Fast Information Streaming Handler (FisH): A Unified Seismic Neural Network for Single Station Real-Time Earthquake Early Warning

• DOI: 10.48550/arxiv.2408.06629
• Publication Date: 2024-08-13
• Authors: Tianning Zhang; Feng Liu; Yuming Yuan; Rui Su; Wanli Ouyang; Lei Bai
• Keywords: Earthquake warning system

Existing EEW approaches often treat phase picking, location estimation, and magnitude estimation as separate tasks, lacking a unified framework. Additionally, most deep learning models in seismology rely on full three-component waveforms are not suitable for real-time streaming data. To address these limitations, we propose novel seismic neural network called Fast Information Streaming Handler (FisH). FisH is designed to process data generate simultaneous results an end-to-end fashion. By integrating tasks within single model, simplifies the overall leverages nonlinear relationships between improved performance. The model utilizes RetNet its backbone, enabling parallel processing during training recurrent handling inference. This capability makes applications, reducing latency systems. Extensive experiments conducted STEAD benchmark dataset provide strong validation effectiveness of our proposed model. demonstrate that achieves impressive performance across multiple event detection characterization tasks. Specifically, it F1 score 0.99/0.96. Also, demonstrates precise earthquake with error only 6.0km, distance 2.6km, back-azimuth 19{\deg}. also exhibits accurate just 0.14. capable generating estimations, providing estimations 8.06km 0.18 mere 3 seconds after P-wave arrives.