On-Site Earthquake Early Warning Using Smartphones.

Area of Science:

• Geophysics
• Computer Science
• Engineering

Background:

• Traditional earthquake early warning systems often rely on centralized networks.
• Crowdsourcing approaches for earthquake early warning are emerging.
• On-site detection offers potential for faster, localized alerts.

Purpose of the Study:

• To develop and evaluate an on-site earthquake early warning system utilizing smartphone accelerometers.
• To classify triggered events as earthquakes and predict peak ground acceleration (PGA).
• To assess the accuracy and lead time of alerts generated by the system.

Main Methods:

• Utilized measured accelerations from a single smartphone for detection and warning.
• Employed a neural network classifier to distinguish earthquake events from non-earthquake triggers.
• Implemented a neural network predictor for real-time PGA estimation post-trigger.
• Validated the system using shake table tests with global and Meinong earthquake datasets.

Main Results:

• The system achieved high accuracy in earthquake classification and PGA prediction.
• 96.9% of issued alerts were correct.
• 61.9% of significant earthquakes triggered an alert before PGA arrival, with most false negatives due to negative lead time.

Conclusions:

• The proposed smartphone-based system provides a promising on-site earthquake early warning solution.
• The approach demonstrates high accuracy and potential for effective pre-shaking alerts.
• Further development could enhance lead times and reduce false negatives.