Atmospheric Impacts of COVID-19 on NO_x and VOC Levels over China Based on TROPOMI and IASI Satellite Data and Modeling

Abstract

China was the first country to undergo large-scale lockdowns in response to the pandemic in early 2020 and a progressive return to normalization after April 2020. Spaceborne observations of atmospheric nitrogen dioxide (NO₂) and oxygenated volatile organic compounds (OVOCs), including formaldehyde (HCHO), glyoxal (CHOCHO), and peroxyacetyl nitrate (PAN), reveal important changes over China in 2020, relative to 2019, in response to the pandemic-induced shutdown and the subsequent drop in pollutant emissions. In February, at the peak of the shutdown, the observed declines in OVOC levels were generally weaker (less than 20%) compared to the observed NO₂ reductions (−40%). In May 2020, the observations reveal moderate decreases in NO₂ (−15%) and PAN (−21%), small changes in CHOCHO (−3%) and HCHO (6%). Model simulations using the regional model MAGRITTEv1.1 with anthropogenic emissions accounting for the reductions due to the pandemic explain to a large extent the observed changes in lockdown-affected regions. The model results suggest that meteorological variability accounts for a minor but non-negligible part (~−5%) of the observed changes for NO₂, whereas it is negligible for CHOCHO but plays a more substantial role for HCHO and PAN, especially in May. The interannual variability of biogenic and biomass burning emissions also contribute to the observed variations, explaining e.g., the important column increases of NO₂ and OVOCs in February 2020, relative to 2019. These changes are well captured by the model simulations.

1. Introduction

The impact of the COVID-19 disruptions on the atmospheric composition is studied by a rapidly growing number of publications, dominated by air quality studies over China [1,2], as this country was the first to undergo large-scale stringent lockdowns starting at the end of January 2020 and because their effects are expected to be very strong due to the high pollution levels. The reduction of social exchanges and economic activities led to sharp changes in air pollution, reported by in situ and satellite observations [3,4,5,6,7,8,9,10,11], and to an estimated reduction of 20–30% in primary pollutant emissions in February in eastern China based on activity data [12]. These widespread disturbances caused cascading changes in secondary pollutants through chemical interactions and feedbacks [13,14]. To disentangle the effects of long-term emission trends, meteorology and chemistry on the observed changes due to the COVID-19 lockdown in China, model studies usually target February and March 2020, when the Chinese economy was most affected, with a particular focus on observations of nitrogen dioxide (NO₂), ozone, and particulate matter [7,10,13,15,16,17]. However, spaceborne observations of other secondary pollutants such as volatile organic compounds (VOCs), are important to complement the analysis regarding the response of the atmospheric system to the pandemic-induced disruptions.

Here we use satellite column densities of four trace gases, NO₂, formaldehyde (HCHO), glyoxal (CHOCHO), and peroxyacetyl nitrate (PAN) in February and May 2019 and 2020 and analyze their distributions against simulations of the regional chemistry/transport model MAGRITTEv1.1 (Model of Atmospheric Composition at Global and Regional scales using Inversion Techniques for Trace gas Emissions, see Section 2.2). The vertical distribution of these compounds presents a pronounced maximum in the boundary layer [18,19] and their relatively short chemical lifetimes make them useful proxies for the emissions of NO_x and VOCs [20,21,22]. The first three compounds are measured by the Tropospheric Monitoring Instrument (TROPOMI, [23]) on board the Sentinel-5 Precursor platform, while PAN is retrieved from the observations of the Infrared Atmospheric Sounding Interferometer (IASI, [24]), carried on the Metop platforms. In addition to February 2020, discussed in many previous studies, we focused also on May 2020, to determine to what extent the lift of the shutdown was accompanied by a return to normality in terms of air composition.

HCHO and CHOCHO are short-lived intermediates in the oxidation of biogenic, pyrogenic, and anthropogenic non-methane VOCs (NMVOCs), and therefore their distributions depend strongly on the NMVOC emissions. Spaceborne HCHO has been frequently used to constrain the emissions of NMVOCs [25,26,27]. During the pandemic, TROPOMI HCHO data showed declining columns in the North China Plain, relative to the same period in 2019 [28]. PAN is formed via oxidation of NMVOCs in the presence of nitrogen oxides (NO_x = NO + NO₂) and constitutes the main atmospheric reservoir for NO_x radicals [19]. Spaceborne CHOCHO and PAN columns during the pandemic are presented and analyzed here for the first time. We conducted simulations over China using the MAGRITTEv1.1 regional model [29] for February and May 2019 and 2020, using state-of-the-art emission datasets, which account for the declining primary pollutant emissions due to the pandemic. The resulting columns are evaluated against the aforementioned satellite data, and are further investigated with the aid of sensitivity scenarios.

2. Materials and Methods

2.1. Satellite Observations from TROPOMI and IASI

The satellite data used are briefly presented here and are thoroughly described in Appendix A, Appendix B, and Appendix C. For all products, we use monthly averages gridded at the model resolution (0.5° × 0.5°).

The NO₂ columns are obtained from TROPOMI [23], which has an overpass local time of ~1:40 p.m. and provides a quasi-global daily coverage at a resolution of 5.5 × 3.5 km². We use L2 data with a quality value higher than 0.75 [30]. Since validation studies have shown underestimations of high columns and overestimation of low columns, we apply a bias correction, i.e., the TROPOMI columns (Ω) are adjusted using Ω′ = (2.12 × Ω −2.12 × 10¹⁵) where Ω and the adjusted column (Ω′) are expressed in molec.cm⁻², based on the linear regression of TROPOMI against ground-based column observations [31].

The TROPOMI HCHO retrieval is described in [32]. To improve the accuracy of the data, we use a modified version of the operational product [33]. In this version, the background correction in the remote Pacific Ocean is updated every day. To avoid possible issues originating from auxiliary data, we use tropospheric columns not corrected for clouds, in combination with a strict cloud fraction filtering (less than 20%) in addition to the usual quality assurance criterion (values higher than 0.5 are retained). The column averaging kernel and the a priori profiles are provided for every observation. In remote regions, the precision of an individual column is estimated at ~5 × 10¹⁵ molec.cm⁻², but for monthly averaged data at 0.5° × 0.5° it is estimated to be lower than 1.0 × 10¹⁵ molec.cm⁻². The systematic uncertainty of the monthly tropospheric column varies between 20% and 50% of the column. The detection limit is estimated as three times the retrieved random component of the uncertainty, multiplied by 2 to account for the underestimation of random errors suggested by comparison with ground-based data [34]. Upon spatiotemporal averaging, the detection limit scales as $1/\sqrt{n}$, with n the number of observations per pixel and per month.

The TROPOMI CHOCHO retrieval is described in [35]. The product includes vertical tropospheric columns derived in the visible spectral range using a stringent cloud fraction filtering (less than 20%), random and systematic errors, and column averaging kernels. The low CHOCHO optical depth implies large random errors (6–10 × 10¹⁴ molec.cm⁻² for dark scenes), which are significantly reduced through spatiotemporal averaging. Systematic errors are estimated at 1–3 × 10¹⁴ molec.cm⁻², corresponding to ca. 30–70% of a typical column in emission regions. Detection limits are calculated as for HCHO, based on the retrieved random uncertainty of the retrieval, but without the factor 2 upscaling.

The PAN total columns are obtained from hyperspectral radiance measurements recorded by the IASI nadir-viewing satellite [24]. IASI operates continuously since 2007 and achieves global coverage twice daily, at 9:30 and 21:30 local time. The retrieval relies on an artificial neural network framework specifically developed for weak infrared absorbers such as ammonia [36,37] and VOCs [38,39,40]. The typical uncertainty on the individual PAN column usually ranges between 1 × 10¹⁵ and 3 × 10¹⁵ molec.cm⁻², and can reach 4 × 10¹⁵ molec.cm⁻² for scenes affected by residual clouds or weak thermal contrast [38]. Single-cell measurements are considered robust when the hyperspectral radiance index (HRI) is higher than 3, which for PAN over Eastern China in February and May corresponds to a median value of 8.9 × 10¹⁵ molec.cm⁻². As for HCHO and CHOCHO, this value is lowered upon spatiotemporal averaging.

2.2. Simulations with the MAGRITTEv1.1 Chemical Transport Model

MAGRITTEv1.1 [29] is run at 0.5° × 0.5° over China and surrounding areas (73–150° E, 17–54° N) for February and May 2019 and 2020. Meteorological fields are from the ERA5 reanalysis [41]. Biomass burning fluxes are from the GFED4s database ([42], https://www.geo.vu.nl/~gwerf/GFED/GFED4, accessed on 21 July 2021), and biogenic emissions from the MEGAN-MOHYCAN model [43,44]. Baseline anthropogenic emissions are provided by the CAMS-GLOB-ANT_v4.2-R1.1 inventory [45,46], which provides emissions from 2000 to 2020. This inventory uses the MEIC v1.3 (http://meicmodel.org, accessed on 21 July 2021) emissions over China between 2008 and 2016. The emissions are extended to 2020 by linear extrapolation based on the emission trends proposed by [47], which account for recent emission decreases. The anthropogenic emissions during the pandemic are obtained from the CONFORM (COVID adjustmeNt Factor fOR eMissions) global dataset, which accounts for the slowdown of human activities through adjustment factors based on activity data, which are applied for all economic sectors and geographical regions [12]. For China, on average, the adjustment factors for the main sectors, i.e., road transport, power generation, industry, and residential are estimated at 0.4, 0.6, 0.65, and 1.1, respectively, in February 2020, and at 1.0, 0.9, 0.8, and 1.0 in May 2020 [12]. The factor for air traffic emissions is estimated at 0.44 in February 2020 and 0.67 in May 2020. Figure 1 illustrates the emission ratio (E₂₀₂₀/E₂₀₁₉) for anthropogenic NO_x and VOC fluxes in February and May, as well as the difference in biomass burning VOC and isoprene fluxes between 2020 and 2019. Besides the average adjustment factors, the CONFORM dataset provides low and high estimates of those factors accounting for uncertainties due to lack of data or limited information for some activity sectors (Figure 2).

The simulated monthly averaged columns at the satellite overpass time (13:30 local time for TROPOMI) are calculated by accounting for the averaging kernels and sampling times of valid data. The performed simulations are described in

Table 1. Description of the model simulations. All simulations use the same emissions for 2019 but differ regarding their 2020 emissions.

Short Name	Description
R1	Use average estimates of CONFORM adjustment factors for anthropogenic 2020 emissions [12]
R1H	Use high estimates of CONFORM adjustment factors. The resulting anthropogenic fluxes for 2020 are higher than in R1.
R1L	Use low estimates of CONFORM adjustment factors. The resulting anthropogenic fluxes for 2020 are lower than in R1.
R2	Use 2020 baseline anthropogenic emissions from CAMS-GLOB-ANT_v4.2-R1.1. These emissions do not account for pandemic-induced disruptions.
R3	Use the same (2019) anthropogenic NOx fluxes in 2019 and 2020
R4	Use the same (2019) anthropogenic VOC fluxes in 2019 and 2020
R5	Use the same (2019) anthropogenic and natural (biomass burning, biogenic) fluxes in 2019 and 2020

.

3. Results

3.1. February 2019 and 2020: Simulated and Observed Changes

Figure 3a,b illustrates the ratio of NO₂ columns in February 2020 to those of February 2019, observed by TROPOMI and simulated by MAGRITTEv1.1 (run R1) using the adjustment factors from the CONFORM dataset. The modeled and observed ratios show a good degree of consistency, both in terms of spatial patterns and percentage changes. Over eastern China, defined as 22–42° N and 108–125° E (Figure 4), the column decrease estimated in the R1 experiment (−42%) is in excellent agreement with the observed decrease (−40%), whereas the simulations R1H and R1L, using high and low CONFORM factors (Figure 2) result in underestimated (−35%) or overestimated (−50%) changes, respectively (

Table 2. Percentage changes of monthly columns between 2020 and the same month in 2019 ((2020-2019)/2019), based on observed and modelled columns from simulations R1 (CONFORM emissions), R1H and R1L (high and low CONFORM emissions), R2 (baseline emissions), R3 (fixed anthropogenic NOx flux), R4 (fixed anthropogenic VOC flux), and R5 (fixed anthropogenic and natural fluxes). All values are calculated for the eastern China region shown in Figure 4 delimited by 22–42° N, 108–125° E.

NO2 Changes
	TROPOMI	R1	R1H	R1L	R2	R3	R4	R5
February	−39.7	−41.9	−35.2	−49.5	−8.4	−2.6	−43.1	−5.2
May	−15.4	−12.3	−5.7	−14.4	−4.4	−2.3	−12.7	−3.2
HCHO Changes
	TROPOMI	R1	R1H	R1L	R2	R3	R4	R5
February	−8.6	−13.7	−11.3	−16.3	−5.7	−10.8	−8.3	−6.3
May	6.0	4.5	6.8	3.1	7.8	5.7	6.9	4.2
CHOCHO Changes
	TROPOMI	R1	R1H	R1L	R2	R3	R4	R5
February	−13.2	−20.7	−14.7	−27.0	0.2	−21.9	2.4	0.5
May	−3.2	−8.4	−0.9	−14.9	−2.1	−9.4	3.6	−1.3
PAN Changes
	IASI	R1	R1H	R1L	R2	R3	R4	R5
February	−17.9	−11.5	−7.6	−15.0	0.6	−7.9	−3.4	0.6
May	−21.2	−19.5	−13.2	−23.0	−11.5	−16.5	−14.2	−9.9

). These declines respond approximately linearly to the anthropogenic NO_x flux decreases of the CONFORM dataset in the R1 simulation (−41%), as well as in the R1H (−33%) and R1L (−51%) simulations. The strongest column reductions are observed in the densely populated North China Plain (NCP, -46%, Figure 4) and in the Hubei-Hunan region (HH, −44%, Figure 4). The latter region encompasses the Hubei province where some of the most severe lockdowns were enforced, and Wuhan, the city where the pandemic was first detected. The R1 experiment predicts very similar strong decreases over these regions (−47% and −45%, respectively), although their spatial patterns show less variability than TROPOMI (Figure 3). Over the Yangtze River Delta (YRD), the modeled change is slightly stronger than observed (−44% in R1 vs. −38% in TROPOMI).

The baseline R2 scenario predicts a column decrease of −8.4%, stronger than the NO_x emission decline (−2% in 2020 with respect to 2019) of the CAMS-GLOB-ANT_v4.2-R1.1 dataset. The R3 (same anthropogenic NO_x flux for 2019 and 2020) and R5 (same emissions in 2019 and 2020) simulations lead to small NO₂ column changes (−3% and −5%, respectively).

We further compared the NO₂ percentage changes between 2019 and 2020 estimated in the standard simulation (R1) and observed by in situ NO₂ data obtained at environmental monitoring stations from the China National Environmental Monitoring Center ([48,49] Figure A1).

Table 3. Percentage changes between 2019 and 2020 in: (i) NO₂ surface concentrations modeled by the standard simulation (R1) and observed at network in situ measurement stations (Figure A1), and (ii) NO₂ columns modeled by R1 and observed by TROPOMI at the locations of the in situ observations. All values are calculated for the regions shown in Figure 4. The number of measurement stations is indicated for each region between brackets.

Region (Number of In Situ Sites)	In Situ vs. R1				TROPOMI Columns vs. R1 (At In Situ Stations)
	February		May		February		May
	In Situ	R1	In Situ	R1	Sat	R1	Sat	R1
Eastern China (1035)	−36	−40	−8	−8	−45	−46	−19	−18
North China Plain (230)	−36	−45	−5	−2	−46	−48	−17	−19
Yangtze River Delta (154)	−41	−39	−2	2	−45	−47	−29	−25
Pearl River Delta (94)	−31	−23	−24	−20	−8	−33	−24	−15
Hubei-Hunan (150)	−39	−46	−18	−8	−48	−49	−11	−11

summarizes this evaluation of surface NO₂ changes and also presents comparisons between TROPOMI and model-simulated NO₂ columns at the locations of the in situ stations. For meaningful comparisons, we consider only in situ data within a two-hour time-window around the TROPOMI overpass time (12 a.m.–2 p.m. local time). The relative decreases in the observed and modeled surface NO₂ concentrations are in good agreement both in February and in May over all regions considered in Figure 4 (less than 10% difference on average). Similarly, the changes in the observed and modeled columns at the locations of the stations show good consistency. With regards to February and May 2019, the in situ data suggest a change of −36% in February and −8% in May 2020 in eastern China (1035 stations), which is well reproduced by the model (−40% in February and −8% in May). Over the same region, the satellite data at the locations of the in situ stations exhibit a change of −45% in February and −19% in May 2020, in excellent agreement with the changes calculated by the model (R1 simulation, Table 3). Similar results are found for the smaller regions of Figure 4, although with more pronounced differences related to the smaller number of measurement sites.

In contrast to NO_x, which have strong direct emissions, HCHO, CHOCHO, and PAN are predominantly secondary in origin. In Northern China, their wintertime columns are lower than in summer, due to reduced photochemical activity and biogenic NMVOC emissions [19,21,22,38]. The low winter columns explain the noisier and more uncertain columns in winter. Despite these limitations, a good agreement between observed and model distributions of the 2020-to-2019 column ratios is found in February (Figure 5). As for NO₂, however, the spatial variability of the changes is much lower in the model than in the observations (Figure 5). Very high and very low values of the satellite-based ratio are often related to columns being near or below the detection limit of the retrievals, especially for glyoxal (in south/central China) and PAN (in western and northeastern China).

The observed HCHO and CHOCHO ratio distributions are relatively similar, with lower-than-1 values in central-eastern China (between ~28° and 38° N), and values higher than 1 in the southern part (<25° N) as well as in northern China (between 40° and 43° N). This pattern is reproduced by the model, with HCHO and CHOCHO column decreases reaching respectively 14% and 21% on average in eastern China (17% and 24% in the YRD region). These declines are largely due to the drop in anthropogenic VOC emissions between 2020 and 2019. The stronger column reduction for CHOCHO than for HCHO is explained by the larger contribution of anthropogenic sources to the CHOCHO budget [50]. The HCHO column changes simulated here are consistent with those reported by another model study focusing on the North China Plain and a different period (between 1–23 January 2020 and 24 January–15 February 2020). In both studies, the spatial patterns are more homogeneous in the models than in the observations (Figure 5 and [13]).

Due to secondary formation, both HCHO and CHOCHO are observed away from the emission source, a well pattern represented in the model (Figure 5). The higher biomass burning emission in 2020 over Myanmar (18–23° N, 96–102° E, Figure 1c) led to modelled HCHO and CHOCHO columns enhancements by 6% and 18%, respectively, in qualitative agreement with the observed enhancements (15% for HCHO and 34% for CHOCHO) but suggesting an underestimation of fire activity. The data also suggest an underestimation of biomass burning emissions over northern Vietnam, as well as an underestimation of CHOCHO formation in agricultural fires, which are a commonplace practice in this region for clearing the fields after harvesting [51]. In eastern China, the use of the CONFORM emissions in R1 leads to a decrease of 8% and to 21% in the HCHO and CHOCHO column ratio, respectively, in comparison with the R2 experiment using baseline emissions for 2020. The stronger impact of emission reduction on CHOCHO than on HCHO is due to the larger contribution of anthropogenic VOCs to the abundance of CHOCHO (20–25% at global scale, [22,52] compared to HCHO (7%, [21]). Similarly, the observations show a pronounced north–south gradient in the PAN column ratios, which is qualitatively well represented in the model (Figure 5), although the observed changes are larger (Table 2). In consistency with the model, the largest decrease is found in the NCP, in response to the anthropogenic emission decreases. The higher column ratios in southern China might be due to a combination of higher biomass burning over Myanmar and higher isoprene emission fluxes over Southeastern China in 2020 (Figure 1d). Isoprene is a major precursor of PAN, responsible for ~37% of its formation on the global scale [19]. While the enhanced fires in Myanmar result in higher modeled PAN columns in February 2020 in this region, this enhancement is overestimated by the model over South China, and the impact of fires is not visible over the source region according to IASI, suggesting that fire events might actually play little role in the PAN formation [19].

3.2. May 2019 and 2020: Simulated and Observed Changes

After the sharp decline in economic activity during the first three months of 2020, China’s economy recovered in the following months, reaching almost normal levels by mid-May, as suggested by numerous indicators, like traffic density, energy consumption and business reopening [12,53]). By the beginning of May 2020, the resumption levels were estimated at ~90%, marking a swift normalization of economic activity, even though some sectors restarted somewhat later, e.g., the services sector [53]. According to the average CONFORM estimate for May 2020, the anthropogenic NO_x and VOC fluxes over China were respectively about 15% and 13.5% lower than in May 2019 (Figure 1), while the high (low) CONFORM estimate suggests decreases of 7% (18%) for NO_x and 5% (20%) for VOCs.

The observed NO₂ column changes and the R1 simulation are in good overall agreement in May (−15% vs. −12% on average, Table 2) even though the observed spatial distribution of the changes is more heterogeneous (Figure 3c,d). Whereas moderate reductions are found in Hubei-Hunan (−7% and −4% according to TROPOMI and MAGRITTE), much higher differences are seen in YRD (−19% and −23%) and NCP (−18% and −14%). Those inter-regional differences are mostly due to meteorological variability, the spatial distribution of the changes being very similar in the R1 and in the R5 simulation (not shown). For example, the 2020/2019 ratio of average columns in the R5 run is positive in Hubei-Hunan (+6%) and negative in both NCP and YRD (−8%). On average for eastern China, the change in NO₂ column due to the COVID-19 restrictions is estimated at about −8% in May (difference between R1 and R2 changes).

The observed distributions of HCHO, CHOCHO, and PAN columns in May 2020 and the corresponding model results are shown in Figure 6, and the ratios of the 2020 columns to those of 2019 are depicted in Figure 7. The comparisons for HCHO columns (Figure 6c,d) show a good agreement in terms of spatial distribution, but also a model overprediction by 20–40% over emission regions. This is in line with the low bias of about 30% of TROPOMI HCHO columns against ground-based FTIR measurements for high HCHO levels (>8 × 10¹⁵ molec.cm⁻²) [34]. Accounting for this data bias would have improved the agreement with the model, but would have had little influence on the difference between the two years.

The observed and modeled HCHO column ratios are slightly higher than unity (Figure 7a,b), except over Yunnan and northern Myanmar, where the low pyrogenic and biogenic VOC emissions in 2020 lead to significant HCHO reductions. Over eastern China, the increase is estimated at 5% on average by R1, in excellent match with the observation (6%, Table 2) although TROPOMI displays a higher variability of column ratios. The observed spatial heterogeneity is not well represented in the model in May, a feature that has been found also in February and also reported in [13]. The column increase derived in spite of the lower 2020 anthropogenic VOC emissions can be attributed to meteorology, as shown by the R5 simulation results displaying a 4% increase in the columns. The estimated change in the R2 scenario using baseline emissions is 8%, suggesting that the impact of the emissions decline due to the pandemic is about −3% for this month. Besides those increases, HCHO decreases are observed in Yunnan and in northern Indochina (18–28° N, 92–104° E), in good agreement with model results (Figure 7a,b). In this region, anthropogenic VOC fluxes are low and HCHO is chiefly formed from biogenic VOC oxidation and fire events. Therefore, the observed column differences stem mostly from differences in these source categories. The emission inventories used in MAGRITTEv1.1 indicate lower biogenic and fire fluxes in May 2020 than in May 2019 in this region, by 15% and 38%, respectively. The negative change calculated by the model (−15%) is weaker than in the observations (−24%), suggesting underestimated differences in biogenic and/or fire fluxes between the two years. This conclusion is comforted by comparison between observed and modeled ratios for CHOCHO over the same region (Figure 7c,d).

The TROPOMI CHOCHO observations in May 2020 are fairly well reproduced by the R1 results over eastern China (Figure 6e,f). Hotspots are found over densely populated megacities and industrial hubs, like the Yangtze River Delta, the Chengdu-Chongqing urban cluster, and the Pearl River Delta, primarily due to anthropogenic VOC precursors, in particular aromatic compounds used in solvents and other industrial processes [54]. Over eastern China, the model does not match the spatial patterns of the observed column ratios (Figure 7c,d). The simulated CHOCHO decreases in central and northern China, due to anthropogenic emission reductions, and the column increases in southern China, due to higher isoprene flux in 2020 (Figure 1d), are not seen in the satellite data (Figure 7c,d). An overestimation of isoprene fluxes in 2020 relative to 2019 in the model was also suggested by HCHO in that area (Figure 7a,b).

IASI detects high PAN levels of about 6–8 × 10¹⁵ molec.cm⁻² over eastern China in May 2020, in good consistency with the model results over land (Figure 6g,h). Note that because of the weak, broadband spectral signatures of PAN and of the lower thermal contrast over sea, oceanic retrievals are more uncertain. The PAN distribution suggests a large anthropogenic contribution, consistent with the strong anthropogenic signals in the distribution of NO₂ and HCHO. The model succeeds in reproducing the observed high PAN levels in Northern China and in the Chengdu-Chongqing area, where human activities are dominant. The observed enhancements over Assam (India) and Myanmar are due to biogenic emissions (over Myanmar) and anthropogenic emissions (over India) (Figure 6g). As shown in Figure 7e,f, the observed and modeled PAN ratios are in fair agreement in both distribution and magnitude. The relative change over eastern China according to IASI (−21%) and the model (−19%) are very similar. According to model results, about half of the PAN decrease is due to meteorological variability (Table 2, runs R1 and R5), whereas the anthropogenic NO_x and VOC reductions led to PAN column declines of about 3% and 5%, respectively (runs R3 and R4 in comparison with R1).

Finally, sensitivity simulations suggest that the reduction in anthropogenic VOC fluxes in 2020 relative to 2019 leads to column reductions of 5% or less for HCHO (difference between R4 and R1 estimates, Table 2), between 12% (May) and 23% (February) for CHOCHO and 5–8% for PAN. The larger changes for CHOCHO are expected because of the strong contribution of anthropogenic VOC oxidation to the CHOCHO budget over China [50,55]. The NO_x level decline due to the pandemic also contributes to the reduction in the columns of HCHO (by 1–3%) and PAN (3–4%), whereas their impact on CHOCHO appears negligible (Table 2, columns R1 and R3).

4. Discussion

When using the CONFORM database, and in agreement with the observations, the simulated combined effect of meteorology and emission reductions is strongest for NO₂ between February 2020 and 2019 (−40% on average over Eastern China), especially in the NCP. The magnitude of this change compares well with recent modelling studies, e.g., the investigation of surface pollutants changes using the CAM-Chem model by [14] and the regression model study based on OMI observations by [7]. Meteorology explains a small part of this change (−5%, run R5), due to warmer and moister conditions in 2020 [6] leading to a shorter chemical NO_x lifetime. The major contribution to the NO₂ reduction originates from the pandemic-induced emission decreases (−34%, difference between R2 and R1 in Table 2). The reduction of VOC emissions is found to have a small offsetting effect (+1.2%, difference between R4 and R1 in Table 2) due to the decreased sink through organic nitrate and peroxyacetyl nitrate formation [13]. Similar results are found in May, with the effect of the reduced emissions explaining about 70% of the column decrease.

In the case of HCHO, the effects of the lockdowns are much smaller than for NO₂, and are estimated at −8% in February and at −3% in May (Table 2, difference between R2 and R1). Meteorology has a significant impact, −6% in February. [28] also reported meteorological impacts (−6% in the NCP) being of the same order as those of emissions, although these conclusions are not strictly comparable to ours since their study was based on January–April averages. In the Beijing area, HCHO columns were higher in February 2020 than in 2019, which is well simulated by R1 (Figure 5). The authors of [28] attribute this increase to enhanced OH levels as a result of NO_x emission decreases. Although boundary layer OH levels were indeed higher in 2020 (>50% compared to 2019) in the R1 run, this effect explains only a small part of the HCHO enhancement in February, as shown by the very similar distribution of the column ratios calculated in the R5 simulation adopting identical emissions in 2019 and 2020 (not shown).

The impact of the emission decreases is more significant for CHOCHO than for HCHO columns. Based on the R1 and R2 simulations, we estimate the column decrease over eastern China at 21% and 6% for February and May, respectively. The model overestimates the column decrease between the two years, which might indicate an overestimation of the anthropogenic contribution to the CHOCHO budget, or issues in the chemical mechanism for glyoxal formation adopted in the model for aromatic compounds.

Finally, based on the sensitivity simulations (Table 2), the lockdowns in 2020 have led to PAN column decreases of −12% and −8% in February and May, respectively, while these changes become smaller when anthropogenic VOC fluxes are kept fixed for both years (−8% in February and −5% in May), and are even smaller when the anthropogenic NO_x emissions are fixed (−3%). This underscores the importance of anthropogenic VOCs in the PAN budget according to the model in this region, and is to a large degree corroborated by the changes in satellite observations between the two years.

5. Conclusions

Spaceborne data of NO₂ and oxygenated VOCs complemented by MAGRITTEv1.1 simulation were used to investigate atmospheric composition changes in China during the stringent lockdown of February 2020 and during the recovery phase in May 2020. The lockdown-related emission reductions from the CONFORM dataset were implemented in these simulations. Relative to 2019, the model succeeds in reproducing the observed NO₂ column changes of −40% in February and −15% in May over eastern China, and goes a long way towards explaining the observed changes of oxygenated VOCs including HCHO, CHOCHO, and PAN. For those species, the observed changes are generally weaker than for NO₂. Besides the response to changes in anthropogenic fluxes, the oxidized VOCs are also affected by biomass burning and biogenic sources. The observed column ratios between 2020 and 2019 reflect primarily the differences in these sources between the two years. In an average sense, these differences are well captured by the model. The observations show a larger variability of column changes, likely due in part to model issues (e.g., due to errors in meteorological fields, in emissions, etc.) but also to noise and biases in the observations.