Modeling protective action decision-making in earthquakes by using explainable machine learning and video data.

Earthquakes pose substantial threats to communities worldwide. Understanding how people respond to the fast-changing environment during earthquakes is crucial for reducing risks and saving lives. This study aims to study people's protective action decision-making in earthquakes by leveraging explainable machine learning and video data. Specifically, this study first collected real-world CCTV footage and video postings from social media platforms, and then identified and annotated changes in the environment and people's behavioral responses during the M7.1 2018 Anchorage earthquake. By using the fully annotated video data, we applied XGBoost, a widely-used machine learning method, to model and forecast people's protective actions (e.g., drop and cover, hold on, and evacuate) during the earthquake. Then, explainable machine learning techniques were used to reveal the complex, nonlinear relationships between different factors and people's choices of protective actions. Modeling results confirm that social and environmental cues played critical roles in affecting the probability of different protective actions. Certain factors, such as the earthquake shaking intensity and number of people shown in the environment, displayed evident nonlinear relationships with the probability of choosing to evacuate. These findings can help emergency managers and policymakers design more effective protective action recommendations during earthquakes.

A unified perspective of seismicity and fault coupling along the San Andreas Fault.

The San Andreas Fault (SAF) showcases the breadth of possible earthquake sizes and occurrence behavior; in particular, the central SAF is a microcosm of such diversity. This section also exhibits the spectrum of fault coupling from locked to creeping. Here, we show that the observations of aseismic slip, temporal clustering of seismicity, and spatial variations in earthquake size distributions are tightly connected. Specifically, the creep rate along the central SAF is shown to be directly proportional to the fraction of nonclustered earthquakes for the period 1984-2020. This relationship provides a unified perspective of earthquake phenomenology along the SAF, where lower coupling manifests in weaker temporal clustering, with repeating earthquakes as an end-member. This new paradigm provides additional justification for characterizing the northwest ∼75 kilometers of the creeping segment as a transition zone, with potential implications for seismic hazard.

3D fault architecture controls the dynamism of earthquake swarms.

The vibrant evolutionary patterns made by earthquake swarms are incompatible with standard, effectively two-dimensional (2D) models for general fault architecture. We leverage advances in earthquake monitoring with a deep-learning algorithm to image a fault zone hosting a 4-year-long swarm in southern California. We infer that fluids are naturally injected into the fault zone from below and diffuse through strike-parallel channels while triggering earthquakes. A permeability barrier initially limits up-dip swarm migration but ultimately is circumvented. This enables fluid migration within a shallower section of the fault with fundamentally different mechanical properties. Our observations provide high-resolution constraints on the processes by which swarms initiate, grow, and arrest. These findings illustrate how swarm evolution is strongly controlled by 3D variations in fault architecture.

The Limits of Earthquake Early Warning Accuracy and Best Alerting Strategy.

We explore how accurate earthquake early warning (EEW) can be, given our limited ability to forecast expected shaking even if the earthquake source is known. Because of the strong variability of ground motion metrics, such as peak ground acceleration (PGA) and peak ground velocity (PGV), we find that correct alerts (i.e., alerts that accurately estimate the ground motion will be above a predetermined damage threshold) are not expected to be the most common EEW outcome even when the earthquake magnitude and location are accurately determined. Infrequently, ground motion variability results in a user receiving a false alert because the ground motion turned out to be significantly smaller than the system expected. More commonly, users will experience missed alerts when the system does not issue an alert but the user experiences potentially damaging shaking. Despite these inherit limitations, EEW can significantly mitigate earthquake losses for false-alert-tolerant users who choose to receive alerts for expected ground motions much smaller than the level that could cause damage. Although this results in many false alerts (unnecessary alerts for earthquakes that do not produce damaging ground shaking), it minimizes the number of missed alerts and produces overall optimal performance.

To catch a quake.

A revolution in seismic detection technology is underway, capturing unprecedented observations of earthquakes and their impacts. These sensor innovations provide real-time ground shaking observations that could improve emergency response following damaging earthquakes and may advance our understanding of the physics of earthquake ruptures.

The limits of earthquake early warning: Timeliness of ground motion estimates.

The basic physics of earthquakes is such that strong ground motion cannot be expected from an earthquake unless the earthquake itself is very close or has grown to be very large. We use simple seismological relationships to calculate the minimum time that must elapse before such ground motion can be expected at a distance from the earthquake, assuming that the earthquake magnitude is not predictable. Earthquake early warning (EEW) systems are in operation or development for many regions around the world, with the goal of providing enough warning of incoming ground shaking to allow people and automated systems to take protective actions to mitigate losses. However, the question of how much warning time is physically possible for specified levels of ground motion has not been addressed. We consider a zero-latency EEW system to determine possible warning times a user could receive in an ideal case. In this case, the only limitation on warning time is the time required for the earthquake to evolve and the time for strong ground motion to arrive at a user's location. We find that users who wish to be alerted at lower ground motion thresholds will receive more robust warnings with longer average warning times than users who receive warnings for higher ground motion thresholds. EEW systems have the greatest potential benefit for users willing to take action at relatively low ground motion thresholds, whereas users who set relatively high thresholds for taking action are less likely to receive timely and actionable information.

Earth tides can trigger shallow thrust fault earthquakes.

We show a correlation between the occurrence of shallow thrust earthquakes and the occurrence of the strongest tides. The rate of earthquakes varies from the background rate by a factor of 3 with the tidal stress. The highest correlation is found when we assume a coefficient of friction of mu = 0.4 for the crust, although we see good correlation for mu between 0.2 and 0.6. Our results quantify the effect of applied stress on earthquake triggering, a key factor in understanding earthquake nucleation and cascades whereby one earthquake triggers others.