pith. sign in

arxiv: 2605.16163 · v1 · pith:5HUPYUGZnew · submitted 2026-05-15 · ⚛️ physics.ao-ph · cs.LG

SwAIther-Precip: Lead-Time-Aware Bias Correction Enables Kilometer-Scale Downscaling of Global AI Precipitation Forecasts over Switzerland

Dan Assouline , Erwan Koch , Federico Amato , Filippo Quarenghi , Daniele Nerini , Thibaut Loiseau , Kyle van de Langemheen , Tom Beucler This is my paper

Pith reviewed 2026-05-19 17:40 UTC · model grok-4.3

classification ⚛️ physics.ao-ph cs.LG
keywords precipitation downscalingAI weather forecastingbias correctiondiffusion modelsU-NetSwitzerlandlead-time dependencekilometer-scale forecasts
0
0 comments X

The pith

A lead-time-conditioned U-Net bias correction enables a diffusion model to produce kilometer-scale probabilistic precipitation forecasts from global AI outputs with 48% lower CRPS over Switzerland.

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

The paper presents SwAIther-Precip, a two-stage framework that first corrects lead-time-dependent biases in coarse 0.25-degree AIFS precipitation forecasts and then adds realistic small-scale variability. A U-Net uses feature-wise linear modulation to adjust systematic errors at the coarse scale, after which a diffusion model generates fine details by training directly on radar-gauge observations. This separation yields forecasts that match observed spatial patterns more closely than raw global outputs while remaining computationally efficient. The method maintains skill through 5-day lead times and improves further at longer ranges when trained across all lead times. Accurate local precipitation at kilometer scales supports better hazard assessment in mountainous areas where unresolved terrain effects dominate.

Core claim

SwAIther-Precip converts coarse-resolution AIFS forecasts into probabilistic km-scale precipitation fields over Switzerland by first applying a lead-time-conditioned U-Net for deterministic bias correction at coarse resolution, which then allows a diffusion model to generate fine-scale variability independently of lead time. Using AIFS forecasts and CombiPrecip observations, the approach reduces CRPS by 48% relative to raw AIFS, reproduces observed spatial variability with spectral fidelity above 0.85 at large scales and 0.88 at small scales, and achieves an effective resolution of approximately 4 km on a 1 km grid for lead times up to 5 days. Training across lead times further improves long

What carries the argument

A deterministic U-Net bias correction at coarse resolution conditioned on lead time via feature-wise linear modulation, which enables a subsequent generative super-resolution diffusion model to train directly on high-resolution observations rather than full atmospheric states.

If this is right

  • SwAIther-Precip reduces CRPS by 48% relative to raw AIFS forecasts for lead times up to 5 days.
  • Generated fields achieve spectral fidelity above 0.85 at large scales and 0.88 at small scales, corresponding to roughly 4 km effective resolution on a 1 km grid.
  • Training the system across all lead times produces a 13% CRPS reduction at 6 days compared with lead-time-specific models.
  • The two-stage separation allows the generative super-resolution stage to operate without the full atmospheric state from the global model.

Where Pith is reading between the lines

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

  • The same lead-time conditioning approach could be tested on other surface variables or over different complex-terrain regions to check whether the bias-correction benefit generalizes.
  • Separating deterministic correction from generative detail addition may lower the data requirements for operational downscaling systems that lack complete reanalysis archives.
  • If the effective 4 km resolution holds in real-time deployment, local hazard models could ingest these fields directly instead of relying on statistical post-processing of coarser inputs.

Load-bearing premise

That a deterministic U-Net bias correction conditioned on lead time at coarse resolution is sufficient to allow a subsequent generative super-resolution model to be trained directly on observations without needing the full atmospheric state.

What would settle it

Applying the full pipeline to a new global AI model or an independent multi-year verification period and finding no CRPS reduction or spectral fidelity below 0.8 at small scales would indicate that the lead-time correction does not reliably enable the observed improvements.

Figures

Figures reproduced from arXiv: 2605.16163 by Dan Assouline, Daniele Nerini, Erwan Koch, Federico Amato, Filippo Quarenghi, Kyle van de Langemheen, Thibaut Loiseau, Tom Beucler.

Figure 7
Figure 7. Figure 7: At 6h, the specialist closely matches SwAIther [PITH_FULL_IMAGE:figures/full_fig_p015_7.png] view at source ↗
read the original abstract

Skillful medium-range precipitation forecasting at kilometer scale remains challenging over complex terrain because precipitation arises from multiscale nonlinear processes that global models cannot explicitly resolve at affordable cost. Global AI weather models can produce skillful medium-range forecasts, but their native 0.25 degrees resolution limits direct use for local hazard applications. Statistical downscaling can help bridge this gap, yet existing approaches often struggle with state-dependent, and especially lead-time-dependent, biases in global forecasts. We introduce SwAIther-Precip, a lead-time-aware downscaling framework that converts coarse-resolution AIFS forecasts into probabilistic km-scale precipitation fields over Switzerland. First, a U-Net conditioned on lead time via feature-wise linear modulation deterministically corrects systematic biases at coarse resolution. This targeted correction enables a cheaper super-resolution stage conditioned only on corrected precipitation, allowing direct training on observations rather than on the full atmospheric state. A diffusion-based model then generates fine-scale spatial variability independently of lead time. Using AIFS forecasts and CombiPrecip radar-gauge observations, SwAIther-Precip reduces CRPS by 48% relative to raw AIFS. The generated fields reproduce observed spatial variability with spectral fidelity above 0.85 at large scales and 0.88 at small scales, corresponding to an effective resolution of approximately 4 km on a 1 km grid for lead times up to 5 days. Training across lead times further improves long-range performance, yielding a 13% CRPS reduction at 6 days relative to lead-time-specific models. These results show that explicitly correcting lead-time-dependent biases before generative super-resolution is key to efficient km-scale probabilistic downscaling of global AI precipitation forecasts.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

1 major / 2 minor

Summary. The manuscript introduces SwAIther-Precip, a two-stage lead-time-aware downscaling framework that converts coarse (0.25°) AIFS precipitation forecasts into probabilistic 1 km fields over Switzerland. Stage 1 applies a deterministic U-Net with FiLM conditioning on lead time to correct systematic biases at coarse resolution; Stage 2 uses a diffusion-based generative super-resolution model conditioned only on the corrected coarse field and trained directly on CombiPrecip radar-gauge observations. The paper reports a 48% CRPS reduction versus raw AIFS, spectral fidelity >0.85 (large scales) and >0.88 (small scales) corresponding to ~4 km effective resolution, and a 13% CRPS gain at 6-day lead times from multi-lead training.

Significance. If the central results hold, the work demonstrates an efficient route to kilometer-scale probabilistic precipitation forecasting from global AI models by separating lead-time-dependent bias correction from generative downscaling. This avoids the need for full atmospheric state variables in the super-resolution stage and enables direct training on observations, which is practically valuable for complex terrain and hazard applications.

major comments (1)
  1. The central claim that the lead-time-conditioned U-Net correction removes systematic biases sufficiently for the subsequent lead-time-independent diffusion model to be trained directly on observations (without full atmospheric state or explicit orography) is load-bearing for both the 48% CRPS reduction and the reported spectral fidelity. The manuscript should provide explicit residual diagnostics (e.g., lead-time-stratified power spectra or CRPS of the corrected coarse field versus observations) to demonstrate that lead-time-specific structure is not passed to the generative stage.
minor comments (2)
  1. Clarify the exact definition and computation of the spectral fidelity metric (e.g., whether it is a normalized power-spectrum correlation or a scale-dependent RMSE) and report it with uncertainty estimates across the validation period.
  2. The abstract states an effective resolution of approximately 4 km on a 1 km grid; include a supporting figure or table showing the scale at which the generated spectra diverge from observations.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their constructive comments and positive assessment of the practical value of our two-stage framework. We address the major comment below and have revised the manuscript to incorporate the requested diagnostics.

read point-by-point responses

  1. Referee: The central claim that the lead-time-conditioned U-Net correction removes systematic biases sufficiently for the subsequent lead-time-independent diffusion model to be trained directly on observations (without full atmospheric state or explicit orography) is load-bearing for both the 48% CRPS reduction and the reported spectral fidelity. The manuscript should provide explicit residual diagnostics (e.g., lead-time-stratified power spectra or CRPS of the corrected coarse field versus observations) to demonstrate that lead-time-specific structure is not passed to the generative stage.

    Authors: We agree that explicit residual diagnostics are needed to substantiate the load-bearing claim. In the revised manuscript we have added lead-time-stratified CRPS and power-spectrum comparisons of the U-Net-corrected coarse fields versus CombiPrecip observations (new Figure 4 and accompanying text in Section 3.2). These show that the lead-time-conditioned correction reduces CRPS by 30-40% relative to raw AIFS across all leads, brings the power spectra into closer agreement with observations at scales >10 km, and leaves residuals without strong lead-time-dependent structure. Because the diffusion model is trained directly on these corrected fields paired with observations, the generative stage adds realistic small-scale variability without inheriting uncorrected lead-time biases. We have also clarified the rationale for omitting full atmospheric state and explicit orography from the super-resolution stage. These additions directly address the referee's concern while preserving the efficiency of the two-stage design. revision: yes

Circularity Check

0 steps flagged

No circularity: staged architecture validated on independent observations

full rationale

The paper describes a two-stage pipeline in which a lead-time-conditioned U-Net first removes systematic biases from coarse AIFS precipitation forecasts, after which a diffusion model performs super-resolution trained directly on CombiPrecip radar-gauge observations. All reported metrics (48% CRPS reduction, spectral fidelity >0.85/0.88) are computed against the raw AIFS input and the same independent observational dataset; no equation reduces to a fitted parameter by construction, no uniqueness theorem is imported from prior self-work, and no ansatz is smuggled via self-citation. The derivation chain is therefore self-contained and externally falsifiable.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The framework rests on the modeling choice that lead-time-dependent biases can be corrected deterministically at coarse scale before generative modeling, with training across lead times assumed to improve generalization.

free parameters (1)
  • lead-time conditioning parameters in U-Net
    Feature-wise linear modulation introduces parameters that are fitted to capture lead-time-dependent bias patterns.
axioms (1)
  • domain assumption Global AI forecast biases are primarily systematic and can be corrected deterministically at coarse resolution using lead-time conditioning.
    This underpins the decision to use a deterministic U-Net stage before the generative super-resolution.

pith-pipeline@v0.9.0 · 5874 in / 1399 out tokens · 62970 ms · 2026-05-19T17:40:28.399578+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

67 extracted references · 67 canonical work pages · 2 internal anchors

  1. [1]

    Communications Earth & Environment , volume=

    Residual corrective diffusion modeling for km-scale atmospheric downscaling , author=. Communications Earth & Environment , volume=. 2025 , publisher=

    work page 2025

  2. [2]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Film: Visual reasoning with a general conditioning layer , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

    work page

  3. [3]

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

    Deep residual learning for image recognition , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    work page

  4. [4]

    arXiv preprint arXiv:2410.07472 , year=

    Exploring the design space of deep-learning-based weather forecasting systems , author=. arXiv preprint arXiv:2410.07472 , year=

    work page arXiv

  5. [5]

    Atmospheric Science Letters , volume=

    One-kilometer resolution forecasts of hourly precipitation over China using machine learning models , author=. Atmospheric Science Letters , volume=. 2025 , publisher=

    work page 2025

  6. [6]

    EnScale: Temporally-consistent multivariate generative downscaling via proper scoring rules

    EnScale: Temporally-consistent multivariate generative downscaling via proper scoring rules , author=. arXiv preprint arXiv:2509.26258 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  7. [7]

    Monthly Weather Review , volume=

    Scale-selective verification of rainfall accumulations from high-resolution forecasts of convective events , author=. Monthly Weather Review , volume=. 2008 , publisher=

    work page 2008

  8. [8]

    Quarterly Journal of the Royal Meteorological Society , year=

    A consistent definition of the fractions skill score for ensemble forecasts , author=. Quarterly Journal of the Royal Meteorological Society , year=

    work page

  9. [9]

    Monthly Weather Review , volume=

    SAL—A novel quality measure for the verification of quantitative precipitation forecasts , author=. Monthly Weather Review , volume=. 2008 , publisher=

    work page 2008

  10. [10]

    Measuring precipitation from space , pages=

    Methods for verifying satellite precipitation estimates , author=. Measuring precipitation from space , pages=. 2007 , publisher=

    work page 2007

  11. [11]

    arXiv preprint arXiv:2406.01465 , year=

    AIFS--ECMWF's data-driven forecasting system , author=. arXiv preprint arXiv:2406.01465 , year=

    work page arXiv

  12. [12]

    arXiv preprint arXiv:2509.18994 , year=

    An update to ECMWF's machine-learned weather forecast model AIFS , author=. arXiv preprint arXiv:2509.18994 , year=

    work page arXiv

  13. [13]

    Environmental modelling & software , volume=

    Automated regression-based statistical downscaling tool , author=. Environmental modelling & software , volume=. 2008 , publisher=

    work page 2008

  14. [14]

    Environmental Modelling & Software , volume=

    SDSM—a decision support tool for the assessment of regional climate change impacts , author=. Environmental Modelling & Software , volume=. 2002 , publisher=

    work page 2002

  15. [15]

    Proceedings of the 26th annual international conference on machine learning , pages=

    Curriculum learning , author=. Proceedings of the 26th annual international conference on machine learning , pages=

    work page

  16. [16]

    arXiv preprint arXiv:2003.12140 , year=

    Metnet: A neural weather model for precipitation forecasting , author=. arXiv preprint arXiv:2003.12140 , year=

    work page arXiv 2003

  17. [17]

    Nature communications , volume=

    Deep learning for twelve hour precipitation forecasts , author=. Nature communications , volume=. 2022 , publisher=

    work page 2022

  18. [18]

    arXiv preprint arXiv:2410.05431 , year=

    Continuous ensemble weather forecasting with diffusion models , author=. arXiv preprint arXiv:2410.05431 , year=

    work page arXiv

  19. [19]

    Management Science , volume=

    Scoring rules for continuous probability distributions , author=. Management Science , volume=. 1976 , publisher=

    work page 1976

  20. [20]

    Weather and Forecasting , volume=

    Decomposition of the continuous ranked probability score for ensemble prediction systems , author=. Weather and Forecasting , volume=

    work page

  21. [21]

    Journal of the American Statistical Association , volume=

    Strictly proper scoring rules, prediction, and estimation , author=. Journal of the American Statistical Association , volume=. 2007 , publisher=

    work page 2007

  22. [22]

    Journal of the Royal Statistical Society: Series B , volume=

    Probabilistic forecasts, calibration and sharpness , author=. Journal of the Royal Statistical Society: Series B , volume=. 2007 , publisher=

    work page 2007

  23. [23]

    Journal of Hydrometeorology , volume=

    Why should ensemble spread match the RMSE of the ensemble mean? , author=. Journal of Hydrometeorology , volume=

    work page

  24. [24]

    Journal of Climate , volume=

    A method for producing and evaluating probabilistic forecasts from ensemble model integrations , author=. Journal of Climate , volume=

    work page

  25. [25]

    Monthly Weather Review , volume=

    Interpretation of rank histograms for verifying ensemble forecasts , author=. Monthly Weather Review , volume=

    work page

  26. [26]

    Monthly Weather Review , volume=

    On the rank histogram of ensemble forecasts , author=. Monthly Weather Review , volume=

    work page

  27. [27]

    Monthly Weather Review , volume=

    Verification of forecasts expressed in terms of probability , author=. Monthly Weather Review , volume=

    work page

  28. [28]

    Econometrica , volume=

    Regression quantiles , author=. Econometrica , volume=. 1978 , publisher=

    work page 1978

  29. [29]

    Monthly Weather Review , volume=

    Variogram-based proper scoring rules for probabilistic forecasts of multivariate quantities , author=. Monthly Weather Review , volume=

    work page

  30. [30]

    Monthly Weather Review , volume=

    Spatial-temporal structure of ensemble forecast uncertainties , author=. Monthly Weather Review , volume=. 2013 , publisher=

    work page 2013

  31. [31]

    Weather and Forecasting , volume=

    Verification of spatial precipitation forecasts using the fractions skill score: the ensemble case , author=. Weather and Forecasting , volume=. 2018 , publisher=

    work page 2018

  32. [32]

    Journal of the Royal Statistical Society: Series A (General) , volume=

    Present position and potential developments: Some personal views statistical theory the prequential approach , author=. Journal of the Royal Statistical Society: Series A (General) , volume=. 1984 , publisher=

    work page 1984

  33. [33]

    arXiv preprint arXiv:2304.12891 , year=

    Latent diffusion models for generative precipitation nowcasting with accurate uncertainty quantification , author=. arXiv preprint arXiv:2304.12891 , year=

    work page arXiv

  34. [34]

    Journal of Advances in Modeling Earth Systems , volume=

    Diffesm: Conditional emulation of temperature and precipitation in earth system models with 3d diffusion models , author=. Journal of Advances in Modeling Earth Systems , volume=. 2024 , publisher=

    work page 2024

  35. [35]

    Hydrology and Earth System Sciences , volume=

    Empirical Mode Decomposition in 2-D space and time: a tool for space-time rainfall analysis and nowcasting , author=. Hydrology and Earth System Sciences , volume=. 2005 , publisher=

    work page 2005

  36. [36]

    0) , author=

    Pysteps: An open-source Python library for probabilistic precipitation nowcasting (v1. 0) , author=. Geoscientific Model Development , volume=. 2019 , publisher=

    work page 2019

  37. [37]

    Atmospheric Science Letters , volume=

    Effective resolution in high resolution global atmospheric models for climate studies , author=. Atmospheric Science Letters , volume=. 2020 , publisher=

    work page 2020

  38. [38]

    Monthly weather review , volume=

    Evaluating mesoscale NWP models using kinetic energy spectra , author=. Monthly weather review , volume=

    work page

  39. [39]

    Science Advances , volume =

    Jaideep Pathak and Yair Cohen and Piyush Garg and Peter Harrington and Noah Brenowitz and Dale Durran and Morteza Mardani and Arash Vahdat and Shaoming Xu and Karthik Kashinath and Michael Pritchard , title =. Science Advances , volume =. 2026 , doi =. https://www.science.org/doi/pdf/10.1126/sciadv.adv0423 , abstract =

    work page doi:10.1126/sciadv.adv0423 2026

  40. [40]

    2024 , eprint=

    AIFS -- ECMWF's data-driven forecasting system , author=. 2024 , eprint=

    work page 2024

  41. [41]

    2025 , eprint=

    CRPS-LAM: Regional ensemble weather forecasting from matching marginals , author=. 2025 , eprint=

    work page 2025

  42. [42]

    2025 , eprint=

    Building Machine Learning Limited Area Models: Kilometer-Scale Weather Forecasting in Realistic Settings , author=. 2025 , eprint=

    work page 2025

  43. [43]

    2023 , eprint=

    Debias Coarsely, Sample Conditionally: Statistical Downscaling through Optimal Transport and Probabilistic Diffusion Models , author=. 2023 , eprint=

    work page 2023

  44. [44]

    2014 international weather radar and hydrology symposium , pages=

    The CombiPrecip experience: development and operation of a real-time radar-raingauge combination scheme in Switzerland , author=. 2014 international weather radar and hydrology symposium , pages=

    work page 2014

  45. [45]

    Smith and Sergey Frolov and Montgomery Flora and Corey Potvin

    Daniel Abdi and Isidora Jankov and Paul Madden and Vanderlei Vargas and Timothy A. Smith and Sergey Frolov and Montgomery Flora and Corey Potvin. HRRRCast: A Data-Driven Emulator for Regional Weather Forecasting at Convection-Allowing Scales. Artificial Intelligence for the Earth Systems. 2026. doi:10.1175/AIES-D-25-0061.1

    work page doi:10.1175/aies-d-25-0061.1 2026

  46. [46]

    Nature , volume=

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

    work page 2023

  47. [47]

    Science , volume =

    Remi Lam and Alvaro Sanchez-Gonzalez and Matthew Willson and Peter Wirnsberger and Meire Fortunato and Ferran Alet and Suman Ravuri and Timo Ewalds and Zach Eaton-Rosen and Weihua Hu and Alexander Merose and Stephan Hoyer and George Holland and Oriol Vinyals and Jacklynn Stott and Alexander Pritzel and Shakir Mohamed and Peter Battaglia , title =. Science...

    work page 2023

  48. [48]

    Journal of Advances in Modeling Earth Systems , volume =

    Rasp, Stephan and Hoyer, Stephan and Merose, Alexander and Langmore, Ian and Battaglia, Peter and Russell, Tyler and Sanchez-Gonzalez, Alvaro and Yang, Vivian and Carver, Rob and Agrawal, Shreya and Chantry, Matthew and Ben Bouallegue, Zied and Dueben, Peter and Bromberg, Carla and Sisk, Jared and Barrington, Luke and Bell, Aaron and Sha, Fei , title =. J...

    work page doi:10.1029/2023ms004019

  49. [49]

    Nature , volume=

    Probabilistic weather forecasting with machine learning , author=. Nature , volume=. 2025 , publisher=

    work page 2025

  50. [50]

    Communications Earth & Environment , volume=

    An artificial intelligence-based limited area model for forecasting of surface meteorological variables , author=. Communications Earth & Environment , volume=. 2025 , publisher=

    work page 2025

  51. [51]

    Regional data-driven weather modeling with a global stretched-grid

    Thomas Nils Nipen and Håvard Homleid Haugen and Magnus Sikora Ingstad and Even Marius Nordhagen and Aram Farhad Shafiq Salihi and Paulina Tedesco and Ivar Ambjørn Seierstad and Jørn Kristiansen and Simon Lang and Mihai Alexe and Jesper Dramsch and Baudouin Raoult and Gert Mertes and Matthew Chantry. Regional data-driven weather modeling with a global stre...

    work page doi:10.1175/aies-d-25-0001.1 2025

  52. [52]

    2025 , editor =

    Gao, Yuan and Wu, Hao and Shu, Ruiqi and Dong, Huanshuo and Xu, Fan and Chen, Rui Ray and Yan, Yibo and Wen, Qingsong and Hu, Xuming and Wang, Kun and Wu, Jiahao and Qing, Li and Xiong, Hui and Huang, Xiaomeng , booktitle =. 2025 , editor =

    work page 2025

  53. [53]

    Environmental Research Letters , volume=

    Improvement of disastrous extreme precipitation forecasting in North China by Pangu-weather AI-driven regional WRF model , author=. Environmental Research Letters , volume=. 2024 , publisher=

    work page 2024

  54. [54]

    arXiv preprint arXiv:2602.18646 , year=

    AI-Based Regional Emulation for Kilometer-Scale Dynamical Downscaling , author=. arXiv preprint arXiv:2602.18646 , year=

    work page arXiv

  55. [55]

    STCast: Adaptive Boundary Alignment for Global and Regional Weather Forecasting

    Stcast: Adaptive boundary alignment for global and regional weather forecasting , author=. arXiv preprint arXiv:2509.25210 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  56. [56]

    Improving Precipitation Forecasts with Convolutional Neural Networks

    Anirudhan Badrinath and Luca Delle Monache and Negin Hayatbini and Will Chapman and Forest Cannon and Marty Ralph. Improving Precipitation Forecasts with Convolutional Neural Networks. Weather and Forecasting. 2023. doi:10.1175/WAF-D-22-0002.1

    work page doi:10.1175/waf-d-22-0002.1 2023

  57. [57]

    and Langguth, M

    Rojas-Campos, A. and Langguth, M. and Wittenbrink, M. and Pipa, G. , TITLE =. Geoscientific Model Development , VOLUME =. 2023 , NUMBER =

    work page 2023

  58. [58]

    Harris, Lucy and McRae, Andrew T. T. and Chantry, Matthew and Dueben, Peter D. and Palmer, Tim N. , title =. Journal of Advances in Modeling Earth Systems , volume =. doi:https://doi.org/10.1029/2022MS003120 , url =. https://agupubs.onlinelibrary.wiley.com/doi/pdf/10.1029/2022MS003120 , note =

    work page doi:10.1029/2022ms003120

  59. [59]

    Proceedings of The 25th International Conference on Artificial Intelligence and Statistics , pages =

    Increasing the accuracy and resolution of precipitation forecasts using deep generative models , author =. Proceedings of The 25th International Conference on Artificial Intelligence and Statistics , pages =. 2022 , editor =

    work page 2022

  60. [60]

    Antonio, Bobby and McRae, Andrew T. T. and MacLeod, David and Cooper, Fenwick C. and Marsham, John and Aitchison, Laurence and Palmer, Tim N. and Watson, Peter A. G. , title =. Journal of Advances in Modeling Earth Systems , volume =. doi:https://doi.org/10.1029/2024MS004796 , url =. https://agupubs.onlinelibrary.wiley.com/doi/pdf/10.1029/2024MS004796 , note =

    work page doi:10.1029/2024ms004796

  61. [61]

    and Jing, H

    Sun, H. and Jing, H. and Dai, Z. and Xiao, S. and Xue, W. and Sun, J. and Lu, Q. , TITLE =. EGUsphere , VOLUME =. 2026 , PAGES =

    work page 2026

  62. [62]

    Enhancing the Spatial Resolution of Medium-Range Precipitation Forecasts Using Super-Resolution Neural Networks

    Ryosuke Arai and Takahiro Sato and Masahiro Imamura. Enhancing the Spatial Resolution of Medium-Range Precipitation Forecasts Using Super-Resolution Neural Networks. Weather and Forecasting. 2025. doi:10.1175/WAF-D-24-0217.1

    work page doi:10.1175/waf-d-24-0217.1 2025

  63. [63]

    2026 , eprint=

    Universal Diffusion-Based Probabilistic Downscaling , author=. 2026 , eprint=

    work page 2026

  64. [64]

    Nature , volume=

    Skilful precipitation nowcasting using deep generative models of radar , author=. Nature , volume=. 2021 , publisher=

    work page 2021

  65. [65]

    Earth's Future , volume =

    Hobeichi, Sanaa and Nishant, Nidhi and Shao, Yawen and Abramowitz, Gab and Pitman, Andy and Sherwood, Steve and Bishop, Craig and Green, Samuel , title =. Earth's Future , volume =. doi:https://doi.org/10.1029/2022EF003291 , url =. https://agupubs.onlinelibrary.wiley.com/doi/pdf/10.1029/2022EF003291 , note =

    work page doi:10.1029/2022ef003291

  66. [66]

    Hydrology , VOLUME =

    Cai, Ninghao and Sun, Hongchuan and Yan, Pengcheng , TITLE =. Hydrology , VOLUME =. 2024 , NUMBER =

    work page 2024

  67. [67]

    Water Resources Management , volume=

    Statistical post-processing for precipitation forecast through deep learning coupling large-scale and local-scale spatiotemporal information , author=. Water Resources Management , volume=. 2025 , publisher=

    work page 2025