ASTUTI-Costa Rica: Fixed Network Smartphone-based Earthquake Early Warning

Abstract:

We test the hypothesis that off-the-shelf smartphones deployed in a fixed network can provide earthquake early warning (EEW) to a population, if adequately informed and sufficiently tolerant of EEW timeliness and accuracy limitations. We assess if alerts can arrive in time to permit protective low-cost actions such as drop-cover-hold on (DCHO). In January 2020 we constructed a network of 82 smartphones distributed throughout Costa Rica with ~ 30 km average spacing. Using on-board accelerometers, we implement a non-parametric detection and alerting algorithm that considers Middle America Trench (MAT) subduction zone earthquakes as the dominant source of shaking hazard and considers that much of the country’s population resides in the interior near the capital, San Jose. Rather than waiting until an earthquake reaches a large threshold size or intensity, we issue an alert when a low acceleration threshold is exceeded for three stations. From more than six months of observations and a beta testing group of 15 people we find data latency (phones to central hub) is ~ 0.4 secs. We simulate the 2012 M7.6 Nicoya earthquake using on-phone vibrations and find median first-alert latency of ~9-13 secs; alert-receipt latency using Amazon Web Services Simple Notification Services is ~4 secs, though we expect more latency when geographically dispersed recipients are considered. P-wave arrival for the event in San Jose was 26-27 secs. This suggests we can expect ~10-13 seconds pre-P warning times for the San Jose population for this scenario event that produced MMI V shaking levels. There have been eight earthquakes of greater than M5.0 near our network. Of these, two events were entirely within the network while the other six occurred outside of it. We detected all eight events and had fewer than ten false alerts depending on detection algorithm. False alerts are limited to a small number of regionally noisy stations. For three of the events we would have provided ~10-15 seconds pre-P warning in San Jose. Three events occurred too far offshore and two events within the late-alert zone, precluding pre-P alert receipt in San Jose. Our results begin to allow operational protective expectations to be set. Provided that users can be sufficiently informed about network limitations, DCHO for San Jose for MAT events may be a reasonable objective.