Skip to main content
Dryad logo

Main Ethiopian Rift THBI Data


Petruska, Jon; Eilon, Zach (2022), Main Ethiopian Rift THBI Data, Dryad, Dataset,


This data contains phase velocity maps in the form of .mat files, and in .xyz text files. Data is saved in the .xyz format as : Latitude, Longitude, phase velocity, standard deviation, and ray density. To be added later: RF's, and joint inversion final velocity profiles, also in the form of .mat files. 


Using the Automated Surface Wave phase velocity Measuring Systems (ASWMS) [jin 2015], we developed phase velocity maps at 25, 32, 40, 50, 60, 80, and 100 s using Rayleigh waves from teleseismic earthquakes to characterize lower crustal and mantle wavespeeds. ASWMS autonomously estimates Rayleigh wave phase delays between nearby seismic stations using inter-station, frequency dependent cross-correlograms on a per-event basis at discrete periods. We used 5741 high quality earthquakes occurring between 2000-03-10 and 2017-10-03, of Mw >= 5.5, with a maximum depth of 250 km, at a distance of 30 -160 degrees from from the center of our study area (12N, 39.5E). We processed vertical seismograms for all stations between 4N and 20N latitude, and 34E to 45E longitude. Our data comprises 415 stations, and 17 public networks dowloaded from the IRIS DMC.

For each earthquake, vertical cross-correlograms, C(t), were computed in the time domain for each possible station pair according to: 
C(t)=S1 *Ws S2

where S1 and S2 are the de-trended, windowed, vertical seismograms of the two stations with instrument response removed, and * denotes cross-correlation. Ws is a 300 s windowing function, with a 75 s Hanning taper. C(t) contains lag and coherency information about the Rayleigh wave energy traveling between the two stations. The dominant energy of the waveform was isolated using another windowing function, Wc(t), and then filtered using a series of Gaussian narrow band filters. By fitting the filtered, windowed cross-correlogram function, Fi*(WcC(t)) using a 5-parameter Gaussian wavelet we extracted frequency-dependent phase delays, tp, for this station pair.

Station pairs with inter-station distance (in km) less than 5, or greater than 3.55 x the period (in seconds), were discarded to reduce the possibility of cycle skipping or spurious measurements. Station pairs with cross-coherency gamma<0.6 were discarded to prevent measurement on poorly correlated cross-correlograms. tp were then inverted using the Eikonal equation to develop inter-station phase travel time surfaces at each period [lin 2009]. Regularization for this step is implemented as in [jin2015].

Events for which more station pairs were discarded than not were also discarded from the stacks. We also culled measurements within earthquake phase velocity maps that had apparent velocities below 3.1 km/s or above 4.8 km/s, as velocities outside these bounds are unrealistic for continental crust between 25 to 100 s. We pieced together temporally and spatially isolated regions by stacking individual pixels from events. Pixels were included in the final stack if the number of rays crossing those pixels from all event maps exceeded 66 (providing on average one ray entering each pixel per km of circumference). Pixels in the same location between different events were assigned a weight, Wg, where Wg = { of rays in event} \ {total rays},  such that events with a greater number of observations contributed more to the final stack phase velocity. Pixels with phase velocities more than three standard deviations from the stack mean were also excluded from the final stack. Maps were smoothed using a linear function with a lengthscale of 0.4x the wavelength.

Usage Notes

The .xyz files are in a very straightforward format: x,y locations, a phase velocity, and the standard deviation, which should be applicable for most any usage.


National Science Foundation, Award: EAR-1723170