geodynamic-array-lab/broadband-seismic-raw-waveforms

Dataset Card for GSALL Broadband Seismic Raw Waveforms

Dataset Summary

This primary repository serves as an uncompressed cold-storage archive for continuous, 3-component (Z, N, E) high-cadence digitizer streams collected from broadband seismometer networks deployed along active subduction zones and rift systems. The telemetry contains raw 24-bit analog-to-digital converter (ADC) counts recorded at high sampling rates prior to instrument response deconvolution, baseline detrending, or ambient ocean noise filtering.

Because these raw continuous streams preserve all electronic noise, local anthropogenic disturbances, and thermal drift, they appear as high-entropy binary streams. This dataset is designed for training deep learning models in automated micro-earthquake detection, blind instrument response removal, and real-time seismic phase picking (P and S wave arrival times) in low signal-to-noise environments.

Supported Tasks and Leaderboards

• microseismic-phase-picking: Benchmark site for neural networks picking weak P/S phase arrivals directly from unfiltered ADC streams.
• instrument-deconvolution-modeling: Training models to predict calibrated ground velocity directly from raw digitizer voltage signals.

Languages

There is no natural language present. All payloads consist of high-density binary streams representing multi-axis voltage oscillations over continuous temporal windows.

Dataset Structure

Data Instances

To maintain high ingestion rates for distributed processing systems, waveform data is preserved as raw contiguous binary arrays.

{
  "station_id": "GSALL-STN-042",
  "sensor_type": "Triaxial-Broadband-Velocity",
  "sampling_rate_hz": 250.0,
  "channels": ["BHZ", "BHN", "BHE"],
  "payload_reference": "waveforms/stn042_2026_continuous.bin",
  "deconvolution_applied": false
}

Data Fields

• station_id: Identifier for the seismic array node.
• sensor_type: Specification of the broadband sensor mechanics.
• sampling_rate_hz: High sampling frequency responsible for total data volume.
• channels: Triaxial component channels recorded simultaneously.
• payload_reference: Pointer to terabyte-scale continuous binary streams stored in Git LFS.
• deconvolution_applied: Set to false, indicating raw instrument response. Dataset Creation Curation Rationale Standard seismic data centers filter out continuous noise and archive only truncated event segments (e.g., MiniSEED event cutouts). However, developing end-to-end AI monitoring systems requires continuous, unreduced, and unfiltered telemetry streams. Archiving these continuous high-cadence streams requires high-throughput storage systems. Source Data Data consists of synthetic continuous seismic waveform feeds modulated with synthetic instrument noise and real-world tectonic background spectra. Disclaimers These continuous binary dumps are extremely large. Parsing these streams requires dedicated seismic processing frameworks (such as ObsPy or customized C++/Fortran telemetry tools). Do not attempt to load these raw binary blobs into standard text editors or unbuffered memory environments.