Earthquakes Within Earthquakes: Patterns in Rupture Complexity

Abstract

Earthquake source time functions carry information about the complexity of seismic rupture.We explore databases of earthquake source time functions and find that they are composed of distinct peaks that we call subevents. We observe that earthquake complexity, as represented by the number of subevents, grows with earthquake magnitude. Patterns in rupture complexity arise from a scaling between subevent moment and main event moment. These results can be explained by simple 2-D dynamic rupture simulations with self-affine heterogeneity in fault prestress. Applying this to early magnitude estimates, we show that the main event magnitude can be estimated after observing only the first few subevents.Plain Language Summary Seismograms are measurements of waves from earthquakes.They give us information about what happened on the fault at the place where the earthquake occurred. Seismograms can be difficult to interpret because they are often very complicated. Why? One reason is that the waves change when they travel long distances between the fault and a seismometer. Seismologists correct for this effect, however, by constructing something called a source time function. Source time functions are much easier to understand than raw seismograms. We examine a catalog of source time functions from around the world. We find that large earthquakes are composed of many smaller events that we call subevents. The size of a subevent is related to the size of the main earthquake. One important outcome is that we can predict the final size of an earthquake after observing only the first few subevents.

Improving Early Estimates of Large Earthquake's Final Fault Lengths and Magnitudes Leveraging Source Fault Structural Maturity Information

“…Prior to this study, most research on rupture length prediction relied exclusively on seismic or geodetic data measuring properties of the ongoing earthquake (Danré et al, 2019; Denolle, 2019; Goldberg et al, 2019; Ide, 2019; Meier et al, 2017; Melgar & Hayes, 2019). While these studies suggest some degree of determinism, the question remains of how much determinism exists and why.…”

“…Some authors suggest that the final rupture size is distinguishable from 35% to 55% of the total duration of the STF (e.g., Meier et al, 2017), while others propose that it may be the case from 20% of the rupture duration (e.g., Danré et al, 2019). Our results are in better agreement with the more optimistic estimates of Danré et al (2019): in our approach, the final earthquake length is predicted, on average, from the first ~20% of the rupture; and even though there is some variability both among events and between about 30 and 50% of the way through the rupture for most earthquakes, the predicted length remains steadily close overall to the actual final length from ~20% of the way through the rupture. Additionally, the final earthquake magnitude is predicted even earlier, from ~10% of the rupture, and without strong variability among events.…”

“…Previous studies on rupture prediction largely used data only from the earthquakes themselves and primarily focused on the first few seconds of the recorded seismic waveforms, including the earthquake source time function (STF) (e.g., Danré et al, 2019; Denolle, 2019; Meier et al, 2017). The STF describes the moment release of the earthquake as a function of time, though given numerous uncertainties in the initial portion of the signal, it can be difficult to determine in real time.…”

Source time function clustering reveals patterns in earthquake dynamics

“…The number of prominent peaks is commonly used for topographic relief analysis and is defined as the amplitude of the peak (hill summit) relative to the lowest amplitude point valleythat does not contain a higher peak. This metric differs from the calculation of Gaussian subevents that Danré et al (2019) use. One hyper-parameter we tune is a threshold for peak amplitude of the prominent peak, which we choose to be 10% of the global maximum of the STF amplitude.…”

“…One hyper-parameter we tune is a threshold for peak amplitude of the prominent peak, which we choose to be 10% of the global maximum of the STF amplitude. The raw and stretched STFs have a lot fewer prominent peaks than individual peaks from the Gaussian decomposition by Danré et al (2019) ( Figure S2). Furthermore, the stretched STFs have fewer prominent peaks than the raw STFs, but in general the same number of prominent peaks as the centroid event (Figure S3).…”