TREAD: A Distance-Aware Transformer for Regional Earthquake Early Warning
Seismological Research Letters · 21.07.2026
DOI: 10.1785/0220260045 ↗Kurz gefasst
Fachartikel in der Zeitschrift Seismological Research Letters, online erschienen am 21.07.2026. Verfasst von Min-Xuan Xu und 1 weiteren Autor:innen. Die Arbeit nennt 36 Literaturangaben.
Abstract (Originalsprache)
Earthquakes can cause severe damage to infrastructure and result in significant casualties. Earthquake early warning (EEW) systems aim to provide timely alerts within seconds after an event begins, enabling protective actions before strong ground shaking arrives. Seismic observations suggest that shaking intensity often exhibits concentric spatial patterns, underscoring the importance of modeling spatial geometry in regional EEW. To this end, we propose Transformer for Earthquake Alert with Distance-awareness (TREAD)—a Transformer-based regional EEW model that incorporates a physically meaningful auxiliary task: regressing distances from seismic stations to target locations where ground motion is predicted. This design explicitly encodes spatial geometry and enables distance-aware interactions. By integrating distance awareness into the modeling process, TREAD improves predictive accuracy. Experiments on the Central Weather Administration and Kyoshin network datasets demonstrate that TREAD outperforms representative baseline methods, achieving higher predictive accuracy and more reliable early warnings across multiple ground-motion thresholds.
Zitieren
APA
Xu, M., & Chen, K. (2026). TREAD: A Distance-Aware Transformer for Regional Earthquake Early Warning. Seismological Research Letters. https://doi.org/10.1785/0220260045
Vancouver
Xu M, Chen K. TREAD: A Distance-Aware Transformer for Regional Earthquake Early Warning. Seismological Research Letters. 2026. doi:10.1785/0220260045
BibTeX
@article{Xu2026,
author = {Xu, Min-Xuan and Chen, Kuan-Yu},
title = {TREAD: A Distance-Aware Transformer for Regional Earthquake Early Warning},
journal = {Seismological Research Letters},
year = {2026},
doi = {10.1785/0220260045}
}
EUVIMED