Urban ozone formation and sensitivities to volatile chemical products, cooking emissions, and NO_x upwind of and within two Los Angeles Basin cities

Volatile chemical products (VCPs) and other non-traditional anthropogenic sources, such as cooking, contribute substantially to the volatile organic compound (VOC) budget in urban areas, but their impact on ozone formation is less certain. This study employs Lagrangian box modeling and sensitivity analyses to evaluate ozone response to sector-specific VOC and nitrogen oxide (NO_x) emissions in two Los Angeles (LA) Basin cities during the summer of 2021. The model simulated the photochemical processing and transport of temporally and spatially gridded emissions from the FIVE-VCP-NEI17NRT inventory and accurately simulates the variability and magnitude of O₃, NO_x, and speciated VOCs in Pasadena, CA. VOC sensitivity analyses show that anthropogenic VOCs (AVOC) enhance the mean daily maximum 8 h average ozone in Pasadena by 13 ppb, whereas biogenic VOCs (BVOCs) contribute 9.4 ppb. Of the ozone influenced by AVOCs, VCPs represent the largest fraction at 45 %, while cooking and fossil fuel VOCs are comparable at 26 % and 29 %, respectively. NO_x sensitivity analyses along trajectory paths indicate that the photochemical regime of ozone varies spatially and temporally. The modeled ozone response is primarily NO_x-saturated across the dense urban core and during peak ozone production in Pasadena. Lowering the inventory emissions of NO_x by 25 % moves Pasadena to NO_x-limited chemistry during afternoon hours and shrinks the spatial extent of NO_x saturation towards downtown LA. Further sensitivity analyses show that using VOCs represented by a separate state inventory requires steeper NO_x reductions to transition to NO_x sensitivity, further suggesting that accurately representing VOC reactivity in inventories is critical to determining the effectiveness of future NO_x reduction policies.

1 Introduction

Decreases in emissions of volatile organic compounds (VOCs) and nitrogen oxides (NO_x= NO + NO₂) have coincided with reduced ozone (O₃) pollution in many United States (US) urban centers (Warneke et al., 2012; Jiang et al., 2018; Kim et al., 2022). As reductions from automotive emissions and other point sources continue, the relative contribution from understudied anthropogenic VOCs (AVOCs), including volatile chemical products (VCPs), to the total emitted VOCs in urban areas continues to rise (McDonald et al., 2018b; Gkatzelis et al., 2021b) and is important for O₃ and acetyl peroxynitrate (PAN) formation (Coggon et al., 2021). Other non-traditional emission sources, including commercial and residential cooking, contribute to the urban VOC (Coggon et al., 2024a; Peng et al., 2022; Wernis et al., 2022) and aerosol (Hayes et al., 2013; Robinson et al., 2006; Robinson et al., 2018) mixture, though the atmospheric impacts and spatial–temporal distributions are not well studied. Attributing O₃ to specific VOC emission sources is challenging, and some methods rely on reactivity estimates (e.g., maximum incremental activity) to predict O₃ formation potential (OFP), which may not reflect realistic atmospheric processing (Gu et al., 2021a; Venecek et al., 2018). Alternative approaches invoke complex source apportionment or tagging methods (Wang et al., 2019; Li et al., 2023; Butler et al., 2020). Models using inventory emissions are a useful tool for O₃ source attribution as they simulate transport and photochemical processing (Abdi-Oskouei et al., 2022; Coggon et al., 2021) while simultaneously enabling sensitivity analyses that directly relate to potential emissions scenarios.

Ozone formation depends on the initial mix and reactivity of VOCs and total NO_x and is known to respond non-linearly to changes in their atmospheric concentrations (Seinfeld, 1989; Kleinman et al., 1997; Kleinman, 2005; Edwards et al., 2014). In regions characterized by high VOC mixing ratios and lower NO_x, O₃ production increases with the addition of NO_x since the rate of radical propagation and NO₂ production is limited by NO. This chemical regime is often termed “NO_x-limited” or “NO_x-sensitive”. In contrast, if NO_x mixing ratios are high relative to VOC reactivity, the opposite dependence on NO_x is observed and O₃ production is considered “NO_x-saturated” (also termed “VOC-limited”, “VOC-sensitive”, or “radical-limited”). In this regime, increasing NO_x suppresses OH radicals and limits the production of peroxy radicals, which propagate ozone by reacting with NO to form NO_2. The addition of VOCs or reduction of NO_x in this chemical regime increases O₃ production. Defining these two distinct chemical regimes in urban areas is critical to understanding effective mitigation strategies, and this paper investigates the sensitivity of ozone mixing ratios to changes in emissions and ozone precursors.

The Los Angeles (LA) Basin in California is among the few remaining regions of the USA where O₃ mixing ratios exceed the National Ambient Air Quality Standard (NAAQS) for ground-level O₃ (8 h average of 70 ppb) with an extreme classification. Significant efforts to mitigate O₃ production in the LA Basin have focused on emissions control. Reductions in AVOCs, carbon monoxide (CO), and NO_x, primarily from the transportation sector, have reduced ground-level O₃ and particulate matter (PM_2.5) for decades (McDonald et al., 2018b; Warneke et al., 2012; McDonald et al., 2013; Pollack et al., 2013). Many of these downward trends have leveled off in recent years, with daily maximum O₃ in some locations actually increasing since 2010 (Jin et al., 2008; Gaudel et al., 2020; Kim et al., 2022). This could be the result of unanticipated trends in emissions (McDonald et al., 2018b), increasing the influence of regional background sources (Silvern et al., 2019); long-range transport (Cooper et al., 2015); changes in atmospheric chemistry (Laughner and Cohen, 2019); and/or consequences of a changing climate with more frequent, longer-lasting, and more intense heat waves in the USA (Habeeb et al., 2015). A clear understanding of O₃ precursors from anthropogenic and biogenic emission sources is key to identifying and targeting NO_x and VOC control strategies in the LA Basin and other major US cities.

The spatiotemporal distribution of emissions and geography of the LA Basin complicate the dependence of O₃ production on NO_x, encompassing both NO_x-limited and NO_x-saturated regimes that are seasonally and spatially dependent. Historically, ozone production in the urban core is NO_x-saturated during peak production (Pollack et al., 2012), whereas locations further downwind transition to NO_x-limited chemistry (Jin et al., 2020; Perdigones et al., 2022; LaFranchi et al., 2011; Pusede and Cohen, 2012). The emission reductions prompted by the COVID-19 pandemic showed mixed results across the LA Basin (Parker et al., 2020; Parker et al., 2022; Schroeder et al., 2022; Pennington et al., 2023); however, when considering meteorology, O₃ formation remains NO_x-saturated in the densely populated areas and becomes NO_x-limited further east (Parker et al., 2022) or on average basin-wide (Schroeder et al., 2022). In general, the extent of the NO_x-saturated regime continues to shrink in many urban areas (Koplitz et al., 2022), but spatial O₃ sensitivities depend on local and upwind emissions, chemical processing, meteorology, and temperature that can vary throughout the day (Nussbaumer and Cohen, 2020).

Determining the O₃ chemical regime, either directly or indirectly, remains challenging. Studies have evaluated changes to O₃ between weekdays and weekends to infer O₃ responses to changes in transportation emissions (termed the “weekday–weekend”, or WD–WE, effect). The WD–WE effect centers on lower NO_x on weekends as commercial transportation declines, while VOC emissions remain similar. WD–WE estimates are limited spatially and temporally and are complicated by reduced overnight O₃ titration impacting O₃ background (Peischl et al., 2023). Other approaches estimate changes in VOCs and NO_x through proxies such as satellite column-integrated formaldehyde (HCHO) to NO₂ ratios, which may fail near transitional regimes and are limited by the satellite spatiotemporal resolution (Schroeder et al., 2017). Direct chemical indicators using termination products of HO_x–NO_x cycling (Sillman, 1995; Martin et al., 2004; Vermeuel et al., 2019) or measured/simulated radical production and loss have also been explored (Kleinman et al., 1997; Mao et al., 2010; Abdi-Oskouei et al., 2022; Sakamoto et al., 2019; Robinson et al., 2021; Kleinman, 2005; Rickly et al., 2023). Analyzing the ozone response to NO_x and VOC perturbations in smog chambers is the most direct method to evaluate O₃ sensitivity (Wu et al., 2022); however, the approach is practically limited by the location and time of sampling and does not account for dilution, mixing, or deposition (Wu et al., 2024).

Chemical transport models (CTMs) and box models have been used to evaluate the O₃ response to changes in precursor emissions and meteorology. Box models require observational constraints to accurately simulate O₃ mixing ratios, while CTMs require substantial computation time to determine the O₃ response to multiple emissions perturbations. In this study, we employ a multi-process Lagrangian box model that simulates the spatial and temporal evolution of emissions in the LA Basin. This model is used to simulate summertime O₃ formation and to evaluate O₃ sensitivities to changing NO_x and VOCs as air masses transit to two locations downwind of downtown Los Angeles: Pasadena and Redlands, CA. To fully represent the processes that impact O₃ production, the model is informed with anthropogenic and biogenic VOCs from state-of-the-science emissions inventories and employs refined chemical mechanisms, ground-site observations, and meteorological variables from chemical transport models to fully represent the processes that impact O₃ production. The model is first evaluated by comparing it to month-long measurements in Pasadena, CA, collected in the summer of 2021. Then, anthropogenic ozone is attributed to sector-specific VOC sources, and the impact of NO_x and VOC perturbations on O₃ formation is quantified to determine the photochemical regime at different locations and times across the LA Basin. Lastly, the O₃ response to total VOCs and their reactivity is evaluated by scaling VOCs to match a separate state anthropogenic emissions inventory.

2 Methods

2.1 Campaign description

The 2021 Southwest Urban NO_x and VOC Experiment (SUNVEx; https://csl.noaa.gov/projects/sunvex/, last access: 1 June 2024) field campaign was conducted at the California Institute of Technology in Pasadena, CA (34.1403° N, 118.1254° W), from 2 August to 6 September 2021. The site has been previously described as a receptor for the pollutants of downtown Los Angeles and other upwind sources (de Gouw et al., 2018; Ryerson et al., 2013). Measurements were conducted either in a trailer or within a stationary mobile laboratory (34.1403° N, 118.1254° W), both of which sampled air from the same 10 m tower. The mobile laboratory was deployed to characterize the spatial distribution and to determine the local sources of emissions for times within two periods (2–6 August and 31 August–3 September 2021). Onboard measurements during hours when the mobile laboratory was away from the ground site were excluded from this analysis. Companion continuous measurements were collected nearby at the Ronald and Maxine Linde Laboratory (34.1364° N, 118.1268° W). Table 1 highlights the comprehensive measurements used in this work. Many of these measurements and techniques are described in detail elsewhere (see references in Table 1) and were used to evaluate box model simulations of speciated VOCs, CO, NO, NO₂, total oxidized nitrogen (NO_y= NO_x+ NO_x reservoirs), acyl peroxynitrates (PANs), nitric acid (HNO₃), and O₃ (see Sect. 3). The Stationary Doppler lidar On a Trailer (StaDOT) was co-located at the Pasadena site and measured the planetary boundary layer height.

Hourly measurements of O₃ and relative humidity at the Dearborn Reservoir (34.059° N, −117.1473° W) air monitoring site, managed by the South Coast Air Quality Management District (SCAQMD), were used to evaluate the modeled O₃ in Redlands, CA. This site was not equipped extensively; therefore model constraints and evaluations are limited.

Table 1 Summary of instrumentation, utilized measurements, and time resolution reported for the SUNVEx campaign in Pasadena, CA.

2.2 Lagrangian box model

A Lagrangian box model was built using the Framework for 0-D Atmospheric Modeling (F0AM) (Wolfe et al., 2016) and simulated the evolution of emissions; the chemistry; and the dilution of VOCs, NO_x, CO, and other trace gases as air was transported across the LA Basin from 7 August–7 September 2021. The box model is designed to complement the Weather Research and Forecasting model coupled with Chemistry (WRF-Chem) simulations described by Zhu et al. (2024a), which was configured using anthropogenic emissions from the FIVE-VCP-NEI17NRT inventory and an updated chemical mechanism named RACM2B-VCP. The WRF-Chem model accurately reproduced O₃ and many speciated VOCs across the LA Basin during the summer of 2021. The box model in this study leverages the WRF-Chem output to constrain transport and meteorological variability (see Sect. 2.2.2) and is configured with the same emissions (see Sect. 2.2.3) and chemistry (see Sect. 2.2.4) as described by Zhu et al. (2024a), with modifications detailed in the following subsections. The box model is intended to evaluate O₃ responses to emissions perturbations and to assess sector contributions to photochemical O₃ observed at two receptor sites. The model can also be used for mechanism development, which is challenging to perform in CTMs. Here, we focus on describing the model and corresponding O₃ responses to anthropogenic VOC and NO_x perturbations.

2.2.1 Model overview

In its basic form, air accumulates emissions that react and are transported over several hours across the LA Basin before reaching either the Pasadena or the Redlands receptor sites. Typically, daytime transport in the LA Basin is that of sea-breeze meteorology, where air parcels originate near the coast (Santa Monica and San Pedro bays) and are advected inland, as shown for representative trajectory paths in Fig. 1 (Washenfelder et al., 2011; Wagner et al., 2012; Hayes et al., 2015). Variability in meteorology and advection can complicate this typical transport pattern; therefore, back trajectories were calculated to estimate a series of air parcel coordinates that move with the wind at 15 min intervals, arriving hourly at each receptor site throughout the month (see Sect. 2.2.2).

For all species, the box model assumes that mixing ratios are controlled by emissions, chemical reactions, and dilution from vertical expansion with background air, and ignores deposition and aerosol uptake. The initial concentrations for most VOCs and NO_x are assumed to be zero, while initial O₃ mixing ratios are set using an air monitoring site near the coast. At each model step coordinate (latitude, longitude, time), the corresponding emissions (E_i) within an 8×8 km area footprint (A) are mixed into a volume defined by the height of the planetary boundary layer (h). Dilution is represented by a first-order rate constant (k_dil) and is calculated based on the changes in boundary layer height between model steps. Chemical processing occurs within each model step, and the resulting concentrations set the starting conditions for each subsequent step. A simplified representation of a moving trajectory path is shown in Fig. S1 in the Supplement, and the basic equation describing this evolution of a given species is provided by Eq. (1):

$$\begin{array}{}\text{(1)}& {\displaystyle \frac{\mathrm{d}{C}_i}{\mathrm{d}t}}={\displaystyle \frac{E_i\left(t\right)}{h\left(t\right)\cdot A}}+\sum r_i\left(t\right)-k_{\mathrm{dil}}\left(t\right)(C_i-C_{\mathrm{b},i}),\end{array}$$

where C_i is the concentration of a given species per volume (molec. cm⁻³), E_i(t) is the emission rate based on the time-varying emissions inventory (molec. s⁻¹; see Sect. 2.2.3), h(t) is the time-varying planetary boundary layer height from Doppler lidar measurements or 3-D-model simulations (cm; see Sect. 2.2.5), A is the footprint area represented by the box model (cm²), r_i(t) denotes the reaction rates (molec. cm⁻³ s⁻¹) that account for the chemical production and loss of C_i, k_dil(t) is the time-varying first-order dilution rate calculated based on boundary layer expansion (s⁻¹), and C_b,i is the background concentration of a given species mixed from outside of the model that represents air from the residual layer or free troposphere entraining into the box. The following sections detail each process represented by Eq. (1).

Figure 1 Example FLEXPART backward-trajectory paths initiated in Pasadena or Redlands, CA. Red dashes indicate the location of the particle cluster center of mass at 15 min intervals overlaid with 4 km spatial grids colored by hourly D5 siloxane emission rates from the FIVE-VCP-NEI17NRT inventory. An example model area boundary (8×8 km perimeter) used to extract emissions for a single model step is shown in yellow.

2.2.2 FLEXPART-WRF back-trajectory analysis

Trajectory paths were generated with a limited-area version of the FLEXible PARTicle dispersion model (FLEXPART) (Stohl et al., 2005) driven with meteorological input provided by the Weather Research and Forecasting v4.2.2 (WRF) numerical weather prediction model (FLEXPART-WRF) (Brioude et al., 2013) and an adapted turbulent parameterization (Verreyken et al., 2019). The dynamical setup from the WRF-Chem model was optimized for the SUNVEx campaign by Verreyken et al. (2025). The model was run on a 4 × 4 km domain covering California and Nevada with the High-Resolution Rapid Refresh (HRRR) (Benjamin et al., 2016) model providing initial and boundary conditions. The boundary layer was simulated using the Mellor–Yamada–Nakanishi–Niino (MYNN) (Nakanishi and Niino, 2009) algorithm with an eddy diffusivity mass flux (EDMF) parameterization (Olson et al., 2019).

The model calculates particle dispersion backwards, releasing 25 000 particles (air parcels) between the surface and 20 m above ground level. For particle masses released hourly in FLEXPART-WRF, trajectories were followed back for 18 h. To simulate the trajectory of air parcels arriving hourly, coordinates (latitude, longitude, altitude) were derived every 15 min as the center of mass of the main particle cluster. Ancillary information (photolysis rates, boundary layer height, temperature, pressure) was obtained at each coordinate from 4-D interpolation along each trajectory from the WRF-Chem simulation (Grell et al., 2005) completed for the SUNVEx campaign period (Zhu et al., 2024a). More information about the WRF-Chem setup can be found in Zhu et al. (2024a).

The FLEXPART-WRF trajectory analysis generated 744 different air mass tracks arriving in Pasadena and Redlands each, every hour from 7 August–7 September (Fig. S2). Meteorology was most consistent midday when the boundary layer was fully developed and sea-breeze meteorology pushed air masses from the coastline northeastwards towards Pasadena, as represented by Fig. 1 and the average trajectory paths shown in Fig. S2. The back-trajectory analysis revealed that air parcels arriving in Pasadena at midday (between 13:00–18:00 PDT) had been transported from the coast for 4–5 h on average. The transport further east towards Redlands, CA, is more complex, and the back-trajectory analysis suggests air parcels often originated further south and not always along the coastline. All backward trajectories from the Pasadena and Redlands receptor sites were traced back 4.5 and 9 h, respectively, and the average hourly paths are shown in Fig. S2 in the Supplement.

2.2.3 Emissions

Anthropogenic emissions of CO, NO_x, and VOCs were injected into each model step along the trajectory path from the spatially resolved (4 × 4 km) hourly FIVE-VCP-NEI17NRT anthropogenic emissions inventory (McDonald et al., 2018a, b; He et al., 2024) for the conterminous USA. The near-real-time (NRT) emission development is extensively detailed in the Supplement of He et al. (2024) and was extracted for the month of August 2021. Briefly, on-road and non-road engine sources are from the Fuel-Based Inventory for Vehicle Emissions (FIVE) and are updated using fuel sales (Harkins et al., 2021; McDonald et al., 2014). Volatile chemical product (VCP) emissions (McDonald et al., 2018b; Coggon et al., 2021) are included and updated using scaling factors calculated from economic activity statistics. VCP sectors include cleaning/degreasing agents, personal care products, adhesives, coatings, inks, and pesticides. Emissions outside of the USA, such as from Mexico and Canada, as well as shipping emissions, are from the Copernicus Atmospheric Monitoring Service (CAMS) (Doumbia et al., 2021). Oil and gas emissions (O&G) are from the Fuel-based Oil and Gas (FOG) inventory of Francoeur et al. (2021). For pollutants where data are available, power-plant emissions are from the Continuous Emissions Monitoring System (CEMS). Point sources and other area source (gridded) emissions not categorized as VCPs include a mix of evaporative sources, industrial processes, and fossil-fuel- and biomass-combustion-related emissions and are taken from the US Environmental Protection Agency (EPA) 2017 National Emissions Inventory (NEI) (https://www.epa.gov/air-emissions-inventories/2017-nationalemissions-inventory-nei-data, last access: 1 June 2024) and are updated with scaling factors calculated from relevant energy and economic metrics. For the purposes of this analysis, point source emissions have been spatially mapped to the same 4 × 4 km grid that the gridded emissions are available on. Area and non-VCP point emissions are grouped with the other fossil fuel sources in this analysis. The fossil fuel sectors comprise diesel exhaust, evaporative gasoline, gasoline exhaust, off-road diesel and gas, commercial marine vessel, power-plant, and fuel-based oil and gas emissions.

Inventory emissions are hourly and produced for three categories, representative of days of the week: weekdays, Saturdays, and Sundays. The diel emission profiles from FIVE for on-road emissions are detailed in McDonald et al. (2014), while those for off-road emissions are found in McDonald et al. (2018a), with updates to marine gasoline provided by Yu et al. (2023). Most VCP diel profiles are sourced from the EPA 2014 NEI, version 2, with personal care updated to reflect the emissions of D5 siloxane as outlined by Coggon et al. (2018). Temporal profiles for other sources were taken from the 2017 NEI (https://gaftp.epa.gov/air/emismod/2017/, last access: 1 June 2024). Further details about the monthly scaling adjustments to generate near-real-time (NRT) emissions to account for rapid changes in human activity, as well as other details of the anthropogenic emissions inventory, are described by He et al. (2024). Beyond the emissions configuration described by He et al. (2024), updates to the chemical speciation of the VOCs from VCP emissions were made and are described in Zhu et al. (2024a). Figure 1 shows an example of the spatially and temporally resolved grids for D5 siloxane (a representative VCP) from FIVE-VCP-NEI17NRT.

Biogenic emissions (mol s⁻¹ km⁻²) were extracted from 4 km spatially and temporally resolved grids retrieved from WRF-Chem model results for biogenic emissions generated using the Biogenic Emission Inventory System (BEIS) v3.14 for the months of August and September 2021. The inventory included urban vegetation related to urban land cover types and leaf area index. BEIS input emissions parameters for August and September 2021 were run through WRF-Chem for the study period using the same meteorology as that used for the FLEXPART-WRF back-trajectory analysis. WRF-Chem uses the BEIS input parameters and meteorology to calculate hourly simulated biogenic emissions specific to each day of the study time period, which are then retrieved from their 4 × 4 km grid. Monoterpene emissions were assumed to be 37 % limonene, 53 % α-pinene, and 10 % eucalyptol (Van Rooy et al., 2021; Zhu et al., 2024a).

Recently commercial and residential cooking emissions were observed to be important contributors to urban emissions and were estimated to represent as much as 21 % of the total anthropogenic VOC emissions in Las Vegas, NV (Coggon et al., 2024a). These emissions include saturated and unsaturated aldehydes (C₃–C₁₀), long-chain fatty acids, monoterpenes, and ethanol. The FIVE-VCP-NEI17NRT inventory was updated to include cooking emissions estimated using source apportionment reported by Coggon et al. (2024a) and was then spatially distributed onto the same 4 × 4 km grid as other anthropogenic emissions using population density. A temporal profile representing human activity was taken from the commercial cooking profile used in the 2017 NEI (Supplement, Fig. S3). All measured C₈ and C₉ saturated aldehydes (octanal and nonanal) were entirely attributed to the cooking sector and served as cooking markers. Ethanol is a major VOC in urban areas, and it is challenging to fully resolve its mass using apportionment techniques. The ethanol attributed exclusively to cooking was lower than that estimated from cooking VOC fluxes from spatially distributed aircraft flux and indoor measurements (Arata et al., 2021; Pfannerstill et al., 2023), and therefore an upper bound was estimated by increasing ethanol to match the median ethanol-to-nonanal ratio from aircraft fluxes in LA (Zhu et al., 2025). Supplement Table S1 shows the speciation profile and chemical mechanism mapping for cooking VOCs used here. There remains some uncertainty in how the cooking VOC mixture and emission rate varies between cities, and Coggon et al. (2024a) discuss that Las Vegas, NV (where the cooking emissions estimate per population was determined) is a major entertainment city and that the number of restaurants per capita may be among the highest in the USA. While the density of cooking emissions might be higher in Las Vegas, the mass spectrum attributed to cooking compared well to a more recent source apportionment analysis in Toronto, ON, Canada (Rivellini et al., 2024).

Total anthropogenic emissions from FIVE-VCP-NEI17NRT are available on an hourly basis. The total emissions of a given species (E_i; molec. s⁻¹) extracted into each model step is outlined by Eq. (2). First, individual emission rate cell grids (e_i; molec. s⁻¹) whose centers were encompassed within the defined domain area of the box (A) at a given air parcel location (latitude, longitude) and time (t) are summed. The total emission is then divided by the product of the number of summed grid cells (n) and the grid resolution of the inventory (a_grid; 4 × 4 km) and then multiplied by the cross-sectional area of the defined box (A; 8 × 8 km). Finally, emission rates are interpolated to a 15 min time basis and implemented into Eq. (1) as E_i(t). This renormalization technique was tested at different box model grid resolutions to confirm that the total mass was conserved.

$$\begin{array}{}\text{(2)}& E_i={\displaystyle \frac{{\sum }_{\mathrm{0}}^n{e}_i}{n\times a_{\mathrm{grid}}}}\times A\end{array}$$

2.2.4 Chemistry

Chemical reactions are represented by RACM2B-VCP, a mechanism described in detail by Zhu et al. (2024a). Briefly, the mechanism is based on the Berkeley Regional Atmospheric Chemistry Mechanism version 2 (RACM2_Berkeley2.0), which incorporates a more complex representation of organic nitrate chemistry (Zare et al., 2018) along with monoterpene and isoprene oxidation (Browne et al., 2014). Reactions for oxygenated VOCs primarily emitted from VCPs are incorporated based on the reactions described by Coggon et al. (2021) and include surrogates for alcohols and glycol ethers (isopropanol, propylene glycol, glycerol). Additional reactions for specific VCP tracers of personal care products (D5 siloxane), solvent-borne coatings (parachlorobenzotrifluoride, PCBTF), adhesives (D4 siloxane), and insecticides (p-dichlorobenzene) (Gkatzelis et al., 2021a) are included to track VCP tracer emissions and chemistry in the chemical transport models. Additional chemistry for long-chain saturated and unsaturated aldehydes was added to reflect the OH oxidation of key cooking emissions and the varying chemical pathways forming O₃ and PAN. This included the addition of lumped saturated and unsaturated cooking aldehydes along with explicit reactions for octanal and nonanal (Coggon et al., 2024a). The reaction rates and product branching ratios for the OH and NO₃ oxidation of lumped saturated cooking aldehydes (CALD), nonanal (NALD), and octanal (OALD) are based on calculations from GECKO-A (https://geckoa.lisa.u-pec.fr/index.php, last access: 1 January 2024)(Aumont et al., 2005), IUPAC (Mellouki et al., 2021), and Atkinson and Arey (2003). The reaction rates and branching ratios for the OH, O₃, and NO₃ oxidation of lumped unsaturated cooking aldehydes (CUALD) are assumed to be generally similar to methacrolein oxidation, with minor updates to the products considering that CUALD has five carbons. The cooking VOC mapping and chemical mechanism updates are outlined in the Supplement, Tables S1 and S2. The distributions of the inventory cooking emissions and the VOC OH reactivity chemically lumped to RACM2B-VCP species are shown in Fig. S4 in the Supplement. The unresolved mass from source apportionment was attributed to hydrocarbons, which typically fragment in PTR-ToF-MS and have been reported in laboratory cooking studies (Schauer et al., 1999).

The RACM2B-VCP used in WRF-Chem simulations during SUNVEx (Zhu et al., 2024a) included a parameterization for the aerosol uptake of isoprene nitrates and monoterpene nitrates and added the uptake of several inorganic species using uptake coefficients, aerosol surface area, and aerosol diameter. In this work, these mechanisms were reverted back to the constant reaction rates for isoprene and monoterpene nitrates employed by RACM2_Berkeley2.0 as estimated by Zare et al. (2018). Additionally, the aerosol uptakes of NO₃, NO₂, and HO₂ were ignored as aerosol parameters were not extracted from the WRF-Chem runs.

2.2.5 Meteorology, dilution, and entrainment

The model used pressure, temperature, and relative humidity from WRF-Chem along the trajectory paths and was constrained with ground-site measurements in Pasadena. Photolysis frequencies for key chemical species needed in RACM-based mechanisms were retrieved along each trajectory from 4-D interpolation of the WRF-Chem simulation. The Tropospheric Ultraviolet and Visible Radiation Model version 5.3.2 (Madronich and Flocke, 1997) was used for photolysis parameterization and mapped to RACM2B-VCP as outlined by Zhu et al. (2024a). Photolysis frequencies account for photon attenuation by clouds or highly absorbing aerosol (e.g., biomass burning smoke) and varied in space and time. The WRF-Chem-calculated photolysis frequencies along each Pasadena trajectory were bias-corrected using observed NO₂ photolysis rate (jNO₂) comparisons to WRF-modeled jNO₂ at the receptor site. This assumption generally lowered upwind daytime photolysis frequencies by 15 %–20 %. Measured jNO₂ was interpolated during periods when shadows blocked instrument retrievals.

An entrainment rate was applied to account for dilution with surrounding air from the expansion of the planetary boundary layer (PBL) between model steps. WRF-Chem-simulated PBL heights were used to account for spatial and temporal changes along each trajectory path. A constant dilution (0.2 h⁻¹) was iteratively determined from a best fit with observations and applied as the boundary layer collapsed from day into night to account for some advection and diffusive mixing. This value is consistent with other urban plume dilution estimates (Dillon et al., 2002; Nunnermacker et al., 1998; Vermeuel et al., 2019). The regional background concentrations for most VOCs were negligible compared to primary emissions in the dense urban core. For species with larger backgrounds including small oxygenates, the box was diluted with regional background mixing ratios estimated from mobile laboratory measurements, regional air quality monitoring stations, or WRF-Chem chemical concentrations retrieved near the coast, and these are included in Table S3 in the Supplement. The background carbon monoxide (CO) mixing ratio increased during the latter half of the campaign as the air mass was impacted by aged wildfire smoke. The initial NO_x concentration was set to zero, while the background NO-to-NO₂ ratio was set to 0.20. As the box transits across the LA Basin, NO_x emissions are injected into the box and then partitioned to NO and NO₂ based on the previous model step. The modeled NO-to-NO₂ ratio closely matched observations (see Sect. 3.2). The background O₃ mixing ratios entrained into the box volume ranged from 20–40 ppb based on mixing ratios observed in the free troposphere above Pasadena in 2023 using the Jet Propulsion Laboratory's Small Mobile Ozone Lidar (https://tolnet.larc.nasa.gov/download, last access: 1 November 2024), which fall into the range of background O₃ estimated near the California coast from both models (Fiore et al., 2014; Dolwick et al., 2015) and measurements (Jaffe et al., 2018).

The initial VOC concentrations were assumed to be zero or taken as regional background concentrations (Supplement, Table S3) Ozone concentrations were initialized using mixing ratios measured at a surface air quality monitoring site near the coast that was along the average trajectory path to Pasadena (Westchester SCAQMD; 33.9551° N, 118.4305° W; labeled as LAX in Fig. 1). The time series and diel profile used as the initial ozone mixing ratio are shown in the Supplement, Fig. S5.

3 Model evaluation

3.1 Ozone

Figure 2a shows the measured and simulated time series of hourly O₃ in Pasadena during the SUNVEx campaign. The box model accurately predicts O₃ variability with a normalized mean bias (NMB) of −0.001 and a coefficient of determination (R²) of 0.77. The NMB is taken as the hourly mean of the model (Mod) relative to the hourly observational mean (Obs) minus 1 as shown in Eq. (3).

$$\begin{array}{}\text{(3)}& \mathrm{NMB}={\displaystyle \frac{\mathrm{Mod}}{\mathrm{Obs}}}-\mathrm{1}\end{array}$$

The box model meteorology simplifications are challenged during boundary layer and meteorological transitions as surface winds and temperatures shift in the early mornings and late evenings. This is evident as the boundary layer collapses and transitions to stagnant winds overnight, which results in VOCs and NO_x accumulation and subsequent O₃ titration that is not always fully captured in the model (Strobach et al., 2024). The model's ability to recreate O₃ production during daylight hours (07:00–19:00 local time, LT) yields an NMB of −0.01 with an average absolute bias of −0.57 ppb. A companion 3-D chemical transport simulation (WRF-Chem) was configured over the LA Basin during SUNVEx with FIVE-VCP-NEI17NRT emissions and a similar RACM2B-VCP mechanism scheme. The 3-D model more explicitly represents complex meteorology and vertical/horizontal transport over broader spatial scales and accurately simulated O₃ in Pasadena and at 11 additional sites across the LA Basin. Zhu et al. (2024a) reported an NMB of +0.08 and an R² of 0.83 in Pasadena. While the box model is constrained to simpler meteorological assumptions, the general agreement between these two models provides confidence that the box model can be used to conduct detailed VOC and NO_x sensitivity analyses (further described in Sect. 4) that are more computationally expensive in a 3-D model.

The time series in Fig. 2a and diel pattern in Fig. 2b show peak O₃ in Pasadena typically occurred between 13:00 and 15:00 LT, whereas peak O₃ in Redlands occurred between 15:00 and 17:00 LT (Fig. S6b). As indicated in Fig. 2, the model overpredicts O₃ during peak model production (NMB = +0.03; absolute bias = +1.7 ppb), which was not obvious from the calculated NMB since the estimate included all daytime hours on the shoulders of peak production. The box model mechanisms primarily include gas-phase chemistry, which differs from the full 3-D-model RACM2B-VCP mechanism used in WRF-Chem. The simplified mechanism used in the box model is missing processes that include HO₂, inorganic, and aerosol loss pathways, but the box model simulation agrees well with the more complex chemical transport model. The day-to-day agreement varies in both models and is likely driven by uncertainty in meteorology, dilution, and vertical distribution.

The FIVE-VCP-NEI17NRT inventory reflects temporal and spatial usage patterns including those that describe differences in emissions on weekdays versus weekend days and heavy-duty vehicle usage impacting NO_x (Yu et al., 2023). The inventory generally does not reproduce VCP and fossil fuel VOC fluctuations driven by temperature that might be expected with evaporative changes. The temperature sensitivity of biogenic VOCs (BVOCs), including isoprene from BEIS, is captured and reflected in the O₃ variability throughout August, as shown by Zhu et al. (2024a). While model–measurement comparisons vary day to day, on average the model accurately reproduces O₃ during primary production, and therefore all trajectories are included in further analysis. The daily maximum 8 h average (MDA8) O₃ is important when considering the NAAQS set at 70 ppb for ground-level ozone. Figure S7 in the Supplement shows the O₃ diel profile for the 8 h moving average during the sampling period, and the average ± standard deviation MDA8 O₃ from the model (59.5 ± 7.2 ppb) compared well with observations (60.4 ± 13.3 ppb) in Pasadena.

Figure 2 (a) The observed ozone time series (grey lines) overlaid with box model output (red markers) in Pasadena, CA (date format: month/day/year). (b) The median diel profile of modeled (red) and measured (grey) ozone. The mean is a dashed line, and the median is marked by the solid line, with shaded regions indicating the 75th- and 25th-quartile ranges. (c) The median diel profile of NO₂ observations (grey) and model output (blue).

The modeled and observed O₃ time series and diel pattern for Redlands, CA, are shown in Fig. S6a and b in the Supplement. The model configuration yields an NMB of −0.02 and an R² of 0.68, and while the model effectively reproduces average O₃, the periods of difference are more frequent than in Pasadena, with a larger spread in the 25th- and 75th-quartile ranges. The NMB and absolute bias during peak production periods (15:00–17:00 LT) were −0.11 and −9 ppb, respectively. The complex meteorology, topography, and uncertain dilution further downwind of the urban core complicate the model's ability to accurately simulate O₃ variability. Additionally, while O₃ is produced from local sources, the transport of O₃, VOCs, and NO_x from upwind sources becomes more important. The population density is lower in the eastern portion of the basin, but NO_x still accumulates to reach mixing ratios comparable to those of downtown LA (Nussbaumer et al., 2023; Chen et al., 2013). The model simulations in Pasadena demonstrate that the emissions and chemistry in the model are accurate; however, in Redlands, applying entrainment rates calculated from PBL height caused overdilution, and instead constant dilution rates (0.1–0.2 h⁻¹) were assumed. While the meteorological constraints on the model for Redlands are more limited, the results provide a useful comparison to the Pasadena site for the O₃ sensitivity analyses outlined in Sect. 4.2.

3.2 NO_x and NO_y

Figure 2c shows the modeled and measured NO₂ diel pattern and demonstrates that the relative amounts of daytime NO₂ are captured well by the model. Both the measurements and the model exhibit an early-morning increase as traffic emissions mix into a shallow boundary layer. The median NO₂ in the model agrees to within measurement uncertainties during peak ozone (13:00–15:00 LT). The diel patterns of NO and NO_x are shown in Fig. 8a–b in the Supplement. NO₂ is less variable than NO during the day, though the observations show a small enhancement of NO₂ around 13:00 LT that is attributed to the transport and arrival of the downtown LA plume (Borbon et al., 2013). The model shows a small yet discernible enhancement in NO₂ around 13:00 LT. The diel profiles for several longer-lived VOCs, predominately associated with the transportation sector (e.g., benzene and carbon monoxide), show a perceptible midday enhancement starting around 13:00 LT (see Supplement, Fig. S9). This suggests that the model captures the downtown LA plume as expected from previous work (de Gouw et al., 2018) but with a small difference in timing compared to the observations. The observations are likely influenced by local anthropogenic sources since the site was located near parking facilities on a university campus in proximity to other commercial buildings and on-campus traffic. The model configuration yields an NMB during daylight hours of +0.30 for NO_x, with a positive bias in both NO₂ and NO (see Fig. S10). The NMB lowers to −0.04 when limiting the calculation to afternoon hours when O₃ is primarily enhanced (12:00–16:00 LT). The model allows the NO-to-NO₂ ratio to vary, and the modeled ratio closely matches observations (Fig. S8c). The diurnal profile of modeled NO_y is shown with the observed NO_y diurnal profile in Fig. S8d in the Supplement. Additional NO_y measurements on top of the Linde Laboratory were higher than those measured at the primary site, and the modeled NO_y falls between the two measurements during peak ozone.

3.3 VOCs

Model evaluation of VOCs showed similar results to that of O₃ and NO_x. Figure S10 in the Supplement summarizes the NMB and coefficients of determination (R²) for observed species that overlap with explicit compounds in RACM2B-VCP on an hourly timescale. For comparisons with ground-site observations, VOC measurements were excluded when the mobile laboratory conducted drives (during times between 31 August–3 September). The NMB calculations are shown for all species during daylight hours when chemistry is important (07:00–19:00 LT) and for the full simulation including high nighttime mixing ratios that might be influenced by local emissions (de Gouw et al., 2018). The chemistry of secondary products is best explained by the model during midday hours when the boundary layer is fully developed and uncertainties due to dilution and dynamic meteorology are reduced. In contrast, primary VOCs undergo chemical oxidation during the daytime that cause lower mixing ratios. Generally, the NMB for most VOCs is within ± 50 %. Biases are likely a result of uncertainties in the emissions inventory as well as of influences of local emission sources on hourly surface observations. While the model cannot reproduce rapid fluctuations in absolute mixing ratios caused by local sources, it does capture concentration trends across the sampling period as shown in Fig. 3a–e. The time series and median diel patterns for selected VOC tracers show that the model over-accumulates many VOCs at night, even when using higher-than-observed PBL heights prescribed overnight by WRF-Chem. Considering the meteorological limitations of the box model and challenges in both measuring and modeling the nocturnal boundary layer (Lee et al., 2023), the agreement within a factor of 2 overnight is reasonable. Relatively, daytime agreement is more important for investigating O₃ production chemical regimes, as will be discussed in Sect. 4.2. The VOC tracers in Fig. 3a–e were chosen to represent various emission source sectors, including toluene for mobile sources, D5 siloxane for VCPs, octanal for cooking, methyl vinyl ketone (MVK) + methacrolein (MACR) for biogenics, and formaldehyde (HCHO) as a secondary oxidation product and a proxy for VOC OH reactivity (Sillman, 1995). Additional diel patterns for overlap species are compiled in Fig. S9 in the Supplement.

Aromatic VOCs emitted predominately by the mobile sector (toluene, benzene) are slightly overpredicted during the day, but the diel patterns, especially for benzene (Fig. S9), capture the afternoon enhancement that represents the LA plume. This suggests that the traditional mobile sources are well represented in the emissions inventory. Oxygenated VOCs from VCPs contribute significantly to the urban VOC makeup and reactivity in large cities (Coggon et al., 2021; McDonald et al., 2018b) and are represented explicitly in the emissions inventory and RACM2B-VCP. This includes the chemistry of several oxygenated VOCs as outlined by Coggon et al. (2021) (methanol, ethanol, isopropanol, ethylene glycol, propylene glycol, glycerol, acetone) and additional VCP tracers added to the mechanism by Zhu et al. (2024a), including D5 siloxane, D4 siloxane, and PCBTF used in personal care products, adhesives, and coatings/paints. These VCP tracers were generally overestimated by the model and similar model–measurement VCP biases were observed when evaluating WRF-Chem simulation with SUNVEx ground-site data. It is challenging to evaluate inventories at a single site, and agreement generally improves when averaged across several ground sites or when compared to aircraft observations, as shown in Zhu et al. (2024a).

Coggon et al. (2024a) showed that octanal and nonanal are emitted by cooking activities and were added to the model inventory and chemical mechanisms, along with additional saturated and unsaturated aldehydes, fatty acids, monoterpenes, and ethanol based on a VOC source apportionment in Las Vegas, NV. The modeled octanal and nonanal agree well with the measurements at night but are low at midday when local lunchtime cooking activity is expected. This suggests that better constraints on the temporal variability in cooking VOCs are needed in the inventory. Ethanol has been shown previously to be emitted by cooking activities indoors (Arata et al., 2021), though at higher amounts than was resolved by source apportionment in Las Vegas, NV. The ethanol diel pattern (Fig. S9) from the base scenario agrees well with observations before noon but remains low in the early afternoon. Additional measurements in urban areas are needed to better constrain the temporal pattern and magnitude of cooking VOCs, including ethanol, in the emissions inventory.

The model using the biogenic inventory BEIS v3.14 underestimated isoprene, MACR, and MVK in Pasadena. Evaluations of BEIS v3.61 with v3.14 suggest leaf canopy and land-use updates increase BVOC emissions across California (Bash et al., 2016). To better account for the influence of biogenic emissions on modeled O₃ production, BEIS v3.14 emissions were increased by 50 % to improve agreement with MACR and MVK, the first-generation oxidative products of isoprene, which have much longer lifetimes compared to isoprene and better represent the upwind atmospheric chemistry of emissions (Warneke et al., 2010). Even after scaling the biogenic emissions up, both isoprene and monoterpenes remain underpredicted compared to observations. Pasadena is more densely vegetated than many areas upwind in the basin, and Coggon et al. (2024b) showed that the ground site was significantly impacted by local vegetation. Therefore, it is not unexpected that observations are higher and may not be representative of the entire basin.

The simulated mixing ratios of secondary products including PAN, nitric acid (HNO₃), formaldehyde (HCHO), and acetaldehyde (CH₃CHO) are biased low but within 25 %. Formaldehyde and acetaldehyde are large contributors to VOC reactivity (VOC_r) in urban areas. Both are directly emitted but are mostly formed chemically through secondary pathways, where major precursors include isoprene and alkenes (Luecken et al., 2012; de Gouw et al., 2018). HCHO serves as a proxy for total VOC_r as the two are often correlated during the daytime, while the HCHO-to-NO₂ ratio has been used as an indicator of O₃ sensitivity (Sillman, 1995; Hong et al., 2021; Duncan et al., 2010). Acetaldehyde is also important for PAN formation, which itself is an important NO_x and radical reservoir. Lastly, HNO₃ represents a dominant sink of radicals and NO_x. Based on this level of model–measurement agreement, the model adequately represents the important emissions and chemical reactions that influence O₃ formation necessary to perform sensitivity analyses to determine meaningful relationships between VOC emission sectors, NO_x, and the O₃ observed at the Pasadena ground site (Sect. 4).

Figure 3 Left panels: the time series of box model output concentrations (ppb) overlaid with observations (grey lines) in Pasadena. The selected VOCs include (a) toluene (mobile sources, blue dots), (b) D5 siloxane (VCPs/personal care, pink dots), (c) octanal (cooking, orange dots), (d) methacrolein plus methyl vinyl ketone (biogenics, green dots), and (e) formaldehyde (VOC_r proxy, maroon dots). Right panels: the median diel pattern for each VOC box model output (colored) overlaid with observations (grey) (date format: month/day/year). Shaded regions indicate the 75th- and 25th-quartile ranges.

4 Sensitivity analysis

4.1 Contribution of anthropogenic and biogenic VOCs to ozone

VOC sensitivity analyses were conducted to evaluate the contribution of anthropogenic and biogenic VOCs to the amount of O₃ produced at the Pasadena and Redlands ground sites. For each VOC sensitivity test, emissions for VOCs from an individual emission sector were set to zero (Sector_i VOCs=0), while NO_x emissions remained constant, and the change in ozone (ΔO₃) was calculated by comparison to the base case simulation with full emissions. The change in MDA8 O₃ from anthropogenic VOCs in each emission sector in the FIVE-VCP-NEI17NRT inventory was summed to estimate the total O₃ from anthropogenic VOCs (termed “AVOC ozone”) following Eq. (4).

$$\begin{array}{}\text{(4)}& \begin{array}{rl}& \text{AVOC ozone}=\\ & {\sum }_i^{\text{no. of sectors}}{\mathrm{Ozone}}_{\text{Base}}-{\mathrm{Ozone}}_{{\mathrm{Sector}}_{i\mathrm{VOCs}=\mathrm{0}}}\end{array}\end{array}$$

In order to test for non-linearity in the response of this approach, additional sensitivity analyses were also conducted by reducing individual source sector VOCs by 10 % rather than zeroing completely and then multiplying the O₃ change by a factor of 10. The results from both tests agree to within 2 % on average for anthropogenic sectors. It is important to note that the anthropogenic O₃ contributions would likely increase for the fossil fuel sectors if both VOCs and NO_x were removed; however, zeroing both VOCs and NO_x from fossil fuels would significantly alter O₃ chemical regimes. Similar analyses were conducted for BEIS to determine the contribution of biogenic VOCs to ozone (termed “BVOC ozone”). In Sect. 4.2 we will show that O₃ production in Pasadena occurs near the transition from NO_x-saturated to NO_x-sensitive chemistry; thus, the analyses described below reflect AVOC and BVOC O₃ produced under predominantly NO_x-saturated (i.e., VOC-limited) conditions.

The pie chart in Fig. 4a shows the source sector AVOC O₃ contributions determined at MDA8 O₃ (total contribution: 13 ± 2 ppb), and Fig. S11b in the Supplement shows the AVOC OH reactivity distribution by emission sector. The distribution shows that AVOCs from VCP sectors account for 45 % (5.8 ± 1.3 ppb) of the mean AVOC MDA8 O₃, while fossil fuels including other area emissions account for 29 % (3.8 ± 0.8 ppb). In contrast, Coggon et al. (2021) showed fossil fuels were 60 % of the AVOC O₃ in New York City (NYC) at midday during an exceedance event. The magnitude and spatial distribution of both VCP and fossil fuel emissions differ between the cities, likely driven by differing populations and traffic densities (see Fig. S7 in Coggon et al., 2021, and Fig. S11). This includes a larger contribution from off-road engines in LA (Kim et al., 2022; Khare and Gentner, 2018). The dominant VCP source sectors in LA were different than in NYC. The construction-related industrial sectors, such as coatings and adhesives, contribute > 70 % of the VCP O₃ compared to < 50 % in NYC. Coatings and adhesives emit aromatics and alkanes, where aromatics are more efficient at producing ozone than the most oxygenated VOCs (Carter, 2010) and are dominant ingredients in household and other consumer sources (e.g., personal care products and cleaning agents) (Coggon et al., 2021). Qin et al. (2021) estimated that 9 ± 2 ppb of MDA8 O₃ in LA in the early summer of 2010 was due to VCPs. This is higher than estimated in this work (5.2 ppb), with a higher relative contribution to the total MDA8 O₃. The absolute MDA8 O₃ likely differed between studies due to the seasonality of biogenic O₃ production, while the absolute VCP VOC contribution differs with usage and regulation changes in the last decade. Seltzer et al. (2022) attributed 4 ppb of midday O₃ (up to 6–8 ppb) to VCPs in Los Angeles using 2016 emission estimates. Coggon et al. (2021) attributed 5–12 ppb of AVOC O₃ from VCPs downwind of NYC, while Abdi-Oskouei et al. (2022) attributed 2.7 ppb to VCPs in Chicago. Overall, this work agrees with previous assessments of VCP contributions to O₃.

Cooking emissions were not considered in any previous analysis, and in Pasadena the contribution to AVOC ozone was comparable to that of fossil fuels (26 %; 3.4 ± 1.5 ppb) and accounts for an important fraction of the AVOC OH reactivity (Fig. S11). Primary cooking emissions include a mix of both saturated and unsaturated aldehydes, including dienals, that have a high reactivity towards OH radicals. Figure S4b in the Supplement shows these aldehydes account for over 70 % of the cooking VOC OH reactivity, and therefore it is not surprising that over a quarter of the anthropogenic ozone is attributed to cooking VOCs, which had not previously been recognized as a significant source. There remains some uncertainty in the O₃ attribution since studies measuring both the composition and the magnitude of cooking emissions in outdoor urban areas are limited (Rivellini et al., 2024; Coggon et al., 2024a). Increasing the source-apportioned cooking ethanol to improve agreement with aircraft flux estimates (see Sect. 2.2.3) produces 0.6 ppb more O₃ from cooking emissions, which provides some measure of uncertainty for O₃ attributed to cooking VOCs. This highlights the need for more measurements to refine emissions estimates in urban areas. This is one of the first studies to show the cooking VOC contribution to O₃ formation in urban areas and demonstrates that cooking emissions are important for AVOC O₃ formation.

Biogenic emissions are dependent on season and temperature (Guenther et al., 1993) and can contribute significantly to VOC reactivity and O₃ formation during summer (Gu et al., 2021b; Liu et al., 2022). It is critical to quantify the impact of urban vegetation and biogenic VOCs on O₃ and air quality as their impact can minimize the effectiveness of VOC emissions control (Schlaerth et al., 2023). The model suggests that BVOC MDA8 O₃ in Pasadena was ∼ 9.4 ppb, or approximately 42 % of total VOC MDA8 O₃ (i.e., AVOC + BVOC MDA8 O₃). Figure 4b shows the average sectoral distribution of O₃ as air is transported from the coast, ending at Pasadena at 15:00 LT. This increase in BVOC contribution to O₃ from the coast inland is consistent with a VOC OH reactivity analysis in the South Coast Air Basin (Liu et al., 2022). Along the entire trajectory, BVOCs account for a significant fraction of the total VOC O₃ and are comparable to AVOC ozone. It is important to note that the BVOC fraction of O₃ is estimated following a 50 % increase in the prescribed BEIS BVOCs, as described in Sect. 3.3. While anthropogenic VOCs have significantly decreased over the past decades, urban vegetation and their biogenic VOC emissions have generally remained constant and will continue to be a major contributor to urban O₃ formation, especially in NO_x-saturated regimes like downtown LA, as will be described below.

The AVOC O₃ distribution from the box model sensitivity analysis in Redlands is consistent with what was shown above for Pasadena (Supplement, Fig. S12a). In Redlands the population density is lower, which might be the reason why the relative O₃ contribution from fossil fuels grows slightly from 29 % to 31 %. There is only a minor decrease in the contribution of cooking, and the BVOC contribution grows slightly (Supplement, Fig. S12b), continuing the increasing BVOC O₃ contribution further inland (Fig. 4b) (Liu et al., 2022). Although the spatial distribution of NO_x and VOCs varies basin-wide, the distribution and the magnitude of photochemically produced O₃ from AVOC and BVOC do not change significantly between Pasadena and Redlands, CA.

Figure 4 (a) The fractional contribution to MDA8 ozone from anthropogenic VOC sources in Pasadena, CA, including VCPs (purple shading), fossil fuels (yellow shading), and cooking (blue) VOC emissions. (b) The absolute contribution of anthropogenic and biogenic VOC sources to ozone averaged along trajectories initialized near the coast and reaching Pasadena at 15:00 LT.

4.2 Spatial and temporal ozone sensitivity to NO_x

Ozone chemical production regimes are often investigated over broad spatial scales (e.g., the entire contiguous USA, large urban areas) during seasons with the highest O₃ (Koplitz et al., 2022; Jung et al., 2022; Jin et al., 2020). Studies and methods to investigate seasonal or diurnal trends are more limited (Jin and Holloway, 2015; Wu et al., 2022, 2024). NO_x, VOCs, and meteorology vary widely across the LA Basin, and the Lagrangian trajectory box model is an effective tool to explore ozone responses to NO_x emissions in space and time upwind and at receptor sites. Here, the O₃ responses to NO_x perturbations are estimated for all of the modeled trajectories within a well-mixed boundary layer, and they therefore do not represent effects due to gradients in shorter-lived species. For each trajectory, NO_x emissions were scaled by a factor of 0.2–3 in 10 % increments from the base condition while VOCs remained constant. The O₃ response relative to the change in emitted NO_x (ΔO${}_{\mathrm{3}}/\mathrm{\Delta }$NO_x) characterizes the chemical regimes as NO_x-limited (ΔO${}_{\mathrm{3}}/\mathrm{\Delta }$NO_x>0); NO_x-saturated (ΔO${}_{\mathrm{3}}/\mathrm{\Delta }$NO_x<0); or at the transition point (ΔO${}_{\mathrm{3}}/\mathrm{\Delta }$NO_x=0), where the response to NO_x turns over. The initial sign of ΔO${}_{\mathrm{3}}/\mathrm{\Delta }$NO_x identifies the existing chemical regime predicted by the box model at a specific location and time. The difference between the NO_x emissions in the base case simulation and where ΔO${}_{\mathrm{3}}/\mathrm{\Delta }$NO_x=0 approximates the change in NO_x required to transition between chemical regimes. For example, if ΔO${}_{\mathrm{3}}/\mathrm{\Delta }$NO_x at a location and time is initially negative in the base case scenario, lowering NO_x results in a transition from a NO_x-saturated to NO_x-limited regime. The change in NO_x emissions required to reach this transition (termed ΔNO${}_{x_{\mathrm{transition}}}$) is calculated as

$$\begin{array}{}\text{(5)}& \mathrm{\Delta }{\mathrm{NO}}_{x_{\mathrm{transition}}}\left(\mathit{\%}\right)={\displaystyle \frac{{\mathrm{NO}}_{x_{\mathrm{\Delta }{\mathrm{O}}_{\mathrm{3}}/\mathrm{\Delta }{\mathrm{NO}}_x=\mathrm{0}}}-{\mathrm{NO}}_{x_{\mathrm{base}}}}{{\mathrm{NO}}_{x_{\mathrm{base}}}}}\times \mathrm{100},\end{array}$$

where NO${}_{x_{\mathrm{\Delta }{\mathrm{O}}_{\mathrm{3}}/\mathrm{\Delta }{\mathrm{NO}}_x=\mathrm{0}}}$ denotes the NO_x emissions at the transition point and NO${}_{x_{\mathrm{base}}}$ denotes those represented by the base case simulation. Hourly trajectories arriving in Pasadena between 12:00 and 20:00 LT are shown in Fig. S13 in the Supplement, where each location along the trajectory path is colored by ΔNO${}_{x_{\mathrm{transition}}}$ as a percentage. Figure 5a shows the average gridded (8 × 8 km) ΔNO${}_{x_{\mathrm{transition}}}$, considering only those grids where the total number of data points exceeds 50 for times between 12:00 and 20:00 LT. Warmer colors indicate that the location is currently NO_x-saturated (negative ΔNO${}_{x_{\mathrm{transition}}}$), cooler colors indicate that it is NO_x-limited (positive ΔNO${}_{x_{\mathrm{transition}}}$), and the status of nearing transition (ΔNO${}_{x_{\mathrm{transition}}}=$ ± 10 %) is indicated in pale yellow. Figure 5a also shows the average trajectory paths to both Pasadena and Redlands during those times. Similarly, Fig. S14 in the Supplement shows the average ΔNO${}_{x_{\mathrm{transition}}}$, bounded within ± 8 km of the average trajectory track in 4 km segments and overlaid with each backward-trajectory coordinate series. Figures 5a and b and S14 show that at locations nearest to the coast or transiting from further south in the basin, where NO_x and VOC emissions are lowest, the regime is primarily NO_x-limited on average. As the air parcels transit across the urban core and through downtown LA, where the density of emissions is high and pollutants increase rapidly, the sensitivity crosses into a NO_x-saturated regime. The typical transport pathway to Redlands does not pass through downtown LA, and ΔNO${}_{x_{\mathrm{transition}}}$ indicates that the upwind trajectory and Redlands itself are situated near or outside the NO_x-saturated core. The chemical regime and the amount of NO_x required to transition vary by upwind location and time of day, as shown in Figs. S13 and S14 in the Supplement.

Figure 5c shows the average diurnal profile of ΔNO${}_{x_{\mathrm{transition}}}$ in Pasadena for all trajectories arriving during sunlit hours during the campaign period. The background color highlights when ozone production is NO_x-saturated (red), NO_x-sensitive (blue), or nearing transition (yellow). The model suggests that Pasadena is primarily NO_x-saturated for most hours of the day and then shifts only slightly into the NO_x-limited chemistry at 15:00 LT. The NO_x diel pattern shown in Fig. S8a in the Supplement shows high NO_x availability for reaction with radicals during early daylight hours, which then decreases rapidly, allowing radical–radical reactions to compete. Although NO concentrations may be high enough to titrate O₃ during these early hours, this process is unlikely to impact O₃ response during primary production periods. The short period of NO_x sensitivity would only require less than a 10 % increase in NO_x emissions to shift back to a NO_x-saturated regime. Pasadena O₃ production exhibiting NO_x-saturated chemistry but near the transitional point supports previous modeling conclusions that the NO_x-saturated regime is shrinking in many urban centers, including the LA Basin (Koplitz et al., 2022), and Pasadena is near this transition. This is also supported by previous observations, such as an analysis of the WD–WE effect, where NO_x is lower during weekend days. Ozone trends in Pasadena from the late 2010s show weekend MDA8 O₃ being only marginally higher than weekday averages, whereas the basin-wide averages are the opposite (Kim et al., 2022). Wu et al. (2024) showed clear WD–WE patterns in NO_x emissions in Pasadena with higher median weekend O₃ than weekday concentrations in both August and September 2021. The results from both of these studies suggest that Pasadena remains NO_x-saturated, though the weekend effect is less pronounced than in earlier decades. In addition, Wu et al. (2024) used direct O₃ response measurements in chamber experiments to show that the O₃ photochemical regime in Pasadena remained NO_x-saturated between 10:00 and 12:00 LT in August 2021. Uncertainty in the modeled response to NO_x is primarily driven by the absolute NO_x emissions designated by the FIVE-VCP-NEI17NRT inventory. Figure S8 shows that the model tends to overestimate NO_x levels (NMB $=+\mathrm{0.30}$), although this bias varies in time. Importantly, evaluating model performance using surface observations impacted by local emissions may not be fully representative. Consequently, the inventory NO_x emissions remain unadjusted and are not constrained to match observations. As a result, the subsequent analysis focuses on understanding the O₃ response to inventory NO_x, which is often used to inform emissions control strategies.

Two additional simulations were conducted to evaluate the sensitivity for scenarios of higher and lower NO_x emissions (Fig. 5c). Increasing the base case NO_x by 25 % shows that Pasadena sits squarely in the NO_x-saturated regime throughout the entire day, which represents NO_x levels ∼ 4.5 years earlier in 2016–2017 assuming a NO₂ reduction rate estimated from 2011–2015 (−5.6 ± 3.6 % yr⁻¹) (Jiang et al., 2018). Reducing NO_x by an additional 25 % pushes Pasadena into a NO_x-limited regime for over 6 h of the day during peak O₃. Figure 5c also shows that O₃ production in Redlands was primarily NO_x-limited during the sampling period. This result is consistent with previous modeling observations, which show that air masses in the NO_x-saturated urban core transition to NO_x-limited further downwind (Koplitz et al., 2022; Schroeder et al., 2022; Jung et al., 2022). These modeled O₃ responses to NO_x in Redlands compare well to WD–WE evaluations in regions further east in the LA Basin (Kim et al., 2022).

Figure 5 (a) The NO_x emissions required to reach the transition between chemical regimes (ΔNO${}_{x_{\mathrm{transition}}}$, %) averaged in 8 × 8 km grids for times between 12:00–20:00 LT, only where n>50. Warmer colors (− %) indicate that the location is currently NO_x-saturated, cooler colors (+ %) indicate that it is NO_x-limited, and pale yellow (neutral) indicates the status of nearing the transitional regime (ΔNO${}_{x_{\mathrm{transition}}}=$ ± 10 %). The average trajectory path to Pasadena and Redlands is indicated by black markers. (b) The average NO_x emissions required to reach the transition within ± 8 km of each average trajectory path as a function of distance from the coastline (km) to Pasadena (top) and Redlands (bottom). (c) The predicted average (solid line) change in NO_x needed to transition between photochemical O₃ chemical regimes in Pasadena, CA. Red shading designates the current chemical regime as NO_x-saturated, and blue shading designates NO_x-limited. The dashed lines simulate the sensitivity as the base case NO_x is increased and reduced by 25 %. The solid grey line designates sensitivity results in Redlands, CA.

4.3 Impact of inventory selection on MDA8 O₃ sensitivity

Ozone is also sensitive to total VOCs and their reactivity. Figure 6 shows the baseline campaign average MDA8 O₃ and its response to changes to NO_x emissions (black line), while Fig. S15 in the Supplement shows the corresponding net ozone production rate (PO₃; ppb h⁻¹) response to NO_x. The black vertical markers show the point where the mean MDA8 O₃ transitions between chemical regimes (ΔMDA8 O${}_{\mathrm{3}}/\mathrm{\Delta }$NO_x≈0), and the hatched line reflects the base case NO_x emissions. The sensitivity analysis shown in Figs. 6 and S15 illustrates that in Pasadena at the transition point, a 10 % to 20 % decrease in NO_x would result in a shift to NO_x-sensitive chemistry based on MDA8 O₃. These model conditions represent the FIVE-VCP-NEI17NRT emissions described in Sect. 3.3, which include cooking VOCs and BVOCs from BEIS v3.14 that were increased by 50 %.

Figure 6 also compares FIVE-VCP-NEI17NRT to other anthropogenic inventories, including the National Emissions Inventory (NEI), which is the standard for determining US emissions standards, represented by the California Emissions Projection Analysis Model (CEPAM2019v1.03) tool used by the California Air Resources Board (CARB) to estimate the average annual anthropogenic emissions by region in California. The emissions prescribed by CEPAM2019v1.03 closely match the 2017 NEI (https://ww2.arb.ca.gov/applications/emissions-air-basin, last access: 6 June 2024). The distributions of VOCs from FIVE-VCP-NEI17NRT and CEPAM2019v1.03 (hereafter designated “CARB CEPAM”) are shown in Fig. S16 in the Supplement for VCPs, on-road and off-road engines, fuel-based oil and gas, and cooking VOCs in the South Coast Air Basin (SoCAB). In Las Vegas, NV, the fraction of cooking VOCs to the total VOC distribution is less than 1 % in the NEI, whereas 21 % of the total VOC mass was determined to be cooking-related emissions by source apportionment (Coggon et al., 2024a). The cooking contribution to AVOCs and speciation determined in Las Vegas was distributed in SoCAB based on population density within the FIVE-VCP-NEI17NRT inventory as described in Sect. 2.2.3. To more realistically represent cooking VOCs in the NEI, the O₃ response to NO_x when zeroing cooking emissions is shown as the blue trace in Fig. 6, which shifts Pasadena into NO_x saturation and lowers the mean MDA8 O₃. Without cooking VOCs, the total VOC reactivity in Pasadena is lower compared to FIVE-VCP-NEI17NRT and will require a NO_x reduction of approximately 30 % to shift into a NO_x-limited photochemical regime. These results show that representing the VOC reactivity from cooking emissions is important for interpreting the impact of NO_x reductions on O₃ formation in Pasadena. Further characterization of cooking VOCs in urban areas and evaluating their representation in models may help inform future O₃ mitigation strategies.

In addition, the VCP emissions from FIVE-VCP-NEI17NRT for 2021 are 4 times higher than CARB CEPAM for 2017 in SoCAB. This is similar to the study by McDonald et al. (2018b), which found that VCP emissions in Los Angeles from FIVE-VCP were a factor of 2.5 higher than those of CARB. Similarly, Qin et al. (2021) showed that scaling the 2011 NEI VCP emissions up by a factor of 3–4 improved agreement with the FIVE-VCP-modeled OH reactivity in California. A recent study by Pfannerstill et al. (2023) also identified mismatches in the spatial distribution of VOC emissions within SoCAB compared to the inventory. While the VOC emissions from on-road engines are comparable, off-road VOCs represented by FIVE-VCP are approximately 50 % lower than those in CARB CEPAM. The red trace in Fig. 6 shows the MDA8 O₃ response to NO_x after scaling FIVE-VCP-NEI17NRT AVOC emissions to reflect what is used in the 2017 CARB CEPAM for the VCP, mobile, and cooking sectors. The VOC adjustments lowered total MDA8 ozone and suggest that even larger model NO_x reductions would be needed to transition to NO_x-limited chemistry in Pasadena (40 % decrease needed to reach ΔMDA8 O${}_{\mathrm{3}}/\mathrm{\Delta }$NO_x≈0). Finally, the yellow trace shows MDA8 O₃ responses with unadjusted BEIS v3.14 BVOCs. As expected, lower biogenic emissions push the model further into a NO_x-saturated regime. This sensitivity analysis is used to compare the updated FIVE-VCP-NEI17NRT emissions to anthropogenic and biogenic emissions inventories commonly used in regulatory models.

The sensitivity analyses highlight the impact of the total VOCs and the composition of anthropogenic VOCs on MDA8 O₃ (and PO₃) and its response to changing NO_x, especially in a VOC-sensitive regime. Emissions inventories are needed to accurately represent all emission sources and total VOC reactivity (Zhu et al., 2025), and the uncertainty between inventories leads to uncertainties in the predicted efficacy of emissions control strategies. NO_x reductions have been effective in reducing O₃, and it is expected that NO_x reductions will continue in the near future (Schroeder et al., 2022; Kim et al., 2022; Koplitz et al., 2022; Wu et al., 2024; Nussbaumer and Cohen, 2020). Figure 6 demonstrates that with an observationally constrained inventory (e.g., FIVE-VCP-NEI17NRT), model simulations would suggest that the LA Basin as a whole is nearing a NO_x-sensitive chemical regime and that the spatial extent of NO_x-saturated chemistry is directionally shrinking towards downtown LA. In these model simulations, Pasadena is nearing a transitional regime. On the other hand, model simulations using emissions representative of the NEI suggest that O₃ in much of the LA Basin is more NO_x-saturated and that a nearly 40 % decrease in NO_x is needed to transition. These differences are significant and could potentially alter conclusions on the efficacy of NO_x reduction policies. More studies are needed to describe the statistical confidence levels in which various scenarios will achieve a consistent NO_x-limited regime in Pasadena, especially in the context of future O₃ exceedance events that drive O₃ design values. This study demonstrates that accurate VOC representation in models can improve the prediction of O₃ formation and is required to inform various NO_x and VOC emission reduction decisions to effectively achieve O₃ levels that satisfy the NAAQS.

Figure 6 The change in the campaign average MDA8 O₃ in Pasadena as NO_x is scaled from its initial mixing ratio (hatched line) for the following scenarios: (1) base model emissions that includes FIVE-VCP-NEI17NRT anthropogenic VOCs, cooking VOCs, and BEIS BVOCs scaled up 50 % (black line); (2) base emissions excluding cooking VOCs (blue); (3) base emissions excluding cooking VOCs with VCP VOCs scaled down and off-road VOCs adjusted to better match the CARB CEPAM inventory emissions (red); and (4) BVOCs scaled down to match the original BVOCs prescribed by BEIS coupled with WRF-Chem (yellow). The location of the transition between photochemical regimes is indicated by the black vertical markers, and VOC_r is taken as the trajectory average.

5 Conclusions

A detailed, multi-process Lagrangian box model was developed to determine O₃ sensitivities to NO_x and VOC emissions from VCPs, fossil fuels, cooking, and biogenics in the Los Angeles Basin. A comparison of the box model output of O₃, NO_x, and speciated VOCs with measurements in Pasadena, CA, during SUNVEx suggests that the FIVE-VCP-NEI17NRT gridded inventory is effective in representing LA emissions in space and time. The box model incorporated variables from a 3-D model together with updated VOC chemistry to accurately simulate ozone (hourly normalized mean bias = −0.001; R²=0.77) and other products of atmospheric chemistry. This included updated chemical mechanisms that better represent the chemistry of oxygenated VOCs emitted from sources such as VCPs and cooking (e.g., saturated and unsaturated aldehydes).

The model proved to be a practical tool to evaluate O₃ response to small perturbations and shows that cooking VOCs contribute nearly as much as fossil fuel VOCs to MDA8 AVOC O₃ in Pasadena (∼ 26 %). The relative contributions in other cities likely vary based on restaurant density, VCP usage, and transportation activity, but this remains one of the first studies to show O₃ production from cooking emissions in an urban area. Additional measurements in other cities are critical to evaluating cooking VOC representation in emissions inventories, both temporally and spatially. The inclusion of cooking VOCs impacts NO_x and VOC control assessments, highlighting the importance of accurately representing VOC reactivity and chemistry in inventories and model simulations. The model suggests BVOCs are still an important O₃ precursor (9.4 ppb) that cannot be easily controlled.

The modeled ozone sensitivity analysis with full inventory emissions suggests that the urban core of Los Angeles, including Pasadena, remains primarily NO_x-saturated during peak O₃ production and shifts to NO_x-limited farther east towards Redlands. Pasadena sits near the transitional point and requires approximately a 10 %–20 % reduction in NO_x to transition, indicating that the spatial extent of NO_x saturation is shrinking across the LA Basin. Scaling FIVE-VCP-NEI17NRT VOC emissions to better reflect those represented by the 2017 NEI and CARB CEPAM lowers absolute O₃ and pushes Pasadena further into NO_x-saturated chemistry, suggesting a 40 % NO_x reduction is needed to shift photochemical regimes. Accurately representing VOCs and their reactivity in emissions inventories is critical to determining absolute O₃ formation and to determining how NO_x reductions may impact O₃.