Using Deep Learning for Rapid Earthquake Parameter Estimation in Single-Station Single-Component Earthquake Early Warning System

Earthquake early warning systems (EEWSs) often rely on fast determination of earthquake source parameters, namely, location, magnitude, and depth. In areas where the seismic network is coarse, the capability to determine source parameters based on data recorded by a single station is desirable. Moreover, being able to use a single component of the seismic data might increase the robustness of the system to sensor malfunction and might save on sensor cost and computation time. Here, we propose a hybrid deep learning (DL) model to estimate source parameters based on single-component data recorded by a single station at 3 s after the P-wave onset. The model, which we call EEWS-311, uses a convolutional neural network (CNN) and bidirectional long short-term memory. It is trained and tested on recordings of more than 14000 events by a single station of the Japanese Hi-net high-sensitivity short-period seismic network. Compared with source parameters obtained by conventional methods, our model achieves excellent performance (average errors in latitude, longitude, magnitude, and depth equal to 0.05°, 0.1°, 0.14 velocity magnitude (Mv), and 5.68 km, respectively). The results demonstrate the suitability of EEWS-311 for earthquake early warning in areas with sufficient training data.

I. Introduction

Earthquake early warning systems (EEWSs) aim at reducing the impact of earthquakes on people and infrastructure by issuing alerts across the impact area before shaking hits [1], [2]. A typical approach is to detect the event and determine its source parameters (location, magnitude, and depth) as soon as possible based on the initial few seconds of P-wave data recorded by the seismic sensors located closest to the epicenter [3], [4], and then use the resulting source information to forecast the shaking impact zone. Early methods aimed at estimating both location and magnitude from single-station recordings [5], [6]. With time, network-based methods (relying on recordings by multiple stations) became prevalent and allowed more accurate source parameter estimation, albeit delivered later. However, single-station EEWS remains advantageous in areas where the seismic network is sparse and in situations where timeliness is more critical than accuracy and can help reduce the size of the unseen zone of EEWS [7]. Moreover, an EEWS based on recordings of a single component of ground motion could be advantageous in terms of station cost and computational time and could be applied to seismic networks that rely on single-component sensors. The present study focuses on EEWS using just 3 s of P-wave recording at a single station and on a single component, which to the best of our knowledge has not yet been examined.