From ShakeAlert to smartphones: Earthquake early warning now and beyond

More than a decade ago, a new Ph.D. student arrived at the University of California, Berkeley to work on earthquakes with Berkeley Seismological Laboratory Director Richard Allen.

“At that time,” says Allen, “a colleague in engineering … had been experimenting with using the accelerometer in phones for various things.” Accelerometers are the hardware that determines a smartphone’s orientation so the screen can rotate as the user moves. As earthquake scientists, Allen and the new student, Qinghai Kong, quickly set their sights on how seismology could exploit these accelerometers. Kong began the project by simply putting phones on shake tables that simulated earthquakes to see what could be measured, says Allen. They presented the seeds of the idea at the annual CITRIS meeting on the Berkeley campus. That idea would blossom into MyShake, an earthquake early warning smartphone app, in the beginning of 2012.

Fast forward to October 2015: Marc Stogaitis, a principal engineer at Google, and his team were engaging in a hack-a-thon in Google’s conference rooms in the heart of Silicon Valley. Stogaitis’ team focused on activity recognition, where they took data from phone sensors like the accelerometer to figure out something about the real world, he recalls. His team asked whether phones could be used to detect car crashes (yes), tornadoes (no), or earthquakes (keep reading). Since that fateful week, he and his team have led Google’s charge, along with Allen and Kong who are Visiting Faculty at Google, in the quest to use phones to help people during earthquakes.

Among a number of collaborators, Kong, now a research scientist at Lawrence Livermore National Laboratory, Allen, Stogaitis and Sarina Patel, a doctoral student in the Berkeley Seismological Laboratory, have been working to get alerts to smartphones. In the United States, they’ve collaborated with the U.S. Geological Survey (USGS) to help disseminate earthquake alerts produced by ShakeAlert, the West Coast’s earthquake early warning system. And the reach of smartphone-based earthquake early warning is increasingly global.

Finding an earthquake

The concept of earthquake early warning is to notify people that they are going to experience shaking from an earthquake before shaking actually reaches them, Patel says. This is possible because earthquakes begin underground, radiating seismic waves away from the initial rupture. In the western U.S., ShakeAlert relies on a dense network of extremely sensitive seismic stations in earthquake-prone areas that constantly monitor the ground for motion. When one of those stations “feels” an earthquake by sensing the first, less damaging P-waves, it sends data to a central computer that asks two basic questions, Patel explains. First, is this really an earthquake signal? Second, have at least three other stations also detected the event?

“All of this happens in the space of a second or two,” says Patel. “The computations are extremely rapid.” Once the earthquake is “created” from the requisite four sensors, algorithms determine the magnitude and location. With this information, a computer issues a forecast about who will feel the earthquake. The computation uses a mathematical relationship between seismic waves and the subsurface through which the waves must travel, she says. The faster the earthquake detection and these calculations, the sooner the alert can go out, hopefully before the more damaging S-waves arrive, and to as many people as possible.  

Using smartphones to deliver alerts

“When an earthquake is detected, there’s an entire mechanism to send out alerts quickly and reliably,” Stogaitis says — and in a way in which users quickly understand what’s going on.

The USGS requires that ShakeAlerts be distributed to smartphones located where shaking is expected to be at least MMI (Modified Mercalli Intensity) III, which corresponds to weak shaking. Also, only earthquakes estimated to be at least magnitude-4.5 trigger alerts.

To distribute these alerts to people, ShakeAlert relies on several systems. For instance, a slew of apps can issue alerts to smartphones, including the now-retired ShakeAlertLA app that was specific to Los Angeles, and the MyShake app that was created by Kong, Allen and the Berkeley Seismological Laboratory team. MyShake, which is supported by the California Governor’s Office of Emergency Services “Earthquake warning California” effort, now delivers ShakeAlert messages to California, Oregon and Washington.

Google also pushes alerts to Android phones but without the use of an app, as an “Earthquake Alert” is built directly into the Android phone’s operating system. Once ShakeAlert has detected an earthquake, a direct connection between the ShakeAlert server and Google allows Android phones to receive the alert with minimal delay, Stogaitis says. (Android users could turn this feature off, but why would you?)

Less high-tech delivery systems include public broadcasting systems, like the emergency broadcasting messages that interrupt television shows. FEMA’s alerting system — the Wireless Emergency Alert system that also disseminates Amber Alerts — can send notifications to any WEA-enabled phone, whether the phone is smart or simply cellular.

ShakeAlert divides the region that might need an alert into eight-sided polygons, or octagons. Everyone situated within each octagon either needs an alert, or they don’t. ShakeAlert sends these coordinates, along with alert itself, to these various systems that deliver the message. The goal is to make the logic for who needs an alert as simple as possible, says Patel.

But for privacy reasons, “we don’t know exactly where you are,” Patel says, so MyShake sorts phones running the app into squares, or boxes. If any part of the ShakeAlert octagon overlaps with a MyShake box, all phones within that square receive an alert.  

Speedy delivery

The MyShake app uses Firebase Cloud Messaging, which is Google’s free cloud service that lets app developers send notifications and messages to users across different platforms, including Android and iOS. When sending out a message simultaneously to the thousands of phones whose users downloaded the app, most cellular networks can handle the load thanks to the number of cell towers over which the alert can be distributed.

But because Android alerts go to all Android users in the shaking region — sometimes numbering several million notifications — Stogaitis says his team had to build a custom, low-latency messaging system to send alerts as quickly as possible. Alerting millions of people at the same time “is an incredibly challenging problem,” Stogaitis says. “A lot of engineering went into [not overloading cell networks].” Of course, if a device is on WiFi, these considerations matter less because the message doesn’t travel through any part of the cellular network.

Patel and Allen reported in a recent paper in Seismological Research Letters that because MyShake is both an official alert provider and a research enterprise, the research team has built in ways to track and find the delays, known as latency, in the system — from receipt of a ShakeAlert message by the MyShake server to delivery to each smartphone. After all, when you’re talking about an alert that gives seconds of warning, an extra second of delay matters. For instance, the Sept. 19, 2020, magnitude-4.5 El Monte earthquake that struck the Los Angeles area at a depth of 17 kilometers resulted in 20,169 alerts sent to smartphones running MyShake. In total, the time between the MyShake servers receiving the ShakeAlert signal and the delivery to 80% of the phones was 3.5 seconds.

The El Monte event, says Patel, was tricky because the earthquake was relatively small but occurred in a populous region, which meant that the alerting area was small, but had a large population that needed to be alerted. About half of alerted users received their notification before the estimated arrival of the S-wave, which is when people begin to feel shaking. The S-wave, which moves slower than the P-wave that ShakeAlert uses for its calculations, causes much more motion on Earth’s surface, and therefore is most likely to cause damage.

Google also looked at its delivery times for the El Monte earthquake. Stogaitis notes that the system messaged more than 2.2 million people, with 80% receiving the alert in less than 5 seconds.  

Style and substance

One advantage to building alerts into the operating system in the way that Google has done for Android phones, says Stogaitis, is that “we can build a full, rich alert of Drop, Cover, and Hold On that was designed very carefully.” People don’t want — or have time — to read a “wall of text” during an earthquake, he says.

For the Android Earthquake Alerts system, the first alert is called “take action.” The alert is loud, and it takes up the entire screen to get people’s attention, and it’s simple so that recipients will know exactly what to do. MyShake also provides a similar alert that will go through your “do not disturb” mode to warn you.

As it turns out, even with mild shaking, people still want alerts, which led Stogaitis’ team to come up with a second “Be Aware” alert that warns of light rattles (expected MMI of between III and V), instead of strong shaking. “It won’t go through your ‘do not disturb’ mode,” he says, and it won’t wake people up in the middle of the night who might sleep through subtle movement.

Seismologists are now partnering with social scientists to figure out exactly what the public wants and will find useful, say Allen and Stogaitis in a recent publication in Science.  

Turning smartphones into seismometers

Built into both Android’s operating system and the MyShake app is the ability for a phone to be used as a seismometer. Google has licensed the MyShake smartphone earthquake early warning patent held by the University of California, Berkeley, which was invented by Allen and Kong. With this technology, any smartphone with the MyShake app, and any Android phone with the Android Earthquake Alert system turned on, can sense earthquakes and collect data.

The process of waiting for an earthquake’s waves to reach the surface and be detected by a seismic station is also the largest delay in the earthquake alerting process, Stogaitis says. This means that if earthquakes can be detected faster, alerts can go out more quickly.

When a phone is not in use and stationary, the sensors that detect your phone’s rotation to keep your screen properly oriented can function as a vibration-detector and listen for earthquake-like motions, says Patel. When a smartphone detects a possible earthquake, the phone sends the trigger to a central server in real-time, much like the ShakeAlert system, she explains.

“Seismic networks are optimized for data transmission,” says Patel. They’re deliberately designed with speed in mind. Their internet connections are wired to eliminate any delay in the transmission of data to ShakeAlert’s computers, limiting any potential data bottlenecks. So, using phones as seismic stations raises the question of how fast wireless or cell networks can convey the first rumblings of an earthquake. A phone is not optimized to transmit information as fast as possible, she says, but there is strength in numbers. Because so many phones transmit the initial inkling of an earthquake, MyShake could detect the earthquake nearly as fast, or potentially even faster, than ShakeAlert, she says. “We do get enough data in a similar timeframe [compared to ShakeAlert].”

The number of triggers coming from MyShake for the El Monte event — meaning the number of smartphones that detected the earthquake — was comparable to the number of ShakeAlert triggers, Patel reported in a presentation at the fall meeting of the American Geophysical Union in 2020. In the earliest seconds, 13 MyShake app triggers arrived before any ShakeAlert triggers made it to a server. This is important for the future, Patel said in the talk, “for thinking about how we can combine these two systems.”

“Currently, MyShake is detecting triggers and feeding them into a backend server,” Patel says. “If we piped those triggers into the ShakeAlert system in real-time, could we make the algorithm produce results faster?”

She points out that one of the barriers to speed is the requirement of four stations to confirm an earthquake is actually occurring. In other words, detecting the earthquake can be a significant delay in the system. “But we have a lot of phones where these seismometers are located, or closer to the source [of the earthquake],” she says, “so you could be shaving seconds off the time it takes to create an alert.” Even saving partial seconds is significant if, end-to-end, it takes between 5 and 10 seconds between an earthquake starting underground to alerts arriving on the majority of phones, she explains.  

People and phones

“Earthquake early warning is really part of the technological revolution that allows us now to communicate and compute information faster than the earthquake waves themselves travel,” says Patel. Earthquake waves travel at approximately the speed of sound, whereas computers can make calculations in milliseconds and data can travel at approximately the speed of light.

Getting that information fast can help people protect themselves with drop, cover and hold on, said Stogaitis. But more than that, it can also help automatically slow trains to avoid derailments. Gas valves can be automatically shut off to prevent fires. Airplane landings can be stopped and directed elsewhere.

Moreover, by turning phones into seismic sensors, earthquake early warning has been expanded dramatically in the past year or so. In 2021, Android Earthquake Alerts began delivering alerts in New Zealand, Greece, Turkey, the Philippines, and central Asia, which has enabled another 150 million people to access potentially lifesaving earthquake early warning alerts. As of publication, more than 90 countries are included in the Android Earthquake Alerts system. According to Allen and Stogaitis, Google’s goal, which is to make Android Earthquake Alerts available globally, “is possible because there are now, for the most part, phones wherever there are people.”