An Attention-based Framework with Multistation Information for Earthquake Early Warnings

TL;DR

The paper addresses the limitation of single-station earthquake early warning by introducing SENSE, an attention-based multistation framework that aggregates signals from $N$ stations to predict regional shaking intensities. It employs an encoder–decoder architecture with a ConvolutionModule for waveform data, a PositionalEncoding for geography, and locality-specific embeddings, followed by a Transformer/Conformer-based feature blending and a flexible PredictionModule that supports discrete PGA-level classification or a continuous Gaussian Mixture Network. Experimental results on the Japan KiK-net and Taiwan networks show that SENSE outperforms or matches state-of-the-art baselines (ISMP, TEAM), with ablations highlighting the benefits of weighting factors and early/late locality encodings. The work demonstrates that multistation attention and locality-aware representations can extend reliable warnings to distant areas and improve robustness across dense seismic networks, supporting faster and more widespread earthquake responses.

Abstract

Earthquake early warning systems play crucial roles in reducing the risk of seismic disasters. Previously, the dominant modeling system was the single-station models. Such models digest signal data received at a given station and predict earth-quake parameters, such as the p-phase arrival time, intensity, and magnitude at that location. Various methods have demonstrated adequate performance. However, most of these methods present the challenges of the difficulty of speeding up the alarm time, providing early warning for distant areas, and considering global information to enhance performance. Recently, deep learning has significantly impacted many fields, including seismology. Thus, this paper proposes a deep learning-based framework, called SENSE, for the intensity prediction task of earthquake early warning systems. To explicitly consider global information from a regional or national perspective, the input to SENSE comprises statistics from a set of stations in a given region or country. The SENSE model is designed to learn the relationships among the set of input stations and the locality-specific characteristics of each station. Thus, SENSE is not only expected to provide more reliable forecasts by considering multistation data but also has the ability to provide early warnings to distant areas that have not yet received signals. This study conducted extensive experiments on datasets from Taiwan and Japan. The results revealed that SENSE can deliver competitive or even better performances compared with other state-of-the-art methods.

Figures (4)

• Figure 1: Model architectures of (a) Transformers, (b) Conformers, and (c) the proposed SENSE.
• Figure 2: A simple example of using a Gaussian mixture model to approximate the probability density at a given location. The blue dashed lines represent three Gaussians with different means and standard deviations. The green curve denotes the result of mixing the three Gaussians. The green area below the green curve is the probability of the earthquake exceeding 8.1%g occurring at the location.
• Figure 3: Japan dataset. (a) Location distribution of stations. (b) Distribution of earthquake event magnitudes.
• Figure 4: Taiwan Dataset. (a) Location distribution of stations. (b) Distribution of earthquake event magnitudes.