Using Seismic Statistical Features and VQ-VAE to Improve Spatiotemporal Seismicity Predictability
Abstract
In this paper we build upon a previous study in which we demonstrated, using XGBoost and earthquake catalogue data from Japan and Chile, that a set of 60 seismic statistical features (SSFs) had much greater predictive value than a set of 428 generic time series features from the tsfresh package. We here extend this previous work in two key ways, focusing on data from Japan as a large dataset is necessary in order to allow for the training of a deep learning (autoencoder) model. First, we move from whole-region prediction (considering, for each candidate event, the likelihood of an event M ≥ 5.0 anywhere in the region in the next 15 days) to localised predictions in which both the region of feature computation and the region of prediction are restricted to a circle of radius 24 km around the candidate event, and we show that performance remains excellent, similar to our previous whole-region study for the same area. Second, we here couple this proven set of SSFs, based on one-dimensional (catalogue) data, with a novel feature based on magnitude-stratified two-dimensional seismic maps, obtained by training a VQ-VAE model to reproduce such maps as output and identifying a measure of its error in doing so with anomalies in magnitude distribution reflecting a localised build-up of crustal stress. We show that while localised prediction based only on a set of SSFs can be effective, with test AUC values as high as those obtained in the case of Japan in our previous whole-region study, the inclusion of the new natively-spatial VQ-VAE-derived feature, top-ranked by SHAP analysis, both improves prediction and points toward a new approach that might allow computationally time-demanding traditional seismic features to be entirely set aside.
Related
- A Hybrid Deep Learning Framework for Tropical Cyclone Intensity Estimation With Spatiotemporal Features and Physical‐Persistence Adjustment
- Development of a Combined Detection and Relocation Workflow (TM2DD) to Characterize Escalating Hydraulic Fracturing Induced Seismicity in the Karnes Trough Fault Zone of the Eagle Ford Shale Play, Texas
- Escalating soil CO ₂ degassing correlates with ground uplift and seismicity during the ongoing unrest of Campi Flegrei (2005–2025)
- Optimizing induced seismicity monitoring for CO2 injection at Trüllikon, Switzerland using risk-informed guidelines
- Tectonic regionality and geophysical controls of the fractal dimension and b-value of seismicity in the Chinese mainland
BibTeX
@misc{quan2026using,
title = {{Using Seismic Statistical Features and VQ-VAE to Improve Spatiotemporal Seismicity Predictability}},
author = {Wei Quan and Denise Gorse},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2606.10069},
eprint = {2606.10069},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2606.10069},
}
RIS
TY - UNPB
TI - Using Seismic Statistical Features and VQ-VAE to Improve Spatiotemporal Seismicity Predictability
AU - Wei Quan
AU - Denise Gorse
PY - 2026
DA - 2026/09/09
PB - arXiv
DO - 10.48550/arxiv.2606.10069
UR - https://arxiv.org/abs/2606.10069
AB - In this paper we build upon a previous study in which we demonstrated, using XGBoost and earthquake catalogue data from Japan and Chile, that a set of 60 seismic statistical features (SSFs) had much greater predictive value than a set of 428 generic time series features from the tsfresh package. We here extend this previous work in two key ways, focusing on data from Japan as a large dataset is necessary in order to allow for the training of a deep learning (autoencoder) model. First, we move from whole-region prediction (considering, for each candidate event, the likelihood of an event M ≥ 5.0 anywhere in the region in the next 15 days) to localised predictions in which both the region of feature computation and the region of prediction are restricted to a circle of radius 24 km around the candidate event, and we show that performance remains excellent, similar to our previous whole-region study for the same area. Second, we here couple this proven set of SSFs, based on one-dimensional (catalogue) data, with a novel feature based on magnitude-stratified two-dimensional seismic maps, obtained by training a VQ-VAE model to reproduce such maps as output and identifying a measure of its error in doing so with anomalies in magnitude distribution reflecting a localised build-up of crustal stress. We show that while localised prediction based only on a set of SSFs can be effective, with test AUC values as high as those obtained in the case of Japan in our previous whole-region study, the inclusion of the new natively-spatial VQ-VAE-derived feature, top-ranked by SHAP analysis, both improves prediction and points toward a new approach that might allow computationally time-demanding traditional seismic features to be entirely set aside.
ER -