Real-Time Earthquake Magnitude Classification from Initial P-Waves: Models, Dataset, and Comparative Analysis for South Asia

Abdullah Al Noman; Md. Abid Ullah Muhib; Md Nasiat Hasan Fahim; Rayhanul Amin Tanvir

arxiv: 2605.22836 · v1 · pith:S5C7LYVMnew · submitted 2026-05-10 · ⚛️ physics.geo-ph · cs.LG

Real-Time Earthquake Magnitude Classification from Initial P-Waves: Models, Dataset, and Comparative Analysis for South Asia

Md Nasiat Hasan Fahim , Md. Abid Ullah Muhib , Rayhanul Amin Tanvir , Abdullah Al Noman This is my paper

Pith reviewed 2026-05-25 00:06 UTC · model grok-4.3

classification ⚛️ physics.geo-ph cs.LG

keywords earthquake magnitude classificationP-wave analysisTransformer modelseismic early warningSouth Asia earthquakesmachine learningreal-time classificationadaptive accuracy

0 comments

The pith

Transformer classifies earthquake magnitudes from 7-second P-waves with 76% accuracy

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

This paper establishes that a Transformer architecture can classify earthquakes into five magnitude classes using only the vertical component of the first 7 seconds of P-wave data recorded at a single station. The authors assembled a dataset of 7318 South Asian events split into slight, light, moderate, strong, and severe categories and tested six machine learning methods. Deep learning models outperformed traditional ones, with the Transformer reaching 76.23 percent standard accuracy and 81.56 percent on a new adaptive accuracy measure that tolerates boundary uncertainty, all at 4.8 milliseconds inference time. These results point to practical viability for real-time magnitude estimation in early warning systems.

Core claim

The authors show that attention mechanisms within a Transformer model extract enough temporal information from short single-station P-wave segments to separate five Richter-scale magnitude classes on their South Asia catalog, delivering higher accuracy than other tested approaches and supporting real-time use despite limited samples of the largest events.

What carries the argument

Transformer architecture with attention applied to 7-second vertical P-wave time series for five-class magnitude classification

Load-bearing premise

The 7318-event South Asia dataset represents the full variety of real earthquake signals and that the initial 7-second single-station P-wave contains enough information to separate the five magnitude classes reliably.

What would settle it

Running the trained model on an independent collection of earthquakes recorded after the dataset period or from a different region and checking whether accuracy remains near 76 percent.

Figures

Figures reproduced from arXiv: 2605.22836 by Abdullah Al Noman, Md. Abid Ullah Muhib, Md Nasiat Hasan Fahim, Rayhanul Amin Tanvir.

**Figure 3.** Figure 3: Earthquake P-wave dataset quality control pipeline showing (a) multi [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗

**Figure 2.** Figure 2: P-wave analysis of a magnitude 3.3 earthquake event. (a) Extracted [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 4.** Figure 4: Confusion matrix showing adaptive performance of the Transformer [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗

**Figure 5.** Figure 5: Confusion matrix showing actual (standard) performance of the [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗

read the original abstract

Rapid earthquake magnitude estimation is crucial for effective early warning systems that can save lives and reduce economic damage. In this paper, we present a comprehensive study of magnitude classification using only the vertical component of the initial 7-second P-wave window from a single station. We compare six machine learning approaches that range from traditional models to state-of-the-art deep learning architectures. We also curated a novel dataset of 7,318 earthquake events in South Asia. The dataset was categorized into five Richter-scale classes: slight (3.0-3.9), light (4.0-4.9), moderate (5.0-5.9), strong (6.0-6.9) and severe (>= 7.0). Our experiments show that deep learning models substantially outperform traditional approaches. Our Transformer based architecture achieved 76.23% standard accuracy and 81.56% adaptive accuracy with 4.8 ms inference latency. The adaptive-accuracy metric is introduced for the inherent uncertainty in magnitude estimation of near class boundaries. These results indicate that the attention mechanisms in Transformers combined with adaptive classification effectively capture the temporal dynamics of seismic signals. The architectural advantage facilitates promising generalization to rare high-magnitude events, despite the inherent data scarcity characteristic of seismic catalogs. The adaptive accuracy provides a more realistic assessment of model performance, and the result suggests viability for real-time deployment.

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

New South Asia dataset and a Transformer at 76% on 7-second P-waves, but the rare-event claims rest on unshown splits and balance checks.

read the letter

The paper's core offering is a fresh catalog of 7318 South Asian events and the concrete result that a Transformer reaches 76.23% accuracy (81.56% adaptive) on the first 7 seconds of vertical P-wave from one station. They also report 4.8 ms inference and beat the other five models they tried. That is the part worth noting first. The adaptive accuracy metric is a sensible addition for boundary uncertainty, and the regional focus plus the latency figure make the work relevant for early-warning pipelines. The dataset itself is new and could be reused by others working in the same area. They do a clean head-to-head across traditional and deep models, which is useful even if the architectures are standard. The soft spots sit in the experimental controls. The abstract gives no class counts, no description of the train-test split, and no mention of temporal or spatial leakage prevention. Seismic catalogs are heavily imbalanced, so without those details it is difficult to know whether the reported numbers hold for the severe class or whether the 7-second window actually separates the bins reliably. The claim of promising generalization to rare high-magnitude events therefore sits on thin evidence. This is the kind of applied ML paper that regional monitoring groups or people building single-station classifiers would read. A reader already working on South Asian catalogs or short-window magnitude estimation would get the most out of the dataset and the metric. It deserves a serious referee to verify the splits, the imbalance handling, and whether the performance numbers survive a proper temporal hold-out. Send it to review.

Referee Report

3 major / 2 minor

Summary. The paper claims that machine learning models, particularly a Transformer architecture, can classify earthquakes into five magnitude classes (slight 3.0-3.9, light 4.0-4.9, moderate 5.0-5.9, strong 6.0-6.9, severe >=7.0) using only the vertical component of the initial 7-second P-wave from a single station. On a newly curated dataset of 7318 South Asia events, the Transformer achieves 76.23% standard accuracy and 81.56% adaptive accuracy with 4.8 ms inference latency, outperforming traditional models and suggesting viability for real-time early warning with promising generalization to rare high-magnitude events.

Significance. If the reported accuracies prove robust under proper validation, the work would contribute a novel South Asia seismic dataset and an adaptive accuracy metric that accounts for boundary uncertainty in magnitude estimation. The low inference latency and focus on attention mechanisms for temporal signal dynamics represent practical strengths for early earthquake warning systems. Dataset curation for a region with limited prior ML studies is a clear positive.

major comments (3)

[Dataset] Dataset section: The distribution of the 7318 events across the five magnitude classes is not reported, nor is any analysis of class imbalance or selection criteria to ensure representativeness of real-world operational conditions. This directly undermines evaluation of the generalization claim to rare severe (>=7.0) events.
[Experiments] Experimental setup: No description is provided of the train-test split strategy (temporal, spatial, or random), cross-validation method, or class-imbalance handling (weighting, oversampling). These details are load-bearing for interpreting whether the 76.23% and 81.56% figures reflect genuine performance or overfitting/selection effects.
[Results] Results: No ablation is shown isolating performance on the severe class or testing whether the 7-second single-station vertical P-wave window suffices to separate classes for high-magnitude events; the generalization claim therefore lacks direct supporting evidence.

minor comments (2)

[Abstract] Abstract: The phrase 'Richter-scale classes' should be clarified, as modern catalogs typically use moment magnitude (Mw) or local magnitude (ML); specifying the exact magnitude type improves precision.
[Methods] Methods: The adaptive accuracy metric is introduced without a formal equation or pseudocode definition, hindering exact reproduction and comparison to standard accuracy.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive comments, which highlight important areas for improving the manuscript's clarity and rigor. We will revise the paper to address the missing details on dataset composition, experimental protocols, and targeted analyses. Our responses to each major comment are provided below.

read point-by-point responses

Referee: [Dataset] Dataset section: The distribution of the 7318 events across the five magnitude classes is not reported, nor is any analysis of class imbalance or selection criteria to ensure representativeness of real-world operational conditions. This directly undermines evaluation of the generalization claim to rare severe (>=7.0) events.

Authors: We agree this information is necessary. The revised manuscript will include a table reporting the exact event counts per class and an analysis of imbalance (noting the expected scarcity of severe events). Selection criteria from the ISC catalog will be detailed, including requirements for clear P-wave signals and magnitude reporting. This will allow readers to evaluate the generalization claims directly. revision: yes
Referee: [Experiments] Experimental setup: No description is provided of the train-test split strategy (temporal, spatial, or random), cross-validation method, or class-imbalance handling (weighting, oversampling). These details are load-bearing for interpreting whether the 76.23% and 81.56% figures reflect genuine performance or overfitting/selection effects.

Authors: We will expand the Experimental Setup section to specify the split (stratified random 80/20 to preserve class ratios), 5-fold cross-validation, and imbalance handling via class-weighted loss functions. These additions will demonstrate that the reported accuracies arise from standard validation practices rather than selection artifacts. revision: yes
Referee: [Results] Results: No ablation is shown isolating performance on the severe class or testing whether the 7-second single-station vertical P-wave window suffices to separate classes for high-magnitude events; the generalization claim therefore lacks direct supporting evidence.

Authors: We will add a dedicated ablation subsection with per-class metrics (including severe-class precision/recall) and a comparison of window lengths (e.g., 3s vs. 7s vs. 10s) focused on high-magnitude events. This will provide direct evidence for the generalization statement while acknowledging data scarcity for the severe class. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical test-set accuracies from standard ML training

full rationale

The paper reports 76.23% standard accuracy and 81.56% adaptive accuracy for a Transformer model trained on a curated 7318-event dataset, evaluated on held-out data. These are conventional empirical metrics obtained after model fitting; no equations, self-citations, or ansatzes are shown that reduce the claimed performance or generalization statements to the inputs by construction. The adaptive accuracy metric is introduced to handle boundary uncertainty but is not demonstrated to be tautological with the training process. The derivation chain consists of data curation followed by standard supervised learning and evaluation, which remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central empirical claim rests on the domain assumption that short single-station P-wave windows suffice for magnitude classification and on the quality and representativeness of the newly curated dataset. No new physical entities are postulated. Model hyperparameters constitute free parameters but are not enumerated in the abstract.

free parameters (1)

Transformer hyperparameters and training settings
Standard deep-learning hyperparameters (learning rate, layers, attention heads, etc.) are tuned to achieve the reported accuracy but are not listed.

axioms (1)

domain assumption The initial 7-second vertical P-wave from one station contains sufficient discriminative information for five-class magnitude classification
This premise is required for the entire experimental design.

pith-pipeline@v0.9.0 · 5802 in / 1395 out tokens · 28131 ms · 2026-05-25T00:06:46.446220+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

20 extracted references · 20 canonical work pages

[1]

Seismic microzonation study of south asian cities and its implications to urban risk resiliency under climate change scenario,

O. Mishra, “Seismic microzonation study of south asian cities and its implications to urban risk resiliency under climate change scenario,” International Journal of Geosciences, vol. 11, no. 4, pp. 197–237, 2020

work page 2020
[2]

Generalized seismic phase detection with deep learning,

Z. E. Ross, M. A. Meier, E. Hauksson, and T. H. Heaton, “Generalized seismic phase detection with deep learning,”Bulletin of the Seismolog- ical Society of America, vol. 108, no. 5A, pp. 2894–2901, 2018

work page 2018
[3]

Magnitude estimation based on integrated amplitude and frequency content of the initial p wave in earthquake early warning applied to tehran, iran,

S. Nazeri, Z. H. Shomali, S. Colombelli, L. Elia, and A. Zollo, “Magnitude estimation based on integrated amplitude and frequency content of the initial p wave in earthquake early warning applied to tehran, iran,”Bulletin of the Seismological Society of America, vol. 107, no. 3, pp. 1432–1438, 04 2017. [Online]. Available: https://doi.org/10.1785/0120160380

work page doi:10.1785/0120160380 2017
[4]

The richter scale: its development and use for determining earthquake source parameters,

D. M. Boore, “The richter scale: its development and use for determining earthquake source parameters,”Tectonophysics, vol. 166, no. 1, pp. 1–14, 1989, quantification of Earthquakes and the Determination of Source Parameters. [Online]. Available: https: //www.sciencedirect.com/science/article/pii/004019518990200X

work page arXiv 1989
[5]

Real-time testing of the on-site warning algorithm in southern cali- fornia and its performance during the july 29 2008 mw5.4 chino hills earthquake,

M. B ¨ose, E. Hauksson, K. Solanki, H. Kanamori, and T. H. Heaton, “Real-time testing of the on-site warning algorithm in southern cali- fornia and its performance during the july 29 2008 mw5.4 chino hills earthquake,”Geophysical Research Letters, vol. 36, no. 5, 2009

work page 2008
[6]

A p wave-based, on-site method for earthquake early warning,

S. Colombelli, A. Caruso, A. Zollo, G. Festa, and H. Kanamori, “A p wave-based, on-site method for earthquake early warning,”Geophysical Research Letters, vol. 42, no. 5, pp. 1390–1398, 2015

work page 2015
[7]

Earthquake early warning system in southern italy: Methodologies and performance evaluation,

A. Zolloet al., “Earthquake early warning system in southern italy: Methodologies and performance evaluation,”Geophysical Research Let- ters, vol. 36, no. 5, 2009

work page 2009
[8]

Presto, the earthquake early warning system for southern italy: Concepts, capabilities and future perspectives,

C. Satriano, L. Elia, C. Martino, M. Lancieri, A. Zollo, and G. Ian- naccone, “Presto, the earthquake early warning system for southern italy: Concepts, capabilities and future perspectives,”Soil Dynamics and Earthquake Engineering, vol. 31, no. 2, pp. 137–153, 2011

work page 2011
[9]

Continuous earthquake detection and classification using discrete hidden markov models,

M. Beyreuther and J. Wassermann, “Continuous earthquake detection and classification using discrete hidden markov models,”Geophysical Journal International, vol. 175, no. 3, pp. 1055–1066, 2008

work page 2008
[10]

Phasenet: A deep-neural-network-based seis- mic arrival-time picking method,

W. Zhu and G. C. Beroza, “Phasenet: A deep-neural-network-based seis- mic arrival-time picking method,”Geophysical Journal International, vol. 216, no. 1, pp. 261–273, 2019

work page 2019
[11]

A machine-learning approach for earthquake magnitude estimation,

S. M. Mousavi and G. C. Beroza, “A machine-learning approach for earthquake magnitude estimation,”Geophysical Research Letters, vol. 47, no. 1, p. e2019GL085976, 2020

work page 2020
[12]

Deep learning for real-time p-wave detection: A case study in indonesia’s earthquake early warning system,

A. Wibowo, L. S. Heliani, C. Pratama, D. P. Sahara, S. Widiyantoro, D. Ramdani, M. B. Fuady Bisri, A. Sudrajat, S. T. Wibowo, and S. R. Purnama, “Deep learning for real-time p-wave detection: A case study in indonesia’s earthquake early warning system,”Applied Computing and Geosciences, vol. 24, p. 100194, 2024. [Online]. Available: https://www.sciencedir...

work page 2024
[13]

Lftnet: A lightweight multi- scale attention network for real-time seismic event detection and phase picking,

J. Guo, J. Tian, Y . Guo, and H. Zhang, “Lftnet: A lightweight multi- scale attention network for real-time seismic event detection and phase picking,”Earth and Space Science, vol. 12, no. 9, p. e2025EA004548, 2025

work page 2025
[14]

An investigation of rapid earthquake characterization using single-station waveforms and a con- volutional neural network,

A. Lomax, A. Michelini, and D. Jozinovi ´c, “An investigation of rapid earthquake characterization using single-station waveforms and a con- volutional neural network,”Seismological Research Letters, vol. 90, no. 2A, pp. 517–529, 2019

work page 2019
[15]

Locating induced earthquakes with a network of seismic stations in oklahoma via a deep learning method,

X. Zhang, J. Zhang, C. Yuan, S. Liu, Z. Chen, and W. Li, “Locating induced earthquakes with a network of seismic stations in oklahoma via a deep learning method,”Scientific Reports, vol. 10, no. 1, pp. 1–12, 2020

work page 2020
[16]

Attention is all you need,

A. Vaswaniet al., “Attention is all you need,” inAdvances in Neural Information Processing Systems, 2017, pp. 5998–6008

work page 2017
[17]

Earthquake transformer—an attentive deep-learning model for simultaneous earthquake detection and phase picking,

S. M. Mousavi, W. L. Ellsworth, W. Zhu, L. Y . Chuang, and G. C. Beroza, “Earthquake transformer—an attentive deep-learning model for simultaneous earthquake detection and phase picking,”Nature Commu- nications, vol. 11, no. 1, pp. 1–12, 2020

work page 2020
[18]

Real-time earthquake magnitude estimation via a deep learning network based on waveform and text mixed modal,

H. Zhang, Z. Xu, X. Jin, and M. Song, “Real-time earthquake magnitude estimation via a deep learning network based on waveform and text mixed modal,”Earth, Planets and Space, vol. 76, no. 1, p. 70, 2024

work page 2024
[19]

Earthquake magnitude and location estimation from real time seismic waveforms with a transformer network,

J. M ¨unchmeyer, D. Bindi, U. Leser, and F. Tilmann, “Earthquake magnitude and location estimation from real time seismic waveforms with a transformer network,”Geophysical Journal International, vol. 226, no. 2, pp. 1086–1104, 2021

work page 2021
[20]

Wavecastnet: An AI- enabled wavefield forecasting framework for earthquake early warning,

D. Lyu, R. Wang, D. Melgar, and V . Sahakian, “Wavecastnet: An AI- enabled wavefield forecasting framework for earthquake early warning,” arXiv preprint arXiv:2405.20516, 2024

work page arXiv 2024

[1] [1]

Seismic microzonation study of south asian cities and its implications to urban risk resiliency under climate change scenario,

O. Mishra, “Seismic microzonation study of south asian cities and its implications to urban risk resiliency under climate change scenario,” International Journal of Geosciences, vol. 11, no. 4, pp. 197–237, 2020

work page 2020

[2] [2]

Generalized seismic phase detection with deep learning,

Z. E. Ross, M. A. Meier, E. Hauksson, and T. H. Heaton, “Generalized seismic phase detection with deep learning,”Bulletin of the Seismolog- ical Society of America, vol. 108, no. 5A, pp. 2894–2901, 2018

work page 2018

[3] [3]

Magnitude estimation based on integrated amplitude and frequency content of the initial p wave in earthquake early warning applied to tehran, iran,

S. Nazeri, Z. H. Shomali, S. Colombelli, L. Elia, and A. Zollo, “Magnitude estimation based on integrated amplitude and frequency content of the initial p wave in earthquake early warning applied to tehran, iran,”Bulletin of the Seismological Society of America, vol. 107, no. 3, pp. 1432–1438, 04 2017. [Online]. Available: https://doi.org/10.1785/0120160380

work page doi:10.1785/0120160380 2017

[4] [4]

The richter scale: its development and use for determining earthquake source parameters,

D. M. Boore, “The richter scale: its development and use for determining earthquake source parameters,”Tectonophysics, vol. 166, no. 1, pp. 1–14, 1989, quantification of Earthquakes and the Determination of Source Parameters. [Online]. Available: https: //www.sciencedirect.com/science/article/pii/004019518990200X

work page arXiv 1989

[5] [5]

Real-time testing of the on-site warning algorithm in southern cali- fornia and its performance during the july 29 2008 mw5.4 chino hills earthquake,

M. B ¨ose, E. Hauksson, K. Solanki, H. Kanamori, and T. H. Heaton, “Real-time testing of the on-site warning algorithm in southern cali- fornia and its performance during the july 29 2008 mw5.4 chino hills earthquake,”Geophysical Research Letters, vol. 36, no. 5, 2009

work page 2008

[6] [6]

A p wave-based, on-site method for earthquake early warning,

S. Colombelli, A. Caruso, A. Zollo, G. Festa, and H. Kanamori, “A p wave-based, on-site method for earthquake early warning,”Geophysical Research Letters, vol. 42, no. 5, pp. 1390–1398, 2015

work page 2015

[7] [7]

Earthquake early warning system in southern italy: Methodologies and performance evaluation,

A. Zolloet al., “Earthquake early warning system in southern italy: Methodologies and performance evaluation,”Geophysical Research Let- ters, vol. 36, no. 5, 2009

work page 2009

[8] [8]

Presto, the earthquake early warning system for southern italy: Concepts, capabilities and future perspectives,

C. Satriano, L. Elia, C. Martino, M. Lancieri, A. Zollo, and G. Ian- naccone, “Presto, the earthquake early warning system for southern italy: Concepts, capabilities and future perspectives,”Soil Dynamics and Earthquake Engineering, vol. 31, no. 2, pp. 137–153, 2011

work page 2011

[9] [9]

Continuous earthquake detection and classification using discrete hidden markov models,

M. Beyreuther and J. Wassermann, “Continuous earthquake detection and classification using discrete hidden markov models,”Geophysical Journal International, vol. 175, no. 3, pp. 1055–1066, 2008

work page 2008

[10] [10]

Phasenet: A deep-neural-network-based seis- mic arrival-time picking method,

W. Zhu and G. C. Beroza, “Phasenet: A deep-neural-network-based seis- mic arrival-time picking method,”Geophysical Journal International, vol. 216, no. 1, pp. 261–273, 2019

work page 2019

[11] [11]

A machine-learning approach for earthquake magnitude estimation,

S. M. Mousavi and G. C. Beroza, “A machine-learning approach for earthquake magnitude estimation,”Geophysical Research Letters, vol. 47, no. 1, p. e2019GL085976, 2020

work page 2020

[12] [12]

Deep learning for real-time p-wave detection: A case study in indonesia’s earthquake early warning system,

A. Wibowo, L. S. Heliani, C. Pratama, D. P. Sahara, S. Widiyantoro, D. Ramdani, M. B. Fuady Bisri, A. Sudrajat, S. T. Wibowo, and S. R. Purnama, “Deep learning for real-time p-wave detection: A case study in indonesia’s earthquake early warning system,”Applied Computing and Geosciences, vol. 24, p. 100194, 2024. [Online]. Available: https://www.sciencedir...

work page 2024

[13] [13]

Lftnet: A lightweight multi- scale attention network for real-time seismic event detection and phase picking,

J. Guo, J. Tian, Y . Guo, and H. Zhang, “Lftnet: A lightweight multi- scale attention network for real-time seismic event detection and phase picking,”Earth and Space Science, vol. 12, no. 9, p. e2025EA004548, 2025

work page 2025

[14] [14]

An investigation of rapid earthquake characterization using single-station waveforms and a con- volutional neural network,

A. Lomax, A. Michelini, and D. Jozinovi ´c, “An investigation of rapid earthquake characterization using single-station waveforms and a con- volutional neural network,”Seismological Research Letters, vol. 90, no. 2A, pp. 517–529, 2019

work page 2019

[15] [15]

Locating induced earthquakes with a network of seismic stations in oklahoma via a deep learning method,

X. Zhang, J. Zhang, C. Yuan, S. Liu, Z. Chen, and W. Li, “Locating induced earthquakes with a network of seismic stations in oklahoma via a deep learning method,”Scientific Reports, vol. 10, no. 1, pp. 1–12, 2020

work page 2020

[16] [16]

Attention is all you need,

A. Vaswaniet al., “Attention is all you need,” inAdvances in Neural Information Processing Systems, 2017, pp. 5998–6008

work page 2017

[17] [17]

Earthquake transformer—an attentive deep-learning model for simultaneous earthquake detection and phase picking,

S. M. Mousavi, W. L. Ellsworth, W. Zhu, L. Y . Chuang, and G. C. Beroza, “Earthquake transformer—an attentive deep-learning model for simultaneous earthquake detection and phase picking,”Nature Commu- nications, vol. 11, no. 1, pp. 1–12, 2020

work page 2020

[18] [18]

Real-time earthquake magnitude estimation via a deep learning network based on waveform and text mixed modal,

H. Zhang, Z. Xu, X. Jin, and M. Song, “Real-time earthquake magnitude estimation via a deep learning network based on waveform and text mixed modal,”Earth, Planets and Space, vol. 76, no. 1, p. 70, 2024

work page 2024

[19] [19]

Earthquake magnitude and location estimation from real time seismic waveforms with a transformer network,

J. M ¨unchmeyer, D. Bindi, U. Leser, and F. Tilmann, “Earthquake magnitude and location estimation from real time seismic waveforms with a transformer network,”Geophysical Journal International, vol. 226, no. 2, pp. 1086–1104, 2021

work page 2021

[20] [20]

Wavecastnet: An AI- enabled wavefield forecasting framework for earthquake early warning,

D. Lyu, R. Wang, D. Melgar, and V . Sahakian, “Wavecastnet: An AI- enabled wavefield forecasting framework for earthquake early warning,” arXiv preprint arXiv:2405.20516, 2024

work page arXiv 2024