California Freight Cleanup → Investigation 7-3
Does wildfire load flip the portfolio ranking?
Thirteen electrification scenarios tested against every observed California wildfire year from 2019 to 2023, from a mild season adding 0.13 µg/m³ of smoke statewide to the catastrophic 2020 season at 3.16 µg/m³. The portfolio ranking held in every case. In 2020, wildfire added 601 excess deaths while the best electrification portfolio avoided 164 — a stark illustration that wildfire prevention is a larger lever in bad fire years, but it doesn’t change which electrification option to choose.
The decision
The California Freight Cleanup portfolio recommends electrification policy allocations based on anthropogenic PM2.5 baselines that exclude wildfire smoke—a deliberate choice because CEC program design addresses sources regulators can actually affect. A reviewer can reasonably ask: does a catastrophic fire year change which policy is optimal? If the portfolio ranking is sensitive to wildfire load, the program may need to account for year-to-year fire variability in its investment sequencing.
Investigation 7-3 stress-tests this directly. Four wildfire PM2.5 scenarios from the Childs V2.0 observational product (2019–2023) are applied to the full 13-scenario electrification menu—bracketing the full observed range of California wildfire PM2.5 from mild to catastrophic.
Methodology
Wildfire PM2.5 increments from four named Childs V2 scenarios are
applied as an additive overlay to the anthropogenic ISRM baseline:
pm25_total = pm25_anthro + pm25_wildfire. The additive assumption is
defensible because wildfire smoke is spatially and temporally distinct from the
combustion sources that electrification policies address.
The same Di et al. 2017 log-linear CRF (β = 0.007046
per µg/m³) used in Investigation 1-1 and Investigation 1-2 is applied to the
total PM2.5 field. Deaths avoided by a policy under wildfire = baseline-with-wildfire
deaths minus policy-with-wildfire deaths. The marginal value ratio (MVR)—deaths
avoided with wildfire / deaths avoided without—tracks the log-linear compression
effect: because wildfire raises both baseline and policy PM2.5 by the
same scalar, the exponential term is multiplied by exp(−β×wf) < 1,
mechanically suppressing computed deaths avoided. MVR is a model-behavior diagnostic,
not a decision variable.
All 13 scenarios (T1–T5 transport; B1–B4 building; four pre-selected combined portfolios) are re-evaluated under each wildfire increment at the 2035 target year. Ranking is determined by deaths avoided in descending order; a ranking change is a boolean flag comparing the ordered list with versus without wildfire.
Findings
| Scenario | WF PM2.5 (µg/m³) | Excess deaths | MVR | Ranking flipped |
|---|---|---|---|---|
| Mild (2019, pre-Dixie) | 0.13 | 27 | 0.999 | No |
| Normal (2019–2023 5-yr mean) | 1.05 | 200 | 0.995 | No |
| Extreme 2020 (August Complex) | 3.16 | 601 | 0.983 | No |
| Extreme 2021 (Dixie/Caldor) | 1.24 | 227 | 0.994 | No |
Under every observed California wildfire year from 2019 to 2023, the optimal portfolio remains T2_accelerated+B2_accelerated (~166 deaths avoided, ranging 164–167 across scenarios), followed by T2_accelerated alone (~130), T4_equity (~119), and the T5+B1 / T1+B1 tie (~110). The five-position ranking is invariant across the full wildfire-load range. MVR floors at 0.983 in the 2020 catastrophic year—the portfolio still captures at least 98% of its deaths-avoided benefit even in the worst observed fire season.
The contrast is stark: in 2020, wildfire smoke added 601 excess deaths while the best-available electrification portfolio avoided 164. Wildfire prevention is quantitatively a larger mortality lever in catastrophic years—see Investigation 4-3—but it does not change which electrification policy is optimal when the program budget is fixed to electrification.
Caveats
- Wildfire PM2.5 is uniform within each region. A single per-region scalar is applied to every cell. Actual smoke concentrations vary orders-of-magnitude within regions depending on fire proximity, topography, and meteorology. The smoothing reduces spatial contrast that might otherwise produce ranking changes.
- Wildfire PM2.5 treated as toxicologically equivalent to anthropogenic PM2.5. Wildfire smoke has a different per-µg health burden (higher organic carbon fraction) in some published literature. A wildfire-specific CRF could materially widen the excess-death counts.
- No Monte Carlo over wildfire scenarios. Each scenario is a deterministic point estimate from the Childs V2 product. Spatial and temporal variability within the ensemble is not propagated. Inv 30 (polynomial chaos expansion) is the path for probabilistic wildfire UQ.
- CRF is Di-only. Investigation 7-3 uses the Di point estimate directly, not the Investigation 6-3 California hierarchical posterior. Ranking robustness is therefore conditional on the Di structural choice.
- Rank-4/rank-5 tie. T5+B1 and T1+B1 are computationally indistinguishable at all wildfire loads (110.1 deaths avoided in mild, 108.3 in extreme 2020). Any narrative ranking must collapse them.
Provenance
| File | Link | Purpose |
|---|---|---|
results.json | Per-scenario, per-wildfire deaths avoided; ranking tables; MVR values | |
analysis.md | Mechanical readout with diff table; ranking matrix top-5 | |
scenario.md | Sticky methodology; MVR explanation; Investigation 4-3 / Investigation 7-2 cross-link notes | |
ranking_matrix.json | Per-policy rank position under each wildfire scenario |
Run provenance: generated 2026-05-01T21:27:48; results.json
sha256 23ab2e3a308e. Upstream: Childs V2.0 BETA wildfire
scenarios (sha256 22f52721a6cd) from
data/processed/wildfire_robustness_scenarios.json.