“ShakeAlert” Algorithms Improve Earthquake Warning Systems

By Lina Sorg

Earthquakes and their resulting tsunamis can often strike with little to no warning, causing serious damage and displacing entire communities. Thus, effective earthquake and tsunami warning systems must rapidly assess conditions and immediately communicate the hazard to communities surrounding the epicenter of seismic activity. Unfortunately, local warning systems in high-risk areas worldwide rarely give residents adequate time to prepare. As a result, early-warning algorithms that rely on geodetic data (coordinate systems and a set of location-based reference points) for quick evaluation of quake sources are becoming more common. In a minisymposium presentation during the recent SIAM Conference on the Mathematics of Planet Earth, Christine Ruhl (University of California, Berkeley) discussed the value of real-time GPS systems that monitor earthquake and tsunami triggers to immediately predict resultant damage. 

75 million Americans in 39 states live in areas prone to significant seismic activity, and frequent movement along the San Andreas Fault makes California particularly susceptible to earthquakes. Researchers in the Berkeley Seismological Lab are currently working on an earthquake early warning (EEW) project exploring how GPS technology can improve warning times. Real-time GPS systems can monitor crustal deformation and strong shaking during seismic activity, and ocean-based sensor networks serve a similar purpose. “GPS is a fundamental tool for warnings and hazards,” Ruhl said.

Ruhl’s lab is involved in the development and testing of ShakeAlert, a GPS-based EEW system for the West Coast of the United States. ShakeAlert identifies and characterizes earthquakes mere seconds after seismic activity begins at the epicenter. The system’s algorithms calculate the projected intensity of shaking and send warnings to affected people and infrastructure; these warnings may range from seconds to even tens of minutes before shaking begins, depending on distance from the epicenter. 

Because large seismic events are relatively rare, simulations are necessary to routinely test the performance of ShakeAlert’s algorithms. Consequently, researchers created the so-called “Fakequake” method as a means of producing the necessary fake earthquakes. The program generates kinematic rupture models and low-frequency synthetic displacements that yield accurate, usable data. It employs ground motion prediction equations (GMPEs), which use factors like magnitude, soil condition, fault mechanism, and source-to-site distance to predict the level of ground shaking and resulting uncertainty at a designated location. The Fakequake system operates via an open-source Python framework, and all of its components are tunable parameters (both static and kinematic). A Karhunen-Loéve expansion creates static slip on arbitrarily complex faults; the different slip patterns become increasingly complex as they move to higher and more forceful movements, and the eigenvectors are actually slip patterns themselves.

The next step for Ruhl and her team is the comparison of Fakequakes to peak ground scaling laws, which help define seismic sources. If the goodness of fit’s coefficient is below 0.7, then their algorithms are behaving successfully. Developers are currently testing ShakeAlert throughout California and using the Fakequake simulations to construct a large catalog of events for future reference.

Ruhl asserted that ShakeAlert can positively impact how high-risk populations respond to earthquakes and tsunamis. In some cases, its advanced warning could even offer sufficient time to slow or halt trains and taxiing planes, maneuver traffic patterns away from bridges and tunnels, and isolate people from machines or dangerous chemicals in work environments. While earthquakes will always be hazardous, ShakeAlert’s implementation could make their outcomes far less devastating. 

Lina Sorg is the associate editor of SIAM News.