Seeing Inside the Storm: Improving Nowcasting by Integrating Meteorological Drivers

Jun Chen; Kaishun Wu; Lin Chen; Minghui Qiu; Shuxin Zhong; Weifeng Chen; Yu Zhang; Zhidan Liu

arxiv: 2605.24067 · v1 · pith:4G3TDYB4new · submitted 2026-05-22 · ⚛️ physics.ao-ph · cs.LG

Seeing Inside the Storm: Improving Nowcasting by Integrating Meteorological Drivers

Minghui Qiu , Jun Chen , Lin Chen , Weifeng Chen , Shuxin Zhong , Zhidan Liu , Yu Zhang , Kaishun Wu This is my paper

Pith reviewed 2026-06-30 14:58 UTC · model grok-4.3

classification ⚛️ physics.ao-ph cs.LG

keywords nowcastingradar reflectivityconvectionstorm predictionmeteorological driversprecipitation forecastingphysics-inspired model

0 comments

The pith

Integrating asynchronous meteorological drivers from radar data boosts storm nowcasting accuracy by 9.7 percent.

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

The paper presents MeteoLogist as a framework that uses radar reflectivity to model the full life cycle of convection by capturing atmospheric precursors such as low-level convergence and latent heating. It tackles the problems of drivers evolving at different times and appearing in scattered locations by splitting radar data into separate physical streams, aligning them over time, and fusing them across space. On nationwide radar records, this yields measurable lifts in identifying intense precipitation, with the largest improvements occurring while storms are still forming rather than after they are already visible.

Core claim

MeteoLogist models the full life cycle of convection from its precursors to organized storm evolution by processing radar echoes into thermodynamic, kinematic, and microphysical streams with Physics-Tailored Encoders, aligning their interactions via causal temporal attention in the Temporal-Phase Aligner, and enforcing spatial coherence with the Cross-Field Spatial Aggregator, which together produce the reported gains in detection metrics.

What carries the argument

Three integrated components: Physics-Tailored Encoders that form distinct dynamical regime streams from radar echoes, Temporal-Phase Aligner that uses causal temporal attention to handle asynchronous driver interactions, and Cross-Field Spatial Aggregator that performs cross-regional fusion to align scattered precursors.

If this is right

High-impact detection (CSI40) improves by 9.7 percent over strong baselines on 3D-NEXRAD data.
Detection gains reach 37.67 percent specifically during the storm-developing stage.
The approach enables sensing of storms before they appear as organized precipitation in radar reflectivity.
The framework covers the entire convection life cycle rather than only mature precipitation events.

Where Pith is reading between the lines

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

The same component structure could be tested on radar datasets from other continents to check whether the alignment and aggregation steps remain effective under different climate regimes.
Combining the extracted driver streams with outputs from numerical weather prediction models might reduce the need for post-hoc tuning.
If the encoders prove robust, the method could be adapted to nowcast other high-impact events such as heavy rainfall leading to flash floods.

Load-bearing premise

Radar echoes alone contain sufficient information about the asynchronous thermodynamic, kinematic, and microphysical drivers.

What would settle it

A replication study on an independent radar archive that shows no improvement in CSI40 during the storm-developing stage would falsify the central claim.

Figures

Figures reproduced from arXiv: 2605.24067 by Jun Chen, Kaishun Wu, Lin Chen, Minghui Qiu, Shuxin Zhong, Weifeng Chen, Yu Zhang, Zhidan Liu.

**Figure 1.** Figure 1: The Framework of MeteoLogist. echoes merely as textures that deform over time. They successfully track where storms move, but ignore the underlying physical drivers, missing the deeper question: when and why do they form? 2.2 Physics-Informed Echo Extrapolation To move beyond surface-level echo tracking, recent work incorporates physics-informed priors. PhyDNet [11] separates physical motion from latent co… view at source ↗

**Figure 2.** Figure 2: evaluates pixel-level error (MAE) and structural accuracy (mCSI) from 20 to 100 minutes forecast horizons. As expected, performance degrades as lead time increases, but the rate of decline varies: recurrent-based models (e.g., ConvLSTM, PredRNN-V2) deteriorate faster (mCSI: –0.022) than recurrent-free ones (–0.014), reflecting weaker spatial generalization over longer horizons. MeteoLogist shows a slower d… view at source ↗

**Figure 3.** Figure 3: Results of the degradation study, showing model performance under varying levels of [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗

**Figure 4.** Figure 4: Qualitative Comparison of Radar Reflectivity Forecasts across Models at Increasing Lead [PITH_FULL_IMAGE:figures/full_fig_p016_4.png] view at source ↗

read the original abstract

Most nowcasting systems, built on radar reflectivity, focus on current precipitation, ignoring the atmospheric precursors -- such as low-level convergence, turbulent eddies, and latent heating -- that offer a fleeting window to foresee storm birth. We introduce MeteoLogist, a physics-inspired radar intelligence framework that models the full life cycle of convection -- from its precursors to organized storm evolution. However, exploiting these precursors is non-trivial: they originate from multiple meteorological drivers -- thermodynamic, kinematic, and microphysical -- that evolve asynchronously (C1) and remain spatially fragmented (C2). To this end, MeteoLogist designs three tightly integrated components. The Physics-Tailored Encoders process radar echoes according to their intrinsic physical scales and semantics, forming thermodynamic, kinematic, and microphysical streams that capture distinct dynamical regimes. The Temporal-Phase Aligner addresses C1 by leveraging causal temporal attention to capture when and how different drivers interact and activate. The Cross-Field Spatial Aggregator addresses C2 through cross-regional fusion, aligning weak and scattered precursors across neighboring cells to expose upstream triggers and enforce spatial coherence. Evaluated on 3D-NEXRAD (2020--2022, US-wide), MeteoLogist boosts high-impact detection (CSI40) by +9.7% over strong baselines, and achieves a remarkable 37.67% gain during the storm-developing stage -- demonstrating true foresight in sensing storms before they appear. The code can be found in the supplementary material.

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.

Desk Editor's Note private letter to a colleague

MeteoLogist adds three targeted components to radar nowcasting and reports clear gains on developing-stage storms, but the abstract leaves the robustness of those gains unproven.

read the letter

The paper's core claim is that standard radar nowcasting misses early convective signals and that a new framework called MeteoLogist can recover them by splitting radar data into thermodynamic, kinematic, and microphysical streams, aligning their timing with causal attention, and fusing scattered spatial signals. The reported +9.7% CSI40 lift and especially the 37.67% gain in the developing stage are the results that would matter for operational use.

The work is new in the specific combination of those three components and in framing the two concrete obstacles (asynchronous drivers and spatial fragmentation) as the main barriers. The design choices line up directly with the stated problems, and the evaluation on three years of US-wide 3D-NEXRAD data is a sensible scale for the task.

The soft spot is the lack of supporting detail. The abstract gives no baseline definitions, no ablation results, no statistical tests, and no discussion of how rare developing-stage events were handled. Without those, the large early-stage improvement cannot be checked for sensitivity to choices in stage labeling or class imbalance. The code is mentioned as available in the supplement, but that does not substitute for the missing controls in the text.

This paper is for researchers and practitioners who build or evaluate nowcasting systems. Anyone working on physics-informed radar models or early-warning applications would find the component breakdown useful even if they end up modifying it. The topic and the coherent architecture are strong enough to justify sending it to peer review; a referee would focus on the missing ablations and baseline comparisons, but the underlying idea is worth that level of scrutiny.

Referee Report

2 major / 1 minor

Summary. The paper introduces MeteoLogist, a physics-inspired framework for radar-based nowcasting that models the full convection life cycle by extracting and aligning thermodynamic, kinematic, and microphysical drivers from radar echoes. It proposes three components—Physics-Tailored Encoders, Temporal-Phase Aligner, and Cross-Field Spatial Aggregator—to address asynchronous evolution (C1) and spatial fragmentation (C2) of precursors. On 3D-NEXRAD data (2020–2022, US-wide), it claims +9.7% improvement in CSI40 over strong baselines and a 37.67% gain during the storm-developing stage.

Significance. If the performance claims hold under rigorous validation, the work could advance nowcasting by enabling earlier detection of storm initiation through precursor signals rather than current precipitation alone. Code availability in supplementary material is a strength for reproducibility.

major comments (2)

[Abstract] Abstract: the central performance claims (+9.7% CSI40 boost and 37.67% developing-stage gain) are presented without any definition of the 'strong baselines,' statistical significance tests, ablation results, or class-imbalance handling; this directly prevents assessment of whether the reported gains are robust or load-bearing for the claim of 'true foresight.'
[Abstract] Abstract: the weakest assumption—that radar echoes alone contain sufficient asynchronous thermodynamic/kinematic/microphysical information for the three named components to extract without additional data sources or post-hoc tuning—is stated but receives no supporting evidence or validation strategy in the provided text.

minor comments (1)

[Abstract] The phrase 'strong baselines' is undefined and should be replaced with explicit model names and references.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments focused on the abstract. We address each point below, clarifying where the full manuscript provides supporting details and indicating revisions where appropriate.

read point-by-point responses

Referee: [Abstract] Abstract: the central performance claims (+9.7% CSI40 boost and 37.67% developing-stage gain) are presented without any definition of the 'strong baselines,' statistical significance tests, ablation results, or class-imbalance handling; this directly prevents assessment of whether the reported gains are robust or load-bearing for the claim of 'true foresight.'

Authors: The abstract is space-constrained, but the full manuscript defines the strong baselines (PredRNN, ConvLSTM, PhyDNet and other radar-only nowcasters) in Section 4.2, reports statistical significance via paired t-tests (p < 0.01) and bootstrap intervals in Section 5.1 and Table 3, presents ablation studies isolating each component in Section 5.3 and Figure 5, and handles class imbalance via the weighted focal loss in Equation (7). These elements underpin the reported gains. We will revise the abstract to add a short clause naming the baseline category and directing readers to the main text for the supporting analyses. revision: partial
Referee: [Abstract] Abstract: the weakest assumption—that radar echoes alone contain sufficient asynchronous thermodynamic/kinematic/microphysical information for the three named components to extract without additional data sources or post-hoc tuning—is stated but receives no supporting evidence or validation strategy in the provided text.

Authors: The assumption is tested directly by the experimental design: all models, including baselines, are trained and evaluated exclusively on 3D-NEXRAD radar reflectivity and radial velocity (Section 4.1), with no auxiliary observations or post-hoc tuning. The 37.67 % gain specifically in the storm-developing stage, where only precursor signals exist, constitutes the primary empirical validation that the Physics-Tailored Encoders, Temporal-Phase Aligner and Cross-Field Spatial Aggregator extract the required asynchronous and fragmented information from radar alone. We can add an explicit sentence in the introduction reiterating the radar-only constraint if the referee finds it helpful. revision: no

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The provided manuscript text consists solely of an abstract describing a new framework (MeteoLogist) with three named components for processing radar data, evaluated on an external 3D-NEXRAD dataset (2020-2022). No equations, fitted parameters, self-citations, or derivation steps are present that would reduce any claimed performance gain (e.g., CSI40 improvements) to the inputs by construction. The reported gains are framed as empirical results on held-out data, with no load-bearing premises that collapse into self-definition, renaming, or ansatz smuggling. The derivation chain is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 4 invented entities

The central claim rests on domain assumptions about driver asynchrony and spatial fragmentation plus the effectiveness of the three newly introduced architectural components; no numerical free parameters are mentioned.

axioms (1)

domain assumption Meteorological drivers evolve asynchronously (C1) and remain spatially fragmented (C2)
Explicitly labeled as the two core challenges the framework is designed to solve.

invented entities (4)

MeteoLogist no independent evidence
purpose: Physics-inspired framework modeling the full convection life cycle from precursors to organized storms
Newly introduced system name and overall architecture.
Physics-Tailored Encoders no independent evidence
purpose: Form thermodynamic, kinematic, and microphysical streams from radar echoes at intrinsic physical scales
One of the three core components of the proposed framework.
Temporal-Phase Aligner no independent evidence
purpose: Use causal temporal attention to capture driver interactions and activation timing
One of the three core components of the proposed framework.
Cross-Field Spatial Aggregator no independent evidence
purpose: Perform cross-regional fusion to align weak precursors and enforce spatial coherence
One of the three core components of the proposed framework.

pith-pipeline@v0.9.1-grok · 5819 in / 1492 out tokens · 46214 ms · 2026-06-30T14:58:37.219347+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

39 extracted references · 1 canonical work pages

[1]

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
[2]

Lindsay J Bennett, Keith A Browning, Alan M Blyth, Douglas J Parker, and Peter A Clark. A review of the initiation of precipitating convection in the united kingdom.Quarterly Journal of the Royal Meteorological Society: A journal of the atmospheric sciences, applied meteorology and physical oceanography, 132(617):1001–1020, 2006

2006
[3]

Crossvit: Cross-attention multi- scale vision transformer for image classification

Chun-Fu Richard Chen, Quanfu Fan, and Rameswar Panda. Crossvit: Cross-attention multi- scale vision transformer for image classification. InProceedings of the IEEE/CVF international conference on computer vision, pages 357–366, 2021

2021
[4]

Datasets.https://rda.ucar.edu/datasets/d841006/, 2025

Datasets. Datasets.https://rda.ucar.edu/datasets/d841006/, 2025. Accessed: 2025- 07-04

2025
[5]

Gate-variants of gated recurrent unit (gru) neural networks

Rahul Dey and Fathi M Salem. Gate-variants of gated recurrent unit (gru) neural networks. In 2017 IEEE 60th international midwest symposium on circuits and systems (MWSCAS), pages 1597–1600. IEEE, 2017

2017
[6]

Flash flood forecasting: An ingredients-based methodology.Weather and forecasting, 11(4):560–581, 1996

Charles A Doswell III, Harold E Brooks, and Robert A Maddox. Flash flood forecasting: An ingredients-based methodology.Weather and forecasting, 11(4):560–581, 1996

1996
[7]

Coalescence microseismic mapping.Geophysical Journal International, 195(3):1773–1785, 2013

Julian Drew, Robert S White, Frederik Tilmann, and Jon Tarasewicz. Coalescence microseismic mapping.Geophysical Journal International, 195(3):1773–1785, 2013

2013
[8]

Substantial convection and precipitation enhancements by ultrafine aerosol particles.Science, 359(6374):411–418, 2018

Jiwen Fan, Daniel Rosenfeld, Yuwei Zhang, Scott E Giangrande, Zhanqing Li, Luiz AT Machado, Scot T Martin, Yan Yang, Jian Wang, Paulo Artaxo, et al. Substantial convection and precipitation enhancements by ultrafine aerosol particles.Science, 359(6374):411–418, 2018

2018
[9]

Simvp: Simpler yet better video prediction

Zhangyang Gao, Cheng Tan, Lirong Wu, and Stan Z Li. Simvp: Simpler yet better video prediction. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 3170–3180, 2022

2022
[10]

Earthformer: Exploring space-time transformers for earth system forecasting.Advances in Neural Information Processing Systems, 35:25390–25403, 2022

Zhihan Gao, Xingjian Shi, Hao Wang, Yi Zhu, Yuyang Bernie Wang, Mu Li, and Dit-Yan Yeung. Earthformer: Exploring space-time transformers for earth system forecasting.Advances in Neural Information Processing Systems, 35:25390–25403, 2022

2022
[11]

Disentangling physical dynamics from unknown factors for unsupervised video prediction

Vincent Le Guen and Nicolas Thome. Disentangling physical dynamics from unknown factors for unsupervised video prediction. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11474–11484, 2020

2020
[12]

Wf-unet: weather fusion unet for precipitation nowcasting.arXiv preprint arXiv:2302.04102, 2023

Christos Kaparakis and Siamak Mehrkanoon. Wf-unet: weather fusion unet for precipitation nowcasting.arXiv preprint arXiv:2302.04102, 2023

work page arXiv 2023
[13]

Interactions between aerosol absorption, thermodynamics, dynamics, and microphysics and their impacts on a multiple-cloud system

Seoung Soo Lee, Zhanqing Li, Jungbin Mok, Myoung-Hwan Ahn, Byung-Gon Kim, Yong- Sang Choi, Chang-Hoon Jung, and Hye Lim Yoo. Interactions between aerosol absorption, thermodynamics, dynamics, and microphysics and their impacts on a multiple-cloud system. Climate Dynamics, 49(11):3905–3921, 2017

2017
[14]

Alphapre: Amplitude-phase disentanglement model for precipitation nowcasting

Kenghong Lin, Baoquan Zhang, Demin Yu, Wenzhi Feng, Shidong Chen, Feifan Gao, Xutao Li, and Yunming Ye. Alphapre: Amplitude-phase disentanglement model for precipitation nowcasting. InProceedings of the Computer Vision and Pattern Recognition Conference (CVPR), pages 17841–17850, June 2025

2025
[15]

Alphapre: Amplitude-phase disentanglement model for precipitation nowcasting

Kenghong Lin, Baoquan Zhang, Demin Yu, Wenzhi Feng, Shidong Chen, Feifan Gao, Xutao Li, and Yunming Ye. Alphapre: Amplitude-phase disentanglement model for precipitation nowcasting. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 17841–17850, 2025

2025
[16]

2025 guizhou floods

NEWS. 2025 guizhou floods. https://watchers.news/2025/07/01/ rongjiang-china-floods-evacuate-120000-six-dead/ , 2025. Accessed: 2025- 07-04. 10

2025
[17]

2025 guizhou floods

NEWS. 2025 guizhou floods. https://www.chinadailyhk.com/hk/article/614690,

2025
[18]

Accessed: 2025-07-04

2025
[19]

Improving nowcasting of convective development by incorporating polarimetric radar variables into a deep-learning model.Geophysical Research Letters, 48(21):e2021GL095302, 2021

Xiang Pan, Yinghui Lu, Kun Zhao, Hao Huang, Mingjun Wang, and Haonan Chen. Improving nowcasting of convective development by incorporating polarimetric radar variables into a deep-learning model.Geophysical Research Letters, 48(21):e2021GL095302, 2021

2021
[20]

Evan Racah, Christopher Beckham, Tegan Maharaj, Samira Ebrahimi Kahou, Mr Prabhat, and Chris Pal. Extremeweather: A large-scale climate dataset for semi-supervised detection, localization, and understanding of extreme weather events.Advances in neural information processing systems, 30, 2017

2017
[21]

Skilful pre- cipitation nowcasting using deep generative models of radar.Nature, 597(7878):672–677, 2021

Suman Ravuri, Karel Lenc, Matthew Willson, Dmitry Kangin, Remi Lam, Piotr Mirowski, Megan Fitzsimons, Maria Athanassiadou, Sheleem Kashem, Sam Madge, et al. Skilful pre- cipitation nowcasting using deep generative models of radar.Nature, 597(7878):672–677, 2021

2021
[22]

Early warning of complex climate risk with integrated artificial intelligence.Nature Communications, 16(1):2564, 2025

Markus Reichstein, Vitus Benson, Jan Blunk, Gustau Camps-Valls, Felix Creutzig, Carina J Fearnley, Boran Han, Kai Kornhuber, Nasim Rahaman, Bernhard Schölkopf, et al. Early warning of complex climate risk with integrated artificial intelligence.Nature Communications, 16(1):2564, 2025

2025
[23]

Onset of convection in a gravitationally unstable diffusive boundary layer in porous media.Journal of Fluid Mechanics, 548:87–111, 2006

A Riaz, M Hesse, HA Tchelepi, and FM Orr. Onset of convection in a gravitationally unstable diffusive boundary layer in porous media.Journal of Fluid Mechanics, 548:87–111, 2006

2006
[24]

The formation, character and changing nature of mesoscale convective systems.Nature Reviews Earth & Environment, 1(6):300–314, 2020

Russ S Schumacher and Kristen L Rasmussen. The formation, character and changing nature of mesoscale convective systems.Nature Reviews Earth & Environment, 1(6):300–314, 2020

2020
[25]

Convolutional lstm network: A machine learning approach for precipitation nowcasting

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

2015
[26]

Deep learning for precipitation nowcasting: A benchmark and a new model

Xingjian Shi, Zhihan Gao, Leonard Lausen, Hao Wang, Dit-Yan Yeung, Wai-kin Wong, and Wang-chun Woo. Deep learning for precipitation nowcasting: A benchmark and a new model. Advances in neural information processing systems, 30, 2017

2017
[27]

Attention is all you need.Advances in neural information processing systems, 30, 2017

Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need.Advances in neural information processing systems, 30, 2017

2017
[28]

Eight kinetically stable but thermodynami- cally activated molecules that power cell metabolism.Chemical reviews, 118(4):1460–1494, 2018

Christopher T Walsh, Benjamin P Tu, and Yi Tang. Eight kinetically stable but thermodynami- cally activated molecules that power cell metabolism.Chemical reviews, 118(4):1460–1494, 2018

2018
[29]

Nuwadynamics: Discovering and updating in causal spatio-temporal modeling

Kun Wang, Hao Wu, Yifan Duan, Guibin Zhang, Kai Wang, Xiaojiang Peng, Yu Zheng, Yuxuan Liang, and Yang Wang. Nuwadynamics: Discovering and updating in causal spatio-temporal modeling. InThe Twelfth International Conference on Learning Representations, 2024

2024
[30]

Predrnn: A recurrent neural network for spatiotemporal predictive learning.IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(2):2208–2225, 2022

Yunbo Wang, Haixu Wu, Jianjin Zhang, Zhifeng Gao, Jianmin Wang, Philip S Yu, and Ming- sheng Long. Predrnn: A recurrent neural network for spatiotemporal predictive learning.IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(2):2208–2225, 2022

2022
[31]

Operational application of optical flow techniques to radar-based rainfall nowcasting.Atmosphere, 8(3):48, 2017

Wang-chun Woo and Wai-kin Wong. Operational application of optical flow techniques to radar-based rainfall nowcasting.Atmosphere, 8(3):48, 2017

2017
[32]

Earthfarsser: Versatile spatio-temporal dynamical systems modeling in one model

Hao Wu, Yuxuan Liang, Wei Xiong, Zhengyang Zhou, Wei Huang, Shilong Wang, and Kun Wang. Earthfarsser: Versatile spatio-temporal dynamical systems modeling in one model. In Proceedings of the AAAI conference on artificial intelligence, volume 38, pages 15906–15914, 2024

2024
[33]

Pastnet: Introducing physical inductive biases for spatio-temporal video prediction

Hao Wu, Fan Xu, Chong Chen, Xian-Sheng Hua, Xiao Luo, and Haixin Wang. Pastnet: Introducing physical inductive biases for spatio-temporal video prediction. InProceedings of the 32nd ACM International Conference on Multimedia, pages 2917–2926, 2024. 11

2024
[34]

Scientific challenges of convective-scale numerical weather prediction.Bulletin of the American Meteorological Society, 99(4):699–710, 2018

Jun-Ichi Yano, Michał Z Ziemia´nski, Mike Cullen, Piet Termonia, Jeanette Onvlee, Lisa Bengts- son, Alberto Carrassi, Richard Davy, Anna Deluca, Suzanne L Gray, et al. Scientific challenges of convective-scale numerical weather prediction.Bulletin of the American Meteorological Society, 99(4):699–710, 2018

2018
[35]

Video prediction by efficient transformers.Image and Vision Computing, 130:104612, 2023

Xi Ye and Guillaume-Alexandre Bilodeau. Video prediction by efficient transformers.Image and Vision Computing, 130:104612, 2023

2023
[36]

Diffcast: A unified framework via residual diffusion for precipitation nowcasting

Demin Yu, Xutao Li, Yunming Ye, Baoquan Zhang, Chuyao Luo, Kuai Dai, Rui Wang, and Xunlai Chen. Diffcast: A unified framework via residual diffusion for precipitation nowcasting. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 27758–27767, 2024

2024
[37]

Skilful nowcasting of extreme precipitation with nowcastnet.Nature, 619 (7970):526–532, 2023

Yuchen Zhang, Mingsheng Long, Kaiyuan Chen, Lanxiang Xing, Ronghua Jin, Michael I Jordan, and Jianmin Wang. Skilful nowcasting of extreme precipitation with nowcastnet.Nature, 619 (7970):526–532, 2023

2023
[38]

A typology of compound weather and climate events.Nature reviews earth & environment, 1(7):333–347, 2020

Jakob Zscheischler, Olivia Martius, Seth Westra, Emanuele Bevacqua, Colin Raymond, Radley M Horton, Bart van den Hurk, Amir AghaKouchak, Aglaé Jézéquel, Miguel D Mahecha, et al. A typology of compound weather and climate events.Nature reviews earth & environment, 1(7):333–347, 2020. A Experiment Setting Details A.1 Detailed Dataset Configurations In our e...

2020
[39]

We adopt the Adam optimizer with the learning rate set to1×10 −3

The weight λs of the pretext loss is set to 1 during the warmup phase and is fixed at 1×10 −4 thereafter. We adopt the Adam optimizer with the learning rate set to1×10 −3. B Extra Experiment Results B.1 Statistical Significance Analysis To ensure empirical rigor and reproducibility, all evaluations are conducted across five independent random seeds. We as...

2006

[1] [1]

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

[2] [2]

Lindsay J Bennett, Keith A Browning, Alan M Blyth, Douglas J Parker, and Peter A Clark. A review of the initiation of precipitating convection in the united kingdom.Quarterly Journal of the Royal Meteorological Society: A journal of the atmospheric sciences, applied meteorology and physical oceanography, 132(617):1001–1020, 2006

2006

[3] [3]

Crossvit: Cross-attention multi- scale vision transformer for image classification

Chun-Fu Richard Chen, Quanfu Fan, and Rameswar Panda. Crossvit: Cross-attention multi- scale vision transformer for image classification. InProceedings of the IEEE/CVF international conference on computer vision, pages 357–366, 2021

2021

[4] [4]

Datasets.https://rda.ucar.edu/datasets/d841006/, 2025

Datasets. Datasets.https://rda.ucar.edu/datasets/d841006/, 2025. Accessed: 2025- 07-04

2025

[5] [5]

Gate-variants of gated recurrent unit (gru) neural networks

Rahul Dey and Fathi M Salem. Gate-variants of gated recurrent unit (gru) neural networks. In 2017 IEEE 60th international midwest symposium on circuits and systems (MWSCAS), pages 1597–1600. IEEE, 2017

2017

[6] [6]

Flash flood forecasting: An ingredients-based methodology.Weather and forecasting, 11(4):560–581, 1996

Charles A Doswell III, Harold E Brooks, and Robert A Maddox. Flash flood forecasting: An ingredients-based methodology.Weather and forecasting, 11(4):560–581, 1996

1996

[7] [7]

Coalescence microseismic mapping.Geophysical Journal International, 195(3):1773–1785, 2013

Julian Drew, Robert S White, Frederik Tilmann, and Jon Tarasewicz. Coalescence microseismic mapping.Geophysical Journal International, 195(3):1773–1785, 2013

2013

[8] [8]

Substantial convection and precipitation enhancements by ultrafine aerosol particles.Science, 359(6374):411–418, 2018

Jiwen Fan, Daniel Rosenfeld, Yuwei Zhang, Scott E Giangrande, Zhanqing Li, Luiz AT Machado, Scot T Martin, Yan Yang, Jian Wang, Paulo Artaxo, et al. Substantial convection and precipitation enhancements by ultrafine aerosol particles.Science, 359(6374):411–418, 2018

2018

[9] [9]

Simvp: Simpler yet better video prediction

Zhangyang Gao, Cheng Tan, Lirong Wu, and Stan Z Li. Simvp: Simpler yet better video prediction. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 3170–3180, 2022

2022

[10] [10]

Earthformer: Exploring space-time transformers for earth system forecasting.Advances in Neural Information Processing Systems, 35:25390–25403, 2022

Zhihan Gao, Xingjian Shi, Hao Wang, Yi Zhu, Yuyang Bernie Wang, Mu Li, and Dit-Yan Yeung. Earthformer: Exploring space-time transformers for earth system forecasting.Advances in Neural Information Processing Systems, 35:25390–25403, 2022

2022

[11] [11]

Disentangling physical dynamics from unknown factors for unsupervised video prediction

Vincent Le Guen and Nicolas Thome. Disentangling physical dynamics from unknown factors for unsupervised video prediction. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11474–11484, 2020

2020

[12] [12]

Wf-unet: weather fusion unet for precipitation nowcasting.arXiv preprint arXiv:2302.04102, 2023

Christos Kaparakis and Siamak Mehrkanoon. Wf-unet: weather fusion unet for precipitation nowcasting.arXiv preprint arXiv:2302.04102, 2023

work page arXiv 2023

[13] [13]

Interactions between aerosol absorption, thermodynamics, dynamics, and microphysics and their impacts on a multiple-cloud system

Seoung Soo Lee, Zhanqing Li, Jungbin Mok, Myoung-Hwan Ahn, Byung-Gon Kim, Yong- Sang Choi, Chang-Hoon Jung, and Hye Lim Yoo. Interactions between aerosol absorption, thermodynamics, dynamics, and microphysics and their impacts on a multiple-cloud system. Climate Dynamics, 49(11):3905–3921, 2017

2017

[14] [14]

Alphapre: Amplitude-phase disentanglement model for precipitation nowcasting

Kenghong Lin, Baoquan Zhang, Demin Yu, Wenzhi Feng, Shidong Chen, Feifan Gao, Xutao Li, and Yunming Ye. Alphapre: Amplitude-phase disentanglement model for precipitation nowcasting. InProceedings of the Computer Vision and Pattern Recognition Conference (CVPR), pages 17841–17850, June 2025

2025

[15] [15]

Alphapre: Amplitude-phase disentanglement model for precipitation nowcasting

Kenghong Lin, Baoquan Zhang, Demin Yu, Wenzhi Feng, Shidong Chen, Feifan Gao, Xutao Li, and Yunming Ye. Alphapre: Amplitude-phase disentanglement model for precipitation nowcasting. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 17841–17850, 2025

2025

[16] [16]

2025 guizhou floods

NEWS. 2025 guizhou floods. https://watchers.news/2025/07/01/ rongjiang-china-floods-evacuate-120000-six-dead/ , 2025. Accessed: 2025- 07-04. 10

2025

[17] [17]

2025 guizhou floods

NEWS. 2025 guizhou floods. https://www.chinadailyhk.com/hk/article/614690,

2025

[18] [18]

Accessed: 2025-07-04

2025

[19] [19]

Improving nowcasting of convective development by incorporating polarimetric radar variables into a deep-learning model.Geophysical Research Letters, 48(21):e2021GL095302, 2021

Xiang Pan, Yinghui Lu, Kun Zhao, Hao Huang, Mingjun Wang, and Haonan Chen. Improving nowcasting of convective development by incorporating polarimetric radar variables into a deep-learning model.Geophysical Research Letters, 48(21):e2021GL095302, 2021

2021

[20] [20]

Evan Racah, Christopher Beckham, Tegan Maharaj, Samira Ebrahimi Kahou, Mr Prabhat, and Chris Pal. Extremeweather: A large-scale climate dataset for semi-supervised detection, localization, and understanding of extreme weather events.Advances in neural information processing systems, 30, 2017

2017

[21] [21]

Skilful pre- cipitation nowcasting using deep generative models of radar.Nature, 597(7878):672–677, 2021

Suman Ravuri, Karel Lenc, Matthew Willson, Dmitry Kangin, Remi Lam, Piotr Mirowski, Megan Fitzsimons, Maria Athanassiadou, Sheleem Kashem, Sam Madge, et al. Skilful pre- cipitation nowcasting using deep generative models of radar.Nature, 597(7878):672–677, 2021

2021

[22] [22]

Early warning of complex climate risk with integrated artificial intelligence.Nature Communications, 16(1):2564, 2025

Markus Reichstein, Vitus Benson, Jan Blunk, Gustau Camps-Valls, Felix Creutzig, Carina J Fearnley, Boran Han, Kai Kornhuber, Nasim Rahaman, Bernhard Schölkopf, et al. Early warning of complex climate risk with integrated artificial intelligence.Nature Communications, 16(1):2564, 2025

2025

[23] [23]

Onset of convection in a gravitationally unstable diffusive boundary layer in porous media.Journal of Fluid Mechanics, 548:87–111, 2006

A Riaz, M Hesse, HA Tchelepi, and FM Orr. Onset of convection in a gravitationally unstable diffusive boundary layer in porous media.Journal of Fluid Mechanics, 548:87–111, 2006

2006

[24] [24]

The formation, character and changing nature of mesoscale convective systems.Nature Reviews Earth & Environment, 1(6):300–314, 2020

Russ S Schumacher and Kristen L Rasmussen. The formation, character and changing nature of mesoscale convective systems.Nature Reviews Earth & Environment, 1(6):300–314, 2020

2020

[25] [25]

Convolutional lstm network: A machine learning approach for precipitation nowcasting

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

2015

[26] [26]

Deep learning for precipitation nowcasting: A benchmark and a new model

Xingjian Shi, Zhihan Gao, Leonard Lausen, Hao Wang, Dit-Yan Yeung, Wai-kin Wong, and Wang-chun Woo. Deep learning for precipitation nowcasting: A benchmark and a new model. Advances in neural information processing systems, 30, 2017

2017

[27] [27]

Attention is all you need.Advances in neural information processing systems, 30, 2017

Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need.Advances in neural information processing systems, 30, 2017

2017

[28] [28]

Eight kinetically stable but thermodynami- cally activated molecules that power cell metabolism.Chemical reviews, 118(4):1460–1494, 2018

Christopher T Walsh, Benjamin P Tu, and Yi Tang. Eight kinetically stable but thermodynami- cally activated molecules that power cell metabolism.Chemical reviews, 118(4):1460–1494, 2018

2018

[29] [29]

Nuwadynamics: Discovering and updating in causal spatio-temporal modeling

Kun Wang, Hao Wu, Yifan Duan, Guibin Zhang, Kai Wang, Xiaojiang Peng, Yu Zheng, Yuxuan Liang, and Yang Wang. Nuwadynamics: Discovering and updating in causal spatio-temporal modeling. InThe Twelfth International Conference on Learning Representations, 2024

2024

[30] [30]

Predrnn: A recurrent neural network for spatiotemporal predictive learning.IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(2):2208–2225, 2022

Yunbo Wang, Haixu Wu, Jianjin Zhang, Zhifeng Gao, Jianmin Wang, Philip S Yu, and Ming- sheng Long. Predrnn: A recurrent neural network for spatiotemporal predictive learning.IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(2):2208–2225, 2022

2022

[31] [31]

Operational application of optical flow techniques to radar-based rainfall nowcasting.Atmosphere, 8(3):48, 2017

Wang-chun Woo and Wai-kin Wong. Operational application of optical flow techniques to radar-based rainfall nowcasting.Atmosphere, 8(3):48, 2017

2017

[32] [32]

Earthfarsser: Versatile spatio-temporal dynamical systems modeling in one model

Hao Wu, Yuxuan Liang, Wei Xiong, Zhengyang Zhou, Wei Huang, Shilong Wang, and Kun Wang. Earthfarsser: Versatile spatio-temporal dynamical systems modeling in one model. In Proceedings of the AAAI conference on artificial intelligence, volume 38, pages 15906–15914, 2024

2024

[33] [33]

Pastnet: Introducing physical inductive biases for spatio-temporal video prediction

Hao Wu, Fan Xu, Chong Chen, Xian-Sheng Hua, Xiao Luo, and Haixin Wang. Pastnet: Introducing physical inductive biases for spatio-temporal video prediction. InProceedings of the 32nd ACM International Conference on Multimedia, pages 2917–2926, 2024. 11

2024

[34] [34]

Scientific challenges of convective-scale numerical weather prediction.Bulletin of the American Meteorological Society, 99(4):699–710, 2018

Jun-Ichi Yano, Michał Z Ziemia´nski, Mike Cullen, Piet Termonia, Jeanette Onvlee, Lisa Bengts- son, Alberto Carrassi, Richard Davy, Anna Deluca, Suzanne L Gray, et al. Scientific challenges of convective-scale numerical weather prediction.Bulletin of the American Meteorological Society, 99(4):699–710, 2018

2018

[35] [35]

Video prediction by efficient transformers.Image and Vision Computing, 130:104612, 2023

Xi Ye and Guillaume-Alexandre Bilodeau. Video prediction by efficient transformers.Image and Vision Computing, 130:104612, 2023

2023

[36] [36]

Diffcast: A unified framework via residual diffusion for precipitation nowcasting

Demin Yu, Xutao Li, Yunming Ye, Baoquan Zhang, Chuyao Luo, Kuai Dai, Rui Wang, and Xunlai Chen. Diffcast: A unified framework via residual diffusion for precipitation nowcasting. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 27758–27767, 2024

2024

[37] [37]

Skilful nowcasting of extreme precipitation with nowcastnet.Nature, 619 (7970):526–532, 2023

Yuchen Zhang, Mingsheng Long, Kaiyuan Chen, Lanxiang Xing, Ronghua Jin, Michael I Jordan, and Jianmin Wang. Skilful nowcasting of extreme precipitation with nowcastnet.Nature, 619 (7970):526–532, 2023

2023

[38] [38]

A typology of compound weather and climate events.Nature reviews earth & environment, 1(7):333–347, 2020

Jakob Zscheischler, Olivia Martius, Seth Westra, Emanuele Bevacqua, Colin Raymond, Radley M Horton, Bart van den Hurk, Amir AghaKouchak, Aglaé Jézéquel, Miguel D Mahecha, et al. A typology of compound weather and climate events.Nature reviews earth & environment, 1(7):333–347, 2020. A Experiment Setting Details A.1 Detailed Dataset Configurations In our e...

2020

[39] [39]

We adopt the Adam optimizer with the learning rate set to1×10 −3

The weight λs of the pretext loss is set to 1 during the warmup phase and is fixed at 1×10 −4 thereafter. We adopt the Adam optimizer with the learning rate set to1×10 −3. B Extra Experiment Results B.1 Statistical Significance Analysis To ensure empirical rigor and reproducibility, all evaluations are conducted across five independent random seeds. We as...

2006