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

by Yannik Behr, John F. Clinton, Carlo Cauzzi, Egill Hauksson, Kristín Jónsdóttir, Craiu G. Marius, Ali Pinar, Jerome Salichon, and Efthimios Sokos

This supplementary material shows a comparison of the ground-motion prediction equation (GMPE) that is part of the Virtual Seismologist (VS) likelihood function, with two other GMPEs representative of the regions discussed in this study.

The predictive equations for P- and S-wave amplitude in the VS likelihood function are mostly based on (and therefore representative of) the seismicity of southern California, although Figure 2 in the main article shows M_VS, the first magnitude estimate provided by the VS implementation of SeisComP3, is suitable also for other test regions in Europe and worldwide, with the exception of deep seismicity and the volcanic earthquakes in Iceland. Possible explanations of this behavior are shown in Figure S1, in which the magnitude scaling (top panel) and distance dependence (bottom panel) of the horizontal peak ground motions predicted by VS are compared with the models of Akkar et al. (2014) and Bradley (2013), representative of Pan-European (Europe and Middle East [EUME]) and New Zealand active shallow crustal seismicity, respectively. Apparent from Figure S1, the median magnitude scaling of Akkar et al. and Bradley always falls within the 1σ bound of the VS prediction, the Akkar et al. model being closer to the median VS model. That is, the models shown in the top panel of Figure S1 are statistically equivalent with 68% confidence.

Similar conclusions are valid for the distance dependence at high (M_w 7) magnitudes, for which the predictive tool by Akkar et al. (2014) shows a remarkable agreement with VS for Joyner–Boore distances (R_JB) smaller than 10 km (Joyner and Boore, 1981). At lower magnitudes (e.g., M_w 7), the predictions of Akkar et al. (2014) and Bradley (2013) are closer to the 84th percentile levels of the VS predictions. These observations are consistent with recent approaches in developing global ground-motion prediction models for homogeneous seismotectonic settings (e.g., Cauzzi et al., 2014; Gregor et al., 2014) where significant regional dependence of strong ground motions (if modeled at all) is typically associated with anelastic attenuation terms (Boore et al., 2014) and therefore becomes apparent only at very large distances from the earthquake source. The bottom panel of Figure S1 seems to suggest that the VS equations will perform better at higher magnitudes, thus enhancing the soundness of the predictive tool for potentially damaging events.

---

Figure

Figure S1. Comparison of the predictive model (16th, 50th, and 84th percentile levels; gray curves) for horizontal peak ground motions implemented in Virtual Seismologist (VS) with the recent Europe and Middle East (EUME) and New Zealand (NZ) models by Akkar et al. (2014) and Bradley (2013), both evaluated for strike-slip events, and V_S30 = 760 m/s (black curves with symbols): (top) magnitude scaling at R_JB = 0, and (bottom) attenuation with distance for low (M_w 4) and high (M_w 7) magnitude values.

---

[ Back ]