A transparent composite, not a black box.

The formula

Each factor is normalized to a 0–100 value, then combined by weighted mean:

ember_index = Σ (factor.value × factor.weight)
              ─────────────────────────────────
                    Σ factor.weight

baseline = same formula using only the 7 static factors
live_delta = ember_index − baseline

Static factors sum to 0.60; live factors sum to 0.40. A parcel with a poor baseline (Very High FHSZ, steep S-facing slope, cured fuel) will read high even on a calm day. A moderate parcel can jump 20+ points during a Red Flag / dry downslope wind event.

Normalization

Each factor is scaled into a 0–100 value using a domain-specific curve tied to the underlying physics or agency rating — not a percentile rank across parcels. FHSZ classes map to fixed anchor points (Moderate → 40, High → 65, Very High → 90). Slope is converted via the Rothermel rate-of-spread multiplier and clamped to 0–100. NDVI is linearly rescaled between 0.50 (green, → 0) and 0.15 (cured, → 100). Wind blends gust (0 mph → 0, 60+ mph → 100) with the dry-downslope-wind score and an RH penalty. Active-fire and hotspot factors combine proximity, wind alignment, size, and containment into a 0–100 composite. The upshot: a "70" on fuel state or wind means the same physical condition in Malibu as in Altadena.

How weights were set

Static weights reflect each factor's established role in fire-agency and insurer risk models. CAL FIRE FHSZ carries the largest static weight (0.15) because it is the only authoritative fuel/terrain hazard rating. CPUC HFTD (0.09) captures utility-driven ignition risk, which recent postmortems (Camp, Kincade, Woolsey) have shown to be a leading real-world ignition source independent of fuel/terrain hazard. Slope, fuel state, and ember drift take mid-weights (0.07–0.10) because they materially change fire behavior but are inferred from remote-sensing and terrain data. Live weights concentrate on wind/RH (0.15) and active-fire proximity (0.12), since those inputs drive the fastest short-term swings in real risk. A self-reported home-hardening credit of up to 32 points is subtracted from both the baseline and the live composite, and the whole result is clamped to 0–100. The weights are engineering choices, not fitted coefficients — they are chosen to keep any single factor from dominating and to keep live conditions capable of moving the score by 20+ points without overwhelming the baseline.

The eleven factors

These labels, weights, and static/live types match the in-app factor breakdown line for line. One caveat: utility ignition risk requires a real CPUC High Fire Threat District reading. Outside California the CPUC has no equivalent map, so that factor is omitted rather than guessed and its 0.09 weight is redistributed across the remaining ten by the weight-sum normalization above.

Bands