Pith. sign in

REVIEW 4 major objections 4 minor 61 references

Temporal Bridges for Spatial Resolution: Enhancing Climate Data Super-Resolution with Bidirectional Alignment

T0 review · 4 major / 4 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 This paper claims that climate-data super-resolution improves when consecutive frames are bidirectionally aligned in a shared latent space, beating single-frame and optical-flow baselines on all five tested variables.

desk verdict A credible new combination of VQ-VAE latent pairing and bidirectional temporal alignment for climate SR, with a real reported gain, but the missing ablation numbers and unspecified VQ-VAE training split block acceptance. read the letter →

arxiv 2608.05981 v1 pith:OD7PP7U3 submitted 2026-08-06 cs.AI cs.LG

Yichen Zhang , Yixiong Xiao , Congxi Xiao , Jingbo Zhou This is my paper · ORCID
classification cs.AIcs.LG
keywords climatesuper-resolutiontemporalalignmentbidirectionalVQ-VAElatentspaceCMIP6ERA5downscaling
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 climate-data super-resolution can be improved by using temporal information from consecutive frames, not just spatial information from a single frame. It proposes a framework that first maps low- and high-resolution climate fields into a shared discrete latent space with paired VQ-VAEs, then trains forward and backward alignment networks to predict the high-resolution latent of a frame from its neighbors, and finally uses those networks as a regularizer in the super-resolution loss. On the CMIP6 (5.625 degrees) to ERA5 (1.40625 degrees) task, the method reports the lowest latitude-weighted RMSE on all five variables considered, for example Z500 at 1077.73 versus 1088.42 for ClimaX and T2m at 6.02 versus 6.11. The central claim is that the gain comes from bidirectional temporal alignment, which transfers temporal structure across resolutions in a way that single-frame and optical-flow methods cannot.

What carries the argument

The central mechanism is the Bidirectional Temporal Alignment module, consisting of two transformer-based networks, $\psi_F$ and $\psi_B$, that predict the high-resolution latent of the current frame from concatenated low-resolution latents of adjacent frames and the adjacent high-resolution latent. These networks are trained in the discrete latent space produced by paired VQ-VAEs with a shared codebook, which unifies feature representations across resolutions; in the super-resolution loss (Equation 4) the high-resolution neighbor is replaced by the super-resolution network's own output. That substitution is what turns the alignment networks from predictors into a temporal regularizer for the super-resolution task.

What would settle it

Run the ablation from Section 4.3 and report the actual RMSE for the configurations with the forward network removed, the backward network removed, and both removed; if eliminating both alignment networks leaves the final RMSE essentially unchanged, or if shuffling the temporal order of the input frames leaves performance intact, the paper's attribution of the gain to bidirectional temporal alignment is falsified.

Watch

Extended reading notes

Core claim

The paper's central claim is that a super-resolution model for gridded climate data performs better when it explicitly models the temporal dynamics between consecutive frames in a learned latent space rather than treating each frame independently. The proposed mechanism trains two alignment networks, a forward network $\psi_F$ that predicts the high-resolution latent of frame $t$ from frames $t-1$ and $t$ plus the previous high-resolution latent, and a backward network $\psi_B$ that does the same with frame $t+1$; these are then folded into the super-resolution loss (Equation 4) as regularizers, with the high-resolution neighbor replaced by the super-resolution model's own prediction. The authors report that this design reduces latitude-weighted RMSE on every one of the five test variables relative to ClimaX, VRT, and SwinIR, with the largest relative gains on Z500 and T2m. They attribute the improvement to the shared-codebook latent space, which lets the alignment networks capture temporal correlation common to both resolutions, and to the bidirectional structure, which the ablation (Figure 3) suggests contributes unevenly across variables.

Load-bearing premise

The load-bearing premise is that temporal correlations between consecutive low-resolution frames share common features with the correlations in high-resolution frames, and that feeding the super-resolution model's predicted high-resolution latents back into the alignment networks still improves training instead of injecting error; the ablation that would verify this is referenced only as Figure 3, with no numerical results reported in the text.

Editorial extensions

If this is right

  • If the alignment regularizer is responsible for the reported gains, climate super-resolution models that ignore temporal correlation are leaving measurable accuracy on the table.
  • The shared-codebook design implies that the alignment networks can be trained once and reused across different upsampling factors or target resolutions, because alignment operates in a resolution-invariant latent space.
  • The comparison with VRT suggests that optical flow, which assumes brightness constancy, is the wrong inductive bias for stochastic atmospheric fields, and learned latent alignment is a viable alternative.
  • The two-step training procedure offers a template for applying latent-space temporal regularization to other gridded geophysical variables beyond the five tested.

Reading between the lines

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

  • A natural extension would be to test whether the temporal bridge survives larger upsampling factors (e.g., 5.625 degrees to 0.25 degrees) or observational station targets, where the shared-codebook assumption may need retraining.
  • The paper itself notes that only adjacent frames are used; extending the alignment to longer temporal windows is a testable direction that might lower RMSE further.
  • The reported asymmetry, with backward alignment mattering more for Z500, T850, and T2m, could be checked against prevailing wind directions at those levels to see whether the model is capturing real advective structure.
  • Because the ablation is referenced but not numerically reported, a clean test of the causal claim would be to replace the alignment inputs with temporally shuffled latents; if the gain disappears, the mechanism is confirmed, and if not, the gain may come from the latent mapping or fine-tuning instead.
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 / 4 minor

Summary. The paper proposes a temporal-enhanced climate super-resolution framework. It first trains paired VQ-VAEs with a shared codebook to map low- and high-resolution climate fields into a common discrete latent space, then trains forward and backward temporal alignment networks, and finally performs two-step super-resolution training (latent-space training followed by fine-tuning in the pixel domain) with the alignment networks used as regularizers. The main empirical claim is in Table 1: on a CMIP6 (5.625 degrees) to ERA5 (1.40625 degrees) task, the proposed method achieves the lowest latitude-weighted RMSE on all five reported variables (Z500, T850, T2m, U10, V10), outperforming ClimaX, VRT, and SwinIR. The authors attribute the gain to the bidirectional temporal alignment mechanism.

Significance. If the mechanism is properly supported, the paper addresses a genuine gap: it attempts to exploit temporal correlations in climate super-resolution without relying on optical flow, which is ill-suited to noisy, stochastic climate fields. The empirical setup follows established benchmarks and the main comparison table reports means and standard deviations over five runs. The most valuable contribution would be a demonstration that learned bidirectional latent alignment transfers across resolutions and improves downstream SR accuracy. However, as submitted, the evidence for the mechanism is incomplete: the key ablation has no reported numbers, the loss in the central equation is not fully specified, and the training splits for the VQ-VAE and alignment networks are not stated, leaving a possible leakage path. These issues directly affect the credibility of both the headline results and the mechanistic interpretation.

major comments (4)
  1. [Section 4.1, Implementation] The manuscript does not state which temporal splits are used to train the VQ-VAE encoders/decoders, the shared codebook, the temporal alignment networks, or the Step-2 fine-tuning. Section 4.1 describes a global split (train 1979-2010, validation 2011-2012, test 2013-2015), but the subsequent Implementation paragraphs only list learning rates and epoch counts. Since Step 2 unfreezes the encoder and decoder, and since the codebook is learned from high-resolution frames, any use of 2011-2015 frames during VQ-VAE pretraining would constitute test-set leakage and could inflate the Table 1 improvements. Please specify the split used for each trained component and confirm that no component sees the test period.
  2. [Section 3.5, Eq. (4)] The loss in Eq. (4) is not well-defined as written: the alignment terms appear as '+·MSE(...)' with no scalar coefficient. The relative weight of the bidirectional temporal regularization is a load-bearing hyperparameter, since the entire claim is that this regularization produces the reported gains. Please define the coefficient, state its value, and provide at least a brief sensitivity analysis over a reasonable range.
  3. [Section 4.3, Figure 3] The central mechanistic claim is supported only by a figure that is not present in the text: 'Figure 3: Ablation study.' appears as a caption with no plot, table, or numeric values. The text states that removing psi_F and psi_B causes 'substantial decline' and that backward alignment matters more for Z500, T850, and T2m, but no numbers are reported. This is the experiment that separates the temporal-alignment mechanism from other components of the pipeline. Please provide a full ablation table with means, standard deviations, and sample sizes, comparable to Table 1.
  4. [Section 3.4] The premise that 'the temporal correlation between consecutive low-resolution time frames shares common features with the correlation in high-resolution frames' is asserted rather than demonstrated. This premise justifies the entire transfer design, and the missing ablation in Section 4.3 is exactly the experiment that would test it. In addition to reporting ablation numbers, please discuss whether the alignment networks learned a transferable correlation or merely memorized resolution-specific statistics.
minor comments (4)
  1. [Section 3.5 vs. Section 4.1] The method description in Section 3.5 and Eq. (4) uses three consecutive time points (t-1, t, t+1), while Section 4.1 states that the model takes four consecutive time points as input and produces outputs at each one. Please clarify the exact temporal context used in the experiments.
  2. [Section 4.2] The text says the method 'significantly outperforms all baselines,' but no statistical significance test is reported; the means and standard deviations over five runs are suggestive but do not by themselves establish significance across all variables. Please report paired tests or confidence intervals.
  3. [Section 4.1, Dataset] The description of the CMIP6 data preparation is given in the appendix, but the main text does not state the CMIP6 model or ensemble member used. Please name the specific source model and confirm that the regridded CMIP6 fields are not directly comparable to ERA5 in a way that biases the SR evaluation.
  4. [Section 4.4, Table 2] The efficiency comparison reports model size as 180 MB for the proposed method, but it is unclear whether this includes the VQ-VAE encoders and decoders or only the SR network. Please state the exact components included in the size and inference-time measurements.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: test RMSE is a held-out external benchmark; missing ablation numbers and unspecified VQ-VAE split are evidence gaps, not circular derivation.

full rationale

I traced the paper's derivation chain. The central empirical claim (Table 1) is evaluated on a temporally held-out test set (2013-2015) defined in Section 4.1, after SR training on 1979-2010 with validation 2011-2012. The reported RMSE is therefore an external benchmark and not a fitted value renamed as prediction. The bidirectional-alignment regularizer in Eq. (4) is an auxiliary training loss: the forward/backward networks of Section 3.4 are themselves trained as predictors of Z_HR(t), but the final inference path (Eq. 7) uses only phi_LR, psi_ST, and omega_HR; the reported metric is computed on I_hat_HR from this path, not on the alignment networks' own outputs. Thus the improvement in Table 1 is not forced by construction. No load-bearing step relies on a self-citation or on a uniqueness theorem by the authors; the cited VQ-VAE (Van Den Oord et al.) and VSR baselines are external, and the ClimaX dataset/protocol citation is used only to define the benchmark. I also flag two non-circular evidence gaps under the reviewing rule: Section 4.3's ablation study is supported only by the caption 'Figure 3: Ablation study.' with no numeric ablation results in the text, so the mechanism claim is under-supported; and Section 4.1 states the SR train/validation/test split but does not state which years were used to train the VQ-VAE and alignment networks, so test-set leakage cannot be excluded. Both are correctness/evidence concerns, not cases where a prediction reduces to its inputs by definition. I therefore find no significant circularity.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

No new physical entities, forces, or conserved quantities are introduced. The "temporal bridge" is a descriptive metaphor for the alignment network regularizer, not a new ontological object.

free parameters (3)
  • lambda, the alignment regularization weight in Equation 4 = not reported
    The temporal alignment terms in Equation 4 are the core of the claimed improvement, and the weight symbol is missing from the text. The benchmark gains depend on how strongly the regularizer is applied, and this value would need to be tuned or guessed to reproduce the results.
  • VQ-VAE codebook size and latent dimension = not reported
    The shared codebook is the backbone of the Paired Latent Mapping module; its size and dimensionality determine the granularity of the latent space and are not stated in the paper.
  • Number of SR training epochs in Step 2 = not reported
    Step 2 of the Temporal Enhanced Super-resolution training runs "for a specified number of epochs" without giving the number, so the stopping criterion is unknown.
assumptions (3)
  • domain assumption Temporal correlations between consecutive LR frames and between HR frames are shared across resolutions.
    Section 3.4 asserts this transferability to justify training alignment networks on low- and high-resolution latent pairs; the entire model design rests on it.
  • domain assumption CMIP6 model output at 5.625 degrees and ERA5 reanalysis at 1.40625 degrees represent the same geophysical field with consistent variable definitions.
    Section 4.1 pairs a climate model's coarse output with a reanalysis product, mixing model bias, observation assimilation, and regridding artifacts; the method treats the pair as clean LR/HR ground truth.
  • domain assumption A codebook trained on HR data provides a valid shared latent space for LR data when frozen.
    Section 3.3 trains the low-resolution encoder against the HR-trained codebook with the codebook frozen, assuming both resolutions quantize onto the same discrete features.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Temporal Bridges for Spatial Resolution: Enhancing Climate Data Super-Resolution with Bidirectional Alignment." pith.science (2026). https://pith.science/paper/OD7PP7U3

@misc{pith2026260805981,
  author       = {Pith},
  title        = {Pith review of: Temporal Bridges for Spatial Resolution: Enhancing Climate Data Super-Resolution with Bidirectional Alignment},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OD7PP7U3}},
  note         = {Machine review of arXiv:2608.05981}
}
read the original abstract

High-resolution climate data is crucial for meteorological predictions and for informing decision support across diverse domains. However, the acquisition of such high-resolution climate information is often prohibitively costly, necessitating the development of data-driven meteorological prediction models. These models aim to generate fine-grained climate data from low-resolution inputs, a process termed climate data super-resolution (SR). Nevertheless, recent advancements in deep learning for climate data SR have primarily focused on leveraging single-frame spatial information, largely neglecting the temporal correlations between different time frames that could enhance SR outcomes. Furthermore, climate data are inherently stochastic and noisy, rendering widely used temporal alignment methods, such as optical flow models, ineffective in this context. Consequently, the development of a framework tailored for climate data SR that effectively captures implicit temporal correlations remains an unresolved challenge. To this end, we propose a novel Temporal-Enhanced framework with bidirectional temporal alignment. In essence, our framework establishes a temporal bridge to enhance spatial resolution in climate data SR through bidirectional alignment, leading to improved SR performance. Within this framework, Paired Latent Mapping achieves spatial alignment and noise reduction by unifying latent spaces. Then a Bidirectional Temporal Alignment captures temporal correlations by training forward and backward networks on consecutive latent frames. Temporal Enhanced Super-resolution then optimizes the entire framework for climate data SR. Experiments on large-scale real-world datasets demonstrated the superior performance of our framework.

Figures

Figures reproduced from arXiv: 2608.05981 by the authors.

Figure 1
Figure 1. Overview of the proposed framework. a. Paired Latent Mapping module maps low and [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Architecture of the proposed framework. a. Low and high-resolution climate data are [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Ablation study. To evaluate the impact of temporal alignment networks on model performance, we conduct an abla￾tion study comparing our method with models under the following configurations: • w/o ψF : removing the forward alignment network and keeping the rest. • w/o ψB: removing the backward alignment network and keeping the rest. • w/o align: removing all temporal alignment networks [PITH_FULL_IMAGE:figures/full… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Examples of Model Output. 12 [PITH_FULL_IMAGE:figures/full_fig_p012_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

61 extracted references · 41 canonical work pages

  1. [1]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Swin transformer: Hierarchical vision transformer using shifted windows , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  2. [2]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Video swin transformer , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  3. [3]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Swinir: Image restoration using swin transformer , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  4. [4]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Basicvsr++: Improving video super-resolution with enhanced propagation and alignment , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    work page

  5. [5]

    Advances in Neural Information Processing Systems , volume=

    Recurrent video restoration transformer with guided deformable attention , author=. Advances in Neural Information Processing Systems , volume=

    work page

  6. [6]

    IEEE Transactions on Image Processing , year=

    Vrt: A video restoration transformer , author=. IEEE Transactions on Image Processing , year=

    work page

  7. [7]

    International Journal of Computer Vision , volume=

    Video enhancement with task-oriented flow , author=. International Journal of Computer Vision , volume=. 2019 , publisher=

    work page 2019

  8. [8]

    Proceedings of the IEEE international conference on computer vision , pages=

    Detail-revealing deep video super-resolution , author=. Proceedings of the IEEE international conference on computer vision , pages=

    work page

Show all 61 references
  1. [9]

    Proceedings of the IEEE International Conference on Computer Vision , pages=

    Robust video super-resolution with learned temporal dynamics , author=. Proceedings of the IEEE International Conference on Computer Vision , pages=

  2. [10]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Real-time video super-resolution with spatio-temporal networks and motion compensation , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  3. [11]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Optical flow estimation using a spatial pyramid network , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  4. [12]

    Wiley Interdisciplinary Reviews: Climate Change , volume=

    Use of stochastic weathergenerators for precipitation downscaling , author=. Wiley Interdisciplinary Reviews: Climate Change , volume=. 2010 , publisher=

    2010

  5. [13]

    Journal of Climate , volume=

    Downscaling of GCM-simulated precipitation using model output statistics , author=. Journal of Climate , volume=. 2014 , publisher=

    2014

  6. [14]

    Journal of Hydrology , volume=

    Statistical downscaling of GCM simulations to streamflow , author=. Journal of Hydrology , volume=. 2001 , publisher=

    2001

  7. [15]

    Advances in neural information processing systems , volume=

    Neural discrete representation learning , author=. Advances in neural information processing systems , volume=

  8. [16]

    Journal of Cloud Computing , volume=

    Challenges in remote sensing based climate and crop monitoring: navigating the complexities using AI , author=. Journal of Cloud Computing , volume=. 2024 , publisher=

    2024

  9. [17]

    Hydrology and Earth System Sciences , volume=

    Nonstationary weather and water extremes: a review of methods for their detection, attribution, and management , author=. Hydrology and Earth System Sciences , volume=. 2021 , publisher=

    2021

  10. [18]

    Atmosphere , volume=

    Machine Learning for Fog-and-Low-Stratus Nowcasting from Meteosat SEVIRI Satellite Images , author=. Atmosphere , volume=. 2023 , publisher=

    2023

  11. [19]

    Pattern Recognition , volume=

    Optical flow and scene flow estimation: A survey , author=. Pattern Recognition , volume=. 2021 , publisher=

    2021

  12. [20]

    Signal Processing: Image Communication , volume=

    A survey of variational and CNN-based optical flow techniques , author=. Signal Processing: Image Communication , volume=. 2019 , publisher=

    2019

  13. [21]

    Computer Vision and Image Understanding , volume=

    Optical flow modeling and computation: A survey , author=. Computer Vision and Image Understanding , volume=. 2015 , publisher=

    2015

  14. [22]

    ACM computing surveys (CSUR) , volume=

    The computation of optical flow , author=. ACM computing surveys (CSUR) , volume=. 1995 , publisher=

    1995

  15. [23]

    Artificial Intelligence Review , volume=

    Optical flow for video super-resolution: A survey , author=. Artificial Intelligence Review , volume=. 2022 , publisher=

    2022

  16. [24]

    Artificial Intelligence Review , volume=

    Video super-resolution based on deep learning: a comprehensive survey , author=. Artificial Intelligence Review , volume=. 2022 , publisher=

    2022

  17. [25]

    Journal of the Atmospheric Sciences , volume=

    On the identification of nonstationary factor models and their application to atmospheric data analysis , author=. Journal of the Atmospheric Sciences , volume=. 2010 , publisher=

    2010

  18. [26]

    arXiv preprint arXiv:2404.17752 , year=

    Generative Diffusion-based Downscaling for Climate , author=. arXiv preprint arXiv:2404.17752 , year=

    arXiv

  19. [27]

    arXiv preprint arXiv:2404.14416 , year=

    Conditional diffusion models for downscaling & bias correction of Earth system model precipitation , author=. arXiv preprint arXiv:2404.14416 , year=

  20. [28]

    Heliyon , volume=

    An improved deep learning procedure for statistical downscaling of climate data , author=. Heliyon , volume=. 2023 , publisher=

    2023

  21. [29]

    Scientific Data , volume=

    Deep learning downscaled high-resolution daily near surface meteorological datasets over East Asia , author=. Scientific Data , volume=. 2023 , publisher=

    2023

  22. [30]

    Proceedings of the National Academy of Sciences , volume=

    Adversarial super-resolution of climatological wind and solar data , author=. Proceedings of the National Academy of Sciences , volume=. 2020 , publisher=

    2020

  23. [31]

    Proceedings of the 28th ACM SIGKDD conference on knowledge discovery and data mining , pages=

    Downscaling earth system models with deep learning , author=. Proceedings of the 28th ACM SIGKDD conference on knowledge discovery and data mining , pages=

  24. [32]

    Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining , pages=

    Climate downscaling using YNet: A deep convolutional network with skip connections and fusion , author=. Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining , pages=

  25. [33]

    Nature , volume=

    Accurate medium-range global weather forecasting with 3D neural networks , author=. Nature , volume=. 2023 , publisher=

    2023

  26. [34]

    arXiv preprint arXiv:2202.11214 , year=

    Fourcastnet: A global data-driven high-resolution weather model using adaptive fourier neural operators , author=. arXiv preprint arXiv:2202.11214 , year=

    arXiv

  27. [35]

    Nature , volume=

    Deep learning and process understanding for data-driven Earth system science , author=. Nature , volume=. 2019 , publisher=

    2019

  28. [36]

    Philosophical Transactions of the Royal Society A , volume=

    Climbing down Charney’s ladder: machine learning and the post-Dennard era of computational climate science , author=. Philosophical Transactions of the Royal Society A , volume=. 2021 , publisher=

    2021

  29. [37]

    Philosophical Transactions of the Royal Society A , volume=

    Can deep learning beat numerical weather prediction? , author=. Philosophical Transactions of the Royal Society A , volume=. 2021 , publisher=

    2021

  30. [38]

    Supporting material of the Intergovernmental Panel on Climate Change, available from the DDC of IPCC TGCIA , volume=

    Guidelines for use of climate scenarios developed from statistical downscaling methods , author=. Supporting material of the Intergovernmental Panel on Climate Change, available from the DDC of IPCC TGCIA , volume=. 2004 , publisher=

    2004

  31. [39]

    Journal of Hydrology , volume=

    Statistical downscaling of hydrometeorological variables using general circulation model output , author=. Journal of Hydrology , volume=. 1998 , publisher=

    1998

  32. [40]

    Climate change 2007: The physical science basis

    Climate models and their evaluation , author=. Climate change 2007: The physical science basis. Contribution of Working Group I to the Fourth Assessment Report of the IPCC (FAR) , pages=. 2007 , publisher=

    2007

  33. [41]

    Journal of Geophysical Research: Atmospheres , volume=

    Introduction to special section: Regional climate modeling revisited , author=. Journal of Geophysical Research: Atmospheres , volume=. 1999 , publisher=

    1999

  34. [42]

    Reviews of geophysics , volume=

    Approaches to the simulation of regional climate change: a review , author=. Reviews of geophysics , volume=. 1991 , publisher=

    1991

  35. [43]

    Climate Research , volume=

    Climate downscaling: techniques and application , author=. Climate Research , volume=

  36. [44]

    Review of downscaling methods in climate change and their role in hydrological studies , author=. Int. J. Environ. Ecol. Geol. Mar. Eng , volume=

  37. [45]

    Nature , volume=

    The quiet revolution of numerical weather prediction , author=. Nature , volume=. 2015 , publisher=

    2015

  38. [46]

    Journal of computational physics , volume=

    The origins of computer weather prediction and climate modeling , author=. Journal of computational physics , volume=. 2008 , publisher=

    2008

  39. [47]

    Agenda , volume=

    Climate change 2007: The physical science basis , author=. Agenda , volume=

    2007

  40. [48]

    Water and Environment Journal , volume=

    A large-scale grid-based hydrological model of the Severn and Thames catchments , author=. Water and Environment Journal , volume=. 1996 , publisher=

    1996

  41. [49]

    Proceedings of the 23rd acm sigkdd international conference on knowledge discovery and data mining , pages=

    Deepsd: Generating high resolution climate change projections through single image super-resolution , author=. Proceedings of the 23rd acm sigkdd international conference on knowledge discovery and data mining , pages=

  42. [50]

    Geoscientific Model Development , volume=

    Overview of the Coupled Model Intercomparison Project Phase 6 (CMIP6) experimental design and organization , author=. Geoscientific Model Development , volume=. 2016 , publisher=

    2016

  43. [51]

    arXiv preprint arXiv:1312.6114 , year=

    Auto-encoding variational bayes , author=. arXiv preprint arXiv:1312.6114 , year=

    arXiv

  44. [52]

    arXiv preprint arXiv:2301.10343 , year=

    Climax: A foundation model for weather and climate , author=. arXiv preprint arXiv:2301.10343 , year=

    arXiv

  45. [53]

    Journal of Advances in Modeling Earth Systems , volume=

    WeatherBench: a benchmark data set for data-driven weather forecasting , author=. Journal of Advances in Modeling Earth Systems , volume=. 2020 , publisher=

    2020

  46. [54]

    Accurate Medium-Range Global Weather Forecasting with

    Bi, Kaifeng and Xie, Lingxi and Zhang, Hengheng and Chen, Xin and Gu, Xiaotao and Tian, Qi , year =. Accurate Medium-Range Global Weather Forecasting with. Nature , volume =. doi:10.1038/s41586-023-06185-3 , urldate =

    doi

  47. [55]

    , year =

    Hertwig, Denise and Ng, Matthew and Grimmond, Sue and Vidale, Pier Luigi and McGuire, Patrick C. , year =. High-Resolution Global Climate Simulations:. International Journal of Climatology , volume =. doi:10.1002/joc.7018 , urldate =

    doi

  48. [56]

    2023 , month = dec, journal =

    Learning Skillful Medium-Range Global Weather Forecasting , author =. 2023 , month = dec, journal =. doi:10.1126/science.adi2336 , urldate =

    2023 doi

  49. [57]

    2019 , month = jul, journal =

    Stochastic Weather and Climate Models , author =. 2019 , month = jul, journal =. doi:10.1038/s42254-019-0062-2 , abstract =

    2019 doi

  50. [58]

    Pathak, Jaideep and Subramanian, Shashank and Harrington, Peter and Raja, Sanjeev and Chattopadhyay, Ashesh and Mardani, Morteza and Kurth, Thorsten and Hall, David and Li, Zongyi and Azizzadenesheli, Kamyar and Hassanzadeh, Pedram and Kashinath, Karthik and Anandkumar, Animas...

  51. [59]

    2013 , month = jun, journal =

    The Influence of Observation Errors on Analysis Error and Forecast Skill Investigated with an Observing System Simulation Experiment , author =. 2013 , month = jun, journal =. doi:10.1002/jgrd.50452 , urldate =

    2013 doi

  52. [60]

    Hersbach, Hans and Bell, Bill and Berrisford, Paul and Hirahara, Shoji and Hor. The. 2020 , month = jul, journal =. doi:10.1002/qj.3803 , urldate =

    2020 doi

  53. [61]

    O'Neill, B. C. and Tebaldi, C. and. The Scenario Model Intercomparison Project (. 2016 , journal =

    2016

Pith tools

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