Wildfire Insurance → Investigation 05

What Drives the Answer?

We varied each parameter one at a time around the baseline and measured the impact on risk classification and expected annual loss. Some parameters barely matter. Others change everything.

10
Parameters Tested
30
MC Draws Each
133%
Baseline EAL Divergence
Sensitivity — Expected Annual Loss

Which Parameters Move the EAL?

We ran a one-at-a-time sensitivity analysis: vary each parameter from its baseline while holding everything else fixed. The chart below shows how much the Expected Annual Loss changes for each perturbation — excluding the damage ratio, which dominates so completely it obscures the rest.

EAL Impact by Parameter (% Change from Baseline)

Model: 200-draw sensitivity run per parameter perturbation. Baseline = Kincade WUI, median weather, 4 ignitions, threshold 0.25, damage ratio 0.30.

Off the Chart: Damage Ratio
+1,380%
Damage Ratio 1.00
+640%
Damage Ratio 0.50

The damage ratio — the fraction of a structure’s value destroyed when fire arrives — overwhelms every other parameter. It isn’t shown in the main chart because it would compress everything else to zero. At 1.00 (total loss), EAL rises 1,380% above baseline. At 0.50, it rises 640%. The next-largest parameter (humidity) moves EAL by 16%.

Finding
The damage ratio dominates everything. Whether 50% or 100% of a burned cell’s value is destroyed changes the EAL by 640–1,380%. This is the most important assumption in any catastrophe model — and it’s the one that requires structure-level data the Rothermel model doesn’t have.
Sensitivity — Risk Classification

Which Parameters Move the Classification?

Loss isn’t the only output that matters. Insurers care about how many properties change risk tiers. We measured the change in the L0→L3 reclassification rate for each parameter perturbation.

Reclassification Delta by Parameter (Percentage Points)
Finding
Humidity is the strongest driver of classification. Wetter conditions reduce reclassification by 21 percentage points — meaning more properties stay in the same tier regardless of fidelity. This makes physical sense: wet fuel doesn’t burn, so the fire model produces less spatial variation.

Reclassification delta = change in fraction of properties reclassified from L0 to L3 relative to baseline. Positive = more reclassification, negative = fewer properties change tier.

Implications

What This Means for Insurers

The sensitivity analysis reveals two tiers of importance:

  • 01 The damage ratio is the whole game. Everything else is noise compared to how much of a structure’s value is destroyed when fire arrives. This is exactly the parameter that zone-based models can’t capture and that structure-level data would improve.
  • 02 Weather conditions drive classification more than model design choices. Humidity matters more than ignition count, threshold choice, or wind speed. An insurer investing in model improvement should prioritize weather data quality over model complexity.

The burn probability threshold (where you draw the line between HIGH and EXTREME) changes the EAL by less than 1%. The classification boundary is not a decision-sensitive parameter — which means the fidelity findings are robust to this arbitrary choice.

← Previous Q4: Where Does the Model Fail? Next → Q6: Cross-Fire Comparison