The observations are taken from the Birmingham NERC air quality supersite during August 2022. This site is located on the University of Birmingham campus (52°27′20.2′′ N, 1°55′44.3′′ W). The site has been in operation for many years and represents an urban background environment. It is influenced by transport emissions from nearby arterial roads and residential emissions from the surrounding area. There are no significant industrial activities within a 4 km radius of the site.

Continuous measurements of NO; NO2; CO; CH4; VOCs; and O3; as well as meteorological parameters including air temperature and pressure, relative humidity, wind speed, and direction were made. Briefly, NO and NO2 were measured by a chemiluminescence-based T200 analyzer (Teledyne API, USA) and the T500U cavity attenuated phase shift (CAPS) analyzer (Teledyne API, USA). The concentration of NOx was then the statistical sum of NO and NO2. The mixing ratio of CO was measured by a laser absorption spectroscopy multi-species continuous emissions monitoring instrument (Enviro Technology Service Ltd., UK) (Li et al., 2020). Manual calibration and span checks for the above instruments were performed every 3 d, and automatic zero calibration was set to be performed on a daily basis. O3 was measured by an O3 analyzer (Model 49i; Thermo Fisher Scientific Inc., USA) with a minimum detection limit (MDL) of 1.0 ppbv. Meteorological parameters, including air temperature and pressure and relative humidity, were obtained by a weather station (WS300-UMB; Luff GmbH, Germany). Additionally, wind speed and direction were measured by a three-axis ultrasonic anemometer (Gill Instruments Ltd., UK) throughout the campaign.

A gas chromatography–flame ionization detection (GC-FID) analysis system (7890A; Agilent Technologies, USA) was used to quantify 38 individual VOC species. Details on instrument settings and quality assurance/quality control methods can be found in Warburton et al. (2023). In brief, the GC-FID system utilizes dual detectors: one detector for C2–C6 non-methane hydrocarbons (NMHCs) and the other detector for the remaining C ≥ 7 hydrocarbons and polar species such as ethers, ketones, and alcohols. Ambient samples were dried at -40 °C using a water trap and then preconcentrated on a carbon adsorbent at the lowest temperature the unit could achieve, which was always lower than -115 °C. Once a 0.5 L sample had been collected, a pre-concentration trap was warmed slightly to -80 °C to purge trapped atmospheric CO2. The trap was then heated to 190 °C for 3 min, with a counter flow of helium thermally desorbing the concentrated VOCs onto a focusing micro-trap held at a temperature lower than -115 °C. The analytes were flash-heated and passed onto a VF-WAX column. The unresolved analytes (C2–C6 NMHCs) were then transferred into a Na2SO4-deactivated Al2O3 porous-layer open-tubular (PLOT) column via a Deans switch for separation and detection by the first FID. The Dean switch then diverted the analytes onto a fused silica internal diameter to balance column flows and subsequently transfer the VF-WAX column-resolved species into the second FID. Generally, the quantification of C2–C6 hydrocarbons was completed by the first FID using 4 ppbv gas standard cylinders (National Physical Laboratory, Teddington, UK). The quantification of C ≥ 7 hydrocarbons and OVOCs was completed by the second FID using an effective carbon number (ECN) with reference to toluene. In this study, the concentration of total VOCs (TVOCs) was defined as the statistical sum of the concentrations of measured individual species, but this is not meant to infer that this represents the total reactive carbon in air, which would always be greater than this value due to unmeasured species. Later in this study, we broadly group species according to their chemical function groups, summing them into alcohols, ketones, alkanes, alkenes, aromatics, aldehydes, and alkynes.

GC-FID system responses were regularly checked by running direct calibration sequences using 4 ppbv gas standard cylinders. It was verified that there was no FID response drift over the analysis period for this study. Additionally, carbon-wise FID responses for all reported species were calculated to verify the use of ECN as a quantification method. Table S1 lists which species were directly calibrated and which used equivalent carbon numbers for quantification. Table S2 shows effective carbon numbers of species which used carbon-wise responses.

Estimates of UK anthropogenic VOC emissions are taken from the NAEI. The NAEI uses a combination of UK-specific methods and default methods as recommended by the European Monitoring and Evaluation Programme (EMEP)/European Environment Agency (EEA) emission inventory guidebook (European Environment Agency, 2016). Further details can be found in NAEI (2021). The VOC inventory is also disaggregated into inventories for each individual VOC species, and details of the speciation process and assumptions can be found in Lewis et al. (2020).

Methods to estimate emissions can be divided into two groups: those using emission factors and those using point source emissions data reported to regulators by the operators of individual industrial sites. The emission factor methods require UK activity data – for example, the consumption of paint, consumption of a fuel, production of steel, or vehicle kilometers traveled. The activity data are then combined with an emission factor, which expresses the total VOC emission that is expected per unit of a given activity. Most total VOC emission factors are taken from the internationally applied EMEP/EEA emission inventory guidebook and so are not necessarily UK-specific. The factors for road transport are directly calculated for the UK, and a particularly detailed approach is used to estimate emissions using emission factors from the Guidebook for many different vehicle types and emission standards, fuels, and road types combined with detailed transport activity from the UK Department for Transport. Government statistics cannot always provide the necessary activity data for other sectors, so industry data are used instead. For instance, NAEI data on the consumption of products containing organic solvents are from industry sources. The alternative point source method can be used for source categories where emissions data can be obtained for all sites within the sector, and this limits the method to source categories such as crude oil refining, steel production, and chemical production. The emissions data reported by the operators of these sites can be based on emissions monitoring although this is not always the case and emissions might instead be estimated, for example, using emission factors.

The NAEI produces updates to the inventory for total VOC mass emissions by source sector each year to achieve a consistent historic time series reflecting trends in UK emissions. Emissions of individual VOC species are estimated using source-specific speciation profiles, which show the mass fraction of each species or, in some cases, groups of species emitted by the source (NAEI, 2021; Passant, 2002). Over 600 individual VOC species or species groups are included in the speciation based on sources in industry, regulators, and, in some cases, literature sources and databases such as the US EPA SPECIATE database. The speciated inventory tends to be more uncertain than the estimation of the total mass of VOC emissions. The inventory for total VOC mass is updated annually, whereas the speciation profiles are only periodically updated when new information becomes available. Thus, trends in a particular species for a sector are a reflection of changes in total VOC emissions for the sector and do not normally reflect any changes over time in the speciation profile of the sector which may have occurred.

The framework for evaluating effects of changing VOC speciation is a zero-dimensional atmospheric modeling chemistry box model (Wolfe et al., 2016) driven by version 3.3.1 of the Master Chemical Mechanism (MCM v3.3.1) (Saunders et al., 2003; Jenkin et al., 2003). The model can be effective in identifying the instantaneous in situ O3 sensitivity to changes in individual VOCs. The measured concentrations of 38 VOC species, NOx, and CO, along with air temperature and pressure and relative humidity, were averaged to a time resolution of 1 h to constrain the model. A 3 d model spin-up, with each 24 h model run constrained by the observational data, was performed in order to initialize the unmeasured compounds and transient radicals. The modeled outputs on the fourth day were taken to represent the steady state of the photochemistry.

Photolysis rates were calculated as a function of the solar zenith angle (Saunders et al., 2003): 1J=lcos⁡χmexp-nsecχ, where J is the photolysis rate in s-1; l, m, and n are constants derived from radiative transfer model runs for clear-sky conditions at an altitude of 0.5 km and literature cross sections/quantum yields; and χ is the solar zenith angle in radians.

The net production rate of O3 (P(O3)) is calculated as the difference in the production rate of O3 and the destruction rate of O3, as in Eq. (2): 2PO3=kHO2+NOHO2NO+∑ikRO2i+NORO2NO-(kO1D+H2OO1DH2O+kO3+OHO3OH+kO3+HO2O3HO2kNO2+OHNO2OH+∑ikRO2i+NO2RO2NO2), where the former part is the rate of O3 production, represented by the rate of NO oxidation by HO2 and RO2 radicals, and the latter part is the destruction rate of O3, calculated by the sum of the rate of O3 photolysis, the rates of the reactions with OH and HO2 radicals, and the rates of NO2 loss through reactions with OH and RO2 radicals.

The sensitivity of O3 to its precursors is quantified by the index of relative incremental reactivity (RIR) (T. Liu et al., 2022), as in Eq. (3): 3RIR=ΔPO3PO3×a-1, where RIR is the relative incremental reactivity (in % / %), ΔP(O3) / P(O3) is the ratio of the change in the O3 production rate to the base O3 production rate, and a is the reduction percentage in the input concentration of O3 precursors – a factor that allows for the effects of changing absolute amounts of VOCs to be evaluated. Here, a value of 30 % was adopted for a.

The time series of O3 and its precursors during August 2022 are shown in Fig. 1, subdivided into periods that will be referred to as the initial period, O3 period, and clear-out period. The three periods covered 1–21 August 2022. Each period included one full week to avoid weekday/weekend differences in NOx and VOC concentrations impacting differently when O3 production is compared between the three periods (de Foy et al., 2020). Ozone showed a generally increasing trend from 1 to 14 August and then returned to relatively low concentrations after 15 August 2022. The daily maximum 8 h average O3 concentrations (MDA8h O3) during the O3 period exceeded the WHO guideline value (100 µg m-3), ranging from 111 to 153 µg m-3. The elevated O3 during the middle of the month corresponded to more intense photochemical formation under hot weather conditions (32.7 °C in maximum) and higher concentrations of O3 precursors (Table S3).

The diurnal profile of NO and NO2 in the three periods generally showed a bimodal pattern, albeit less pronounced in the initial and clear-out periods (Fig. 2). The two peaks likely arise as a consequence of increased traffic volumes at the start and end of the day coupled with boundary layer height changes in the early morning and into the evening (Lee et al., 2020). The average concentrations of NO2 during 05:00–10:00 UTC+1 were 10.8 ppbv in the O3 period, which was considerably higher than the concentration of 3.9 ppbv in the initial period and 3.4 ppbv in the clear-out period. The low level of NO in the O3 period highlights the rapid consumption of NO via photochemical processes. The oxidation of CO is an important source of HO2 in the atmosphere (Chen et al., 2020), which is in the range of 82.5 to 134.2 ppbv here, with little difference between periods. The diurnal profiles of O3 peaked at 15:00, with maximum hourly concentrations of 31.6, 67.2, and 30.4 ppbv in the initial, O3, and clear-out periods, respectively. Slight decreases in O3 were observed during the nighttime (00:00–05:00 UTC+1), indicating enhanced NO titration effects.

The detailed VOC composition in the three periods is presented in Fig. S1. Concentrations of TVOCs were 19.4 ± 8.4, 48.0 ± 18.8, and 23.5 ± 12.5 µg m-3 in the three periods, respectively. Alcohols, represented mainly by methanol and ethanol, were the predominant group that contributed 40.3 %–47.4 % of over-measured VOC mass. This was followed by alkanes (21.4 %–24.6 %) and ketones (16.3 %–17.3 %). Contributions of aldehyde (acetaldehyde), aromatics, alkenes, and acetylene were low, ranging from 1.0 %–9.4 % of the total mass. The average mixing ratios of the top 10 species in selected periods at the Birmingham supersite are listed in Table S4. The top 10 species were represented by methanol, acetone, ethanol, acetaldehyde, and C2–C4 alkanes across the initial period, O3 period, and clear-out period. The top individual species contributing to the total VOCs were methanol (10.3 %–33.6 %) and acetone (15.5 %–17.1 %) regardless of the subdivided periods. The results highlight large emissions of ethane, propane, n-butane, and i-butane associated with natural gas (NG), liquefied petroleum gas (LPG), and propellant use and fuel combustion and evaporation. Ambient VOCs largely influenced by combustion-related sources (i.e., vehicle exhaust and coal combustion) generally show an alkane-dominated composition (Wu and Xie, 2017). Here, the composition and amount of VOCs observed were most likely influenced by non-combustion processes, such as volatile chemical product usage and industrial processes (Gkatzelis et al., 2021). Methanol was the most abundant VOC, with an average concentration of 4.1 ppbv, followed by acetone (2.0 ppbv), ethane (1.9 ppbv), ethanol (1.8 ppbv), and acetaldehyde (1.0 ppbv). The average ratio of ethene / ethane was 0.2 ± 0.1 over the campaign, which is considerably lower than those seen in polluted locations, e.g., Hong Kong SAR (China) (0.7 ± 0.1) (Wang et al., 2018) and Seremban (Malaysia) (1.1) (Zulkifli et al., 2022).

The general diel profiles for all selected VOCs, except for ethane, showed a bimodal pattern (Fig. S2). Concentrations were much higher during the night and lower during the day due to them being subject to photochemical losses during the daytime. The bimodal pattern is less apparent for methanol and acetone as they are abundant species originating from many anthropogenic sources in urban areas. For example, methanol was the most abundant species measured at a roadside in the UK using thermal desorption–gas chromatography coupled with flame ionization detection (TD-GC-FID) (Cliff et al., 2023). A separate study on gasoline and diesel vehicle exhausts reported that methanol and acetone were the largest OVOCs emitted (Wang et al., 2022). Gkatzelis et al. (2021) conducted a positive matrix factorization (PMF) analysis based on an observed VOC data set in New York City and concluded that acetone was the second most abundant species in the measurements and was mostly attributed to volatile consumer product emissions (90 %). (See Sect. 3.3 for further discussions on anthropogenic sources of OVOCs.)

The in situ O3 formation sensitivity was examined via the reaction rates of ozone precursors and an OH radical (OH reactivities; k(OH)) and RIR scales of ozone precursors, along with the chemical budgets of O3 formation and loss. In initial and clear-out periods, k(OH) exhibited consistent diurnal patterns, ranging from 2.4 to 5.9 s-1 (Fig. S3). In the O3 period, k(OH) reached 9.0 and 8.7 s-1 at approximately 07:00 and 20:00 UTC+1, respectively. A rapid increase in k(OH) was observed in the early morning (00:00–06:00 UTC+1). VOCs and model-generated species represented 60.5 %, 65.7 %, and 56.7 % of the total k(OH) in the three periods, respectively. NOx and CO only contributed 10.2 %–27.9 % to the total k(OH). Among the VOC groups, alcohols exhibited the largest k(OH) in all periods, accounting for 5.0 %–6.9 % of the total k(OH). The diurnal production and loss of O3 are shown in Fig. S4. The oxidation and photolysis of VOCs promoted the production of RO2, and NO+RO2 contributed 47.7 % of the O3 production pathways in the O3 period and 36.2 % and 39.8 % in the initial and clear-out periods, respectively. Considering O3 destruction, OH + NO2 was the most important pathway during morning (08:00–12:00 UTC+1), accounting for 73.5 %, 55.4 %, and 59.4 % of the O3 destruction pathways in the three periods. The dominant OH + NO2 contribution to O3 destruction suggested that the in situ O3 productions in all three periods were sensitive to VOC emissions to some extent.

In order to understand contributions to O3 formation from direct emissions and secondary formations of OVOCs, we developed two modeling scenarios where (1) all OVOC species were constrained to the observed mixing ratio and (2) all OVOC species were unconstrained. Scenario 2 allowed for secondary formations of OVOCs by oxidations of their precursor VOCs. As shown in Fig. S5, secondary formations of OVOCs had little impact on O3 formation in all periods. The simulation of O3 production using the box model without constraining observed OVOCs slightly underestimated the average daily maximum O3 mixing ratio and P(O3) compared to the scenario with all observed OVOC species constrained. The underestimation for the average daily maximum mixing ratio of O3 was 4.8 %, 6.9 %, and 5.1 % in initial period, O3 period, and clear-out period, respectively. In this case, the underestimation of the average daily maximum P(O3) was 5.1 %, 6.0 %, and 9.3 % in the three periods, respectively. The results demonstrated that in the Birmingham case study, primary emissions of OVOCs played a central role in the in situ ozone production.

The relative incremental reactivity of NOx, CO, and anthropogenic VOCs (AVOCs, i.e., all measured VOCs except for isoprene) is shown in Fig. 3. The in situ O3 production was most sensitive to anthropogenic VOCs with the highest positive RIR values (0.44–0.49). This is as anticipated, given earlier analyses demonstrating their role in determining k(OH) and O3 production. The low RIR (0.03–0.07) for CO in all three periods indicated a minor contribution of CO oxidation to O3 production. The high RIR (0.24) for NOx was only observed in the O3 period. Acetaldehyde showed the highest positive RIR (0.17–0.19) among the AVOCs, suggesting that the photolysis and oxidation of acetaldehyde were a limiting factor for O3 formation. The important role of carbonyl compounds in atmospheric photochemistry has also been reported in previous studies, contributing up to 59.3 % of the O3 formation in ambient environments in China, the United States, and Brazil (Qin et al., 2021; Q. Liu et al., 2022; Edwards et al., 2014). Alkanes and alcohols exhibited lower RIR values (0.02–0.04) despite their high mass concentrations.

The trends in anthropogenic VOC emissions from 1990–2019 estimated by the NAEI are shown in Fig. S6. Over that period, the annual national emissions decreased by ∼ 69.0 %, from 2941 kt in 1990 to 911 kt in 2019. The reduction is partly attributed to more stringent controls for gasoline vehicle emissions, both tailpipe and evaporative/fugitive. In 2019, VOC emissions from on-road transport and fuel fugitive losses accounted for only 3.3 % and 13.7 % of the total mass of VOC emissions compared to 29.1 % and 26.9 % in 1990. Efforts have also been directed towards controlling industrial processes, commercial solvent usage, and combustion emissions, resulting in reductions of 66.8 %, 48.9 %, and 20.7 %, respectively, over that period. However, contributions from solvent usage to total VOC emissions over 1990–2019 showed only modest reductions in the 1990s and 2000s and, indeed, small increases in the most recent years (Fig. 4b). This slight growth in solvent usage is due to increasing emissions from solvent use in consumer products such as decorative products, aerosols, personal-care products, and detergents (NAEI, 2021). Solvent usage has become the largest contributory sector (33.7 %) to VOC emissions by 2019, followed by industrial processes (16.0 %).

As shown in Fig. 4a, the VOC speciation over the 1990–2019 period was dominated in mass terms by contributions from alkanes and alcohols, the former decreasing as gasoline sources declined and the other increasing as non-industrial solvent and food and drink industry processes' emissions followed a different pattern. Alkane emissions fell from 46.6 % to 30.6 % over that period. Further reductions in alkane emissions are expected from policies that aim to phase-out sales of new internal combustion engine vehicles in the UK (and in many other places) by 2035. The growth in the relative contributions of alcohols was primarily driven by increases in emissions of methanol and ethanol and, to a lesser extent, in 1-butanol and 2-propanol (Fig. 4c).

Figure S7 shows UK emission trends of individual species from different VOC classes. These highlight a national trend, present since 1990, of decreasing emissions of ethane associated with natural gas leakage; reductions in toluene and propane associated with on-road gasoline evaporation; and reductions in benzene, ethene, and acetylene associated with tailpipe exhaust. Meanwhile, the reduced emissions have been accompanied with increases in emissions of methanol and ethanol. The increase in methanol is largely attributed to increased emissions from car-care products (i.e., non-aerosol products) (Monks et al., 2020). The increase in ethanol is due to increased domestic and industrial solvent usage.

The NOx, CO, and VOC speciation within the NAEI for each of the six major emission sectors was used to assign proportional sectoral contributions to the VOCs observed in Birmingham and hence to ozone production in the case study (Table S5). Isoprene was excluded from the analysis as it is assumed to have an entirely biogenic source. The six sectors are road transport (both of on-road exhaust emission and evaporative losses of fuel vapor), industrial processes, combustion, solvent usage, and fuel fugitive and agriculture emissions. This forms a key assumption that the VOCs at the observation site are affected directly and in the same proportion as VOCs reported in national amounts in the NAEI. We make this assumption since it provides a reasonable starting point for understanding how each VOC sector may influence O3 production during a case study event; however, ozone formation might be sensitive to differing regional distribution in speciation. The attribution of VOC sources based on the NAEI data can be thought of as representative of this case study as a typical urban environment, but it might not hold for cities near large industrial VOC sources (i.e., oil refinery and industrial production sites) since they can significantly affect the composition and chemical reactivity of ambient VOCs.

Figure S8 shows the modeled RIRs for these sources in the initial, O3, and clear-out periods. All the sources generally showed higher RIR values in the O3 period. Road transport exhibited the highest positive RIR values in all periods (0.30–0.36), followed by industrial processes (0.06–0.09) and solvent usage (0.05–0.07). Despite being a relatively minor contributor to the mass of national VOC emissions (only 3.3 % of the total in 2019), road transport VOCs still played the most important role in local ozone photochemical chemistry in this case study.

Figure 5a shows the changes in P(O3) during the O3 period from 08:00–16:00 UTC+1, which might arise as a result of reductions in the individual sectors described above. For this analysis, emission changes in these sectors can be obtained by reducing model-constrained concentrations of VOCs, NOx, and CO according to their contributions arising from individual emission sectors. This is a thought experiment, where under 2019 general observed atmospheric conditions (for, e.g., NOx and CO), each of the VOC source sectors is then further reduced in isolation (from 2019 levels) and the effects on ΔP(O3) are evaluated. Based on these scenarios, reducing emissions from the individual sectors all resulted in decreased P(O3), as would be anticipated. Reducing ozone precursors arising from road transport would lead to a decreased P(O3) of ∼ 1.71 ppbv h-1 if that sector could be 100 % abated in the case study. This is expected because road transport is a source of photochemically reactive VOCs, including aromatics, aldehyde, and short-chain alkanes/alkenes. Other sectors showed more modest effects, with reductions in solvent-related VOCs being the next most significant lever to control ozone. Fully abating emissions of all industrial and solvent process emissions only resulted in a decreased P(O3) of ∼ 0.35 ppbv h-1 largely because they are dominated by ethanol and methanol with relatively low RIR values. Considering the real-world changes in VOC emissions over the period of 1990–2019, the very major reductions in road transport emissions have led to the largest effects on reducing P(O3) (Fig. 5b). Whilst there have also been some very large reductions (94.6 %) in fuel fugitive emissions, the impact on P(O3) reduction is modeled to have been relatively modest, being similar to industrial processes and solvent usage. It is important to acknowledge the limitation of this analysis. In the real world, reductions in ambient VOCs from the NAEI sectors are affected by photochemical loss rate and advection processes, potentially altering the proportion of VOCs that would be observed with each sector reduction at the measurement site. This would potentially be an important consideration if instantaneously radical budgets were being evaluated, but it is a less significant issue when integrating ozone production effects.

Further model runs were performed to better understand the impacts of the shift between alkanes and alcohol species on P(O3), given trends showing decreasing alkane emissions and increasing alcohol emissions during the 1990–2019 period (Fig. 4). The modeled alkane concentrations in the case study were reduced by 10 %, 30 %, 50 %, 70 %, 90 %, and 99 %. This represents a downward trajectory in alkane emissions that would be anticipated as gasoline vehicles are slowly retired. Two further scenarios were then developed to sit alongside these reductions in alkanes. Firstly, the concentration of ethanol was increased to keep the overall total VOC concentrations in the model under case study conditions unchanged. Second, the concentrations of both ethanol and methanol were scaled upwards to keep the total VOC concentration unchanged. As shown in Fig. 6, reductions in alkanes alone resulted in decreased P(O3) to a maximum of ∼ 0.26 ppbv h-1 if fully abated. If that alkane reduction was balanced with increased ethanol and methanol, then ΔP(O3) is reduced by a maximum of 0.19 ppbv h-1. If alkane reductions were balanced by increasing ethanol alone, then P(O3) still decreases but only up to 0.07 ppbv h-1.