Pith. sign in

REVIEW 4 major objections 5 minor 47 references

Physics-Guided Learning of Meteorological Dynamics for Weather Downscaling and Forecasting

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Physics-guided module sharpens weather forecasts with just 55K parameters.

desk verdict Serious engineering contribution, but the physics-loss benefit is confounded by data augmentation and the physical-consistency claims outrun the evidence. read the letter →

arxiv 2505.14555 v2 pith:K4M2SXXW submitted 2025-05-20 cs.LG cs.AI

Yingtao Luo , Shikai Fang , Binqing Wu , Qingsong Wen , Liang Sun This is my paper
classification cs.LGcs.AI
keywords physics-informedneuralnetworkweatherdownscalingforecastinglatentforceparameterizationPDEdiscoverycontinuouscoordinaterepresentationmodelfine-tuningautomaticdifferentiation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper argues that a weather model's predictions should be held to a learned dynamical law rather than to data alone. It proposes a framework that represents weather as a continuous function of coordinates, uses automatic differentiation to build a library of physical terms, and completes that library with a neural 'latent force' that absorbs processes missing from the dataset. The recovered equation then serves two purposes: it lets the model generate super-resolution weather at arbitrary coordinates, and it acts as a physics regularizer when fine-tuning any pretrained forecasting model. If the approach works as claimed, a 55K-parameter add-on can make existing large models both cheaper and more physically consistent.

What carries the argument

The central object is the parameterized dynamics equation du/dt = Phi(u)Xi + Q_pi(x,y,t), which combines a sparse library of explicit PDE terms (advection, diffusion, etc.) with a latent force neural network that absorbs processes absent from the data. Automatic differentiation of the coordinate-based surrogate f_theta(x,y,t) provides the derivative terms; regression on the residual fits the sparse coefficients Xi; and the residual du/dt - Phi Xi - Q_pi defines the physics loss L_phy. This same equation does double duty: it regularizes the downscaling surrogate during training, and later constrains the outputs of any pretrained forecasting model g_omega via finite-difference derivatives, transferring the recovered dynamics without adding model capacity.

What would settle it

Generate a synthetic dataset from a known PDE (e.g., advection-diffusion with a prescribed time-varying source), train PhyDL-NWP, and compare the recovered coefficient vector Xi and latent force Q_pi against the ground-truth values; if the latent force absorbs terms the library can express, or the recovered coefficients drift with random seeds, the claimed recovery and transfer of physical dynamics fails.

Watch

Extended reading notes

Core claim

PhyDL-NWP claims that the gap between data-driven weather models and physical consistency can be closed by explicitly parameterizing a governing equation of the form du/dt = Phi(u)Xi + Q_pi, where Phi is a library of interpretable PDE terms computed from the predicted fields, Xi is a sparse coefficient vector learned by regression, and Q_pi is a neural network representing unmeasured forces. Training the model against the residual of this equation, through a physics loss added to the data loss, produces a continuous coordinate-based surrogate that supports resolution-free downscaling without paired super-resolution labels, and yields a transferable dynamic constraint. When this constraint is applied to fine-tune a pretrained forecaster, the authors report consistent improvements in RMSE and ACC over the same model without physics guidance, across regional reanalysis, real observational, and global benchmark datasets.

Load-bearing premise

The load-bearing premise is that the combination of a sparse physics library and a neural-network latent force actually recovers the true unmeasured dynamics, rather than the latent force simply memorizing whatever the explicit terms fail to fit.

Editorial extensions

If this is right

  • Downscaling becomes resolution-free: once the surrogate is trained, any coordinate can be queried, so 2x, 4x, or finer outputs come from the same model without paired low- and high-resolution training data.
  • The physics module is an add-on, not a replacement: it fine-tunes ClimaX, FourcastNet, and GraphCast with only 55K parameters and improves their RMSE and ACC, with gains increasing toward seven-day lead times.
  • The learned equations are interpretable: e.g., temperature evolves by 10m-wind advection plus a latent term, making visible which unmeasured processes (vertical motion, diffusivity, sources) are being parameterized.
  • The inference cost advantage is structural: derivatives for the forecasting loss use finite differences on super-resolved output, avoiding per-frame surrogate training, which yields 55 to 170 times faster epochs than standalone models.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the latent force captures genuinely transferable physics rather than dataset-specific residuals, the same parameterization could be reused as a learned closure for subgrid processes in dynamical cores or coupled regional models, a use the paper does not test.
  • The resolution-free property suggests an application beyond post-processing downscaling: adaptive or variable-mesh simulation, where a single continuous function supplies fields at dynamically chosen grid points during integration.
  • A synthetic identifiability test would settle whether Xi and Q_pi separate correctly: generate data from a known PDE with known forcing, fit the framework, and check the recovered coefficients; the real-world experiments do not fully distinguish true equation recovery from residual memorization.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes PhyDL-NWP, a physics-guided framework for weather downscaling and forecasting. In the downscaling stage, a coordinate-based MLP surrogate f_theta is trained to predict weather variables as a continuous function of (x,y,t), while a sparse linear PDE library Phi*Xi and a latent force network Q_pi are jointly learned to explain the temporal tendency du/dt (Eqs. 2-5). In the forecasting stage, the fitted Xi and Q_pi are used to add a physics residual loss beta*Lphy to the fine-tuning of a pretrained forecasting model g_omega, and the surrogate f_theta is used to generate super-resolved training inputs (Eqs. 6-8, Algorithm 1). Experiments on downscaling (Huadong) and forecasting (Ningbo, Ningxia, WeatherBench) report consistent RMSE/ACC improvements over vanilla baselines, and Table 6 compares the learned equations with textbook meteorology.

Significance. The central idea—that a small plug-in module can extract a parameterized PDE from data and then use it as a transferable regularizer for large pretrained forecasters—is appealing and potentially practical. The availability of code, multiple datasets, and the surprisingly small 55K parameter footprint are concrete strengths. If substantiated, the claimed 55-170x speed advantage and the ability to fine-tune arbitrary base models with a physics loss would be a useful engineering contribution. However, the paper currently does not isolate the effect of the physics loss from the simultaneous data augmentation and fine-tuning, and the learned physics is largely represented by a flexible black-box latent force, so the two foundational claims—improved forecasting from physically grounded guidance, and physical interpretability—rest on evidence that is not yet conclusive.

major comments (4)
  1. [§4.1, Eq. (4)] The claim that "PhyDL-NWP learns from coarse-granular inputs alone ... without labeled training outputs" is directly contradicted by the training objective. Eq. (4) defines Ldata as the mean squared error between f_theta and the high-resolution fields u on the full grid, so the surrogate is explicitly trained against fine-granular labels. The downscaling results in Table 2 therefore do not demonstrate label-free super-resolution. Please either correct the claim to state that the method requires high-resolution training data but can be queried at arbitrary coordinates, or provide a version trained with only coarse-resolution supervision to support the original claim.
  2. [§3.4, Eqs. (6)-(8), Tables 3-5] The reported '+' models differ from their vanilla counterparts in two coupled ways: the training inputs are replaced by f_theta-super-resolved data (Algorithm 1), and the objective adds beta*Lphy. Since no beta=0 control (physics loss omitted, augmentation retained) and no augmentation-only control are reported, the observed improvements cannot be attributed to the recovered physics. The gain could come entirely from the high-resolution training inputs or from the extra fine-tuning procedure itself. Please add at least one ablation per dataset/baseline family that separates the effect of the physics loss from the effect of the data augmentation.
  3. [§3.2, Eq. (2)] The decomposition of the tendency into explicit PDE terms Phi*Xi and the latent force Q_pi is non-identifiable in the form presented. Because Q_pi is a flexible MLP (8 layers of 100 neurons, per Appendix A.2), any temporal residual for any choice of Xi can be absorbed into Q_pi, making the separation between 'physics' and 'latent force' unfalsifiable from the training loss alone. The physics residual of Eq. (5) is therefore a self-consistent regression rather than an independent physical law. To support the physical-consistency claim, provide independent validation, for example by showing that the learned Xi and Q_pi predict tendencies on held-out time periods or on a withheld variable/dataset without refitting, or by demonstrating sensitivity to sparsity constraints.
  4. [Table 6] The learned equations reported in Table 6 place essentially all of the physically informative terms—vertical advection, diffusion, pressure gradient, and source terms—into Q_pi, and the explicit terms use U10 to advect an 850 hPa temperature field. This configuration makes the claimed 'alignment with physics' weak evidence: the explicit part captures only a small subset of the true equation, and the comparison is largely a test of whether a flexible network can absorb the omitted terms. Please report quantitative measures of what fraction of the tendency variance is explained by the explicit terms versus Q_pi, and discuss whether the recovered explicit coefficients have physically plausible magnitudes and signs.
minor comments (5)
  1. [§4.1] The description of the downscaling setup as '2x and 4x' is ambiguous: the input resolutions are 0.5 and 1 degree and the output is 0.25 degree, so the linear scale factors are 2 and 4, but the pixel count increases by 4x and 16x. Please clarify the convention used.
  2. [Table 1] The comparison of time cost per epoch is not apples-to-apples: PhyDL-NWP is a plug-in module, not a standalone model, and the reported 7.8s likely reflects only the module's fine-tuning cost, whereas the other rows are full-model training epochs. Please report inference time per forecast as well, and state the hardware and batch-size settings under which the timings were measured.
  3. [§3.4, Eq. (8)] The finite-difference approximation of the temporal derivative in Eq. (8) is applied on 6-hourly WeatherBench data; at such coarse sampling, a first-order difference may be a poor approximation of the true tendency. Please justify the step size choice or test sensitivity to the finite-difference order.
  4. [§4.1] The sentence 'Since PhyDL-NWP can provide infinite resolution results given continuous coordinates, we believe that it will be accurate for higher resolution downscaling' is speculative; consider showing results at 8x or a scale-analysis plot instead.
  5. [§4.3, Figure 5] Figure 5 is presented as evidence that Q and Phi*Xi substitute for each other and match dT/dt, but only a single example year is shown. Please quantify the match (e.g., R^2 or relative error on a held-out period) and show for multiple variables.

Circularity Check

1 steps flagged · score 6.0 of 10

Physical-consistency claim is self-consistent by construction; forecasting and downscaling results retain independent held-out content.

  1. self definitional [Section 3.3, Eqs. (2)-(5)]
    "The joint optimization of fθ(x, y, t), φ(û)i, and Qπ(x, y, t) allows the model to simultaneously learn to predict weather variables and approximate the underlying physical dynamics. These inferred terms, together with Qπ, in turn guide the training of fθ, ensuring physical consistency in the learned mapping."

    The 'underlying physical dynamics' are not an external law: they are the fitted right-hand side of Eq. (2), with Qπ a flexible MLP and Ξ fitted by the same optimization. Because a universal Qπ can absorb any residual, Eq. (2) can represent essentially any observed tendency, so the physics residual in Eq. (5) is minimizable by construction on the training distribution. The claim that fθ is 'physically consistent' is therefore self-referential—the model is constrained to satisfy a regression fitted to the model's own outputs. The sparse library does not restore identifiability, since the latent force can always compensate, making the split between 'physics' and 'latent force' a bookkeeping choice rather than a discovered law.

full rationale

The downscaling and forecasting numbers (Tables 2-5) are obtained on held-out chronological splits, so the reported RMSE/ACC improvements are empirically independent of the physics-loss construction and do not reduce by definition. The circularity is confined to the paper's physical-consistency interpretation: the governing dynamics are fitted to fθ and then re-imposed on fθ (Eqs. 2-5), and the same fitted decomposition is later called 'recovered physics' when regularizing gω (Eq. 8). Because Qπ is a universal MLP, the decomposition is not identifiable from the training loss; Table 6's 'alignment' with textbook PDEs is therefore an interpretation of a fitted decomposition, not a confirmation of an independently derived law. No load-bearing self-citation or imported uniqueness theorem appears: Refs. [4,22,23] are related-work citations, not premises of the derivation. A separate empirical confound—the '+' models differ by both super-resolution data augmentation and βLphy, with no β=0 ablation (Tables 3-5)—weakens the causal attribution to the physics term but is a missing control, not a circular reduction, and is not counted in the score beyond the partial attribution.

Assumptions & free parameters 3 free parameters · 4 assumptions · 1 invented entities

The paper introduces a latent force term and fits the PDE coefficients from data, so the central physical model relies on fitted quantities. The candidate library and the continuity assumption are domain assumptions. The latent force is an invented entity with no external validation.

free parameters (3)
  • PDE coefficients Xi = learned via linear regression per variable
    Coefficients for each candidate term in Eq (2) are fitted to the training data; the learned dynamics in Table 6 reflect these fits.
  • Latent force network Q_pi parameters = learned weights of MLP
    Q_pi is an MLP with 8x100 units trained to capture residuals; it absorbs all missing physics, so its parameters are fitted to data.
  • Loss weights alpha and beta = alpha=10, beta=1e-2
    Chosen by grid search on validation set, affecting the balance between data and physics losses.
assumptions (4)
  • domain assumption The true dynamics are expressible as a linear combination of the candidate library Phi plus a latent force Q_pi
    Eq (2) assumes this form; the library includes only terms such as advection and diffusion, and all unmodeled physics is folded into Q_pi.
  • domain assumption The weather fields are continuous functions of (x,y,t) that can be accurately approximated by a feedforward neural network
    The surrogate f_theta is an MLP; continuity and smoothness are assumed for auto-diff derivatives to be meaningful.
  • domain assumption Physics learned on the downscaling domain transfers to the forecasting model
    The Xi and Q_pi learned on Huadong/WeatherBench are applied to regularize g_omega on other time periods; this assumes stationarity of the underlying dynamics.
  • ad hoc to paper The latent force model can represent all missing physical processes without overfitting
    Q_pi is chosen to absorb vertical velocity, friction, radiation, etc.; there is no independent evidence that an MLP captures these correctly.
invented entities (1)
  • Latent force field Q_pi
    purpose: Represents unobserved physical forces (vertical velocity, friction, heat sources) not present in the dataset, to close the PDE in Eq (2)
    Q_pi is a neural network fitted to the residual of the data-driven PDE; no independent measurement or prediction is provided to validate its physical content.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Physics-Guided Learning of Meteorological Dynamics for Weather Downscaling and Forecasting." pith.science (2026). https://pith.science/paper/K4M2SXXW

@misc{pith2026250514555,
  author       = {Pith},
  title        = {Pith review of: Physics-Guided Learning of Meteorological Dynamics for Weather Downscaling and Forecasting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/K4M2SXXW}},
  note         = {Machine review of arXiv:2505.14555}
}
read the original abstract

Weather forecasting is essential but remains computationally intensive and physically incomplete in traditional numerical weather prediction (NWP) methods. Deep learning (DL) models offer efficiency and accuracy but often ignore physical laws, limiting interpretability and generalization. We propose PhyDL-NWP, a physics-guided deep learning framework that integrates physical equations with latent force parameterization into data-driven models. It predicts weather variables from arbitrary spatiotemporal coordinates, computes physical terms via automatic differentiation, and uses a physics-informed loss to align predictions with governing dynamics. PhyDL-NWP enables resolution-free downscaling by modeling weather as a continuous function and fine-tunes pre-trained models with minimal overhead, achieving up to 170x faster inference with only 55K parameters. Experiments show that PhyDL-NWP improves both forecasting performance and physical consistency.

Figures

Figures reproduced from arXiv: 2505.14555 by the authors.

Figure 1
Figure 1. Schematic diagram of PhyDL-NWP for downscaling. First, given a continuous input coordinate (𝑥, 𝑦, 𝑡), the surrogate model 𝑓𝜃 approximates the weather data. Then, based on PyTorch’s auto-differentiation and the existing meteorology theory, we calculate the derivatives for the construction of physical mechanisms driven by PDE. Last, based on linear regression, we learn the PDE that fits the weather data well to provid… view at source ↗
Figure 2
Figure 2. To calculate differential terms efficiently, we propose to use [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 2
Figure 2. Schematic diagram of PhyDL-NWP for forecasting. We first use pre-trained surrogate model for weather downscaling to perform data augmentation, which is a necessity for aligning weather data resolution in the forecasting model. Then, we take the augmented historical data and use a pre-trained state-of-the-art forecasting model to predict future data. Based on the spatio-temporal coordinates of the predicted data, we … view at source ↗
Figures from the paper (3 more)
Figure 3
Figure 3. Figure 3: Model comparison in Ningxia dataset before and after physics guidance for a variety of forecasting ranges on the [PITH_FULL_IMAGE:figures/full_fig_p006_3.png]
Figure 4
Figure 4. Figure 4: Example of comparison of 7-day weather forecast results in Ningxia dataset. AFNO+ are closer to the ground truth. [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: The latent force and PDE for the temperature vari [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

47 extracted references · 29 canonical work pages

  1. [1]

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

    2015

  2. [2]

    Kaifeng Bi, Lingxi Xie, Hengheng Zhang, Xin Chen, Xiaotao Gu, and Qi Tian. 2023. Accurate medium-range global weather forecasting with 3D neural networks. Nature (2023), 1–6

    2023

  3. [3]

    Kang Chen, Tao Han, Junchao Gong, Lei Bai, Fenghua Ling, Jing-Jia Luo, Xi Chen, Leiming Ma, Tianning Zhang, Rui Su, et al. 2023. FengWu: Pushing the Skillful Global Medium-range Weather Forecast beyond 10 Days Lead. arXiv preprint arXiv:2304.02948 (2023)

    arXiv 2023

  4. [4]

    Yuntian Chen, Yingtao Luo, Qiang Liu, Hao Xu, and Dongxiao Zhang. 2022. Sym- bolic genetic algorithm for discovering open-form partial differential equations (SGA-PDE). Physical Review Research 4, 2 (2022), 023174

    work page 2022

  5. [5]

    John Guibas, Morteza Mardani, Zongyi Li, Andrew Tao, Anima Anandkumar, and Bryan Catanzaro. 2021. Adaptive fourier neural operators: Efficient token mixers for transformers. arXiv preprint arXiv:2111.13587 (2021)

    arXiv 2021

  6. [6]

    Jindong Han, Hao Liu, Hengshu Zhu, Hui Xiong, and Dejing Dou. 2021. Joint air quality and weather prediction based on multi-adversarial spatiotemporal networks. In Proceedings of the AAAI Conference on Artificial Intelligence , Vol. 35. 4081–4089

    work page 2021

  7. [7]

    Yan Han, Lihua Mi, Lian Shen, CS Cai, Yuchen Liu, Kai Li, and Guoji Xu. 2022. A short-term wind speed prediction method utilizing novel hybrid deep learning algorithms to correct numerical weather forecasting. Applied Energy 312 (2022), 118777

    work page 2022

  8. [8]

    Hans Hersbach, Bill Bell, Paul Berrisford, Shoji Hirahara, András Horányi, Joaquín Muñoz-Sabater, Julien Nicolas, Carole Peubey, Raluca Radu, Dinand Schepers, et al. 2020. The ERA5 global reanalysis. Quarterly Journal of the Royal Meteoro- logical Society 146, 730 (2020), 1999–2049

    work page 2020

Show all 47 references
  1. [9]

    Shuai Hu, Yue Xiang, Da Huo, Shafqat Jawad, and Junyong Liu. 2021. An improved deep belief network based hybrid forecasting method for wind power. Energy 224 (2021), 120185

    2021

  2. [10]

    Hao Jiang and Li Chen. 2022. An Efficient Content-aware Downsampling-based Video Compression Framework. In 2022 IEEE International Conference on Visual Communications and Image Processing (VCIP) . IEEE, 1–5

    2022

  3. [11]

    Renhe Jiang, Zhaonan Wang, Jiawei Yong, Puneet Jeph, Quanjun Chen, Ya- sumasa Kobayashi, Xuan Song, Shintaro Fukushima, and Toyotaro Suzumura

  4. [12]

    George Em Karniadakis, Ioannis G Kevrekidis, Lu Lu, Paris Perdikaris, Sifan Wang, and Liu Yang. 2021. Physics-informed machine learning. Nature Reviews Physics 3, 6 (2021), 422–440

    2021

  5. [13]

    Dmitrii Kochkov, Janni Yuval, Ian Langmore, Peter Norgaard, Jamie Smith, Griffin Mooers, Milan Klöwer, James Lottes, Stephan Rasp, Peter Düben, et al . 2024. Neural general circulation models for weather and climate. Nature (2024), 1–7

    2024

  6. [14]

    Remi Lam, Alvaro Sanchez-Gonzalez, Matthew Willson, Peter Wirnsberger, Meire Fortunato, Alexander Pritzel, Suman Ravuri, Timo Ewalds, Ferran Alet, Zach Eaton-Rosen, et al . 2022. GraphCast: Learning skillful medium-range global weather forecasting. arXiv preprint arXiv:2212.12...

    2022 arXiv

  7. [15]

    Yu Li, Fei Tang, Xin Gao, Tongyan Zhang, Junfeng Qi, Jiarui Xie, Xinang Li, and Yuhan Guo. 2022. Numerical weather prediction correction strategy for short-term wind power forecasting based on bidirectional gated recurrent unit and XGBoost. Frontiers in Energy Research 9 (2022...

    2022

  8. [16]

    Zongyi Li, Nikola Kovachki, Kamyar Azizzadenesheli, Burigede Liu, Kaushik Bhattacharya, Andrew Stuart, and Anima Anandkumar. 2020. Fourier neural oper- ator for parametric partial differential equations. arXiv preprint arXiv:2010.08895 (2020)

    2020 arXiv

  9. [17]

    Zongyi Li, Hongkai Zheng, Nikola Kovachki, David Jin, Haoxuan Chen, Burigede Liu, Kamyar Azizzadenesheli, and Anima Anandkumar. 2024. Physics-informed neural operator for learning partial differential equations. ACM/JMS Journal of Data Science 1, 3 (2024), 1–27

    2024

  10. [18]

    Yuxuan Liang, Haomin Wen, Yutong Xia, Ming Jin, Bin Yang, Flora Salim, Qing- song Wen, Shirui Pan, and Gao Cong. 2025. Foundation Models for Spatio- Temporal Data Science: A Tutorial and Survey. arXiv preprint arXiv:2503.13502 (2025)

    2025 arXiv

  11. [19]

    Haitao Lin, Zhangyang Gao, Yongjie Xu, Lirong Wu, Ling Li, and Stan Z Li. 2022. Conditional local convolution for spatio-temporal meteorological forecasting. In Proceedings of the AAAI Conference on Artificial Intelligence , Vol. 36. 7470–7478

    2022

  12. [20]

    Yumin Liu, Auroop R Ganguly, and Jennifer Dy. 2020. Climate downscaling using YNet: A deep convolutional network with skip connections and fusion. In Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining . 3145–3153

    2020

  13. [21]

    Andrew C Lorenc. 1986. Analysis methods for numerical weather prediction. Quarterly Journal of the Royal Meteorological Society 112, 474 (1986), 1177–1194

    1986

  14. [22]

    Yingtao Luo, Qiang Liu, Yuntian Chen, Wenbo Hu, Tian Tian, and Jun Zhu. 2023. Physics-guided discovery of highly nonlinear parametric partial differential equations. In Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining . 1595–1607

    2023

  15. [23]

    Yingtao Luo, Chang Xu, Yang Liu, Weiqing Liu, Shun Zheng, and Jiang Bian

  16. [24]

    Tung Nguyen, Johannes Brandstetter, Ashish Kapoor, Jayesh K Gupta, and Aditya Grover. 2023. ClimaX: A foundation model for weather and climate.arXiv preprint arXiv:2301.10343 (2023)

    2023 arXiv

  17. [25]

    Tung Nguyen, Jason Jewik, Hritik Bansal, Prakhar Sharma, and Aditya Grover

  18. [26]

    Sungwon Park, Karandeep Singh, Arjun Nellikkattil, Elke Zeller, Tung Duong Mai, and Meeyoung Cha. 2022. Downscaling earth system models with deep learning. In Proceedings of the 28th ACM SIGKDD conference on knowledge discovery and data mining. 3733–3742

    2022

  19. [27]

    Linsey S Passarella, Salil Mahajan, Anikesh Pal, and Matthew R Norman. 2022. Reconstructing high resolution ESM data through a novel fast super resolution convolutional neural network (FSRCNN). Geophysical Research Letters 49, 4 (2022), e2021GL097571

    2022

  20. [28]

    arXiv preprint arXiv:2307.01909 (2023)

    ClimateLearn: Benchmarking Machine Learning for Weather and Climate Modeling. arXiv preprint arXiv:2307.01909 (2023)

    2023 arXiv

  21. [29]

    Ilan Price, Alvaro Sanchez-Gonzalez, Ferran Alet, Tom R Andersson, Andrew El-Kadi, Dominic Masters, Timo Ewalds, Jacklynn Stott, Shakir Mohamed, Peter Battaglia, et al. 2023. Gencast: Diffusion-based ensemble forecasting for medium- range weather. arXiv preprint arXiv:2312.157...

    2023 arXiv

  22. [30]

    Maziar Raissi, Paris Perdikaris, and George E Karniadakis. 2019. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. J. Comput. Phys. 378 (2019), 686–707

    2019

  23. [31]

    Jaideep Pathak, Shashank Subramanian, Peter Harrington, Sanjeev Raja, Ashesh Chattopadhyay, Morteza Mardani, Thorsten Kurth, David Hall, Zongyi Li, Kamyar Azizzadenesheli, et al . 2022. Fourcastnet: A global data-driven high- resolution weather model using adaptive fourier neu...

    2022 arXiv

  24. [32]

    Sumanta Chandra Mishra Sharma and Adway Mitra. 2022. ResDeepD: A residual super-resolution network for deep downscaling of daily precipitation over India. Environmental Data Science 1 (2022), e19

    2022

  25. [33]

    Xingjian Shi, Zhourong Chen, Hao Wang, Dit-Yan Yeung, Wai-Kin Wong, and Wang-chun Woo. 2015. Convolutional LSTM network: A machine learning approach for precipitation nowcasting. Advances in neural information processing systems 28 (2015)

    2015

  26. [34]

    Stephan Rasp, Peter D Dueben, Sebastian Scher, Jonathan A Weyn, Soukayna Mouatadid, and Nils Thuerey. 2020. WeatherBench: a benchmark data set for data-driven weather forecasting. Journal of Advances in Modeling Earth Systems 12, 11 (2020), e2020MS002203

    2020

  27. [35]

    Thomas Vandal, Evan Kodra, Sangram Ganguly, Andrew Michaelis, Ramakrishna Nemani, and Auroop R Ganguly. 2017. Deepsd: Generating high resolution climate change projections through single image super-resolution. InProceedings of the 23rd acm sigkdd international conference on k...

    2017

  28. [36]

    Yogesh Verma, Markus Heinonen, and Vikas Garg. 2023. ClimODE: Climate and Weather Forecasting with Physics-informed Neural ODEs. In The Twelfth International Conference on Learning Representations

    2023

  29. [37]

    Roland B Stull. 1988. An introduction to boundary layer meteorology . Vol. 13. Springer Science & Business Media

    1988

  30. [38]

    Jonathan A Weyn, Dale R Durran, and Rich Caruana. 2020. Improving data- driven global weather prediction using deep convolutional neural networks on a cubed sphere. Journal of Advances in Modeling Earth Systems 12, 9 (2020), e2020MS002109

    2020

  31. [39]

    Haixu Wu, Hang Zhou, Mingsheng Long, and Jianmin Wang. 2023. Interpretable weather forecasting for worldwide stations with a unified deep model. Nature Machine Intelligence (2023), 1–10

    2023

  32. [40]

    Bin Wang, Jie Lu, Zheng Yan, Huaishao Luo, Tianrui Li, Yu Zheng, and Guangquan Zhang. 2019. Deep uncertainty quantification: A machine learning approach for weather forecasting. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining...

    2019

  33. [41]

    Wanghan Xu, Fenghua Ling, Wenlong Zhang, Tao Han, Hao Chen, Wanli Ouyang, and Lei Bai. 2024. Generalizing Weather Forecast to Fine-grained Temporal Scales via Physics-AI Hybrid Modeling. arXiv preprint arXiv:2405.13796 (2024)

    2024 arXiv

  34. [42]

    Xiaoying Yang, Shuai Yang, Mou Leong Tan, Hengyang Pan, Hongliang Zhang, Guoqing Wang, Ruimin He, and Zimeng Wang. 2022. Correcting the bias of daily satellite precipitation estimates in tropical regions using deep neural network. Journal of Hydrology 608 (2022), 127656

    2022

  35. [43]

    Zonghan Wu, Shirui Pan, Guodong Long, Jing Jiang, Xiaojun Chang, and Chengqi Zhang. 2020. Connecting the dots: Multivariate time series forecasting with graph neural networks. In Proceedings of the 26th ACM SIGKDD international conference on knowledge discovery & data mining . 753–763

    2020

  36. [46]

    Tingzhao Yu, Qiuming Kuang, Jiangping Zheng, and Junnan Hu. 2021. Deep precipitation downscaling. IEEE Geoscience and Remote Sensing Letters 19 (2021), 1–5. KDD ’25, August 3–7, 2025, Toronto, ON, Canada. Yingtao Luo, Shikai Fang, Binqing Wu, Qingsong Wen, and Liang Sun A Appe...

    2021

  37. [500]

    In the forecasting experiments, we divide each dataset into train, validation, and test sets using an 8:1:1 ratio in chronological order

    For variable𝑡 (temperature), the atmosphere level is 850. In the forecasting experiments, we divide each dataset into train, validation, and test sets using an 8:1:1 ratio in chronological order. The scores of the average of variables reported in Table 5 are com- puted based o...

    2025

  38. [2022]

    In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining

    Learning differential operators for interpretable time series modeling. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. 1192–1201

  39. [2023]

    In Proceedings of the AAAI Conference on Artificial Intelligence , Vol

    Spatio-temporal meta-graph learning for traffic forecasting. In Proceedings of the AAAI Conference on Artificial Intelligence , Vol. 37. 8078–8086

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.