Publication Date:
2015-06-09
Description:
A global validation dataset of 116 seismic events and 20,977 associated Pn and P arrivals is used to assess travel-time prediction and event location accuracy for the global-scale, 3D, P -wave velocity model called LLNL-G3Dv3 ( Simmons et al. , 2012 ). Strong regional trends that are observed for ak135 travel-time residuals are largely removed when LLNL-G3Dv3 is used for prediction. The 25th–75th quantile spread of travel-time residuals is reduced by 30%–40% at teleseismic distances, and the spread is reduced by ~60% at regional distances (〈16°). Epicenter error decreases when more data are used to constrain event locations until more than ~40 arrivals times are used. At which point, epicenter error reduction tends to plateau. Median epicenter errors for the ak135 and LLNL-G3Dv3 models plateau at ~8.0 and ~5.5 km, respectively, for teleseismic P datasets. Median epicenter errors for the ak135 and LLNL-G3Dv3 models plateau at ~12.0 and ~4.0 km, respectively, for regional Pn datasets. We demonstrate that spatially correlated travel-time residual errors for the ak135 model lead to increasing epicenter error when ~40 to ~100 Pn arrivals are used to constrain the location. The effect of correlated error is mitigated by LLNL-G3Dv3, for which epicenter error steadily decreases to ~4 km when 100 Pn arrivals are used. The median area of 0.95 epicenter probability bounds for ak135 and LLNL-G3Dv3 are 1811 and 758 km 2 , respectively. The ak135 ellipses are inflated to achieve the desired rate of true events occurring inside the probability region, whereas LLNL-G3Dv3 error ellipses based on empirical residual distributions cover the true location at the expected rate because location bias is minimal.
Print ISSN:
0037-1106
Electronic ISSN:
1943-3573
Topics:
Geosciences
,
Physics
