Machine Learning for Emulation of Seismic-Phase Travel Times in 3D Earth Models
Date: 4/24/2019
Time: 09:15 AM
Room: Cascade I
Seismic-phase travel time prediction errors are the predominant source of bias in estimates of event locations. Seismic-phase travel times computed using a global 3-D Earth model can reduce travel-time prediction error to approximately 0.6 seconds on average, leading to median event epicenter error of approximately 6 km for a global or regional network with azimuth gap less than 120°. However, using a generalized 3-D model can be cumbersome because computation of travel times through a 3-D model can take 0.1 to 1.0 seconds, which is orders of magnitude too slow for real-time monitoring systems. Alternatively, travel times can be precomputed, but this approach is difficult to implement for a network comprised of a constantly evolving station set.
In an effort to make 3-D models easier to utilize, we develop a machine learning approach to emulate seismic-phase travel time calculation through a 3-D model. Our goal is to establish a computationally efficient way to implement 3-D models in real-time monitoring systems and enable routine utilization of 3-D models in basic research. We demonstrate this approach by training a gradient-boosted regressor using travel times computed through the LLNL-G3D model. The training set is millions of travel times from randomly selected event locations to each network station, as well as randomly selected station locations to provide predictive capability to new stations. Preliminary tests find that machine learning effectively captures global effects like ellipticity and event depth. The effects of the 3-D model can be emulated with average errors of approximately 0.38 seconds which is well below the error of the 3-D model itself. With computation time on the order of 10 micro-seconds, the machine learning emulator may be a practical way to utilize 3-D models in event locators. Prepared by LLNL under Contract DE-AC52-07NA27344.
Presenting Author: Stephen Myers
Authors
Stephen Myers myers30@llnl.gov Lawrence Livermore National Laboratory, Livermore, California, United States Presenting Author
Corresponding Author
|
Derek Jensen Jensen54@llnl.gov Lawrence Livermore National Laboratory, Livermore, California, United States |
Nathan Simmons simmons27@llnl.gov Lawrence Livermore National Laboratory, Livermore, California, United States |
Machine Learning for Emulation of Seismic-Phase Travel Times in 3D Earth Models
Category
From Drifting to Anchored: Advances in Improving Absolute Hypocenter Location Accuracy for Natural, Induced and Explosion Seismic Events