Relocating earthquakes with empirical traveltimes

A strategy is proposed to incorporate effects of 3-D velocity variations on earthquake locationsusing empirical traveltimes (ETTs). Traveltime residuals are interpolated from those predictedusing a 1-D velocity reference, mapped on to the hypocentres of corresponding earthquakesfor each station in a network. First, station corrections are computed by averaging the residualsover a fixed scale. Then, summary-ray variograms are used to estimate uncertainty and that,in turn, is used to tune a local fit to neighbouring residuals to refine the corrections. Resulting3-D traveltime estimates are then used as a description of the forward problem in a nonlineargrid-search relocation. Data are weighted according to the estimated uncertainty. Data fromthe Icelandic Southern Iceland Lowlands (SIL) national seismic network are used to test thestrategy. ETTs are estimated forP- andS-waves at 65 stations in the SIL network, basedon four million arrival time readings of 300.000 events registered between 1990 and 2012.ETTs are strongly correlated for the two wave types. The spatial variations of the predictedcorrections are consistently comparable or somewhat less forS-waves thanP-waves. Thisfeature suggests variations of theVP/VSratio in the Icelandic crust. Error estimates are alsostrongly correlated for the two wave types and between nearby stations. Relocations aretested by comparison with explosions and small populations of well-located events withindenser subnetworks. Relocations result in modestly enhanced clustering of explosions andearthquakes and significantly improved depth estimates. Estimates of the random relocationerror are statistically better behaved than those of the SIL system. They are in general reduced,as is expected since 3-D heterogeneity is now partly taken into account.