We estimated the emissions of NMVOCs from 1990 to 2017 following the bottom-up framework of the MEIC model (available at http://www.meicmodel.org/, last access: 9 January 2019). The emissions were calculated based on a technology-based methodology, as described in detail by earlier papers (Zhang et al., 2009; Zheng et al., 2014, 2018; Liu et al., 2015; Li et al., 2017). Briefly, emissions for stationary sources were estimated based on the “emission factor” method following Eq. (1) (Zhang et al., 2009): 1Ei=∑j∑kAi,j,k∑mXi,j,k,mEFj,k,m, where i denotes the administrative unit of China, j is the source category in the classification system, k represents the fuel type for combustion-related sources or products for industrial processes, and m is the technology of fuel combustion or industrial production. E represents the estimated emissions, which are integrated by A (activity rate), X (technology distribution ratio) and the EF (emission factor) for each emitting source. The EF is determined based on the raw unabated emission factor (EFraw), the penetration ratio (Cn) and the removal efficiency of the control technology ηn as follows: 2EF=EFraw∑nCn(1-ηn).

For power plants, NMVOC emissions were derived from the China coal-fired Power plant Emissions Database (CPED; Liu et al., 2015), which is developed based on detailed information on fuel type, fuel quality, combustion technology and pollutant abatement facilities for > 7600 power generation units nationwide in China. The improved emissions for the on-road transportation sector developed by Zheng et al. (2014) were integrated into the framework of MEIC, which estimated the vehicle population and emission factors at a county level.

A detailed four-level source classification system, representing sector, fuel and/or product, technology and/or solvent type, and end-of-pipe pollutant abatement facilities, was established by including over 700 emitting sources in the MEIC model. Only anthropogenic sources, excluding open biomass burning, aviation and international shipping, were considered. Emissions of biofuel burning in households were estimated in this inventory. We present a total of five sectors (power, industry, residential, solvent use and transportation) and 15 subsectors by combining 109 NMVOC emitting sources by fuel type, industrial product, solvent use, vehicle type and diesel engine in Table 1. The detailed source categories, activity rates, emission factors and references are given in Table S1. Both combustion processes using fossil fuel and biofuel were considered for boilers and stoves. The subsector of “oil production, distribution and refinery” includes the evaporative emissions during oil production, transfer, refining and refueling in oil stations.

Paint use was further divided from the solvent use sector and includes the paint use applied to architecture, vehicles, wood and other industrial purposes. The inter-annual market shares of waterborne and solvent-based paint were further taken into account for each source category. Printing ink, pharmaceutical production, pesticide use, glue use and domestic solvent use were separately calculated and grouped into the subsector of “solvent use other than paint”.

For on-road transportation, we set up a process-based calculation framework for gasoline and diesel vehicles classified into eight types covering both trucks and passenger cars and four performance categories (high-duty, medium-duty, light-duty and mini). Each emitting process, including pollutant exhaust in the running mode and NMVOC evaporation, was considered. China's emission standards matching pre-Euro I and Euro I to Euro V emissions standards during 1990–2017 were applied for each vehicle type, as listed in Zheng et al. (2018).

Following Li et al. (2014), we developed emissions by individual chemical species based on the profile-assignment approach. First, we established a “composite” profile database for China by integrating the local profiles and supplementing them with the SPECIATE v4.5 database for absent sources (Simon et al., 2010; available at https://www.epa.gov/air-emissions-modeling/speciate-version-45-through-40, last access: 21 April 2018). Then, regarding OVOCs, we reviewed the profiles for all combustion-related sources, including the combustion of coal, fuel oil, gasoline and diesel in the power, industry, residential and transportation sectors, and corrected the incomplete profiles that were absent from the OVOC fractions. Especially, OVOCs account for more than 30 % for the residential coal (31 %), biofuel use (23 %–33 %) and exhaust from heavy-duty diesel vehicles (32 %). Finally, we assigned the composite profile to each source by setting up the source linkage between the profiles and the inventory. The selected source profiles used in this work are presented in Table S1.

The detailed procedure for developing the composite profile database is illustrated in Li et al. (2014). Briefly, for sources for which local profiles are available and for which there are significant differences in technology or legislation between China and western countries, only local profiles are used; otherwise, all corresponding profiles are listed as “candidate” ones and included for further compilation of the composite profile database. The gathered local profiles cover the following major contributing sources: biofuel combustion (Tsai et al., 2003; Liu et al., 2008; Wang et al., 2009; Mo et al., 2016); coal combustion (Liu et al., 2008; Shi et al., 2015); asphalt production (Liu et al., 2008); oil production, handling and refinery (Liu et al., 2008); vehicle varnish paint (Yuan et al., 2010); printing ink (Yuan et al., 2010; Zheng et al., 2013; H. Wang et al., 2014); gasoline evaporation (Liu et al., 2008; Zhang et al., 2013; Wu et al., 2015); polypropylene production (Mo et al., 2015); gasoline vehicles (Duffy et al., 1999; Schauer et al., 2002; Liu et al., 2008); and diesel vehicles (Schauer et al., 1999; Liu et al., 2008; Yao et al., 2015; Mo et al., 2016). In addition to the source profiles used for speciation in Li et al. (2014), we updated profiles that were newly added in SPECIATE v4.5 and local profiles measured in recent years (Zhang et al., 2013; H. Wang et al., 2014; Shi et al., 2015; Wu et al., 2015; Yao et al., 2015; Mo et al., 2015, 2016). Profiles for approximately 60 sources were updated, including power plants, paint production, industrial coal use, gasoline evaporation, coke production, biofuel combustion in residential stoves, glue use, paint use, vehicles and off-road diesel engines. Compared to Li et al. (2014), the update in source profiles results in higher mass fractions for alkenes, alkynes and OVOCs and lower contributions for aromatics.

Due to the improper sampling and analysis method used in profile measurements, several local profiles lack significant OVOC fractions (Li et al., 2014). We extended the revision to all combustion processes and corrected their profiles by appending the component of OVOCs with fractions derived from the “complete” profiles for the same source. After OVOC correction, all candidate profiles were averaged by species to establish the composite profile database. The equation of the OVOC revision is as follows: 3Xrevised(i,j)=Xori(i,j)∑jXori(i,j)×(1-Xovoc(i)‾), where Xrevised is the mass fraction of species j in the revised profile for source i and Xori is the original mass fraction; Xovoc‾ represents the calculated mean of the OVOC proportion for all candidate profiles that have measured OVOCs.

OFP is a widely used scale to investigate the potential ozone production due to emissions of NMVOCs and has been applied to guide the establishment of most cost-effective ozone mitigation measures (e.g., Song et al., 2007; Zheng et al., 2009). OFP for individual chemical species is calculated based on the mass and maximum incremental reactivity (MIR), which scales the ozone production potential for corresponding species: 4OFPi,j,k=EVOCi,k×Xi,j×MIRj, where i, j and k represent the source, chemical species and province, respectively. OFP is the ozone formation potential, EVOC is the total NMVOC emission estimate, X represents the mass fraction for species j emitted from source i, derived from the composite profiles in this study, and MIR is the maximum incremental reactivity scale for species j (Carter, 1994, 2010).

China's NMVOC emissions have shown a continuously increasing trend. NMVOC emissions were estimated to be 9.8 Tg in 1990; then they increased to 14.5 Tg in 2000, 20.3 Tg in 2005, 25.1 Tg in 2010 and 28.5 Tg in 2017, with annual growth rates of 4.0 % (1990–2000), 7.0 % (2000–2005), 4.3 % (2005–2010) and 1.8 % (2010–2017). Emissions by sectors and subcategories for each sector during 1990–2017 are shown in Figs. 1 and 2, respectively (details in Table 1). Industry (+6.0 Tg, +368 %) and solvent use (+10.7 Tg, +850 %) are the main sectors driving the total emission increase during 1990–2017. The transportation emissions first increased rapidly and then decreased, peaking at 6.5 Tg in 2008. Residential emissions gradually decreased during the last decade, dominated by biofuel combustion, leading to a 24 % emission decline in 2017 compared to 1990. As a result, the proportions by sector for national emissions have changed, with growing contributions from industry (17 %–27 % for 1990–2017) and solvent use (13 %–42 %), shrinking contributions from residential use (55 %–14 %), and stable contributions from the transportation sector (15 %–17 %).

The emission trends by subcategory for each major sector (industry, residential, solvent use and transportation) are further illustrated in Fig. 2. Coal combustion, the chemical industry and oil-related processes are the main contributors to the industrial emission changes. Despite the gradual decline in industrial coal use since 2012, industrial processes still show a continuously increasing trend, driven by the chemical industry. The rapid emission growth of solvent use can be attributed to several sources, including paint use and other various solvent applications. Both the residential and transportation sectors have started to decrease in recent years. The significant reductions in crop residue use in residential stoves are consistent with the socioeconomic development in rural areas of China. Vehicular emissions show a sharp increase from 1990 to 2008, driven by the increase in vehicle ownership, then decrease gradually because of the stage-by-stage implementation of volatile organic compound (VOC) abatement measures, especially for gasoline-fueled passenger cars.

Emissions by individual chemical species were developed based on the total NMVOC emissions (as illustrated above) and the mass fractions derived from source profiles. Figure 3d presents the sectorial emissions of the top 30 species in OFP-descending order in 2017, which together account for > 80 % of the total OFP. Toluene is the largest contributor to OFP, with emissions estimated to be 3.4 Tg (12.0 % of the total), followed by m- and p-xylene (1.4 Tg, 5.1 %), ethylene (1.2 Tg, 4.3 %), o-xylene (0.71 Tg, 2.5 %), propylene (0.41 Tg, 1.4 %) and formaldehyde (0.41 Tg, 1.4 %; in OFP-descending order). The different distribution patterns between mass and OFP are attributed to the variations among chemical species in the reactivity scales of MIR. In 2017, toluene, xylene (including all isomers of xylene, i.e., m- and p-xylene and o-xylene), and ethylbenzene were dominated by solvent use, while ethylene, propylene and formaldehyde were mainly contributed by industrial and residential sources.

We present the emissions by sector for these 30 species in 1990, 2000 and 2010 with the calculated OFP as references in Fig. 3a–c. Notably, aromatics, including toluene and xylene, showed dramatic emission increases, whereas alkenes (ethylene and propylene) and OVOCs (formaldehyde and acetaldehyde) showed moderate changes. For the first 10-year period (1990–2000), ethylene contributed approximately 10 % by mass and 20 %–25 % by OFP, ranking first among all identified species. Transportation drove up emissions of all chemical species during 1990–2000. In the second 10-year stage (2000–2010), toluene surpassed ethylene, becoming the largest contributor to the total emissions, with an increase in proportion from 6 % to 9 %. Similar increasing trends were estimated for other aromatic species linked to the solvent use and industry sectors. Solvent use has continuously driven up related species such as toluene and xylene since 2010 due to the increasing demand and relatively limited penetration of control measures. Meanwhile, ethylene, acetylene and formaldehyde started to decrease as a result of reductions in residential biofuel use and vehicle exhaust.

Figure 4 further illustrates the emission and OFP trends for six representative species (ethane, ethylene, toluene, xylene, formaldehyde and acetylene) of chemical groups (alkanes, alkenes, aromatics, OVOCs and alkynes). Apart from ethane and acetylene, all species are identified with high contributions to ozone formation during 1990–2017 (see Fig. 4b). Sharp growth was estimated for toluene and xylene, with 6-fold (> 2 Tg) higher emissions in 2017 than in 1990, mainly driven by solvent use as illustrated above. The emissions of ethane first increased then showed slight decrease and experienced a 39 % (+178 Gg) increase in 2017 compared to 1990. Ethylene emissions rose rapidly in the first 16 years and then declined, increasing by 11 % (+117 Gg) from 1990 to 2017. The declining trend of ethylene in recent years can be attributed to the residential combustion activities. In contrast to the overall growing trend, formaldehyde decreased by 25 % (-137 Gg), leading to even lower emissions in 2017 than in 1990 because of the reduced use of biofuel in residential stoves.

As shown in Fig. 4, emission fractions by chemical group changed significantly, with reduced proportions of alkenes and OVOCs and increased shares of aromatics and alkanes. In 2017, aromatics were the largest contributing chemical group to emissions, accounting for 33 % of the total. The mass fractions for alkenes and OVOCs gradually decreased from 20 % and 23 % in 1990 to 13 % and 15 % in 2017, respectively. Figure 5 decomposes the driving forces of emission changes by chemical group and sector from 1990 to 2017. During 1990–2000, the 48 % (+4.6 Tg) emission increase was mainly attributed to alkanes (+1.8 Tg), aromatics (+1.3 Tg) and alkenes (+0.91 Tg) contributed by the transportation sector. Since 2000, activity rates from solvent use and industry have grown rapidly along with the economic development, leading to large emission increases in alkanes and aromatics. For the period of 2000–2010, aromatics and alkanes accounted for 36 % and 26 % of the total emission growth, respectively, dominated by solvent use and industrial sources. Solvent use and industrial processes continuously promoted the emissions of aromatics (+2.8 Tg) and alkanes (+1.5 Tg) in recent years (2010–2017), but the increasing trend was lowered by the declined emissions of transportation and residential sectors (-2.9 Tg) that benefited from the penetration of control measures and transition of fuel types. As a result, a much lower emission growth ratio of 13 % was estimated for 2010–2017, compared to a > 40 % increase for previous decades (see Fig. 5).

The national OFP shows a persistent increasing trend from 38 Tg of O3 to 100 Tg of O3, at a growth factor of 2.6, during 1990–2017 (Fig. 1). Due to their large emission amounts and high ozone-producing chemical reactivity (scaled by MIR), ethylene, toluene, xylene and propylene are estimated to be the key NMVOC precursors in ozone formation during the last decades (see Fig. 3). The rankings of OFP contribution by individual species have changed over time, with increasingly important roles played by reactive aromatic species (toluene, xylene and ethylbenzene) and decreasing contributions from alkenes (ethylene, propylene and butenes) and OVOCs (formaldehyde and acetaldehyde). Specifically, during 1990–2017, toluene, xylene, 2-methyl-2-butene, ethylbenzene, 2-butene, 1,2,4-trimethylbenzene, propylene, 2-pentene and formaldehyde contributed most to the OFP trend. Transportation, solvent use and industry sectors were the main contributors to the OFP increase and partly suppressed by the decline in biofuel use.

As presented in Fig. 4, xylene (sum of m-xylene, p-xylene and o-xylene) and toluene surpassed ethylene to become the two largest OFP contributors at present. In 2000, alkenes contributed 47 % to the total OFP, while aromatics accounted for 24 %. Driven by the increasing emissions of aromatics since 2000, the OFP contributions of aromatics (43 %) are now even higher than those of alkenes (37 %). Alkenes and aromatics together represented 80 % of the total OFP in 2017. Among the top 30 species contributors, the OFP proportions of aromatics are even higher, increasing from 20 % in 1990 to 50 % in 2017. The significant role of aromatics highlights the importance of setting up corresponding measures to suppress ozone formation.

We present the components driving the OFP growth by chemical group and sector from 1990 to 2017 in Fig. 5. As illustrated above, during 1990–2000, the rapidly increasing number of vehicles along with economic development introduced large quantities of reactive alkenes and aromatics as well as OFP (+45 %). From 2000 to 2010, solvent use became the largest contributor to the OFP change, accompanied by the boom in aromatic emissions. During this period, on-road gasoline vehicles and industry also played important roles in the OFP increase caused by alkenes. The OFP contribution of alkanes was small even though the emission increase was significant; this is due to the low chemical reactivity of alkanes. Notably, during the most recent years (2010–2017), emissions of several key source categories have started to stabilize or even decrease, significantly mitigating the increased OFP caused by solvent use and industrial processes. OFP reductions are mainly attributed to alkenes (-3.3 Tg of O3) and OVOCs (-2.5 Tg of O3) associated with the sectors of residential biofuel combustion and transportation.

By allocating the emissions into grids based on spatial surrogates, we depicted the spatial distributions of OFP at a spatial resolution of 0.25∘×0.25∘ in Fig. 6. Power plant locations derived from the CPED and verified using Google Earth were used to distribute the corresponding emissions; rural population density and road maps were used as proxies for residential biofuel combustion and on-road vehicles, respectively; urban and total population datasets were used to allocate the emissions of other sources. The spatial patterns of OFP are in general consistent with the observed ozone maps (see Fig. 6). Significant signals of urbanization are especially prevalent over China, demonstrating the potential severe ozone pollution in densely populated regions such as eastern China and the Sichuan basin. We further analyzed the OFP trends for four key regions with severe ozone pollution in China, i.e., Beijing–Tianjin–Hebei (BTH), the Yangtze River Delta (YRD), the Pearl River Delta (PRD) and the Sichuan Basin (SCB) for 2013–2017. Compared to the values of 2013, OFP for BTH, YRD, PRD and SCB showed minor growth ratios of 2 %, 8 %, 9 % and 2 % in 2017, respectively. Especially, the mitigation measures covering various sources implemented in PRD achieved a 7 % decrease in OFP in 2014–2017. This suggests that due to the absence of effective mitigation measures regarding NMVOCs, the potential ozone production for megacity clusters has been stable in recent years in contrast to the dramatic reductions of NOx and primary PM2.5.

We depicted the trend of summer mean (June–July–August) maximum daily 8 h average (MDA8) ozone concentrations observed by the ground monitoring network over China since 2013 in Fig. 6. In contrast to the flat trend of OFP during the same period, significant increases in ozone were observed in northern, central and southwestern China. Previous studies indicated that VOC-limited conditions prevail in megacities of China, where more ozone can be produced as a result of dramatic NOx emission decline (Gao et al., 2017; K. Li et al., 2018). The effective reduction of PM2.5 in recent years favoring the penetration of ultraviolet light to the surface may also lead to greater ozone production. Based on model simulation, K. Li et al. (2018) demonstrates that the sharp decrease in NOx emissions (∼ 21 %) in VOC-limited regions, and the ∼ 40 % reduction of PM2.5 concentrations slowing down the aerosol sink of hydroperoxy radicals, drove the rise of surface ozone in China during 2013–2017. As indicated by above analyses, the lagging mitigation measures of NMVOCs compared to other criteria for pollutants in most of urban China have advanced ozone production through nonlinear chemistry in the gas phase and/or multiphase chemistry between gases and aerosols. Designing cost-effective mitigation measures for NMVOCs accompanying NOx, CO and PM2.5 is quite urgent and crucial for ozone control in the near future.

The NMVOC emissions in China estimated in our work are compared with previous estimates in Fig. 7. The increasing pattern since 2000 is generally consistent among different long-term emission inventories (Kurokawa et al., 2013; Wu et al., 2016). Our estimates are slightly lower than the values of Wu et al. (2016), Wu and Xie (2017), and the Regional Emission inventory in ASia version 2.1 (REAS v2.1; Kurokawa et al., 2013) but are higher than those of Wei et al. (2014) and Bo et al. (2008). The emissions estimated by the various inventories for the most recent years, i.e., since 2010, agree relatively well, with variations of 10 %–22 %. For regional emission inventories, as the data sources of activity rates are generally consistently obtained from official statistics, we can attribute the emission differences to the distinct source classification system and assignment of emission factors. For the emission inventory at a global scale, the calculated growth rate of NMVOC emissions in the Emissions Database for Global Atmospheric Research (EDGAR; EC-JRC/PBL, 2011; Crippa et al., 2018) is much slower than our estimates. EDGAR emissions are slightly lower than our estimates after 2000 but much higher in the 1990s. The reasons for these differences are complicated and should include inconsistency in source categories, data sources of activity rates and emission factors (M. Li et al., 2018).

The uncertainties in the total NMVOC emissions in China were estimated to be at a moderate level of ±68 %–±78 % (Zhang et al., 2009; Kurokawa et al., 2013), mainly arising from the lack of reliable data for scattered areal sources. Solvent use and industrial processes are sectors with high uncertainty according to Wu et al. (2016). On the other hand, selection and application of source profiles can lead to differences of over a magnitude of 3 in quantifying individual species, representing the largest uncertainty sources during speciation (Li et al., 2014).

Uncertainties in the composite profile are related to the accuracy of individual profiles applied for speciation. We calculated the uncertainties in the composite ones via the propagation of errors approach. For each species included in sources, the standard error (SE) for all profiles that had measurements was calculated to represent the “true” mass fraction error based on limited samples. If only one profile is used, expert judgment was used to estimate the profile error. We assumed the coefficients of variation (CVs; i.e., standard deviation divided by the mean) of the SPECIATE v4.5 profiles with overall qualities of A–E to be ±5 %–500 % and assigned local source profiles at CVs of ±5 %–±15 % according to the measurement year based on expert judgment. Errors were added linearly for sources that shared profiles and then combined in quadrature into subsectors. Then the profile uncertainties were calculated to be 1.96 times the CV at the 95 % confidence interval.

Source profiles contribute large uncertainties in determining emissions of individual chemical species and further species-specific OFP (see Fig. S1). Despite using similar sampling and analytical methods, the measured profiles show significant diversity among different studies, varying with fuel type, combustion technology, end-of-pipe control facilities, solvent components, etc. For abundant components of toluene, ethylene, m- and p-xylene, and propylene, the uncertainties in the mass fractions are in the range of 7 %–453 %, with averages of 74 %–101 %, showing comparable accuracy to the emission estimates. Profiles of vehicular sources and paint use show low uncertainties for all chemical species, including on-road gasoline, on-road diesel, off-road diesel and industrial paint use, because of the inclusion of reliable local source profiles. The uncertainty matrix shown in Fig. S1 highlights the need for more measurements and further analyses for important sources (species), especially chemical industry (o-xylene and benzene), other industrial processes (toluene, xylene, formaldehyde and 2-methyl-2-butene), residential biofuel combustion (toluene, xylene, ethylbenzene, cis-2-butene and butyl cellosolve) and waste treatment (xylene, ethylene and formaldehyde). Yet, it is difficult to quantify the uncertainties for trace gases such as polycyclic aromatic hydrocarbons (PAHs) and halohydrocarbons that have high molar mass but have not been included in standard sampling and analytical protocols in current measurements.

The inter-annual variations in profiles are yet to be investigated, and more local source profiles with complete information of chemical species are still in need. An improved linkage between the source profile matrix and the source, province and temporal information will be important in improving the accuracy of emission estimates.

It should be addressed that our inventory only includes anthropogenic sources (including biofuel) and excludes open biomass burning, which may introduce bias for analyses covering all source types. Based on the most recent work (Yin et al., 2019), emissions of open biomass burning in China are 1.12–2.16 Tg of NMHC, corresponding to 2.90–5.60 Tg of NMVOCs by applying an averaged OVOC fraction of 61.4 % (8753 in SPECIATE 4.5; Andreae and Merlet, 2001) during 2003–2017. Compared to 17.6–28.5 Tg of NMVOCs from anthropogenic sources during the same period, we captured > 83 % of the total emissions, and the total NMVOC emissions would be 32.8 Tg in 2017, with large emission decreases (-7 %) for 2015–2016.

Ozone pollution has become increasingly severe in China, especially in megacities (Gao et al., 2017; Lu et al., 2018). Both ground measurements and satellites have detected an increasing trend of tropospheric ozone concentrations over recent decades due to emissions of precursors, tropospheric chemistry, penetration from the stratosphere and meteorological changes (Verstraeten et al., 2015; Wang et al., 2016; K. Li et al., 2018).

Our estimates of the speciated NMVOC emissions and underlying indications will be important for establishing the most cost-effective mitigation measures for ozone. The effective actions to control fine PM in China have gained significant achievements since 2013, while the ozone problem has not been fully addressed by the Chinese government. Based on our estimates, the implementation of control measures for vehicles, industrial and residential sources has led to emission reductions of alkenes, alkanes and aldehydes. However, due to the absence of effective control for evaporative sources, large amounts of aromatics have been emitted from the condensed phase into the atmosphere since 2000. Paint use, the chemical industry, petroleum production and distribution, and other solvent use sources are the main contributors to the changes since 2010 and remain inefficiently controlled nationwide. These sources should be addressed and controlled more stringently in the next step. As urban China mainly has VOC-limited conditions, mitigation of NMVOC emissions will suppress ozone formation effectively. China urgently needs to set up goals and enact more stringent legislation to control NMVOC emissions and NOx, CO and PM2.5 emissions to prevent further potential ozone pollution and adverse effects on human health.