arxiv: 2605.14606 · v1 · submitted 2026-05-14 · 💻 cs.CV

Recognition: 2 theorem links

· Lean Theorem

MambaRain: Multi-Scale Mamba-Attention Framework for 0-3 Hour Precipitation Nowcasting

Chunlei Shi , Cui Wu , Xiang Xu , Hao Li , Ni Fan , Xue Han , Yongchao Feng , Yufeng Zhu

show 5 more authors

Boyu Liu Zengliang Zang Hongbin Wang Yanlan Yang Dan Niu

Authors on Pith no claims yet

Pith reviewed 2026-05-15 04:52 UTC · model grok-4.3

classification 💻 cs.CV

keywords precipitation nowcastingMambaself-attentionmulti-scale modelingspatiotemporal forecastingradar nowcastingdeep learningstate space models

0 comments

The pith

MambaRain combines Mamba blocks with self-attention to extend accurate precipitation nowcasting to three hours.

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

The paper presents MambaRain, a multi-scale encoder-decoder that pairs Mamba's linear-complexity selective state space modeling for long-range temporal dynamics with self-attention modules for spatial correlations in radar observations. Existing deterministic nowcasting methods lose skill after roughly 90 minutes because they cannot sustain long-range spatiotemporal dependencies in chaotic precipitation fields. The hybrid design plus a spectral loss term that counters blurring is intended to maintain fine-scale motion details and push reliable forecasts out to the full 0-3 hour window, with the largest gains appearing in the 2-3 hour range. A reader would care because better short-term rainfall predictions directly support flood alerts, traffic management, and agricultural decisions.

Core claim

MambaRain is a hybrid architecture whose core is the synergistic integration of Mamba blocks that model global temporal dynamics across extended sequences with linear complexity and self-attention modules that explicitly capture spatial correlations within precipitation fields. This combination, embedded in a multi-scale encoder-decoder and trained with an additional spectral loss to preserve fine-scale details, enables comprehensive spatiotemporal representation learning that extends the viable forecasting horizon to 2-3 hours while delivering substantial accuracy gains over prior deterministic approaches.

What carries the argument

Multi-scale encoder-decoder that uses Mamba blocks for selective-state temporal modeling and self-attention for explicit spatial correlation capture, regularized by a spectral loss.

If this is right

Forecast skill remains usable through the full 0-3 hour window instead of collapsing after 90 minutes.
Particularly large gains appear in the 2-3 hour range where prior deterministic models degrade most rapidly.
The spectral loss term reduces blurring and preserves small-scale motion features essential for nowcasting accuracy.
Linear-complexity temporal modeling allows longer sequences without the quadratic cost of full attention.

Where Pith is reading between the lines

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

The same hybrid block pattern could be tested on other chaotic spatiotemporal fields such as cloud optical depth or wind vectors.
Operational nowcasting pipelines might adopt the architecture once the added attention overhead is quantified on GPU hardware.
Extending the horizon beyond three hours would require checking whether the multi-scale design continues to scale or saturates.
Replacing the attention modules with cheaper spatial operators could be explored to lower inference latency while retaining accuracy.

Load-bearing premise

The assumption that Mamba's sequential processing and attention's spatial modeling will together capture the chaotic multi-scale structure of precipitation without introducing new artifacts that require heavy post-hoc correction.

What would settle it

Head-to-head evaluation on standard radar nowcasting benchmarks (e.g., MRMS or similar) showing that MambaRain does not improve or even degrades skill scores relative to strong baselines such as PredRNN or Earthformer specifically in the 120-180 minute lead-time band.

Figures

Figures reproduced from arXiv: 2605.14606 by Boyu Liu, Chunlei Shi, Cui Wu, Dan Niu, Hao Li, Hongbin Wang, Ni Fan, Xiang Xu, Xue Han, Yanlan Yang, Yongchao Feng, Yufeng Zhu, Zengliang Zang.

**Figure 1.** Figure 1: (a) Overview of the MambaRain model architecture featuring multi-scale encoder-decoder structure with MFormer [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗

**Figure 3.** Figure 3: Echo intensity distribution above 10dBZ for both [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

**Figure 2.** Figure 2: Study Area Training Setup. MambaRain is trained for 250 epochs with batch size 16 on four NVIDIA RTX 3090 GPUs. The Adam (a) Xinjiang Province - Northwest China 34.5°-49.5°N, 73.5°-96.5°E (b) Southeast China 20°-40°N, 100°-120°E [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 4.** Figure 4: Threshold-dependent performance comparison of CSI ( [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗

**Figure 5.** Figure 5: Temporal evolution of CSI at the 30 dBZ threshold over the 0–3 hour forecasting horizon on the Southeast China [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗

**Figure 6.** Figure 6: Visualization comparison of different models results at T+30 to T+180 min on Southeast China dataset. Rows correspond [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗

**Figure 7.** Figure 7: Visualization comparison of different models results at T+30 to T+180 min on Xinjiang Province dataset. Rows [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗

read the original abstract

Accurate precipitation nowcasting over extended horizons (0-3 hours) is essential for disaster mitigation and operational decision-making, yet remains a critical challenge in the field. Existing deterministic approaches are predominantly constrained to shorter prediction windows (0-2 hours), exhibiting severe performance degradation beyond 90 minutes owing to their inherent difficulty in capturing long-range spatiotemporal dependencies from radar-derived observations. To address these fundamental limitations, we propose MambaRain, a novel multi-scale encoder-decoder architecture that synergistically integrates Mamba's linear-complexity long-range temporal modeling with self-attention mechanisms for explicit spatial correlation capture. The core innovation lies in a hybrid design paradigm wherein Mamba blocks leverage selective state space mechanisms to model global temporal dynamics across extended sequences with computational efficiency, while self-attention modules explicitly characterize spatial correlations within precipitation fields - a capability inherently absent in Mamba's sequential processing paradigm. This complementary synergy enables comprehensive spatiotemporal representation learning, effectively extending the viable forecasting horizon to 2-3 hours with substantial accuracy improvements. Furthermore, we introduce a spectral loss formulation to mitigate blurring artifacts characteristic of chaotic precipitation systems, thereby preserving fine-scale motion details critical for nowcasting accuracy. Experimental validation demonstrates that MambaRain substantially outperforms existing deterministic methodologies in 0-3 hour nowcasting tasks, with particularly pronounced performance gains in the challenging 2-3 hour prediction range.

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

MambaRain mixes Mamba with attention in a multi-scale setup to target 3-hour nowcasting, but the outperformance claims rest on results that need to be checked.

read the letter

The main point is that this paper takes the Mamba state-space model, pairs it with self-attention in a multi-scale encoder-decoder, and adds a spectral loss to keep fine details in precipitation forecasts out to three hours. That combination is the actual new piece: Mamba for efficient long-range time modeling and attention to handle the spatial structure that pure sequential models often drop. The motivation lines up with known limits in radar nowcasting, where most deterministic methods lose skill after about 90 minutes because they cannot track chaotic multi-scale motion well. The architecture description stays consistent with how these blocks are usually put together, and the spectral loss is a straightforward way to penalize blur without extra parameters that would need heavy tuning. The paper does a clean job explaining the design choices and why they should extend the horizon. The soft spot is the experimental side. The abstract states clear gains over baselines in the 2-3 hour window, but the provided text gives no datasets, metrics, error bars, or ablation numbers to back that up. If the full paper shows reproducible tables with standard nowcasting benchmarks and proper controls, the claim holds; otherwise it stays unverified. This work is for people who build spatiotemporal models for weather or similar fields. A reader who follows new sequence architectures or radar forecasting would find the hybrid useful to consider. It deserves peer review because the core idea is coherent and the problem matters, even if the results section will likely need tightening.

Referee Report

2 major / 3 minor

Summary. The paper proposes MambaRain, a multi-scale encoder-decoder architecture integrating Mamba blocks for linear-complexity long-range temporal modeling with self-attention modules for explicit spatial correlation capture in radar-derived precipitation fields. It further introduces a spectral loss to mitigate blurring and preserve fine-scale details. The central claim is that this hybrid design extends accurate deterministic nowcasting to the 2-3 hour horizon with substantial outperformance over existing methods.

Significance. If the reported gains are robust, the work could meaningfully advance operational nowcasting by addressing the well-known degradation beyond 90 minutes through efficient sequence modeling and spatial attention, with direct relevance to disaster mitigation. The choice of Mamba for temporal dynamics is timely and computationally motivated, while the spectral loss provides a targeted handle on multi-scale chaotic features.

major comments (2)

[§4.3] §4.3, spectral loss: the weighting coefficient is listed as a free hyperparameter with no ablation study, sensitivity analysis, or reported value; because the loss is central to the claim of reduced blurring and preserved motion details, the absence of this analysis leaves the contribution of the spectral term unquantified.
[§5.2] §5.2, Tables 1-2: performance metrics for the 2-3 hour range are presented without error bars, multiple-run statistics, or significance tests; this directly affects the strength of the claim that gains are 'particularly pronounced' and reproducible.

minor comments (3)

[Abstract] Abstract: the phrase 'substantial accuracy improvements' is not accompanied by any numerical values or specific metrics; adding the key quantitative results would make the summary self-contained.
[§3.1] §3.1: the multi-scale encoder-decoder diagram (Figure 2) would benefit from explicit labeling of the Mamba block and attention module placements to match the textual description.
[Related work] Related work: the original Mamba paper (Gu et al., 2023) is cited but the discussion of prior spatiotemporal nowcasting models could more explicitly contrast computational complexity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. The comments highlight opportunities to strengthen the quantification of the spectral loss contribution and the statistical robustness of the reported gains. We address each point below and will revise the manuscript accordingly.

read point-by-point responses

Referee: [§4.3] §4.3, spectral loss: the weighting coefficient is listed as a free hyperparameter with no ablation study, sensitivity analysis, or reported value; because the loss is central to the claim of reduced blurring and preserved motion details, the absence of this analysis leaves the contribution of the spectral term unquantified.

Authors: We agree that the spectral loss weighting coefficient requires explicit reporting and sensitivity analysis to quantify its impact. In the revised manuscript we will state the exact value employed during training and add an ablation study (varying the coefficient over a small grid while holding all other hyperparameters fixed) that measures its effect on blurring metrics and fine-scale detail preservation for the 2-3 h horizon. revision: yes
Referee: [§5.2] §5.2, Tables 1-2: performance metrics for the 2-3 hour range are presented without error bars, multiple-run statistics, or significance tests; this directly affects the strength of the claim that gains are 'particularly pronounced' and reproducible.

Authors: We acknowledge that the absence of error bars and significance testing weakens the reproducibility claim. In the revised version we will rerun the key experiments with multiple random seeds, report mean and standard deviation in Tables 1-2 for the 2-3 h range, and add paired statistical significance tests against the strongest baselines to substantiate that the observed improvements are statistically meaningful. revision: yes

Circularity Check

0 steps flagged

No significant circularity; architecture and loss are independently motivated

full rationale

The paper defines MambaRain via an explicit hybrid encoder-decoder that pairs Mamba blocks (for linear-complexity temporal modeling) with self-attention modules (for spatial correlations) plus a spectral loss term. These components are introduced as design choices motivated by the complementary limitations of each mechanism, not derived from equations that reduce to fitted parameters or prior self-citations. No load-bearing step equates a claimed prediction to an input fit, renames a known result, or imports uniqueness via author-overlapping citations. Experimental claims are presented as falsifiable benchmarks against existing methods, keeping the derivation self-contained.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

The central claim rests on the untested assumption that Mamba's linear-complexity state-space modeling plus attention will jointly solve long-range spatiotemporal dependencies in precipitation; no free parameters, axioms, or invented entities are quantified in the abstract.

free parameters (1)

spectral loss weighting coefficient
Introduced to preserve fine-scale details but value and fitting procedure not specified.

axioms (1)

domain assumption Mamba blocks can model global temporal dynamics across extended radar sequences with linear complexity
Invoked to justify extension beyond 90-minute horizon.

invented entities (1)

MambaRain hybrid architecture no independent evidence
purpose: To combine Mamba temporal modeling with spatial self-attention for precipitation fields
New proposed model whose performance claims lack external validation in the abstract.

pith-pipeline@v0.9.0 · 5582 in / 1370 out tokens · 50867 ms · 2026-05-15T04:52:47.230251+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

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

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Mamba blocks leverage selective state space mechanisms to model global temporal dynamics... self-attention modules explicitly characterize spatial correlations... spectral loss formulation to mitigate blurring artifacts
IndisputableMonolith/Foundation/ArithmeticFromLogic.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

MambaRain... multi-scale encoder-decoder... 8-tick period nowhere mentioned; no φ, J(x), or 8-period clock

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

38 extracted references · 38 canonical work pages · 1 internal anchor

[1]

Wavec2r: Wavelet-driven coarse-to-refined hierarchical learning for radar retrieval,

C. Shi, H. Xu, Y . Li, Y .-L. Wei, Y . Feng, Y . Zhang, and D. Niu, “Wavec2r: Wavelet-driven coarse-to-refined hierarchical learning for radar retrieval,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 40, no. 11, 2026, pp. 8951–8959

work page 2026
[2]

Alphapre: Amplitude-phase disentanglement model for precipitation nowcasting,

K. Lin, B. Zhang, D. Yu, W. Feng, S. Chen, F. Gao, X. Li, and Y . Ye, “Alphapre: Amplitude-phase disentanglement model for precipitation nowcasting,” inProceedings of the Computer Vision and Pattern Recog- nition Conference, 2025, pp. 17 841–17 850

work page 2025
[3]

Lmcast: A pretrained language model guided long-term memory transformer for precipitation nowcasting,

F. Gao, C. Luo, G. Deng, X. Li, B. Zhang, D. Yu, and Y . Ye, “Lmcast: A pretrained language model guided long-term memory transformer for precipitation nowcasting,”Neural Networks, p. 108168, 2025

work page 2025
[4]

Pimmnet: In- troducing multi-modal precipitation nowcasting via a physics-informed perspective,

D. Yu, W. Du, K. Lin, X. Li, Y . Ye, C. Luo, and X. Chen, “Pimmnet: In- troducing multi-modal precipitation nowcasting via a physics-informed perspective,” inProceedings of the 33rd ACM International Conference on Multimedia, 2025, pp. 11 522–11 531

work page 2025
[5]

End-to-end data-driven weather prediction,

A. Allen, S. Markou, W. Tebbutt, J. Requeima, W. P. Bruinsma, T. R. Andersson, M. Herzog, N. D. Lane, M. Chantry, J. S. Hoskinget al., “End-to-end data-driven weather prediction,”Nature, vol. 641, no. 8065, pp. 1172–1179, 2025

work page 2025
[6]

M4caster: Multi-source, multi-spatial, multi-temporal modeling for precipitation nowcasting,

D. Niu, C. Shi, T. Zhang, H. Wang, Z. Zang, M. Jiang, and J. Yang, “M4caster: Multi-source, multi-spatial, multi-temporal modeling for precipitation nowcasting,”Neurocomputing, vol. 648, p. 130621, 2025

work page 2025
[7]

Future extreme precipitation amplified by intensified mesoscale moisture con- vergence,

P. Chang, D. Fu, X. Liu, F. S. Castruccio, A. F. Prein, G. Danabasoglu, X. Wang, J. Bacmeister, Q. Zhang, N. Rosenbloomet al., “Future extreme precipitation amplified by intensified mesoscale moisture con- vergence,”Nature Geoscience, vol. 19, no. 1, pp. 33–41, 2026

work page 2026
[8]

D-pra: A dynamic two- step real-time precipitation retrieval algorithm based on geostationary satellite observation,

M. Cui, L. Jia, J. Lu, C. Zheng, and D. Ji, “D-pra: A dynamic two- step real-time precipitation retrieval algorithm based on geostationary satellite observation,”IEEE Transactions on Geoscience and Remote Sensing, vol. 63, pp. 1–16, 2025

work page 2025
[9]

How effective are time-series models for precipitation nowcasting? a comprehensive benchmark for gnss- based precipitation nowcasting,

Y . Zhang, S. Xiong, H. Wang, W. Yin, J. Peng, Y . Zhang, C. Zhou, H. Chen, Q. Zhao, and P. Duan, “How effective are time-series models for precipitation nowcasting? a comprehensive benchmark for gnss- based precipitation nowcasting,”IEEE Transactions on Geoscience and Remote Sensing, vol. 64, pp. 1–16, 2026

work page 2026
[10]

Detec- tion accuracy of high-resolution infrared satellite precipitation estimates over mainland china: A multiperspective assessment of fengyun-4a,

D. Qian, Y . Lyu, Z. Shen, H. Wu, R. Huang, B. Yong, and H. Su, “Detec- tion accuracy of high-resolution infrared satellite precipitation estimates over mainland china: A multiperspective assessment of fengyun-4a,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 18, pp. 7264–7280, 2025

work page 2025
[11]

Joint intensity and spatio-temporal representation learning for extreme precipitation nowcasting,

Z. Pan, R. Hang, Q. Liu, C. Shi, Z. Xu, and X.-T. Yuan, “Joint intensity and spatio-temporal representation learning for extreme precipitation nowcasting,”IEEE Journal of Selected Topics in Applied Earth Ob- servations and Remote Sensing, vol. 18, pp. 18 905–18 921, 2025

work page 2025
[12]

Precipitation retrieval integrating multiple satellite observations: A dataset and a framework,

Z. Wang, B. He, C. Wang, B. Xu, and C. Bai, “Precipitation retrieval integrating multiple satellite observations: A dataset and a framework,” IEEE Transactions on Geoscience and Remote Sensing, vol. 63, pp. 1– 15, 2025

work page 2025
[13]

Learnable optical flow network for radar echo extrapolation,

C. Zhang, X. Zhou, X. Zhuge, and M. Xu, “Learnable optical flow network for radar echo extrapolation,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 14, pp. 1260– 1266, 2020

work page 2020
[14]

Na- tionwide radar-based precipitation nowcasting—a localization filtering approach and its application for germany,

R. Reinoso-Rondinel, M. Rempel, M. Schultze, and S. Tr ¨omel, “Na- tionwide radar-based precipitation nowcasting—a localization filtering approach and its application for germany,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 15, pp. 1670–1691, 2022

work page 2022
[15]

Convolutional lstm network: A machine learning approach for precipitation nowcasting,

X. Shi, Z. Chen, H. Wang, D. Y . Yeung, W. K. Wong, and W. C. Woo, “Convolutional lstm network: A machine learning approach for precipitation nowcasting,” inProc. Adv. Neural Inf. Process. Syst. (NeurIPS), vol. 28, 2015, pp. 802–810

work page 2015
[16]

Deep Learning for Precipitation Nowcasting: A Benchmark and A New Model,

X. Shi, Z. Gao, L. Lausen, H. Wang, D. Y . Yeung, W. K. Wong, and W. C. Woo, “Deep Learning for Precipitation Nowcasting: A Benchmark and A New Model,” inProc. Adv. Neural Inf. Process. Syst. (NeurIPS), 2017, pp. 5617–5627

work page 2017
[17]

Aa-transunet: Attention augmented tran- sunet for nowcasting tasks,

Y . Yang and S. Mehrkanoon, “Aa-transunet: Attention augmented tran- sunet for nowcasting tasks,” inProc. Int. Joint Conf. Neural Netw. (IJCNN), 2022, pp. 01–08

work page 2022
[18]

Earthformer: Exploring space-time transformers for earth system forecasting,

Z. Gao, X. Shi, H. Wang, Y . Zhu, Y . B. Wang, M. Li, and D.- Y . Yeung, “Earthformer: Exploring space-time transformers for earth system forecasting,” inProc. Adv. Neural Inf. Process. Syst. (NeurIPS), vol. 35, 2022, pp. 25 390–25 403

work page 2022
[19]

Skilful nowcasting of extreme precipitation with nowcastnet,

Y . Zhang, M. Long, K. Chen, L. Xing, R. Jin, M. I. Jordan, and J. Wang, “Skilful nowcasting of extreme precipitation with nowcastnet,”Nature, vol. 619, no. 7970, pp. 526–532, 2023

work page 2023
[20]

Diffcast: A unified framework via residual diffusion for precipitation nowcasting,

D. Yu, X. Li, Y . Ye, B. Zhang, C. Luo, K. Dai, R. Wang, and X. Chen, “Diffcast: A unified framework via residual diffusion for precipitation nowcasting,” inProc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), 2024, pp. 27 758–27 767

work page 2024
[21]

Cascast: Skillful high-resolution precipitation nowcasting via cascaded modelling,

J. Gong, L. Bai, P. Ye, W. Xu, N. Liu, J. Dai, X. Yang, and W. Ouyang, “Cascast: Skillful high-resolution precipitation nowcasting via cascaded modelling,”arXiv preprint arXiv:2402.04290, 2024

work page arXiv 2024
[22]

Extreme precipitation nowcasting using multi-task latent diffusion models,

L. Chaorong, L. Xudong, Y . Qiang, Q. Fengqing, and H. Yuanyuan, “Extreme precipitation nowcasting using multi-task latent diffusion models,”IEEE Trans. Geosci. Remote Sens., 2024

work page 2024
[23]

Fsrgan: A satellite and radar-based fusion prediction network for precipitation nowcasting,

D. Niu, Y . Li, H. Wang, Z. Zang, M. Jiang, X. Chen, and Q. Huang, “Fsrgan: A satellite and radar-based fusion prediction network for precipitation nowcasting,”IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens., 2024

work page 2024
[24]

Simcast: Enhancing precipitation nowcasting with short-to-long term knowledge distillation,

Y . Yin, S. Chen, Y . Li, L. Wang, R. Jin, W. Cui, and S. Xiang, “Simcast: Enhancing precipitation nowcasting with short-to-long term knowledge distillation,”arXiv preprint arXiv:2510.07953, 2025

work page arXiv 2025
[25]

Syncast: Synergizing contradictions in precipitation nowcasting via diffusion sequential preference optimization,

K. Xu, J. Gong, W. Zhang, B. Fei, L. Bai, and W. Ouyang, “Syncast: Synergizing contradictions in precipitation nowcasting via diffusion sequential preference optimization,”arXiv preprint arXiv:2510.21847, 2025

work page arXiv 2025
[26]

Multifactor spatial downscal- ing of satellite precipitation based on vegetation index and elevation,

Z. Zeng, N. Peleg, H. Chen, and L. Zhuo, “Multifactor spatial downscal- ing of satellite precipitation based on vegetation index and elevation,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 19, pp. 3260–3273, 2026

work page 2026
[27]

A framework of multi-source precipitation data fusion in the yellow river basin based on climate and terrain partitioning,

Z. Jian, Q. Yang, H. Liu, and J. Shao, “A framework of multi-source precipitation data fusion in the yellow river basin based on climate and terrain partitioning,”IEEE Transactions on Geoscience and Remote Sensing, pp. 1–1, 2026

work page 2026
[28]

Improving the spatiotemporal resolution of satellite remote sensing precipitation in complex terrain—based on the random forest method,

X. Luo, J. Liao, H. Wang, T. Zhang, Q. Zeng, T. Yu, and Z. Li, “Improving the spatiotemporal resolution of satellite remote sensing precipitation in complex terrain—based on the random forest method,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 19, pp. 10 687–10 700, 2026

work page 2026
[29]

Resolving the precipitation microphysical variability induced by orographic enhancement in com- plex terrain over the san francisco bay area,

H. Chen, R. Cifelli, and V . Chandrasekar, “Resolving the precipitation microphysical variability induced by orographic enhancement in com- plex terrain over the san francisco bay area,” inIGARSS 2020 - 2020 IEEE International Geoscience and Remote Sensing Symposium, 2020, pp. 5415–5418

work page 2020
[30]

Mamba: Linear-Time Sequence Modeling with Selective State Spaces

A. Gu and T. Dao, “Mamba: Linear-time sequence modeling with selective state spaces,”arXiv preprint arXiv:2312.00752, 2023

work page internal anchor Pith review Pith/arXiv arXiv 2023
[31]

Mamba-unet: Dual-branch mamba fusion u-net with multiscale spatio- temporal attention for precipitation nowcasting,

S. Zhao, F. Wang, X. Huang, X. Yang, N. Jiang, J. Peng, and Y . Ban, “Mamba-unet: Dual-branch mamba fusion u-net with multiscale spatio- temporal attention for precipitation nowcasting,”IEEE Transactions on Industrial Informatics, vol. 21, no. 6, pp. 4466–4475, 2025

work page 2025
[32]

Mambacast: An efficient precipitation nowcasting model with dual-branch mamba,

H. Jin, Y . Ye, C. Liu, and F. Gao, “Mambacast: An efficient precipitation nowcasting model with dual-branch mamba,”IEEE Geoscience and Remote Sensing Letters, vol. 23, pp. 1–5, 2026

work page 2026
[33]

Adnm-unet: An asymmetric dual-branch noncausal mamba u-net with multiscale attention enhancement for cloud mask nowcasting,

M. Li, X. Huang, F. Wang, X. Yang, J. Peng, Y . Ban, and N. Jiang, “Adnm-unet: An asymmetric dual-branch noncausal mamba u-net with multiscale attention enhancement for cloud mask nowcasting,”IEEE Transactions on Geoscience and Remote Sensing, vol. 63, pp. 1–15, 2025

work page 2025
[34]

Weathergen: A unified diverse weather generator for lidar point clouds via spider mamba diffusion,

Y . Wu, Y . Zhu, K. Zhang, J. Qian, J. Xie, and J. Yang, “Weathergen: A unified diverse weather generator for lidar point clouds via spider mamba diffusion,” in2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025, pp. 17 019–17 028

work page 2025
[35]

Hi-rsmamba: Hierarchical mamba for remote sensing image restoration under adverse weather,

X. He, J. Li, T. Song, and X. Chen, “Hi-rsmamba: Hierarchical mamba for remote sensing image restoration under adverse weather,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 19, pp. 7373–7388, 2026

work page 2026
[36]

Mdtnet: Multi-scale deformable transformer network with fourier space losses toward fine-scale spatiotemporal precipitation nowcasting,

Z. Zhao, X. Dong, Y . Wang, J. Wang, Y . Chen, and C. Hu, “Mdtnet: Multi-scale deformable transformer network with fourier space losses toward fine-scale spatiotemporal precipitation nowcasting,”IEEE Trans. Geosci. Remote Sens., 2024

work page 2024
[37]

Fourier amplitude and correlation loss: Beyond using l2 loss for skillful precipitation nowcasting,

C.-W. Yan, S. Q. Foo, V . H. Trinh, D.-Y . Yeung, K.-H. Wong, and W.- K. Wong, “Fourier amplitude and correlation loss: Beyond using l2 loss for skillful precipitation nowcasting,” inProc. Adv. Neural Inf. Process. Syst. (NeurIPS), vol. 37, 2024, pp. 100 007–100 041

work page 2024
[38]

Swin-umamba: Mamba-based unet with imagenet-based pretraining,

J. Liu, H. Yang, H.-Y . Zhou, Y . Xi, L. Yu, C. Li, Y . Liang, G. Shi, Y . Yu, S. Zhanget al., “Swin-umamba: Mamba-based unet with imagenet-based pretraining,” inInternational conference on medical image computing and computer-assisted intervention. Springer, 2024, pp. 615–625

work page 2024