Members: Avichai Mishnayot, Gabi Rubanovich

Supervisor: Itzhak Lior

Earthquake early warning (EEW) systems aim to provide advance notice of impending ground motions, enhancing safety and increasing response times. It does so by on-site recording of P-waves (the less damaging wave of the earthquake) and then giving a prediction of how damaging the S-wave (the more damaging wave of the earthquake) is going to be. Distributed acoustic sensing (DAS) utilizes existing optical fiber infrastructure to detect and measure earthquake-induced strain, converting it to actionable ground motion data for improved earthquake early warning systems. However, the vast majority of seismic sensors are located on-land, which delays detection of offshore earthquakes and limits the effectiveness of EEW systems in providing timely alerts. Therefore, using existing fiber optics can solve this challenge by providing better coverage. Here we demonstrate a use of DAS information for real-time on-site damage prediction of earthquakes. We built an empiric model and convolutional neural network to generate predictions of high damage earthquakes, and compared them in order to determine which one results in a more accurate prediction. We managed to convert the data acquired by the optic fiber – strain and strain rate – to ground acceleration and found a linear relationship between the S-wave and the P-wave Based on this relationship we created our models and found that the convolutional neural network gives a better prediction of the most damaging part of the earthquake. The implication of this project can be an EEW system using existing infrastructure that is able to provide important extra time in order to prepare for a high-damage earthquake. This solution is cheaper and more robust than any existing today.