WF-5 · Satellite

Satellite — event-based Etna

A new Etna event triggers multi-source data fusion and two indicative 12-hour atmospheric trajectories.

Active · event-drivenChecked every 6 hours; posted only for a valid new eventGeospatial · conditional forecast
Vertical card with ~700 and 500 hPa trajectories, corridors, regime, thermal/seismic context and safety gate.
1 · In one sentence

GVP event, INGV sources, earthquakes, FIRMS and winds are combined into an educational potential ash-transport scenario.

2 · Why it exists

Move from a public event to a traceable geospatial picture without presenting it as an official alert.

3 · What it produces

Vertical card with ~700 and 500 hPa trajectories, corridors, regime, thermal/seismic context and safety gate.

How to read the latest output: Lines follow forecast winds at two levels; bands are indicative corridors. They do not represent concentration, deposition or hazard.

Lines follow forecast winds at two levels; bands are indicative corridors. They do not represent concentration, deposition or hazard.

4 · How it works

From sources to output.

Workflow pipeline diagram
Pipeline diagram: from public sources through checks and computation to the published cards, with human oversight.
1New event
2INGV gate
3Multi-data
42-level winds
5Trajectories
6Publication

5 · Data used

Data used
SourceVariableFrequencyLimitation
Smithsonian GVPPublic dated Etna eventEvery checkFeed delay and granularity
INGV/OEOfficial notices and safety checkEvery eventConservative parsing; uncertainty blocks
NASA FIRMS VIIRS24-hour thermal anomaliesEvery eventThermal proxy, not proof of ash
Earthquake catalogue24-hour events and magnitudeEvery eventCoverage and thresholds
Open-MeteoHourly winds at 700 and 500 hPaEvery eventInterpolated field; not an ash model
6 · Models and rules

Models and rules

Hourly Lagrangian advection converts wind speed and direction into geographic displacement. Two pressure levels show vertical differences; mean speed, variability and directional shear feed an MLP regime classifier. DBSCAN groups FIRMS anomalies; Isolation Forest can flag unusual configurations.

7 · Controls

Controls

Fail-closed official safety gate, freshness, provenance, stale-image checks, exact and perceptual hashes, anti-duplicate state and two independent bilingual posts.

8 · Results and metrics

Results and metrics

Run outcomes (published, skipped, blocked), source availability and duplicates prevented. No operational dispersion skill is claimed.

9 · Limits

Limits

No source term, real plume height, grain size, turbulence, sedimentation, deposition, concentration, visibility or aviation hazard.

Possible applications of the pattern

Event and geospatial fusion

Possible applications: event-driven workflows for environmental phenomena, distributed infrastructure or territorial monitoring. Observations, simulations and official communications must remain separate.

Transferable components

  • GVP trigger and INGV safety gate
  • earthquakes, FIRMS and winds
  • 700/500 hPa advection
  • bilingual anti-duplicate state

Research questions

  • uncertainty propagation
  • comparison with dispersion models
  • event-trigger quality
  • regime validation
Application mapShared technical details

Workflow technical dossier

IMPLEMENTEDDescribes behaviour present in the workflow.EXTENSIONIndicates a possible check or evolution, not an operating feature.

Implementation detail and assessment criteria.

This section connects the visible output to data-engineering, modelling, validation and delivery choices present in the repository.

Architecture and data

  • Trigger on a new dated Etna event in the GVP feed, followed by a fail-closed INGV/OE official gate.
  • Fusion of seismicity, FIRMS hotspots and Open-Meteo winds; typed objects with Pydantic.
  • JSON state with event, fingerprint, image hash and URIs of two standalone publications.

Features and methods

  • Hourly trajectory integration through Lagrangian advection at roughly 700 and 500 hPa.
  • Indicative corridors derived from central paths; shear and geometry feed regime classification.
  • DBSCAN for FIRMS clusters, Isolation Forest for unusual configurations and MLPClassifier/optional PyTorch.

Validation and failure modes

  • Checks on event age, official source, current data, timestamps and non-reuse of previous images.
  • Tests for safety, captions, events, seismicity, guards, release contracts and bilingual publishing.
  • The method does not model concentration, sedimentation, grain size or aviation hazard.

Runtime and delivery

  • Installable Python package with Hatch CLI, pytest and Ruff; dependencies locked in uv.lock.
  • Beautiful Soup/pypdf, requests, pandas, scikit-learn, matplotlib/Pillow and GitHub Actions.
Compare all workflows Research and development

10 · Operating timeline

Operating timeline

  • Status: Active · event-driven
  • Frequency: Checked every 6 hours; posted only for a valid new event
  • Page updated: July 2026

11 · Technical detail

Method and assumptions

Hourly Lagrangian advection converts wind speed and direction into geographic displacement. Two pressure levels show vertical differences; mean speed, variability and directional shear feed an MLP regime classifier. DBSCAN groups FIRMS anomalies; Isolation Forest can flag unusual configurations.

Main failure modes

No source term, real plume height, grain size, turbulence, sedimentation, deposition, concentration, visibility or aviation hazard.

Publication governance

Fail-closed official safety gate, freshness, provenance, stale-image checks, exact and perceptual hashes, anti-duplicate state and two independent bilingual posts.

See real outputs

The Publications page reads the public Bluesky feed and shows up to three recent runs, including multiple images in one post.

Open publications