← Back to Gallery

WITHDRAWN Rapid Earthquake Magnitude Estimation From P-wave Strains: Comparing Borehole Strain Meters and DAS

WITHDRAWN Distributed acoustic sensing (DAS) is a promising sensing technology for improving earthquake early warning (EEW) in offshore settings where maintaining seismic networks remains prohibitively expensive. Ascertaining precise earthquake locations using DAS data, however, remains a difficult task given the poor azimuthal coverage of these typically semi-linear arrays, and estimating earthquake magnitudes with DAS strain measurements is an unresolved problem requiring more research. Here, we introduce a method for discriminating large, potentially damaging, earthquakes from smaller events based on the first 4 seconds of strain waveform after a P-wave arrival without calculating earthquake locations. We train a suite of ensemble random forest machine learning models to predict whether borehole strain meter waveforms generated by local earthquakes (3.5<=M<=7.1) are from M>=5.4 events, then test the predictive models on DAS data (3.5<=M<=7). Our random forest features include the median, max, range, and minimum of continuous wavelet transforms scalograms filtered at three different bandwidths (0.2-0.5 Hz, 0.5-2 Hz, and 2-5 Hz) calculated over 4-second sliding windows and using 10 wavelet families (variations of Gaussian and Morlet wavelets), as well as the max and range of the strain amplitude, and whether the maximum strain is >= 10^5, 10^6, or 10^7. We discover that the power of coefficients of continuous wavelet transforms filtered at the lowest (0.2-0.5 Hz) frequencies is the strongest predictor of larger magnitudes, capturing both long-period dynamic strains and near-field, static deformation. The reliability and precision of our offline results are comparable to real-time EEW operational systems and could be a useful strategy for employing DAS for EEW in both terrestrial and offshore environments.

Session: 

Type: 

Room: 

Date: 

Presentation Time: (local time)

Presenting Author: Theresa Sawi

Student Presenter: No

Invited Presentation: 

Poster Number:

Additional Authors