Improvement of the satellite-derived NO_x emissions on air quality modeling and its effect on ozone and secondary inorganic aerosol formation in the Yangtze River Delta, China

We developed a top-down methodology combining the inversed chemistry transport modeling and satellite-derived tropospheric vertical column of NO₂ and estimated the NO_x emissions of the Yangtze River Delta (YRD) region at a horizontal resolution of 9 km for January, April, July, and October 2016. The effect of the top-down emission estimation on air quality modeling and the response of ambient ozone (O₃) and inorganic aerosols (SO${}_{\mathrm{4}}^{\mathrm{2}-}$, NO${}_{\mathrm{3}}^{-}$, and NH${}_{\mathrm{4}}^{+}$, SNA) to the changed precursor emissions were evaluated with the Community Multi-scale Air Quality (CMAQ) system. The top-down estimates of NO_x emissions were smaller than those (i.e., the bottom-up estimates) in a national emission inventory, Multi-resolution Emission Inventory for China (MEIC), for all the 4 months, and the monthly mean was calculated to be 260.0 Gg/month, 24 % less than the bottom-up one. The NO₂ concentrations simulated with the bottom-up estimate of NO_x emissions were clearly higher than the ground observations, indicating the possible overestimation in the current emission inventory, attributed to its insufficient consideration of recent emission control in the region. The model performance based on top-down estimate was much better, and the biggest change was found for July, with the normalized mean bias (NMB) and normalized mean error (NME) reduced from 111 % to −0.4 % and from 111 % to 33 %, respectively. The results demonstrate the improvement of NO_x emission estimation with the nonlinear inversed modeling and satellite observation constraint. With the smaller NO_x emissions in the top-down estimate than the bottom-up one, the elevated concentrations of ambient O₃ were simulated for most of the YRD, and they were closer to observations except for July, implying the VOC (volatile organic compound)-limited regime of O₃ formation. With available ground observations of SNA in the YRD, moreover, better model performance of NO${}_{\mathrm{3}}^{-}$ and NH${}_{\mathrm{4}}^{+}$ was achieved for most seasons, implying the effectiveness of precursor emission estimation on the simulation of secondary inorganic aerosols. Through the sensitivity analysis of O₃ formation for April 2016, the decreased O₃ concentrations were found for most of the YRD region when only VOC emissions were reduced or the reduced rate of VOC emissions was 2 times of that of NO_x, implying the crucial role of VOC control in O₃ pollution abatement. The SNA level for January 2016 was simulated to decline 12 % when 30 % of NH₃ emissions were reduced, while the change was much smaller with the same reduced rate for SO₂ or NO_x. The result suggests that reducing NH₃ emissions was the most effective way to alleviate SNA pollution of the YRD in winter.

1 Introduction

Nitrogen oxides (NO_x = NO₂ + NO) play an important role in the formation of ambient ozone (O₃) and inorganic aerosols (SO${}_{\mathrm{4}}^{\mathrm{2}-}$, NO${}_{\mathrm{3}}^{-}$, and NH${}_{\mathrm{4}}^{+}$, SNA). The NO_x emission inventory is a necessary input of the air quality model (AQM) and has a great influence on NO₂, O₃, and SNA simulation (Zhou et al., 2017; Chen et al., 2019a). Moreover, it is crucial for exploring the sources of atmospheric pollution of O₃ and fine particles (particles with aerodynamic diameter smaller than 2.5 µm, PM_2.5) with the AQM. The inventories were usually developed with a bottom-up method, in which the emissions were calculated based on the activity data (e.g., fuel consumption and industrial production) and emission factors (the emissions per unit of activity data) by source category and region. Bias existed commonly in the bottom-up inventories, due mainly to the uncertainty of economic and energy statistics and to the fast changes in the emission control measures, especially in developing countries like China (Granier et al., 2011; Saikawa et al., 2017; Zhang et al., 2019).

To improve the emission estimation, an inversed top-down method has been developed based on satellite observations and the AQM (Martin et al., 2003; Zhao and Wang, 2009; Zyrichidou et al., 2015; Yang et al., 2019a). The emissions were corrected based on the difference between the modeled and observed tropospheric vertical column densities (TVCDs) of NO₂ and the response coefficient of NO₂ TVCDs to emissions (Martin et al., 2003; Cooper et al., 2017). With higher temporal and spatial resolution than other instruments, the NO₂ TVCDs from the Ozone Monitoring Instrument (OMI) were frequently used (Kurokawa et al., 2009; Gu et al., 2014; de Foy et al., 2015; Kong et al., 2019; Yang et al., 2019a).

Currently, top-down methods are mainly developed at the global or national scale with relatively coarse horizontal resolution (Martin et al., 2003; Miyazaki et al., 2012; Jena et al., 2014; Gu et al., 2014). At the global scale, for example, Martin et al. (2003) and Miyazaki et al. (2012) estimated the NO_x emissions at the horizontal resolution of 2^∘ × 2.5^∘ and 2.8^∘ × 2.8^∘, respectively. Martin et al. (2003) found that the satellite-derived NO_x emissions for 1996–1997 were 50 %–100 % larger than the bottom-up estimates in the Po Valley, Tehran, and Riyadh urban areas. Miyazaki et al. (2012) suggested that the bottom-up method underestimated the NO_x emissions over eastern China, the eastern United States, southern Africa, and central–western Europe. At the national scale with the horizontal resolution of 0.5^∘ × 0.5^∘, the annual NO_x emissions in India 2015 derived with the top-down method were 7 %–60 % smaller than various bottom-up estimates (Jena et al., 2014). With the TVCDs from OMI and another instrument (Global Ozone Monitoring Experiment, GOME), the difference in national NO_x emissions for China was quantified to be 0.4 Tg N/yr (5.8 % relative to OMI) at the resolution of 70 km × 70 km (Gu et al., 2014). Compared to national and regional ones, limited estimates were available at the regional scale with finer resolution. In China, great differences exist in the levels and patterns of air pollution across the regions, attributed partly to a big variety of air pollutant sources across the country. To achieve the target of air quality improvement required by the central government, varied air pollution control plans were usually developed and implemented at the city/provincial level. Therefore, the top-down estimates in NO_x emissions at finer horizontal resolution are greatly needed for understanding the primary sources of NO₂ pollution and demonstrating the effect of emission control at the regional scale.

At present, the reliability and rationality of the top-down emission estimates are commonly evaluated with the AQM and satellite observations. For example, the bias between the NO₂ TVCDs from OMI observations and the AQM based on the top-down NO_x emission estimation was −30.8 ± 69.6 × 10¹³ molec. cm⁻² in winter in India (Jena et al., 2014). The linear correlation coefficient (R²) between OMI and the AQM with the top-down emission estimates could reach 0.84 in Europe (Visser et al., 2019). Compared to the satellite observations with relatively large uncertainty (Yang et al., 2019b; Liu et al., 2019), surface concentrations that better represent the effect of air pollution on human health and the ecosystems were less applied in the evaluation of the top-down estimates of NO_x emissions. Limited assessments were conducted at the national scale. For example, Liu et al. (2018) found that the normalized mean error (NME) between the observed and simulated NO₂ concentrations based on the top-down estimate of NO_x emissions could reach 32 % in China at the resolution of 0.25^∘ × 0.25^∘. Besides NO₂, the estimation of NO_x emissions also plays an important and complicated role in secondary air pollutant simulation including O₃ and SNA, and the response of secondary pollution to the primary emissions was commonly nonlinear. The simulated O₃ concentrations in Shanghai (the most developed city in eastern China) could increase over 20 %, with a 60 % reduction in NO_x emissions in summer 2016, implying a clear VOC (volatile organic compound)-limited pattern for the O₃ formation in the megacity (Wang et al., 2019). For the response of SNA to NO_x emissions, the NH${}_{\mathrm{4}}^{+}$ and SO${}_{\mathrm{4}}^{\mathrm{2}-}$ concentrations at an urban site in another megacity Nanjing in eastern China were simulated to increase 1.9 % and 2.8 %, with a 40 % abatement of NO_x emissions in autumn 2014, respectively, due to the weakened competition of SNA formation against SO₂ (Zhao et al., 2020). To the best of our knowledge, however, the relatively new information from the inversed modeling of NO_x emissions has not been sufficiently incorporated into the SNA and O₃ analyses with the AQM in China.

Located in eastern China, the Yangtze River Delta (YRD) region, including the city of Shanghai and the provinces of Anhui, Jiangsu, and Zhejiang, is one of the most developed and heavy-polluted regions in the country. The air quality for most cities in the YRD failed to meet the National Ambient Air Quality Standard (NAAQS) Class II in 2016 (MEPPRC, 2017). NO_x emissions made great contributions to the severe air pollution in the region. Based on an offline sampling and measurement study, for example, the annual average of the NO${}_{\mathrm{3}}^{-}$ mass fraction to total PM_2.5 reached 19 % in Shanghai in 2014, and it was significantly elevated in the pollution event periods (Ming et al., 2017). In this study, we chose the YRD to estimate the NO_x emissions with the inversed method and to explore their influence on the air quality modeling. The top-down estimates of NO_x emissions were firstly obtained with the nonlinear inversed method and OMI-derived NO₂ TVCDs for 2016. The advantage of the top-down estimation against the bottom-up one was then evaluated with the AQM and abundant ground-based NO₂ concentrations. The influences of the top-down estimation in NO_x emissions were further detected on O₃ and SNA modeling. Sensitivity analyses were conducted by changing the emissions of precursors to investigate the sources and potential control approaches of O₃ and SNA pollution for the region.

2 Data and methods

2.1 The top-down estimation of NO_x emissions

The top-down estimation of NO_x emissions was conducted for January, April, July, and October of 2016, representing the situations of the four seasons in the YRD region, and the horizontal resolution was 9 km ×9 km. The inversed method assumed a nonlinear and variable correlation between NO_x emissions and NO₂ TVCDs (Cooper et al., 2017), and the a posterior daily emissions (top-down estimates) were calculated with the following equation:

$$\begin{array}{}\text{(1)}& E_{\mathrm{t}}=E_{\mathrm{a}}\left(\mathrm{1}+{\displaystyle \frac{{\mathrm{\Omega }}_{\mathrm{o}}-{\mathrm{\Omega }}_{\mathrm{a}}}{{\mathrm{\Omega }}_{\mathrm{o}}}}\mathit{\beta }\right),\end{array}$$

where E_t and E_a represent the a posterior and the a prior daily NO_x emissions, respectively; Ω_o and Ω_a represent the observed and simulated NO₂ TVCDs, respectively; β represents the response coefficient of the simulated NO₂ TVCDs to a specific change in emissions and was calculated based on the simulated changes in TVCDs (ΔΩ) from a 10 % change in emissions (ΔE).

The inversed method assumed that the daily emissions were similar. For a given month, the a posterior daily emissions were used as the a priori emissions of the next day, and the monthly top-down estimate of the NO_x emissions was scaled from the average of the a posterior daily emissions of the last 3 d in the month, as the top-down estimate of daily NO_x emissions usually converged within a 1-month simulation period (Zhao and Wang, 2009; Yang et al., 2019b). In our previous work (Yang et al., 2019b), we demonstrated the robustness of the method, by applying the “synthetic” TVCDs from air quality simulation based on a hypothetical “true” emission inventory, instead of those from satellite observations. We found that sufficient iteration times could result in a relatively constant emission estimate (the top-down estimate) close to the “true” emission input. From a bottom-up perspective, the difference in NO_x emissions between weekdays and the weekend was within 5 % in the YRD region (Zhou et al., 2017), indicating an insignificant bias from the ignorance of the daily variation of emissions.

The NO₂ TVCDs were from OMI on board the Aura satellite. It crosses the Equator at 13:30 local time. The horizontal resolution of OMI was 24 km × 13 km at nadir (Levelt et al., 2006), one of the finest resolutions available for NO₂ TVCD observations before October 2017. We applied the Peking University Ozone Monitoring Instrument NO₂ product (POMINO v1; Lin et al., 2014, 2015) to constrain the NO_x emissions. POMINO v1 modified the retrieval methodology of the Dutch Ozone Monitoring Instrument NO₂ product (DOMINO v2) in China and provided better linear correlation of NO₂ TVCDs between the satellite and available ground-based observations using multi-axis differential optical absorption spectroscopy (MAX-DOAS) (Lin et al., 2015). The original NO₂ TVCDs from POMINO v1 (level 2) were resampled into an 18 km × 18 km grid system based on the area weight method and then downscaled to 9 km × 9 km with Kriging interpolation. As an example, the NO₂ TVCDs for July 2016 in the YRD are shown in Fig. S1 in the Supplement, and larger TVCDs were found in the east–central YRD.

Figure 1 The modeling domain and locations of meteorological and air quality monitoring sites. The map data provided by the Resource and Environment Data Cloud Platform are freely available for academic use (http://www.resdc.cn/data.aspx?DATAID=201, last access: 20 January 2020). © Institute of Geographic Sciences & Natural Resources Research, Chinese Academy of Sciences.

2.2 Model configuration

The Models-3 Community Multi-scale Air Quality (CMAQ) version 5.1 was used to conduct the inversed modeling of NO_x emission estimation and to simulate the ground-level concentrations of NO₂, O₃, and SNA. As a three-dimensional Eulerian model, CMAQ includes complex interactions of atmospheric chemistry and physics and is one of the most widely applied AQMs to evaluate the sources and processes of air pollution in China (UNC, 2012; Xing et al., 2015; Zheng et al., 2017). As shown in Fig. 1, the two nested modeling domains were applied with their horizontal resolutions set to 27 km and 9 km, respectively. The mother domain (D1, 177 cells × 127 cells) included most parts of China, and the second (D2, 118 × 121 cells) covered the YRD region. The model included 28 vertical layers, and the height of the first layer (ground layer) was approximately 60 m. The carbon bond gas-phase mechanism (CB05) and AERO6 aerosol module were used in the CMAQ. The initial concentrations and boundary conditions for the D1 were derived from the default clean profile, while those of D2 were extracted from the CMAQ Chemistry Transport Model (CCTM) outputs of its mother domain. The first 5 d of each simulated month were chosen as the spin-up period. Details on model configuration were described in Zhou et al. (2017) and Yang and Zhao (2019).

The Multi-resolution Emission Inventory for China (MEIC; http://www.meicmodel.org/, last access: 20 January 2020) for 2015 was applied as the initial input of anthropogenic emissions in D1 and D2, with an original horizontal resolution of 0.1^∘ × 0.1^∘. In this study, the MEIC emissions from residential sources were downscaled to the horizontal resolution of 9 km × 9 km, based on the spatial density of population, and those from power, industry, and transportation based on the spatial distribution of gross domestic product (GDP). The NO_x emissions from soil were originally obtained from Yienger and Levy (1995) and were doubled as advised by Zhao and Wang (2009). The emissions of Cl, HCl, and lightning NO_x were collected from the Global Emissions Initiative (GEIA; Price et al., 1997). Biogenic emissions were derived from the Model Emissions of Gases and Aerosols from Nature developed under the Monitoring Atmospheric Composition and Climate project (MEGAN MACC; Sindelarova et al., 2014).

Meteorological fields were provided by the Weather Research and Forecasting (WRF) model version 3.4, a state-of-the-art atmospheric modeling system designed for both numerical weather prediction and meteorological research (Skamarock et al., 2008). The simulated parameters from WRF for D2 in January, April, July, and October 2016 were compared with the observation dataset of the US National Climate Data Center (NCDC), as summarized in Table S1 in the Supplement. The index of agreement (IOA) of wind speed for the 4 months between the two datasets was larger than 0.8. The root mean square error (RMSE) of wind directions for the 4 months was smaller than 40^∘, and the index of agreement (IOA) of temperature and relative humidity between the two datasets was larger than 0.8 and 0.7, respectively. The simulated meteorological parameters in D2 could reach the benchmarks derived from Emery et al. (2001) and Jiménez et al. (2006).

The hourly NO₂ and O₃ concentrations were observed at 230 state-operated stations of air quality monitoring in 41 cities within the YRD region, and they were applied to evaluate the model performance. Locations of the stations are indicated in Fig. 1, and the observation data were derived from the China National Environmental Monitoring Center (CNEMC; http://www.cnemc.cn/, last access: 20 January 2020). The observations of SO${}_{\mathrm{4}}^{\mathrm{2}-}$, NO${}_{\mathrm{3}}^{-}$, and NH${}_{\mathrm{4}}^{+}$ (SNA) concentrations in PM_2.5 for the YRD region during 2015–2017 were collected and applied to evaluate the influence of the top-down estimation of NO_x emissions on SNA simulation. In particular, the hourly SNA concentrations of PM_2.5 at the Jiangsu Provincial Academy of Environmental Science, an urban site in the capital city of Jiangsu Province, Nanjing (JSPAES; Chen et al., 2019b), were observed with the Monitor for Aerosols and Gases in ambient Air (MARGA; Metrohm, Switzerland) for January, April, July, and October 2016. Meanwhile, the daily average concentrations of SNA were also available from MARGA measurement for the 4 months at the Station for Observing Regional Processes and the Earth System, a suburban site in eastern Nanjing (SORPES; Ding et al., 2019). In addition, the seasonal average concentrations of SNA were available at another four sites in the YRD, including the Nanjing University of Information Science & Technology site in Nanjing (NUIST; Zhang, 2017), and three sites respectively in the cities of Hangzhou (HZS; Li, 2018), Changzhou (CZS; Liu et al., 2018), and Suzhou (SZS; Wang et al., 2016). Details of the collected SNA measurement studies are summarized in Table S2 in the Supplement, and the locations of those sites are illustrated in Fig. 1.

2.3 Scenario setting of sensitivity analysis

In general, there are two categories of chemical regimes (VOC-limited and NO_x-limited) in O₃ formation (Wang et al., 2010; Jin et al., 2017). In the VOC-limited regime, growth in O₃ concentrations occurs with increased VOC emissions and declined NO_x emissions, while the increased NO_x emissions result in enhancement of O₃ concentrations in the NO_x-limited regime. To explore the sources and potential control approaches of O₃ pollution, the sensitivity of O₃ formation to its precursor emissions for April was analyzed with CMAQ modeling in the YRD region. In the YRD, the peaking time of O₃ concentration has gradually moved from summer to late spring, and the mean observed O₃ concentration in April was 72.5 µg/m³, slightly higher than that in July (71.9 µg/m³). In addition, the model performance of O₃ was better for April than that for July in this work (see details in Sect. 3.2). Therefore, we selected April to explore the sensitivity analysis of O₃ formation in the region. As summarized in Table S3 in the Supplement, eight cases were set besides the base scenario with the top-down NO_x estimates for April 2016. Cases 1 and 6 reduced only the NO_x emissions by 30 % and 60 %, and Cases 2 and 7 reduced only the VOC emissions by 30 % and 60 %, respectively. To explore the co-effect of VOCs and NO_x emission controls on O₃ concentrations, cases with different reduction rates of VOCs and NO_x emissions were designed. The emissions of NO_x and VOCs in Case 4 were decreased by 30 % and 60 % and in Case 5 by 60 % and 30 %, respectively. Both NO_x and VOC emissions were reduced 30 % and 60 % in Cases 3 and 8, respectively.

The response of SNA concentrations to the changes in precursor emissions was influenced by various factors including the abundance of NH₃, atmospheric oxidation, and the chemical regime of O₃ formation (Wang et al., 2013; Cheng et al., 2016; Zhao et al., 2020). To explore the sensitivity of SNA formation to its precursor emissions, four cases were set besides the base scenario for January 2016, the month with the largest observed SNA concentrations. As shown in Table S4 in the Supplement, the emissions of NO_x, SO₂, and NH₃ were reduced by 30 % in Cases 9–11, respectively, and the emissions of NO_x, SO₂, and NH₃ were simultaneously decreased by 30 % in Case 12.

Figure 2 The bottom-up and top-down estimates of NO_x emissions by month for the YRD region in 2016.

3 Results and discussion

3.1 Evaluation of the bottom-up and top-down estimates of NO_x emissions

Figure 2 compares the magnitude of the NO_x emissions estimated based on the bottom-up (MEIC) and top-down methods by month in the YRD region. The top-down estimates were smaller than the bottom-up ones for all the 4 months concerned, and the average of the monthly NO_x emissions was calculated to be 260.0 Gg/month for 2016 with the top-down method, 24 % smaller than the bottom-up estimation. The comparison indicates a probable overestimation in NO_x emissions with the current bottom-up methodology, attributed partly to the insufficient consideration of the effect of recent control on emission abatement. Stringent measures have gradually been conducted to improve the local air quality in the YRD region. For example, the “ultra-low” emission policy for the power sector started in 2015, requiring the NO_x concentration in the flue gas of coal-fired units to be the same as that of gas-fired units. Technology retrofitting on power units has been widely conducted, significantly improving the NO_x removal efficiencies of selective catalytic reduction (SCR) systems. These detailed changes in emission control, however, could not be fully incorporated in a timely manner into the national emission inventory that relies more on the routinely reported information and policy of environmental management over the country. With the online data from continuous emission monitoring systems (CEMS) incorporated, NO_x emissions from the power sector were estimated to be 53 % smaller than MEIC for China in 2015 in our previous work (Zhang et al., 2019). The bias between the top-down and bottom-up estimates could be larger in earlier years and reduced more recently. According to Yang et al. (2019b) and Qu et al. (2017), for example, the top-down NO_x emissions were 44 % and 31 % smaller than bottom-up ones for the YRD region and the whole of China in 2012. Benefiting from the better data availability, the bottom-up inventory has been improved with the inclusion of more information on individual power and industrial plants for recent years (Zheng et al., 2018).

Figure 3 The spatial differences between the bottom-up and top-down estimates of NO_x emissions for January, April, July, and October 2016 (top-down minus bottom-up, in units of mol N/s).

The differences in the spatial distribution of NO_x emissions between the bottom-up and top-down estimates are illustrated by month for the YRD in Fig. 3. The top-down estimates were commonly smaller than the bottom-up ones in the east–central YRD, with intensive manufacturing industry and high population, and larger than those in most of Zhejiang Province, with more hilly and suburban regions. The bias might result from the following issues. From a bottom-up perspective, on the one hand, more stringent control measures were preferentially conducted for power and industrial plants in regions with heavier air pollution like the east–central YRD. As mentioned above, the effects of such actions were difficult to fully track in the bottom-up inventory, leading to the overestimation of emissions for those regions. Due to the lack of precise locations of individual industrial plants (except for large point sources), moreover, the spatial allocation of the emissions relied commonly on the densities of population and economy, assuming a strong correlation with emissions for them. Such an assumption, however, would not still hold in recent years, as a number of factories in the relatively developed region have been moved to less developed suburban regions (e.g., southern Zhejiang) for both environmental and economic purposes. The insufficient consideration of the moving of emission sources is thus expected to result in the overestimation of emissions for developed regions and underestimation for the less developed. On the other hand, the satellite-derived TVCDs were relatively small in southern Zhejiang (Fig. S1), and a larger error in the satellite retrieval, and thereby emissions constrained with the inversed modeling, was expected.

Figure 4 The observed and simulated hourly NO₂ concentrations based on the bottom-up and top-down NO_x emissions for January, April, July, and October 2016.

Figure 4 illustrates the observed and simulated hourly NO₂ concentrations using the bottom-up and top-down estimates of NO_x emissions in the CMAQ by month. The NO₂ concentrations simulated with the bottom-up estimates were clearly larger than the observations in all the 4 months concerned, with the largest and smallest normalized mean bias (NMB) reaching 111 % and 34 % for July and January, respectively. The result suggests again the overestimation in NO_x emissions in the current bottom-up inventory for the YRD. The model performance based on the top-down estimates was much better than that based on the bottom-up ones, indicating that the inversed modeling with the satellite observation constraint effectively improved the estimation of NO_x emissions. The biggest improvement was found for July, with the NMB reduced from 111 % to −0.4 % and the NME reduced from 111 % to 33 %. As shown in Fig. 2, a relatively big reduction from the bottom-up to top-down estimation in NO_x emissions was found for July compared to most of the other months.

Scatter plots of the annual means of the observed and simulated surface NO₂ concentrations are shown in Fig. S2 in the Supplement. The slope between the observations and simulation with the top-down estimate (0.99) was much closer to 1 than that with the bottom-up one (1.57), indicating clearly the advantage of the top-down method for constraining the magnitude of the total emissions in the YRD region. The difference in the two slopes implies that the surface NO₂ concentrations simulated with the bottom-up estimation were over 50 % larger than those based on top-down ones. As a comparison, the total emissions in the bottom-up inventory were only 30 % larger than the top-down estimation for the whole YRD region. The larger overestimation in the concentrations than the emissions from the bottom-up inventory could result partly from the bias of the locations of state-operated ground observation sites. Most of these sites were located in the urban areas where excess emissions were allocated according to the high density of economy and population, and elevated concentrations were thus simulated compared to rural areas. The similar correlation coefficients (R) suggested that the spatial distribution of NO_x emissions was not greatly improved in the top-down estimation on an annual basis of urban observations. Uncertainty existed in the satellite observations: the NMB between NO₂ TVCDs in POMINO and available ground-based MAX-DOAS observations was 21 % on cloud-free days (Liu et al., 2019). Due mainly to the NO_x transport, moreover, a bias of 13 %–33 % on the spatial distribution of emissions was estimated for the inversed method at the horizontal resolution of 9 km or finer (Yang et al., 2019b). Inclusion of more available observations in rural areas helps improve the comprehensive evaluation of emission estimation.

Figure 5 The spatial distribution of the simulated monthly mean NO₂ concentration with the top-down estimates and differences between the simulations with the top-down and bottom-up NO_x emissions in January, April, July, and October 2016 (top-down minus bottom-up).

Figure 5 illustrates the spatial distribution of monthly mean NO₂ concentrations simulated based on the top-down estimates and the differences between the simulations with the top-down and bottom-up ones. The larger NO₂ concentrations existed in the east–central YRD for all the months (left column in Fig. 5), and the difference in the spatial distribution of NO₂ concentrations (right column in Fig. 5) was similar to that in NO_x emissions (Fig. 3). Larger reduction in NO₂ concentrations based on the top-down estimates was commonly found in the east–central YRD, while the increased concentrations were found in most of Zhejiang.

Figure 6 The observed and simulated hourly O₃ concentrations with the bottom-up and top-down NO_x emission estimates for January, April, July, and October 2016.

3.2 Evaluation of the O₃ simulation based on the top-down NO_x estimates

Figure 6 shows the observed and simulated hourly O₃ concentrations based on the bottom-up and top-down estimates of NO_x emissions by month. Indicated by the smaller NMBs and NMEs, the model performance of O₃ based on the top-down estimates was better than that based on the bottom-up ones for most months. It suggests that the constrained NO_x emissions with satellite observations could play an important role in the improvement of O₃ simulation. The largest improvement was found in January, for which the NMB and NME were changed from −44 % and 49 % to 13 % and 40 %, respectively, attributed to the biggest change in NO_x emissions between the top-down and bottom-up estimates for the month. The worse O₃ modeling performance was found for July when the top-down estimate instead of the bottom-up one was applied in the simulation, indicated by the increased NMB and NME. Since the top-down estimation of NO_x emissions was justified by the improved NO₂ simulation in July (Fig. 4c), the worse O₃ simulation might result from the uncertainty in emissions of the volatile organic compounds (VOCs) and the chemical mechanism of the AQM in summer. As suggested by Li (2019), the biogenic VOC (BVOC) emissions of the YRD region could be overestimated by 121 % in summer attributed to ignoring the effect of droughts, and such overestimation might elevate the O₃ concentrations in the AQM. In order to explore the influence of the uncertainty of BVOC emissions on O₃ model performance, we conducted an extra case in which the BVOC emissions were cut by 50 % in CMAQ. As shown in Fig. S3 in the Supplement, the NMB between the observed and simulated O₃ based on the top-down estimate of NO_x emissions and the reduced BVOC emissions declined 27 % in July. However, it was still larger than the NMB at 1.1 % when the bottom-up estimate of NO_x emissions was applied (Fig. 4c). This comparison thus suggested that the complicated mechanism for summer O₃ formation was insufficiently considered in the current model. A recent study conducted an intercomparison of surface-level O₃ simulation from 14 state-of-the-art chemical transport models and implied that the larger overestimation of summer O₃ than winter for eastern China resulted possibly from the uncertainty in the photochemical treatment in models (Li et al., 2019).

Figure 7 The spatial distribution of the simulated monthly mean O₃ concentration with the top-down NO_x estimates and the spatial differences between the simulations with the top-down and bottom-up NO_x emissions in January, April, July, and October 2016 (top-down minus bottom-up).

Table 1 The model performance statistics of daily maximum 8 h averaged (MDA8) O₃ concentrations in January, April, July, and October 2016 with the bottom-up and top-down NO_x emissions.

Table 1 summarizes the observed and simulated daily maximum 8 h averaged (MDA8) O₃ concentrations based on the bottom-up and top-down estimates of NO_x emissions and summarized by month for the YRD region. The MDA8 O₃ concentrations simulated with the top-down estimates were larger than those with the bottom-up ones and were closer to the observations for most months. As most of the YRD was identified as a VOC-limited region (Li et al., 2012; Zhou et al., 2017), the reduced NO_x emissions with the top-down method enhanced the O₃ levels in the AQM. Similar to the hourly concentrations, the most significant improvement for MDA8 was found in January, with the NMB and NME reduced from −35 % and 39 % to 11 % and 28 %, respectively. Moreover, the improvement of April and October for MDA8 was larger than that for the hourly concentrations, indicating that the improved NO_x emissions were more beneficial for the simulation of daytime peak O₃ concentrations in spring and winter. Figure 7 illustrates the spatial distribution of the monthly mean O₃ concentrations simulated based on the top-down NO_x estimates and the differences between the simulations with the top-down and bottom-up estimates by month. In contrast to NO₂, the smaller O₃ concentrations existed in the east–central YRD for most months, as it was identified as a VOC-limited region with a relatively high NO₂ level (Wang et al., 2019). Larger O₃ concentrations were found for the surrounding regions in the YRD, e.g., southern Zhejiang, attributed partly to the relatively abundant BVOC emissions (Li, 2019). An exception existed for July, with clearly larger O₃ concentrations in the east–central YRD. With the largest population density and the most developed economy in the YRD, the area contains a large number of chemical industrial plants and solvent storage, transportation, and usage (Zhao et al., 2017). High temperature in summer promoted the volatilization of chemical products and solvent and thereby enhanced the seasonal VOC emissions more significantly compared to other less developed YRD regions. Moreover, the lowest NO₂ concentration found in summer helped increase the O₃ concentration for the region (Gu et al., 2020). Regarding the simulation difference with two emission estimates, application of the top-down estimates instead of the bottom-up ones elevated the O₃ concentrations in most of the YRD region. In particular, the big reduction in NO_x emissions for the east–central YRD (Fig. 3) was expected to be responsible for the evident growth in O₃ concentrations. As the east–central YRD was identified as a VOC-limited region in terms of O₃ formation, the O₃ concentration in the region would be elevated along with the reduced NO_x emissions, reflecting the negative effect of NO_x control on O₃ pollution alleviation (Wang et al., 2019).

Table 2 Comparison of observed and simulated NO${}_{\mathrm{3}}^{-}$, NH${}_{\mathrm{4}}^{+}$, and SO${}_{\mathrm{4}}^{\mathrm{2}-}$ concentrations by site and season in 2016 (unit: µg/m³). The information on SNA observation sites is provided in Table S2 in the Supplement. BU and TD indicate the CMAQ modeling with the bottom-up and top-down estimate of NO_x emissions, respectively.

3.3 Evaluation of SNA simulation based on the top-down NO_x estimates

Shown in Table 2 is the comparison between the observed and simulated SNA (SO${}_{\mathrm{4}}^{\mathrm{2}-}$, NO${}_{\mathrm{3}}^{-}$, and NH${}_{\mathrm{4}}^{+}$) concentrations by season. Larger observed and simulated SNA concentrations were found in winter and spring, and smaller concentrations were found in summer and autumn. For most seasons, the simulations of NO${}_{\mathrm{3}}^{-}$ concentrations were moderately improved with the top-down estimates of NO_x emissions for all the YRD cities concerned, with an exception of Nanjing in autumn. The largest improvement was found in summer, with the mean bias between the simulation and observations reduced 35 % for all the cities involved. Compared to the bottom-up inventory, the commonly smaller NO_x emissions in the top-down estimates limited the NO₂ concentration and suppressed the formation of NO${}_{\mathrm{3}}^{-}$, while the enhanced O₃ from the reduced NO_x emissions promoted it (Cai et al., 2017; Huang et al., 2020). In summer, the former dominated the process, with the most evident improvement in NO₂ simulation (Fig. 4); thus the reduced NO${}_{\mathrm{3}}^{-}$ concentrations that were closer to observations were simulated for all the cities.

The simulations with both top-down and bottom-up estimates of NO_x emissions underestimated the NH${}_{\mathrm{4}}^{+}$ concentrations for most cases, and such underestimation was slightly corrected with the application of the top-down estimates except for summer. The average change in NH${}_{\mathrm{4}}^{+}$ concentrations was 2.3 %, much smaller than that of NO${}_{\mathrm{3}}^{-}$ at 14 %. The moderate improvement in NH${}_{\mathrm{4}}^{+}$ simulation with the reduced NO_x emissions in the top-down estimates resulted partly from the enhancement of the simulated O₃ concentrations and thereby the promoted NH${}_{\mathrm{4}}^{+}$ formation. In summer, however, the significant drop in the simulated NO₂ concentration was assumed to reduce the NO${}_{\mathrm{3}}^{-}$ and NH${}_{\mathrm{4}}^{+}$ formation and to weaken the consistency between the simulated and observed NH${}_{\mathrm{4}}^{+}$. The difference between the simulated SO${}_{\mathrm{4}}^{\mathrm{2}-}$ with the bottom-up and top-down NO_x emission estimates was small for most seasons, implying a limited benefit of improved NO_x emissions on SO${}_{\mathrm{4}}^{\mathrm{2}-}$ modeling. Besides emission data, the chemical mechanisms included in the model should be important for the model performance. For example, adding SO₂ heterogeneous oxidation in the model could largely improve the sulfate simulation in Nanjing (Sha et al., 2019).

Figure 8 The spatial differences between the simulated SNA concentrations with the bottom-up and top-down NO_x emission estimates for January, April, July, and October 2016 (top-down minus bottom-up).

Figure 8 shows the differences in the spatial distribution of SNA concentrations simulated with the bottom-up and top-down estimates of NO_x emissions by month. In most of the region, the differences of NO${}_{\mathrm{3}}^{-}$ concentrations were larger than those of NH${}_{\mathrm{4}}^{+}$ and SO${}_{\mathrm{4}}^{\mathrm{2}-}$ for all seasons, and they were mainly controlled by the changed ambient NO₂ or O₃ level. The difference in spatial pattern of NO${}_{\mathrm{3}}^{-}$ was similar to that of O₃ for January, and the larger growth attributed to the application of the top-down estimates was found in northern Anhui and eastern Zhejiang (Fig. 8a). The result implies that the change in NO${}_{\mathrm{3}}^{-}$ concentration in winter could result partly from the improved O₃simulation; i.e., the elevated O₃ was an important reason for the enhanced formation of SNA in winter (Huang et al., 2020). Similarly, the increased NO${}_{\mathrm{3}}^{-}$ was found for more than half of the YRD region in April, along with the growth of O₃ concentrations (Fig. 8d). For July, however, the difference in spatial pattern of NO${}_{\mathrm{3}}^{-}$ (Fig. 8g) was similar to NO₂ (Fig. 5g), and the larger reduction attributed to the application of the top-down estimates was found in the northern YRD. The result suggests that the declining NO_x emissions and thereby NO₂ concentration dominated the reduced NO${}_{\mathrm{3}}^{-}$ formation in summer. This was mainly because the reduction of the top-down NO_x emission estimate from the bottom-up emission inventory was much larger for July compared to spring or autumn (Fig. 2). In addition, the VOC-limited mechanism in O₃ formation was found to be weaker in summer than winter (see Fig. 7e and g), resulting in less O₃ formation and thereby nitrate aerosol through oxidation. In October, the growth in NO${}_{\mathrm{3}}^{-}$ concentrations was found again in most of the YRD when the top-down estimates were applied (Fig. 8j). The growth in the north resulted mainly from the increased O₃ level, while that in the south was associated with the increased NO₂. The differences in spatial patterns of simulated NH${}_{\mathrm{4}}^{+}$ concentrations were similar to those of NO${}_{\mathrm{3}}^{-}$ for the 4 months, suggesting that the change in NH${}_{\mathrm{4}}^{+}$ was associated with formation and decomposition of NH₄NO₃. However, the changes of spatial distribution of SO${}_{\mathrm{4}}^{\mathrm{2}-}$ were similar to those of O₃ concentration. Since NH${}_{\mathrm{4}}^{+}$ was preferred to react with SO${}_{\mathrm{4}}^{\mathrm{2}-}$ rather than NO${}_{\mathrm{3}}^{-}$ (Wang et al., 2013), the formation of SO${}_{\mathrm{4}}^{\mathrm{2}-}$ was mainly influenced by the atmospheric oxidizing capacity when only NO_x emissions were changed.

Figure 9 The observed and simulated hourly NO${}_{\mathrm{3}}^{-}$ concentrations with the bottom-up and top-down NO_x emission estimates for January, April, July, and October 2016 at JSPAES.

Figure 9 illustrates the observed and simulated hourly NO${}_{\mathrm{3}}^{-}$ concentrations based on the bottom-up and top-down estimate of NO_x emissions by month at JSPAES. The NMBs and NMEs for simulation with the top-down emissions were smaller than those with bottom-up ones in January and July, implying the benefit of the improved NO_x emissions on hourly NO${}_{\mathrm{3}}^{-}$ concentration simulation in winter and summer. The best model performance with the top-down estimates was found in January, with the hourly variation commonly caught with the AQM. However, the NO${}_{\mathrm{3}}^{-}$ concentration was seriously overestimated, and the model failed to catch the hourly variations in summer indicated by the large NMB and NME. As shown in Fig. S4 in the Supplement, both the NO₂ and O₃ concentrations at JSPAES were significantly overestimated for July except O₃ with the bottom-up NO_x emission estimate, and this partly explained the elevated NO${}_{\mathrm{3}}^{-}$ level from the CMAQ simulation.

Figures S5 and S6 in the Supplement compare the observed and simulated hourly concentrations at JSPAES by month for NH${}_{\mathrm{4}}^{+}$ and SO${}_{\mathrm{4}}^{\mathrm{2}-}$, respectively. The NMBs and NMEs for NH${}_{\mathrm{4}}^{+}$ simulation with the top-down estimates were smaller than those with the bottom-up ones for most months, while the changes in SO${}_{\mathrm{4}}^{\mathrm{2}-}$ concentration were small. The NH${}_{\mathrm{4}}^{+}$ and SO${}_{\mathrm{4}}^{\mathrm{2}-}$ concentrations were largely underestimated with the top-down estimates in January, indicated by the NMB at −44 % and −38 %, respectively. Meanwhile, as shown in Fig. S7 in the Supplement, the SO₂ concentrations were overestimated by 61 % at the site. The results thus imply a great uncertainty in the gas–particle partitioning of (NH₄)₂SO₄ formation in the model in winter, attributed probably to the missed oxidation mechanisms of SO₂ (Chen et al., 2019c).

Table 3 The changed percentages of ozone concentration based on the sensitivity analysis for April 2016.

3.4 Sensitivity analysis of O₃ and SNA formation in the YRD region

Table 3 summarizes the relative changes in the simulated O₃ concentrations for April 2016 in different cases. The mean O₃ concentration would decline by 8.9 % and 19.5 % with 30 % and 60 % VOC emissions off (Cases 2 and 7), while it would increase by 14.2 % and 23.7 % with 30 % and 60 % NO_x emissions off (Cases 1 and 6), respectively. The result confirmed the VOC-limited regime of O₃ formation in the YRD region: controlling VOC emissions was an effective way to alleviate O₃ pollution, while reducing NO_x emissions alone would aggravate O₃ pollution.

Table 4 The changed percentages of NO${}_{\mathrm{3}}^{-}$, NH${}_{\mathrm{4}}^{+}$, and SO${}_{\mathrm{4}}^{\mathrm{2}-}$ concentrations based on the sensitivity analysis for January 2016.

The growth of O₃ concentrations was also found when the reduction rate of NO_x emissions was equal to or larger than that of VOCs. The O₃ concentration would increase by 7.1 % and 14.5 % respectively when both NO_x and VOC emissions were reduced by 30 % and 60 % (Cases 3 and 8), and it would increase by 19.8 % when NO_x and VOC emissions were respectively declined by 60 % and 30 % (Case 5). In contrast, small abatement of O₃ concentrations (2.1 %) was achieved from the 30 % and 60 % reduction of emissions respectively for NO_x and VOCs (Case 4), implying that the O₃ level could be restrained when the reduction rate of VOCs was 2 times that of NO_x or more. To control the O₃ pollution effectively and efficiently, therefore, the magnitude of VOC and NO_x emission reduction should be carefully planned and implemented. In actual fact, controlling VOCs is more difficult than NO_x. Compared to NO_x that come mainly from fossil fuel combustion (Zheng et al., 2018), it is more complicated to identify the sources of specific VOC species that are most active in O₃ formation (Wei et al., 2014; Zhao et al., 2017). Moreover, substantial VOC emissions are from fugitive sources, for which emission control technology can hardly be effectively applied. Therefore, it is a big challenge to control O₃ pollution by reducing more VOCs than NO_x.

Figure 10 The spatial differences of monthly mean O₃ concentrations between the simulations based on the base case (top-down estimates) and sensitivity cases in April 2016 (sensitivity case minus base case).

Figure 10 illustrates the differences in spatial patterns of the simulated monthly mean O₃ concentrations between the base and sensitivity cases in April. The O₃ concentrations were expected to decline for the whole YRD region in the cases of 30 % and 60 % VOC emissions off (Fig. 10b and d), indicating the VOC-limited regime of O₃ formation for the entire YRD. For other cases, the O₃ concentrations were clearly elevated in the central–eastern YRD with relatively large population and developed industry, particularly for the cases with NO_x control only (Fig. 10a and c) or relatively large NO_x abatement together with VOC control (Fig. 10f and g). Even for the case with 60 % of VOC reduction and 30 % of NO_x (Fig. 10h), there was still a small increase in O₃ concentration in the central–eastern YRD, in contrast to the slight O₃ reduction found for most areas of the YRD. These results reveal the extreme difficulty in O₃ pollution control for the region. In southwestern Zhejiang, the O₃ concentrations were found to decline in the cases with large abatement of NO_x emissions (Fig. 10c, f and g), suggesting a shift from a VOC-limited to a NO_x-limited region for the O₃ formation.

Table 4 summarizes the change in the simulated monthly means of SNA (NO${}_{\mathrm{3}}^{-}$, NH${}_{\mathrm{4}}^{+}$, and SO${}_{\mathrm{4}}^{\mathrm{2}-}$) concentrations between the base case and sensitivity cases in January. The SNA concentrations were decreased in most cases, implying that the reduction in precursor emissions was useful for mitigating the SNA pollution. Compared to that of precursor emissions, however, the reduction rate of SNA was much smaller, attributed to the strong nonlinearity of SNA formation. The largest reductions were found at 11.7 % and 12.4 % when emissions of NH₃ and all the three precursors were decreased by 30 % (Cases 11 and 12), respectively. In contrast, the SNA concentrations declined slightly by 1 % and increased by 0.5 % when NO_x and SO₂ emissions were reduced by 30 % (Cases 9 and 10), respectively. The results suggest that most of the YRD was in an NH₃-neutral or even NH₃-poor condition in winter, consistent with the judgment through the AQM based on an updated NH₃ emission inventory (Zhao et al., 2020), as the NH₃ volatilization in winter was much smaller than other seasons. Reducing NH₃ emissions was the most efficient way to control SNA pollution for the region in winter. In Case 11 with NH₃ control only, the reduction in NO${}_{\mathrm{3}}^{-}$ and NH${}_{\mathrm{4}}^{+}$ concentrations was much larger than that of SO${}_{\mathrm{4}}^{\mathrm{2}-}$. As NH₃ reacted with SO₂ prior to NO_x, NH₄NO₃ was assumed easier to decompose than (NH₄)₂SO${}_{\mathrm{4}}^{}$when NH₃emissions were reduced. The growth of NO${}_{\mathrm{3}}^{-}$ concentrations was found for Case 10 (SO₂ control only), since the free NH₃ from the reduced SO₂ emissions could react with NO_x in the NH₃-poor condition. Similarly, the SO${}_{\mathrm{4}}^{\mathrm{2}-}$ concentrations increased for Case 9 (NO_x control only), as the elevated O₃ attributed to the reduction of NO_x emissions promoted the SO${}_{\mathrm{4}}^{\mathrm{2}-}$ formation.

4 Summary

From a top-down perspective, we have estimated the monthly NO_x emissions for the YRD region in 2016, based on nonlinear inversed modeling and NO₂ TVCDs from POMINO, and the bottom-up and top-down estimates of NO_x emissions were evaluated with the AQM and ground NO₂ observations. Due to insufficient consideration of improved controls on power and industrial sources, the NO_x emissions were probably overestimated in the current bottom-up inventory (MEIC), resulting in significantly higher simulated NO₂ concentrations than the observations. The simulated NO₂ concentrations with the top-down estimates were closer to the observations for all four seasons, suggesting the improved emission estimation with satellite constraint. Improved O₃ and SNA simulations with the top-down NO_x estimates for most months indicate the importance role of precursor emission estimation in secondary pollution modeling for the region. Through the sensitivity analysis of O₃ formation, the mean O₃ concentrations were found to decrease for most of the YRD when only VOC emissions were reduced or the reduced rate of VOCs was 2 times that of NO_x, and the result indicates the effectiveness of controlling VOC emissions in O₃ pollution abatement for the region. For part of southern Zhejiang, however, the O₃ concentrations were simulated to decline with the reduced NO_x emissions, implying the shift from a VOC-limited to a NO_x-limited region. Compared to reducing NO_x or SO₂ only, larger reduction in SNA concentrations was found when 30 % of emissions were cut for NH₃ or all the three precursors (NO₂, NH₃, and SO₂). The result suggests that reducing NH₃ emissions was crucial to alleviate SNA pollution of the YRD in winter.

Limitations remain in this study. Due to the limited horizontal resolution of OMI, a relatively big bias existed in the spatial distribution of the constrained NO_x emissions at the regional scale compared to national or continental ones, and the uncertainty could exceed 30 % for the YRD region (Yang et al., 2019b). Therefore the improvement in the top-down estimates of NO_x emissions can be expected when more advanced and reliable products of satellite observations become available at a finer horizontal resolution (e.g., TROPOspheric Monitoring Instrument, TROPOMI). In addition, more SNA observations from online measurement are recommended for a better space coverage and temporal resolution to explore more carefully the response of SNA to the changes in emissions of NO_x and other precursors.