Spaceborne tropospheric nitrogen dioxide (NO₂) observations from 2005–2020 over the Yangtze River Delta (YRD), China: variabilities, implications, and drivers

Nitrogen dioxide (NO₂) is mainly affected by local emission and meteorology rather than long-range transport. Accurate knowledge of its long-term variabilities and drivers is significant for understanding the evolution of economic and social development, anthropogenic emission, and the effectiveness of pollution control measures on a regional scale. In this study, we quantity the long-term variabilities and the underlying drivers of NO₂ from 2005–2020 over the Yangtze River Delta (YRD), one of the most densely populated and highly industrialized city clusters in China, using OMI spaceborne observations and the multiple linear regression (MLR) model. We have compared the spaceborne tropospheric results to surface in situ data, yielding correlation coefficients of 0.8 to 0.9 over all megacities within the YRD. As a result, the tropospheric NO₂ column measurements can be taken as representative of near-surface conditions, and we thus only use ground-level meteorological data for MLR. The inter-annual variabilities of tropospheric NO₂ vertical column density (NO₂ VCD_trop) from 2005–2020 over the YRD can be divided into two stages. The first stage was from 2005–2011, which showed overall increasing trends with a wide range of (1.91 ± 1.50) to (6.70 ± 0.10) × 10¹⁴ molec. cm⁻² yr⁻¹ (p<0.01) over the YRD. The second stage was from 2011–2020, which showed overall decreasing trends of (−6.31 ± 0.71) to (−11.01 ± 0.90) × 10¹⁴ molec. cm⁻² yr⁻¹ (p<0.01) over each of the megacities. The seasonal cycles of NO₂ VCD_trop over the YRD are mainly driven by meteorology (81.01 %–83.91 %), except during winter when anthropogenic emission contributions are pronounced (16.09 %–18.99 %). The inter-annual variabilities of NO₂ VCD_trop are mainly driven by anthropogenic emission (69.18 %–81.34 %), except for a few years such as 2018 which are partly attributed to meteorology anomalies (39.07 %–91.51 %). The increasing trends in NO₂ VCD_trop from 2005–2011 over the YRD are mainly attributed to high energy consumption associated with rapid economic growth, which causes significant increases in anthropogenic NO₂ emission. The decreasing trends in NO₂ VCD_trop from 2011–2020 over the YRD are mainly attributed to the stringent clean air measures which either adjust high-energy industrial structure toward low-energy industrial structure or directly reduce pollutant emissions from different industrial sectors.

1 Introduction

As a major tropospheric pollutant, nitrogen dioxide (NO₂) not only threatens human health and crop growth but is also involved in a series of atmospheric photochemical reactions (Yin et al., 2019; Wang et al., 2011; Geddes et al., 2012). NO₂ is a crucial precursor in the formation of ozone (O₃), particulate matter (PM), acid rain, and photochemical smog in the troposphere (Yin et al., 2021a; Lu et al., 2019a, b; Sun et al., 2018c). Since severe NO₂ pollution increases the risk of respiratory disease and is highly associated with mortality (Meng et al., 2021; MacIntyre et al., 2014; Tao et al., 2012), many countries take the NO₂ level as an important pollution indicator of air quality (Xue et al., 2020). The sources of tropospheric NO₂ are mainly from anthropogenic emissions through high temperature combustion, like transportation (vehicles, ships, and airplanes) and industrial facilities (petrochemicals and power plants) (Zheng et al., 2018b; Chi et al., 2021; van Geffen et al., 2015). Additional minor sources of NO₂ are attributed to natural emissions from the biogeochemical reactions in soil, volcanic eruption, and lightning (Bond et al., 2001; Zhang et al., 2003; Lu et al., 2021). The dominant sink of tropospheric NO₂ is attributed to a chemical destruction, which first converts NO₂ into nitric acid (HNO₃) and peroxyacetyl nitrate (PAN), which then are removed by dry or wet deposition (Browne et al., 2013). Due to a short lifetime of a few hours, tropospheric NO₂ is heavily affected by local emission and meteorology rather than long-range transport (Kim et al., 2015; Cheng et al., 2012; Ji et al., 2019, 2021).

Many scientists have used a suite of active and passive observation technologies on board ground-based, vehicle-based, ship-based, airborne, or spaceborne platforms to assess the temporal–spatial variabilities of NO₂ and identify their driving forces in different regions around the globe (Richter et al., 2005; Jiang et al., 2018; Liu et al., 2018; Zhang et al., 2021; Schreier et al., 2015; Shaiganfar et al., 2017). Among all observation technologies and platforms, spaceborne remote sensing observations have their unique features. By validation using ground-based remote sensing or balloon observations, spaceborne observations can provide a global NO₂ dataset with reasonable accuracy. Typical spaceborne instruments include SCIAMACHY, GOME, OMI, and TROPOMI, which have been widely used in scientific investigations of the global nitrogen cycle, O₃ formation regime, and regional pollution and transport, the quantification of NO₂ emissions from biomass burning regions, megacities, and industrial facilities, and the validation of shipborne observations and atmospheric chemical transport models (CTMs) (Richter et al., 2005; Bechle et al., 2013; Boersma et al., 2011; Ghude et al., 2009; Lamsal et al., 2008). Using spaceborne observations to derive long-term trends of NO₂ and their drivers not only provides valuable information for evaluation of regional emissions, but also improves our understanding of atmospheric evolution. Richter et al. (2005) first investigated the inter-annual variabilities of tropospheric NO₂ vertical column density (NO₂ VCD_trop) from space with GOME and SCIAMACHY observations during 1996–2004. Richter et al. (2005) found substantial reductions in NO₂ VCDs over some areas of Europe and the United States but a highly significant increase of about 50 % – with an accelerating trend in annual growth rate – over the industrial areas of China. In a subsequent study, Ghude et al. (2009) found the same phenomena as Richter et al. (2005) with GOME and SCIAMACHY observations from 1996–2006, which disclosed that NO₂ VCD_trop showed increasing trends over the rapidly developing regions (China: 11 ± 2.6 % yr⁻¹, South Asia: 1.76 ± 1.1 % yr⁻¹, Central East Africa: 2.3 ± 1 % yr⁻¹) and decreasing or leveling off trends over the developed regions (United States: −2 ± 1.5 % yr⁻¹, Europe: 0.9 ± 2.1 % yr⁻¹). With multiple satellite platforms including GOME, SCIAMACHY, OMI, and GOME-2, Hilboll et al. (2013) also found 5 %–10 % yr⁻¹ of increasing trends for NO₂ VCD_trop over eastern Asia during 1996–2011. With OMI observations, Lamsal et al. (2015) quantified the NO₂ trend from 2005 to 2013 over the United States, and Krotkov et al. (2016) investigated the NO₂ trends over different countries for the period of 2005–2014.

Along with the great advances in social and economic development in recent decades, air quality in China has changed dramatically (Sun et al., 2018a, b, 2017, 2020, 2021c, 2022; Yin et al., 2020, 2021c, d; Liu et al., 2022). China has implemented a series of clean air measures in different stages to tackle air pollution across China. One of the landmark clean air measures could be the Action Plan on the Prevention and Control of Air Pollution implemented in 2013, which launched many stringent measures to improve air quality across China. These measures include the reduction of air pollutant emissions, the adjustment of industrial structure and energy mix, the establishment of early-warning systems and monitoring for air pollution, and other compulsive policies (China State Council, 2013). Both spaceborne and ground-based observations have witnessed the effectiveness of these successful policies. The OMI NO₂ VCD_trop decreased by 21 % from 2011–2015 over 48 cities of China (Liu et al., 2017). The national averaged surface NO₂ recorded by the China National Environmental Monitoring Center (CNEMC) network significantly decreased from (16.68 ± 4.82) ppbv in 2013 to (11.29 ± 3.25) ppbv in 2020 (Lin et al., 2021).

In this study, we use NO₂ VCD_trop from 2005–2020 provided by OMI to comprehensively evaluate the long-term trends, implications, and underlying drivers of NO₂ over the Yangtze River Delta (YRD; including Anhui, Jiangsu, Shanghai, and Zhejiang provinces; Table S1). In addition to anthropogenic emission, meteorology also drives NO₂ variability by affecting emissions, transport, chemical production, and scavenging. The relationships of NO₂ against meteorological variables are complex and are region- and time-dependent. In the present work, we separate the contributions of meteorology and anthropogenic emission to the NO₂ variability using the multiple linear regression (MLR) model over the major cities (Hefei, Nanjing, Suzhou, Shanghai, Hangzhou, Ningbo) within the YRD. As one of the three most densely populated and highly industrialized city clusters in China, the YRD has long been identified as a key region for air pollution mitigation. This study can not only improve our understanding of temporal–spatial NO₂ evolution in the atmosphere, but it also provides valuable information for future clean air policy. We introduce detailed descriptions of OMI and ground-level NO₂ products in Sect. 2.1, and meteorological fields in Sect. 2.2. The method for separating contributions of meteorology and anthropogenic emission is presented in Sect. 2.3. Section 3.1 and 3.2 analyze the temporal–spatial variabilities of tropospheric NO₂ from 2005–2020 over the YRD on provincial and megacity levels, respectively. A comparison between the OMI NO₂ product and the ground-level measurements is performed in Sect. 3.3. We discuss the implications and underlying drivers of the variabilities of tropospheric NO₂ from 2005–2020 over the YRD in Sect. 4. We conclude this study in Sect. 5.

2 Data and method

2.1 Observation data

2.1.1 OMI NO₂ product

OMI is a hyperspectral atmospheric composition detection instrument on board the National Aeronautics and Space Administration (NASA) Aura Earth Observing System (EOS) satellite, launched in July 2004 (Boersma et al., 2007). The EOS satellite flies over a low-Earth orbit at an altitude of about 710 km. The local overpass time (LT) of OMI satellite is about 13:30 in the early afternoon. The retrieval micro-window for NO₂ VCDs lies between 405 and 465 nm, with a spectral resolution of about 0.5nm (Marchenko et al., 2015). The spatial resolution of OMI measurements is 13 × 24 km² at nadir. OMI provides observations of O₃, NO₂, SO₂, aerosol, cloud, HCHO, BrO, and OClO, with nearly daily global coverage (Levelt et al., 2006). The daily LV3 NO₂ VCD_trop data product (GES DISC; http://disc.sci.gsfc.nasa.gov, last access: 1 September 2021), which is a gridded dataset with a 0.25^∘ × 0.25^∘ spatial resolution, is used in this study. The NO₂ VCD_trop values are calculated by the stratosphere–troposphere separation (STS) scheme proposed by numerous previous studies (Bucsela et al., 2013; Lamsal et al., 2014; Goldberg et al., 2017). The STS scheme first subtracts the stratospheric NO₂ slant column densities (SCDs) from the total NO₂ SCDs, and then it divides the resulting tropospheric NO₂ SCDs by the tropospheric air mass factor (AMF). The formulation for calculating NO₂ VCD_trop is as follows:

$$\begin{array}{}\text{(1)}& {\mathrm{VCD}}_{\mathrm{trop}}={\displaystyle \frac{{\mathrm{SCD}}_{\mathrm{total}}-{\mathrm{SCD}}_{\mathrm{strat}}}{{\mathrm{AMF}}_{\mathrm{trop}}}},\end{array}$$

where AMF is defined as the ratio of the SCD to the VCD (Solomon et al., 1987),

$$\begin{array}{}\text{(2)}& {\mathrm{AMF}}_{\mathrm{trop}}={\displaystyle \frac{{\mathrm{SCD}}_{\mathrm{trop}}}{{\mathrm{VCD}}_{\mathrm{trop}}}}.\end{array}$$

The tropospheric AMF is calculated by NO₂ profiles simulated by the Global Modeling Initiative (GMI) chemistry transport model with a horizontal resolution of 1^∘ × 1.25^∘ (Rotman et al., 2001). Separation of stratospheric and tropospheric columns is achieved by the local analysis of the stratospheric field over unpolluted areas (Bucsela et al., 2013). The OMI NO₂ VCD_trop dataset has been used in many studies to investigate the O₃ formation regime and regional pollution and transport (Lin et al., 2010; Zhang et al., 2017; Duncan et al., 2013; Liu et al., 2016). In this study, only the LV3 data product collected with cloud radiance fractions of less than 30 % is used (Streets et al., 2013).

2.1.2 Ground-level NO₂ data

We extract ground-level NO₂ data over the YRD from the China National Environmental Monitoring Center (CNEMC) network (http://www.cnemc.cn/en/, last access: 26 November 2021). The CNEMC network has operated more than 3000 monitoring sites that covered almost all major cities over China by 2020. The CNEMC datasets have been used in many studies for the evaluation of regional atmospheric pollution and transport (Li et al., 2019, 2021; Lu et al., 2019a, 2020; Sun et al., 2021a; Yin et al., 2021a; Zhao et al., 2016; He et al., 2017). As one of the six key atmospheric pollutants (CO, SO₂, NO₂, PM₁₀, O₃, and PM_2.5) routinely measured by the CNEMC network, ground-level NO₂ measurements at 188 sites in 40 cities over the YRD have been available since 2014. In this study, comparisons between the OMI NO₂ data product and the ground-level NO₂ measurements are only performed over six key megacities, i.e., Shanghai, Nanjing, Hangzhou, Suzhou, Ningbo, and Hefei, within the YRD. The population, geolocation, number of measurement sites, and data range of each city are summarized in Table 1. The number of measurement sites in each city ranges from 8 to 11, the altitude ranges from 3 to 50 m (above sea level, a.s.l.), and the population ranges from 0.9 to 2.5 million. All ground-level NO₂ data at each station are measured by active differential absorption ultraviolet (UV) analyzers. We use a data quality control method following previous studies to remove unreliable NO₂ data (Lu et al., 2019a, 2020; Sun et al., 2021a; Yin et al., 2021a). Specifically, we first convert all hourly measurements into Z scores, and we then remove the measurement if its Z score meets one of the following rules: (1) Z_i is larger or smaller than the previous value Z_i−1 by 9 (|Z_i-$Z_{i-\mathrm{1}}|\mathit{>}\mathrm{9}$); (2) the absolute value of Z_i is greater than 4 ($\left|Z_i\right|\mathit{>}\mathrm{4}$); or (3) the ratio of the Z value to the third-order center moving average is greater than 2 ($\frac{\mathrm{3}{Z}_i}{Z_{i-\mathrm{1}}+Z_i+Z_{i+\mathrm{1}}}\mathit{>}\mathrm{2}$), where i represents the ith hourly measurement data. After removing outliers with the above filter criteria, we finally average NO₂ data at all measurement sites in each city to form a city-representative NO₂ dataset.

Table 1 Geolocation, the number of measurement sites, and population for the six megacities within the YRD. Population statistics are based on the seventh nationwide population census in 2020 provided by the National Bureau of Statistics of China.

2.2 Meteorological fields

We obtain meteorological fields during 2005–2020 from the second Modern-Era Retrospective analysis for Research and Applications (MERRA-2) (Gelaro et al., 2017). This dataset is produced by the NASA Global Modeling and Assimilation Office (https://gmao.gsfc.nasa.gov/reanalysis/MERRA-2/, last access: 1 August 2021) with a spatial resolution of 0.5^∘ × 0.625^∘ and temporal resolutions of 1 h for boundary layer height and surface meteorological variables and 3 h for other variables. Previous studies have verified that meteorological fields provided by MERRA-2 match well with the meteorological parameters observed by Chinese weather stations (Song et al., 2018; Carvalho, 2019; Wang et al., 2017; Kishore Kumar et al., 2015; Zhou et al., 2017). In order to match OMI observations which are available at about 13:30 LT, the average for meteorological data is only performed between 13:00 and 14:00 LT.

2.3 Multiple linear regression (MLR) model

We establish a multiple linear regression (MLR) model to quantify the contributions of meteorology and anthropogenic emission to the long-term variabilities of NO₂ VCD_trop during 2005–2020 over the YRD. Similar MLR methodologies have been used in previous studies to estimate the contributions of meteorology and emission to the variabilities of O₃ and PM_2.5 in North America, Europe, and China (Li et al., 2019, 2020; Xu et al., 2011; Zhai et al., 2019; Zhao and Wang, 2017). The meteorological parameters used in our MLR model are elaborated in Table 2.

Table 2 Meteorological parameters used in the MLR model.

In order to highlight the variabilities of NO₂ VCD_trop, we follow the method of previous studies and calculate NO₂ VCD_trop anomalies (y_anomaly) by subtracting a reference value (y_reference) from all tropospheric NO₂ observations (y_individual) (Hakkarainen et al., 2016, 2019; Mustafa et al., 2021). The formulation of this method is expressed as

$$\begin{array}{}\text{(3)}& {\mathit{y}}_{\mathrm{anomaly}}={\mathit{y}}_{\mathrm{individual}}-{\mathit{y}}_{\mathrm{reference}}.\end{array}$$

In this study, we take the average of all NO₂ VCD_trop values from 2005–2020 (i.e., the 16-year mean) as the reference value. The MLR model for each city is explained as

$$\begin{array}{}\text{(4)}& \mathit{y}={\mathit{\beta }}_{\mathrm{0}}+{\sum }_{k=\mathrm{1}}^{\mathrm{11}}{\mathit{\beta }}_k{\mathit{x}}_k,\end{array}$$

where y is the regression result for monthly OMI NO₂ VCD_trop anomalies, β₀ is the intercept, and x_k (k∈[1, 11]) is the meteorological variables. The regression coefficients β_k are calculated by nonlinear least squares fitting. This MLR model finds the optimal regression result by minimizing the sum of squares of the fitting residual and then solves regression coefficients β_k by the following equation:

$$\begin{array}{}\text{(5)}& {\mathit{\beta }}_k={\left(\sum {\mathit{x}}_k{\mathit{x}}_k^T\right)}^{-\mathrm{1}}\left(\sum {\mathit{x}}_k{\mathit{y}}_k\right).\end{array}$$

The regression results y represent the meteorology-induced contributions to the variabilities of NO₂ VCD_trop. Since both soil and lighting NO_x are meteorology-dependent, the effects of soil and lighting NO_x on NO₂ variability are also attributed to meteorology contribution. The difference y^′ between the monthly OMI NO₂ VCD_trop anomalies y_anomaly and y calculated as Eq. (6) represents the portion that cannot be explicitly explained by the meteorological influence.

$$\begin{array}{}\text{(6)}& {\mathit{y}}^{\prime }={\mathit{y}}_{\mathrm{anomaly}}-\mathit{y}\end{array}$$

By subtracting the meteorological influence from the total NO₂ amounts, the y^′ is referred to as the aggregate contribution of anthropogenic emission. Positive y and y^′ indicate that meteorology and anthropogenic emission cause NO₂ VCD_trop above the reference value (i.e., the 16-year mean), respectively. In contrast, negative y and y^′ indicate that meteorology and anthropogenic emission cause NO₂ VCD_trop below the reference value, respectively.

Since the meteorological parameters listed in Table 2 differ in their units and magnitudes, which could lead to unstable performance of the model, we normalized all meteorological parameters via Eq. (7) before using them in regression. This normalization preprocessing procedure can also speed up the convergence of the MLR model.

$$\begin{array}{}\text{(7)}& {\mathit{z}}_k={\displaystyle \frac{{\mathit{x}}_k-{\mathit{u}}_k}{{\mathit{\sigma }}_k}},\end{array}$$

where u_k and σ_k are the average and 1σ standard deviation (SD) of x_k, and z_k is the normalized value for parameter x_k.

3 Temporal–spatial variabilities of NO₂ VCD_trop over the Yangtze River Delta

3.1 Variabilities at provincial level

We present the temporal–spatial distribution of the annual averaged NO₂ VCD_trop over the YRD from 2005–2020 in Fig. 1. The major pollution areas for NO₂ VCD_trop over the YRD are located in the south of Jiangsu Province and the north of Zhejiang Province. In addition, NO₂ pollution in the eastern Anhui Province showed an increasing trend during 2005–2013 and became one of the major pollution areas within the YRD during 2010–2013. The amplitudes of NO₂ VCD_trop over the YRD showed large year-to-year variabilities from 2005–2020, but spatial extensions of the major pollution areas are almost constant over years. Among all the pollution areas, the heaviest pollution regions are uniformly located in the densely populated and highly industrialized megacities, such as Shanghai, Nanjing, Suzhou, Hangzhou, Ningbo, and Hefei.

Figure 1 Temporal–spatial variabilities of NO₂ VCD_trop provided by the OMI satellite over the YRD from 2005–2020. The three provinces (Anhui, Jiangsu, and Zhejiang) and six key megacities (Hefei, Nanjing, Suzhou, Shanghai, Hangzhou, and Ningbo) are marked.

The annual means and seasonal cycles of NO₂ VCD_trop over the YRD during 2005–2020 at Province or municipality level, i.e., Anhui Province, Jiangsu Province, Zhejiang Province, and Shanghai municipality, are presented in Fig. 2. The NO₂ VCD_trop over each province is calculated by averaging all observations within the boundary of each province. For seasonal variability, clear seasonal features over the whole YRD region and each province are observed (Fig. 2a): (1) high levels of NO₂ VCD_trop occur in late winter to spring, and low levels of NO₂ VCD_trop occur in later summer to autumn; (2) the 1σ SDs in late winter to spring are larger than those in later summer to autumn; and (3) seasonal cycles of NO₂ VCD_trop over Jiangsu, Zhejiang, and the whole YRD region show bimodal patterns; i.e., two seasonal peaks occur around March and December or January, and one seasonal trough occurs around September. However, over Anhui these cycles show a unimodal pattern and do not have a peak around March. The NO₂ VCD_trop has a maximum monthly mean value of (1.93 ± 0.21), (2.40 ± 0.25), (1.61 ± 0.16), and (1.91 ± 0.16) × 10¹⁶ molec. cm⁻² in January or December over Anhui, Jiangsu, Zhejiang, and the whole YRD region, respectively. The minimum monthly mean values over Anhui, Jiangsu, Zhejiang, and the whole YRD region occur in July, with values of (0.35 ± 0.05), (0.83 ± 0.07), (0.57 ± 0.06), and (0.39 ± 0.01) × 10¹⁶ molec. cm⁻², respectively.

Figure 2 (a) Monthly averaged NO₂ VCD_trop over the whole YRD region (green dots and lines), Anhui Province (black dots and lines), Zhejiang Province (blue dots and lines), and Jiangsu Province (yellow dots and lines). (b) Same as (a) but for annual average. The vertical error bar is 1σ standard deviation (SD) within that month or year.

Except for a few anomalies such as the year-to-year decrease in 2005–2006, and the increases in 2016–2017 and 2018–2019, the overall inter-annual variabilities of NO₂ VCD_trop over the YRD can be divided into two stages (Fig. 2b). The first stage was from 2005–2011, which showed overall increasing trends in NO₂ VCD_trop over the YRD. During 2005–2009 of this stage, change rates of NO₂ VCD_trop were less pronounced, where the 2009 levels relative to 2005 only increased by (0.33 ± 0.02) × 10¹⁵ (3.96 ± 0.25) %, (1.05 ± 0.11) × 10¹⁵ (8.55 ± 0.08) %, and (0.46 ± 0.03) × 10¹⁵ molec. m⁻² (5.05 ± 0.32) % over Anhui, Jiangsu, and the whole YRD region, respectively, and thy leveled off over Zhejiang. However, NO₂ VCD_trop in 2011 relative to 2009 showed significantly increments of (2.88 ± 0.23) × 10¹⁵ (33.78 ± 2.70) %, (3.81 ± 0.32) × 10¹⁵ (29.01 ± 2.45) %, (2.08 ± 0.18) × 10¹⁵ (27.97 ± 2.43) %, and (2.10 ± 0.19) × 10¹⁵ molec. m⁻² (21.59 ± 1.95) % over Anhui, Jiangsu, Zhejiang, and the whole YRD region, respectively. The second stage was from 2011–2020, which showed overall decreasing trends in NO₂ VCD_trop over the YRD. The total decrements over Anhui, Jiangsu, Zhejiang, and the whole YRD region in 2020 relative to 2011 are (4.91 ± 0.39) × 10¹⁵ (41.48 ± 3.30) %, (4.82 ± 0.31) × 10¹⁵ (43.25 ± 2.72) %, (3.78 ± 0.36) × 10¹⁵ (40.47 ± 4.12) %, and (4.82 ± 0.35) × 10¹⁵ molecule m⁻² (43.26 ± 3.07) %, respectively.

We have followed the methodology of Li et al. (2020) and used the linear regression model to estimate the inter-annual trends of NO₂ VCD_trop over the YRD (Table 3). During 2005–2011, inter-annual trends of NO₂ VCD_trop over the YRD region and each province spanned a wide range of (1.74 ± 0.72) × 10¹⁴ molec. cm⁻² yr⁻¹ (p=0.02) to (5.94 ± 1.01) × 10¹⁴ molec. cm⁻² yr⁻¹ (p<0.01), indicating the regional representativity of each dataset. In contrast, inter-annual trends of NO₂ VCD_trop over the YRD region and each province from 2011–2020 varied over (−4.86 ± 0.49) to (−8.16 ± 0.82) × 10¹⁴ molec. cm⁻² yr⁻¹ (p<0.01). For the aggregate trends during 2005–2020, NO₂ VCD_trop values over the whole YRD region and each province are negative. The largest and lowest decreasing trends are observed in Jiangsu and Anhui, with values of (−1.92 ± 0.30) × 10¹⁴ molec. cm⁻² yr⁻¹ (p<0.01) and (−0.92 ± 0.26) × 10¹⁴ molec. cm⁻² yr⁻¹ (p<0.01), respectively.

Table 3 Inter-annual trends of NO₂ VCD_trop over each province within the YRD and the whole YRD region during 2005–2011, 2011–2020, and 2005–2020.

3.2 Variabilities at megacity level

The annual means and seasonal cycles of NO₂ VCD_trop over the major megacities within the YRD during 2005–2020 are presented in Fig. 3. Similar to the derivation of provincial level NO₂, NO₂ VCD_trop over each megacity is calculated by averaging all observations within the boundary of each megacity. The results show that the amplitudes and variabilities of NO₂ VCD_trop at megacity level are basically coincident with those at the corresponding provincial levels. Overall, the amplitudes and 1σ SDs of NO₂ seasonal cycles in cold seasons are larger than those in warm seasons, and the inter-annual NO₂ variabilities at megacity level can also be divided into two stages, i.e., an overall increasing stage during 2005–2011 and a decreasing stage during 2011–2020. As a result, it is feasible to select these major megacities as representatives for mapping the drivers of NO₂ variabilities over the YRD.

Figure 3 (a) Monthly averaged NO₂ VCD_trop over Hefei (black dots and lines), Nanjing (blue dots and lines), Shanghai (yellow dots and lines), Suzhou (red dots and lines), Hangzhou (green dots and lines), and Ningbo (cyan dots and lines). (b) Same as (a) but for annual average. The vertical error bar is 1σ standard variation within that month or year.

Specifically, megacity levels of NO₂ VCD_trop show seasonal maxima in December and seasonal minima in July. Seasonal maxima over Hefei, Shanghai, Nanjing, Suzhou, Hangzhou, and Ningbo are (2.03 ± 0.15), (2.80 ± 0.23), (2.62 ± 0.25), (2.66 ± 0.16), (1.83 ± 0.18), and (2.27 ± 0.21) × 10¹⁶ molec. cm⁻², and seasonal minima are (0.34 ± 0.04), (0.83 ± 0.11), (0.58 ± 0.06), (0.62 ± 0.05), (0.32 ± 0.02), and (0.38 ± 0.03) × 10¹⁶ molec. cm⁻², respectively. The seasonal maxima are on average (82.27 ± 2.34) %, (67.19 ± 1.56) %, (71.06 ± 2.32) %, (83.33 ± 3.05) %, (77.62 ± 2.89) %, and (70.84 ± 2.76) % higher than the seasonal minima over respective megacities. As commonly observed, the seasonal variability of NO₂ VCD_trop with respect to their annual means spanned a wide range of −55.1 % to 103.5 %, depending on season and measurement time (Fig. 3a).

The NO₂ VCD_trop in all megacities shows maximum values in 2011, where the maximum values over Hefei, Shanghai, Suzhou, Ningbo, Nanjing, and Hangzhou are (1.41 ± 0.25), (2.18 ± 0.23), (1.81 ± 0.17), (1.39 ± 0.12), (1.88 ± 0.18), and (1.19 ± 0.14) × 10¹⁶ molec. cm⁻², respectively (Fig. 3b). In terms of the increments relative to the 2005 levels, Hefei and Shanghai from 2005 to 2011 have the largest and lowest increments of (5.37 ± 0.51) × 10¹⁵ molec. cm⁻² (61.77 ± 5.87) % and (2.62 ± 0.27) × 10¹⁵ molec. cm⁻² (14.68 ± 1.51) %, respectively. The increments over other cities varied over (3.31 ± 0.32) × 10¹⁵ molec. cm⁻² (31.20 ±  3.02) % to (5.21 ± 0.41) × 10¹⁵ molec. cm⁻² (38.40 ± 3.02) %. In terms of the decrements relative to the 2011 levels, Shanghai and Hangzhou from 2011 to 2020 have the largest and lowest decrements of (9.77 ± 0.82) × 10¹⁵ molec. cm⁻² (46.89 ± 3.94) and (5.28 ± 0.45) × 10¹⁵ molec. cm⁻² (45.43 ± 3.87) %, respectively. The decrements over other cities are also evident and varied over (6.33 ± 0.58) × 10¹⁵ molec. cm⁻² (45.53 ± 4.18) % to (9.05 ± 0.98) × 10¹⁵ molec. cm⁻² (48.12 ± 5.21) %. A few anomalies are also observed in some megacities and are in good agreement with the corresponding provincial levels. For example, NO₂ VCD_trop over Hefei and Suzhou had increased by (0.09 ± 0.01) × 10¹⁵ molec. cm⁻² (0.77 ± 0.09) % and (0.80 ± 0.07) × 10¹⁵ molec. cm⁻² (4.90 ± 0.43) % in 2013 relative to 2012 levels, respectively. In addition, NO₂ VCD_trop over Hefei, Shanghai, Nanjing, Hangzhou, and Suzhou had increased by (0.65 ± 0.12) × 10¹⁵ (8.41 ± 1.55) %, (0.35 ± 0.02) × 10¹⁵ (2.66 ± 0.15) %, (0.86 ± 0.18) × 10¹⁵ (8.16 ± 1.71) %, (0.55 ± 0.08) × 10¹⁵ (8.68 ± 1.26) %, and (0.29 ± 0.05) × 10¹⁵ molec. cm⁻² (2.52 ± 0.43) % in 2019 relative to 2018 levels, respectively.

The inter-annual trends of NO₂ VCD_trop during 2005–2011 over all cities are positive and span a wide range of (1.91 ± 1.50) to (6.70 ± 0.10) × 10¹⁴ molec. cm⁻² yr⁻¹ (p<0.01) (Table 4). In contrast, the inter-annual trends of NO₂ VCD_trop during 2011–2020 over all cities are negative. The largest and lowest decreasing trends are observed in Nanjing and Hangzhou, with values of (−11.01 ± 0.90) and (−6.31 ± 0.71) × 10¹⁴ molec. cm⁻² yr⁻¹ (p<0.01), respectively. For the aggregate trends during 2005–2020, NO₂ VCD_trop over all cities is negative. The largest and lowest decreasing trends are observed in Shanghai and Hefei, with values of (−4.58 ± 0.43) × 10¹⁴ molec. cm⁻² yr⁻¹ (p<0.01) and (−0.30 ± 3.43) × 10¹⁴ molec. cm⁻² yr⁻¹ (p=0.385), respectively.

Table 4 Inter-annual trends of NO₂ VCD_trop over each city within the YRD during 2005–2011, 2011–2020, and 2005–2020.

3.3 Comparisons with the CNEMC data

In order to investigate if satellite column measurements can represent the near-surface variabilities, we have compared the OMI NO₂ VCD_trop data over the six megacities within the YRD with the ground-level NO₂ data provided by the CNMEC (Fig. 4). The comparisons over all megacities were performed on a monthly basis between June 2014 and December 2020. Ground-level NO₂ concentrations were taken as the average of all CNMEC stations in each city. The NO₂ VCD_trop values were taken as the average of all OMI observed grids within the scope of each city. Considering the overpass time of OMI is at about 13:30 LT, we only average the ground-level NO₂ data between 13:00 and 14:00 LT for comparison, which ensures that the temporal differences between the CNMEC and OMI dataset are all within ± 30 min. With these rules, there are over 700 matching samples in each city available for comparison.

Figure 4 Correlation of OMI NO₂ VCD_trop against ground-level observations data over Hefei, Nanjing, Shanghai, Suzhou, Hangzhou, and Ningbo. We fitted both datasets directly without unifying their units, which does not affect the investigation with respect to the agreement of the two datasets in terms of variabilities. Blue lines are linear fitted lines, and black lines are the 1 : 1 line.

Correlation plots of OMI NO₂ VCD_trop data against the CNMEC ground-level NO₂ measurements are shown in Fig. 4. The results show that the NO₂ variabilities observed by OMI and the CNMEC are in good agreement over all megacities, with correlation coefficients (r²) of 0.88, 0.81, 0.89, 0.88, 0.86, and 0.83 for Hangzhou, Hefei, Nanjing, Ningbo, Shanghai, and Suzhou, respectively. The discrepancies between OMI and CNMEC data can be mainly attributed to their differences in temporal–spatial resolutions. OMI averages NO₂ concentration at about 13:30 LT over a large coverage due to its relatively coarse spatial resolution (Wallace and Kanaroglou, 2009; Zheng et al., 2014). The CNMEC data represent the averaged point concentrations between 13:00 and 14:00 LT around the measurement site. NO₂ is a short-lifetime species and is characterized by large temporal–spatial variabilities. Any temporal–spatial inhomogeneity in NO₂ concentration could affect the comparison (Meng et al., 2010; Wallace and Kanaroglou, 2009). Considering the above differences, the correlations of the two datasets over all megacities are satisfactory. The tropospheric NO₂ column measurements can be used as representatives of near-surface conditions. As a result, to simplify calculations, we only use ground-level meteorological data for MLR.

Over a polluted atmosphere, the NO₂ column measurements can be taken as representative of near-surface conditions because tropospheric NO₂ has a vertical distribution that is heavily weighted toward the surface (Kharol et al., 2015; Zhang et al., 2017; Duncan et al., 2013, 2016; Kramer et al., 2008). Many studies have taken advantage of this favorable vertical distribution of NO₂ to derive surface emissions of NO₂ from space (Silvern et al., 2019; Boersma et al., 2009; Streets et al., 2013; Anand and Monks, 2017; Lu et al., 2015; Ghude et al., 2013; Cooper et al., 2020). Meanwhile, the use of NO₂ column measurements to explore tropospheric O₃ sensitivities has been the subject of several past studies, which disclosed that this diagnosis of O₃ production rate (PO₃) is consistent with the findings of surface photochemistry (Baruah et al., 2021; Choi and Souri, 2015; Jin et al., 2017, 2020; Jin and Holloway, 2015; Schroeder et al., 2017; Souri et al., 2017; Sun et al., 2021b, 2018c; Yin et al., 2021b).

4 Implications and drivers

We incorporate the 11 meteorological parameters listed in Table 2 into the MLR model to fit the time series of monthly averaged NO₂ VCD_trop from 2005–2020 over the six megacities within the YRD (Fig. S1). Correlation plots of the MLR results and the satellite tropospheric NO₂ data are shown in Fig. 5. The results show that the MLR model can reproduce the seasonal variabilities of tropospheric NO₂ VCDs over each city well, with correlation coefficients of 0.85 to 0.90. We separate the contributions of meteorology and anthropogenic emission to the NO₂ variability over the six megacities with the methodology described in Sect. 2.3. Figure 6 shows monthly averaged tropospheric NO₂ VCDs along with the meteorological-driven contributions and the anthropogenic-driven contributions in each city. Figure 7 is the same as Fig. 6, but the statistics are based on annual average.

Figure 5 Correlations of OMI NO₂ VCD_trop against the MLR model results over Hefei, Nanjing, Shanghai, Suzhou, Hangzhou, and Ningbo. Blue lines are linear fitted lines, and black lines are the 1 : 1 line.

4.1 Drivers of seasonal cycles of NO₂ VCD_trop

As shown in Fig. 6 for all megacities, the seasonal variabilities of meteorological contributions are consistent with those of NO₂ VCD_trop except the period from February to March, and the anthropogenic contributions varied around zero throughout the year except in December and February. This means that the seasonal variabilities of tropospheric NO₂ over the YRD are mainly determined by meteorology (81.01 %–83.91 %) and also influenced by anthropogenic emission in December and February. Meteorological contributions are larger than zero in winter and lower than zero in summer, indicating that meteorology increases the NO₂ level in winter and decreases the NO₂ level in summer. This contrast in meteorological contribution is associated with the seasonal cycle of temperature. Similarly, anthropogenic contributions are larger than zero in December and lower than zero in February, representing anthropogenic emission increases NO₂ level in December and decreases NO₂ level in February. The enhanced anthropogenic contributions in December are mainly attributed to more extensive anthropogenic activities such as residential heating in megacities in this period, which usually results in more anthropogenic NO₂ emission due to the increase in energy and fuel consumption. The decreased anthropogenic contributions in February are attributed to the Spring Festival. We elaborate the analysis below.

Figure 6 Monthly averaged NO₂ VCD_trop (red dots and lines) along with the meteorological-driven portions (blue dots and lines) and the anthropogenic-driven portions (black dots and lines) over each city within the YRD. The vertical error bar is 1σ standard deviation (SD) within that month.

As shown in Fig. S2, the vast majorities of meteorological contributions over all megacities are from temperature, and additional minor contributions over some cities such as Nanjing, Shanghai, and Suzhou are attributed to relative humidity, pressure, or surface incoming shortwave flux (SWGDN) (Agudelo–Castaneda et al., 2014; Parra et al., 2009). Significant negative correlations between temperature and NO₂ VCD_trop are observed in all megacities (Fig. S3, Table 5). Higher temperature tends to decrease NO₂ VCD_trop and vice versa. This is because higher temperature conditions could accelerate the chemical reaction that destructs NO₂ in the troposphere (Pearce et al., 2011; Yin et al., 2021a). In addition, surface pressure shows high positive correlation, and both surface relative humidity and SWGDN show negative correlations with NO₂ VCD_trop, but their contribution levels are much lower than the temperature. All other meteorological variables only have weak correlations with NO₂ VCD_trop (Table 5).

Table 5 Correlations of monthly averaged observations against each meteorological parameter from 2005–2020.

In all cities except Hefei, there is a significant increase in NO₂ level from February to March. The maximum and minimum increments occur in Shanghai and Nanjing, with values of (3.28 ± 0.29) × 10¹⁵ molec. cm⁻² (16.37 ± 1.45) % and (0.47 ± 0.05) × 10¹⁵ molec. cm⁻² (2.60 ± 0.28) %, respectively. In contrast, the meteorological contributions show decreased change rates in the same period. As a result, this increase in NO₂ level from February to March could be attributed to anthropogenic emission rather than meteorology. Indeed, anthropogenic contributions show significant increases of (3.95 ± 0.32) to (6.53 ± 0.55) × 10¹⁵ molec. cm⁻² over all megacities from February to March. The most important festival in China – the Spring Festival – typically occurs in February, when a large number of migrants in megacities return to their hometowns for holiday, and most industrial production is shut down, which could cause significant reductions in anthropogenic emission. In March, these migrants get back to work, and all industrial enterprises resumed production, which could cause a rebound in anthropogenic emission. The seasonal maxima of NO₂ in March are not observed in Hefei because the anthropogenic-emission-induced increases are offset by meteorology-induced decreases.

The year 2020 is a special year compared to all other years, when a large-scale lockdown occurred in February, and some regional travel restrictions occasionally occurred in other seasons across China due to the COVID-19 pandemic. In the comparison, we removed all NO₂ measurements in 2020 to eliminate the influence of COVID-19. The monthly averaged NO₂ VCD_trop values from 2005–2019 along with the meteorological contributions and the anthropogenic contributions in each city are shown in Fig. S4. Figures S5 and S6 are the same as Figs. 2 and 3, respectively, but for 2011–2019. We obtained the same conclusion as that from Fig. 6, indicating that the drivers of seasonal cycles of NO₂ VCD_trop deduced above are consistent over the years.

4.2 Drivers of inter-annual variabilities of NO₂ VCD_trop

As shown in Fig. 7 for all megacities, the inter-annual variabilities of anthropogenic contributions are in good agreement with those of NO₂ VCD_trop, indicating that inter-annual variabilities of NO₂ VCD_trop are mainly driven by anthropogenic emission. The same as those of NO₂ VCD_trop, the inter-annual anthropogenic contributions over each city can also be divided into two stages, i.e., an overall increasing stage during 2005–2011 and a decreasing stage during 2011–2020. For the first stage (2005–2011), anthropogenic contributions account for 84.72 %, 92.96 %, 93.52 %, 79.06 %, 97.12 %, and 90.21 % of the increases in NO₂ VCD_trop, while meteorological contributions account for 15.28 %, 7.04 %, 6.48 %, 20.94 %, 2.88 %, and 9.79 % over Hangzhou, Hefei, Nanjing, Ningbo, Shanghai, and Suzhou, respectively. The annual averaged meteorological contributions over each city varied around zero in all years, except for a few anomalies in some years. For example, meteorological contributions over all cities are larger than zero in 2005 and 2011 but lower than zero after 2014. Pronounced anomalies include the enhancements that occurred in 2011 in all cities and the decrements in 2015 over Suzhou, in 2018 over Hangzhou, and in 2016 over other cities. All these anomalies in meteorological contributions are highly correlated with temperature anomalies (Fig. S7). As shown in Figs. S8 and S9, the temperature in all cities is lower than the reference value (i.e., the 16-year mean) in 2005 and 2011 and larger than the reference value after 2014. As a result, in addition to anthropogenic emission, the NO₂ enhancements in 2011 are partly attributed to the lower temperature in this year. Meanwhile, higher temperature in the YRD region in recent years favors the decrease in NO₂ VCD_trop. For the second stage (2011–2020), anthropogenic contributions account for 70.15 %, 65.22 %, 66.97 %, 73.45 %, 74.43 %, and 73.84 % of the decreases in NO₂ VCD_trop, while meteorological contributions account for 29.85 %, 34.78 %, 33.03 %, 26.55 %, 25.57 %, and 26.16 % over Hangzhou, Hefei, Nanjing, Ningbo, Shanghai, and Suzhou, respectively.

Figure 7 The same as Fig. 6 but for annual average.

Since anthropogenic NO₂ emissions are highly related to economic and industrial activities (Lin and McElroy, 2011; Russell et al., 2012; Vrekoussis et al., 2013; Guerriero et al., 2016), to understand the inter-annual variabilities of NO₂ VCD_trop, we have investigated the inter-annual variabilities of gross domestic product (GDP) over the YRD from the primary sector, secondary sector, and tertiary sector (http://www.stats.gov.cn/, last access: 1 August 2021) from 2005–2020. The primary sector includes agriculture, forestry, animal husbandry, and fishery. The secondary industry includes mining, manufacturing, power, heat, gas and water production and supply, and construction. The tertiary industry, namely the service industry, refers to all industries excluding the primary industry and the secondary industry. The secondary industry is more related to energy and fuel consumption, and it thus dominates the anthropogenic NO₂ emission. Figure S10 shows the time series of GDP over the YRD from 2005–2020, and Fig. S11 is the same as Fig. S10 but for a year-to-year increment, i.e., the increase in GDP for a given year relative to its previous year. The results show that the GDP of each province within the YRD increased over time starting from 2005, but the relative contribution of each industry sector is different from year to year. The primary-sector-related GDP is relatively constant, but both the secondary-sector- and tertiary-sector-related GDPs show significant increasing trends from 2005–2020.

During 2009–2011, the GDPs increased significantly by CNY 198.45, CNY 483.86, CNY 656.40, and CNY 327.05 billion over Shanghai, Zhejiang, Jiangsu, and Anhui, where the secondary sector contributions account for 46.50 %, 53.64 %, 48.99 %, and 60.34 % respectively. Before 2011, much of China's economic growth still relied on the high-carbon fossil energy system, and efforts to control atmospheric pollution were relatively small. These significant increases in GDP could cause significant increases in anthropogenic NO₂ emission. After 2011, China implemented a series of clean air measures to tackle air pollution across China. These measures include the reduction of industrial pollutant emissions, the adjustment of industrial structure and energy mix, and other compulsory policies (China State Council, 2013). Zheng et al. (2018a) estimated China's anthropogenic emission trends from 2010–2017 with the bottom-up emission inventory and found that, as a consequence of the clean air measures, anthropogenic NO_x emission across China during 2010–2017 decreased by 17 %. In Fig. S12, we further analyzed the variabilities of NO_x emissions over the YRD region from 2008–2017 using the categories provided by the Multi-resolution Emission Inventory for China (MEIC) inventory, including motor vehicle emissions, major industrial emissions, resident emissions, and power emissions (http://meicmodel.org, last access: 25 February 2022) (Li et al., 2017; Zheng et al., 2018a). The results show that the decreases in Tro_NO₂ over the YRD during 2011–2013 are attributed to the reductions of industrial and power emissions, during 2013–2014 are mainly attributed to the reductions of motor vehicle emissions and power emissions, and after 2014 are attributed to the reductions of motor vehicle emissions, power emissions, and industrial emissions.

Although the total GDPs over all megacities still increased over time after 2011, most of these increases are from the tertiary sector, indicating the effectiveness of the adjustment of industrial structure and energy mix. The largest anthropogenic NO₂ producer from the tertiary sector is attributed to the transportation industry, such as traffic and cargo transport. The Chinese government implemented stringent restrictions on vehicle exhaust emissions after 2011 (Ministry of Ecology and Environment of the People's Republic of China, 2011, 2016). For example, they implemented the fourth and the fifth national motor vehicle pollutant emissions standards in 2011 and 2018, respectively, which mandate 30 % and 60 % reductions in vehicle NO_x emissions relative to the third national standard (Ministry of Ecology and Environment of the People's Republic of China, 2007, 2018). These stringent measures could significantly reduce anthropogenic NO₂ emissions from the tertiary sector. Overall, the decreasing trends in NO₂ VCD_trop from 2011–2020 over all megacities within the YRD are mainly attributed to the stringent clean air measures in this period, which either adjust high-energy industrial structure toward low-energy industrial structure or directly reduce pollutant emissions from different industrial sectors.

5 Conclusions

In this study, we quantified the long-term variabilities and the underlying drivers of NO₂ VCD_trop from 2005–2020 over the Yangtze River Delta (YRD) using the OMI LV3 NO₂ data product and MLR. The major pollution areas for NO₂ VCD_trop over the YRD are located in the south of Jiangsu Province and north of Zhejiang Province. In addition, NO₂ pollution in the eastern Anhui Province showed an increasing trend during 2005–2013 and became one of the major pollution areas within the YRD during 2010–2013. The amplitudes of NO₂ VCD_trop over the YRD showed large year-to-year variabilities from 2005–2020, but spatial extensions of the major pollution areas are almost constant over the years. Among all the pollution areas, the heaviest pollution regions are uniformly located in the densely populated and highly industrialized megacities such as Shanghai, Nanjing, Suzhou, Hangzhou, Ningbo, and Hefei. For six megacities, the spaceborne tropospheric results have been compared to surface in situ data, yielding correlation coefficients between 0.8 and 0.9.

Clear seasonal features and inter-annual variabilities of NO₂ VCD_trop over the YRD region are observed. Overall, the amplitudes and 1σ SDs of NO₂ seasonal cycles in cold seasons are larger than those in warm seasons, and the inter-annual NO₂ variabilities at megacity level can be divided into two stages, i.e., an overall increasing stage during 2005–2011 and a decreasing stage during 2011–2020. We have used the MLR to quantify the drivers of NO₂ VCD_trop from 2005–2020 over all megacities within the YRD. The seasonal cycles of NO₂ VCD_trop over the YRD are mainly driven by meteorology (81.01 %–83.91 %), except in winter when anthropogenic emission contributions are also pronounced (16.09 %–18.99 %). The inter-annual variabilities of NO₂ VCD_trop are mainly driven by anthropogenic emission (69.18 %–81.34 %), except in a few years, such as 2018, which are partly attributed to meteorology anomalies (39.07 %–91.51 %).

The increasing trends in NO₂ VCD_trop from 2005–2011 over the YRD are mainly attributed to high energy consumption associated with rapid economic growth, which causes significant increases in anthropogenic NO₂ emission. The decreasing trends in NO₂ VCD_trop from 2011–2020 over the YRD are mainly attributed to the stringent clean air measures in this period, which either adjust high-energy industrial structure toward low-energy industrial structure or directly reduce pollutant emissions from different industrial sectors. This study does not only improve our knowledge with respect to long-term evolution of economic and social development, anthropogenic emission, and the effectiveness of pollution control measures over the YRD, but it also has positive implications for forming future clean air policies in this key region.