce its landfall intensity

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

The paper tests whether small, targeted changes can alter the trajectory and strength of an extreme atmospheric river using an AI weather model. Researchers locate sensitive areas with mathematical instability measures and introduce idealized cloud seeding to mimic heat release. In a California storm case, these changes shift moisture transport downstream and can weaken the event at landfall when conditions align. This approach hints at using the atmosphere's sensitivity for managing extreme weather risks.

Core claim

Perturbations induce coherent downstream shifts in moisture transport, reducing intensity at landfall under favorable kinematic conditions. The response is nonlinear and contingent on the local flow geometry.

What carries the argument

Finite-time Lyapunov exponents and jet-eddy interaction criteria for identifying intervention locations, applied with an idealized cloud-seeding operator in the Aurora AI model.

If this is right

  • The effectiveness of steering depends on the local flow geometry and kinematic conditions.
  • Nonlinear responses imply that perturbation outcomes vary and require careful selection.
  • This method offers a computational way to explore risk mitigation for extreme events.
  • Similar techniques might apply to other weather phenomena with identifiable instabilities.

Where Pith is reading between the lines

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

  • AI weather foundation models can act as efficient testbeds for investigating potential weather control strategies.
  • Future work could involve comparing model predictions with high-resolution traditional simulations or observations to check realism.
  • The chaotic sensitivity of the atmosphere might allow for minimal interventions to achieve meaningful risk reductions if the approach generalizes.

Load-bearing premise

The idealized cloud-seeding operator and the Aurora model's response accurately represent real atmospheric dynamics under small perturbations.

What would settle it

If applying similar perturbations in a different weather model or to observed events does not produce coherent shifts in moisture transport, the steering effect would be called into question.

Figures

Figures reproduced from arXiv: 2604.18906 by Moyan Liu, Qin Huang, Upmanu Lall.

Figure 1
Figure 1. Figure 1: Selected Perturbation Sites for Atmospheric River Intervention 3.2 Perturbations Result Each candidate perturbation is applied once over a single 6-hour model step within a 300 km radius, targeting the lower-to-mid troposphere (925–700 hPa). In this config￾uration, an idealized 30% freeze efficiency is applied to the local vapor field, with 50% of the frozen condensate removed as precipitation. Within the … view at source ↗
Figure 2
Figure 2. Figure 2: The time evolution of Z500, 500 hPa wind, and IVT anomaly for two seeding experiments. (a) Favorable site (⋆, 30.8◦N, 166.5◦E): the seeding location lies within a cycloni￾cally shearing, jet-adjacent flow regime with high FTLE. The imposed perturbation is amplified through the flow, leading to a reduction of IVT in the California target region during landfall. (b) Unfavorable site (×, 28.5◦N, 171.8◦E): the… view at source ↗
Figure 3
Figure 3. Figure 3: Time evolution and spatial structure of IVT response to targeted perturbations during an atmospheric river event. Top panel: Area-mean IVT over the target region for ERA5 (observed), Aurora control, and Aurora perturbed simulations, with the shaded region indicat￾ing the landfall window. Middle row: Percent change in IVT for AR pixels (250 kg m−1 s −1 ) at selected forecast steps. Bottom row: Percent chang… view at source ↗
read the original abstract

Advances in deep learning methods for weather forecasting are creating opportunities to computationally explore the potential for steering or control of extreme weather trajectories for societal risk reduction. We present initial investigations into the feasibility of redirecting extreme atmospheric rivers (ARs) through small, instability-aware perturbations. Using the Aurora AI weather foundation model, we identify sensitive upstream locations using finite-time Lyapunov exponents and jet-eddy interaction criteria. We apply an idealized cloud-seeding operator that mimics latent heat release to assess whether these Lyapunov-guided interventions can influence downstream evolution. In a case study of a severe California AR, perturbations induce coherent downstream shifts in moisture transport, reducing intensity at landfall under favorable kinematic conditions. The response is nonlinear and contingent on the local flow geometry. These initial results suggest that the atmosphere's intrinsic chaotic sensitivity could be leveraged for dynamical control, offering a new research direction for extreme event risk mitigation.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The manuscript explores the feasibility of steering extreme atmospheric rivers using small, instability-aware perturbations in the Aurora AI weather foundation model. Guided by finite-time Lyapunov exponents and jet-eddy interaction criteria, an idealized cloud-seeding operator is applied at sensitive upstream locations in a California AR case study. The authors report that these perturbations produce coherent downstream shifts in moisture transport and reduce intensity at landfall under favorable kinematic conditions, with the response being nonlinear and dependent on local flow geometry.

Significance. If the central claims hold under further scrutiny, this work opens a promising new direction for computationally exploring dynamical control of extreme weather events via AI foundation models, leveraging the atmosphere's intrinsic chaotic sensitivity for societal risk reduction. The approach of combining Lyapunov-based targeting with idealized perturbations in a data-driven model is timely and innovative, providing a proof-of-concept that could inspire hybrid AI-physics studies.

major comments (3)
  1. [Abstract and Results] Abstract and Results: The claims that perturbations 'induce coherent downstream shifts in moisture transport, reducing intensity at landfall' are stated qualitatively without supporting quantitative metrics (e.g., changes in IVT, precipitation totals, or coherence measures), error analysis, or statistical significance testing, making the magnitude and robustness of the effect difficult to assess.
  2. [Methods and Validation] Methods and Validation: The idealized cloud-seeding operator and Aurora model's nonlinear response to small perturbations are not validated against physics-based NWP models, ensemble forecasts, or observations; without such cross-checks or perturbation-size sensitivity tests, it remains unclear whether the reported coherence reflects real atmospheric dynamics or model-specific artifacts.
  3. [Case study] Case study: The contingency of the response on 'favorable kinematic conditions' and 'local flow geometry' is noted but not systematically explored across multiple cases or flow regimes, limiting the generalizability of the steering feasibility conclusion.
minor comments (2)
  1. [Methods] Clarify the precise computation of finite-time Lyapunov exponents and jet-eddy criteria within the Aurora model, including any approximations or hyperparameters used.
  2. [Figures] Improve figure captions to explicitly describe perturbation locations, scales, and the specific fields shown (e.g., moisture transport anomalies).

Simulated Author's Rebuttal

3 responses · 1 unresolved

We thank the referee for their insightful comments and positive evaluation of the potential impact of our work. We have revised the manuscript to incorporate quantitative metrics, additional sensitivity analyses, and an expanded case discussion to address the concerns raised, while preserving the initial proof-of-concept nature of the study.

read point-by-point responses

  1. Referee: [Abstract and Results] Abstract and Results: The claims that perturbations 'induce coherent downstream shifts in moisture transport, reducing intensity at landfall' are stated qualitatively without supporting quantitative metrics (e.g., changes in IVT, precipitation totals, or coherence measures), error analysis, or statistical significance testing, making the magnitude and robustness of the effect difficult to assess.

    Authors: We agree that providing quantitative support is essential for assessing the effect's magnitude. In the revised version, we have included specific metrics such as the reduction in peak IVT (15% in the example case), changes in 24-hour precipitation accumulation, and a coherence metric defined as the correlation coefficient of the perturbation-induced moisture flux anomaly with the control. Error analysis is now presented using standard deviation from an ensemble of 10 perturbation realizations, and we discuss the challenges of statistical significance testing in this single-event context, including bootstrap-derived intervals. revision: yes

  2. Referee: [Methods and Validation] Methods and Validation: The idealized cloud-seeding operator and Aurora model's nonlinear response to small perturbations are not validated against physics-based NWP models, ensemble forecasts, or observations; without such cross-checks or perturbation-size sensitivity tests, it remains unclear whether the reported coherence reflects real atmospheric dynamics or model-specific artifacts.

    Authors: This is a valid concern for establishing the physical relevance of the results. We have added perturbation-size sensitivity tests showing that the steering effect remains coherent for heating perturbations up to 10% of the background latent heat release rate, beyond which the response becomes disorganized. Full validation against NWP models or observations is not included in this revision due to the high computational cost and the focus on AI model exploration; however, we have added a dedicated paragraph in the Discussion section outlining plans for such cross-validation in future work and acknowledging the possibility of model-specific behaviors. revision: partial

  3. Referee: [Case study] Case study: The contingency of the response on 'favorable kinematic conditions' and 'local flow geometry' is noted but not systematically explored across multiple cases or flow regimes, limiting the generalizability of the steering feasibility conclusion.

    Authors: We recognize the limitation in generalizability. To partially address this, the revised manuscript includes a brief comparison with a second AR event from a different year, where similar upstream perturbations under analogous jet-eddy configurations produce comparable downstream shifts. A comprehensive exploration across multiple flow regimes would require a larger ensemble of cases and is noted as an important avenue for subsequent research in the Conclusions section. revision: partial

standing simulated objections not resolved
  • Full cross-validation of the idealized operator and results against physics-based numerical weather prediction models, as this exceeds the scope and resources of the current proof-of-concept study.

Circularity Check

0 steps flagged

No circularity: results are direct outputs of numerical experiments in an external AI model

full rationale

The paper conducts case-study simulations inside the Aurora foundation model, applying Lyapunov-guided perturbations via an idealized cloud-seeding operator and reporting the resulting moisture-transport shifts. No closed-form derivation, parameter fitting, or self-referential definition is present; the central claim is simply the observed model response under stated conditions. No self-citation chains, uniqueness theorems, or ansatzes imported from prior author work are invoked as load-bearing steps. The work is therefore self-contained as a numerical exploration and does not reduce any result to its own inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim depends on the fidelity of the Aurora model and the realism of the idealized operator; no free parameters are explicitly named in the abstract.

axioms (1)
  • domain assumption The Aurora AI weather foundation model faithfully reproduces the response of real atmospheric rivers to small upstream perturbations.
    The entire feasibility demonstration rests on this untested assumption about model accuracy.

pith-pipeline@v0.9.0 · 5450 in / 1094 out tokens · 26329 ms · 2026-05-10T02:31:35.809554+00:00 · methodology

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

32 extracted references · 17 canonical work pages

  1. [1]

    2022 , month =

    Clarke, Ben and Otto, Friederike and Stuart-Smith, Rupert and Harrington, Luke , title =. 2022 , month =. doi:10.1088/2752-5295/ac6e7d , url =

    work page doi:10.1088/2752-5295/ac6e7d 2022

  2. [2]

    Climate Extremes are Becoming More Frequent, Co-occurring, and Persistent in Europe , journal =

    Pradhan, Prajal and Seydewitz, Tobias and Zhou, Bin and L. Climate Extremes are Becoming More Frequent, Co-occurring, and Persistent in Europe , journal =. 2022 , doi =

    2022

  3. [3]

    Nature Communications , volume =

    Newman, Robert and Noy, Ilan , title =. Nature Communications , volume =. 2023 , doi =

    2023

  4. [4]

    Annual Review of Resource Economics , volume =

    Ferreira, Susana , title =. Annual Review of Resource Economics , volume =. 2024 , doi =

    2024

  5. [5]

    2021 , month =

    Doss-Gollin, James and Farnham, David J and Lall, Upmanu and Modi, Vijay , title =. 2021 , month =. doi:10.1088/1748-9326/ac0278 , url =

    work page doi:10.1088/1748-9326/ac0278 2021

  6. [6]

    npj Natural Hazards , volume =

    Hwang, Jisun and Lall, Upmanu , title =. npj Natural Hazards , volume =. 2024 , doi =

    2024

  7. [7]

    and Lucic, Ana and Stanley, Megan and Allen, Anna and Brandstetter, Johannes and Garvan, Patrick and Riechert, Maik and Weyn, Jonathan A

    Bodnar, Cristian and Bruinsma, Wessel P. and Lucic, Ana and Stanley, Megan and Allen, Anna and Brandstetter, Johannes and Garvan, Patrick and Riechert, Maik and Weyn, Jonathan A. and Dong, Haiyu and Gupta, Jayesh K. and Thambiratnam, Kit and Archibald, Alexander T. and Wu, Chun-Chieh and Heider, Elizabeth and Welling, Max and Turner, Richard E. and Perdik...

    2025

  8. [8]

    Lang, Simon and Alexe, Mihai and Chantry, Matthew and Dramsch, Jesper and Pinault, Florian and Raoult, Baudouin and Clare, Mariana C. A. and Lessig, Christian and Maier-Gerber, Michael and Magnusson, Linus and Bouall. arXiv preprint arXiv:2406.01465 , year =. doi:10.48550/arXiv.2406.01465 , url =. 2406.01465 , archivePrefix =

    work page doi:10.48550/arxiv.2406.01465

  9. [9]

    and Sun, M., 2025

    Bell, Aaron and Aides, Amit and Helmy, Amr and Muslim, Arbaaz and Barzilai, Aviad and Slobodkin, Aviv and Jaber, Bolous and Schottlander, David and Leifman, George and Paul, Joydeep and Sun, Mimi and Sherman, Nadav and Williams, Natalie and Bjornsson, Per and Lee, Roy and Alcantara, Ruth and Turnbull, Thomas and Shekel, Tomer and Silverman, Vered and Gigi...

    work page doi:10.48550/arxiv.2510.18318

  10. [10]

    arXiv preprint arXiv:2508.09376 , year =

    Huang, Qin and Liu, Moyan and Lall, Upmanu , title =. arXiv preprint arXiv:2508.09376 , year =. doi:10.48550/arXiv.2508.09376 , url =. 2508.09376 , archivePrefix =

    work page doi:10.48550/arxiv.2508.09376

  11. [11]

    EGUsphere , year =

    Liu, Moyan and Huang, Qin and Lall, Upmanu , title =. EGUsphere , year =. doi:10.5194/egusphere-2025-3997 , url =

    work page doi:10.5194/egusphere-2025-3997 2025

  12. [12]

    arXiv preprint arXiv:2510.26005 , year =

    Liu, Moyan and Huang, Qin and Lall, Upmanu , title =. arXiv preprint arXiv:2510.26005 , year =. doi:10.48550/arXiv.2510.26005 , url =. 2510.26005 , archivePrefix =

    work page doi:10.48550/arxiv.2510.26005

  13. [13]

    , title =

    Das, Puja and Posch, August and Barber, Nathan and Hicks, Michael and Duffy, Kate and Vandal, Thomas and Singh, Debjani and van Werkhoven, Katie and Ganguly, Auroop R. , title =. npj Climate and Atmospheric Science , volume =. 2024 , doi =

    2024

  14. [14]

    Probabilistic weather forecasting with machine learning , journal =

    Price, Ilan and S. Probabilistic weather forecasting with machine learning , journal =. 2025 , doi =

    2025

  15. [15]

    and Eiras-Barca, J

    Garaboa-Paz, D. and Eiras-Barca, J. and P\'erez-Mu\ nuzuri, V. , TITLE =. Earth System Dynamics , VOLUME =. 2017 , NUMBER =

    2017

  16. [16]

    arXiv preprint arXiv:2505.08529 , year =

    Zhao, Shan and Xiong, Zhitong and Zhao, Jie and Zhu, Xiao Xiang , title =. arXiv preprint arXiv:2505.08529 , year =. doi:10.48550/arXiv.2505.08529 , url =. 2505.08529 , archivePrefix =

    work page doi:10.48550/arxiv.2505.08529

  17. [17]

    A Multimethod Attribution Analysis of Spain's 2024 Extreme Precipitation Event , journal =

    Barriopedro, David and Jim. A Multimethod Attribution Analysis of Spain's 2024 Extreme Precipitation Event , journal =. 2025 , doi =

    2024

  18. [18]

    The latent heating feedback on the mid-latitude circulation , journal =

    Auestad, H. The latent heating feedback on the mid-latitude circulation , journal =. 2025 , doi =

    2025

  19. [19]

    and Adames Corraliza,

    Lin, Q.-J. and Adames Corraliza,. Hadley Cell instability and its link to tropical depression-type waves , journal =. 2025 , doi =

    2025

  20. [20]

    and Coauthors , title =

    DeFlorio, Michael J. and Coauthors , title =. Bulletin of the American Meteorological Society , volume =. 2024 , doi =

    2024

  21. [21]

    , title =

    Lorenz, Edward N. , title =. Tellus A , volume =. doi:https://doi.org/10.1111/j.1600-0870.1984.tb00230.x , url =. https://onlinelibrary.wiley.com/doi/pdf/10.1111/j.1600-0870.1984.tb00230.x , year =

    work page doi:10.1111/j.1600-0870.1984.tb00230.x 1984

  22. [22]

    Garrido-Perez and Juan Emmanuel Johnson and Ricardo García-Herrera

    David Barriopedro and Bernat Jiménez-Esteve and Soledad Collazo and Jose M. Garrido-Perez and Juan Emmanuel Johnson and Ricardo García-Herrera. A Multimethod Attribution Analysis of Spain’s 2024 Extreme Precipitation Event. Bulletin of the American Meteorological Society. 2025. doi:10.1175/BAMS-D-25-0049.1

    work page doi:10.1175/bams-d-25-0049.1 2024

  23. [23]

    Journal of Geophysical Research: Machine Learning and Computation , volume =

    Sahu, Pankaj Lal and Sandeep, Sukumaran and Kodamana, Hariprasad , title =. Journal of Geophysical Research: Machine Learning and Computation , volume =. doi:https://doi.org/10.1029/2025JH000594 , url =. https://agupubs.onlinelibrary.wiley.com/doi/pdf/10.1029/2025JH000594 , note =

    work page doi:10.1029/2025jh000594

  24. [24]

    2025 , eprint=

    MAUSAM: An Observations-focused assessment of Global AI Weather Prediction Models During the South Asian Monsoon , author=. 2025 , eprint=

    2025

  25. [25]

    and Clark, Peter A

    Dacre, Helen F. and Clark, Peter A. , title =. npj Climate and Atmospheric Science , year =. doi:10.1038/s41612-025-00942-z , url =

    work page doi:10.1038/s41612-025-00942-z

  26. [26]

    Bell and Claudia Langford Brown and Kathryn Conlon and Stephanie Herring and Kenneth E

    Jesse E. Bell and Claudia Langford Brown and Kathryn Conlon and Stephanie Herring and Kenneth E. Kunkel and Jay Lawrimore and George Luber and Carl Schreck and Adam Smith and Christopher Uejio , title =. Journal of the Air & Waste Management Association , volume =. 2018 , publisher =. doi:10.1080/10962247.2017.1401017 , note =

    work page doi:10.1080/10962247.2017.1401017 2018

  27. [27]

    2026 , eprint=

    Evaluating the Predictability of Selected Weather Extremes with Aurora, an AI Weather Forecast Model , author=. 2026 , eprint=

    2026

  28. [28]

    Valerio Lembo and Simona Bordoni and Emanuele Bevacqua and Daniela I. V. Domeisen and Christian L. E. Franzke and Vera M. Galfi and Chaim I. Garfinkel and Christian M. Grams and Assaf Hochman and Roshan Jha and Kai Kornhuber and Frank Kwasniok and Valerio Lucarini and Gabriele Messori and Duncan Pappert and Iago Perez-Fernandez and Jacopo Riboldi and Emma...

    work page doi:10.1175/bams-d-24-0145.1 2024

  29. [29]

    doi: 10.1038/s41612-018-0013-0

    Vitart, Frédéric and Robertson, Andrew W. , title =. npj Climate and Atmospheric Science , year =. doi:10.1038/s41612-018-0013-0 , url =

    work page doi:10.1038/s41612-018-0013-0

  30. [30]

    Daniela I. V. Domeisen and Christopher J. White and Hilla Afargan-Gerstman and Ángel G. Muñoz and Matthew A. Janiga and Frédéric Vitart and C. Ole Wulff and Salomé Antoine and Constantin Ardilouze and Lauriane Batté and Hannah C. Bloomfield and David J. Brayshaw and Suzana J. Camargo and Andrew Charlton-Pérez and Dan Collins and Tim Cowan and Maria del Ma...

    work page doi:10.1175/bams-d-20-0221.1 2022

  31. [31]

    Perez, G. M. P. and Vidale, P. L. and Klingaman, N. P. and Martin, T. C. M. , TITLE =. Weather and Climate Dynamics , VOLUME =. 2021 , NUMBER =

    2021

  32. [32]

    , title =

    Finkel, Justin and O’Gorman, Paul A. , title =. Journal of Advances in Modeling Earth Systems , volume =. doi:https://doi.org/10.1029/2025MS005456 , url =. https://agupubs.onlinelibrary.wiley.com/doi/pdf/10.1029/2025MS005456 , note =

    work page doi:10.1029/2025ms005456