Is the Phase 1 wildfire baseline stable enough for 2040 projections?
Phase 1 Inv 12 found that a 10% wildfire PM2.5 reduction ($1.65B fuel management, Portfolio C) avoids 1,447 deaths/yr — wildfire treatment is on the efficient frontier at every budget level. That ranking assumed the wildfire baseline is stationary through 2040.
Atmospheric science says otherwise. California summer VPD anomaly rose from −0.2 hPa (1980) to +1.9 hPa (2023). Burned area scales exponentially in VPD (Abatzoglou–Williams 2016). If the historical trend or CMIP6 projections continue, the 2050 wildfire regime diverges sharply from the baseline Phase 1 rested on — and the climate envelope may swamp the policy signal.
This investigation asks: under plausible climate trajectories from a 6-GCM CMIP6 subset, is the 10% wildfire policy still the right choice, and at what fidelity level does climate uncertainty start to dominate fuel-management uncertainty?
Five levels of climate–fire coupling
Each level adds a distinct piece of physics: from stationarity, to trend extrapolation, to GCM ensemble uncertainty, to coupled chemistry, to full Earth-system feedback.
Anchor: 2020 observed burned area 4.40 Macres at VPD anomaly +1.6 hPa. Abatzoglou–Williams coefficient β=0.33 per hPa. Smoke-days scale linearly with burned area (Aguilera et al. 2021).
The 6-GCM fan at 2050
Three SSP2-4.5 GCMs (moderate mitigation) and three SSP5-8.5 GCMs (high
emissions) from the NEX-GDDP-CMIP6 downscaled ensemble. The BA multiplier
is exp(0.33 × (VPD2050 − 1.6)); the 2050
deaths column is the multiplier times the Phase 1 baseline.
| GCM | SSP | VPD 2040 | VPD 2050 | BA 2050 | Deaths 2050 |
|---|---|---|---|---|---|
| GFDL-ESM4 | SSP245 | +2.6 hPa | +3.0 hPa | 1.59× | 22,971 |
| MPI-ESM1-2-HR | SSP245 | +2.3 hPa | +2.7 hPa | 1.44× | 20,805 |
| MRI-ESM2-0 | SSP245 | +2.2 hPa | +2.5 hPa | 1.35× | 19,477 |
| EC-Earth3 | SSP585 | +3.2 hPa | +4.1 hPa | 2.28× | 33,023 |
| CESM2 | SSP585 | +3.1 hPa | +3.9 hPa | 2.14× | 30,914 |
| ACCESS-CM2 | SSP585 | +2.9 hPa | +3.5 hPa | 1.87× | 27,091 |
NEX-GDDP-CMIP6 medians; LOCA2 and MACA downscaled products give ±0.3 hPa spread at 2050, roughly 10% additional deaths/yr uncertainty beyond what the fan shows.
California burned area 2000–2023
The historical record is the main evidence that stationarity (L1) is wrong. Burned area jumped >4× between 2000–2010 and 2018–2021. Williams et al. (2019) attribute the dominant share to anthropogenic warming via rising vapor-pressure deficit.
| Year | Burned area (acres) |
|---|---|
| 2000 | 295,000 |
| 2005 | 222,538 |
| 2010 | 108,742 |
| 2015 | 893,362 |
| 2018 | 1,975,086 |
| 2019 | 259,823 |
| 2020 | 4,397,809 |
| 2021 | 2,568,948 |
| 2022 | 363,939 |
| 2023 | 332,722 |
CALFIRE official records; 2020 value used as L2/L3 reference anchor.
Climate dominates policy noise — pair Portfolio C with monitors
The Phase 1 10% wildfire reduction (Portfolio C, $1.65B) still scales with the regime: at L3 median, 10% of 25,031 deaths = 2,503 lives/yr avoided, nearly 2× the Phase 1 number. But because the fan width (11,828 deaths) is 8.2× the policy wedge, the value of execution precision is dwarfed by the value of narrowing the climate corridor.
This sets up Inv 27: adaptive monitor placement. The biggest lever on fire-policy EVSI is not a better statistical fire model, but a denser PM2.5 monitoring network that discriminates between SSP2-4.5 and SSP5-8.5 trajectories as they unfold. A commitment to “10% wildfire reduction” without a commitment to adapt as climate reveals itself leaves most of the EVSI on the table.
Recommendation: Commit to Portfolio C at $1.65B; add $30–50M/yr for a PM2.5 + smoke-tracer monitoring program (Inv 27) that can discriminate the climate fan by 2035. Couple to Inv 21 hierarchical CRF so monitor data updates both the exposure field and the concentration-response function.
Where this model could be wrong
- Abatzoglou–Williams was fit on western US forest; CA chaparral/oak-woodland may respond with a smaller β (0.25–0.35/hPa bracket).
- Downscaling choice (NEX-GDDP vs LOCA2 vs MACA) adds ±0.3 hPa spread at 2050; we use NEX-GDDP medians only.
- L4 SOA uplift is uniform; real secondary-aerosol formation is fuel-type-dependent.
- L5 earth-system feedback is literature estimate; sign is contested (Heilmann et al. 2023 argues vegetation dieback could stabilize).
- Population growth and aging held at 2024 levels; accounting for CA's aging population adds ~15% deaths/µg uplift at 2050.
Sources: Abatzoglou & Williams 2016 (PNAS), Williams et al. 2019 (Earth's Future), Goss et al. 2020 (ERL), Aguilera et al. 2021 (Nature Comm), Hurteau 2019 (Earth's Future), NASA NEX-GDDP-CMIP6, CALFIRE historical records.