Two subduction-related heterogeneities beneath the Eastern Alps and the Bohemian Massif imaged by high-resolution P-wave tomography

. We present high-resolution tomographic images of the upper mantle beneath the E. Alps and the adjacent Bohemian Massif (BM) in the North based on data from the AlpArray-EASI and AlpArray Seismic Networks. The 15 tomography locates the Alpine high-velocity perturbations between the Periadriatic Lineament and the Northern Alpine Front. The northward-dipping lithosphere keel is imaged down to ~200-250 km depth, without signs of delamination, and we associate it with the Adriatic plate subduction. Detached high-velocity heterogeneity, sub-parallel to and distinct from the E. Alps heterogeneity is imaged at ~100 - 200km depths beneath the southern part of the BM. We associate this heterogeneity with the western end of a SW-NE striking heterogeneity beneath the south-eastern part of the BM, imaged in models of 20 larger extent. The strike, parallel with the Moldanubian/Brunovistulian mantle-lithosphere boundary in the BM and with the westernmost part of the Carpathian front, lead us to consider potential scenarios relating the heterogeneity to (1) a remnant of the delaminated European plate, (2) a piece of continental-and-oceanic lithosphere mixture related to the building of the BM, particularly to the closure of the old Rheic ocean during the MD/BV collision or (3) a lithospheric fragment going through to the NW

The TimePicker 2017 enables highly-accurate automatic measuring of absolute arrival times of teleseismic Pwaves based on waveform cross-correlations and beam-forming.Assuming a waveform similarity of the teleseismic P-waveforms within an array, we cross-correlate and stack shifted traces of all the stations (grey traces), and create a low-noise beam trace of an event (black trace).Times of extremes on the station signal (red trace) are measured by three different methods.First, times of the station signal extremes are picked (red P1, P2).Second, we crosscorrelate the station signal with the long P-waveform beam (~2.5 signal period) and determine the long correlation picks (black P1, P2).Third, we cross-correlate the station signal with the short P-waveform beam (~0.5 signal period) around each of the extremes and determine the short correlation picks (blue P1, P2).Each of the red, black and blue picks is complemented by its error estimate.Time error of the red extreme depends on a signal noise level (see cyan basins, their height is given by a noise magnitude -red dashed lines), errors of the black and blue correlation picks come from coherence of the signal with the beam.The final times of the extremes (green P1, P2) are probabilistic combinations of the three partial picks (red, black and blue P1, P2), which assures coherent measurements in case of more complex waveforms.
The extreme with the lowest error estimate (green P2) is chosen for computation of the absolute arrival time of the P-wave at the station.For that we determine (1) the arrival time on the beam (P_abs) at a moment when the beam leaves its noise channel and (2) time of the beam extreme corresponding to the green P2 extreme.The time difference between them is then subtracted from the final time of the chosen station signal extreme (green P2).

Figure S1 :
Figure S1: Distribution of teleseismic earthquakes used in the study and ray-path coverage (a).Distributions of extended set of earthquakes arriving from northern and southern back-azimuths in fans of 60° (b).

Figure S3 :
Figure S3: Maps of Moho depth (left) and thickness (center) and velocities (right) of sediments used to correct P-wave residuals for the crust relative to IASP'91 model.

Figure S4 :Figure S5 :Figure S6 :
FigureS4: Data and model variance trade-off curve evaluated for various values of damping of the isotropic-velocity perturba ons and numbers of itera ons.Red arrow marks variances for the final model.The data variance and model variance are squared norms of the me residuals and velocity perturba ons, respec vely.The data uncertain es are included in the evalua on of the data variance.Data variance in the code is 1/(N-1)*sum((residual -average of residuals)*weight)**2, i.e., instead of dividing by sigma, which is some me used as well, the numerator is mul plied by a unitless weight.

Figure S7 :
Figure S7: Pairs of horizontal slices (a) and ver cal cross-sec on (b) through the checkerboard model (right images in the pairs) and retrieved perturba ons (le images in the pairs), plo ed for all inverted node levels.The same mask as in Fig. 3 is used for shading in part (b).

Figure S8 :
Figure S8: Velocity perturbations along the N-S cross-section in the center of the array from real data (upper row) and from synthetic data (middle row) calculated for models of the steep detached slab beneath the E. Alps (lower row).The top of the heterogeneity migrates upward (bottom row) from 150 km (detachment as in Paffrath et al., 2021) to 60 km (representing no detachment).The slab detachment larger than the 30km grid would be revealed in the upper 200 km of the EASI-AA model.A potential leakage does not overprint the images.