The Virtual Seismologist in SeisComP3: A New Implementation Strategy for Earthquake Early Warning Algorithms

Abstract

The feasibility of earthquake early warning (EEW) is now widely recognized. However, EEW systems that are in operation or under evaluation worldwide have significant variations and are usually operated independently of routine earthquake monitoring. We introduce a software that allows testing and evaluation of a well‐known EEW algorithm directly within a widely used earthquake monitoring software platform. In the long term, we envision this approach can lead to (1) an easier transition from prototype to production type EEW implementations, (2) a natural and seamless evolution from very fast EEW source parameter estimates with typically large uncertainties to more delayed but more precise estimates using more traditional analysis methods, and (3) the capability of seismic networks to evaluate the readiness of their network for EEW, and to implement EEW, without having to invest in and maintain separate, independent software systems. Using the Virtual Seismologist (VS), a popular EEW algorithm that has been tested in real time in California since 2008, we demonstrate how our approach can be realized within the widely used monitoring platform SeisComP3. Because this software suite is already in production at many seismic networks worldwide, we have been able to test the new VS implementation across a wide variety of tectonic settings and network infrastructures. Using mainly real‐time performance, we analyze over 3200 events with magnitudes between 2.0 and 6.8 and show that, for shallow crustal seismicity, 68% of the first VS magnitude estimates are within ±0.5 magnitude units of the final reported magnitude. We further demonstrate the very significant effect of data communication strategies on final alert times. Using a Monte Carlo simulation approach, we then model the best possible alert times for optimally configured EEW systems and show that, for events within the dense parts of each of the seven test networks, effective warnings could be issued for magnitudes as small as M 5.0.