The τc-Pd Onsite algorithm is one of the methods included in the current ShakeAlert prototype system. More information on the basics of the method can be found Here.
Running an algorithm continuously in real-time provides a wealth of new challenges and observations compared with the controlled off-line testing usually carried as part of initial algorithm research. During the years that Onsite has been running as part of the ShakeAlert system, methods and parameters have been monitored and updated to improve the real-time performance.
Some examples of observed unexpected behavior and changes include:
- Split events:
- A single earthquake leads to multiple EEW alerts. While the first is often reasonably accurate, the later alerts are usually grossly incorrect in terms of location and sometimes magnitude. This behavior arises when P-wave arrivals at more distant stations, or incorrectly included S-wave arrivals at nearer stations, are treated as new events.
- Onsite’s association criteria have been updated to try to improve the likelihood that phases are correctly associated with an event. A short note with examples and changes made is available Here.
- False alerts from environmental noise:
- The aim of Onsite is still to work with single station alerts. However, due to the high likelihood of false alerts from noisy stations, it currently requires two stations to trigger before a ShakeAlert message is issued. There have still been a few occasions when environmental noise on nearby stations has coincided and a false alert issued; there were 7 such cases between 1 Jan 2015 and 13 March 2016. In all instances, Onsite went on to cancel the event alert, usually within a few seconds.
- It was observed that most of these environmental noise triggers were caused by signals at strong motion stations; these data were only introduced to full Onsite processing at the start of 2015. Analysis of noise information collected by the ShakeAlert system led to changes to trigger thresholds used for different sensor types.
- False alerts from regional or teleseismic events:
- Onsite (like the other ShakeAlert algorithms) contains various waveform checks to try to discriminate between local earthquake signals and those from regional or teleseismic events. Occasionally these discrimination techniques are unsuccessful, particularly for larger magnitude events just beyond the network borders, leading to one or multiple false alerts; for example an event in south-eastern Nevada during 2015 led to 3 false alerts by Onsite. Seeing several event alerts within a short time period (seconds of each other) but geographically widely distributed might (but not necessarily!) be an indication of this kind of false alert.
- Overcoming this kind of false alert is a particular challenge, as the ‘noise’ source is a real earthquake. It is an ongoing research topic to investigate which waveform measurements can be used to differentiate these signals.
Ongoing Research and Development
Various parts of the τc-Pd Onsite algorithm are the subject of ongoing research and development. Below is a non-exhaustive list:
- Location method: the existing method of determining event epicenter from a limited number of stations performs poorly for areas that are out-of-network or have sparse station coverage. Alternative methods are being investigated.
- Phase discrimination: the current Onsite method is designed to work with P-wave arrival data; S-wave arrivals incorrectly processed can lead to erroneous event parameters or to split/false alerts. Onsite currently uses two different methods to assess the likelihood that a signal is a P-wave, but additional discrimination methods are also being investigated.
- Magnitude determination: increasing the accuracy of the earliest Onsite alerts, and of alerts for events near the detection threshold, is another area of active research.