EUVIMEDEUVIMED entdecken
Crossref

Detection of the 2019 Hikurangi subduction zone slow event using the transformer deep learning model

Tieding Lu, Shuyi Lan, Zhengdong Luo

Acta Geodynamica et Geomaterialia · 2026

DOI: 10.13168/agg.2026.0027 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Acta Geodynamica et Geomaterialia, 2026. Verfasst von Tieding Lu und 2 weiteren Autor:innen. Die Arbeit nennt 41 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The study of slow slip events (SSEs) is of great significance for understanding interplate coupling in subduction zones and revealing the mechanisms governing plate motion. In this study, Global Navigation Satellite System (GNSS) time series from 46 stations in the Hikurangi subduction zone, New Zealand, in 2019 were analyzed. Trend terms, periodic terms, and nonstationary deformation signals were removed to improve the signal-to-noise ratio of slow slip time series, and a Transformer-Attention model was constructed for the automatic detection of SSE occurrence intervals. The results show that, when detecting preprocessed slow slip time series, the model achieved Accuracy values higher than 0.984, Precision values higher than 0.906, Recall values higher than 0.914, and F1-score values higher than 0.948. When detecting raw GNSS time series containing other tectonic deformation signals, the model achieved Accuracy values higher than 0.953, Precision values higher than 0.903, Recall values higher than 0.707, and F1-score values higher than 0.828. These results indicate that the Transformer-Attention model can identify slow slip signals and effectively suppress interference from tectonic deformation signals. Compared with the PatchTST, BiLSTM-Attention, CNN-Attention and TCN models, the Transformer-Attention model improved R² by 42.5 %, 46.1 %, 53.1 %, and 54.3 %, reduced RMSE by 58.3 %, 59.8 %, 62.1 %, and 54.3 %, and reduced MAE by 68.9 %, 70.9 %, 73.3 %, and 77.9 %, respectively. Therefore, the Transformer-Attention model can effectively identify slow slip signals in GNSS time series and provide an important observational basis for investigating the spatiotemporal evolution characteristics of slow slip events.

Zitieren

APA

Lu, T., Lan, S., & Luo, Z. (2026). Detection of the 2019 Hikurangi subduction zone slow event using the transformer deep learning model. Acta Geodynamica et Geomaterialia. https://doi.org/10.13168/agg.2026.0027

Vancouver

Lu T, Lan S, Luo Z. Detection of the 2019 Hikurangi subduction zone slow event using the transformer deep learning model. Acta Geodynamica et Geomaterialia. 2026. doi:10.13168/agg.2026.0027

BibTeX
@article{Lu2026,
  author = {Lu, Tieding and Lan, Shuyi and Luo, Zhengdong},
  title = {Detection of the 2019 Hikurangi subduction zone slow event using the transformer deep learning model},
  journal = {Acta Geodynamica et Geomaterialia},
  year = {2026},
  doi = {10.13168/agg.2026.0027}
}
Kostenlos starten →