← Back to Sessions

Computation of High-Precision, Deep Magnitude Earthquake Catalogs on a Massive Scale

Description: 

The tremendous growth and availability of continuous seismic data over the past decades have posed challenges to building comprehensive catalogs of reliable earthquake parameters. But it also sparked the development of new methods for efficiently computing such catalogs at unprecedented resolution and scales, including millions of earthquakes and billions of measurements, and spanning decades of observations at local to plate tectonic scales. Here we show results from using supervised machine learning (ML) and waveform cross-correlation for (sub-noise) event detection and phase arrival and delay time measurement, unsupervised ML for event characterization and discrimination, inversion and grid search methods for accurate absolute hypocenter location, and double-difference methods for precise relative location. These tools are tailored and combined in computational workflows that efficiently handle massive amounts of data, both for retro-active and real-time processing, in regions of high-rate/high-density (e.g., volcano, induced) and low-rate/low-density seismic activity (e.g., stable continental regions). We focus here on one of the most challenging tasks, the assessment of robustness and effective resolution of the resulting catalogs. It is often unfeasible to obtain formal errors at these scales, but we show how statistical resampling methods can be used to evaluate the sensitivity of location estimates to measurement errors and choices of apriori information, and how the complementary nature of our data can be harnessed to evaluate estimates of measurement uncertainties. Finally, the accuracy of any earthquake catalog should be evaluated with respect to the scientific questions it is used to address, therefore additional information about the source (e.g., for studies of earthquake interaction) or tectonic setting (for fault studies) need to be incorporated. Several of the methods we discuss here are extensively used worldwide, and ongoing efforts within the SCOPED project focus on implementing these workflows on a hybrid cloud+HPC computational platform for use by the broader scientific community.

Session: It’s All About Relocation, Relocation, Relocation

Type: Oral

Date: 4/20/2023

Presentation Time: 05:00 PM (local time)

Presenting Author: Felix Waldhauser

Student Presenter: No

Invited Presentation: 

Authors