Fire-aware Pareto framework

An optimization layer built over USDA's burn probability raster. Combines per-cluster fire risk with the cost layer in a weighted objective that lets operators or policymakers explicitly tune the tradeoff.

#Overview

Unlike the harvest cost surrogate and the transport circuity model, the fire-aware framework is not a new prediction layer. It is an optimization layer over data FRED consumes from elsewhere: the USDA FSim 2024 burn probability raster.

FRED's contribution is the framework itself, the explicit weighting parameter, and the analysis of the leverage curve across California cluster supply.

#USDA FSim as the input

FSim is the USDA Forest Service wildfire simulator. It produces a 30-meter-resolution raster of annual burn probability across the contiguous United States, based on millions of stochastic fire simulations under historical climate, weather, and fuel conditions.

FRED uses the 2024 release of the dataset, area-weighted to each cluster polygon. The result is a per-cluster annual burn probability ranging from approximately 0 to 0.08.

NOTE
FSim's burn probability is a long-run statistical estimate, not a forecast for any specific year. Treat it as a comparative measure across clusters, not an absolute prediction of imminent ignition.

#The objective function

With per-cluster burn probability available, FRED frames procurement as a weighted optimization. The objective for each cluster, given a weighting parameter alpha:

NOTE
score = (1 - alpha) · cost_norm - alpha · fire_norm

where:

  • cost_norm is the cluster's harvest cost normalized to the 5th-95th percentile range across all California clusters.
  • fire_norm is the cluster's burn probability normalized the same way.
  • alpha is the user-specified weighting (0 = cost only, 1 = fire only).

Normalization uses percentile bounds rather than raw min-max to prevent statewide outliers (a single very-low-cost cluster, for example) from compressing the distribution. This is a small technical detail with material effect on the leverage analysis.

#Computing the Pareto curve

For a given facility and demand, FRED computes the Pareto curve by sweeping alpha from 0 to 1, solving the weighted procurement optimization at each value, and recording the resulting cost and fire-risk reduction. The default sweep uses 21 alpha values non-uniformly spaced to give more resolution near the leverage knee.

Each alpha point produces a procurement plan: a set of clusters selected to meet demand. The plan's weighted-average cost gives the cost axis; its weighted-average burn probability gives the fire axis. The curve is monotonic and concave by construction.

#The leverage knee

The leverage knee is the point on the Pareto curve where the leverage ratio (fire reduction percentage divided by cost premium percentage) is maximized. Empirically across California, this knee sits near alpha = 0.05, with leverage ratio approximately 5.8x.

Specifically at the knee:

  • Cost premium versus the cost-only baseline: about +20.5 percent
  • Procurement-weighted fire reduction: about 118.9 percent
  • Leverage ratio: roughly 5.8x

Past the knee, leverage drops rapidly. Pure fire optimization (alpha = 1) gives only marginally better fire reduction than alpha = 0.3 while costing substantially more.

#The structural ceiling

A counterintuitive but important finding: even at maximum fire weighting (alpha = 1), approximately 57 percent of procurement stays in the Very Low fire-risk tier.

This is a structural ceiling, not a modeling limit. It exists because California's biomass supply and high-fire-risk areas overlap only partially. Some of the most fire-prone parts of the state have low biomass density (chaparral, oak woodland). Some of the densest biomass is in lower-fire-risk montane forests. When alpha pushes procurement toward fire-prone areas, the available cluster pool shrinks rapidly.

The ceiling matters for policy design. Programs that incentivize fire-risk biomass procurement are bounded by where biomass and fire risk geographically coincide. FRED quantifies that bound; it is informative for shaping realistic incentive structures.

TIP
The full leverage curve and ceiling analysis is documented in Chapter 6 of the thesis, with per-county and per-watershed breakdowns for researchers and policymakers needing geographic granularity.