Recently they had a significant country wide false alarm in Israel at 3AM... There was a emergency alert cell broadcast (similar to amber alert), which caused everyone to move their phone at the same time, which was falsely detected as an earthquake, which caused an Android earthquake alert to be sent to all phones in Israel 30 seconds later. I guess they didn't plan for this scenario
Edit: Arstechina article seems to mention this: "only three were false positives. One of those was triggered by a different system sending an alert that vibrated a lot of phones"
That would be quite an implementation flaw if it didn't account for the phone's own vibrations. Lots of countries use widespread emergency alert messages frequently.
It's obvious to anyone that's written software for hardware sensors to ensure the hardware itself is not going to interfere with readings of other sensors. Don't tell the motor to run if the motor is already running type of issues. It's a fairly common bozo check
Typical Google product. Reminds me of a person who put a bunch of phones in a car and drove which caused Google maps to wrongly show traffic in that area. It was deliberately done though as an experiment
Note that those are three completely false events. The survey results Google published show 15% of people not feeling any shaking (neither strong nor light). That's still a good figure, but reading there were only 3 false positives gave me the impression that you're basically always in for a ride when you get the alert and it's not that miraculously accurate either
Surprised they don't do some signal processing on all the IMU signals to see if they correlate (to within a rotation matrix), and if the timing of shaking at different locations is consistent with the distance to some (solvable) epicenter.
The whole country moving phones in random directions at exactly the same time isn't what an earthquake signature should look like.
I don't get it. I thought earthquake alerts were meant to trigger _before_ the earthquake arrives. If it happens 30 secondes after detecting vibrations, not considering the false positive, it can only mean "hey, you just felt, or are feeling an earthquake, hope you're sheltered".
This is how I experienced the Google Earth Alerts in Wellington, NZ which experiences frequent earthquakes. Earthquake happens. 30 seconds to five minutes later I get an alert from Google.
The best earthquake indicator is an old house with wooden windows that you'll hear rattling five seconds before you get to feel it.
A faster moving small quake (p-wave) will precede the bigger, more damaging quake. This system detects the p-wave and alert people hoping they can get out in time before the big quake hit.
This is when the earthquake is a few hundred kilometers away, such as in Mexico City, where most earthquakes occur off the coast and the waves take a few minutes to reach the city.
Somewhat relatedly, I support a service with global scale traffic. Whenever there's an earthquake in APAC, we get a traffic spike, like 100x normal for that time of night. I'm pretty sure it's incoming alerts waking people up / checking where the epicenter is and if they need to run from any tsunami or flooding, but it's still really hard to scale that kind of thing up for regional demand when it spikes that hard!
Edit: Arstechina article seems to mention this: "only three were false positives. One of those was triggered by a different system sending an alert that vibrated a lot of phones"