Evaluation of Anthropogenic Emissions and Ozone Pollution in the North China Plain : 1 Insights from the Air Chemistry Research in Asia ( ARIAs ) Campaign

Abstract. To study the air pollution in the North China Plain (NCP), the Air Chemistry Research in Asia (ARIAs) campaign conducted airborne measurements of air pollutants including O3, CO, NO and NO2 in spring 2016. High concentrations of pollutants, > 100 ppbv of O3, > 500 ppbv of CO, and > 10 ppbv of NO2, were observed throughout the boundary layer during the campaign. CMAQ simulations with the 2010 EDGAR emissions can capture the basic spatial and temporal variations of ozone and its major precursors such as CO, NOx and VOCs, but significantly underestimate their concentrations. Observed emission enhancements of CO and NOx with respect to CO2 suggest the existence of combustion with high emissions such as biomass burning in the NCP. The comparison with emission factors from the 2010 EDGAR emission inventory indicates that the contribution of combustion with high emissions has been overestimated. Differences between CMAQ simulations with 2010 emissions and satellite observations in 2016 can reflect the change in anthropogenic emissions. NOx emissions decreased in megacities such as Beijing and Shanghai confirming the effectiveness of recent control measures in China, while in other cities and rural areas NOx emissions slightly increased, e.g., CMAQ predicts only ~ 80 % of NOx observed in the aircraft campaign area. CMAQ also underestimates HCHO (a proxy of VOCs, by ~ 20 %) and CO (by ~ 60 %) over the NCP, suggesting adjustments of the 2010 EDGAR emissions are needed to improve the model performance. HCHO/NO2 column ratios derived from OMI measurements and CMAQ simulations show that VOC-sensitive chemistry dominates the ozone photochemical production in eastern China, suggesting the importance of tightening regulations on VOCs emissions. We adjusted EDGAR emissions based on satellite observations, conducted sensitivity experiments of CMAQ, and achieved better model performance in simulating ozone, but underestimation still exists. Because of the VOC-sensitive environment in ozone chemistry over the NCP, future study and regulations should focus on VOCs emissions with the continuous controls on NOx emissions in China.



Introduction
With rapid economic growth in the past three decades, the consumption of energy in China increased dramatically (Zhang and Cheng, 2009;Guan et al., 2018;Shan et al., 2018).
Fossil fuels dominate total energy consumption, with coal still accounting for more than 50% of the carbon dioxide (CO 2 ) emissions in China (Shan et al., 2018).This drastic increase in fossil fuel energy consumption is accompanied with deterioration of air quality (Chan and Yao, 2008;Fang et al., 2009), posing a threat to public health (Tie et al., 2009;Kan et al., 2012;Chen et al., 2013;Lelieveld et al., 2015).Particulate matter (PM) pollution, especially PM 2.5 in the North China Plain (NCP), drew public concern and governmental actions (He et al., 2001;Ye et al., 2003;Wang et al., 2005;Sun et al., 2006;Yang et al., 2011;Zhang et al., 2012;Zhang et al., 2013).PM pollution also has complex interactions with the planetary boundary layer (PBL) and its evolution, which can further degrade the air quality (Guo et al., 2016;Li et al., 2017b).Recent studies showed that tropospheric ozone (O 3 ) pollution increased in China which exacerbated its complex air pollution problem (Xue et al., 2014;Verstraeten et al., 2015;Wang et al., 2017b;Ni et al., 2018).
Elevated ozone concentrations have adverse impacts on both human health (WHO, 2003;Anderson, 2009;Jerrett et al., 2009) and the ecosystem (Adams et al., 1989;Chameides et al., 1999;Ashmore, 2005).Tropospheric ozone absorbs thermal radiation and acts as the third most important anthropogenic contribution to radiative forcing of climate (Ramanathan and Dickinson, 1979;Lacis et al., 1990;IPCC, 2014).In the lower troposphere, the photolysis of ozone is an important source of atmospheric hydroxyl (OH) radicals that control the lifetimes of atmospheric species such as CO and volatile organic compounds (VOCs) (Logan et al., 1981;Thompson, 1992;Finlayson-Pitts and Pitts, 1999).Tropospheric ozone has a relatively long lifetime of several days to weeks (Stevenson et al., 2006;Young et al., 2013), leading to significant long-range transport of ozone and its precursors (Jacob et al., 1999;Derwent et al., 2004;Lin et al., 2008).Thus, investigation of ozone pollution in China is essential to support the national and international policy decision for air quality and the climate.Tropospheric ozone is produced through complex photochemical reactions of precursors including nitrogen oxides (NO x = NO + NO 2 ) and VOCs in the presence of sunlight (Haagensmit, 1952;Crutzen, 1974;Fishman et al., 1979;Seinfeld and Pandis, 2006).In China, sectors of power generation, industry, and transportation dominates the NO x emissions (Streets et ARIAs research flights and the A 2 BC surface observations provide a comprehensive dataset to thoroughly study the tropospheric ozone pollution and emissions of its precursors in China. In this study, we evaluate anthropogenic emissions and the ozone pollution in the NCP using a combination of aircraft measurements, surface monitoring data, satellite observations, and modeling results.The Environmental Protection Agency (EPA) Community Multiscale Air Quality (CMAQ) model was used to simulate the atmospheric chemistry for the ARIAs campaign.We evaluate the emission data by comparing with the aircraft measurements and satellite products, and then adjust emissions to improve the CMAQ performance.Lastly, we investigate the sensitivity of ozone production derived from CMAQ simulations and OMI observations and discuss the future ozone pollution in China.

Aircraft campaign in the NCP
With more than 250 million tons of iron and steel produced in 2016 (data from http://data.stats.gov.cn,accessed in September 2018), Hebei Province in the NCP is the most industrialized area in China.Due to its high emissions and proximity to megacities Beijing and Tianjin, the Beijing-Tianjin-Hebei area has experienced severe air pollution in the past decade (Zhao et al., 2013b;Wang et al., 2014b).In May and June 2016, the ARIAs aircraft campaign was conducted over Hebei Province to investigate the emissions, chemical evolution, and transport of air pollutants.The airborne campaign was coordinated with the A 2 BC field campaign in Xingtai (XT,37.18 °N,114.36 °E,182 m above sea level, ASL) and the NASA KORUS-AQ campaign to expand the study to East Asia.A Harbin Y12 research airplane (similar to the de Havilland Twin Otter) was employed to measure concentrations of air pollutants including O 3 , carbon monoxide (CO), CO 2 , NO 2 , and aerosol optical properties.The research airplane was located in Luancheng airport, hereafter referred to as LC (LC,37.91 °N,114.59 °E,58 m ASL), south of Shijiazhuang, the capital city of Hebei province with a population of 10 million.Eleven research flights were conducted during the ARIAs campaign (Fig. 1a).Vertical profiles of air pollutants from near surface (~100 m above ground level, AGL) to the free troposphere (> 3000 m) were conducted over LC, XT (the supersite of the A 2 BC campaign), Julu (JL,37.22 °N,115.02 °E,30 m ASL),and Quzhou (QZ,36.76 °N,114.96 °E,40 m ASL).
The airborne measurements of ozone were conducted using a commercially available analyzer (Model 49C, Thermo Environmental Instruments, TEI, Franklin, Massachusetts) (Taubman et al., 2006).NO 2 was measured using a modified commercially available cavity ringdown spectroscopy (CRDS) detector (Castellanos et al., 2009;Brent et al., 2013).Concentrations of CO and CO 2 were monitored with a 4-channel Picarro CRDS instrument (Model G2401-m, Picarro Inc., Santa Clara, CA), calibrated with CO/CO 2 standards certified at the National Institute of Standards and Technology (Ren et al., 2018).All the instruments were routinely serviced, calibrated and used for airborne measurements in the United States and China (Taubman et al., 2006;Dickerson et al., 2007;Hains et al., 2008;He et al., 2012;He et al., 2014;Ren et al., 2018;Salmon et al., 2018).Measurements of ambient air pollutants were made at 1 Hz frequency and synchronized with time, geolocation and altitude from the Global Position System (GPS).
In the ARIAs research flights, 28 whole air samples (WAS) were collected in vertical spirals at different altitudes from ~400 m to ~3500 m.The WAS were analyzed using gas chromatography (GC) with Flame Ionization Detection (FID) and Mass Spectroscopy (MS) by the College of Environmental Sciences and Engineering at Peking University.74 species of alkanes, alkenes/alkynes, aromatics, and halocarbons were identified and quantified for a study on ozone photochemical chemistry (see details in Benish et al., 2019).Detection limits for the compounds ranged from 2 to 50 pptv.Surface observation of trace gases including O 3 , CO, NO, and NO x were measured at the A 2 BC Xingtai supersite using analyzers manufactured by Ecotech (Wang et al., 2018b); detailed description of the analyzers is discussed in Zhu et al. (2016).
Surface HCHO concentrations were monitored using a formaldehyde analyzer (AERO LASER, Germany, Model 4021) based on fluorometric Hantzsch reactions (Gilpin et al., 1997;Rappenglück et al., 2010).All surface observations were collected as 1-min averaged data and processed to hourly mean values.

Satellite products
To evaluate the emissions and atmospheric chemistry in the NCP and greater East Asia, we used satellite observations of CO, NO 2 , and HCHO for May and June 2016.The Measurements of Pollution In the Troposphere (MOPITT) instrument onboard the NASA Terra satellite retrieved CO column contents with ~10:30 am local overpass time (Deeter et al., 2003).We used the latest version 7 MOPITT Level 3 daily gridded average products (1° × 1° spatial resolution, available at https://eosweb.larc.nasa.gov/project/mopitt/mop03j_v007)for the ARIAs campaign period (MOPITT Science Team, 2013).MOPITT thermal-infrared and near-infrared (TIR + NIR) products shows improved sensitivity to near surface CO in China (Worden et al., 2010).We used MOPITT near surface CO (~ 900 hPa) products and related averaging kernels (AKs) to evaluate the CMAQ results (Deeter et al., 2012).OMI, onboard the NASA Aura satellite, is a UV/Vis solar backscatter spectrometer in a polar sun-synchronous orbit with a ~1:35 pm local overpass time.With high spatial resolution (13 km × 24 km for the center at nadir) and nearly daily coverage, OMI has provided monitoring of trace gases and aerosol properties since 2005 (Levelt et al., 2006).The Version 3 OMI Level 2 NO 2 products (https://disc.gsfc.nasa.gov/datasets/OMNO2_V003/summary)(Krotkov et al., 2018) were used to evaluate the emissions and atmospheric chemistry in East Asia.Under clear sky, tropospheric NO 2 columns from OMI has precision of ~0.5 × 10 16 molecules cm -2 and an accuracy of ±30% (Krotkov et al., 2017).OMI HCHO Version 3 Smithsonian Astronomical Observatory (SAO) (https://disc.gsfc.nasa.gov/datasets/OMHCHO_V003/summary)Level 2 products were used in this study (Chance, 2007;González Abad et al., 2015).The precision of column HCHO is ~1.0 × 10 16 molecules cm -2 and SAO products have an accuracy of ±25-30% without cloud (Millet et al., 2006;Boeke et al., 2011).Data in OMI pixels affected by the row anomaly and contaminated by clouds were filtered out using quality flags for both NO 2 and HCHO columns.

Model set-up
We used CMAQ version 5.2 (EPA, 2017) to simulate atmospheric chemistry for the ARIAs campaign.The Weather Research and Forecasting (WRF) model Version 3.8.1 (Skamarock et al., 2008) was driven by the European Centre for Medium-Range Weather Forecasts (ECMWF) ERA-Interim products (ds627.0,https://rda.ucar.edu/datasets/ds627.0) (Dee et al., 2011) to simulate meteorological fields.Two domains with spatial resolution of 36 km and 12 km (Fig. 1b) were used to cover East Asia, with 35 layers from the surface to 50 hPa and ~20 layers in the lowest 2 km.Major physical options in WRF include the Rapid Radiative Transfer Model (RRTM) radiation scheme (Clough et al., 2005), the Pleim-Xiu surface layer and land surface model (Pleim and Xiu, 1995;Xiu and Pleim, 2001), the Asymmetric Convective Model (ACM2) boundary layer scheme (Pleim, 2007), the Kain-Fritsch cumulus scheme (Kain, 2004), and the WRF Single-Moment 6 (WSM-6) microphysics (Hong and Lim, 2006).The National Centers for Environmental Prediction (NCEP) ADP Global Surface and Upper Air Observational Weather Data (ds461.0and ds351.0,https://rda.ucar.edu)were used to perform observational and analysis nudging on all domains following the method developed for NASA aircraft campaigns (He et al., 2014;Mazzuca et al., 2016).WRF outputs were processed by the EPA Meteorology-Chemistry Interface Processor Version 4.3 (MCIP v4.3, released in November 2015) for emission processing and CMAQ simulations.

Anthropogenic emissions were from the Emissions Database for Global Atmospheric
Research Version 4.2 (EDGAR v4.2, 0.1° × 0.1° resolution) of year 2010, which are widely used for chemical transport modeling (European Commission, 2011).We used the EPA Sparse Matrix Operator Kernel Emissions (SMOKE) modeling system Version 4.5 (UNC, 2017) to project EDGAR emissions to the modeling domain.Emissions of air pollutants were speciated into Carbon Bond 05 chemical mechanism (Yarwood et al., 2005) and updated AERO6 aerosol module (Appel et al., 2013).The EDGAR v4.2 inventory has emissions for energy, industry, residential, and transport sectors.Without stack height information for power plants in the energy sector, we followed the approach developed in He et al. (2012) to locate these anthropogenic emissions at ~200 m above the surface as an approximation for averaged stack height and plume rise.We used the United States Geological Survey (USGS) 24 category land use dataset combined with the Biogenic Emission Inventory System (BEIS) emission factors table to generate the input files for the CMAQ inline biogenic emissions modeling.Biogenic emissions were estimated using the BEIS module inline in CMAQ (EPA, 2017).CMAQ v5.2 uses the updated Carbon Bond 6 (CB6r3) chemical mechanism (Yarwood et al., 2010) including improved chemistry mechanism for organic nitrates and peroxyacyl nitrates (PAN) chemistry and will lead to better performance for simulating Secondary Organic Aerosols (SOA) and tropospheric ozone in the United States (Appel et al., 2016).CMAQ was run with a coarse domain and a nested domain (Fig. 1b).Chemical initial and boundary conditions for the coarse domain were obtained from the default concentration profiles built in CMAQ (EPA, 2017).Results from the CMAQ coarse domain were used to generate boundary conditions for the nested domain.The WRF-CMAQ system was run from mid-April to June with the first 2 weeks as spin-up.Hourly concentrations of air pollutants were saved for further analysis and model evaluation.38.5°N, 114.0-115.5°E hereafter).We conducted vertical spirals over XT (the A 2 BC supersite), LC (the airport in south of Shijiazhuang), and two rural areas (JL and QZ) during the ARIAs research flights.Figure 3a summarizes vertical distributions and the mean profiles of air pollutants over XT, with mean O 3 concentrations of 80 ppbv in the lower atmosphere.We observed isolated plumes with >10 ppbv of NO 2 , >1000 ppbv of CO, and >440 ppmv of CO 2 over XT, usually with a secondary maximum between 800 and 1200 m.These plumes aloft can play an important role in long-range transport of air pollutants to downwind regions.Profiles over LC (Fig. 3b) show higher O 3 concentrations (>100 ppbv) and relatively moderate NO 2 (~3 ppbv) and CO (~250 ppbv).The rural areas, JL and QZ, have relatively clean environment with <80 ppbv of O 3 , <2 ppbv of NO 2 , and <300 ppbv of CO (Fig. 3c and 3d).Even the concentrations of air pollutants over the rural region in the NCP are comparable or higher than values in urban areas in North America and Europe.In summary, we found the south-north and east-west gradients of air pollution, i.e., higher concentrations of air pollutants in the west XT-LC corridor near the mountain as compared with east side of JL and QZ.Thus, the ARIAs research flights have good coverage of regions with both high and moderate concentrations of air pollutants and can fairly represent the regional nature of air pollution over the NCP.

Air Pollution in the NCP and CMAQ performance
Comparison of the surface trace gas observations at the Xingtai supersite and the CMAQ simulations driven by the EDGAR inventory (named baseline CMAQ case hereafter) reveals that CMAQ generally underestimates concentrations of major air pollutants (Fig. S1 in the supplementary material).The baseline CMAQ run successfully captures the diurnal and daily variations of surface ozone in Xingtai, although consistently underpredicts its concentrations.For Following the approach described in Goldberg et al. (2016), we calculated the 10-min average O 3 , CO, NO, and NO 2 concentrations from aircraft measurements and compared them with the baseline CMAQ simulations (Fig. 4) and found similar underestimation (50% to 75% for all air pollutants) as compared with surface measurements (Fig. S1 in the supplementary material).
CMAQ overestimates NO y but substantially underestimates NO and NO 2 , which suggests that a significant amount of reactive nitrogen compounds could exist in the format of organic nitrates or nitrate aerosols in the model.Figure 5 compares total VOCs concentrations from WAS samples and CMAQ simulations, indicating that VOCs levels are significantly underestimated by 80%.The model evaluation with surface and aircraft measurements suggest that in the NCP ozone pollution has been significantly underestimated in the baseline CMAQ run, which could be due to the uncertainty introduced by using the 2010 EDGAR emissions to simulate the 2016 ARIAs campaign period.Thus, we need to evaluate the emissions inventory data to improve the CMAQ performance and investigate the sensitivity of ozone production.

Evaluation of emissions inventory in the NCP
The EDGAR v4.2 emission inventory in East Asia was created based on the 2010 MIX emission inventory (Li et al., 2017a), so substantial changes were anticipated when used for the ARIAs campaign in 2016.Anthropogenic emission inventories are usually based on the "bottomup" approach, which relies on the statistics of fossil fuel usage and emission factors (EFs) for each sector defined as the ratio of the amount of air pollutants released by a unit of CO 2 emissions, e.g.CO/CO 2 and NO x /CO 2 .To evaluate the emission inventory data in the NCP, we used a 60-s moving window and conducted linear regression of observed air pollutant (CO, NO x , etc.) concentrations vs. CO 2 concentrations, i.e.ΔCO/ΔCO 2 and ΔNO x /ΔCO 2 , defined as emission enhancements (EEs).Through only selecting EEs that are in the PBL (below 1.5 km AGL in this study) and statistically significant (R 2 > 0.6), the values of EEs can act as a proxy of EFs in the air mass observed (Halliday et al., 2018).
EEs observed during the research flights have a broad range of values.ΔCO/ΔCO 2 ranges from below 1%, a typical value of modern automobile emissions, to higher than 10%, a value indicating fossil fuel combustion with high emissions such as biomass burning (Fig. 6a and 6b).
The mean of observed EE for CO (3.7%) is close to that calculated from the EDGAR inventory (4.0%) in the aircraft campaign area.Observed ΔNO x /ΔCO 2 ratios also have isolated high values (>0.1%) with a mean value of 0.05%, which is substantially higher than the EF (~0.03%) derived from the EDGAR inventory.Since estimation of anthropogenic CO 2 flux in an urban/suburban area is challenging (Cambaliza et al., 2014;Heimburger et al., 2017), the underestimation of CO and NO x in the NCP could be caused by either underestimated EFs or uncertainty in anthropogenic CO 2 emission data used in the 'bottom-up' approach.
To further investigate the characteristics of air pollutant emissions in the NCP, we conducted a similar analysis of ΔNO x /ΔCO, which are usually co-emitted in combustion processes.Since around half of the CO and NO x are from mobile sources in the EDGAR emission inventory, this ratio can approximately represent the emission characteristic of mobile sources in the NCP.The mean observed ΔNO x /ΔCO ratio is ~1.3%, significantly lower than 5.6% based on the EDGAR emission inventory (Fig. 6c).These results suggest that the EDGAR emission inventory substantially overestimates the ratios of NO x /CO, while the automobile emissions over the NCP in 2016 have been greatly improved due to recent regulations focusing on NO x , i.e., EDGAR overestimates the contribution from combustion with high emissions.It is worth noting that we only evaluated the emission ratios (EEs or EFs) in the EDGAR inventory, while the underestimation of CO and NO x emissions could be caused by inaccurate CO 2 emissions which have not been examined in this study.

Evaluation of CO, NO x , and VOCs emissions using satellite data
Satellite observations are widely used to evaluate the anthropogenic emissions in East Asia sometimes supplemented by model simulations, e.g., CO emissions using the MOPITT CO products (Jiang et al., 2015;Zheng et al., 2018), anthropogenic NO x emissions using OMI NO 2 products (Wang et al., 2012;de Foy et al., 2015;Qu et al., 2017), and VOCs emissions using OMI HCHO products (Stavrakou et al., 2016).In this study, we used measurements from multiple satellite instruments to evaluate the CMAQ performance of NO 2 , HCHO, and CO.Since NO 2 and HCHO can be treated as proxy of NO x and VOCs emissions, we can further improve the 2010 EDGAR emissions over the NCP base on satellite data.
We followed the approach developed in Canty et al. (2015) to compare the tropospheric column contents of NO 2 from OMI products and CMAQ simulations.Level 2 OMI NO 2 swatch information including row anomaly and quality flags were used to sample NO 2 vertical profiles from CMAQ outputs, and then CMAQ NO 2 column was calculated using the OMI averaging kernel (AK).Lastly, we averaged OMI and CMAQ NO 2 column contents to create daily 0.25° × 0.25° Level 3 products (see details in Canty et al., 2015).A similar approach was used to integrate HCHO column contents from CMAQ simulations based on OMI HCHO retrievals (see details in Ring et al., 2018) and construct daily 0.25° × 0.25° Level 3 HCHO products.For tropospheric CO, we selected the CO concentrations at ~ 900 hPa in CMAQ and averaged them to 1.0° × 1.0° daily products using MOPITT CO averaging kernel (MOPITT Science Team, 2013).All gridded daily data of satellite and CMAQ were averaged in May and June 2016 for comparison.
Figure 7a shows strong signals over the NCP of the OMI NO 2 observations.CMAQ underestimates NO 2 columns over the aircraft campaign area, and only predicts 81% of NO 2 column as compared with OMI observations.However, in urban regions such as Beijing, the Yangtze River Delta (YRD), and the PRD, CMAQ substantially overestimates column NO 2 by up to 30%.Because the baseline CMAQ simulations used the 2010 anthropogenic emission data, these differences should reflect the changes in NO x emissions due to recent air pollution regulations.The comparison of NO 2 column suggests that NO x pollution of megacities in China has been substantially improved after 2010 while NO x pollution in smaller cities and rural area has worsened, consistent with results from independent studies using OMI (Duncan et al., 2016;Krotkov et al., 2016).OMI HCHO retrievals also show high values over the NCP in spring when plants' photosynthetic activity is relatively low, reflecting that the domination of anthropogenic VOCs emissions in north China (Zhao et al., 2017).CMAQ has good agreement with OMI HCHO within the aircraft campaign area (<20% underestimation), but substantially underestimates HCHO columns in south China where biogenic VOCs dominate (Fig. 7b).The MOPPIT products show high near-surface CO concentrations over the eastern China (Fig. 7c), Using NO 2 and HCHO as proxies of NO x and anthropogenic VOCs emissions, the comparison of satellite observations and the baseline CMAQ simulations suggests that both NO x and VOCs emissions in the aircraft campaign area need to be adjusted for a better simulation of tropospheric ozone.Also, the underestimation of CO, as an important precursor, can lead to underprediction of tropospheric ozone.We calculated the model/satellite ratios of NO x , HCHO, and CO in East Asia (Fig. 8) and used these ratios to adjust their anthropogenic emissions in CMAQ.The results will be discussed in Section 3.4.

Tropospheric ozone production sensitivity from OMI and CMAQ
Photochemical production of tropospheric ozone is highly non-linear and dependent on concentrations of NO x and VOCs (Kleinman, 1994;Sillman, 1999;Kleinman, 2000).A maximum rate of ozone production can be achieved with an optimal VOCs/NO x ratio.With other VOCs/NO x ratios, ozone production can be either in the VOC-sensitive regime (the rate of ozone production is controlled by VOCs concentrations) or in the NO x -sensitive regime (the rate of production is controlled by NO x concentrations).Different pollution control strategies can be implemented to reduce the tropospheric ozone levels in these two regimes.For instance, in a VOC-sensitive environment, reducing NO x emissions will lead to limited effects until the ozone production has been changed to a NO x -sensitive environment with the continuous removal of NO x from the atmosphere.Duncan et al. (2010) developed an approach using OMI HCHO/NO 2 column ratio to estimate the ozone production sensitivity as: 1) HCHO/NO 2 <1: VOC-sensitive regime; 2) HCHO/NO 2 1~2: transition regime; 3) HCNO/NO 2 > 2: NO x -sensitive regime.
Studies show that urban areas in the U.S. such as Los Angeles, New York City and Houston are in VOC-sensitive or transition regimes, which lead to difficulty in local regulation of air quality (Duncan et al., 2010;Mazzuca et al., 2016;Ring et al., 2018).Recent studies suggest new threshold values of HCHO/NO 2 ratios between VOC-sensitive, transition, and NO x -sensitive regimes in the U.S. (Jin et al., 2017;Schroeder et al., 2017).
Using the Duncan et al. (2010) approach, studies using OMI products suggest large areas of eastern China are either in VOC-sensitive regime (mostly megacities such as Beijing) or in transition regime (Jin and Holloway, 2015;Jin et al., 2017;Xing et al., 2018).We follow the approach described in Ring et al. (2018) to calculate the column HCHO/NO 2 ratios from OMI observations and CMAQ simulations for East Asia.OMI column HCHO/NO 2 ratios suggest that the ozone photochemical production is VOC-sensitive or in transition region over the NCP and other major urban areas such as YRD and PRD (Fig. 9a) if the Duncan et al. (2010) approach is applicable for these areas.CMAQ successfully captured the spatial distribution of the regional nature of ozone production sensitivity in eastern China but predicted that the rate of ozone production is controlled more by VOCs with the CMAQ HCHO/NO 2 ratio lower than 1.0 in Beijing, YRD, and PRD (Fig. 9b).The VOC-sensitive environment suggests the rate of ozone photochemical production in the NCP is controlled not only by NO x emissions, but also by VOCs emissions which currently lack regulations in China.With continuous reduction of anthropogenic NO x emissions in China, VOCs controls might be efficient in these VOC-sensitive regions.

Improvements of tropospheric ozone simulation using satellite products
Results of the previous two sections show that the baseline CMAQ run substantially underestimates the concentrations of ozone and its major precursors in the NCP.To identify the individual and combined effects of the emission discrepancy of impacting major ozone precursors in the NCP, we designed a series of sensitivity experiments with emissions adjusted to satellite observations.Unlike the top-down approach using global chemical transport models such GEOS-Chem (Lin et al., 2010b;Qu et al., 2017), here we simply applied the ratios of air pollutant column contents from satellite observations and CMAQ simulations on each 0.25 degree grids (Fig. 8) as: CO CMAQ /CO MOPITT , NO 2CMAQ /NO 2OMI , and HCHO CMAQ /HCHO OMI ratios for anthropogenic CO, NO x , and VOCs emissions, respectively.To estimate the contribution from biogenic VOCs emissions, we conducted one more run with the in-line BEIS module turned off.Table 1 shows the emission adjustments for the five sensitivity experiments.CMAQ was run for the nested 12 km domain (D02) with the same meteorology, initial conditions, and boundary conditions derived from the coarse domain simulations.CMAQ simulations are ~1700 ppbv while surface observations have CO peaks higher than 6000 ppbv (Fig. 10b).The adjustments of the emission inventory have improved the model simulations of NO 2 /NO (Fig. 10c and 10d) and HCHO (Fig. 10e).During the ARIAs flights, we observed various sources of emissions in the aircraft campaign area such as small factories and biomass burning, which are not included in the EDGAR emission inventory.Thus, the reason for the model underestimation could be that the spatial resolution (12 km) of the nested CMAQ domain cannot represent the detailed emissions and resolve the local air pollution hotspots.
However it is worth noting that even our CMAQ system is still not capable to reproduce the surface air quality at Xingtai, the adjustments of EDGAR emissions based on satellite observations reduce the underestimation.
The ARIAs flights covered a large area (~10 4 km 2 ) in Hebei Province, which represent the regional nature of air pollution over the NCP.A case comparison of CMAQ_All case and Y12 measurements on June 11, 2016 (Fig. 11) shows better results in both concentrations and vertical gradient of air pollutants (compared with Fig. S2 in the supplementary material), indicating the effectiveness of improving the emission inventories based on satellite observations.Table 2 summarizes the model performance of CMAQ as compared with aircraft measurements.The adjustments of the EDGAR emissions with satellite observations moderately improved simulations of ozone pollution, with the root mean square error (RSME) decreasing from 25.1 ppbv (the baseline case) to 21.2 ppbv (CMAQ_All case) and the mean ratio of CMAQ simulations to aircraft observations increasing from 0.75 to 0.82.The model performance of CO has also been improved, with the RMSE decreasing from 247.0 ppbv to 203.6 ppbv and the mean ratio increasing from 0.40 to 0.66.For nitrogen compounds including NO 2 , NO, and NO y , the adjustments of EDGAR emissions have small impacts on improving the CMAQ performance.
The reason could be that the ozone photochemistry is mainly VOC-sensitive over the NCP, so the adjustments of NO x emissions have limited impacts close to sources.

Conclusions and Discussion
The ARIAs campaign conducted aircraft measurements over the NCP and observed high concentrations of air pollutants including O 3 , CO, and NO x .CMAQ simulations driven by the 2010 EDGAR emissions substantially underestimate the levels of ozone and its precursors in the campaign region.Analysis of emission enhancements of CO and NO x with respect to concurrent shows that the EDGAR VOCs and CO emissions could also be underestimated in the NCP.HCHO/NO 2 column ratio from OMI observations indicates tropospheric ozone production is mainly in the VOC-sensitive regime in the NCP, which is also confirmed by CMAQ simulations.
To test a hypothesis that the poor model performance is due to emission biases, we adjusted the EDGAR emissions over East Asia based on satellite observations.Better performance of simulating ozone and its precursors is achieved, while underestimation still exists.
Both satellite observations and CMAQ simulations indicate that the VOC-sensitive chemistry dominates the ozone photochemical production in eastern China, so the rate of local ozone production is mainly controlled by the VOCs emissions.In the past few years, despite implementation of control measures mainly on SO 2 and NO x , ozone concentrations have increased in China.Our study indicated that high NO x concentrations were pervasive in the PBL over rural areas of the NCP, where anthropogenic VOCs were also abundant.Reducing NO x emissions is essential to control ozone on the regional scale, but our model simulations indicated that reducing VOCs emissions can lower the rate of photochemical smog production.
Currently, studies and regulations on anthropogenic VOCs emissions in China are lacking, so with expectation of further decreasing NO x emissions in China, more severe ozone pollution could be anticipated.It is worth noting that while VOCs controls can have beneficial impact on the local rate of ozone production in the VOC-sensitive regime, the ozone levels will not decrease until NO x emissions are substantially lower, i.e., regulations on VOCs are needed as well as the continuous controls on NO x emissions in China.These results can also partially explain why ozone pollution emerged in the past few years while PM 2.5 pollution has been substantially improved with strict regulations on anthropogenic emissions.New datasets such as the updated 'bottom-up' emissions inventory for East Asia and high resolution satellite observations such as TROPOMI and GEMS products are needed to improve the modeling of ozone pollution in China, which can provide scientific evidence for future national and international regulations on air quality.

Figure 2
Figure 2 summarizes all aircraft measurements of O 3 , NO 2 , CO, and CO 2 over the NCP from eleven research flights.Generally, we observed high concentrations of trace gases, such as >100 part per billion by volume (ppbv) of O 3 , >20 ppbv of NO 2 , >500 ppbv of CO, and >450 part per million by volume (ppmv) of CO 2 , in the aircraft campaign area (defined as 36.5- CO and NO x , two important ozone precursors, CMAQ substantially underestimates their concentrations in Xingtai by more than 50% and especially fails to capture the extremely high values such as 6~7 ppmv of CO and ~100 ppbv of NO x .This underestimation could be caused by local sources poorly represented in the 12-km model simulations.For ambient HCHO, an Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2019-248Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 29 March 2019 c Author(s) 2019.CC BY 4.0 License.important byproduct of VOC oxidization in ozone photochemical production, the baseline CMAQ run captures the variations, but substantially underestimates its concentrations.These results suggest that the underestimation of ozone precursors in CMAQ could lead to the poor model performance of simulating tropospheric ozone and other pollutants.Similar analyses were conducted to investigate air pollutant concentrations in the lower troposphere over the NCP observed by the aircraft.A case of the research flight on June 11, 2016 (Fig. S2 in the supplementary material) shows that CMAQ well captures the vertical gradient of air pollutants, while substantially underestimates concentrations of all trace gases except NO y .
Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2019-248Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 29 March 2019 c Author(s) 2019.CC BY 4.0 License.while the baseline CMAQ run substantially underestimates CO concentrations over north China and only predicts 42% of the CO over the aircraft campaign area.

Figure 10
Figure 10 presents the evaluation of surface observations with respect to two sensitivity experiments (CMAQ_baseline and CMAQ_all, comparison with all CMAQ runs are presented in Fig. S3 in the supplementary material).CMAQ still might not capture the extreme high values of surface O 3 and CO (Fig. 10a and 10b).For instance, the maximum CO concentration from Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2019-248Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 29 March 2019 c Author(s) 2019.CC BY 4.0 License.CO 2 measurements suggests that the usage of the 2010 EDGAR emissions for the 2016 ARIAs campaign could introduce substantial uncertainty due to the recent changes of anthropogenic emissions in China.Comparison of atmospheric columns of NO 2 from CMAQ simulations and satellite observations suggests that NO x emissions decreased in megacities such as Beijing and Shanghai but increased in rural areas from 2010 to 2016.Similar analysis of HCHO and CO Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2019-248Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 29 March 2019 c Author(s) 2019.CC BY 4.0 License.

Figure 1 .
Figure 1.ARIAs flights over the NCP and the WRF-CMAQ domains.Eleven Research Flights

Figure 2 .
Figure 2. Summary of air pollutant concentrations in the NCP observed by Y12 aircraft.a) O 3 , b)

Figure 4 .
Figure 4. Comparison of 10-min averaged aircraft data and CMAQ simulations from 11 ARIAs research flights.a) O 3 , b) CO, c) NO, and d) NO 2 .Black line shows the 1:1 ratio; red line stands for the linear regression fitting line.M_Diff: mean difference; R: correlation; NMB: normalized mean bias; NME: normalized mean error; RMSE: root-mean square error; M_Ratio: mean ratio.a)

Figure 5 .
Figure 5.Comparison of total VOCs concentrations from WAS samples and CMAQ simulations.Values are in unit of parts per billion Carbon (ppbC).Black line shows the 1:1 ratio; red line stands for the linear regression fitting line.M_Diff: mean difference; R: correlation; NMB: normalized mean bias; NME: normalized mean error; RMSE: root-mean square error; M_Ratio: mean ratio.

Figure 6 .
Figure 6.Comparison of emission enhancements (EEs) from the ARIAs campaign and emission factors (EFs) from the EDGAR emission inventory.a) ΔCO/ΔCO 2 , b) ΔNO x /ΔCO 2 , c) ΔNO x /ΔCO.Blue histogram shows the distribution of EEs observed by the Y12 aircraft; red line shows the ratio calculated using the EDGAR anthropogenic emissions.a)

Figure 7 .
Figure 7.Comparison of air pollutants from satellite observations and CMAQ simulations.a) OMI NO 2 column (left) and the difference between OMI and CMAQ (right), Unit: Dobson Unit (1 DU = 2.69 × 10 20 molecules/cm 2 ); b) OMI HCHO column (left) and the difference between OMI and CMAQ (right), Unit: DU; c) MOPITT near surface CO (left) and the difference between MOPITT and CMAQ (right), Unit (ppbv).a)

Figure 9 .
Figure 9. Column HCHO/NO 2 ratios over East Asia in spring 2016.a) Ratio derived from collocated OMI HCHO and NO 2 observation; b) Ratio calculated from CMAQ simulations with OMI quality information and averaging kernel (AK).a)

Figure 10 .
Figure 10.Comparison of surface hourly observations of air pollutants and CMAQ simulations at the Xingtai supersite from May to mid-June 2016.a) O 3 , b) CO, c) NO 2 * , d) NO x , and e)