Full pipeline · Under 30 seconds

From an address to a
complete risk report

Every Satelife report runs through a 14-step AI pipeline that pulls real satellite data, historical fire records, and on-the-ground imagery — then scores your property across six risk dimensions.

14 steps  ·  6 data layers  ·  1 actionable report
Phase 01

Address resolution

Every run starts with a property address. We resolve it to exact parcel coordinates and validate coverage before any data is fetched.

01

Address input

The user submits a US property address — street, city, state. Batch mode supports up to hundreds of parcels per request for portfolio analysis.

Entry point
02

Geocoding & parcel resolution

The address is resolved to latitude/longitude coordinates via Nominatim (OpenStreetMap), with Google Maps as fallback. The result is cached for 30 days to avoid repeat lookups.

Nominatim · Google Maps
Phase 02

Multi-source data collection

Seven independent data streams run in parallel — imagery, terrain, fire history, vegetation, weather, wind, and road access. Each feeds a separate scored layer.

03

Satellite & street imagery

Nine image sources fire in parallel: Google Maps Static, MapTiler, Esri World Imagery, NAIP aerial, and a four-direction Street View sweep (N/E/S/W). All stored in Supabase Storage.

Google · MapTiler · Esri · NAIP
04

Terrain & topography

Elevation, slope, and aspect are derived from USGS Digital Elevation Models. Steep south-facing slopes with uphill wind exposure receive higher hazard scores.

USGS DEM
05

Fire hazard zone classification

CAL FIRE's public ArcGIS REST service returns the official Fire Hazard Severity Zone (Moderate / High / Very High) for the parcel's jurisdiction — State, Local, or Federal Responsibility Area.

CAL FIRE · FHSZ
06

Fire history analysis

Historical fire perimeters from CAL FIRE and NIFC are queried for the surrounding area. Proximity and recency of past fires directly affect the risk score.

CAL FIRE · NIFC
07

Vegetation & fuels assessment

Vegetation density and fuel continuity are measured at 30 m and 100 m aggregation from NLCD and California Forest Observatory data. Dense, continuous fuel loads elevate the score significantly.

NLCD · California Forest Observatory
08

Weather & fire-weather data

Real climate data is fetched for both California and Portugal adapters. The climate factor is a composite of Fire Weather Index percentile (45%), fire-season mean wind (30%), aridity index (15%), and special-event bonus (10%).

FWI · ERA5 · NOAA
09

Wind analysis

Prevailing wind direction and speed are pulled from historical records and current forecasts. Wind rose data feeds both the climate factor and the terrain exposure calculation.

Wind rose · Seasonal patterns
10

Road access & brigade response

Fire station locations are fetched from OpenStreetMap Overpass and routed to the property via OSRM. The estimated brigade-response ETA (≤5 min → score 10; no station → score 95) is a direct input to the risk model.

Overpass · OSRM
Phase 03

AI analysis

GPT-Vision analyses the property imagery and the RiskEngine combines all scored layers into a single 0–100 risk score calibrated per jurisdiction.

11

Defensible space analysis

GPT-Vision reviews the Street View and aerial imagery to assess vegetation clearance across three zones — Zone 0 (0–5 ft), Zone 1 (5–30 ft), and Zone 2 (30–100 ft) — around the structure.

GPT-Vision · Street View
12

Home hardening assessment

A second Vision pass evaluates structural fire resistance: roof material, vents, deck construction, eaves, and window type. A confidence score triggers human review if below the 0.7 threshold.

GPT-Vision · Admin review gate
13

Risk score computation

The RiskEngine applies the formula Risk = Hazard × Exposure × Vulnerability, combining all 11 scored layers into a single 0–100 score and a risk tier (Low / Moderate / High / Very High). The engine uses swappable regional adapters — California and Portugal are live.

RiskEngine · Regional adapters
Six layers of intelligence

Every score is built from real data

The RiskEngine combines 11 composable data layers. Each layer is scored independently and weighted by its contribution to wildfire risk at the parcel level.

Property & location

Parcel boundary, jurisdiction, address resolution, and regional adapter selection.

Foundation layer

Terrain & topography

Slope, elevation, aspect, and uphill wind-exposure score from USGS DEM data.

Hazard component

Wildfire hazard & fire history

Official FHSZ classification combined with proximity and recency of historical fire perimeters.

Hazard component

Vegetation & fuels

Fuel density and continuity at 30 m and 100 m radius from NLCD and Forest Observatory datasets.

Hazard component

Weather, wind & climate

Fire Weather Index, aridity, seasonal wind patterns, and special fire-weather event bonuses.

Hazard + exposure

Defensible space & home hardening

AI-assessed vegetation clearance across three zones plus structural fire-resistance scoring via GPT-Vision.

Vulnerability component
Phase 04

Report generation

The final step packages everything into a structured JSON payload and a four-page PDF report, then sends a notification email with a secure download link.

Step 14 — PDF report

A four-page PDF is rendered with the risk score, tier, per-layer breakdown, satellite imagery, and a prioritised mitigation plan. Stored securely in Supabase Storage with a signed URL.

Structured JSON output

Every report produces a machine-readable JSON payload with risk_score, risk_tier, per-layer scores, imagery URLs, and the full mitigation plan — ready for API consumers and portfolio tools.

Email notification

A transactional email via Resend delivers the "your report is ready" notification with a direct link to the dashboard download. The full pipeline completes in under 30 seconds for cached areas.

Ready to see your
risk score?

Enter your property address and get your first Satelife wildfire risk report — free, in under 30 seconds.

Get your free report →