Skip to content

Image Function

For image functions this version of Qseek relies heavily on machine learning pickers delivered by SeisBench.

SeisBench Image Function

SeisBench offers access to a variety of machine learning phase pickers pre-trained on various data sets.

Citation PhaseNet

Zhu, Weiqiang, and Gregory C. Beroza. "PhaseNet: A Deep-Neural-Network-Based Seismic Arrival Time Picking Method." arXiv preprint arXiv:1803.03211 (2018).

SeisBench Module

PhaseNet image function. For more details see SeisBench documentation.

model: PhaseNet | EQTransformer | OBSTransformer | LFEDetect

The model to use for the image function. Currently supported models are PhaseNet, EQTransformer, GPD, OBSTransformer, LFEDetect.

pretrained: cascadia | cms | diting | dummy | ethz | geofon | instance | iquique | jcms | jcs | jms | lendb | mexico | nankai | neic | obs | obst2024 | original | original_nonconservative | san_andreas | scedc | stead | volpick

SeisBench pre-trained model to use. Choose from ethz, geofon, instance, iquique, lendb, neic, obs, original, scedc, stead. For more details see SeisBench documentation

window_overlap_samples: 1500

Window overlap in samples.

torch_use_cuda

Use CUDA for inference. If True use default device, if int use the specified device.

torch_cpu_threads: 4

Number of CPU threads to use if only CPU is used.

batch_size: 128

Batch size for inference, larger values can improve performance.

stack_method: avg | max

Method to stack the overlaping blocks internally. Choose from avg and max.

rescale_input: 1.0

Upscale input by factor. This augments the input data from e.g. 100 Hz to 50 Hz (factor: 2). Can be useful for high-frequency microseismic events.

phase_map

Phase mapping from SeisBench PhaseNet to Lassie phases.

weights

Weights for each phase.

JSON for SeisBench
{
  "image": "SeisBench",
  "model": "PhaseNet",
  "pretrained": "original",
  "window_overlap_samples": 1500,
  "torch_use_cuda": true,
  "torch_cpu_threads": 4,
  "batch_size": 128,
  "stack_method": "avg",
  "rescale_input": 1.0,
  "phase_map": {
    "P": "cake:P",
    "S": "cake:S"
  },
  "weights": {
    "P": 1.0,
    "S": 1.0
  }
}