The data behind FireRisk.ai
Every figure in our maps, reports, and briefings is grounded in official, interagency wildfire data — the same feeds fire managers and emergency services rely on. We don’t guess, infer, or let AI invent numbers. A shared numeric validator rejects any AI report, briefing, or threat read that contains a figure we can’t trace to an official source. That’s how you know what you’re reading is true.
Our anti-fabrication guarantee
When we write an AI daily update, we hand the model a fact sheet of official figures and forbid it from using anything else — no web, no prior knowledge, only data we verified. Then every briefing passes a numeric validator before it can publish:
- ✓ Every 3+ digit number (acreage, personnel, cost, year) is cross-checked against the fact sheet.
- ✓ If even one number doesn’t trace to an official figure, we reject the entire briefing and publish nothing rather than something slightly wrong.
- ✓ Casualty, structure-loss, and evacuation counts can never appear unless an official source states them — a hard guard the prompt can’t override.
Most automated fire sites paraphrase whatever they scrape. We do the opposite: when in doubt, we say less — and only what’s sourced.
Where our data comes from
Eight official feeds, plus clearly-labeled news for context. Each refreshes on its own cadence. (For how our 0–100 property risk score is built from federal hazard models, see the risk-score methodology.)
Every active U.S. wildfire: name, acres, % contained, discovery date, cause, personnel, cost, primary fuel, county/state, coordinates — plus official perimeters on the map.
Refresh: Queried live; cached at most 2 min for threat checks, ~10 min on incident pages. · Why: The authoritative interagency source incident commanders themselves use.
Past fire perimeters (~2010–present) for burn-scar history and repeat-fire context.
Refresh: Historical; changes infrequently. · Why: Shows whether an area has burned before and how close.
Red Flag Warnings & Fire Weather Watches, plus gridpoint forecasts: temperature, relative humidity, wind speed and gusts, and a multi-day outlook at the fire.
Refresh: Live; ~5 s timeout per query. · Why: The official source for fire weather — what drives rapid spread.
Ground-monitor Air Quality Index (AQI) and PM2.5 near a fire or address. Falls back to keyless OpenAQ when needed — never a false “clean.”
Refresh: Live; no caching. · Why: Smoke is the most widespread wildfire health hazard.
Official incident-team publications (terrain, tactics, closures) and incident photos — used verbatim, entity-decoded, never paraphrased into new claims.
Refresh: ~10 min cache. · Why: The crew’s own words, straight from the source.
The national 1–5 mobilization level from the National Interagency Coordination Center.
Refresh: ~1 h cache; omitted entirely if unavailable. · Why: Context for how stretched national firefighting resources are.
Near-real-time satellite heat detections (hotspots) filtered to likely fire, shown on the map.
Refresh: ~15 min. · Why: Independent satellite confirmation between official updates.
California’s own incident database and year-to-date season totals — labeled precisely as CAL FIRE’s tracked major incidents, never relabeled as “all wildfires.”
Refresh: Live. · Why: California-specific detail NIFC doesn’t always carry.
Headlines that name a specific fire — used only for context and citations, never as a source of numbers.
Refresh: ~5 min. · Why: Local reporting sometimes surfaces closures before feeds catch up — shown as secondary, with the outlet named.
What the numbers really mean
The same figures shown elsewhere are easy to misread. Here’s how we interpret them.
Acreage = perimeter, not verified burn
NIFC acreage is the area inside the current estimated perimeter — it can include unburned islands and is revised up or down as crews map the fire. We always pair it with the change since the last report.
Containment % = perimeter secured, not “out”
Containment is the share of the perimeter held with line, water, or retardant. A fire can still burn inside secured line. 100% contained ≠ extinguished — we never call a fire “out” or “controlled.”
Personnel = everyone assigned
The personnel figure includes firefighters, aircraft crews, logistics, command, and medics — not only people on the line. We report it as NIFC publishes it.
Perimeter ≠ live flame front
A perimeter is everywhere a fire has burned, not where it is actively burning right now. Active flame is inferred from satellite heat, not the perimeter polygon.
Official vs. unofficial
Government feeds (NIFC, NWS, AirNow, NASA, InciWeb) get the highest trust. News and “from the web” items are clearly labeled, cited, and never used to generate numbers.
Wildfire vs. prescribed fire
We track wildfires (unplanned). Planned prescribed burns are filtered out of alerting so a controlled burn never triggers a false alarm.
What we deliberately leave out
Restraint is a feature. We exclude anything we can’t ground — and tell you where to get it officially.
Casualty, injury & missing counts
NIFC doesn’t release these during active incidents, and news estimates are unverified. Our validator would reject them anyway — so we never publish them.
Real-time “homes threatened/destroyed”
Structure counts are preliminary and come late in damage surveys. We link to county emergency management instead of repeating a moving estimate.
Specific evacuation orders & road closures
These change by the hour and are a county/CalTrans call. We point you to the official source for your exact address rather than risk a stale order.
Spread predictions
We don’t forecast which town a fire will reach or when — that needs incident meteorology. We report the current state and trend only.
Social-media rumor
No viral claims. Our news feed cites named outlets that corroborate official data, never replace it.
Freshness & how we handle stale data
Live threat checks query NIFC, NWS, and AirNow on every request (with ~4–5 second timeouts so a page never hangs). The national pulse refreshes about every 5 minutes; per-fire and per-state daily reports regenerate every few hours; satellite hotspots about every 15 minutes. If a feed is unreachable, we omit that piece rather than show stale or guessed data — and if an AI briefing can’t be grounded, we publish nothing at all.
Awareness only — not an emergency service
FireRisk.ai provides situational awareness from official data. We are not a replacement for official emergency alerts, an evacuation-order source, or a real-time incident-command resource. Data can lag real conditions, and a fire seen on local news may not yet appear in NIFC. If you are in or near a fire zone, sign up for your county’s emergency alerts, check your county emergency-management page and InciWeb, follow Watch Duty and local authorities — and if told to evacuate, leave immediately. In an emergency, call 911.