NCREE Studies Machine Learning for Soil Radon Seismic Precursor Analysis
Frontier AI models · Structures
NCREE researchers combined soil radon monitoring across Taiwan fault zones with ensemble machine learning to detect pre-earthquake anomalies.
Researchers at the National Center for Research on Earthquake Engineering integrated geochemical soil gas monitoring with machine learning to evaluate seismic precursors in Taiwan. Using monitoring stations along active fault zones including Hsinchu, Tainan, Taipei, and Yilan Jiaoxi, the team built a stacked ensemble model integrating CatBoost, LightGBM, XGBoost, and Random Forest. After systematically removing meteorological and seasonal trends, preliminary tests using Jiaoxi station data demonstrated improved identification of subtle pre-seismic radon anomalies over single models.
Key takeaways
- The stacked ensemble integrates CatBoost, LightGBM, XGBoost, and Random Forest to process complex geophysical data.
- The workflow removes temperature, humidity, and barometric trends before applying anomaly detection to purified residuals.
Source：國家地震工程研究中心｜地震工程研究簡訊 Read the original ↗
Prepared, checked and automatically published by AI; not individually reviewed by a human. Translation checked by AI.