explain_silence¶

Science¶

The negative-evidence check. For every station that was delivering but has no pick on the referenced origin, the PMC model asks whether a detection was expected. Stations fall into unexplained (verdict should_have_detected) or explained with a reason.

below_noise means detection_probability < 0.5 — the fitted model puts the odds against a detection — not a statement that the instantaneous signal sits below the noise floor. has_pick, silent, absent, out_of_range, no_waveform and no_data_in_window are the other reasons. A station marked delivering in the daily table but with no archive bytes in the predicted window is no_waveform or no_data_in_window, not silent, and is not negative detection evidence.

log_likelihood is the combined detected/silent term of the PMC model, comparable across events, so thresholds can be fitted rather than guessed. When n_trials is 0 it is 0.0 — no trials, not perfect consistency.

One unexplained silence can be a local noise burst. Several unexplained silences at stations closer than those that detected is the strongest single piece of evidence this set of tools can give that the solution is wrong. A value origin is accepted and then the result is not catalogue-bearing.

References¶

Schorlemmer, D. and Woessner, J. (2008). Probability of detecting an earthquake. Bulletin of the Seismological Society of America. doi:10.1785/0120070133

Contract¶

Generated from the live registry (ToolSpec). Field names, types and defaults come from the input and payload models; they are not hand-typed.


Group	`forward`
Class	3
Tool version	`1.2.0`
Highest declared tier	F
Timeout	30 s
Budget key	`none`
`readOnlyHint`	true
`idempotentHint`	true

Input¶

Field	Type	Required	Notes
`origin_ref`	object (kind `catalog` \| `solution` \| `value`)	yes	discriminated on `kind`
`stations`	array[string] \| null	no	default `null`

Payload¶

Field	Type	Required	Notes
`checked`	integer	yes
`unexplained`	array[object]	yes
`unexplained[].station_id`	string	yes
`unexplained[].distance_km`	number	yes
`unexplained[].r_hyp_km`	number \| null	no	default `null`
`unexplained[].predicted_p_time_utc`	string \| null	yes
`unexplained[].detection_probability`	number	yes
`unexplained[].expected_snr_db`	number	yes
`unexplained[].measured_noise_db`	number	yes
`unexplained[].verdict`	const `should_have_detected`	yes
`explained`	array[object]	yes
`explained[].station_id`	string	yes
`explained[].reason`	string ∈ {`has_pick`, `below_noise`, `silent`, `absent`, `out_of_range`, `no_waveform`, `no_data_in_window`}	yes
`explained[].detection_probability`	number \| null	no	default `null`
`explained[].r_hyp_km`	number \| null	no	default `null`
`log_likelihood`	number	yes
`n_trials`	integer \| null	no	default `null`
`model_version`	string	yes

Example request¶

{
  "origin_ref": {
    "kind": "catalog",
    "event_id": 1001
  }
}

Example payload¶

{
  "checked": 28,
  "unexplained": [
    {
      "station_id": "YY.STA3",
      "distance_km": 55.0,
      "r_hyp_km": 56.1,
      "predicted_p_time_utc": "2024-06-15T08:12:12.100Z",
      "detection_probability": 0.91,
      "expected_snr_db": 11.2,
      "measured_noise_db": -147.0,
      "verdict": "should_have_detected"
    }
  ],
  "explained": [
    {
      "station_id": "ZZ.STA9",
      "reason": "below_noise",
      "detection_probability": 0.12,
      "r_hyp_km": 312.4
    },
    {
      "station_id": "XX.STA1",
      "reason": "has_pick",
      "detection_probability": null,
      "r_hyp_km": 43.0
    }
  ],
  "log_likelihood": -3.42,
  "n_trials": 22,
  "model_version": "pmc-2024-06"
}

What this tool does not tell you¶

What this tool does not tell you: that one unexplained silence means a wrong solution — one can be a local noise burst. Several unexplained silences at stations closer than those that detected is the strongest single piece of evidence available that the solution is wrong.

Warnings named for this tool¶

Code	When
`partial_result`	a requested station was not scored
`availability_stale`	the availability table has no row for the origin's UTC day