California Freight Cleanup → Investigation 4-2
Does cutting freight NOx lower ozone or raise it?
Standard model: -1.84 ppb (benefit) • Regime-aware: +2.10 ppb (disbenefit) • AQS sign confirmed: 3/3 basins
The widely-used ISRM framework is calibrated on national annual-mean linear sensitivities and does not encode VOC/NOx chemistry regimes. In VOC-limited basins like the LA Basin, this produces a directional error: cut NOx, and ambient ozone goes up, not down. We climb five fidelity levels to show where the error enters, validate the sign against AQS monitoring data, and correct the Phase 1 portfolio co-benefit claim.
Decision context
Phase 1 of this study used ISRM linear response factors to estimate ozone co-benefits from NOx reductions in the diesel freight scenario. The ISRM does not encode VOC/NOx chemistry regimes. In VOC-limited basins, this produces a directional error: the model predicts an ozone reduction when the atmospheric chemistry produces an ozone increase.
Adding chemistry fidelity changes the sign for the LA Basin — not just the scale. For the San Joaquin Valley and Sacramento, only the scale shifts; both retain an ozone benefit at higher fidelity. Because the LA Basin holds 72% of the three-basin population, its sign flip controls the aggregate portfolio outcome.
The five-rung fidelity ladder
We climb five fidelity levels for each of the three major California air basins under the canonical T2 scenario (20% NOx reduction, 5% VOC co-benefit):
| Level | Method | Key limitation |
|---|---|---|
| L1 | Photostationary equilibrium (Seinfeld-Pandis 2016, Ch. 6) | No VOC/NOx nonlinearity; all basins show large spurious reductions (−7.8 to −12.4 ppb) |
| L2 | ISRM linear (Goodkind et al. 2019, Heo et al. 2017) | Annual-mean linear sensitivity only; regime-blind. This is the fidelity used in Phase 1. |
| L3 | InMAP reduced-complexity CTM (Tessum et al. 2017) | Adds spatial resolution over ISRM; shares L2 annual-mean linearity limitation. L2→L3 gap is negligible (<0.2 ppb across all basins). |
| L4 | CMAQ episodic stub with regime-aware isopleth correction (Jin et al. 2020) | Piecewise-linear isopleth correction; not hour-resolved. Sign validated against AQS. Policy-relevant fidelity. |
| L5 | WRF-Chem meteorological stochastic envelope | Stochastic draw, not real WRF-Chem output; represents met-coupling uncertainty conceptually. Not fused (stochastic only). |
Levels L1–L4 are fused via Kennedy-O’Hagan AR1 to produce a posterior mean and sigma per basin. L5 is excluded because it is stochastic, not a deterministic model level. The fusion posteriors are informational; downstream investigations consume the L4 scenario deltas directly.
Basin results
| Basin | VOC/NOx | Regime | L2 delta (ppb) | L4 delta (ppb) | Gap | Sign flip | Phase-1 error |
|---|---|---|---|---|---|---|---|
| LA Basin (South Coast) | 3.2 | VOC-limited | -1.84 | +2.10 | +3.94 | YES | 214% |
| San Joaquin Valley | 11.5 | NOx-limited | -2.36 | -3.40 | −1.04 | no | 44% |
| Sacramento Valley | 6.8 | transitional | -1.70 | -0.98 | +0.72 | no | 43% |
The LA Basin holds 17.9 M of the 24.8 M three-basin population (72%), so the LA sign flip controls the portfolio aggregate. Population-weighted ozone delta: -1.92 ppb (L2, benefit) to +0.82 ppb (L4, disbenefit).
AQS validation
We validate L4 sign predictions against observed 2023→2024 design-value
deltas from 90 California AQS monitors (version 2 pre-generated files, EPA AQS
daily_44201_2024.zip, anchored 2026-04-25, sha256
53e19767a0e6).
| Basin | Monitors | Obs median (ppb) | L4 delta (ppb) | L4 sign match | Bias (ppb) |
|---|---|---|---|---|---|
| LA Basin | 31 | +1.67 | +2.10 | YES | +0.44 |
| San Joaquin Valley | 39 | -0.33 | -3.40 | YES | -3.07 (gate fail) |
| Sacramento Valley | 20 | -2.67 | -0.98 | YES | +1.69 |
Pre-registered gates: sign agreement (≥ 2/3 basins) and magnitude agreement (within ±1.5 ppb for all basins). Sign agreement: 3/3 pass. Magnitude gate: FAIL at SJV (bias −3.07 ppb). The validated claim is sign agreement, not absolute magnitude. Any downstream use of the SJV L4 magnitude should treat it as directional-only.
Cobenefit monetization and Phase 1 correction
Regime-aware ozone deltas monetized via Turner et al. 2016 CRF and $11.6M VSL:
| Field | Value |
|---|---|
| Phase 1 (L2 linear) co-benefit | +$7,273.8M (health benefit) |
| Phase 2 (L4 regime-aware) co-benefit | −$3,132.8M (health disbenefit) |
| Co-benefit lost (modeling-error correction) | $10,406.6M |
The $10.4B is a modeling-error correction, not an estimate of actual harm imposed. It represents the portion of the Phase 1 portfolio benefit claim that disappears when ISRM linear approximation is replaced by regime-aware chemistry. Any portfolio benefit summary that used the Phase 1 ozone figure must subtract $10.4B from the ozone channel and add a disbenefit of $3.1B in its place.
Findings
Phase 1 ozone cobenefit is directionally wrong for the largest basin
The ISRM linear framework predicts an LA Basin ozone benefit from a 20% NOx cut. L4 regime-aware chemistry predicts the opposite—and AQS monitoring confirms the sign (31 monitors, +1.67 ppb observed vs. +2.10 ppb modeled). The 17.9M-person LA Basin is not a co-benefit recipient under T2; it bears additional ozone burden. Any portfolio ranking that added Phase 1 ozone co-benefits for LA is directionally wrong.
The fidelity gap is chemistry-driven, not resolution-driven
L2 and L3 agree to within 0.16 ppb despite using different spatial resolutions and source-receptor matrices. The 3.94 ppb LA gap is explained almost entirely by the L3→L4 regime correction. For LA Basin ozone response, the right investment is regime classification—not higher spatial resolution.
Equity conclusions from Phase 1 are compromised for the LA DAC population
Phase 1 treated LA Basin ozone as a benefit. The LA Basin holds California’s highest-density DAC populations. Any DAC equity analysis that attributed an ozone co-benefit to LA under T2 must be reversed: the LA DAC population receives an ozone disbenefit. Downstream equity investigations must use the L4 signed delta (+2.10 ppb LA), not the Phase 1 linear estimate (−1.84 ppb).
SJV and Sacramento retain benefit, but SJV magnitude is not validated
SJV (NOx-limited, VOC/NOx = 11.5) shows a larger benefit at L4 than at L2 (-3.40 vs. -2.36 ppb); the NOx-limited isopleth amplifies the response. Sacramento (transitional, VOC/NOx = 6.8) narrows but stays negative (-0.98 vs. -1.70 ppb). Both basins retain benefit direction at all fidelity levels. The SJV L4 magnitude fails the pre-registered gate (bias -3.07 ppb); direction only is validated for SJV.
Caveats
- L4 is a piecewise-linear isopleth stub, not a real CMAQ episodic run. The sign flip is correct and validated against AQS trends; the magnitude is a parametric approximation. Real CMAQ episodic output would resolve hour-by-hour nonlinearity the piecewise-linear correction cannot capture.
- VOC reactivity is lumped. The model does not distinguish anthropogenic from biogenic VOC. On high-temperature days with elevated isoprene, the effective LA Basin VOC/NOx ratio could shift above 4 into a transitional regime. The sign-flip conclusion is robust; the magnitude is not.
- L5 meteorological envelope is stochastic, not WRF-Chem. Met-coupling uncertainty (5–9% of the delta) is conceptually bounded but not empirically validated. L5 is informational only; no downstream investigation consumes it.
- SJV magnitude gate failure is a pre-registered finding. SJV L4 bias of −3.07 ppb against −0.33 ppb observed exceeds the ±1.5 ppb gate—a consequence of SJV’s episodic ozone character (harvest-burn events, inversions) and the annual vs. episodic comparison mismatch. Sign agreement is the validated claim; magnitude requires episodic CMAQ runs.
- KO-AR1 fusion posteriors are pulled low by L1. The fusion posteriors (LA 66.7 ppb, SJV/Sac ~52.5 ppb) sit well below the L4 values because the photostationary L1 extreme anchors the low end of the AR1 prior. These posteriors are informational about cross-rung variance; the L4 scenario deltas are the correct policy inputs.
-
Investigation 1-1 / 6 / 7 ozone code path not yet Sillman-updated.
The
scenario_o3code path inrfaq.analysis.concentration_modelused by Investigation 1-1, 6, and 7 does not yet incorporate the Sillman (1995) thresholds applied here. Element 1 ozone offsets are directionally correct but not calibrated against the Sillman boundary. This is a documented cleanup item; Investigation 4-1 carries the policy-facing number.
Provenance
| Field | Value |
|---|---|
| Investigation | 18: Atmospheric chemistry MFMC overlay (ozone) |
| Run timestamp | 2026-05-01T21:27:49 |
| Script sha256 | 6bd352eea99acd1d |
| Results sha256 | 28d5083ca123 |
| Last human review | 2026-04-27 |
| AQS anchor |
data/processed/v2_carb_ozone_dv_2024.npz
(sha256 53e19767a0e6, acquired 2026-04-25)
|
| Key references |
Sillman S. (1995), J. Geophys. Res. 100(D7):14175–14188 (NOy/HNO3 indicator thresholds — method anchor); Sillman S. (1999), Atmos Environ 33:1821–1845 (regime classification — broader review);
Goodkind AL et al. (2019), PNAS 116(18):8775–8780 (ISRM L2);
Tessum CW, Hill JD, Marshall JD (2017), “InMAP: A model for air pollution interventions.” PLoS One 12(4):e0176131 (model description); Heo J, Adams PJ, Gao HO (2017), Environment International 106:119–126, doi:10.1016/j.envint.2017.06.006 (APSCA reduced-form source-receptor model);
Jin X et al. (2020), Environ Sci Technol 54(9):5339–5347 (CA isopleths L4);
Turner MC et al. (2016) (national CRF; California-specific ozone-mortality dose-response is assumed equal to the national estimate, a transfer assumption that is conventional but not separately validated for the SoCal photochemical regime), Am J Respir Crit Care Med 193:1134–1142 (O3 CRF);
Kennedy MC & O’Hagan A (2000), Biometrika 87(1):1–13 (AR1 multi-fidelity emulator);
EPA AQS daily_44201_2024.zip (V-2, 90 CA monitors)
|
| Upstream inputs | No upstream California Freight Cleanup investigation results consumed; anchor is the processed AQS file only. |
| Downstream consumers | Investigation 4-1 (ozone-channel Sobol GSA) • Investigation 1-1 / 6 / 7 narrative framing • Investigation M-1 portfolio frontier ozone-channel correction |
| Source files |