Figure 3 An example of the P-wave onset time picking by PickNet (Wang et al., 2019) using vertical-component records of the 10 February 2019 MW5.0 event in the border region between Myanmar and India (epicentre and station locations shown by the black dot and triangles in Figure 1). The station name is given in the beginning of each trace, and the red vertical line shows the picked P-wave onset time. (a) 600-s long records starting from the earthquake original time. (b) Zoom-in views of the records in (a) in 20-s-long windows starting from 10 s before the theoretical P-wave arrival times in AK135 model (Kennett et al., 1995). All waveforms are self-normalized by the maximum amplitudes.
To evaluate the coverage of our dataset, we calculate the number of the first-arriving P-wave ray paths through each 0.2° × 0.2° cell at different depths, and the results are displayed in Figure 4. The ray path density distribution shows that both the lower crust and uppermost mantle (depth range 45-70 km) are generally well sampled by the dataset, with large parts of the region having ray paths of more than 150 in each cell. The density of the ray paths increases from the lower crust to the uppermost mantle. With the decrease of the takeoff angles for longer distances, sampling points at greater depths increase, particularly in the uppermost mantle where the ray paths are nearly horizontal, as shown in Figure 2. This implies that our dataset may provide better resolution in the uppermost mantle than that in the lower crust.