Input file specification

The application requires a MiniSEED Waveform (or any subtype) containing seismograms on which the wave arrival time should be detected. The waveform has to contain E, N and Z channels for each station.

Figure 1. Application input files specification

Filling form values

The application form consists of five parameters - see Figure 2. The first parameter allows you to choose the model used for the the detection - GPD or PhaseNet (see [1] and [2], respectively). Each model was originally trained on a set of data described in the respective publication, the weights obtained based on that training are set as default, however, different weights may be chosen depending on the characteristic of the input data by modifying the Pretrained model weights field. The Original option should be set to use the default, original weights published by the authors of the models, while to use one of the weights obtained for public data described in [3] use one of the ETHZ, GEOFON, INSTANCE, Iquique, LenDB, NEIC, SCEDC and STEAD options. Additionally, for the PhaseNet model, there are also weights obtained by training on BOGDANKA and LGCD¹ episodes available. See Figure 3 for detailed view of the available weights.

To configure the sensitivity of the phase detection, use the P phase threshold and the S phase threshold (see Figure 2), which determine the threshold of the probability for which we assume a detection. 

Figure 2. Application form with values filled

Figure 3. Available values of the Pretrained model weights

⁽¹⁾ The weights for the LGCD episode are not well suited for S phase detection.

Produced output

The main output of the application is a QuakeML file containing picks marking the phase detections. The file is named after the input MiniSEED file with -picks.xml suffix added. The picks are displayed on the original input file within the application output view (Figure 4).

Figure 4. Application output - MiniSEED waveform with resulting picks displayed

As an additional information, the application provides also the detection probability values, saved to the file named after the input MiniSEED with -annotations.mseed suffix added. The probability can be examined (see Figure 5 and Figure 6) after expanding the Wave detection probability option (bottom of Figure 4).

Figure 5. Application output - wave detection probability

Figure 6. Details of detection probability plots, for phase P, phase S and noise

Related publications

[1] Ross, Meier, Hauksson and Heaton (2018). Generalized Seismic Phase Detection with Deep Learning. doi: 10.1785/0120180080.

[2] Zhu, W., & Beroza, G. C. (2019). PhaseNet: a deep-neural-network-based seismic arrival-time picking method. doi: 10.1093/gji/ggy423

[3] Woollam, Münchmeyer, Tilmann et al. (2022). SeisBench — A Toolbox for Machine Learning in Seismology. doi: 10.1785/0220210324.

Back to top

Related Documents

• Page:
  P and S Waves Detection Tool user guide
• Page:
  Seismic Catalog Recalculation user guide
• Page:
  Waveform-based seismic event location user guide
• Page:
  Spectral Analysis user guide
• Page:
  Mechanism: Shear-Tensile crack user guide