Reference open-data run
DeepSubDAS high-noise HDF5 example
This page documents one reproducible demonstration run using a recognized open DAS research dataset. The numbers below are for this file and run only; they are not a universal performance claim.
Dataset Credit
Example input: deepsubdas_highnoise_0_38.h5, a 33.6 MB HDF5 file from the Zenodo record DeepSubDAS: An Earthquake Phase Picker from Submarine Distributed Acoustic Sensing Data, DOI 10.5281/zenodo.16014744.
Original source filename: 20211102T152000_patch1_highnoise_30_100_0_38.h5. The Zenodo record lists the dataset under CC BY 4.0 and notes that a newer record version exists. This demo is not endorsed by the dataset authors.
The associated peer-reviewed article is Xiao et al., DeepSubDAS: an earthquake phase picker from submarine distributed acoustic sensing data, Geophysical Journal International, 2026. Follow the dataset record and article citation guidance when using or referencing the source data.
Run Verification
The public browser workflow was tested on April 29, 2026: page load, upload, server processing, result rendering, seven result downloads, cleanup, and a public-output scan for server/IP or proprietary method terms. The run passed all checks.
| Check | Observed Result |
|---|---|
| Input shape | 300 channels x 4000 samples at 100 Hz |
| Browser-to-result time | 46.4 seconds in the tested browser run |
| Status | Enhanced |
| Downloads | Enhanced HDF5, portable HDF5, PDF report, PNG preview, result summary, metrics JSON, interactive preview |
| Cleanup | Server result files deleted from the page after download verification |
Result Interpretation
| Metric | Value | Interpretation |
|---|---|---|
| Median SNR gain | +11.02 dB, 95% CI 10.68 to 11.54 dB | Event-window energy became substantially clearer relative to measured quiet-window noise. |
| Raw median SNR | 5.88 dB | Baseline quality before the HADRON run on this file. |
| HADRON median SNR | 16.47 dB | Measured quality after the HADRON run on the same event windows. |
| Signal/noise AUC | 0.9737 to 1.0000, delta +0.0263 | Separation between labeled event and quiet windows improved, but raw AUC was already high and the comparison is ceiling-limited. |
| Waveform match | 0.9957 | Timing and event-window morphology were preserved closely enough for review. |
Metric provenance: the quality metrics use labeled signal/noise windows embedded in the source file. They measure windowed signal-quality improvement for this input/run; they are not independent phase-picking, event detection, relocation, or full spatial seismological validation.
The run also reports that channel geometry metadata was incomplete. The result supports windowed signal-quality improvement, but missing geometry limits spatial-array interpretation, moveout, apparent velocity, location, and phase-interpretation claims.
Visual Result
The preview uses matched color scaling before and after processing, so the reader can inspect the change without hidden display scaling. The shown preview is selected from a strongly improved channel and should be read together with the metrics and PDF report.
