AI based 1D and 3D P- and S-Wave Velocity Models for the Alpine Mountain Chain from Local Earthquake Data

The increase in seismic data availability as well as the improvement of automated picking algorithms allows us to reassess the seismicity and velocity structure in many regions around the globe with higher accuracy. Using the seismic recordings from a total of more than 1100 stations of the AlpArray Seismic Network and other permanent and temporary stations within the area we work towards a comprehensive 3D P- and S-wave crustal velocity model for the European Alpine region using Local Earthquake Tomography. Phase arrival times of local seismicity are determined by the widely used deep neural network PhaseNet.
We present first a P- & S-wave minimum 1D model of the Greater Alpine region computed with the established linearized inversion algorithm VELEST and compare it to our new 1D model using a bayesian Markov chain Monte Carlo (McMC) algorithm exploring a broader model space. Pg and Pn phase arrivals in the epicentral distance ranges from 0-130km and 300-600km, respectively, are included while picks within the triplication zone from 130-300km are not considered due to difficult phase identification. Both models match within the error margin of the McMC result, while the discrepancy is largest in the lower crust where the resolution decreases due to the chosen epicentral distance ranges. 
With our minimum 1D model as starting model we compute a 3D P-wave model using the SIMULPS code. As the remaining residual distributions of the 1D and 3D model show, the removal of outliers in the pick catalog is more accurate when based on the 3D residuals due to insufficient incorporation of velocity variations along epicentral distance and backazimuth in the 1D model. The most prominent first order structures of the 3D model are in agreement with previous local studies of the area and the model already can be used to consistently improve crustal correction terms on an orogenic scale for teleseimic tomographies and thus sharpen the seismic image of the upper mantle. Furthermore, it will allow to the associate the phases Pg, Pmp & Pn to picked onset times in the crustal triplication zone more accurately. Due to their ray paths these picks are of special importance to the resolution in the lower crust and will contribute significantly to the final 3D P- and S-wave model. Absolute velocities along the Moho interface are higher than in previous studies and therefore in better accordance with values expected from petrology.