The seismic alert system (sas) for Mexico City has now been in operation for about 15 years. The sas takes advantage of the fact that the city is located more than 300 km from the foci of many of the potentially damaging earthquakes. The system consists of 15 accelerometers located along the coast of the State of Guerrero, above a segment of subduction plate boundary that is a mature seismic gap. An algorithm estimates the magnitudes of earthquakes from the near-source accelerograms and issues public and restricted alerts for earthquakes with M ≥6 and 5 ≤ M < 6, respectively. An evaluation of the sas’s performance during 1991–2004 reveals a surprisingly high rate of failure and false alerts. This poor performance results from an inadequate detection algorithm and a limited areal coverage by the sas. This renders the alert system of limited use.
In this article we propose an alternative strategy for detecting earthquakes potentially damaging to Mexico City that differs substantially from the one presently implemented by the sas. Although our analysis is based on close-to-source accelerograms of 45 Mexican earthquakes and the corresponding peak accelerations recorded at a reference site in Mexico City (cu), there is no restriction on the distance to the field station, except that its location should provide sufficient alert time to Mexico City. Based on these data, an attenuation relation is derived to compute expected peak acceleration at cu (Ared) from root-mean-square acceleration (Arms) at a field station. The relation permits specification of an Ared threshold, given the peak acceleration at cu (Acu) for which an alert is desired along with prescribed probabilities of failure and false alert. We find that the use of bandpass-filtered (0.2–1.0 Hz) accelerograms leads to an improved performance of the sas. The choice of the filter is guided by the frequency band of amplification of seismic waves in the lake-bed zone of Mexico City. We think that a single level of general public alert may be the best option. A good choice appears to be an alert for Acu ≥ 2 gal (for 0.2–1.0 Hz bandpass-filtered accelerograms) with 1% probability of failure. To accomplish this we must set Ared ≥ 0.8 gal. The data since 1985 suggest that such an alert would occur about once or twice a year and the event will be felt by most persons in the lake-bed zone. The proposed algorithm, along with an array of sensors located 30 to 40 km apart and distributed in a roughly semicircular arc of 310-km radius centered at Mexico City, should considerably improve the areal coverage and performance of the sas and potentially save thousands of lives.