Dataset: Earthquake Seismogram Denoiser (EQS-Denoiser)

This dataset contains preprocessed earthquake event and noise waveforms, supplementary to the publication:

"Towards Operational Earthquake Seismogram Denoising"
by Nikolaj Dahmen, John Clinton, Men-Andrin Meier, Luca Scarabello
Bulletin of the Seismological Society of America

Waveform data are provided in physical units [m/s],  Z12 components and with a sampling rate of 100 Hz. Event signals have been preprocessed according to the methods described in the publication to reduce background noise (see Figure 2). Event signals have 18k samples with P arrival aligned at sample 6000, noise signals have 8k samples. Dataset statistics for event and noise waveforms are summarized in Figures S1 and S2. The dataset is separated into training, validation, and testing dataset, and are here available for publicly accessible networks. Metadata includes information on recording station, earthquake, and identified phase arrivals. A subset of the test dataset with fixed event and noise waveform pairs is provided in the benchmark dataset. 

Data are provided in HDF5 format with accompanying metadata, compatible with SeisBench.

Proposed dataset structure:

• Event waveforms:

  • Event_data/waveforms.hdf5

  • Event_data/metadata.csv

• Noise waveforms:

  • Noise_data/waveforms_noise.hdf5

  • Noise_data/metadata_noise.csv

Importing dataset with SeisBench:

import seisbench.data as sbd

# load noise dataset
noise_ds = sbd.WaveformDataset("/Noise_data/", sampling_rate=100, component_order="Z12")
# Training examples: 274,240
# Development examples: 81,416
# Test examples: 40,096

# load event dataset
event_ds = sbd.WaveformDataset("/Event_data/", sampling_rate=100, component_order="Z12")
# Training examples: 19,022
# Development examples: 5,385
# Test examples: 2,695