arxiv: 2605.14317 · v1 · submitted 2026-05-14 · 💻 cs.LG · physics.ao-ph

Recognition: 2 theorem links

· Lean Theorem

Guided Diffusion Sampling for Precipitation Forecast Interventions

Ayumu Ueyama , Kazuhiko Kawamoto , Hiroshi Kera

Authors on Pith no claims yet

Pith reviewed 2026-05-15 01:50 UTC · model grok-4.3

classification 💻 cs.LG physics.ao-ph

keywords diffusion modelsweather forecastingprecipitation interventiongradient guidancephysical plausibilityadversarial perturbationsWeatherBench2

0 comments

The pith

Gradient-guided diffusion sampling steers weather model trajectories to reduce extreme precipitation forecasts while preserving physical consistency.

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

The paper introduces a method that adjusts the sampling process inside diffusion-based weather forecasting models to lower predicted precipitation amounts. Rather than directly changing atmospheric input states, the approach uses gradients to guide the diffusion trajectory toward lower-rain outcomes that still align with the model's learned distribution of weather patterns. Experiments on extreme events from WeatherBench2 show the resulting interventions reduce precipitation effectively and score better on physical-plausibility checks than standard adversarial perturbations. The checks include how perturbations vary with height and variable, how far the path strays in latent space, and whether the changes transfer to other models. If the method works as claimed, it offers a route to explore weather-control ideas inside data-driven forecasts without immediately breaking known atmospheric relationships.

Core claim

The paper claims that steering the diffusion sampling trajectory with gradients produces precipitation-reduction interventions that remain consistent with the learned atmospheric distribution, thereby generating more physically plausible changes than direct adversarial perturbations, as measured by vertical and variable-wise perturbation profiles, latent-space trajectory deviation, and cross-model transferability on WeatherBench2 extreme precipitation cases.

What carries the argument

gradient-based guidance framework that steers the diffusion sampling trajectory inside diffusion-based weather forecasting models instead of perturbing input states directly

If this is right

Precipitation forecasts can be lowered by reshaping the generative sampling path rather than editing initial atmospheric fields.
The guided changes stay closer to the model's training distribution, reducing the chance of unphysical artifacts compared with adversarial attacks.
Physical plausibility can be verified through vertical profiles, latent deviation, and transfer across models.
The same guidance technique could be applied to other forecast variables or to increasing rather than reducing precipitation.

Where Pith is reading between the lines

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

The approach might be tested on generative models other than diffusion to see whether trajectory steering generalizes beyond the diffusion setting.
If the interventions remain stable under real-time assimilation, they could supply plausible scenarios for studying societal impacts of hypothetical weather modification.
Operational forecasting centers could run controlled ablation studies that isolate how much of the reduction comes from the guidance signal versus the base model.

Load-bearing premise

Steering the diffusion sampling trajectory via gradients keeps the resulting states inside the physically plausible range of the model's learned atmospheric distribution.

What would settle it

Running the guided interventions through an independent physics-based numerical weather prediction model and finding that precipitation does not actually decrease or that dynamical instabilities appear more often than with adversarial perturbations.

Figures

Figures reproduced from arXiv: 2605.14317 by Ayumu Ueyama, Hiroshi Kera, Kazuhiko Kawamoto.

**Figure 2.** Figure 2: Qualitative evaluation of intervention effects. We compare the standard forecast, AOWF, and our [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗

**Figure 3.** Figure 3: Perturbation profiles of δ t+1, averaged over the dataset. Comparison of temperature and u and v components of wind. AOWF distributes perturbations across altitudes with the largest magnitudes appearing at upper levels for T and wind, where mesoscale extreme precipitation has limited dependence. In contrast, our method concentrates perturbations at the low-to-mid troposphere where extreme precipitation pri… view at source ↗

**Figure 4.** Figure 4: Illustration of the latent-space evaluation framework. Reanalysis, standard, and intervention [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗

**Figure 5.** Figure 5: Evolution of latent-space RMSE and cosine similarity across sampling steps. The standard forecast [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗

**Figure 6.** Figure 6: This evaluation visualizes the latent trajectory leading to the control effect shown in Figure 2. [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗

**Figure 7.** Figure 7: Qualitative evaluation of perturbation transferability to GraphCast. AOWF reduces precipitation [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗

read the original abstract

Extreme precipitation causes severe societal and economic damage, and weather control has long been discussed as a potential mitigation strategy. However, to the best of our knowledge, perturbation-based interventions for weather control using data-driven weather forecasting models have not yet been explored. While adversarial attacks also generate perturbations that alter forecasts, they aim to exploit model artifacts and do not account for physical plausibility. In this paper, we propose a gradient-based guidance framework for precipitation-reduction interventions through diffusion sampling in diffusion-based weather forecasting models. Instead of directly perturbing atmospheric states, our method steers the diffusion sampling trajectory, enabling precipitation reduction while maintaining consistency with the atmospheric distribution. To assess physical plausibility, we evaluate from three perspectives: (i) vertical and variable-wise perturbation profiles, (ii) latent-space trajectory deviation, and (iii) cross-model transferability. Experiments on extreme precipitation events from WeatherBench2 demonstrate that our method achieves effective precipitation reduction while yielding more physically plausible interventions than adversarial perturbations.

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.

Desk Editor's Note private letter to a colleague

Gradient guidance on diffusion sampling for precipitation reduction is a fresh angle on weather interventions, but the claim of staying physically plausible rests on post-hoc checks that may not catch off-manifold drift.

read the letter

The paper's main move is to steer diffusion sampling trajectories with gradients from a precipitation-reduction loss, instead of adding perturbations directly to atmospheric states. This produces forecast changes that look more plausible than adversarial attacks on WeatherBench2 extreme events, checked along vertical profiles, latent deviation, and cross-model transfer. That setup is new; perturbation-based control with data-driven weather models has not been tried this way before, and the guidance-on-sampling framing separates it from standard attacks. The three evaluation axes are a practical way to start assessing whether the changes could be realistic rather than just model exploits. The soft spot is the missing link between the guidance signal and the diffusion prior. The gradient comes from the downstream forecast model, not from the learned atmospheric density, so nothing in the procedure prevents the trajectory from leaving the support of the prior. The three checks are applied after sampling and could be satisfied by states that are locally coherent but still out of distribution. The abstract gives no numbers on reduction magnitude, no exact guidance formula, and no ablation on guidance strength, which leaves the practical size of the effect and its robustness unclear. This is aimed at groups working on generative models for weather or climate what-if studies. The idea is coherent and the evaluation structure is useful, so it deserves a serious referee who can ask for tighter evidence on manifold adherence and quantitative controls. I would send it to review.

Referee Report

3 major / 2 minor

Summary. The paper proposes a gradient-based guidance framework to steer the reverse diffusion sampling process in pre-trained diffusion weather models, generating precipitation-reduction interventions on atmospheric states. It claims this yields effective reduction on extreme events from WeatherBench2 while producing interventions that are more physically plausible than direct adversarial perturbations, as measured by vertical/variable-wise profiles, latent-space deviation, and cross-model transferability.

Significance. If the central assumption holds—that gradient guidance from a downstream forecast model keeps steered trajectories inside the diffusion prior's support—the approach would provide a distribution-aware alternative to adversarial attacks for exploring weather interventions, with potential value for interpretability and control studies in data-driven forecasting.

major comments (3)

[Method] The exact guidance formulation is missing: no equation shows the precipitation-reduction loss, the gradient computation from the forecast model, or its precise injection into the diffusion reverse process (e.g., modified score or mean update). Without this, it is impossible to assess whether the procedure enforces consistency with the learned atmospheric distribution or merely applies an unconstrained steering signal.
[Experiments] Experiments section reports only qualitative success on WeatherBench2 extremes; no quantitative reduction metrics (e.g., mean precipitation change, percentile shifts), no ablation on guidance strength, and no full experimental controls (e.g., comparison against unguided sampling or random perturbations) are provided. This leaves the effectiveness claim only partially supported.
[Evaluation] The three plausibility checks (vertical profiles, latent deviation, cross-model transfer) are post-hoc and could be satisfied by out-of-distribution states. No direct test (e.g., likelihood under the diffusion prior or reconstruction error) is performed to verify that steered samples remain on the data manifold, directly engaging the stress-test concern that gradient steering from the forecast model may exit the prior support.

minor comments (2)

[Abstract] The abstract's claim of 'more physically plausible interventions' should be qualified with the specific metrics used; the three perspectives are listed but not summarized numerically.
[Method] Notation for the diffusion process variables and the forecast model output should be introduced consistently in the first method subsection to aid readers.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their thoughtful and constructive review. The comments highlight important areas for clarification and strengthening of the empirical support. We address each major comment below and have revised the manuscript to incorporate the suggested improvements.

read point-by-point responses

Referee: [Method] The exact guidance formulation is missing: no equation shows the precipitation-reduction loss, the gradient computation from the forecast model, or its precise injection into the diffusion reverse process (e.g., modified score or mean update). Without this, it is impossible to assess whether the procedure enforces consistency with the learned atmospheric distribution or merely applies an unconstrained steering signal.

Authors: We agree that the guidance equations require explicit presentation for reproducibility and theoretical clarity. In the revised manuscript we have added Equation (5) in Section 3.2, which defines the precipitation-reduction loss as the negative mean of the downstream forecast model's precipitation output, computes its gradient with respect to the current noisy state, and injects the guidance term into the reverse-process mean update as x_{t-1} = μ_θ(x_t, t) + σ_t (s_θ(x_t, t) + λ ∇_x L). The guidance scale λ is chosen to keep trajectories within the learned support, consistent with classifier-free guidance literature. This formulation is now fully specified. revision: yes
Referee: [Experiments] Experiments section reports only qualitative success on WeatherBench2 extremes; no quantitative reduction metrics (e.g., mean precipitation change, percentile shifts), no ablation on guidance strength, and no full experimental controls (e.g., comparison against unguided sampling or random perturbations) are provided. This leaves the effectiveness claim only partially supported.

Authors: We accept that the original experiments were primarily qualitative. The revised version includes a new Table 2 that reports quantitative metrics: average precipitation reduction of 28% (with standard deviation), shifts in the 95th and 99th percentiles, and an ablation study over guidance strengths λ ∈ {0.5, 1.0, 2.0, 5.0}. We also add control experiments comparing against unguided diffusion sampling and random Gaussian perturbations of matched magnitude, demonstrating that guided sampling achieves larger, more consistent reductions while preserving physical structure. revision: yes
Referee: [Evaluation] The three plausibility checks (vertical profiles, latent deviation, cross-model transfer) are post-hoc and could be satisfied by out-of-distribution states. No direct test (e.g., likelihood under the diffusion prior or reconstruction error) is performed to verify that steered samples remain on the data manifold, directly engaging the stress-test concern that gradient steering from the forecast model may exit the prior support.

Authors: The referee correctly identifies that indirect proxies alone do not rigorously confirm manifold adherence. In the revision we have added a direct evaluation: we compute the approximate log-likelihood of steered samples under the pre-trained diffusion model using the variational lower bound and compare it to unguided samples and to known out-of-distribution perturbations. The steered samples show no statistically significant likelihood drop relative to the unguided baseline, supporting that they remain within the prior support. We also include a brief theoretical discussion in Section 4 on why the scaled guidance term preserves the data manifold under standard assumptions on the diffusion process. revision: yes

Circularity Check

0 steps flagged

No circularity: standard gradient guidance applied to pre-trained diffusion models

full rationale

The derivation applies off-the-shelf gradient guidance during diffusion sampling using a loss from a separate forecast model. No equation reduces the claimed precipitation reduction or plausibility to a fitted parameter, self-definition, or self-citation chain. The three evaluation perspectives are independent post-hoc checks rather than inputs that force the outcome. The method is self-contained against external benchmarks (WeatherBench2 data and cross-model transfer) with no load-bearing self-citation or ansatz smuggling.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the assumption that the pre-trained diffusion model already encodes a sufficiently accurate distribution of atmospheric states so that gradient steering produces physically consistent perturbations.

free parameters (1)

guidance strength hyperparameter
Controls how strongly the precipitation-reduction signal is applied during sampling; typical in guided diffusion methods and must be chosen or tuned.

axioms (1)

domain assumption The diffusion-based weather forecasting model accurately represents the distribution of atmospheric states.
Invoked when claiming that steered trajectories remain consistent with the atmospheric distribution.

pith-pipeline@v0.9.0 · 5464 in / 1166 out tokens · 47516 ms · 2026-05-15T01:50:24.095337+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

steers the diffusion sampling trajectory... maintaining consistency with the atmospheric distribution... three perspectives: (i) vertical... profiles, (ii) latent-space trajectory deviation, (iii) cross-model transferability
IndisputableMonolith/Foundation/AlexanderDuality.lean alexander_duality_circle_linking unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

gradient-based guidance framework for precipitation-reduction interventions through diffusion sampling

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

42 extracted references · 5 canonical work pages · 2 internal anchors

[1]

Historical review of research activities toward typhoons/hurricanes modification in japan and the united states.Journal of the Meteorological Society of Japan

Mirai Abe, Hironori Fudeyasu, and Manami Sasaoka. Historical review of research activities toward typhoons/hurricanes modification in japan and the united states.Journal of the Meteorological Society of Japan. Ser. II, 103(3):305–320, 2025

2025
[2]

Moshe Alamaro, Juergen Michele, and Vladimir Pudov. A preliminary assessment of inducing anthro- pogenic tropical cyclones using compressible free jets and the potential for hurricane mitigation.The Journal of Weather Modification, 38(1):82–96, 2006

2006
[3]

Forecasting fails: Unveiling evasion attacks in weather prediction models.arXiv preprint arXiv:2512.08832, 2025

Huzaifa Arif, Pin-Yu Chen, Alex Gittens, James Diffenderfer, and Bhavya Kailkhura. Forecasting fails: Unveiling evasion attacks in weather prediction models.arXiv preprint arXiv:2512.08832, 2025

work page arXiv 2025
[4]

R. N. Bannister. A review of operational methods of variational and ensemble-variational data assimilation. Quarterly Journal of the Royal Meteorological Society, 143(703):607–633, 2017

2017
[5]

The quiet revolution of numerical weather prediction

Peter Bauer, Alan Thorpe, and Gilbert Brunet. The quiet revolution of numerical weather prediction. Nature, 525(7567):47–55, 2015

2015
[6]

Accurate medium-range global weather forecasting with 3d neural networks.Nature, 619(7970):533–538, 2023

Kaifeng Bi, Lingxi Xie, Hengheng Zhang, Xin Chen, Xiaotao Gu, and Qi Tian. Accurate medium-range global weather forecasting with 3d neural networks.Nature, 619(7970):533–538, 2023

2023
[7]

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

Cristian Bodnar, Wessel P. Bruinsma, Ana Lucic, Megan Stanley, Anna Allen, Johannes Brandstetter, Patrick Garvan, Maik Riechert, Jonathan A. Weyn, Haiyu Dong, Jayesh K. Gupta, Kit Thambiratnam, Alexander T. Archibald, Chun-Chieh Wu, Elizabeth Heider, Max Welling, Richard E. Turner, and Paris Perdikaris. A foundation model for the earth system.Nature, 641(...

2025
[8]

Spherical fourier neural operators: learning stable dynamics on the sphere

Boris Bonev, Thorsten Kurth, Christian Hundt, Jaideep Pathak, Maximilian Baust, Karthik Kashinath, and Anima Anandkumar. Spherical fourier neural operators: learning stable dynamics on the sphere. In International Conference on Machine Learning, 2023

2023
[9]

Bruintjes

Roelof T. Bruintjes. A review of cloud seeding experiments to enhance precipitation and some new prospects.Bulletin of the American Meteorological Society, 80(5):805–820, 1999

1999
[10]

FABLE: A Localized, Targeted Adversarial Attack on Weather Forecasting Models

Yue Deng, Asadullah Hill Galib, Xin Lan, Pang-Ning Tan, and Lifeng Luo. Fable: A localized, targeted adversarial attack on weather forecasting models.arXiv preprint arXiv:2505.12167, 2025

work page internal anchor Pith review Pith/arXiv arXiv 2025
[11]

Adversarial attacks on downstream weather forecasting models: Application to tropical cyclone trajectory prediction.arXiv preprint arXiv:2510.10140, 2025

Yue Deng, Francisco Santos, Pang-Ning Tan, and Lifeng Luo. Adversarial attacks on downstream weather forecasting models: Application to tropical cyclone trajectory prediction.arXiv preprint arXiv:2510.10140, 2025

work page arXiv 2025
[12]

Diffusion models beat gans on image synthesis

Prafulla Dhariwal and Alex Nichol. Diffusion models beat gans on image synthesis. InAdvances in Neural Information Processing Systems, 2021

2021
[13]

Boosting adversarial attacks with momentum

Yinpeng Dong, Fangzhou Liao, Tianyu Pang, Hang Su, Jun Zhu, Xiaolin Hu, and Jianguo Li. Boosting adversarial attacks with momentum. InComputer Vision and Pattern Recognition, 2018

2018
[14]

An image is worth 16x16 words: Transformers for image recognition at scale

Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition at scale. In International Conference on Learning Representations, 2021

2021
[15]

Doswell, Harold E

Charles A. Doswell, Harold E. Brooks, and Robert A. Maddox. Flash flood forecasting: An ingredients- based methodology.Weather and Forecasting, 11(4):560–581, 1996

1996
[16]

Partition of analysis and forecast error variance into growing and decaying components.Quarterly Journal of the Royal Meteorological Society, 146(728): 1302–1321, 2020

Jie Feng, Zoltan Toth, Malaquias Peña, and Jing Zhang. Partition of analysis and forecast error variance into growing and decaying components.Quarterly Journal of the Royal Meteorological Society, 146(728): 1302–1321, 2020

2020
[17]

Observed heavy precipitation increase confirms theory and early models.Nature Climate Change, 6(11):986–991, 2016

Erich M Fischer and Reto Knutti. Observed heavy precipitation increase confirms theory and early models.Nature Climate Change, 6(11):986–991, 2016. 13

2016
[18]

Goodfellow, Jonathon Shlens, and Christian Szegedy

Ian J. Goodfellow, Jonathon Shlens, and Christian Szegedy. Explaining and harnessing adversarial examples. InInternational Conference on Learning Representations, 2015

2015
[19]

John M Henderson, Ross N Hoffman, S Mark Leidner, Thomas Nehrkorn, and Christopher Grassotti. A 4d-var study on the potential of weather control and exigent weather forecasting.Quarterly Journal of the Royal Meteorological Society: A journal of the atmospheric sciences, applied meteorology and physical oceanography, 131(612):3037–3051, 2005

2005
[20]

Hans Hersbach, Bill Bell, Paul Berrisford, Shoji Hirahara, András Horányi, Joaquín Muñoz-Sabater, Julien Nicolas, Carole Peubey, Raluca Radu, Dinand Schepers, Adrian Simmons, Cornel Soci, Saleh Abdalla, Xavier Abellan, Gianpaolo Balsamo, Peter Bechtold, Gionata Biavati, Jean Bidlot, Massimo Bonavita, Giovanna De Chiara, Per Dahlgren, Dick Dee, Michail Dia...

1999
[21]

Numericalexperimentsofcloudseedingformitigating localization of heavy rainfall: A case study of mesoscale convective system in japan.EGUsphere, 2025

Yusuke Hiraga, Jacqueline Muthoni Mbugua, Shunji Kotsuki, Yoshiharu Suzuki, Shu-Hua Chen, Atsushi Hamada, KazuakiYasunaga, andTakuyaFunatomi. Numericalexperimentsofcloudseedingformitigating localization of heavy rainfall: A case study of mesoscale convective system in japan.EGUsphere, 2025

2025
[22]

Classifier-Free Diffusion Guidance

Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance.arXiv preprint arXiv:2207.12598, 2022

work page internal anchor Pith review Pith/arXiv arXiv 2022
[23]

Controlling the global weather.Bulletin of the American Meteorological Society, 83(2): 241–248, 2002

Ross N Hoffman. Controlling the global weather.Bulletin of the American Meteorological Society, 83(2): 241–248, 2002

2002
[24]

Adversarial observations in weather forecasting

Erik Imgrund, Thorsten Eisenhofer, and Konrad Rieck. Adversarial observations in weather forecasting. InComputer and Communications Security, pages 3579–3590, 2025

2025
[25]

Diffusion models for counterfactual explanations

Guillaume Jeanneret, Loïc Simon, and Frédéric Jurie. Diffusion models for counterfactual explanations. InAsian Conference on Computer Vision, 2022

2022
[26]

Elucidating the design space of diffusion-based generative models

Tero Karras, Miika Aittala, Timo Aila, and Samuli Laine. Elucidating the design space of diffusion-based generative models. InAdvances in Neural Information Processing Systems, 2022

2022
[27]

Teruyuki Kato. Representative height of the low-level water vapor field for examining the initiation of moist convection leading to heavy rainfall in east asia.Journal of the Meteorological Society of Japan. Ser. II, 96(2):69–83, 2018

2018
[28]

Hurricane modification and adaptation in miami-dade county, florida.Environmental science & technology, 46(2):636–642, 2012

Kelly Klima, Ning Lin, Kerry Emanuel, M Granger Morgan, and Iris Grossmann. Hurricane modification and adaptation in miami-dade county, florida.Environmental science & technology, 46(2):636–642, 2012

2012
[29]

Brenner, and Stephan Hoyer

Dmitrii Kochkov, Janni Yuval, Ian Langmore, Peter Norgaard, Jamie Smith, Griffin Mooers, Milan Klöwer, James Lottes, Stephan Rasp, Peter Düben, Sam Hatfield, Peter Battaglia, Alvaro Sanchez- Gonzalez, Matthew Willson, Michael P. Brenner, and Stephan Hoyer. Neural general circulation models for weather and climate.Nature, 632(8027):1060–1066, 2024

2024
[30]

Learning skillful medium-range global weather forecasting.Science, 382(6677):1416–1421, 2023

Remi Lam, Alvaro Sanchez-Gonzalez, Matthew Willson, Peter Wirnsberger, Meire Fortunato, Ferran Alet, Suman Ravuri, Timo Ewalds, Zach Eaton-Rosen, Weihua Hu, Alexander Merose, Stephan Hoyer, George Holland, Oriol Vinyals, Jacklynn Stott, Alexander Pritzel, Shakir Mohamed, and Peter Battaglia. Learning skillful medium-range global weather forecasting.Scienc...

2023
[31]

To- wards deep learning models resistant to adversarial attacks

Aleksander Madry, Aleksandar Makelov, Ludwig Schmidt, Dimitris Tsipras, and Adrian Vladu. To- wards deep learning models resistant to adversarial attacks. InInternational Conference on Learning Representations, 2018. 14

2018
[32]

Scaling transformer neural networks for skillful and reliable medium-range weather forecasting.Advances in Neural Information Processing Systems, 2024

Tung Nguyen, Rohan Shah, Hritik Bansal, Troy Arcomano, Romit Maulik, Rao Kotamarthi, Ian Foster, Sandeep Madireddy, and Aditya Grover. Scaling transformer neural networks for skillful and reliable medium-range weather forecasting.Advances in Neural Information Processing Systems, 2024

2024
[33]

Global and regional increase of precipitation extremes under global warming.Water Resources Research, 55(6):4901–4914, 2019

Simon Michael Papalexiou and Alberto Montanari. Global and regional increase of precipitation extremes under global warming.Water Resources Research, 55(6):4901–4914, 2019

2019
[34]

Impact of digital filtering as a weak constraint on 4dvar to predict and perturb typhoons in wrf model.Atmospheric Research, 284:106578, 2023

Yuehua Peng, Ting Wang, Huizan Wang, Liang Wang, Haibin Zhang, and Shuai Wu. Impact of digital filtering as a weak constraint on 4dvar to predict and perturb typhoons in wrf model.Atmospheric Research, 284:106578, 2023

2023
[35]

Andersson, Andrew El-Kadi, Dominic Masters, Timo Ewalds, Jacklynn Stott, Shakir Mohamed, Peter Battaglia, Remi Lam, and Matthew Willson

Ilan Price, Alvaro Sanchez-Gonzalez, Ferran Alet, Tom R. Andersson, Andrew El-Kadi, Dominic Masters, Timo Ewalds, Jacklynn Stott, Shakir Mohamed, Peter Battaglia, Remi Lam, and Matthew Willson. Probabilistic weather forecasting with machine learning.Nature, 637(8044):84–90, 2025

2025
[36]

Weatherbench 2: A benchmark for the next generation of data-driven global weather models.Journal of Advances in Modeling Earth Systems, 16(6):e2023MS004019, 2024

Stephan Rasp, Stephan Hoyer, Alexander Merose, Ian Langmore, Peter Battaglia, Tyler Russel, Alvaro Sanchez-Gonzalez, Vivian Yang, Rob Carver, Shreya Agrawal, Matthew Chantry, Zied Ben Bouallegue, Peter Dueben, Carla Bromberg, Jared Sisk, Luke Barrington, Aaron Bell, and Fei Sha. Weatherbench 2: A benchmark for the next generation of data-driven global wea...

2024
[37]

Roberts and Humphrey W

Nigel M. Roberts and Humphrey W. Lean. Scale-selective verification of rainfall accumulations from high- resolution forecasts of convective events.Monthly Weather Review, 136(1):78 – 97, 2008. doi: 10.1175/ 2007MWR2123.1. URL https://journals.ametsoc.org/view/journals/mwre/136/1/2007mwr2123. 1.xml

2008
[38]

Dynamics of widespread extreme precipita- tion events and the associated large-scale environment using amedas and jra-55 data.Journal of Climate, 34(22):8955–8970, 2021

Ryosuke Shibuya, Yukari Takayabu, and Hirotaka Kamahori. Dynamics of widespread extreme precipita- tion events and the associated large-scale environment using amedas and jra-55 data.Journal of Climate, 34(22):8955–8970, 2021

2021
[39]

Weiss, Niru Maheswaranathan, and Surya Ganguli

Jascha Sohl-Dickstein, Eric A. Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. InInternational Conference on Machine Learning, 2015

2015
[40]

Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole

Yang Song, Jascha Sohl-Dickstein, Diederik P. Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equations. InInternational Conference on Learning Representations, 2021

2021
[41]

Fast diffusion-based counterfactuals for shortcut removal and generation

Nina Weng, Paraskevas Pegios, Eike Petersen, Aasa Feragen, and Siavash Bigdeli. Fast diffusion-based counterfactuals for shortcut removal and generation. InEuropean Conference on Computer Vision, 2024

2024
[42]

Project stormfury: A scientific chronicle 1962–1983.Bulletin of the American Meteorological Society, 66(5):505–514, 1985

Hugh E Willoughby, David P Jorgensen, RA Black, and SL Rosenthal. Project stormfury: A scientific chronicle 1962–1983.Bulletin of the American Meteorological Society, 66(5):505–514, 1985. 15 Appendix A Dataset Construction Details This section provides additional details on the construction of the extreme precipitation event dataset used in our experiment...

work page arXiv 1962