Significant changes in the chemical compositions and sources of PM2.5 in Wuhan since the city lockdown as COVID-19.

Wuhan was the first city to adopt the lockdown measures to prevent COVID-19 spreading, which improved the air quality accordingly. This study investigated the variations in chemical compositions, source contributions, and regional transport of fine particles (PM2.5) during January 23-February 22 of 2020, compared with the same period in 2019. The average mass concentration of PM2.5 decreased from 72.9 μg m-3 (2019) to 45.9 μg m-3 (2020), by 27.0 μg m-3. It was predominantly contributed by the emission reduction (92.0%), retrieved from a random forest tree approach. The main chemical species of PM2.5 all decreased with the reductions ranging from 0.85 μg m-3 (chloride) to 9.86 μg m-3 (nitrate) (p < 0.01). Positive matrix factorization model indicated that the mass contributions of seven PM2.5 sources all decreased. However, their contribution percentages varied from -11.0% (industrial processes) to 8.70% (secondary inorganic aerosol). Source contributions of PM2.5 transported from potential geographical regions showed reductions with mean values ranging from 0.22 to 4.36 μg m-3. However, increased contributions of firework burning, secondary inorganic aerosol, road dust, and vehicle emissions from transboundary transport were observed. This study highlighted the complex and nonlinear response of chemical compositions and sources of PM2.5 to air pollution control measures, suggesting the importance of regional-joint control.

Introduction

The air quality in China and other countries has improved a lot after the outbreak of the COVID-19 (Sharma et al., 2020; Tobías et al., 2020; Zhang et al., 2020a). For instance, after two weeks of lockdown, the urban air pollutant concentrations in Barcelona markedly decreased by 28– 51% (Tobías et al., 2020). Similarly, the air pollutant concentrations decreased by 18– 43% during the lockdown period compared to previous years in India (Sharma et al., 2020).

In China, the quarantine period began since 2020-01-23 and lasted until the end of February. Due to the restriction measures (Huang et al., 2020) and Spring Festival effects (Kong et al., 2015), many factories were shut down, the traffic volume on roads and construction activities reduced obviously, and the fireworks burning was banned especially in urban regions, etc. The substantial reductions of human activities have reduced the anthropogenic emissions and thus improved the air quality (Fig. S1). Despite obvious emission reductions due to COVID-19, air pollutions still occurred in the North China Plain due to the unfavorable meteorological conditions (Wang et al., 2020). Further research found that the haze events during the lockdown were driven by the enhancement of secondary pollution as the increase of atmospheric oxidizing ability (Huang et al., 2020). Till now, the variations of chemical compositions and local or regional-transported sources of PM2.5 were still unclear for the lockdown period.

Previous studies have revealed the role of pollution control measures on air quality improvement (Cai et al., 2017; Kong et al., 2018; Wang et al., 2019; Zheng et al., 2019a). Elemental carbon (EC), organic matter (OM), sulphate, nitrate, and ammonium (SNA) levels in autumn and winter decreased by >50% from 2013 to 2017 in Beijing, due to the implementation of the Air Pollution Prevention Control and Action Plan in China (Wang et al., 2019). Another factor influencing the variations in aerosol compositions and levels is meteorology, which impacts the transportation, dispersion, accumulation, and deposition of air pollutants (Whiteaker et al., 2002; Zhang et al., 2015; Bei et al., 2016). Generally, emission dominates the interannual variability of air pollutants (Chen et al., 2019; Zhai et al., 2019; Zhang et al., 2019a; Zheng et al., 2020), while meteorology dominates the day-to-day variations of them (Zhang et al., 2012; He et al., 2016, He et al., 2017). It was estimated that the anthropogenic emission reductions explained 91% of the PM2.5 decrease in China from 2013 to 2017 (Zhang et al., 2019a). Meteorology, however, explained >70% variances of daily average pollutant levels over China during 2014 and 2015 (He et al., 2017). The impacts of strictest lockdown measures and meteorology on PM2.5 levels, compositions, and sources were still unknown.

This study used the online monitoring datasets to investigate the impacts of lockdown and meteorology on the chemical components and sources of PM2.5 for two periods (one-month (01/23– 02/22) since the lockdown of Wuhan in 2020 and the same period in 2019). Positive matrix factorization (PMF) model was adopted to apportion PM2.5 sources. A random forest tree method was used to separate the contributions of emission reduction and meteorology to the changes of PM2.5 concentrations, chemical species, and sources. A backward trajectory-based method was used to study the impact of regional transport on PM2.5 in Wuhan.

Materials and methods

Observation datasets

Hourly observations of PM2.5 chemical compositions including water-soluble ions (WSIs), trace elements (TE), organic carbon (OC), and EC were conducted using on-line instruments during 2020/01/23 00:00 to 2020/02/22: 23:00 at an environmental monitoring supersite in Wuhan. The measurements for the same period in 2019 were also conducted and more details about the instruments were described in Text S1. The site (114.38° E, 30.52° N) is located in the urban center, which is a residential/commercial site, with no obvious industrial emissions at surroundings. Time series of the main chemical species for the two periods are shown in Fig. S2. Additionally, synchronous in-situ observations of six criteria air pollutants (CO, NO2, O3, PM10, PM2.5, and SO2) were conducted. CO, NO (NO = NO + NO2), O3, and SO2 were measured with a correlation infrared absorption analyzer (TAPI, model: 300E, USA), a chemiluminescence trace level NO-NO2-NO analyzer (Casella, model: ML9841B, UK), an UV photometric O3 analyzer (TEI, model: 49i, USA), and a pulsed UV fluorescence SO2 analyzer (Casella, model: ML9850B, UK), respectively. PM10 and PM2.5 were measured by an oscillating balance analyzer (TH-2000Z, China) with two separate inlets. Time series of the six criteria air pollutants for the two periods are shown in Fig. S3. In-situ meteorological parameters including ambient temperature (Temp), atmospheric pressure (P), wind speed (WS), wind direction (WD), and relative humidity (RH) are shown in Fig. S4. The mean and standard deviations of air pollutants, chemical compositions, and meteorological parameters for the two periods are listed in Table 1 .

Table 1

Mean and standard deviation (SD) of the six criteria air pollutants (μg m−3), main PM2.5 chemical species including water-soluble ions (μg m−3), trace elements (ng m−3), carbonaceous components (μg m−3), and meteorological parameters including ambient temperature (Temp, °C), atmospheric pressure (P, hPa), wind speed (WS, m s−1), relative humidity (RH, %), mixing layer height (MLH, m), and solar irradiation (SI, W m−2) for the one-month lockdown period of Wuhan in 2020 and the same period in 2019.

Variables	2019	2020	Variables	2019	2020
	Mean ± SD	Mean ± SD		Mean ± SD	Mean ± SD
CO	1.02 ± 0.37	0.98 ± 0.31	Fe	292 ± 235	153 ± 135
NO2	40.7 ± 28.6	18.1 ± 15.6	Cu	26.6 ± 28.3	11.1 ± 6.48
O3	29.7 ± 21.1	40.9 ± 18.0	Zn	90.2 ± 70.7	32.5 ± 24.2
PM10	76.0 ± 42.4	50.4 ± 28.0	As	11.8 ± 10.9	4.99 ± 4.87
PM2.5	72.9 ± 35.4	45.9 ± 26.9	Se	5.00 ± 3.16	2.36 ± 2.12
SO2	6.15 ± 6.10	4.52 ± 3.30	Ag	5.80 ± 2.54	3.62 ± 2.07
Na+	0.21 ± 0.17	0.24 ± 0.08	Cd	4.11 ± 2.12	4.47 ± 2.25
NH+4	14.4 ± 6.65	9.59 ± 6.06	Ba	78.7 ± 128	36.0 ± 55.7
Mg2+	0.21 ± 0.26	0.01 ± 0.01	Hg	1.95 ± 0.80	1.13 ± 0.87
K+	2.02 ± 2.13	1.22 ± 1.11	Pb	47.7 ± 36.8	17.3 ± 12.1
Ca2+	0.48 ± 0.39	0.08 ± 0.10	OC	10.4 ± 4.21	8.09 ± 3.52
Cl−	2.82 ± 1.85	1.96 ± 1.62	EC	2.19 ± 1.23	1.15 ± 0.70
NO-3	23.9 ± 12.3	14.1 ± 9.49	Temp	4.12 ± 3.51	8.97 ± 3.90
SO2–4	13.4 ± 6.40	10.3 ± 6.48	P	1023 ± 5.34	1019 ± 5.12
K	2038 ± 2026	1163 ± 1053	WS	2.08 ± 1.53	0.99 ± 0.79
Ca	173 ± 222	105 ± 104	RH	84.3 ± 15.1	73.2 ± 19.3
Cr	2.73 ± 2.44	1.28 ± 1.81	MLHa	318 ± 232	415 ± 381
Mn	23.3 ± 19.3	8.17 ± 5.78	SIa	126 ± 140	195 ± 180

MLH and SI were derived from the HYSPLIT calculation.

De-weather model

The variations of air pollutants are controlled by both meteorological conditions and emissions. To isolate the trend resulted from emissions, an R package “normalweatherr” was used to remove the influence of meteorology (Grange et al., 2018; Vu et al., 2019; Zhang et al., 2020b). The package used a boosted regression tree (RF) approach to model the air pollutant concentrations. The algorithm solved the relationship between air pollutant levels and their predictors including meteorological parameters and time variables such as day of the year (Julian day), day of the week (Monday to Sunday), and hour of a day (0– 23) (Grange et al., 2018). The input dataset was randomly divided into a training dataset (i.e., 70% of input dataset) for constructing the RF model and a testing dataset (30% of input dataset) for testing the performance of the RF model with an unseen dataset. After the building of the RF model, the weather normalized technique was used for predicting the air pollutant level at a certain measured time point (i.e., 2020/01/23 10:00) with randomly selected meteorological parameters for 1000 times. For each prediction, the independent variables (meteorological parameters and time variables except for the trend term) were sampled without replacement and were randomly allocated to a dependent variable observation. The meteorologically normalized trend at a specific observation time was then calculated as the arithmetic mean of 1000 predictions. More details about the model can be found elsewhere (https://github.com/skgrange/normalweatherr).

In this study, the RF model was independently applied to the datasets for the study period in 2019 and 2020, respectively. The six criteria air pollutants, main PM2.5 chemical species including chloride, sulphate, nitrate, ammonium, trace elements, OC, EC, and source contributions derived from the receptor model (see Section 2.3 and Section 3.3 for details) were meteorologically normalized. The predictors (independent variables) included meteorological parameters (ambient temperature, atmospheric pressure, wind speed, relative humidity, mixing layer height, and solar irradiation) and temporal variables such as an hour of a day, day of a week, and day of a year.

Source apportionment

Positive matrix factorization (PMF 5.0) has been widely used in PM2.5 source apportionment (Bressi et al., 2014; Kong et al., 2018; Zheng et al., 2019a; Zhang et al., 2019b). WSIs, OC, EC, and several trace elements were input into the model. Species were grouped into strong, weak, and bad according to their single-noise ratio and percentages below detection limits (Callén et al., 2014; Zheng et al., 2018). The species classification is listed in Table S1. Due to the instrument maintenance or incomplete chemical compositions of PM2.5, a small part of samples was excluded from the PMF analysis. Finally, a dataset of 728 (sample) × 21 (species) for 2019 and a dataset of 719 (sample) × 18 (species) for 2020 were input into the PMF model, respectively.

To find the optimal factor number, it was tested from 3 to 9. The diagnostic parameters such as the Pearson correlation coefficient, Q , and Q /Q were calculated to co-determine the optimal solution. Q /Q is an indicator of the optimal solution that if the factor number is properly estimated, it approaches 1. Q /Q is an indicator of the number of an outlier with scaled residuals larger than 4 and it is equal to 1 when there are no outliers. As shown in Table S2, the Pearson correlation coefficient between the observed and modelled concentrations increased with the factor number increasing, suggesting better fitting results as a large factor number. The Q /Q decreased with the increasing of factor number and it approached 1 for the 7-factor solution. To evaluate the uncertainty associated with random and rotational ambiguity, the bootstrap-displacement (BS-DISP) method was used. In this study, the percentage of cases accepted in BS-DISP decreased with the factor number increasing. The cases accepted in BS-DISP for 7-factor was higher than 6-factor and 8-factor solutions for the dataset in 2020, while it was lower than 6- and 8-factor solutions for the dataset in 2019. To better compare the source apportionment results, 7-factor was considered as the optimal solution. Corresponding results were reported in this study. More details about PMF running and result diagnosis can be found elsewhere (Bressi et al., 2014; Zheng et al., 2018) and in Text S2.

Potential geographical source regions

Hourly backward trajectories reaching the monitoring site were calculated using the HYSPLIT model (Stein et al., 2015) driven by the GDAS_1 reanalysis meteorological field. The start height was set as 100 m above the ground level. 72 h backward air mass trajectories were calculated. The air masses reaching the monitoring site were mainly from north and south directions in this study (Fig. S5). The potential geographical source regions were separated into North China (NC), South China (SC), and Local according to the following procedures: (1) the endpoint with a height >2500 m were excluded due to few impacts of surface emissions on air masses (Kanaya et al., 2016); (2) the endpoints of each trajectory were projected to one of the three regions; (3) the number of trajectory endpoints projected in above regions were counted; (4) the trajectory was classified into SC, NC, and Local in turn, starting with SC because the air masses transported from south direction were less compared to other three regions. If a trajectory spent over 10 h within SC, it was allocated to SC. Similarly, the trajectory was allocated to NC if it spent over 10 h within NC. The air mass was attributed to Local if it spent over 52 h within local areas. This method was used elsewhere (Liu et al., 2019a). The thresholds were somewhat arbitrarily set according to the sensitive analysis of the threshold values on air mass classification and the combination used here showed the best classification (Fig. S6).

Statistical analysis and data visualization

Analysis of variance (ANOVA) was used to test whether the differences of meteorological parameters, chemical compositions, and source contributions for the two periods were significant or not at the 95% confidence interval. Data analysis and visualization were realized by R language (R Core Team, 2018) with packages such as “ggplot2”, “openair”, “lubridate”, “reshap2”, and “normalweatherr”, etc. All these packages can be accessed via the Comprehensive R Archive Network (CRAN: https://cran.r-project.org2).

Results

Changes in air pollutants

The national mean change rates of CO, NO2, O3, PM10, PM2.5, and SO2 were −13.0%, −35.7%, 21.9%, −32.5%, −15.0%, and −20.1%, respectively (Fig. S1). The results suggested a substantial improvement of air quality in China during the lockdown period due to COVID-19. Most regions showed a decreasing in PM2.5 and an enhancement in O3. The reverse variations of PM2.5 and O3 were also reported elsewhere during the lockdown periods (Sharma et al., 2020; Tobías et al., 2020). Previous studies suggested that the O3 increase was related with its enhanced production rate, as the PM2.5 reduction impairing its role as a scavenger of hydroperoxy radicals (Li et al., 2019a, Li et al., 2019b). In this study, the change rates of NO2 (−55.6%), PM10 (−33.7%), PM2.5 (−37.0%), and SO2 (−26.5%) at the observational site were lower than the national averages of them, suggesting larger impacts of lockdown on air quality in Wuhan.

Variations in chemical compositions

Details about the mass concentrations of measured chemical species for the two periods are listed in Table 1. Fig. 1A and B shows the mass concentrations and percentages of main PM2.5 chemical species. SNA were the dominant species, totally accounting for 74.1% and 72.8% of the main chemical compositions of PM2.5 in 2019 and 2020, respectively. Compared to those in 2019, the mean mass concentrations of nitrate in 2020 decreased most (9.86 μg m−3), followed by ammonium (4.73 μg m−3), sulphate (2.90 μg m−3), OC (2.22 μg m−3), TE (1.17 μg m−3), EC (1.02 μg m−3), and chloride (0.85 μg m−3). The decrease of TE and EC verified the primary emission reduction due to the lockdown of Wuhan city.

Fig. 1

Differences in the main chemical species and source contributions of PM2.5 derived from PMF model for the two periods. A and B are the main PM2.5 chemical compositions. C and D are PMF source contribution results. BB, CC, FW, IP, RD, SIA, and VE represent biomass burning, coal combustion, firework burning, industrial processes, road dust, secondary inorganic aerosol, and vehicle emissions, respectively.

The PM2.5 chemical compositions associated with air masses from different potential geographical regions for the two periods are compared in Fig. S7. For chemical species associated with local air masses, the mean sulphate level in 2020 (14.0 ± 6.19 μg m−3) was higher than that in 2019 (12.3 ± 3.85 μg m−3) (p = 0.82), while the other species showed reductions compared to 2019. For air masses from NC, the OC and sulphate in 2020 were higher than those in 2019. For other species, they showed lower mass concentrations during the study period in 2020.

Variations in source contributions

Seven sources including biomass burning (BB), coal combustion (CC), firework burning (FW), industrial processes (IP), road dust (RD), secondary inorganic aerosol (SIA), and vehicle emissions (VE) were apportioned for both the two periods. Their source profiles and diurnal variations derived from the PMF model are shown in Fig. 2, Fig. 3 , respectively. Prior to further analysis, the similarity of the source profiles was checked by Pearson coefficients (r), index of agreement (IOA), and coefficient of efficiency (COE). As shown in Table S3, r, IOA, and COE varied in the ranges of 0.72– 0.98, 0.54– 0.85, and −0.01– 0.77, respectively. If the source profiles are similar, the r, IOA, and COE approach 1. According to this rule, the profiles of BB, CC, RD, SIA, and VE showed high similarity, while the FW and IP showed less similarity. The source identification results were detailedly described as follows.

Fig. 2

Source profiles of biomass burning (BB), coal combustion (CC), firework burning (FW), industrial processes (IP), road dust (RD), secondary inorganic aerosol (SIA), and vehicle emissions (VE) for 01/23 00:00– 02/22 23:00 in 2019 (left panel) and 2020 (right panel) derived from PMF model. The boxplots are constructed according to the bootstrap results (n = 100).

Fig. 3

Diurnal variations of seven PM2.5 sources derived from the PMF model during 01/23 0:00– 02/22 23:00 in 2019 (dashed line) and 2020 (solid line). The error bars represent the 95% confidence intervals of the mean.

Biomass burning

K+ is widely used as a tracer of biomass burning (Chen et al., 2017; Zhou et al., 2018; Zheng et al., 2019a). In addition to K+, abundant OC, EC, Na+, Cl−, OC, and EC are also reported for the source profile of biomass burning (Reid et al., 2005), although they vary according to fuel type and combustion efficiency (Bressi et al., 2014). In this study, high loadings of K+ (19.2 ± 6.52% in 2019 and 14.9 ± 15.6% in 2020, the same sequence for the following descriptions) and Cl− (59.5 ± 15.6% and 72.3 ± 16.1%) were found in this factor. The diurnal variation of this source peaked at noon and night, with lower values occurred at early morning and late afternoon (Fig. 3). Biomass burning was prohibited in urban areas and the source may be related to regional transportation from north China in 2019 and suburban regions of Wuhan in 2020, according to the evidences from conditional probability function (CPF) analysis in Fig. S8. Biomass burning was an important contributor to PM2.5 in north China in winter, with the contributions of 7– 12% (Zhang et al., 2013; Shang et al., 2018). In the suburban of Wuhan and its surrounding regions, biomass fuels are still adopted for heating and cooking. Outdoor biomass burning was frequently observed in the surrounding areas of Wuhan (Mehmood et al., 2020). Compared to the same period in 2019, the fire spot numbers detected by MODIS in 2020 reduced (Fig. S9). The contributions of biomass burning also decreased as shown in Fig. 1C.

Coal combustion

High loadings of As (86.2 ± 11.8% and 92.2 ± 8.11%), Se (47.7 ± 10.0% and 42.9 ± 7.51%), and Pb (36.9 ± 10.0% and 38.8 ± 4.96%) were found in this factor, which have been widely reported as the tracers of coal combustion (Tian et al., 2015; Liu et al., 2018; Zhou et al., 2018; Liu et al., 2019b). Additionally, this factor was positively correlated with SO2 (r = 0.65, p < 0.01 for 2019 and r = 0.52, p < 0.01 for 2020). Therefore, it was reasonable to attribute this factor to coal combustion.

Firework burning

Firework burning was characterized by high loadings on K (49.5 ± 4.63% and 44.3 ± 5.71%), K+ (49.6 ± 4.38% and 52.3 ± 10.5%), Ba (77.1 ± 6.73% and 68.6 ± 8.26%), and Cu (53.5 ± 6.20% and 39.5 ± 6.34%) (Kong et al., 2015; Rai et al., 2020). K+ was mainly from the furnish paper burning; Cu and Ba were additives for colourful flame (Kong et al., 2015). Higher contributions were found around the New Year's Eve and the Lantern Festival (Fig. S10), with the maximum contributions occurring at ~02:00. Firework burning is prohibited in the urban areas of Wuhan. The high contributions of this source occurred with wind speed higher than 3 m s−1 (Fig. S8). Therefore, this factor was attributed to firework burning and it was mainly associated with regional transportation.

Industrial processes

This factor was characterized by high loadings on Cr (43.9 ± 23.7% and 91.9 ± 23.1%), Mn (48.8 ± 16.8% and 21.3 ± 4.92%), Fe (33.5 ± 11.5% and 25.5 ± 6.02%), and Zn (27.9 ± 10.4% and 15.5 ± 2.75%) (Taiwo et al., 2014). PM2.5 emitted from iron and steel industry has high fractions of Fe and Mn (An et al., 2015). Cr and Mn are abundant in particles from stainless steel production and Zn is abundant for particles emitted from Zn metallurgy emissions (Querol et al., 2007). This factor can be attributed to local industrial processes in Wuhan. It was verified by the high CPF values under low wind speed (i.e., <1 m s−1) in east (2019) and south directions (2020) (Fig. S8), where locates plenty of industrial plants (Fig. S11). Two peak contributions at daytime and lower contributions at night of this factor in 2019 were also related to the daily production activities of industrial processes (Merico et al., 2020).

Road dust

This factor was highly loaded on crustal elements, like Ca (91.6 ± 8.21% and 94.9 ± 13.0%), Fe (29.5 ± 8.30%, and 37.6 ± 6.43%), and Mn (20.1 ± 12.4% and 27.7 ± 6.04%). A previous study conducted in Wuhan identified a similar source as fugitive dust due to high loadings on Ba, Ca, Fe, and Mn (Lyu et al., 2016). Fe and Mn are also abundant in brake wear and tyre wear dusts (Garg et al., 2000; Adachi and Tainosho, 2004). According to these markers, road dust was widely identified around the world (Richard et al., 2011; Pancras et al., 2013; Tao et al., 2014; Waked et al., 2014; Lyu et al., 2016; Liu et al., 2019b). Therefore, this factor was identified as road dust in this study.

Secondary inorganic aerosol

SO2 – 4 (50.6 ± 9.60% and 52.2 ± 10.8%), NO– 3 (47.3 ± 11.6% and 50.9 ± 15.0%), and NH+ 4 (46.8 ± 8.76% and 51.5 ± 10.3%) had higher loadings on this factor (Lyu et al., 2016; Zhang et al., 2013; Zheng et al., 2019a). Excepted for their primary emissions from coal combustion (Dai et al., 2018; Zhang et al., 2018; Yan et al., 2020), SIA can be formed from the atmospheric homogenous and heterogeneous reactions of their precursors (Seinfeld and Pandis, 2006). As shown in Fig. S8, high contributions of this factor were found for both low and high wind speeds, which favoured their formation locally or through regional transportation.

Vehicle emissions

This factor had higher loadings on OC (41.7 ± 12.1% and 45.7 ± 11.5%) and EC (37.5 ± 12.9% and 34.5 ± 6.21%) and moderate loadings on Cu (13.8 ± 3.79%, 23.0 ± 10.5%) and Pb (10.4 ± 3.59% and 19.6 ± 7.66%). A similar factor was resolved in previous studies, which was identified as vehicle emissions (Xia and Gao, 2011; Yao et al., 2016; Liu et al., 2018; Zheng et al., 2019a). Significant positive correlations (r = 0.39, p < 0.001 in 2019 and r = 0.35, p < 0.001 in 2020) between this factor and NO2 were found. Additionally, the diurnal variations showed morning and evening peaks in 2019, which reflected the daily variation of traffic activities. The diurnal pattern of this source in 2020 was not as clear as that in 2019, which should be related to the prohibition running of private cars during the city lockdown period (Fig. 3).

The mass concentrations and percentages of the seven sources are shown in Fig. 1C and D. The top three contributors to PM2.5 mass were SIA, BB, and VE. Industrial processes exhibited the highest reduction of mass concentration contribution (8.02 μg m−3), followed by BB (5.50 μg m−3), SIA (5.00 μg m−3), VE (2.99 μg m−3), CC (2.43 μg m−3), and FW (1.54 μg m−3). The percentage contributions of CC, IP, and FW also showed reductions of 3.10%, 11.0%, and 0.08%, respectively, while the percentage contribution of SIA increased by 8.70%. The reverse variations of these sources suggested the decreased primary emissions but enhanced secondary formation rate during the lockdown.

The diurnal variations of PM2.5 source contributions showed differences for the two periods (Fig. 3). For industrial processes, it showed double peaks in 2019 while it monotonically decreased from about 05:00 to 16:00 during the study period in 2020. The diurnal variation of VE indicated that the peak value at 10:00 (local time) in 2020 was impaired compared to 2019. Previous studies concerning the impacts of short-term pollution control measures for mega-events on air pollutant variations showed decreases of air pollutants, while the diurnal patterns of air pollutants during control period were similar to the non-control period (Wang et al., 2015; Liu et al., 2015, Liu et al., 2017). The results here implied that the strictest lockdown measures not only reduced the primary emissions obviously but also changed the diurnal variation patterns of air pollutants. The universal reductions of air pollutants from various sources indicated a possibility of precise control of PM2.5 at a specific time of a day.

Fig. S7 shows the mean PM2.5 source contributions associated with different air mass classifications. For local air masses, the SIA mass contributions in 2020 (25.5 ± 12.8 μg m−3) were higher than that in 2019 (17.5 ± 12.5 μg m−3), while other sources showed obvious decreases compared to 2019. For air masses from NC, the mass contributions of the seven sources decreased in 2020 compared with those in 2019. For air masses from SC, the mass contributions of VE (12.7 ± 4.07 μg m−3) and RD (1.23 ± 1.07 μg m−3) in 2020 were higher than those (10.9 ± 4.65 μg m−3 for VE and 0.57 ± 0.57 μg m−3 for RD) in 2019.

Discussions

Role of emission on reduction

As shown in Table 1, higher ambient temperature, mixing layer height, solar irradiation, lower relative humidity, and wind speed in 2020 compared with those for 2019 were found. Due to the conflicting roles of these meteorological conditions on air pollutant formation and removal processes (Zhang et al., 2015; Chen et al., 2020), it was difficult to discuss the influences of emission reduction on air pollutant concentrations. Therefore, a weather normalized technic was adopted to analyze the influence of emission on air pollutant variations. After the weather normalized procedures, the pollutant levels can represent the anthropogenic emissions (Li et al., 2019a; Zhai et al., 2019). The differences between the observed and de-weather levels of air pollutants can be regarded as meteorology related variations. As shown in Fig. S12, meteorological conditions were favorable for reducing air pollutant concentrations for both the two periods. The pollutant emission dominated the reduction of air pollutants during the study period of 2019 and 2020 (Fig. 4 ).

Fig. 4

Changes of air pollutants, chemical species, and source contributions of PM2.5 between the study period in 2020 and 2019 due to emission (EMI) and meteorology (MET).

The observed PM2.5 decreased by 27.0 μg m−3 in the study period of 2020 (45.9 μg m−3), compared to that of 2019 (72.9 μg m−3). The emission reduction contributed to 24.8 μg m−3 (92.0%) and meteorology contributed only 2.20 μg m−3 (8.00%) to PM2.5 decrease. For the main chemical species of PM2.5, the variations due to emission were in the range of −0.87 μg m−3 (Cl−) to −9.50 μg m−3 (NO– 3). Similarly, emission-related variations were −0.37 μg m−3 (RD) to −8.09 μg m−3 (IP) for PM2.5 source contributions. The reduced mass concentrations for air pollutants caused by emission control were higher than those caused by meteorology related variations (Fig. 4). Previous studies suggested that meteorology dominated the day-to-day variations of air pollutants (He et al., 2016, He et al., 2017). However, the city lockdown experiment showed that substantial emission reductions dominated the reductions in mass concentrations of PM2.5 and its chemical compositions. The air quality improvement is a long-way journey. The results here implicated that air pollution control measures should be continuously taken step by step to improve the air quality in regions with serious air quality issues.

Enhanced contributions from secondary formation

OC, sulphate, nitrate, and ammonium are from both primary emissions and secondary formation (Huang et al., 2014), while EC is from primary emissions (Bond et al., 2013; Cao et al., 2013). As shown in Fig. 5 , the ratios of OC, sulphate, nitrate, and ammonium for the comparison between 2020 and 2019 were lower than 1, suggesting the reduction of primary emissions in 2020. The atmospheric oxidation ability (AOA), however, enhanced in Wuhan and northern China with the evidence from the increased Ox (O3 + NO2) (Fig. S1). The ratios of OC, sulphate, nitrate, and ammonium to EC were generally higher than 1 during a day (Fig. 5), suggesting a substantial enhancement of secondary formation (Huang et al., 2020). The sulphate and nitrate oxidation rates (SOR and NOR) also showed enhancements in 2020 (Fig. S13), implying higher fractions of sulphate and nitrate from secondary formation. For OC, sulphate, and ammonium, the ratios to EC ranged from about 5% to ~85% (Fig. 5A, B, D), suggesting stronger formation. The diurnal variations of the scaled sulphate, nitrate, ammonium, and OC (SNAO) by EC for the two periods further showed the enhancement of secondary formation (Fig. S14). The SNAO enhancement from 5:00 to 18:00 coincided with the diurnal cycle of O3, indicating the role of O3 oxidization in secondary formation during the daytime (Brown, 2006; Hallquist et al., 2009).

Fig. 5

Scatter plots between organic carbon (OC) (A), sulphate (B), nitrate (C), ammonium (D), and elemental carbon (EC) ratios between the observational periods in 2019 and 2020. The colour represents the hour of a day and the dot size represents the scaled levels of Ox (O3 + NO2) by EC in 2020 to those in 2019. The black solid lines represent the 1:1 ratio and the grey dash lines represent the ranges of ratio changes for the two periods.

This unique lockdown experiment can be regarded as an extremely sensitive test of primary emission reduction on air quality. Results here revealed that the primary emission reduced while the secondary formation increased. The lockdown measures yielded a significant decreasing of traffic volumes which subsequently reduced the NO levels. Reduction of NO emissions is important for O3 and inorganic species controls (South Coast Air Quality Management District, 2013). A recent study suggested that reducing NO will reduce nitrate, while increase the secondary organic aerosol (SOA) levels (Zhao et al., 2017) due to the enhancement of AOA (Praske et al., 2018; Huang et al., 2020). The impacts of NO reduction on AOA or SOA yield have a tipping-point. Before this tipping-point, reducing NO enhances the AOA and promotes the SOA yield, while after this point, further reducing NO emissions will impair AOA and limit the yield of SOA (Zhao et al., 2017; Huang et al., 2020). Such significant NO reduction during the lockdown still enhanced the AOA in Wuhan. In the short-term, the NO emissions will recover in China with the alleviate of COVID-19 (Zhang et al., 2020a) and the AOA should decrease. However, considering the historical decreasing trends of NO emission in developed countries (Duncan et al., 2016; Jiang et al., 2018), the NO emissions would be decreasing in China continuously. Decreasing volatile organic compounds (VOCs) would offset the secondary formation enhancement as NO emission reduction, considering the non-linear relationship between them (Huang et al., 2020). Therefore, further air pollution control would be challenging in China, considering the combination pollution of PM2.5 and O3 (Zeng et al., 2019). A two-pollutant strategy to solve PM2.5 and ozone problems should be carried out with a focus on VOCs and NO emission reduction in further air pollution control policies (Li et al., 2019b).

Reduced emission from different geographical source regions

It was estimated that CO, NO, SO2, PM2.5, BC, and OC emissions decreased by 13– 41%, 29– 57%, 15– 42%, 9– 34%, 13– 54%, and 3– 42%, respectively in the first two months in 2020 compared to 2019 (Huang et al., 2020). The differential concentration-weighted trajectory (DCWT) results (Text S3) showed reductions of source contributions for air masses transported from potential geographical regions affecting Wuhan for the two periods (Fig. 6 ). Generally, the DCWT values of the seven sources showed negative mean (± standard deviation) values of −3.29 ± 6.76 μg m−3 (BB), −1.28 ± 1.61 μg m−3 (CC), −0.26 ± 3.71 μg m−3 (FW), −3.87 ± 5.28 μg m−3 (IP), −0.22 ± 0.87 μg m−3 (RD), −4.36 ± 11.0 μg m−3 (SIA), and −2.23 ± 4.23 μg m−3 (VE), suggesting substantial emission reductions of these sources along the transportation routes to the receptor site. The large standard deviations, however, indicated the spatial variances, especially for BB, FW, and SIA.

Fig. 6

Differential concentration-weighted trajectory (DCWT, μg m−3) values of source contributions for the observational period in 2020 compared with those in 2019.

The DCWT values of FW increased by 27.5 μg m−3 at most for the air masses transported from the eastern Shandong and northern Jiangsu, suggesting the influences of regional transportation of firework burning on air quality in Wuhan, especially for the Spring Festival period. The DCWT values of SIA showed positive values for air masses transported from two routes: central Shanxi and Inner Mongolia as well as the southern Anhui and Jiangsu, with a maximum increase of 30.2 μg m−3 and 33.4 μg m−3, respectively. The increased DCWT values for SIA were similar to the results of a recent study that the enhanced secondary formation in the Beijing-Tianjin-Hebei and Yangtze River Delta regions (Huang et al., 2020). The positive DCWT values of SNA scaled by CO, especially for sulphate (Fig. S15) also verified the secondary formation enhancement for the two transport routes.

High SO2 emissions were found in northern China and the lower reach of the Yangtze River (van der A et al., 2017; Kourtidis et al., 2018). The air masses reaching Wuhan during winter were mainly from two transportation routes including north and northeast directions (Zheng et al., 2019b). High emissions and enhanced atmospheric oxidative ability contributed to the secondary formation of sulphate during air mass transportation from the two directions to Wuhan. This phenomenon was further supported by the enhancement of SOR from 0.45 (2019) to 0.68 (2020) for the air masses transported from NC (Fig. S16). The decreased atmospheric oxidative ability (Fig. S1) and less SO2 emissions in south China resulted in less secondary formation of sulphate, which was proved by the reduced SOR from 0.22 (2019) to 0.20 (2020) for air masses transported from SC (Fig. S16).

Conclusion

In this study, the variations of chemical species and sources of PM2.5 in Wuhan were investigated, where the strictest lockdown measures were adopted as COVID-19. Compared to the same period in 2019 (72.9 μg m−3), PM2.5 levels in the following one month after the lockdown of Wuhan in 2020 (45.9 μg m−3) decreased by 27 μg m−3, mainly due to the emission reduction (92.0%). The mass concentrations of main PM2.5 components showed reductions ranging from 0.85 (chloride) to 9.86 μg m−3 (nitrate). The mass contributions of various sources decreased in 1.54 (firework burning)–8.02 μg m−3 (industrial processes). The decreases of trace elements (0.65%) and elemental carbon (0.67%) suggested the reduction of primary emissions due to lockdown. The increased mass percentages of sulphate (3.05%), organic carbon (2.48%), and secondary inorganic aerosol (8.70%), however, suggested the enhanced secondary formation of PM2.5. The air masses reaching Wuhan from different potential geographical regions showed substantial reductions of the mass contributions from various sources, excepted for biomass burning, firework burning and secondary inorganic aerosols from several regions. This study highlighted that the lockdown measures not only reduced the mass concentrations of air pollutants and associated chemical compositions, but also modified the diurnal variation patterns of PM2.5 sources. Further researches should deeply investigate the emission reduction on atmospheric oxidation ability and secondary aerosol formation mechanisms.

CRediT authorship contribution statement

Huang Zheng: Conceptualization, Visualization, Writing - original draft, Writing - review & editing. Shaofei Kong: Conceptualization, Funding acquisition, Writing - review & editing, Supervision. Nan Chen: Data curation, Resources. Yingying Yan: Writing - review & editing. Dantong Liu: Writing - review & editing. Bo Zhu: Data curation. Ke Xu: Data curation. Wenxiang Cao: Data curation. Qingqing Ding: Data curation. Bo Lan: Data curation. Zhouxiang Zhang: Data curation. Mingming Zheng: Data curation. Zewei Fan: Validation. Yi Cheng: Validation. Shurui Zheng: Writing - review & editing. Liquan Yao: Writing - review & editing. Yongqing Bai: Conceptualization. Tianliang Zhao: Conceptualization. Shihua Qi: Resources.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.