ADRIZ → INGV Etna Monitoring live wildfire + volcanic status

Live monitoring
Proof of value

Proof of value — check us against your own reports

This page exists to show, not claim. For each event we place what ADRIZ independently produced next to INGV's own public report for the same date, with a live link so you can verify every number yourself. We include the cases where we missed or disagreed — credibility comes from showing those too. Every rate carries a sample size n and a 95% confidence interval; read the CI, not the point estimate.

Primary INGV source for the volcanic rows: the INGV-OE weekly bulletin, mirrored verbatim in English by the Smithsonian GVP (Etna 211060). Navigate the GVP Etna archive to the week label in each row. Reproduction method per claim: ingv_proof_of_value.md · raw data: proof_of_value.json.

1. Event timeline — ADRIZ vs INGV's own report (with live links)

DateINGV's own report (live link)What ADRIZ independently produced Independent corroborationVerdict
Loading…

2. Continuous agreement scoreboard (n + 95% CI)

Auto-generated from comparison.json: INGV-OE weekly bulletin (via GVP 211060) vs ADRIZ satellite-derived calls.

3. Model proof — tested against INGV's own camera frames

4. Honest limits — where we miss

Stated plainly, not buried

    Layers
    Volcanic (SO2)
    Wildfire active
    Quiet AOI
    FIRMS fire (size = FRP)

    What you are seeing (for the INGV reader)

    This is the system running live on real data over Etna. The orange dots on the map are NASA FIRMS NRT active-fire detections (VIIRS/MODIS, ~3–6 h latency), refreshed every 30 min by a serverless worker — not a static snapshot. The coloured AOI markers are our per-area verdict (volcanic / wildfire / quiet) with our score, confidence and source timestamp. Below, the system auto-grades itself against your own weekly bulletin (via GVP 211060). For the curated, verifiable head-to-head — our detections next to your own reports, with links and honest misses — open the “Proof of value” tab above.

    Per-AOI status — OUR assessment

    Loading…

    ADRIZ vs INGV — auto self-grading

    Live agreement against the INGV-OE bulletin

    We ingest INGV-OE's public weekly reporting on Etna (via Smithsonian GVP, volcano 211060) into a machine-readable ground-truth feed and AUTO-COMPARE it against our own satellite-derived calls. Each rate carries a sample size n and a Wilson 95% CI. Read the CI, not the point estimate — the overlap is thin and historical.

    Loading comparison…
    DimensionAgreement95% CIn

    per-date detail (our call vs the INGV weekly window)
    Obs dateINGV week / stateOur SO2 / fireMatch

    What we ingest from INGV-OE

    Loading INGV feed…

    SO2 here is a plume-presence cue, not an eruptive-state classifier. Active fire is FIRMS NRT (~3–6 h). INGV bulletins are weekly; our signals are per-overpass, so per-day calls and the weekly window legitimately differ at episode boundaries.

    Capability slots

    Data latency / freshness

    LayerObservedAgeTypical

    Operating model

    Real-time (auto-refreshed):

    • Active fire (FIRMS NRT) — pulled by a serverless Worker on a 30-min cron and served from edge storage; the page re-checks it every 5 min. Source latency ~3–6 h.
    • SO2 plume status — refreshed each Sentinel-5P overpass (~daily) by the feed processor (this preview reads an archived per-scene series).

    File-fed (update when their result file is published): detector alerts, multi-source thermal/FRP, escalation flag, burn-scar, fuel-danger. These are the heavy model layers — they show pending until their result file lands, then light up. They are not live-polled.