PlumeSentinel AI
Architecture

Three-Layer Architecture

From operational forecasts, through agentic AI, to decisions and public guidance.

Three-Layer Architecture

PlumeSentinel AI is structured in three layers that work together at operational tempo:

Physical Layer
Operational Forecasts
Federal smoke & fire forecast products
Satellite observations
Complementary research models
Operations Layer
Agentic AI Platform
Grounded fusion of forecasts & observations
Provenance & confidence
Scenario analysis
Human review & escalation
Public-Sector Layer
Decision & Communication
Wildfire smoke briefings
Ranked action options
Health-protective recommendations
Public messaging drafts
Independent validation continuously compares forecasts to observations and informs the operations layer.
  • Physical Forecast Layer — federal smoke and fire forecast products plus complementary research models.
  • Agentic AI Platform — grounded, agentic workflows that fuse observations and forecasts into a current incident state.
  • Decision Support and Communication — wildfire smoke briefings, ranked actions, public messaging.

Satellite Imagery

Two complementary satellite systems anchor the observational backbone — polar-orbiting VIIRS (Visible Infrared Imaging Radiometer Suite) for high-resolution spatial detail, and geostationary GOES ABI (Geostationary Operational Environmental Satellite — Advanced Baseline Imager) for continuous temporal tracking.

VIIRS true-color satellite image showing a dramatic spiral of wildfire smoke over the Northwest Territories, September 2023

VIIRS: Wildfire Smoke Plume Detection

A Smoky Cloud Swirl over Northwest Territories

VIIRS true-color image from NOAA-20 (the second satellite of the Joint Polar Satellite System), September 16, 2023. Smoke from boreal wildfires forms a dramatic spiral spanning hundreds of kilometers, driven by upper-level wind shear. VIIRS captures both the spatial extent of the plume and the thermal anomalies that generate it.

Credit: NASA Earth Observatory / Lauren Dauphin, using VIIRS data from NASA EOSDIS (Earth Observing System Data and Information System) LANCE (Land, Atmosphere Near real-time Capability for EOS), GIBS (Global Imagery Browse Services)/Worldview, and JPSS (Joint Polar Satellite System). science.nasa.gov →

GOES-18 geostationary satellite image showing Canadian wildfire smoke sweeping across multiple U.S. states, May 2023

GOES ABI: Continental-Scale Smoke Tracking

Smoke Fills North American Skies

GOES-18 geostationary image, May 15, 2023. Canadian wildfire smoke sweeps across multiple U.S. states. Unlike polar-orbiting VIIRS (~2 passes/day), GOES captures imagery every 5–15 minutes — real-time plume tracking at the cadence PlumeSentinel AI needs.

Credit: NASA Earth Observatory / Lauren Dauphin, using GOES-18 imagery courtesy of NOAA (National Oceanic and Atmospheric Administration) and NESDIS (National Environmental Satellite, Data, and Information Service). science.nasa.gov →

Operational Forecasts

National Weather Service smoke and fire forecast products provide the physical backbone:

  • Operational regional smoke forecasts at high spatial resolution
  • National air-quality forecasting at synoptic scales
  • Next-generation experimental smoke and dust products
  • Upstream fire detection and emissions inputs

Independent Validation

Forecasts are continuously compared against observations:

  • Polar-orbiting and geostationary satellite aerosol products
  • Surface PM2.5 reference monitoring
  • Operational hazard mapping analyses
  • Confidence diagnostics for every forecast product in use

Agentic AI Platform

The core operations layer uses grounded, agentic workflows to:

  • Integrate live observations and forecast outputs into a current incident state
  • Track provenance, confidence, and uncertainty across products
  • Run scenario analyses and rank response options
  • Coordinate human review and escalation

Decision and Communication

The decision-support layer converts incident intelligence into public-sector action:

  • Operator-facing wildfire smoke briefings
  • Ranked action options for agencies
  • Health-protective recommendations
  • Audience-specific public messaging drafts
  • Auditable records for after-action learning

The 15-Minute Promise

Reduce the path from plume detection to decision-ready briefing from hours of manual synthesis to under 15 minutes.