Stress drop estimation from source spectra: Resolution limits and consistency between different studies.

Abstract:

The average stress drop during an earthquake is a parameter fundamental to ground motion prediction and earthquake source physics, but it has proved hard to measure accurately, as demonstrated by the lack of agreement between different studies of the same earthquakes. This has limited our understanding of earthquake rupture, as well as the spatiotemporal variations of fault strength. In this study, we investigate the resolution limits of spectral analysis by applying different methods and data-selection constraints to both synthetic spectra, and observed earthquakes.

We use synthetic tests to investigate the resolution of two common approaches: (1) a joint spectral fitting method (“JS”, simultaneously fitting corner frequency and kappa) to define the resolution limit of corner frequency (fc), and (2) a stacking approach based on spectral decomposition (“SNSS”) (Chen & Abercrombie, 2020). These tests show that the SNSS approach can recover the true input stress drop if the fc is within the resolution limit.

We then investigate the effect of using different methods, different waves and window lengths and different data selection criteria to analyze induced earthquakes from the Guthrie and Prague sequences in central Oklahoma (M2-4) and tectonic earthquakes on the San Andreas Fault at Parkfield, California (M1-3). We compare stress drop results from our analyses with those from previously published studies of the same earthquakes. We find both random and systematic variations, and also some consistency.

The comparison suggests that: (1) for the same dataset, two methods obtain similar median values, but the SNSS method has much smaller scatter and a stronger correlation with ground motion event terms; (2) lowering signal-to-noise-ratio (SNR) criteria tends to result in significantly larger scatter and a tendency to suggest scaling with magnitude (3) for the same sequence from different studies, the relative variabilities are often consistent, however, the absolute stress drop values can vary over two orders of magnitude – studies with longer time window (including coda and lower SNR data) or using joint spectral fitting often have much lower stress drop values. For a single M4 earthquake, different methods (SNSS, spectral ratio, finite slip inversion) all obtain similar stress drop values.