Lessons from recent earthquakes
- Why are earthquakes sometimes so powerful? Stress interactions can trigger a domino effect, causing multiple ruptures and leading to devastating earthquakes, like those in Turkey in 2023.
- Insights into the interlocked multi-fault ruptures and regional seismic hazard assessment: the 2019 Ridgecrest sequence.
- Unveiling the hidden threat: How a seemingly Mw 7.5 earthquake concealed a slow rupture episode which triggers a powerful global-spreading tsunami.
There was a pair of major earthquakes in Turkey this year that caused a lot of destruction and fatalities (one of the deadliest disasters in the region in historic times). The sequence was unusual because the two earthquakes were almost of the same size (called a “doublet”), which compounded the damage. The earthquakes occurred on known faults, and in this sense were expected. What was unexpected was their size - they were much bigger than any known past earthquakes on the same faults. This happened because of an unlikely “cascade” that broke through various fault bends and step-overs that normally would act as barriers stopping the rupture propagation. While conventional studies and responses usually focus on particular data and aspects of earthquakes, we unveiled a holistic synthesis of diverse and rich data sets and state-of-the-art earthquake interpretation tools, ends up with finding both earthquakes branched and jumped in unexpected ways, in which stress interactions and static and dynamic triggering worked together to result in a cascade of rupture with a larger than usual total rupture length and greater destruction potential. The unusually complex dynamics of the Turkey earthquakes call for reassessment of common assumptions built into seismic hazard assessments. Additionally, there are strong similarities between the fault system that ruptured in Turkey and many other places with similar fault architecture in the world, such as California. See our paper for more details.
In July 2019, a significant earthquake sequence struck the Ridgecrest region, breaking Southern California's seismic quiescence and highlighting the need to integrate knowledge of earthquake source processes and regional fault structures for hazard response. We found that the Mw 6.4 foreshock on July 4 consisted of three subevents on interlocked orthogonal faults, while the Mw 7.1 mainshock on July 6 ruptured along bifurcating faults toward the northwest and southeast, and halted at horsetail faulting structures. The foreshock ruptured toward the mainshock’s hypocenter but was stopped by a 4-km gap which was gradually filled by small earthquakes, suggesting that the foreshock triggered seismicity which eroded this barrier and induced the mainshock. By combining a rich dataset of ground shaking (strong motion data, broadband seismograms) and deformation (high rate GPS, InSAR) with realistic 3D structure considered, we found that the ruptures are substantially simpler at depth than near the surface, with four faults explaining all the data reasonably well. Interestingly, both the foreshock and the mainshock started at fault junctions and then ruptured multiple faults. Where the two events overlapped, their slip patterns are largely complementary to each other. See our Science and GRL papers for more details.
The 2021 August South Sandwich Island Mw 8.2 earthquake was a surprise, because it was initially reported as a magnitude 7.5 event at a deep depth (47 km) but generated a global-spreading tsunami that would only be expected for a larger and shallower event. By using seismic data with period as long as 500s, we revealed a hidden Mw 8.16 shallow slow event that happened between clusters of regular ruptures in the beginning and end. Although the slow event contributed 70% of the seismic moment, lasted three minutes, and ruptured a 200-km section of the plate interface, it is essentially invisible at short or intermediate periods, which explains its anomalously low body-wave and surface-wave magnitudes. The 2021 South Sandwich Island earthquake represents an extreme example of the broad spectral behaviors of subduction zone earthquakes and calls for attention in the research and warning of similar events. See our paper for more details.
Dynamics in the deep Earth
- Mega-annum-term subduction slab dynamics shape the local temperature and governs the deep earthquake rupture properties.
- Transition from metastable olivine transformational faulting to shear melting controls large deep earthquakes in subducting slabs.
The Fiji-Tonga subduction zone accounts more than 75% of global deep earthquakes, but lacked large ones (M~8). This puzzling deficit was changed by a pair of Fiji-Tonga M8 earthquakes in two weeks, 2018. We analyzed the focal mechanisms, rupture processes, aftershock seismicity and thermal properties, and found that the doublet events may have ruptured in two distinct slabs, even though they are only ~250 km from each other! The August M8.2 event occurred in the Tonga slab while the September M7.9 event ruptured in a warm detached Australian slab leaning on the Tonga slab. The 2018 Fiji M8 doublet reflects the complex interaction of slabs near the bottom of the mantle transition zone, and also emphasizes local slab temperature's primary control on deep earthquake ruptures and aftershocks. See our EPSL and GRL paper for more details.
Deep earthquakes at 500 to 700 km depth are subject to a physical environment that is prohibitive of typical brittle failures. Despite decades of observations, a coherent mechanism that can explain events of varying magnitudes from different subduction zones is yet to be identified. Our study systematically analyzes 40 M>7 deep earthquakes that occurred since 1990. We resolve their kinematic rupture processes and compare the source parameters with slab structures from improved thermal simulations. We find that most large deep earthquakes are likely controlled by two mechanisms: the earthquakes initiate from metastable olivine transformation faulting and rupture beyond the slab core due to shear melting. The final earthquake magnitude hinges upon the transition of the mechanisms, and the transition is critical in creating larger deep earthquakes (M>7.5). This transition causes greater moment release, increased geometric complexity, and a notable reduction in rupture length, which in turn produces greater stress drops. Consequently, stress drops of deep earthquakes scale with earthquake magnitude, in contrast to shallow crustal earthquake behaviors. Our proposed mechanism can coherently explain deep earthquakes from the coldest to the warmest slabs, such as the Tonga and South American subduction zones. Additionally, our source models provide unique new constraints on the stress and dynamics within subducting slabs. See our paper for more details.
Natural hazard mitigation
- Enhancing tsunami hazard mitigation: application on the complex 2024 Mw 7.5 Noto Peninsula earthquake.
- Improving ground motion assessments in the Los Angeles Basin: a high-resolution 3D velocity model from new dense seismic arrays
Large earthquakes can pose a significant challenge for understanding its tsunami generation due to their complex rupture behaviors, and the 2024 Mw 7.5 Noto Peninsula earthquake at Japan is a remarkable case. Our study offers valuable insights into tsunami hazard mitigation by analyzing this event using advanced models that incorporate a 6-subevent rupture sequence derived from Bayesian analysis of seismic data. We discover two distinct rupture phases: an initial onshore rupture to the southwest, followed by a delayed re-rupture at the original hypocenter, which caused substantial seafloor uplift to the northeast. This uplift was crucial in generating the tsunami. By constructing a multi-fault uplift model that matches the region's known fault system and validating it with geodetic data, we accurately simulated the tsunami’s wave height, timing, and direction: all key components for improving current tsunami early warning systems. Our analysis demonstrates the need to account for complex earthquake source behaviors in tsunami modeling. See our paper for more details.
The shallow velocity structure of the Los Angeles (LA) Basin plays an important role in the seismic hazard of this populated area. But most existing velocity models of the LA Basin have limited resolution due to the sparsity of seismic stations, or the restricted coverage of sonic logging and industry reflection profiles. We take a step forward by combining the aperture of broadband seismic stations and density of industrial nodal arrays (>16,000 stations), and derive a 3D shear wave velocity model, covering a large portion of the central LA Basin for the depths shallower than 3 km. We find that the small scale heterogeneities are better resolved compared with the conventional models. We also capture the presence of the Newport-Inglewood fault by a NW-SE trending high velocity belt. Our model could better predict the variations in the shallow crustal amplifications, which can improve strong ground motion simulations. See our paper for more details.
Technical development for sustainable resilience
- Subevent MCMC inversion: towards bridging point sources and finite slip models for geometrically complex earthquakes.
- Bayesian differential moment tensor inversion for more accurate focal mechanisms.
- Determination of small earthquake rupture directivities with energy envelope deconvolution
- Deep learning-based automated dispersion picking for big data seismic tomography.
Rapid and accurate characterization of the earthquake sources helps to mitigate seismic hazards and to understand the earthquake physics. In the past decades, seismologists developed routine moment tensor inversions and finite fault slip models, which well characterized the simplicity and complexity of most large earthquakes, respectively. However, point source moment tensor does not contain earthquake rupture finiteness, while finite fault slip inversions need prior information of rupture kinematics and fault geometry, thereby being challenged by some complex large earthquakes that ruptured multiple/unknown fault segments. To fill this gap between point sources and finite slip models, we propose a subevent inversion framework to resolve the rupture and geometrical complexity of large earthquakes simply yet flexibly. By parameterizing a complex event with a series of simple subevents, our method can reflect unknown ruptures on unknown faults without assuming a specific fault geometry or kinematic history. We also incorporate Bayesian analysis for uncertainty assessments. We validated our subevent inversions using a series of examples, including synthetic model test, a composite event from real earthquakes, and a complex large earthquake. These results illustrate that our subevent inversion can provide rapid and standardized constraints on the first order rupture complexity of large earthquakes.
Moment tensors are key to seismic discrimination but often require accurate assumption on the Earth structure for estimation. This limits the data availability and the precision of moment tensor inversions. To overcome this difficulty, we develop a differential moment tensor inversion (diffMT) method that uses relative measurements to remove the Earth's structural effects shared by clustered events, thereby improves the accuracy of source parameters. In a Bayesian framework, we invert the body- and surface-wave amplitude ratios of an event pair for refined moment tensors. Applications to three North Korea nuclear tests from 2013 to 2016 demonstrate that diffMT reduces the uncertainties substantially compared with the traditional waveform-based moment tensor inversion. Our results suggest high percentages of explosive component with similar double-couple components for the North Korea nuclear tests. See our paper for more details.
Earthquake faults often rupture in a single direction, which can be detected by measuring the Doppler shift in their seismic radiation, i.e., that seismic stations in the direction of rupture record shorter pulses than those observed at other stations. These directivity effects are easily seen for large earthquakes but are challenging to measure for small events because their apparent pulse widths are biased by scattering from small-scale crustal structure. Here we develop a new approach that uses seismogram envelope functions rather than the original seismic waves, while also removing distortions from the paths and site effects. We show that it provides more robust directivity results than more standard methods. Application to 69 aftershocks of the 2019 Ridgecrest earthquakes reveals a complex network of faulting behavior. Our method makes it possible to study rupture patterns in smaller earthquakes, which improves our understanding of their rupture dynamics and helps to better assess regional seismic hazards. See our paper for more details.
Dispersive surface waves have been important and extensively used for understanding the Earth's structure, since surface waves can be easily extracted from ambient noise correlations and does not depend on a suitable earthquake distribution. However, picking the dispersion curves can be very labor-intensive, particularly when dealing with dense arrays. Here, we develop a machine learning based algorithm to automatically extract surface wave dispersion curves. We first convert the standard frequency-time analysis of seismograms into images, then use a convolutional neural network to drive a supervised learning. Our results show that the machine classification is nearly identical to the human picked phases. See our paper for more details.