Reducing False Positives in the On-Site and Regional Earthquake Early Warning Systems

Date: 4/25/2019

Time: 06:00 PM

Room: Grand Ballroom

The on-site earthquake early warning (EEW) systems are an efficient means to reduce seismic risk to the critical infrastructure. In addition, a collection of the on-site nodes can form a backbone of the sparse regional EEW network. Oftentimes, the sensors are located in noisy urban areas, which increases the probability of a false positive. We developed a method to reduce this probability by extracting from the real-time records of the ground motion a set of features that are likely to describe the P wave of a large seismic event, and analyzing them in real time. Besides, rather than performing a point measurement, we create an on-site multi-sensor array which allows to compute statistics of the extracted features, which further reduces the false positives. We use recursive Bayesian estimation to compute the parameters of the seismic event from the on-site data and apply the same approach to the sparse seismic network. Since the on-site EEW system operates on its own LAN, there are virtually no delays in the data transmission over the local network. Remote nodes in a sparse seismic network communicate only the updates of the event parameters for the recursive Bayesian estimation, which improves the speed of the issued alarm.

Our results demonstrated that this algorithm works efficiently for the set of the on-site EEW systems installed during the last 10 years in south-western British Columbia, producing virtually no false positives. Some of the sensors are installed near the major highways, busy city streets, and railroads. The most recent effort to improve the reliability of the systems consists of redundant solutions, where the entire data acquisition path, from sensors to the on-site processing embedded computer, is parallelized, and even consists of the heterogeneous elements with different failure modes. This is particularly important for the mission-critical facilities with the high cost of downtime.

Presenting Author: Anton G. Zaicenco

Authors