SeisMamba: Low-Latency Single-Station Seismic Magnitude Estimation for Spatially Distributed Earthquake Early Warning

Quenton Yeo; Zhaoge Bi; Linghan Huang; Luke Stephen Higgins; Flora Salim; Huaming Chen

cs.LGAug 25, 2026

SeisMamba: Low-Latency Single-Station Seismic Magnitude Estimation for Spatially Distributed Earthquake Early Warning

Authors: Quenton Yeo, Zhaoge Bi, Linghan Huang, Luke Stephen Higgins, Flora Salim, Huaming Chen

Organizations: The University of Sydney Sydney, Australia · Accenture Sydney, Australia · The University of New South Wales Sydney, Australia

Abstract

Rapid earthquake magnitude estimation is central to earthquake early warning, yet many operational systems depend on dense regional seismic networks and region-specific calibration. This creates a spatial coverage barrier for high-risk areas with sparse sensing infrastructure. Single-station learning offers a lower-cost alternative, but existing models often face an accuracy--latency trade-off and may degrade under regional distribution shift. We present SeisMamba, a lightweight Mamba-based architecture for low-latency magnitude estimation from minimally processed three-component seismic waveforms recorded at a single station. SeisMamba combines hierarchical convolutional encoding, sparse selective state-space modelling, multi-scale feature fusion, and an auxiliary temporal prediction head to support efficient long-sequence waveform analysis. On the STEAD benchmark, SeisMamba achieves the best MSE, RMSE, and $R^2$ among tested baselines while requiring only 0.55 ms for a batch of 32 waveforms on an NVIDIA T4 GPU, making it about three times faster than transformer-based baselines. We further conduct a Chile--Taiwan regional hold-out experiment as a diagnostic test of cross-region deployment, where SeisMamba retains useful performance on geographically unseen seismic regions. These results suggest that selective state-space waveform modelling provides a promising accuracy--latency backbone for spatially distributed, low-cost earthquake early warning.

Figures & tables

Figure 1. SeisMamba architecture overview. Local convolutions extract short-range waveform morphology, sparse Mamba blocks model longer temporal context, and the output heads produce scalar and time-resolved magnitude estimates. A pipeline diagram showing a three-component seismic waveform processed by a convolutional encoder, sparse Mamba blocks, multi-scale feature fusion, and scalar and temporal prediction heads.

Model	MSE	RMSE	MAE	$R^{2}$	Time
PhaseNet	0.1316	0.3627	0.2724	0.8638	0.50
MagNet	0.1245	0.3528	0.2561	0.8712	0.97
EQTransformer	0.0860	0.2932	0.2016	0.9237	1.70
AMAG	0.0681	0.2609	0.1467	0.9389	1.58
U-Mamba	0.0683	0.2613	0.1793	0.9282	1.39
SeisMamba	0.0628	0.2506	0.1566	0.9443	0.55

Table 1. Benchmark comparison on STEAD. Time is measured in ms per batch of 32 waveforms on an NVIDIA T4 GPU. Best values are bold.

Figure 2. Spatial hold-out evaluation setup. Events from Chile and Taiwan are excluded from training and reserved only for geographically unseen testing, while the remaining STEAD events are used for training and validation. World map showing global STEAD training events and held-out test regions in Chile and Taiwan for regional transfer evaluation.

Setting	MSE	RMSE	MAE	$R^{2}$	Time
Regional hold-out
Chile–Taiwan	0.1669	0.4085	0.2808	0.8518	0.58
Ablation on STEAD
Encoder only	0.0763	0.2762	0.1776	0.9323	0.56
Wider	0.0669	0.2587	0.1608	0.9406	0.55
Deeper	0.0660	0.2569	0.1608	0.9414	1.09

Table 2. Regional hold-out and ablation results. Time is measured in ms per batch of 32 waveforms.

Seismology

Seismology studies earthquakes and seismic waves to understand Earth’s structure, monitor faults, and assess geological and infrastructure hazards.

See all 43 papers

Real-Time Earthquake Magnitude Classification from Initial P-Waves: Models, Dataset, and Comparative Analysis for South Asia
May 10, 2026Md Nasiat Hasan Fahim, Md. Abid Ullah Muhib, Rayhanul Amin Tanvir +1Time Series Classification Seismology
Department of Computer Science and Engineering Shahjalal University of Science and Technology Sylhet, Bangladesh
Event-Centric Earthquake Monitoring with a Multi-Station Transformer
Oct 12, 2026Haoyu Wang, Weiqiang ZhuSeismology
Department of Earth & Planetary Science, University of California, Berkeley · Berkeley Seismology Laboratory, University of California, Berkeley
Data-Driven Forecasting of three-Component Seismograms Using Transformer Architectures
Jun 1, 2026Waleed Esmail, Stuart Russell, Jana Klinge +2Multivariate Time Series Forecasting Time Series Forecasting
Institut für Kernphysik, Universität Münster, Wilhelm-Klemm-Straße 9, 48149, Münster · Institut für Geophysik, Universität Münster, Corrensstraße 24, 48149, Münster · James Cook University, 1 James Cook Drive, Douglas, Queensland, 4814, Australia +1

May 10, 2026J/K move · Enter open · S save

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

Department of Computer Science and Engineering Shahjalal University of Science and Technology Sylhet, Bangladesh

Time Series Classification Seismology

Open paper AskPDF arXiv

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.

Abstract

Seismology

Explore similar work

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

Event-Centric Earthquake Monitoring with a Multi-Station Transformer

Data-Driven Forecasting of three-Component Seismograms Using Transformer Architectures