To determine precise earthquake hypocenters, we adopted a relocation procedure that comprises three stages: (i) inversion in the 1D background velocity model obtained in (16), (ii) inversion in the 3D background velocity model obtained in (16), and (iii) relocation with the DD method (35). Stages (i) and (ii) closely followed the methodology of (16), except that, here, we fixed the velocity model to those obtained in that study—i.e., the velocity models were not updated during the relocation. Below, we discuss the outcome of each stage in more detail.

In stage (i), we computed earthquake locations by least-squares inversion of P- and S-arrival times in a 1D model. We performed picking in Seisan (36), and the relocation was done with the program Hypocenter (37) included in Seisan. For each event, we visually checked that picks were correctly assigned and removed picks with residuals larger than 6 s. We estimated absolute location errors from the solution variance (arrival time misfit) as described in (16, 37). The hypocenter solutions obtained at this stage have average location errors of 5.1 km in the horizontal direction and 7.5 km in the vertical direction. The results were used to identify events with deep origin (>35 km depth, taking into account vertical error bars) that will be retained for further processing.

In stage (ii), we relocated individual earthquakes by iterative least-squares inversion of P- and S-arrival times in a 3D model to obtain more accurate locations for the events retained in the previous stage. The initial locations used for the inversion are those obtained in stage (i). We calculated arrival times in the 3D velocity model with SIMULR16 (38), using the same ray tracer, parametrization, and distance- and residual-dependent weighting scheme as in (16). This inversion yields hypocenter solutions with considerably smaller absolute location errors averaging 1.3 km in the horizontal direction and 1.8 km in the vertical direction.

In stage (iii), we relocated events by minimizing differential arrival times between event pairs, which enhances the resolution of relative hypocenters for clustered events. The differential arrival times used as inputs were obtained via both catalog hypocenters and CCs of seismic waveforms. A first list of inputs was built by compiling differential arrival times for all pairs of catalog hypocenters from stage (ii) that are less than 20 km apart. This results in 1,851,141 P-differential arrival times and 1,695,296 S-differential arrival times from 265 stations. We then identified pairs from the catalog list that are less than 12 km apart. For these, we computed a second list of differential arrival times by cross-correlating waveforms with the EQcorrscan package and the ObsPy toolbox (39). The waveforms we used for CC were 1.0 s long, starting 0.3 s before the P and S phases, and were band pass–filtered between 2.5 and 8 Hz. To retain a differential time pick, we required a normalized CC coefficient of at least 0.7 and a resulting value that differs by less than 0.4 s from the catalog value. This yields a second list of 173,173 P-differential arrival times and 142,128 S-differential arrival times from 133 stations. The differential arrival times from the two lists were then inverted simultaneously via a DD algorithm with the hypoDD program [version 2.1b (33)].

HypoDD’s DD algorithm involves an iterative inversion in which we can introduce a stricter weighting on the input data at each new iteration. This was done by progressively increasing the weight for CC differential times (from 0.02 to 1), by reducing the maximum distance across which event pairs are compared (from 12 km down to 3 km for CC-based inputs), and by removing data outliers (from a complete dataset in the first iteration to the removal of data points with residuals larger than six times the standard deviation (SD) of all residuals in the final iteration). In the final iteration, we were thus left with 51% of catalog and 32% of CC differential arrival times to estimate the relative locations of clustered hypocenters. At this stage, we allowed event pairs to form a cluster when they are connected through at least eight catalog differential times and eight CC differential times. To address the variable data coverage during the entire time period (2006–2017), we varied inversion parameters and verified that hypocenters relocate consistently for overlapping data subsets. This led us to adopt a relatively high damping value of 600 to achieve a reasonable system condition number of 40 to 60 for the entire dataset. We found that active structures outlined by hypocenters from the high-coverage period (June 2006 to October 2007 for the Tripoli cluster) were also well constrained and appeared more complete when including hypocenters from the complete time period.

The results of the DD relocation are presented in Figs. 1 to 3 and are available in external data file S2. As there are no large gaps in seismicity that would interrupt the clustering chain within the slab, most slab seismicity was contained within one large cluster. This remains true regardless of the existence of localized seismicity gaps updip of the interface vents, as hypocenters remain connected away from these zones. Separate clusters occurred, e.g., at large depth and in the deep overriding crust (below the Gulf of Patras). Shallower earthquakes in the overriding crust were not processed. As the DD relocation mainly improves relative hypocenter locations, here, we estimated relative errors [rather than absolute errors as in stages (i) and (ii)] through jackknife resampling of the dataset (35, 40). For this, we reran the DD inversion 1000 times with a reduced dataset in which 10% of the differential arrival times were randomly removed. The errors were then estimated with the general “delete-j” jackknife estimator described in (40). We find that, on average, the relative location error is 0.19 km in the horizontal direction and 0.21 km in the vertical direction. The relative errors of earthquakes within ±50 km of the cross section in Fig. 3A are shown in fig. S7.

Note: The content above has been extracted from a research article, so it may not display correctly.



Q&A
Please log in to submit your questions online.
Your question will be posted on the Bio-101 website. We will send your questions to the authors of this protocol and Bio-protocol community members who are experienced with this method. you will be informed using the email address associated with your Bio-protocol account.



We use cookies on this site to enhance your user experience. By using our website, you are agreeing to allow the storage of cookies on your computer.