Rooted in academia
- Eisermann, A.S., A. Ziv, and G.H. Wust-Bloch, 2018. Array-based earthquake location for regional earthquake early warning: Case studies from the Dead Sea Transform, Bull. Seism. Soc. Am.
- Lior, I., and A. Ziv, 2018. The relation between ground motion, earthquake source parameters and attenuation: Implications for source parameter inversion and ground motion prediction equations, J. Geophys. Res.
- Eisermann, A.S., A. Ziv, and G.H. Wust-Bloch, 2015. Real-time back azimuth for earthquake early warning, Bull. Seismol. Soc. Am.
- Netanel, M., Eisermann, A.S. and Ziv, A., 2021. Off‐Network Earthquake Location by Earthquake Early Warning Systems: Methodology and Validation.Bull. Seismol. Soc. Am.
Our pilot early warning systems have been installed in Canada, Turkey, Israel and India, and another system is being deployed in the Philippines. We have detected and located in real time dozens of events using only one or two stations.
- M5.3 in Turkey – epicenter 100km offshore using only two stations
- M4.7 in Istanbul – offshore using a single station in a noisy environment
- M3.4 in Canada – located in under 3 seconds of origin time using a single station
- M2.9 in Canada – location off-network using two stations
- M2.8 in Canada – location off-network using two stations
The “AGU talk – Real-time results from array-based earthquake early warning pilots in Canada, Turkey and Israel”
The core challenge of EEW systems is to generate reliable predictions based on only a few seconds of waveform data. Seismic arrays are systems of linked sensors spread over hundreds of meters that can extract more information from incoming waveforms compared to a single-sensor station. The additional information provided by array-based measurements, such as slowness and back azimuth, can lead to better ground motion predictions as well as reduce the number of stations needed for a reliable location. We use seismic arrays alongside traditional single-sensor stations to provide an optimal solution.
In many places around the world, large magnitude earthquakes can happen outside the seismic network, such as offshore or in neighboring countries. Thanks to the back azimuths calculated by the arrays in under 300ms from the P-wave arrivals, our system can reliably locate events outside the seismic network using only two stations..
S-Wave Detection and Utilization
Most earthquake early warning systems disregard the arrival of the S-wave, since it is difficult to detect. Our system leverages array seismology, namely seismic slowness, to accurately identify the S-wave. This enables us to include S-wave picks and not-yet-arrived S-wave data in our location algorithm. S-wave picks strongly constrain the hypocentral distance and can be extremely beneficial if the earthquake is close to one of the arrays. Not-yet-arrived S-wave data is used to constrain the minimal hypocentral distance for each station (i.e. the longer it takes for the S-wave to arrive after the P-wave, the further away we know the event is occurring). This minimal distance is fed to the regional system, which updates the location estimate every 0.2 seconds.
On-site Non Seismic Signals Filtering
We continuously calculate the horizontal (or apparent) seismic slowness of incoming waveforms, as well as the signal coherency and amplitude decay across the array. These measurements enable us to identify non-earthquake signals and avoid false picks. This is done locally at the array site using an on-site processing unit that sends only necessary information after a P-wave has been detected, yielding efficient low latency network performance.
The horizontal seismic slowness measured during the P-wave arrival can be used to backward raytrace along the ray path to estimate the event depth. This can be done in the distance domain using the epicentral distance calculated from crossing two P-wave back azimuths, or in the time domain if one P-wave pick or one S-wave pick is available, depending on which is available first. To avoid real-time raytracing, we pre-calculate the results for the entire possible range of inputs, allowing us to determine depth in real time with zero computational cost.