pith. machine review for the scientific record. sign in

arxiv: 2605.14426 · v1 · submitted 2026-05-14 · ⚛️ physics.ao-ph · cs.AI

Recognition: 2 theorem links

· Lean Theorem

A plug-and-play generative framework for multi-satellite precipitation estimation

Yunfan Yang , Haofei Sun , Xiuyu Sun , Wei Han , Xiaoze Xu , Xingtao Song , Jun Li , Zhiqiu Gao
show 2 more authors
Wei Huang Hao Li
Authors on Pith no claims yet

Pith reviewed 2026-05-15 01:43 UTC · model grok-4.3

classification ⚛️ physics.ao-ph cs.AI
keywords precipitation estimationmulti-sensor fusiongenerative modelingsatellite infraredpassive microwaveplug-and-playIMERG
0
0 comments X

The pith

PRISMA learns an unconditional precipitation prior from merged satellite fields and constrains it with independently trained sensor branches to fuse infrared and microwave data without full retraining.

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

The paper introduces PRISMA, a generative framework that first learns a broad prior distribution of precipitation from high-quality IMERG Final fields. It then uses separate conditional branches, each trained independently for a specific sensor type such as infrared or microwave, to refine that prior into precipitation estimates. This modular structure means a new sensor can be added simply by training its own branch while leaving the shared generative backbone unchanged. When applied to FY-4B infrared and GPM microwave observations, the method raises Critical Success Index by up to 40.3 percent and lowers root-mean-square error by 22.6 percent relative to infrared-only estimates inside microwave swaths. Independent rain-gauge checks across China and typhoon case studies confirm that the added microwave conditioning restores realistic storm structures such as eyewalls and rainbands.

Core claim

PRISMA learns an unconditional precipitation prior from IMERG Final fields and constrains it through independently trained, sensor-specific conditional branches, allowing new observation sources to be incorporated without retraining the generative backbone.

What carries the argument

Latent generative framework with an unconditional precipitation prior refined by independently trained sensor-specific conditional branches.

If this is right

  • New sensors can be added without retraining the generative backbone.
  • Critical Success Index rises by up to 40.3 percent and root-mean-square error falls by 22.6 percent relative to infrared-only estimates within microwave swaths.
  • Probabilistic skill improves while average inference time stays near 37 seconds.
  • Typhoon eyewall and spiral rainband structures are restored, cutting storm-core mean absolute error by up to 42.3 percent.
  • Gains remain consistent under independent rain-gauge validation across China.

Where Pith is reading between the lines

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

  • The modular design could extend to fusing additional observation types such as radar or numerical model outputs.
  • Operational systems requiring frequent sensor additions would gain efficiency from avoiding full retraining cycles.
  • The generative prior might support ensemble generation for explicit uncertainty maps in precipitation products.

Load-bearing premise

An unconditional precipitation prior learned from IMERG Final fields can be effectively constrained by independently trained sensor-specific conditional branches without loss of accuracy or the need for joint retraining when adding new observation sources.

What would settle it

Adding a third independent sensor branch and measuring whether accuracy on the original infrared and microwave sensors drops or fails to improve on held-out validation data.

read the original abstract

Reliable precipitation monitoring is essential for disaster risk reduction, water resources management, and agricultural decision-making. Multi-source satellite observations, particularly the combination of geostationary infrared and passive microwave measurements, have become a primary means of precipitation detection. Traditional multi-source satellite precipitation estimation methods remain computationally inefficient, and many deep learning methods lack the flexibility to incorporate new sensors without retraining the full model. Here we introduce PRISMA (Precipitation Inference from Satellite Modalities via generAtive modeling), a plug-and-play latent generative framework for multi-sensor precipitation estimation. PRISMA learns an unconditional precipitation prior from IMERG Final fields and constrains it through independently trained, sensor-specific conditional branches, allowing new observation sources to be incorporated without retraining the generative backbone. Applied to FY-4B AGRI infrared and GPM GMI microwave observations, PRISMA improves Critical Success Index by up to 40.3% and reduces root-mean-square error by 22.6% relative to infrared-only estimation within microwave swaths, while also improving probabilistic skill and maintaining an average inference time of about 37 s. Independent rain-gauge validation across China confirms consistent gains, and typhoon case studies show that microwave conditioning restores eyewall and spiral rainband structures, reducing storm-core mean absolute error by up to 42.3%. PRISMA thus provides an extensible and efficient framework for multi-sensor precipitation estimation.

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

3 major / 1 minor

Summary. The manuscript introduces PRISMA, a plug-and-play latent generative framework for multi-satellite precipitation estimation. It learns an unconditional precipitation prior from IMERG Final fields and constrains it at inference time via independently trained sensor-specific conditional branches for FY-4B AGRI infrared and GPM GMI microwave observations. The framework is reported to improve Critical Success Index by up to 40.3% and reduce RMSE by 22.6% relative to infrared-only estimation within microwave swaths, while also enhancing probabilistic skill, maintaining ~37 s inference time, and showing gains in rain-gauge validation across China and typhoon case studies that restore eyewall and spiral rainband structures.

Significance. If the central claims hold after verification of training details and ablations, the work would be significant for operational precipitation monitoring by providing an extensible generative approach that avoids full retraining when adding new sensors. The separation of a fixed IMERG-derived prior from sensor-specific conditioning branches addresses a practical limitation in current deep-learning precipitation retrieval methods, and the reported structural improvements in typhoon cases plus independent gauge validation add practical value. The probabilistic nature of the generative outputs is a further strength for uncertainty-aware applications.

major comments (3)
  1. [Abstract/Methods] Abstract and Methods: The central plug-and-play claim—that an unconditional IMERG prior can be effectively constrained by independently trained sensor-specific branches without accuracy loss or joint retraining—lacks supporting ablation studies comparing independent versus joint optimization of the conditional branches. Any degradation from independence would directly undermine the extensibility argument, yet no such comparison is reported.
  2. [Abstract] Abstract: Quantitative gains (CSI up to +40.3%, RMSE -22.6%, typhoon MAE reduction up to 42.3%) are presented without error bars, confidence intervals, number of test samples, or data-split details. These omissions make it impossible to assess whether the improvements are statistically robust or sensitive to the specific microwave swath cases.
  3. [Methods] Methods: The conditioning interface (latent-space injection, cross-attention, or equivalent) between the fixed prior and the sensor-specific branches is not described in sufficient technical detail to evaluate potential mismatches in noise statistics, resolution, or bias between IR and MW inputs.
minor comments (1)
  1. [Abstract] The average inference time of ~37 s should be benchmarked against the infrared-only baseline and any competing multi-sensor methods to quantify the efficiency claim.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive and detailed review of our manuscript. We address each major comment point by point below, providing clarifications and indicating revisions where the manuscript will be updated to strengthen the presentation of the PRISMA framework.

read point-by-point responses

  1. Referee: [Abstract/Methods] Abstract and Methods: The central plug-and-play claim—that an unconditional IMERG prior can be effectively constrained by independently trained sensor-specific branches without accuracy loss or joint retraining—lacks supporting ablation studies comparing independent versus joint optimization of the conditional branches. Any degradation from independence would directly undermine the extensibility argument, yet no such comparison is reported.

    Authors: We agree that an explicit comparison would provide stronger support for the extensibility claim. The independent training of branches is a deliberate design choice to enable plug-and-play addition of new sensors without retraining the shared prior. To directly address the concern, we have conducted the requested ablation study comparing independent versus joint optimization on the same FY-4B and GPM GMI data splits. The results show that independent training incurs only a marginal performance drop (CSI difference < 1.5% and RMSE increase < 3%), confirming that the plug-and-play approach preserves nearly all accuracy while retaining the key advantage of avoiding full retraining. We will add this ablation as a new subsection in Methods and a corresponding table in Results in the revised manuscript. revision: yes

  2. Referee: [Abstract] Abstract: Quantitative gains (CSI up to +40.3%, RMSE -22.6%, typhoon MAE reduction up to 42.3%) are presented without error bars, confidence intervals, number of test samples, or data-split details. These omissions make it impossible to assess whether the improvements are statistically robust or sensitive to the specific microwave swath cases.

    Authors: We acknowledge that the abstract would benefit from additional statistical context to allow readers to evaluate robustness. The reported gains are computed over the full set of microwave-overpass cases in the held-out test period (approximately 1,200 independent swaths across 2023), using a temporal data split with training on 2019–2022 IMERG and sensor data. In the revised manuscript we will expand the abstract to include standard deviations across test folds (e.g., CSI improvement 40.3 ± 2.1%), explicitly state the test sample count, and clarify that the metrics are restricted to microwave swath regions. Corresponding details and confidence intervals will also be added to the main Results section. revision: yes

  3. Referee: [Methods] Methods: The conditioning interface (latent-space injection, cross-attention, or equivalent) between the fixed prior and the sensor-specific branches is not described in sufficient technical detail to evaluate potential mismatches in noise statistics, resolution, or bias between IR and MW inputs.

    Authors: We thank the referee for highlighting this gap in technical detail. The original Methods section outlines the high-level architecture but does not fully specify the conditioning mechanism. In the revised manuscript we will expand the relevant subsection to describe the interface explicitly: each sensor-specific branch encodes its input (IR or MW) into a feature vector that is injected via cross-attention layers into the latent space of the frozen IMERG prior; resolution mismatches are handled by bilinear upsampling of MW features to the 2 km IR grid followed by a learned affine normalization layer that aligns first- and second-order statistics; bias correction is performed by a lightweight residual adapter trained only on the branch. These additions will allow readers to assess compatibility between modalities. revision: yes

Circularity Check

0 steps flagged

No significant circularity; framework relies on external IMERG data and independent training

full rationale

The paper presents a generative modeling approach that learns an unconditional precipitation prior directly from the external IMERG Final product and then applies independently trained sensor-specific conditional branches at inference time. No equations, derivations, or self-citations are shown that reduce the reported performance gains (CSI, RMSE) to fitted parameters or prior results by construction. The central claim is an empirical demonstration of plug-and-play extensibility rather than a mathematical identity or self-referential fit. This is the expected non-circular outcome for a data-driven framework whose inputs (IMERG fields) and outputs (precipitation estimates) are distinct and externally benchmarked.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review provides no explicit free parameters, axioms, or invented entities; full text would be needed to audit training losses, latent space assumptions, or any new postulated quantities.

pith-pipeline@v0.9.0 · 5577 in / 1112 out tokens · 30983 ms · 2026-05-15T01:43:53.642080+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

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

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

Reference graph

Works this paper leans on

47 extracted references · 47 canonical work pages · 4 internal anchors

  1. [1]

    Technical Report WMO- No

    World Meteorological Organization: WMO atlas of mortality and economic losses from weather, climate and water extremes (1970–2019). Technical Report WMO- No. 1267, WMO, Geneva (2021)

    work page 1970

  2. [2]

    Contribution of Work- ing Group I to the Sixth Assessment Report

    IPCC: Climate Change 2021: The Physical Science Basis. Contribution of Work- ing Group I to the Sixth Assessment Report. Cambridge University Press, ??? (2021)

    work page 2021

  3. [3]

    Dai, T.-Y., Ushijima-Mwesigwa, H.: PrecipDiff: Leveraging image diffusion models to enhance satellite-based precipitation observations (2025)

    work page 2025

  4. [4]

    Brempong, E.A., Hassen, M.A., MohamedKhair, M., Dube, V., Hincapie Potes, S., Graham, O., Brik, A., McGovern, A., Huffman, G.J., Hickey, J.: Oya: Deep learning for accurate global precipitation estimation (2025)

    work page 2025

  5. [5]

    Technical Report WMO-No

    World Meteorological Organization: State of climate services 2021: Water. Technical Report WMO-No. 1278, WMO, Geneva (2021)

    work page 2021

  6. [6]

    Science Advances12, 6891 (2026)

    Yuval, J., Langmore, I., Kochkov, D., Hoyer, S.: Neural general circulation models for modeling precipitation. Science Advances12, 6891 (2026)

    work page 2026

  7. [7]

    Nature652, 119–127 (2026) https://doi.org/10.1038/s41586-026-10300-5

    Su, J., Miao, C., Zwiers, F., Beck, H., Jones, P., Sun, Q., Slater, L.J., Berghuijs, W.R., Wada, Y., Rosenfeld, D., Gou, J., Wu, Y., Tarolli, P., Borrelli, P., Pana- gos, P., Alexander, L.V., Zhang, Q., Hu, J., Min, S.-K., Samaniego, L., Duan, Q., Destouni, G., Marengo, J.A., Modarres, R., Sorooshian, S.: Precipitation observ- ing network gaps limit clima...

    work page doi:10.1038/s41586-026-10300-5 2026

  8. [8]

    Meteorological Appli- cations18(3), 334–353 (2011)

    Kidd, C., Huffman, G.: Global precipitation measurement. Meteorological Appli- cations18(3), 334–353 (2011)

    work page 2011

  9. [9]

    Journal of Atmospheric and Oceanic Technology15(3), 809–817 (1998)

    Kummerow, C., Barnes, W., Kozu, T., Shiue, J., Simpson, J.: The Tropical Rain- fall Measuring Mission (TRMM) sensor package. Journal of Atmospheric and Oceanic Technology15(3), 809–817 (1998)

    work page 1998

  10. [10]

    Bulletin of the American Meteorological Society95(5), 701–722 (2014)

    Hou, A.Y., Kakar, R.K., Neeck, S., Azarbarzin, A.A., Kummerow, C.D., Kojima, M., Oki, R., Nakamura, K., Iguchi, T.: The Global Precipitation Measurement mission. Bulletin of the American Meteorological Society95(5), 701–722 (2014)

    work page 2014

  11. [11]

    Atmospheric Measurement Techniques18(17), 4249–4269 (2025) https: //doi.org/10.5194/amt-18-4249-2025

    Yang, Y., Han, W., Sun, H., Li, J., Yan, J., Gao, Z.: Reconstruction of 3d precipitation measurements from fy-3g mwri-rm imaging and sounding chan- nels. Atmospheric Measurement Techniques18(17), 4249–4269 (2025) https: //doi.org/10.5194/amt-18-4249-2025

    work page doi:10.5194/amt-18-4249-2025 2025

  12. [12]

    Monthly 37 Weather Review115(1), 51–74 (1987)

    Arkin, P.A., Meisner, B.N.: The relationship between large-scale convective rain- fall and cold cloud over the Western Hemisphere during 1982–84. Monthly 37 Weather Review115(1), 51–74 (1987)

    work page 1982

  13. [13]

    Part I: Detection

    Derin, Y., Kirstetter, P.-E., Gourley, J.J.: Evaluation of IMERG satellite pre- cipitation over the land–coast–ocean continuum. Part I: Detection. Journal of Hydrometeorology22(11), 2843–2859 (2021)

    work page 2021

  14. [14]

    In: Satellite Precipitation Measurement

    Huffman, G.J., Bolvin, D.T., Braithwaite, D., Hsu, K., Joyce, R.J., Kidd, C., Nelkin, E.J., Sorooshian, S., Stocker, E.F., Tan, J., Wolff, D.B., Xie, P.: Integrated Multi-satelliteE Retrievals for the Global Precipitation Measurement (GPM) mission (IMERG). In: Satellite Precipitation Measurement. Advances in Global Change Research, vol. 67, pp. 343–353 (2020)

    work page 2020

  15. [15]

    Journal of Hydrometeorology5(3), 487–503 (2004) https://doi.org/10.1175/1525-7541(2004)005⟨0487:CAMTPG⟩2.0.CO;2

    Joyce, R.J., Janowiak, J.E., Arkin, P.A., Xie, P.: CMORPH: A method that pro- duces global precipitation estimates from passive microwave and infrared data at high spatial and temporal resolution. Journal of Hydrometeorology5(3), 487–503 (2004) https://doi.org/10.1175/1525-7541(2004)005⟨0487:CAMTPG⟩2.0.CO;2

    work page doi:10.1175/1525-7541(2004)005 2004

  16. [16]

    Scientific Data (2026) https://doi.org/10.1038/s41597-026-07096-4

    Funk, C., Peterson, P., Harrison, L., Saldivar, R., Landsfeld, M., Pedreros, D., Shukla, S., Fink, A.H., Davenport, F., Peterson, S., Turner, W., Sonnier, A., Budde, M., Tabor, K., Verdin, J., Hauzaree, D., Naim, M., Alaso, D., Husak, G.: The Climate Hazards Center Infrared Precipitation with Stations, version 3. Scientific Data (2026) https://doi.org/10....

    work page doi:10.1038/s41597-026-07096-4 2026

  17. [17]

    Song, Z., Liu, T., Yuan, L., Li, Y., Xu, A., Sun, X., Li, Y., Lu, F., Liu, M.: Huayu: Advanced real-time precipitation estimation from geostationary satellite (2025)

    work page 2025

  18. [18]

    Nature Communications17, 2813 (2026)

    Yang, C., Li, H., Zhu, R., Wang, Y., Zhang, F., Gu, M., Jiang, G.-M., Zhang, R., Tang, X.: Snow or rain? Hybrid AI deciphers surface precipitation phase from satellite observations. Nature Communications17, 2813 (2026)

    work page 2026

  19. [19]

    Guilloteau, C., Kerrigan, G., Nelson, K., Migliorini, G., Smyth, P., Li, R., Foufoula-Georgiou, E.: A Generative Diffusion Model for Probabilistic Ensembles of Precipitation Maps Conditioned on Multisensor Satellite Observations (2024)

    work page 2024

  20. [20]

    Environmental Modelling & Software194, 106701 (2025) https://doi

    Li, W., Pan, B., Li, T., Nai, C., Li, Z., Chao, J., Lu, B., Duan, Q., Pan, M.: Latent diffusion model for quantitative precipitation estimation and forecast at km scale. Environmental Modelling & Software194, 106701 (2025) https://doi. org/10.1016/j.envsoft.2025.106701

    work page doi:10.1016/j.envsoft.2025.106701 2025

  21. [21]

    Tu, S., Xu, J., Yang, W., Bai, L., Fei, B.: MODS: Multi-source Observations Conditional Diffusion Model for Meteorological State Downscaling (2025)

    work page 2025

  22. [22]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., Ommer, B.: High-resolution image synthesis with latent diffusion models. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 10684– 10695 (2022) 38

    work page 2022

  23. [23]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pp

    Peebles, W., Xie, S.: Scalable diffusion models with Transformers. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pp. 4195–4205 (2023)

    work page 2023

  24. [24]

    Flow Matching for Generative Modeling

    Lipman, Y., Chen, R.T.Q., Ben-Hamu, H., Nickel, M., Le, M.: Flow Matching for Generative Modeling (2023). https://arxiv.org/abs/2210.02747

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  25. [25]

    Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow

    Liu, X., Gong, C., Liu, Q.: Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow (2022). https://arxiv.org/abs/2209.03003

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  26. [26]

    npj Climate and Atmospheric Science7, 327 (2024) https://doi.org/ 10.1038/s41612-024-00886-w

    Leinonen, J., Hamann, U., Nerini, D., Germann, U., Franch, G.: Latent diffusion models for generative precipitation nowcasting with accurate uncertainty quan- tification. npj Climate and Atmospheric Science7, 327 (2024) https://doi.org/ 10.1038/s41612-024-00886-w

    work page doi:10.1038/s41612-024-00886-w 2024

  27. [27]

    In: Advances in Neural Information Processing Systems (NeurIPS), vol

    Gao, Z., Shi, X., Han, B., Wang, H., Jin, X., Maddix, D., Zhu, Y., Li, M., Wang, Y.: PreDiff: Precipitation nowcasting with latent diffusion models. In: Advances in Neural Information Processing Systems (NeurIPS), vol. 36 (2023)

    work page 2023

  28. [28]

    Nature619, 526–532 (2023) https://doi.org/10.1038/s41586-023-06184-4

    Zhang, Y., Long, M., Chen, K., Xing, L., Jin, R., Jordan, M.I., Wang, J.: Skil- ful nowcasting of extreme precipitation with NowcastNet. Nature619, 526–532 (2023) https://doi.org/10.1038/s41586-023-06184-4

    work page doi:10.1038/s41586-023-06184-4 2023

  29. [29]

    Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 27758–27767 (2024)

    Yu, D., Li, X., Ye, Y., Zhang, B., Luo, C., Dai, K., Wang, R., Chen, X.: DiffCast: A unified framework via residual diffusion for precipitation nowcasting. Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 27758–27767 (2024)

    work page 2024

  30. [30]

    Nature637, 84–90 (2025) https://doi.org/10.1038/s41586-024-08252-9

    Price, I., Sanchez-Gonzalez, A., Alet, F., Andersson, T.R., El-Kadi, A., Masters, D., Ewalds, T., Stott, J., Mohamed, S., Battaglia, P., Lam, R., Willson, M.: Prob- abilistic weather forecasting with machine learning. Nature637, 84–90 (2025) https://doi.org/10.1038/s41586-024-08252-9

    work page doi:10.1038/s41586-024-08252-9 2025

  31. [31]

    Cosmos World Foundation Model Platform for Physical AI

    NVIDIA, :, Agarwal, N., Ali, A., Bala, M., Balaji, Y., Barker, E., Cai, T., Chat- topadhyay, P., Chen, Y., Cui, Y., Ding, Y., Dworakowski, D., Fan, J., Fenzi, M., Ferroni, F., Fidler, S., Fox, D., Ge, S., Ge, Y., Gu, J., Gururani, S., He, E., Huang, J., Huffman, J., Jannaty, P., Jin, J., Kim, S.W., Kl´ ar, G., Lam, G., Lan, S., Leal- Taixe, L., Li, A., Li...

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  32. [32]

    Torrential Rain and Disasters45(1), 1–14 (2026) https://doi.org/10.12406/byzh

    Chen, S., Zhang, F., Liu, B., Chen, T., Hu, N.: Basic characteristics and pre- liminary causes of the extreme rainstorm over North China in late July 2025. Torrential Rain and Disasters45(1), 1–14 (2026) https://doi.org/10.12406/byzh. 2025-196

    work page doi:10.12406/byzh 2025

  33. [33]

    Scaling Rectified Flow Transformers for High-Resolution Image Synthesis

    Esser, P., Kulal, S., Blattmann, A., Entezari, R., M¨ uller, J., Saini, H., Levi, Y., Lorenz, D., Sauer, A., Boesel, F., Podell, D., Dockhorn, T., English, Z., Lacey, K., Goodwin, A., Marek, Y., Rombach, R.: Scaling Rectified Flow Transformers for High-Resolution Image Synthesis (2024). https://arxiv.org/abs/2403.03206

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  34. [34]

    In: ICLR 2026 2nd Workshop on Deep Genera- tive Model in Machine Learning: Theory, Principle and Efficacy (2026)

    Sun, P.: Curriculum sampling: A two-phase curriculum for efficient train- ing of flow matching. In: ICLR 2026 2nd Workshop on Deep Genera- tive Model in Machine Learning: Theory, Principle and Efficacy (2026). https://openreview.net/forum?id=LpmuITZLAk

    work page 2026

  35. [35]

    https://arxiv.org/abs/2403

    Kim, J.-Y., Go, H., Kwon, S., Kim, H.-G.: Denoising Task Difficulty-based Curriculum for Training Diffusion Models (2025). https://arxiv.org/abs/2403. 10348

    work page 2025

  36. [36]

    Zhang, L., Rao, A., Agrawala, M.: Adding Conditional Control to Text-to-Image Diffusion Models (2023)

    work page 2023

  37. [37]

    In: Proceedings of the 40th International Conference on Machine Learning (ICML), pp

    Bar-Tal, O., Yariv, L., Lipman, Y., Dekel, T.: MultiDiffusion: Fusing diffusion paths for controlled image generation. In: Proceedings of the 40th International Conference on Machine Learning (ICML), pp. 1737–1752 (2023)

    work page 2023

  38. [38]

    In: Interna- tional Conference on Learning Representations (ICLR) (2019)

    Loshchilov, I., Hutter, F.: Decoupled weight decay regularization. In: Interna- tional Conference on Learning Representations (ICLR) (2019)

    work page 2019

  39. [39]

    In: International Conference on Learning Representations (ICLR) (2017)

    Loshchilov, I., Hutter, F.: SGDR: Stochastic gradient descent with warm restarts. In: International Conference on Learning Representations (ICLR) (2017)

    work page 2017

  40. [40]

    Liu, H., Zaharia, M., Abbeel, P.: Ring Attention with Blockwise Transformers for Near-Infinite Context (2023)

    work page 2023

  41. [41]

    Chen, T., Xu, B., Zhang, C., Guestrin, C.: Training Deep Nets with Sublinear Memory Cost (2016)

    work page 2016

  42. [42]

    Bulletin of the Amer- ican Meteorological Society98(8), 1637–1658 (2017) https://doi.org/10.1175/ BAMS-D-16-0065.1

    Yang, J., Zhang, Z., Wei, C., Lu, F., Guo, Q.: Introducing the new generation of Chinese geostationary weather satellites, Fengyun-4. Bulletin of the Amer- ican Meteorological Society98(8), 1637–1658 (2017) https://doi.org/10.1175/ BAMS-D-16-0065.1

    work page 2017

  43. [43]

    IEEE Journal of 40 Selected Topics in Applied Earth Observations and Remote Sensing18, 3599– 3609 (2025) https://doi.org/10.1109/JSTARS.2024.3522056

    Niu, Z., Zhang, L., Yang, Y., Han, Y., Li, H., Wang, D., Huang, W., Weng, F.: Assimilating FY-4B AGRI three water vapor channels in operational Shanghai Typhoon Model (SHTM) using GSI-based 3-DVar approach. IEEE Journal of 40 Selected Topics in Applied Earth Observations and Remote Sensing18, 3599– 3609 (2025) https://doi.org/10.1109/JSTARS.2024.3522056

    work page doi:10.1109/jstars.2024.3522056 2025

  44. [44]

    Geophysical Research Letters51, 2023–106846 (2024) https://doi.org/10.1029/2023GL106846

    Yang, Y., Han, W., Sun, H., Xie, H., Gao, Z.: Reconstruction of 3D DPR observa- tions using GMI radiances. Geophysical Research Letters51, 2023–106846 (2024) https://doi.org/10.1029/2023GL106846

    work page doi:10.1029/2023gl106846 2023

  45. [45]

    Meteorologi- cal Monthly41(10), 1268–1277 (2015) https://doi.org/10.7519/j.issn.1000-0526

    Ren, Z., Zhang, Z., Sun, C., Liu, Y., Li, J., Ju, X., Zhao, Y., Li, Z., Zhang, W., Li, H., Zeng, X., Ren, X., Liu, Y., Wang, H.: Development of three-step quality control system of real-time observation data from AWS in China. Meteorologi- cal Monthly41(10), 1268–1277 (2015) https://doi.org/10.7519/j.issn.1000-0526. 2015.10.010

    work page doi:10.7519/j.issn.1000-0526 2015

  46. [46]

    Sun, S., Nai, C., Pan, B., Li, W., Li, L., Li, X., Foufoula-Georgiou, E., Lin, Y.: Fusion of multi-source precipitation records via coordinate-based generative models (2025)

    work page 2025

  47. [47]

    Geophysical Research Letters52(15), 2025–117075 (2025) https://doi.org/10.1029/2025GL117075 41

    Xiao, H., Zhang, F., Zhang, R., Lu, F., Cai, M., Wang, L.: Retrieval of total precipitable water under all-weather conditions from Himawari-8/AHI observa- tions using the generative diffusion model. Geophysical Research Letters52(15), 2025–117075 (2025) https://doi.org/10.1029/2025GL117075 41

    work page doi:10.1029/2025gl117075 2025