Supervised Machine Learning of High Rate GNSS Velocities for Earthquake Strong Motion Signals
Abstract
High rate GNSS processed time series capture a broad spectrum of earthquake ground motion signals but experience regular sporadic noise that can be difficult to distinguish from true seismic signals. Noise events often have similar amplitude signatures to true signals. The range of possible signal frequencies amidst a relatively high, location- and time-varying noise floor makes filtering difficult to generalize. Existing methods for automatic detection rely on external inputs to mitigate unacceptable false alert rates, which will limit their usefulness. For these reasons seismic signal detection within GNSS velocity time series makes for a compelling candidate for data-driven machine learning classification. We generated and labeled through visual inspection a dataset of 1701 5Hz GNSS time differenced carrier phase velocity (TDCP) time series concurrent in space and time with expected seismic surface waves from 77 events ranging from Mw4.8 to Mw8.2. The events observed occur over nearly 20 years across a range of fault and sensor network geometries.
A supervised random forest classifier model outperforms the existing geodetic event detection methods in stand-alone mode. We employed a nested cross validation approach to minimize testing bias of our dataset and evaluated its real-time performance on unseen events. TDCP velocity processing has increased sensitivity relative to traditional geodetic displacement processing without requiring sophisticated corrections. This geodetic approach does not replace the sensitivity of traditional seismic infrastructure to the earliest phase picks or weaker signals. The improved performance of this classifier encourages the intelligent inclusion of these valuable signals of opportunity in operational ground motion observations and models. Capturing these unsaturated peak dynamics from geodetic observations are of particular value in larger magnitude events and in complementing sparse seismic networks. We believe this model encourages future work integrating these GNSS velocity measurements with the wide array of developing earthquake monitoring ML techniques using traditional seismic measurements.- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2022
- Bibcode:
- 2022AGUFM.S53A..06D