Abstract:
To address persistent threats posed by earthquakes and their secondary disasters to human society, major earthquake-monitoring countries and regions worldwide have deployed dense seismic monitoring networks. These networks have evolved into comprehensive observation systems integrating diverse geophysical observables, including seismic waves, crustal deformation, electromagnetic fields, and gravitational fields. Traditional multi-parameter observation systems are built around permanent fixed stations, which offer limited flexibility and incur high deployment costs. Such systems struggle to support post-earthquake emergency observations and temporary field surveys. Therefore, integrating multiple geophysical sensors via an edge-side server to build a low-cost, lightweight integrated geophysical observation platform carries substantial research and practical value.
As a core technical framework for geophysical hazard monitoring, earthquake early warning systems exploit the velocity difference between less-damaging compressional waves (P-waves) and more destructive shear (S-waves) to deliver warning windows ranging from several seconds to tens of seconds for regions near the epicenter. Conventional earthquake early warning systems adopt centralized computing architectures, where edge devices only perform data acquisition while central servers generate warning results. This architecture suffers from two key drawbacks. First, it relies heavily on stable broadband networks for continuous real-time data transmission. In communication-constrained environments such as uninhabited zones and post-earthquake emergency response, limited bandwidth and transmission outages may cause data loss and warning failure. Second, centralized processing of massive datasets imposes heavy computational loads on central servers, degrading system adaptability and stability under complex operational conditions.
This study designed and implemented an edge-side geophysical multi-parameter monitoring and intelligent earthquake early warning system. A Radio Determination Satellite Service (RDSS)-based warning communication mechanism was developed for communication-limited scenarios to enable integrated parallel multi-parameter monitoring. The DiTing seismogram foundation model was deployed locally, and an intelligent earthquake early warning scheme built upon this model was constructed. Developed with a collaborative Qt-Python architecture, the system adopts a multi-threaded parallel processing framework and comprises three core modules. First, the system integrates data acquisition and partitioned management for multiple instruments including geomagnetic sensors, seismometers, cameras, and environmental sensors, supporting parallel monitoring and time–frequency visualization of seismic and geomagnetic data. Second, the DiTing seismogram foundation model is deployed on local hardware via the lightweight FastAPI framework. Using three-component seismic data, the model performs localized real-time P/S-wave phase picking, removing the reliance of intelligent analysis on cloud networks. Third, an RDSS satellite communication module is integrated to establish a monitoring pipeline of “data acquisition–local intelligent analysis–warning information transmission”, enabling transmission of small volumes of ground-motion parameters or warning messages under communication-constrained conditions.
The proposed system was validated through performance and early-warning tests on an edge host with 8 GB RAM and a quad-core Intel Core i5-10300H @2.5 GHz, requiring no cloud services. Under test datasets and simulated streaming inputs, all performance metrics stayed within predefined thresholds, including average and peak CPU utilization, packet loss rate, local API invocation latency, and P-wave response latency. The foundation-model API achieved an inference latency of approximately 1 s; latency can be reduced to roughly 0.5 s when only the early warning module is retained. Operators can select suitable operating modes according to hardware capability, application scenarios, and operational requirements.
Under routine monitoring conditions, the system achieved favorable detection performance, with an API false-trigger rate of merely 0.029 2% and an average of about 1.27 false triggers per day. Under abnormal instrument states such as baseline drift and data jitter, as well as strong anthropogenic disturbances from human activities, vehicle traffic and subway operations, the false-trigger rate rose noticeably. Nevertheless, the system retained anti-interference capacity and did not suffer total failure. Furthermore, the system supports RDSS satellite links and is promising for deployment in uninhabited regions and emergency response after major earthquakes.
This study preliminarily verified the feasibility of the proposed system for continuous operation, real-time seismic phase picking, and local information transmission. For near-field earthquake records in the test dataset, the system can provide effective early warning windows of several to tens of seconds for monitored areas.