Estimating earthquake magnitude and location from prompt elasto-gravity signal using deep learning

Abstract:

Mass redistribution during large earthquakes produces a prompt elasto-gravity signal (PEGS) that travels at the speed of light and can be observed on seismograms before the arrival of P-waves. Numerical simulations show that PEGS carries information about earthquake magnitude and the temporal evolution of seismic moment. Therefore, PEGS could be used to both improve the accuracy of current early source estimation systems (which are based on P-waves and known to produce biased estimates of Mw for large earthquakes) and speed-up early warning. However, PEGS has been detected for only a handful of very large earthquakes so far, and its potential use for operational early warning remains to be established.

Here we show that Mw can be estimated before the arrival of P-waves below the previously identified limit of 8.0, using PEGS data from subduction earthquakes recorded by broadband seismometers. We tailor a deep learning algorithm to focus on the Japan subduction zone which experienced one of the most destructive subduction earthquakes (2011, Mw=9.1 Tohoku earthquake). Given the paucity of PEGS observations, we train a regression, double-branched convolutional neural network (CNN) on a database of synthetic seismograms for Mw and location estimation from PEGS. We augment the database with empirical noise in order to simulate more realistic waveforms at about 80 station locations from the Japanese F-Net network and from networks with data available through IRIS. Under this experimental setup, we find that Mw can be estimated in the Mw range 7.0 - 9.5 with a standard deviation of ~0.5. The application of our model to real data from eight subduction earthquakes in Japan shows that uncertainties on Mw are proportional to the level of background noise in the recordings. For data with low noise amplitudes, the model is able to predict Mw even for moderate magnitude events (Mw<7.5).

Our results highlight the potential of PEGS to enhance the performance of existing earthquake early warning systems for large earthquakes (Mw above 7.0).