California Freight Cleanup → Investigation 1-2

Which building electrification scenario wins under full uncertainty?

Four scenarios for phasing out residential gas in California: current code enforcement, an aggressive $2B retrofit, new-construction-only, and a cooking-first urban equity variant. The health benefits are real but small relative to uncertainty, and the cost comparison makes current enforcement the clear baseline.

The decision

California’s building sector burns roughly 700 million therms of natural gas per year in residential end-uses. That combustion is a diffuse PM_2.5 and NO₂ source spread across every neighborhood, compounding traffic and industrial pollution. The concrete decision: stay with current-code enforcement (B1), fund an aggressive $2B retrofit (B2), restrict new construction only and let the stock turn over (B3), or target gas cooking appliances as an urban co-benefit lever concentrated in DAC neighborhoods (B4). Because deaths-averted magnitudes are small relative to the P5–P95 uncertainty band, the analysis must determine whether scenario differences exceed the noise floor before recommending a more expensive tier.

Methodology

The same 32-million-resident population grid (21,164 cells, 22.9% DAC share) and health infrastructure used in Investigation 5 is applied here with one key difference: the emission-fraction vectors target the residential components of the ISRM sector-marginal fields rather than on-road. Each building scenario produces a 16-dimensional vector for four target years (2025, 2030, 2035, 2040). The residential components (indices 5–9, one per region) encode the fraction of gas combustion remaining after appliance replacement, capped at 10% minimum. Regional adoption multipliers reflect faster uptake in Bay Area and LA Basin than in SJV and rural California.

The gas-replacement rates by scenario are:

B1 (current policy): 5%, 12%, 22%, 35% at 2025/2030/2035/2040
B2 (aggressive, $2B): 10%, 28%, 50%, 70%
B3 (new construction only): 2%, 6%, 12%, 20%
B4 (cooking-first, $0.5B): same statewide rates as B1 with 1.35× multiplier in Bay Area, 1.30× in LA Basin, 0.65× in SJV

A grid-feedback term captures the emissions consequence of increased electricity demand as buildings switch from gas to electric appliances. The marginal gas fraction of California’s grid declines from 60% in 2025 to 15% in 2040 per SB 100 targets; this schedule is parametric rather than dispatch-modeled.

Ten thousand shared Monte Carlo draws (seed 43) are run across all four scenarios at each target year. CRF and VSL distributions are identical to Investigation 5. Ozone disbenefit is computed deterministically for each scenario–year pair via the same regime-aware scenario_o3 function. VOI uses the rfaq.voi.building_decision module with EVPI computed over the four B1–B4 alternatives at 2035.

Findings

Building scenarios at 2035: PM_2.5, ozone, and net deaths avoided
Scenario	Deaths avoided vs. baseline	O₃ offset %	Net-neg. cells (pop.)	Cost ($B)	VOI opt. prob.
B1 — Current enforcement	14582	15.5%	249 (178K)	$0	98.9%
B2 — Aggressive, $2B	82	15.7%	252 (178K)	$2.0	1.1%
B3 — New construction only	22	15.5%	249 (178K)	$0	0.0%
B4 — Cooking-first, $0.5B	39	16.6%	264 (185K)	$0.5	0.0%

The BAU baseline at 2035 is 14,582 deaths/yr (P5 7,747–P95 22,740). All four scenarios produce real but small benefits—22 to 82 deaths avoided against a 15,000-death uncertainty band. The VOI result is correspondingly decisive: B1 is optimal in 98.9% of draws, EVPI $5.5M. The quantitative answer to “should we spend $2B on aggressive retrofit?”: no—unless indoor air quality co-benefits from gas cooking (Inv 19) materially change the net-benefit calculation.

The ozone offset for building electrification (15–17%) is substantially lower than for transport (50–93%). Residential NOx is more diffuse, emitted in a different diurnal pattern than on-road traffic—so the regime-shift signal is weaker. Fewer than 300 grid cells (178K residents) reach net-negative territory under any building scenario. B4’s slightly higher ozone offset (16.6% vs. B1’s 15.5%) reflects its urban-concentrated adoption multipliers pushing more gas-combustion reduction into VOC-limited LA Basin and Bay Area cells. DAC share does not shift meaningfully across B1–B4; building electrification’s DAC impact runs through the indoor-air pathway in Inv 19, not the outdoor-concentration model here.

Caveats

The baseline is frozen across all four target years. Deaths-avoided figures represent reductions from a static counterfactual; they overstate each scenario’s marginal contribution in later years if baseline pollution improves autonomously through economy-wide decarbonization or fleet turnover.
Scenario vectors cannot separate end-use within residential. B4 (cooking-first) is modeled only as a spatially-concentrated adoption pattern, not as a true reduction targeting only cooking appliances. Cooking accounts for ~9.8% of residential gas (RECS); the B4 outdoor-air benefits reported here overstate what a cooking-only program would achieve.
Grid feedback is parametric, not dispatch-modeled. The marginal gas fraction schedule (60% → 15%) follows SB 100 targets; actual marginal generation in high-electrification years depends on grid capacity, storage, and demand response. Overestimating marginal gas fraction inflates grid-feedback NOx and suppresses net benefits.
Ozone disbenefit is deterministic. The ozone offset carries no confidence interval even though O₃ CRF uncertainty is comparable to PM_2.5. Net deaths-avoided uncertainty is understated as a result. scenario_o3 also does not yet implement Sillman (1995) indicator thresholds; regime classifications near the VOC/NOx boundary may be incorrect. PM_2.5 figures are unaffected.
No indoor air quality pathways. The framework captures outdoor PM_2.5, O₃, and NO₂ only. Indoor NO₂ from gas cooking—the primary rationale for B4—is outside this model. Inv 19 adds that pathway as a post-hoc multiplier.
VOI uses undiscounted net benefit. EVPI is computed at 2035 without discounting program costs or health benefits. A discounted calculation would likely strengthen B1’s position further.

Provenance

File	Link	Purpose
`results.json`	Full MC summary, ozone disbenefit, VOI, all scenarios × years
`analysis.md`	Mechanical readout, diff-from-previous-run table, stubs for human interpretation
`scenario.md`	Sticky methodology, key anchors, downstream dependency map

Run provenance: generated 2026-05-04T07:37:14; results.json sha256 99dcff1b4dfb. Five downstream investigations read this investigation’s outputs: Investigation 1-3 (combined), Inv 11 (CRF-conditional), Investigation 4-3 (wildfire vs. electrification), Investigation M-1 (portfolio frontier), Inv 19 (indoor-air coupling).