Pith. sign in

REVIEW 5 major objections 5 minor 25 references

A machine-learning ensemble trained on reanalysis and satellite precipitation can beat the operational GEFS at forecasting rain up to 15 days out.

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

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

T0 review · deepseek-v4-flash

2026-08-04 08:22 UTC pith:BIF6VM6J

load-bearing objection A real architecture and a genuine two-month evaluation, but the A.2 control shows the IMERG advantage is likely target-matching, not generalizable skill. the 5 major comments →

arxiv 2510.20769 v2 pith:BIF6VM6J submitted 2025-10-23 physics.ao-ph cs.LG

CSU-PCAST: A Dual-Branch Transformer Framework for medium-range ensemble Precipitation Forecasting

Tianyi Xiong , Haonan Chen , Kelly Mahoney , Jingyin Tang , Tim Smith , Janice Bytheway This is my paper
classification physics.ao-ph cs.LG
keywords ensemble precipitation forecastingdeep learning weather predictionSwin TransformerIMERGGEFSmedium-range forecastCRPSautoregressive model
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper presents CSU-PCAST, a 30-member ensemble precipitation forecasting system built on a Swin Transformer with stochastic noise conditioning, and claims it outperforms the operational GEFS ensemble in medium-range precipitation forecasts. Using 21 years of ERA5 atmospheric data with IMERG satellite precipitation as labels, the model is initialized from GFS analyses and predicts 6-hourly rainfall out to 15 days. Evaluated on January and July 2023, it reports higher Critical Success Index at thresholds from 0.1 to 20 mm, lower RMSE, lower CRPS, and better Brier Scores across lead times. If correct, this shows that a data-driven ensemble can match or exceed a traditional numerical ensemble for a high-impact, hard-to-predict variable like precipitation.

Core claim

CSU-PCAST consistently outperforms GEFS when both are initialized from GFS analyses and verified against IMERG. In January and July 2023, the model achieves higher CSI at 0.1, 1, 5, 10, and 20 mm thresholds, lower precipitation RMSE, and lower CRPS across the full 15-day window. The largest gains appear at moderate-to-heavy rainfall and at longer lead times, where GEFS skill degrades quickly. Brier Score differences are also favorable through about day 10, and a Typhoon Sanba case shows improved spatial structure and exceedance probabilities. The paper argues that a purpose-built precipitation branch, trained with a combined CRPS and weighted log1p MSE loss, is key to this advantage.

What carries the argument

The central architecture is a patch-based Swin Transformer V2 backbone whose feature representations are modulated by conditional layer normalization (FiLM) driven by both time embeddings and injected Gaussian noise. This noise conditioning generates ensemble spread without perturbing initial conditions. A dual-branch decoder separates the total-precipitation channel from the 57 other atmospheric and surface variables, allowing the precipitation branch to be trained with a precipitation-specific loss that emphasizes rain above 5 and 10 mm via intensity weighting. The model is trained autoregressively, first on non-precipitation variables, then fine-tuned for precipitation, and during inferen

Load-bearing premise

IMERG Final (version 07) precipitation is treated as the ground truth for both training labels and verification, so reported skill is skill at reproducing IMERG, which the model was trained to match, not necessarily skill at reproducing true precipitation.

What would settle it

Run CSU-PCAST and GEFS forecasts for a full year and verify against an independent precipitation analysis not used in training, such as a dense gauge network over the United States or a reanalysis-independent satellite product. If the CSI and CRPS advantages shrink or reverse, the reported skill is likely an artifact of matching IMERG's error structure rather than a genuine improvement in precipitation forecasting.

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

Share X Bluesky LinkedIn Reddit HN

If this is right

  • If the claimed skill generalizes, data-driven ensembles could provide operationally useful precipitation guidance at a fraction of the computational cost of traditional ensemble NWP.
  • The noise-conditioning approach offers a lightweight alternative to diffusion or perturbed-initial-condition ensembles for generating spread.
  • The dual-branch decoder suggests a general recipe for handling heavy-tailed variables: train a shared backbone for smooth variables, then attach a specialized branch with a tailored loss.
  • The reported improvements at heavier thresholds and longer lead times are exactly where operational GEFS is weakest, so CSU-PCAST could complement existing guidance.
  • Consistent with the paper's own statement, both systems remain underdispersive, so the reliability gain is relative, not absolute.

Where Pith is reading between the lines

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

  • The evaluation covers only two months of one year, so the claim of general 15-day superiority is an extrapolation; testing across all seasons would clarify whether the advantage holds year-round.
  • Because the model is trained to match IMERG, the comparison against GEFS may partly measure how well each system reproduces IMERG's retrieval biases; verification against independent gauge or radar data would test this.
  • The intensity-weighted log1p loss could be transferred to other extreme-event forecasting problems, such as heat waves or ocean wave heights, where rare large values need emphasis.
  • If the model's spread is produced purely by noise conditioning, it may be possible to tune spread on the fly without retraining, enabling cheap calibration for downstream users.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. The paper presents CSU-PCAST, a dual-branch Swin Transformer framework for 15-day, 30-member ensemble precipitation forecasting. The model is trained on ERA5 atmospheric fields with IMERG precipitation as labels, initializes from operational GFS analyses at inference, and uses an autoregressive rollout with stochastic noise for ensemble diversity. The authors evaluate deterministic and probabilistic precipitation skill against GEFS for January and July 2023, reporting higher CSI, lower RMSE and CRPS, and improved Brier Scores. An appendix control experiment switches training labels and verification reference to ERA5 precipitation.

Significance. If the headline result held, CSU-PCAST would be a notable contribution: an ML ensemble at 0.25° that beats an operational NWP ensemble in precipitation skill. The architecture is well described, the evaluation is temporally out-of-sample, and the authors honestly report a control experiment in Appendix A.2. However, that same control experiment sharply limits the claim: when the training/verification reference is changed from IMERG to ERA5 and the model is initialized from GFS, CSU-PCAST becomes marginally worse than GEFS. Since the main evaluation uses IMERG as both training target and verification reference, the reported skill advantage may reflect target-product matching rather than a robust forecasting improvement. In addition, the abstract claims a full-year 2023 evaluation while the body evaluates only January and July, and no uncertainty quantification is provided for score differences.

major comments (5)
  1. [Abstract; Section 2; Section 4.3] The abstract states evaluation 'over the full year of 2023,' but Sections 2 and 4.3 describe evaluation only for January and July 2023 (two representative months). The conclusion also refers to 'both winter and summer cases.' The abstract's claim is not supported by the body. Either perform a full-year evaluation or revise the abstract to match the two-month evidence.
  2. [Section 3.1; Section 2; Eqs. (3)–(5); Appendix A.2] The training objective (Eqs. 3–5) minimizes CRPS and log1pMSE against IMERG, and all main precipitation metrics are verified against the same IMERG product. GEFS, by contrast, is an independent NWP system. The control experiment in Appendix A.2 is the key test: with ERA5 precipitation as label and reference, and with GFS initialization, CSU-PCAST performs marginally worse than GEFS (Figs. 9–11). This indicates that the main reported advantage is tied to the identity of the training/verification product, not an intrinsic forecast-skill gain. The authors need to either evaluate against an independent reference (e.g., gauge-adjusted products or a second satellite product) or substantially weaken the central claim and discuss this circularity explicitly.
  3. [Section 3.2] GEFS forecasts at native 0.5° resolution are bilinearly interpolated to 0.25° prior to verification, while CSU-PCAST is trained and evaluated at 0.25°. Bilinear interpolation smooths the GEFS precipitation field, which can inflate apparent CSU-PCAST advantages in categorical metrics such as CSI at high thresholds and in CRPS. The authors should quantify this effect, e.g., by also evaluating CSU-PCAST on the 0.5° grid after conservative remapping, or by reporting both native- and common-grid comparisons.
  4. [Section 2; Figures 1–4] All claims of 'consistently higher' or 'consistently lower' skill rely on point estimates over two months. No confidence intervals, significance tests, or block-bootstrap intervals are provided, despite strong spatial and temporal correlations in precipitation fields. The CSI differences in Figure 1 are often small (e.g., <0.05 at several lead times/thresholds), and the BS differences in Figure 4 cross zero at long lead times. The authors should add uncertainty quantification to support the strength of the claims.
  5. [Section 1] The introduction states that CSU-PCAST 'outperforms ECMWF and GEFS ensembles,' but the paper only evaluates against GEFS. No ECMWF ENS comparison is presented anywhere in the manuscript. This unsupported claim should be removed or qualified to 'GEFS' only.
minor comments (5)
  1. [Abstract; Section 1] Typos: 'NVIDA' in Section 2 and 'Cooperative Intitute' in Acknowledgments. The abstract says 'Precipitation Forecasting' but the title uses 'precipitation Forecasting'; use consistent capitalization.
  2. [Section 3.1, Table 1] '2-meter dewpoint temperature' is abbreviated 2D, but the standard abbreviation is 2Td or 2D; clarify. Also 'Total column water vapor' is TCWV; fine. Consider defining abbreviations in the table caption.
  3. [Figures 1, 2, 4] Figure panels are small and the text refers to 'POD' and 'FAR' in Section 2.1, but these metrics are not shown in the figures or explicitly defined in Section 4.3. Either add the figures or remove the reference.
  4. [Section 4.2.3] The statement that 'precipitation inherently follows an autoregressive dependency' and therefore one-step training is sufficient is plausible but not demonstrated. The model is rolled out 60 steps without multi-step precipitation training; the paper should cite evidence or provide an ablation (e.g., fine-tuning with 2–3 steps) to support stability.
  5. [General] The manuscript does not mention code or data availability. Given the reproducibility expectations for ML-based forecasting papers, please add a statement or repository link.

Circularity Check

1 steps flagged

Training target (IMERG) is also the verification reference, and the Appendix A.2 control shows the headline advantage disappears when the label/reference is switched to ERA5 under GFS initialization.

specific steps
  1. fitted input called prediction [Section 3.1; Section 4.2.3 (Eqs. 4–5); Section 2.2 (Eq. 8, Fig. 4); Appendix A.2]
    "IMERG precipitation (version 07) is adopted as the label dataset... Model training minimizes a hybrid loss combining the Continuous Ranked Probability Score (CRPS) and a weighted log1p mean squared error (log1pMSE)... L=LCRPS( ˆXt+1,X t+1) +λL log1pMSE( ˆPt+1,P t+1)... The CRPS results further demonstrate the advantages of CSU-PCAST... However, when the model was initialized with GFS, CSU-PCAST performed marginally worse than GEFS across all evaluation metrics."

    The training loss (Eqs. 3–5) directly minimizes CRPS and log1pMSE with IMERG as ground truth, while the evaluation (Eqs. 6–9) computes CSI/RMSE/CRPS/BS against the same IMERG product. The training CRPS is the same scoring rule used for headline probabilistic verification, so the model is fitted to the reference against which it is then judged. GEFS is not fitted to IMERG, making the comparison asymmetric. The Appendix A.2 control confirms the dependence: with ERA5 labels/reference and GFS initialization, CSU-PCAST is marginally worse than GEFS, so the main advantage is tied to the choice of training/verification product. This is partial, not full, circularity because 2023 is out-of-sample and GFS analyses provide an independent initialization.

full rationale

The core reported result—lower CRPS and higher CSI than GEFS—rests on an evaluation reference (IMERG) that is also the training label. Since the model's loss explicitly optimizes the same scoring rule (CRPS) and the same reference product used in verification, the comparison is biased in the model's favor. The Appendix A.2 control is the decisive internal evidence: when both label and verification reference are changed to ERA5 while keeping GFS initialization, the model no longer beats GEFS. Thus the headline skill is partly a target-matching effect rather than a label-independent property. No self-citation or uniqueness-theorem circularity is present; the architecture borrows from FuXi but that is not load-bearing. Non-precipitation variables are similarly trained and evaluated on ERA5, reinforcing the same pattern, though the main claim is about precipitation.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 0 invented entities

The central claim rests on two external data products: ERA5 as atmospheric state and IMERG as precipitation truth, plus an autoregressive rollout assumption. No new physical entities are introduced. The loss weights, intensity weights, and CRPS surrogate constants are hand-chosen and shape the trained distribution.

free parameters (5)
  • lambda (loss balance) = 5
    Eq. 4: weight on the log1pMSE precipitation term relative to CRPS in the precipitation training stage.
  • omega intensity weights = 1 (<5 mm), 2 (5-10 mm), 3 (>10 mm)
    Eq. 5: ad hoc grid-point weights to emphasize moderate and heavy precipitation in the log1pMSE loss.
  • alpha in CRPS surrogate loss = 0.95
    Eq. 3: coefficient used to set epsilon = (1 - alpha)/M in the CRPS loss approximation.
  • training ensemble size M = 2
    CRPS loss uses 2 sampled members during training, while inference uses 30 members; the mismatch is not analyzed.
  • architectural capacity = 48 Swin blocks, C=1536
    Hand-chosen model size adopted from FuXi; not fitted to data but essential to the reported performance.
axioms (5)
  • domain assumption IMERG Final precipitation is sufficiently accurate as ground truth for both training and verification.
    Section 3.1 selects IMERG as the label and Section 2 uses it as the reference. If IMERG biases are not representative of true rainfall, the reported improvements may not transfer.
  • domain assumption ERA5 reanalysis fields are a valid training distribution for a model initialized from GFS analyses.
    Training uses ERA5 while inference ingests operational GFS analyses (Section 4.2.4). The GFS fine-tuning experiment did not help, so the distribution shift is managed implicitly, not corrected.
  • ad hoc to paper A 60-step autoregressive rollout is stable even though the precipitation branch is trained only one step ahead.
    Section 4.2.3 asserts single-step precipitation training is sufficient and that autogressive rolling is physically consistent; no multi-step precipitation training or ablation is shown.
  • domain assumption GEFS precipitation interpolated from 0.5 to 0.25 degrees is a fair baseline for grid-scale precipitation verification.
    Section 3.2 bilinearly interpolates GEFS to the ERA5/IMERG grid. Interpolation smooths the baseline and may affect threshold-based scores.
  • domain assumption Gaussian noise injected via FiLM/conditional layer norm produces a statistically meaningful ensemble without calibration.
    The paper does not calibrate the ensemble and notes both systems remain underdispersive (Section 2.2), so the probabilistic claims rest on the noise injection design.

pith-pipeline@v1.3.0-alltime-deepseek · 17417 in / 17254 out tokens · 152618 ms · 2026-08-04T08:22:31.639649+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of CSU-PCAST: A Dual-Branch Transformer Framework for medium-range ensemble Precipitation Forecasting." pith.science (2026). https://pith.science/paper/BIF6VM6J

@misc{pith2026251020769,
  author       = {Pith},
  title        = {Pith review of: CSU-PCAST: A Dual-Branch Transformer Framework for medium-range ensemble Precipitation Forecasting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BIF6VM6J}},
  note         = {Machine review of arXiv:2510.20769}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Accurate medium-range precipitation forecasting is essential for hydrometeorological risk management but remains challenging for both numerical weather prediction (NWP) systems and data-driven models. We present CSU-PCAST, a deep learning-based ensemble forecasting framework for global precipitation prediction. The model is trained using ERA5 atmospheric and surface variables at 0.25{\deg} resolution with precipitation labels from NASA's IMERG dataset. CSU-PCAST uses 57 prognostic variables and static geographical fields to predict both atmospheric states and 6-h accumulated precipitation. The framework employs a Swin Transformer backbone with stochastic noise conditioning, temporal embeddings, and a dual-branch decoder for precipitation and non-precipitation variables. During inference, CSU-PCAST is initialized from operational GFS analyses and generates 30 ensemble members out to 15 days using an autoregressive strategy. Evaluation against GEFS over the full year of 2023 shows improved precipitation forecast skill at short lead times, including higher Critical Success Index (CSI) and lower RMSE during the first several forecast days. CSU-PCAST also reduces GEFS wet bias for light precipitation and dry bias at heavier precipitation thresholds. Probabilistic verification demonstrates lower CRPS, higher Brier Skill Scores at several thresholds, and improved ensemble reliability relative to GEFS, although both systems remain underdispersive. A case study of the Sanba extreme precipitation event further shows improved spatial structure and exceedance-probability guidance. These results demonstrate the potential of CSU-PCAST for short-to-medium-range ensemble precipitation forecasting while highlighting remaining challenges in extreme precipitation prediction and ensemble calibration.

Figures

Figures reproduced from arXiv: 2510.20769 by Haonan Chen, Janice Bytheway, Jingyin Tang, Kelly Mahoney, Tianyi Xiong, Tim Smith.

Figure 1
Figure 1. Figure 1: CSI scores of the precipitation forecast (6 hr) from CSU-PCAST model and GEFS at different lead times and precipitation intensities during Jan 2023 and Jul 2023: (a) Jan CSU-PCAST; (b) Jan GEFS; (c) Jan CSI difference between CSU-PCAST and GEFS; (d) Jul CSU-PCAST; (e) Jul GEFS; (f) Jul CSI difference between CSU-PCAST and GEFS [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: RMSE of the precipitation forecast (6 hr) from the CSU-PCAST model and GEFS at different lead times, evaluated against IMERG precipitation: (a) RMSE between CSU-PCAST and GEFS during January 2023; (b) Relative RMSE difference between CSU-PCAST and GEFS during January 2023; (c) RMSE between CSU-PCAST and GEFS during July 2023; (d) Relative RMSE difference between CSU-PCAST and GEFS during July 2023 midlatit… view at source ↗
Figure 3
Figure 3. Figure 3: Precipitation forecasts initialized at 2023-07-06 00UTC with a forecast lead time of 60 hours. The top row shows forecasts from the CSU-PCAST model (left) and GEFS (right), while the bottom panel shows IMERG ground-truth observations. maintains a tighter distribution around the truth, reflecting reduced uncertainty. The BS is applied to binary events defined by precipitation thresholds (e.g., exceeding 0.1… view at source ↗
Figure 4
Figure 4. Figure 4: Brier Score (BS) differences and CRPS for precipitation forecasts in January and July 2023, against IMERG. Panels (a.1–e.1) show BS differences at thresholds of 0.1, 1, 5, 10, and 20 mm during January 2023, and panels (a.2-e.2) show BS differences during Jul 2023, respectively. The blue horizontal line denotes the GEFS baseline (0), while the red curves represent the relative BS of the CSU-PCAST model comp… view at source ↗
Figure 5
Figure 5. Figure 5: Panel a: overall architecture of CSU-PCAST. The model is composed of patch embedding, a U-Transformer backbone, and fully connected layers. Panel b: the details of Swin layers, the transformer block has 4 Swin layers, each Swin layer contains 12 Swin blocks. 8 [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Panel a (first and third rows): RMSE of the ensemble mean for T2M, T850, U10, V10, and Z500 from CSU-PCAST and GEFS during January 2023, both evaluated against ERA5; Panel b (second and fourth rows): Relative RMSE differences between CSU-PCAST and GEFS during January 2023. 14 [PITH_FULL_IMAGE:figures/full_fig_p014_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Panel a (first and third rows): RMSE of the ensemble mean for T2M, T850, U10, V10, and Z500 from CSU-PCAST and GEFS during July 2023, both evaluated against ERA5; Panel b (second and fourth rows): Relative RMSE differences between CSU-PCAST and GEFS during July 2023. 15 [PITH_FULL_IMAGE:figures/full_fig_p015_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Panel a (first and second rows): ACC of the ensemble mean for T2M, T850, U10, V10, and Z500 from CSU-PCAST and GEFS during January 2023, both evaluated against ERA5; Panel b (third and fourth rows): ACC of the ensemble mean from CSU-PCAST and GEFS during July 2023, both evaluated against ERA5 16 [PITH_FULL_IMAGE:figures/full_fig_p016_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: CSI scores of 6-hour precipitation forecasts from the CSU-PCAST model and GEFS at different lead times and precipitation intensities during January 2023. The first row corresponds to results initialized with GFS, while the second row corresponds to results initialized with ERA5. Panels (a–c) show the CSI of CSU-PCAST, GEFS, and their differences for the GFS-initialized experiment, and panels (d–f) show the… view at source ↗
Figure 10
Figure 10. Figure 10: BS differences for precipitation forecasts initialized with GFS and ERA5, verified against ERA5 during January 2023. Panels (a–c) show BS differences at precipitation thresholds of 5 mm, 10 mm, and 20 mm, respectively, for the GFS-initialized experiment; panels (d–f) show the corresponding results for the ERA5-initialized experiment. The blue horizontal line denotes the GEFS baseline (0), while the red cu… view at source ↗
Figure 11
Figure 11. Figure 11: Comparison of CSU-PCAST and GEFS precipitation forecast skill against ERA5 during January 2023. Panels (a–b) show the CRPS and RMSE for forecasts initialized with GFS, while panels (c–d) show the corresponding results for forecasts initialized with ERA5. The red curves represent CSU-PCAST, and the blue curves represent GEFS. 19 [PITH_FULL_IMAGE:figures/full_fig_p019_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Precipitation forecasts initialized at 2023-07-06 00UTC with a forecast lead time of 60 hours. The top row shows forecasts from the CSU-PCAST model initialized with GFS (left) and GEFS (right), while the bottom row shows the forecasts from the CSU-PCAST initialized with ERA5 and ERA5 ground truth (right). 20 [PITH_FULL_IMAGE:figures/full_fig_p020_12.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

25 extracted references · 5 linked inside Pith

  1. [1]

    Gfs-powered machine learning weather prediction: A comparative study on training graphcast with noaa’s gdas data for global weather forecasts.Preprint, 2025

    S Sadeghi Tabas, J Wang, W Lei, et al. Gfs-powered machine learning weather prediction: A comparative study on training graphcast with noaa’s gdas data for global weather forecasts.Preprint, 2025

    2025

  2. [2]

    M Leutbecher and T. N. Palmer. Ensemble forecasting.Journal of Computational Physics, 227(7):3515–3539, 2008

    2008

  3. [3]

    Fact sheet: Ensem- ble weather forecasting

    European Centre for Medium-Range Weather Forecasts (ECMWF). Fact sheet: Ensem- ble weather forecasting. https://www.ecmwf.int/en/about/media-centre/focus/2017/ fact-sheet-ensemble-weather-forecasting, 2017. [Accessed: 17 Aug 2025]

    2017

  4. [4]

    Pangu-weather: A 3d high-resolution model for fast and accurate global weather forecast.arXiv preprint arXiv:2211.02556, 2022

    K Bi, L Xie, H Zhang, et al. Pangu-weather: A 3d high-resolution model for fast and accurate global weather forecast.arXiv preprint arXiv:2211.02556, 2022

    Pith/arXiv arXiv 2022

  5. [5]

    Fourcastnet: A global data-driven high-resolution weather model using adaptive fourier neural operators.arXiv preprint arXiv:2202.11214, 2022

    J Pathak, S Subramanian, P Harrington, et al. Fourcastnet: A global data-driven high-resolution weather model using adaptive fourier neural operators.arXiv preprint arXiv:2202.11214, 2022

    Pith/arXiv arXiv 2022

  6. [6]

    Probabilistic weather forecasting with machine learning.Nature, 637(8044):84–90, 2025

    I Price, A Sanchez-Gonzalez, F Alet, et al. Probabilistic weather forecasting with machine learning.Nature, 637(8044):84–90, 2025

    2025

  7. [7]

    Fuxi-ens: A machine learning model for medium-range ensemble weather forecasting.arXiv preprint arXiv:2405.05925, 2024

    X Zhong, L Chen, H Li, et al. Fuxi-ens: A machine learning model for medium-range ensemble weather forecasting.arXiv preprint arXiv:2405.05925, 2024. 12 CSU-PCAST: Medium-range Ensemble Precipitation Forecasting

    Pith/arXiv arXiv 2024

  8. [8]

    The era5 global reanalysis.Quarterly Journal of the Royal Meteorological Society, 146(730):1999–2049, 2020

    H Hersbach, B Bell, P Berrisford, et al. The era5 global reanalysis.Quarterly Journal of the Royal Meteorological Society, 146(730):1999–2049, 2020

    1999

  9. [9]

    Dreary state of precipitation in global models

    Graeme L Stephens, Tristan L’Ecuyer, Richard Forbes, Andrew Gettelmen, Jean-Christophe Golaz, Alejandro Bodas-Salcedo, Kentaroh Suzuki, Philip Gabriel, and John Haynes. Dreary state of precipitation in global models. Journal of Geophysical Research: Atmospheres, 115(D24), 2010

    2010

  10. [10]

    The quiet revolution of numerical weather prediction.Nature, 525(7567):47–55, 2015

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

    2015

  11. [11]

    A review of global precipitation data sets: Data sources, estimation, and intercomparisons.Reviews of geophysics, 56(1):79–107, 2018

    Qiaohong Sun, Chiyuan Miao, Qingyun Duan, Hamed Ashouri, Soroosh Sorooshian, and Kuo-Lin Hsu. A review of global precipitation data sets: Data sources, estimation, and intercomparisons.Reviews of geophysics, 56(1):79–107, 2018

    2018

  12. [12]

    Gencast: Diffusion-based ensemble forecasting for medium-range weather.arXiv preprint arXiv:2312.15796, 2023

    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. Gencast: Diffusion-based ensemble forecasting for medium-range weather.arXiv preprint arXiv:2312.15796, 2023

    Pith/arXiv arXiv 2023

  13. [13]

    An introduction to multivariate probabilistic forecast evaluation.Energy and AI, 4:100058, 2021

    Mathias Blicher Bjerregård, Jan Kloppenborg Møller, and Henrik Madsen. An introduction to multivariate probabilistic forecast evaluation.Energy and AI, 4:100058, 2021

    2021

  14. [14]

    The era5 global reanalysis.Quarterly journal of the royal meteorological society, 146(730):1999–2049, 2020

    Hans Hersbach, Bill Bell, Paul Berrisford, Shoji Hirahara, András Horányi, Joaquín Muñoz-Sabater, Julien Nicolas, Carole Peubey, Raluca Radu, Dinand Schepers, et al. The era5 global reanalysis.Quarterly journal of the royal meteorological society, 146(730):1999–2049, 2020

    1999

  15. [15]

    Huffman, David T Bolvin, Dan Braithwaite, Kuolin Hsu, Robert Joyce, Christopher Kidd, Eric J Nelkin, Soroosh Sorooshian, Jackson Tan, and Pingping Xie

    George J. Huffman, David T Bolvin, Dan Braithwaite, Kuolin Hsu, Robert Joyce, Christopher Kidd, Eric J Nelkin, Soroosh Sorooshian, Jackson Tan, and Pingping Xie. Algorithm Theoretical Basis Document (ATBD) Version 06, NASA Global Precipitation Measurement (GPM) Integrated Multi-satellitE Retrievals for GPM (IMERG), March 2019

    2019

  16. [16]

    Performance of imerg as a function of spatiotemporal scale.Journal of Hydrometeorology, 18(2):307–319, 2017

    Jackson Tan, Walter A Petersen, Pierre-Emmanuel Kirstetter, and Yudong Tian. Performance of imerg as a function of spatiotemporal scale.Journal of Hydrometeorology, 18(2):307–319, 2017

    2017

  17. [17]

    Global forecast system (gfs)

    National Centers for Environmental Prediction (NCEP). Global forecast system (gfs). https://www.emc.ncep. noaa.gov/emc/pages/numerical_forecast_systems/gfs.php, 2025. [Accessed: 2 Sep 2025]

    2025

  18. [18]

    Global ensemble forecast system (gefs)

    National Centers for Environmental Prediction (NCEP). Global ensemble forecast system (gefs). https: //www.emc.ncep.noaa.gov/emc/pages/numerical_forecast_systems/gefs.php, 2025. [Accessed: 2 Sep 2025]

    2025

  19. [19]

    Fuxi: a cascade machine learning forecasting system for 15-day global weather forecast.npj climate and atmospheric science, 6(1):190, 2023

    Lei Chen, Xiaohui Zhong, Feng Zhang, Yuan Cheng, Yinghui Xu, Yuan Qi, and Hao Li. Fuxi: a cascade machine learning forecasting system for 15-day global weather forecast.npj climate and atmospheric science, 6(1):190, 2023

    2023

  20. [20]

    Videomae: Masked autoencoders are data-efficient learners for self-supervised video pre-training.Advances in neural information processing systems, 35:10078–10093, 2022

    Zhan Tong, Yibing Song, Jue Wang, and Limin Wang. Videomae: Masked autoencoders are data-efficient learners for self-supervised video pre-training.Advances in neural information processing systems, 35:10078–10093, 2022

    2022

  21. [21]

    Film: Visual reasoning with a general conditioning layer

    Ethan Perez, Florian Strub, Harm De Vries, Vincent Dumoulin, and Aaron Courville. Film: Visual reasoning with a general conditioning layer. InProceedings of the AAAI conference on artificial intelligence, volume 32, 2018

    2018

  22. [22]

    Gnn-film: Graph neural networks with feature-wise linear modulation

    Marc Brockschmidt. Gnn-film: Graph neural networks with feature-wise linear modulation. InInternational Conference on Machine Learning, pages 1144–1152. PMLR, 2020

    2020

  23. [23]

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

    2023

  24. [24]

    Pytorch fsdp: experiences on scaling fully sharded data parallel.arXiv preprint arXiv:2304.11277, 2023

    Yanli Zhao, Andrew Gu, Rohan Varma, Liang Luo, Chien-Chin Huang, Min Xu, Less Wright, Hamid Shojanazeri, Myle Ott, Sam Shleifer, et al. Pytorch fsdp: experiences on scaling fully sharded data parallel.arXiv preprint arXiv:2304.11277, 2023. 13 CSU-PCAST: Medium-range Ensemble Precipitation Forecasting A Appendix A.1 Additional Results for Non-Precipitation...

    Pith/arXiv arXiv 2023

  25. [2023]

    The blue horizontal line denotes the GEFS baseline (0), while the red curves represent the relative BS of the CSU-PCAST model compared with GEFS

    Panels (a–c) show BS differences at precipitation thresholds of 5 mm, 10 mm, and 20 mm, respectively, for the GFS-initialized experiment; panels (d–f) show the corresponding results for the ERA5-initialized experiment. The blue horizontal line denotes the GEFS baseline (0), while the red curves represent the relative BS of the CSU-PCAST model compared wit...

    2023