Rapid Magnitude Assessment of Large Earthquakes From Recurrent Neural Networks

Date: 4/25/2019

Time: 06:00 PM

Room: Grand Ballroom

Earthquake early warning (EEW) systems provide seconds to minutes of warning to both people and “automated systems” before strong shaking occurs at their location. Rapid and accurate magnitude estimation is critical to the success of EEW systems, however, fast magnitude determination for large (M_w8+) earthquakes is very challenging for modern systems because of the two factors: 1. the limitations of inertial-based seismic equipment and 2. the initial features of the small and large earthquakes are unable to be distinguished by the current method. Here, we demonstrate that by applying a recurrent neural network (RNN) to Global Positioning System (GPS) waveforms generated from several hundred synthetic Cascadia megathrust earthquakes, we are able to rapidly determine the moment magnitude (M_w) from the synthetic GPS data before the rupture has ended. On average, we can determine the final earthquake magnitude roughly halfway through the rupture process, which strongly supports the idea of weak earthquake determinism. This method can improve currently operating earthquake and tsunami early warning systems. We first focus on large Cascadia Subduction Zone ruptures (M8+), where there is a pressing need for such an algorithm and will ultimately expand to global subduction zones.

Presenting Author: Jiun-Ting Lin

Authors