A novel earthquake early warning system (EEWS) has been proposed by researchers Jawad Fayaz, Rodrigo Astroza, and Sergio Ruiz in a recent study published in Engineering. The system, named HEWFERS (Hybrid Earthquake Early Warning Framework for Estimating Response Spectra), integrates deep learning techniques with seismological principles to provide real-time predictions of ground shaking intensity, offering a significant advancement in the field of seismic risk reduction.

Earthquakes pose a substantial threat to global safety, often resulting in significant loss of life and infrastructure damage. According to the study, between 1990 and 2019, earthquakes caused over 1.3 million fatalities worldwide, and in 2020, they accounted for 34% of global insured losses from natural catastrophes. To mitigate these devastating effects, the development of effective EEWSs has become a critical area of research. The United Nations’ Sendai Framework for Disaster Risk Reduction 2015–2030 underscores the importance of enhancing community resilience against natural disasters, including earthquakes.

HEWFERS represents a step forward in this direction by leveraging a hybrid approach that combines a domain-informed variational autoencoder (VAE) for physics-based latent variable extraction, a feed-forward neural network (FFNN) for on-site prediction, and Gaussian process regression (GPR) for spatial prediction. This integration allows the system to provide both on-site and regional early warnings, significantly improving the accuracy and timeliness of ground motion intensity predictions.

The system operates in two phases. In Phase I, immediately after an earthquake rupture, the framework uses the first seismic station to detect the initial waves and estimate latent variables representing the ground motion spectrum. These estimates are then used to provide a regional early warning through spatial regression. In Phase II, once the seismic waves reach the target station, the system updates the predictions using Bayesian methods, incorporating on-site data to refine the estimates.

The study validates the effectiveness of HEWFERS using a comprehensive database of approximately 14,000 recorded ground motions from the K-Net and KiK-Net networks in Japan. The results demonstrate that the system can provide accurate predictions of ground motion intensity, with the posterior estimates from Phase II showing a high degree of accuracy and reduced uncertainty compared to the prior estimates from Phase I.

A key innovation of HEWFERS is the use of explainable artificial intelligence (XAI) techniques, specifically Shapley additive explanations (SHAP), to interpret the predictive mechanisms of the system. This transparency allows stakeholders to understand the factors influencing the predictions, thereby enhancing the credibility and usability of the system for decision-making.

The authors highlight that HEWFERS is designed to be scalable and adaptable to different regions. While the model was trained primarily on Japanese data, its data-driven nature allows it to be retrained for other geographic areas, capturing localized seismic characteristics. This adaptability aligns with the UN’s goals of promoting global resilience against seismic threats.

HEWFERS offers a robust and accurate tool for earthquake early warning, capable of providing timely and reliable predictions of ground shaking intensity. Its integration of advanced machine learning techniques with seismological principles not only enhances the precision of EEWSs but also provides a transparent and interpretable framework for seismic risk management.

The paper “An Interpretable and Domain-Informed Real-Time Hybrid Earthquake Early Warning for Ground Shaking Intensity Prediction,” is authored by Jawad Fayaz, Rodrigo Astroza, Sergio Ruiz. Full text of the open access paper: https://doi.org/10.1016/j.eng.2025.03.009. For more information about Engineering, visit the website at https://www.sciencedirect.com/journal/engineering.