Does Ambient Secondary Conversion or the Prolonged Fast Conversion in Combustion Plumes Cause Severe PM_2.5 Air Pollution in China?

Abstract

The ambient formation of secondary particulate matter (ambient FSPM) is commonly recognized as the major cause of severe PM_2.5 air pollution in China. We present observational evidence showing that the ambient FSPM was too weak to yield a detectable contribution to extreme PM_2.5 pollution events that swept northern China between 11 and 14 January 2019. Although the Community Multiscale Air Quality (CMAQ) model (v5.2) reasonably reproduced the observations in January 2019, it largely underestimated the concentrations of the PM_2.5 during the episode. We propose a novel mechanism, called the “in-fresh-stack-plume non-precipitation-cloud processing of aerosols” followed by the evaporation of semi-volatile components from the aerosols, to generate PM_2.5 at extremely high concentrations because of highly concentrated gaseous precursors and large amounts of water droplets in fresh cooling combustion plumes under poor dispersion conditions, low ambient temperature, and high relative humidity. The recorded non-precipitation-cloud processing of the aerosols in fresh stack combustion plumes normally lasts 20–30 s, but it prolongs as long as 2–5 min under cold, humid, and stagnant meteorological conditions and expectedly causes severe PM_2.5 pollution events. Regardless of the presence of the natural cloud in the planetary boundary layer during the extreme events, the fast conversion of air pollutants in water droplets and the generation of the PM_2.5 through the non-precipitation-cloud processing of aerosols always occur in fresh combustion plumes. The processing of aerosols is detectable using a nano-scan particle sizer assembled on an unmanned aerial vehicle to monitor the particle formation in stack plumes. In-fresh-stack-plume processed aerosols under varying meteorological conditions need to be studied urgently.

1. Introduction

As a result of great efforts to reduce air pollutant emissions in China, a large drop in the mass concentration of atmospheric particles with diameters smaller than 2.5 µm (PM_2.5) has been reported in inland and coastal atmospheres in comparison with the observations since 2013 [1,2,3,4,5,6,7,8,9,10,11,12,13]. For instance, the reported annual average mass concentration of the PM_2.5 in Beijing was 42 µg m⁻³ in 2019 and it decreased by approximately 50% relative to the value in 2013 (http://www.xinhuanet.com/politics/2020-01/03/c_1125419813.htm, accessed on 23 March 2022). However, severe PM_2.5 air pollution events, with 24 h averages of mass concentration exceeding 200 µg m⁻³, still occur in northern China, including Beijing [10,14,15]. Surprisingly, severe PM_2.5 pollution events continued to occur amid the peak period of the COVID-19 outbreak in 2020 [16,17]. Many hypotheses on the formation of secondary particulate matter (FSPM) in ambient air—referred to as “ambient FSPM” in this paper—have been suggested to interpret the severe events over the past decade [1,3,5,6,7,16,17,18,19,20,21,22]. The ambient FSPM is physically detectable in clean or slightly polluted atmospheres [23,24,25,26]. However, whether the mechanisms can yield a significant contribution to severe PM_2.5 pollution events is controversial [1,13,16,17,18,19,20,21,22].

In addition to the ambient FSPM, the primary particulate matter emitted from stack combustion sources has also been recognized as an important contributor to the ambient PM_2.5 [2,3,4,27,28,29,30,31]. To explore the contribution of the ambient FSPM to severe PM_2.5 air pollution, we introduce a metric, that is, PM_2.5/CO. CO is chemically stable in ambient air at the ground level and is an ideal indictor of the accumulation extent [3,29], although CO may not be necessarily derived from the same sources of PM_2.5. A combination of the mass concentrations of the PM_2.5 and the ratios of the PM_2.5/CO allowed us to reasonably identify the contribution of accumulating and/or mixing the primary particulate matter to largely increase the PM_2.5 [29] and then examining whether the contribution of the ambient FSPM to a largely increased PM_2.5 is negligible or not. In addition, we highlight three technical terms, that is, (1) ambient FSPM, (2) primary particulate matter which has been generated prior to being released to ambient air, and (3) FSPM in fresh stack combustion plumes, which we refer to as the “in-fresh-stack-plume non-precipitation-cloud processing of aerosols” that lasts for dozens of seconds to several minutes with the rapid cooling of water vapor in seconds, the fast chemical conversion of chemicals in droplets, and the evaporation of water droplets to generate the PM_2.5 afterward (see Figure 1a–d, Supplementary Materials Videos S1–5, and Feng et al. [27]). The FSPM in the fresh stack combustion plumes cannot be treated as the primary particulate matter while it is also largely different from the ambient FSPM. Although several studies in the literature reported the formation and dispersion of primary and secondary pollutants [32,33,34,35], the rapid cooling of water vapor and in-fresh-stack-plume non-precipitation-cloud processing of aerosols have never been considered. Figure 2a–h show how the PM_2.5 and PM_2.5/CO change under two contrasting conditions, that is, accumulation and dilution, in the presence or absence of the two formation mechanisms. The details are provided in Materials and Methods.

With the above-mentioned technical terms in mind, an extreme PM_2.5 pollution event that occurred on 11–14 January 2019 in the inland and coastal atmospheres across northern China provides an ideal example to explore the cause of severe PM_2.5 pollution events in present-day China. In addition, the measurements and simulations during other periods were also used to facilitate the analysis.

2. Materials and Methods

2.1. Data Sources

The hourly average mass concentrations of air pollutants were obtained from air quality monitoring networks in China, where thousands of air quality monitoring sites have been established in hundreds of cities. The data for each day can also be downloaded from https://www.iqair.com/china, accessed on 23 March 2022; however, the web-based raw data have no quality control and need to be processed before analysis. Note that the hourly average of each air pollutant in Baoding reflects the mean value of its concentrations at tens of local stations; the same is true in the case of Langfang (Figure S1). Local time is used in this study, for example, the hourly average mass concentration of a pollutant at 01:00 represents the average from 01:00 to 01:59, and this definition is applied for the whole study. The meteorological data in Tiantan (39.875° N, 116.434° E) and Wanliu (39.993° N, 116.315° E), including ambient temperature (T), relative humidity (RH), wind speed (WS), wind direction (WD), and precipitation, can be obtained from http://data.cma.cn/data/detail/dataCode/A.0012.0001.html (accessed on 23 March 2022) after registration. The real-time aerosol chemical species (including organic, SO₄²⁻, NO₃⁻, and NH₄⁺) in PM_2.5 at an urban site in Beijing (39.974° N, 116.371° E, marked as U-site in Supplementary Materials Figure S2) were measured from 30 December 2018 to 15 January 2019 using a PM_2.5 time-of-flight aerosol chemical speciation monitor (ToF-ACSM), and the data have been published by Lei et al. [36]. The closest air monitoring station, i.e., Aoti (40.003° N, 116.407° E) was ~5 km from the urban site (Supplementary Materials Figure S2).

Data collected at three sites in Qingdao, a coastal megacity, were used for the analysis in this study (Supplementary Materials Figure S3). Site 1 and Site 2 were situated at a new high-technique zone within ~1 km. At Site 1, various instruments were housed in a research lab. The sample inlets were directly connected to ambient air outside the window, and the sampling height was approximately 15 m above the ground level. A fast mobility particle size (FMPS 3091, TSI) downstream of a dryer (TSI) operating at 1 s intervals is used to measure the concentration of atmospheric particles from 5.6 to 560 nm. In addition, the condensation particle counter (CPC 3785, TSI) was used to measure the total particle number concentration. Electricity-free sampling tubes (TSI) with a length of ~2 m were used for the particle measurements. Details about the instruments used in this study can be found in Zhu et al. [37]. A multiple log-normal function was used to fit the number particle size distribution [37]. The geometric median mode diameter (GMMD) of the fitted accumulation mode at 100–300 nm was used for analysis. Kelvin effects are expected to be negligible in the size range and they should have a larger growth rate than those in particles beyond 300 nm because of the mass transfer effect. An additional stainless-steel sampling tube was connected to a URG-9000D Ambient Ion Monitor-Ion chromatography (AIM-IC, Thermofisher, Waltham, MA, USA) installed with a PM_2.5 cyclone operating at a rate of 3 L/min. The AIM-IC reported semi-continuous concentrations of chemically reactive gases including atmospheric ammonia and SO₂ together with their particulate partners in 1 h time resolution. The AIM-IC included an ICS-1100 Ion Chromatograph (Thermofisher, Waltham, MA, USA), which was equipped with an analytical column (Ion Pac CS17A (Thermofisher, Waltham, MA, USA) (2 × 250 mm) for cations and Ion Pac AS11-HC (Thermofisher, Waltham, MA, USA) (2 × 250 mm) for anions) and a guard column (CG17A (2 × 50 mm) for cations and AG11-HC (Thermofisher, Waltham, MA, USA) (2 × 50 mm) for anions). More detailed information on AIM-IC analysis can be found in our previous study [38].

A local air quality monitoring station is situated at Site 2. All criteria air pollutants are regularly measured according to the standard protocol adopted for air quality monitoring networks in China. However, the data need access permission. All meteorological data, including T, RH, WS, WD, and precipitation, were also obtained from the sampling site. Site 3 is approximately 30 km away from Site 1 and Site 2. The planetary boundary layer (PBL) height and the extinction of aerosol coefficients are obtained by a micropulse Lidar (EV-Lidar-CAM, Beijing Everise Technology Ltd., http://www.everisetech.com.cn/, accessed on 23 March 2022) deployed on the roof of a fourth-floor building at Site 3. The instrument measured the aerosol back-scattered signal from 105 m up to 10 km every 2 min, with a vertical resolution of 15 m. The extinction coefficients were obtained based on the Fernal method [39]. This instrument operates at a wavelength of 532 nm with a single output energy of 10 μJ and a repetition rate of 2500 Hz.

A nano-Scan Particle Sizer (SMPS, Model 3910, TSI corporation, Upper Marlboro, MD, USA) was assembled on an unmanned aerial vehicle (UAV) developed in 2017 (PloughUAV Corporation, Shanghai, China) to monitor particle formation in stack plumes. The measurements were performed in the morning and afternoon on December 22, 2017 (300 m height stacks in Shanghai, 31.460° N, 121.421° E) in Shanghai, China. Each cruise lasted for approximately 15 min, including lifting, suspending measurements, and landing. The other UAV (Phantom-4, DJI Corporation, Shenzhen, China) with a camera was used to record the flying process with the UAV carrying the SMPS. The SMPS data are uncorrected because sampling efficiencies are still under investigation.

2.2. Scenario Analysis

Scenario A, as shown in Figure 2a: Supposing that ambient FSPM and FSPM in fresh stack combustion plumes are too weak to yield a detectable contribution to PM_2.5 in mass concentration, it is evident that PM_2.5 comes mainly from primary anthropogenic sources. Under poor dispersion conditions, primary anthropogenic air pollutants, including PM_2.5 and CO, may accumulate to a high level, thereby increasing the concentrations of both species with time. However, the ratio of PM_2.5/CO remained constant with the accumulation effect of both species to be canceled out. Note that the CO was just an indicator of accumulation and may not necessarily come from the same sources of PM_2.5 and the precursors.

Scenario B, as shown in Figure 2b: When the polluted air is diluted solely by clean air (assuming that the clean air has negligible levels of PM_2.5 and CO), the concentrations of PM_2.5 would decrease with time. However, PM_2.5/CO remained constant with the dilution effect of both species to be canceled out.

Scenario C, as shown in Figure 2c: Supposing that ambient FSPM occurs concurrently with the accumulation of primary anthropogenic air pollutants, both the concentrations of PM_2.5 and ratios of PM_2.5/CO would increase with time.

Scenario D, as shown in Figure 2d: Supposing that ambient FSPM occurred concurrently with the dilution of highly concentrated primary anthropogenic air pollutants by the clean air, the ratios of PM_2.5/CO always increase. The concentrations of PM_2.5 generally decrease with time, except on occasions where the rates of ambient FSPM equal or exceed the dilution rate of PM_2.5 (not shown).

Scenario E, as shown in Figure 2e: Supposing that the aerosols are mainly derived from FSPM in fresh stack combustion plumes and accumulate under poor dispersion conditions, the concentrations of PM_2.5 observed at the downwind site, where water droplets in the plumes have been completely dissipated, would increase with time. The ratios of PM_2.5/CO at the same site may remain constant or decrease with time, depending on the evaporation extent of semi-volatile species from PM_2.5.

Scenario F, as shown in Figure 2f: Supposing that the aerosols are mainly derived from FSPM in fresh stack combustion plumes and have accumulated to a high extent. The aerosols at high concentrations are diluted by the clean air, and the concentrations of PM_2.5 always decrease with time. The ratios of PM_2.5/CO may remain constant or decrease with time, depending on the evaporation extent of semi-volatile species from PM_2.5.

Scenario G, as shown in Figure 2g: Supposing that the less polluted ambient air (or clean air) is replaced by the polluted plume, both the concentrations of PM_2.5 and the ratios of PM_2.5/CO would increase with time. In this scenario, an excellent linear correlation between PM_2.5 and PM_2.5/CO usually exists. Further, assuming the occurrence of ambient FSPM at various formation rates, the ratios of PM_2.5/CO would be distributed in the red shadowed area. Assuming the occurrence of evaporation of semi-volatile species from PM_2.5 at various evaporation rates, the ratios of PM_2.5/CO would be distributed in the gray shadowed area.

Scenario H, as shown in Figure 2h: Supposing that the polluted plume is replaced by a less polluted plume (or clean air), both the concentrations of PM_2.5 and the ratios of PM_2.5/CO would decrease with time. Further, assuming the occurrence of ambient FSPM at various formation rates, the ratios of PM_2.5/CO would be distributed in the red shadowed area. Assuming the occurrence of evaporation of semi-volatile species from PM_2.5 at various evaporation rates, the ratios of PM_2.5/CO would be distributed in the gray shadowed area.

In ambient air, one above-mentioned scenario may sometimes dominate. In these cases, the accumulation, dilution, replacement, and ambient FSPM can be easily identified. In general, a combination of different scenarios may affect the ratios of PM_2.5/CO. In these cases, the ratios should be analyzed cautiously.

2.3. Meteorological and Air Quality Model

The Weather Research and Forecasting (WRF) Model (v3.9) was used to generate meteorological fields for air quality modeling in this study. The meteorological modeling system includes two domains: Domain 1, a mother domain containing 164 × 97 grid cells at a 36 km spatial resolution for China, and Domain 2, a sub-domain containing 142 × 220 grid cells at a 12 km resolution for eastern China (Supplementary Materials Figure S4). The Community Multiscale Air Quality (CMAQ) model (v5.2) was applied for the simulation of concentrations of air pollutants in January 2019 with the episodic events [40]. The model’s vertical resolution included 14 layers from the surface to the tropopause, with the first layer height about 35 m above ground level. The CB05TUCl chemical mechanism was chosen to simulate the gas-phase chemistry, while the aerosol module version 6 (AERO6) was chosen to simulate the particulate-phase chemistry. The 2017 Multi-resolution Emission Inventory for China (MEIC) was used in this study. The initial and boundary conditions for Domain 1 were generated from the GEOS-Chem global chemistry model, and those for Domain 2 were extracted from Domain 1. In this study, the one month was simulated using an initial one-week spin-up period to minimize the influence of initial conditions. More details are provided in the Supporting Information.

Two parameters, i.e., normalized mean bias (NMB) and normalized mean error (NME), were used to evaluate the performance of the CMAQ modeling results. The two parameters were defined as:

$$\mathrm{NMB}=\frac{{{\displaystyle \sum }}_{\mathrm{i}=1}^{\mathrm{n}}\left({\mathrm{S}}_{\mathrm{i}}-{\mathrm{O}}_{\mathrm{i}}\right)}{{{\displaystyle \sum }}_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{O}}_{\mathrm{i}}} \times 100\%$$

(1)

$$\mathrm{NME}= \frac{{{\displaystyle \sum }}_{\mathrm{i}=1}^{\mathrm{n}}\left|{\mathrm{S}}_{\mathrm{i}}-{\mathrm{O}}_{\mathrm{i}}\right|}{{{\displaystyle \sum }}_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{O}}_{\mathrm{i}}} \times 100\%$$

(2)

where S_i is the modeled atmospheric concentration of PM_2.5, O_i is the observed atmospheric concentration of PM_2.5, ${\displaystyle \sum }_{\mathrm{i}=1}^{\mathrm{n}}$ is the sum of difference between S_i and O_i, n is the data number. Based on the US EPA recommended evaluation standards, the modeled concentrations are accepted with NMB ≤ ±30% and NME below 75% [41].

3. Results

3.1. Evidence for Negligible FSPM during Severe PM_2.5 Air Pollution in Inland Atmospheres

An extreme PM_2.5 pollution event on 11–14 January 2019 swept over northern China, ~2000 km in length and ~300 km in width (Figure 2i), and it affected approximately 400 million people. The maximum 24 h average PM_2.5 mostly exceeded 200 µg m⁻³, and the increased PM_2.5 concentrations started to be observed as early as 9 January (Figure 2j and Supplementary Materials Figure S5a–b) in some cities. Beijing was the most vulnerable, with the recorded hourly average of the PM_2.5 exceeding 500 µg m⁻³ over 4–10 h at various urban sites (Figure 3a,d, and Supplementary Materials Figure S6a–i), the maximum 24 h average exceeding 270 µg m⁻³. We started the analysis from Beijing, followed by the analysis of observations made in the neighboring cities of Langfang and Baoding as well as the coastal megacity of Qingdao, to explore the formation mechanism of the PM_2.5 during this extreme pollution event. The observations at the Tiantan site in Beijing were selected for illustration, where large increases in the PM_2.5 were observed during three separate periods on 11–14 January 2019 (Figure 3a). However, the ambient FSPM was too weak to be detectable, as evidenced below and corresponded to various scenarios above-mentioned:

(1) Evidence 1a: The hourly averages of the PM_2.5 increased by 137% from 18:00 on 13 January (PM_2.5 = 126 µg m⁻³ at 18:00–18:59) to 05:00 on 14 January (PM_2.5 = 299 µg m⁻³ at 05:00–05:59), with almost invariant ratios of the PM_2.5/CO around 93 ± 2 µg m⁻³/mg m⁻³ (median value ± standard deviation, 2/93 ≈ 2.2%; see the black markers in Figure 3a). This indicates that the largely increased levels of the PM_2.5 were probably caused solely by the accumulation of air pollutants. At 2–7 h immediately before the increase in the PM_2.5 (that is, 12:00–16:00 on 13 January), the hourly averages of the PM_2.5 and the ratios of the PM_2.5/CO stood on a relatively low and stable platform and fluctuated narrowly around 11 ± 0.8 µg m⁻³ (average value ± standard deviation) and 37 ± 3 µg m⁻³/mg m⁻³, respectively. It took 2 h for the ratio to jump from the low platform (37 ± 3) to the high platform (93 ± 2), with a clear 1 h transition at 17:00 on 13 January (PM_2.5/CO = 54 µg m⁻³/mg m⁻³). The extreme increase in the rate of the ratios solely during these 2 h (but not before or after) clearly reflects a rapid replacement process of the cleaner ambient air by the polluted air mass rather than by any detectable contribution from the ambient FSPM. In the replacement process, an excellent correlation was observed between the two variables, that is, PM_2.5/CO = 0.54 × PM_2.5 + 29, R² = 0.998 (p < 0.05) when the data at 16:00 (start), 17:00 (middle), and 18:00 (end) were used.

(2) Evidence 1b: A similar process recurred from 22:00 on 11 January to 04:00 on 12 January (red marker in Figure 3a) when the PM_2.5/CO stood on the platform of 104 ± 3 µg m⁻³/mg m⁻³ (3/104 = 2.8%) with an increase in the hourly average of the PM_2.5 by 27%, that is, from 285 to 362 µg m⁻³. The increase in the PM_2.5 was also probably determined solely by accumulation. At 3–7 h before the accumulation (from 16:00 to 19:00 on 11 January), the hourly averages of the PM_2.5 and the ratios of the PM_2.5/CO stood on a low platform, narrowly fluctuating around 75 ± 5 µg m⁻³ and 54 ± 1 µg m⁻³/mg m⁻³ (1/54 = 1.9%), respectively. The PM_2.5/CO took 3 h to jump on the higher platform of 104 ± 3 µg m⁻³/mg m⁻³, but no detectable ambient FSPM can be observed before 19:00 and after 22:00 on 11 January. The replacement alone was responsible for the excellent correlation between two variables, that is, PM_2.5/CO = 0.22 × PM_2.5 + 40, R² = 0.94 (p < 0.05) when the data at 19:00 (start), 20:00 (middle), 21:00 (middle), and 22:00 (end) were used.

The meteorological conditions for the time periods examined in Evidence 1a and 1b (Figure 3c) were a higher ambient relative humidity (RH > 60%) and a lower ambient temperature (T < −4 °C). The wind speed fluctuated approximately 1 m·s⁻¹ (Supplementary Materials Figure S7a–b), favoring the accumulation of air pollutants. In addition, no precipitation was recorded between 11 and 14 January (Supplementary Materials Figure S7c).

(3) Evidence 1c: When similar meteorological conditions recurred in the late afternoon on 12 January (Figure 3c and Supplementary Materials Figure S7a–d) with an even higher ambient RH lasting for a longer time, a longer process including the accumulation, replacement, and evaporation of the PM_2.5 was observed from 11:00 on 12 January to 08:00 on 13 January. The accumulation was first observed from 11:00 to 14:00 on 12 January (blue markers in Figure 3a), leading to a large increase in the PM_2.5 by 63%, that is, from 114 to 186 µg m⁻³. However, the PM_2.5/CO fluctuated narrowly around 71 ± 2 µg m⁻³/mg m⁻³ (2/71 = 2.8%).

(4) Evidence 2: The observation commenced at 21:00 on 12 January and ended at 03:00 on 13 January (purple markers in Figure 3a). During this period, the hourly averages of the PM_2.5 fluctuated narrowly around 507 ± 6 µg m⁻³ (6/507≈1%), indicating no detectable ambient FSPM. In contrast, the ratios of the PM_2.5/CO decreased by 16%, that is, from 139 to 116 µg m⁻³/mg m⁻³ with CO accumulation. On the PM_2.5 platform, the decreasing ratios of the PM_2.5/CO were likely caused by a fraction of semi-volatile species evaporating from the PM_2.5 under the condition of decreased ambient RH (Figure 3c), and the evaporation effect seemingly canceled out the accumulation effect of the PM_2.5.

(5) Evidence 3: We further plotted a theoretical dilution curve in Figure 3b (gray dashed line), assuming that the most polluted plume (that is, PM_2.5 = 583 µg m⁻³ and PM_2.5/CO = 149 µg m⁻³/mg m⁻³ observed at 19:00 on 12 January) was physically diluted by the moderately polluted plume at the end of the dilution (that is, PM_2.5 = 158 µg m⁻³ and PM_2.5/CO = 88 µg m⁻³/mg m⁻³ observed at 08:00 on 13 January) in the absence of ambient FSPM and new emissions. Assuming that the ambient FSPM took place during the dilution period, the predicted values would be distributed throughout the red shadowed area in Figure 3b above the theoretical dilution curve. This is untrue, and the null hypothesis, that is, the detectable contribution from the ambient FSPM to the increased PM_2.5, was thus rejected. The theoretical replacing curve during the increasing period is also presented in Figure 3b (pink dashed line with two endpoints at the maximum PM_2.5 and PM_2.5/CO, and the smaller values obtained at 14:00 on 12 January). The observed values generally stood on the curve, allowing us to reject the null hypothesis.

(6) Evidence 4: The CMAQ modeling results were further compared with the observations at the Tiantan site on 1–31 January 2019 (Figure 4a). The NME and NMB were 67 and −3%, respectively, indicating that the modeling performance was acceptable according to the US EPA recommended evaluation standards [41]. However, the modeling results were clearly classified into two periods, i.e., an overestimation period covering from 00:00 on 1 January to 20:00 on 11 January and from 00:00 on 15 January to 23:00 on 31 January of 2019 (NME = 66% and NMB = 36%), and an underestimation period covering from 21:00 on 11 January to 18:00 on 14 January of 2019 (NME = 70% and NMB = −69%). Combining the positive NMB during the overestimation period and the largely negative NMB during the underestimation period allows to infer missing mechanisms of the fast formation of the PM_2.5 during the underestimation period [13,14,15,16,18], when the missing mechanisms overwhelmingly determined the concentrations of the PM_2.5. The similar comparison results can be obtained in Qingdao (Figure 4b), and we will return to this later.

(7) Additional evidence: On 12–13 January, at the Wanliu site in Beijing, a negligible contribution from the ambient FSPM to the largely increased PM_2.5 was also confirmed during the pollution event (Figure 3d–f) and is further discussed in Supporting Information. The negligible contribution from the ambient FSPM can be confirmed during the events when the analysis was applied to the measurements at other sites in Beijing, shown in Supplementary Materials Figure S6a–e.

The negligible contribution from the ambient FSPM to the largely increased PM_2.5 was confirmed in Langfang (~50 km south of Beijing; Supplementary Materials Figure S5a–b) and the analysis was detailed in Supporting Information. The contribution from the ambient FSPM to the significant increase in the PM_2.5 was also too weak to be detectable in Baoding (~140 km southwest of Beijing, Figure 2j) on 9–13 January. The maximum value (PM_2.5 = 507 µg m⁻³) occurred at 08:00 on 12 January in Baoding, which was 12–13% smaller than the maximum values observed at the Taintan and Wanliu sites in Beijing 11–12 h later. The PM_2.5/CO corresponding to the maximum concentration of the PM_2.5 in Baoding was only 77 µg m⁻³/mg m⁻³ and approximately half of the values at the two sites in Beijing. The events in Baoding are further detailed as below.

The hourly averages of the PM_2.5 increased largely by 408% from 13:00 on 9 January (PM_2.5 = 49 µg m⁻³) to 07:00 on 11 January (PM_2.5 = 249 µg m⁻³) with the ratios of the PM_2.5/CO, which fluctuated around 60 ± 6 µg m⁻³/mg m⁻³ (6/60 = 10%), with no increasing trend. The ratios of the PM_2.5/CO even narrowed at approximately 61 ± 4 µg m⁻³/mg m⁻³ (4/61 = 6.6%, red markers in Figure 2j), excluding the data measured at 05:00–13:00 on 10 January. After 07:00 on 11 January, with a several-hour transition, the ratios of the PM_2.5/CO reached a higher level (black markers) and fluctuated narrowly at approximately 87 ± 4 µg m⁻³/mg m⁻³ (4/87 = 4.6%). On the higher level, the hourly averages of the PM_2.5 increased largely by 85% (from 267 to 495 µg m⁻³) and then decreased by 26% (from 495 to 361 µg m⁻³). The increase and decrease in the hourly average of the PM_2.5 are probably caused by the accumulation effect and the dilution effect, with a negligible contribution from the ambient FSPM. When the concentration of the PM_2.5 increased by 15% from 04:00 (PM_2.5 = 441 µg m⁻³) to 08:00 on 12 January (PM_2.5 = 506 µg m⁻³), the ratios of the PM_2.5/CO stood on another platform and fluctuated narrowly around 76 ± 2 µg m⁻³/mg m⁻³ (2/76 = 2.6%, green markers). The increase in the hourly average of the PM_2.5 was also probably caused by the accumulation alone. Overall, the switch of the three platforms, but with invariant ratios of the PM_2.5/CO inside each, should reflect the replacement of different plumes. Based on the analysis of a negligible increase in the PM_2.5/CO driven by the ambient FSPM in Baoding, the negligible ambient FSPM was also expected during the long-range transport of the polluted air masses from Baoding to downwind zones.

To examine whether the contribution from the ambient FSPM to the largely increased PM_2.5 has always been too weak to be detectable in northern China, the observational results on 13–14 January in Qingdao, approximately 500 km southeast of Beijing (see maps in Supplementary Materials Figures S1 and S3), were studied (Figure 5a–g).

3.2. Evidence for Negligible FSPM during Severe PM_2.5 Air Pollution in Coastal Atmospheres

On 13 January, the hourly averages of the PM_2.5 increased by 243% from 04:00 (purple dashed line, PM_2.5 = 82 µg m⁻³) to 12:00 (green dashed line, PM_2.5 = 281 µg m⁻³; Figure 5a) in Qingdao. At the same period, the ratios of the PM_2.5/CO increased by 148%. However, the geometric median mode diameter (GMMD) of the accumulation mode particle (>100 nm) number concentration fitted by log-normal functions exhibited a stepwise change, that is, it fluctuated narrowly around 175 ± 8 nm from 00:00 to 07:41, stepping down to a platform around 158 ± 8 nm from 07:42 to 11:08, and returning to a platform around 178 ± 12 nm from 11:09 to 12:59 (black line in Figure 5b). A lack of detectable growth in the size of the accumulation mode particles allowed us to confirm that there was no detectable gas–particle condensation formation of particle matter [19,22]. Alternatively, the number concentration of atmospheric particles with diameters larger than 100 nm (N_{>100 nm}) increased (Figure 5c) with increasing mass concentrations of the PM_2.5. The two variables exhibited a moderately good correlation (R² = 0.81, p < 0.01; Supplementary Materials Figure S8a). The good correlation together with a lack of the detectable accumulation mode particle growth in size strongly suggested that the accumulation solely caused the largely increased levels of the PM_2.5.

The particles primarily emitted from coal combustion in China reportedly consist of appreciable chloride salts as well as sulfate and nitrate salts [28,42,43]. Similar to biomass burning and sea-salt aerosols, chloride is expected to be replaced by secondarily formed nitrate and sulfate in aerosol aging processes [44,45] and chloride depletion would consequently lead to a decrease in the molar ratios of Cl⁻/SO₄²⁻ and/or Cl⁻/NO₃⁻. However, both the ratios of the Cl⁻/SO₄²⁻ and Cl⁻/NO₃⁻ increased with increasing concentrations of Cl⁻, SO₄²⁻, and NO₃⁻ during the PM_2.5 increasing period (Figure 5a,f), allowing us to reject the null hypothesis of a detectable contribution from ambient FSPM to largely increase the PM_2.5. Moreover, the mass ratios of the NH₄⁺/PM_2.5 exhibited a decreasing trend during the PM_2.5 increasing period (Figure 5a, g). This implies that there is an increasing contribution from non-ammonium aerosols, for example, carbonaceous species, which are mainly derived from primary combustion emissions [3,36,42].

During the period of the PM_2.5 increase from 04:00–12:00, the PM_2.5 exhibits an excellent correlation with the PM_2.5/CO, that is, [PM_2.5/CO] = 0.42 × [PM_2.5] + 25, R² = 0.98, p < 0.01 (Supplementary Materials Figure S8b) as the slightly polluted plume was replaced by the heavily polluted plume. From 13:00 to 16:00, the PM_2.5/CO decreased by 9% (that is, from 149 to 136 µg m⁻³/mg m⁻³), while the concentration of the PM_2.5 increased by 22% (that is, from 277 to 338 µg m⁻³). The GMMD of the accumulation mode (Figure 5b) slightly decreased by ~30 nm with increasing N_{>100 nm} (Figure 5c) during this period. This indicates that the accumulation was accompanied by the evaporation of semi-volatile species from the particles to some extent, supported by decreasing ratios of Cl⁻/SO₄²⁻ and Cl⁻/NO₃⁻, similar to Scenario E in Figure 2e. This was caused by the increase in the mixing layer based on the observed radar signals (Figure 5e) and meteorological conditions (Figure 5d and Supplementary Materials Figure S9a–c).

The CMAQ modeling results were also evaluated against the observations at the Aoshan site in Qingdao on 1–31 January 2019 (Figure 4b). The calculated NMB and NME indicated an acceptable modeling performance according to the US EPA recommended evaluation standards [41]. However, a worsened modeling performance accompanied by increased PM_2.5 concentrations can be found during the first-half period than in the second-half period, i.e., NME = 46 %, NMB = −23 %, and R = 0.57 (first-half period) versus NME = 36 %, NMB = −2 %, and R = 0.78 (second-half period). An even worsened modeling performance and larger underestimation of the PM_2.5 concentrations occurred during the period from 04:00 on 13 January to 05:00 on 15 January with an NMB as low as −49 %, when high concentrations of the PM_2.5 were observed. Missing mechanisms of the fast formation of the PM_2.5 may also overwhelmingly determine the concentrations of the PM_2.5 during the short period.

Overall, the ambient FSPM is confirmed to be too weak to contribute to largely increased PM_2.5 during severe PM_2.5 pollution events at various locations on different days in northern China. To meet the law of mass conservation, a mechanism that can generate mass concentrations of PM_2.5 exceeding the observations was required to interpret the event by considering evaporation and dispersion of PM_2.5 in the ambient air.

4. Discussion

4.1. Impractical Conversion Rate of SO₂ to SO₄²⁻ Absence of Non-Precipitation Cloud during the Events

To study the chemical formation of secondary ions such as SO₄²⁻, NO₃⁻, and NH₄⁺ in the PM_2.5, the chemical components in the PM_2.5 measured at the U-site and the routine air quality observations at the nearby Aoti site were combined during the events (Figure 6a–d). At the Aoti site, the PM_2.5 concentrations increased from 118 ± 9 µg m⁻³ at 12:00–14:00 to 548 ± 20 µg m⁻³ at 20:00–22:00 on 12 January when the concentrations of the SO₄²⁻, NO₃⁻, NH₄⁺, and organics in the PM_2.5 at the U-site increased by 169, 66, 72, and 159 µg m⁻³, respectively (Figure 6a–b). The large increase in the SO₄²⁻ concentrations led to the mass ratio of (SO₄²⁻ + NO₃⁻ + NH₄⁺) to organics that were as high as 1.6 ± 0.2 from 16:00 on 12 January to 05:00 on 13 January (blue shadow in Figure 6b), contrasting to 1.1 ± 0.2 from 16:00 on 11 January to 09:00 on 14 January by excluding the blue shadow period. Thus, we further calculated the conversion rate of SO₂ to SO₄²⁻ during the blue shadow period. Suppose that no ambient conversion of SO₂ to SO₄²⁻ occurred, the ratios of the SO₂/CO should be constant through the blue shadow period. Thus, the ratios of the SO₂/CO at the initial few hours, i.e., 4.1 ± 0.0 µg m⁻³/mg m⁻³ at 12:00–14:00 on 12 January, were combined with the CO concentrations measured afterward to estimate the theoretically accumulated SO₂ concentration absent of secondary conversions (dashed line in Figure 6c). The difference between the dashed line and the observed concentrations of SO₂ (black cross) can be the assumed converted amount of SO₂ to SO₄²⁻ and the conversion rate of the SO₂ was as high as 55% per hour during the increasing period from 12:00–14:00 to 20:00–22:00 on 12 January. Even with such high conversion rates, the converted SO₂ can only account for ~30% of the increased SO₄²⁻ concentrations in the PM_2.5 during the increasing period. In this case, an additional hypothesis was required to interpret the lack of the mass conservation between the increased concentrations of SO₄²⁻ in the PM_2.5 and the theoretically converted SO₂, i.e., (1) the ratio of the SO₂/CO near the SO₂ sources should be larger than 4.1 ± 0.0 µg m⁻³/mg m⁻³; (2) when the SO₂ transported from the sources to the observational site, a large fraction of sticky SO₂ has been scavenged through atmospheric deposition; (3) however, it is not the case for the SO₄²⁻ in the PM_2.5 because of its smaller deposition velocity. In the absence of the non-precipitation-cloud processing conversion of the SO₂, such high conversion rates were practically impossible according to previously reported results in combustion plumes [46,47,48,49]. The same analysis can be applied to the conversion of NO₂ to particulate NO₃⁻ as shown in Figure 6c. The 11% conversion of NO₂ to NO₃⁻ per hour during the increasing period was also practically impossible absent of the non-precipitation-cloud processing conversion of NO₂.

In addition, the different temporal variation patterns of the PM_2.5 mass concentration at various sites in Beijing (Figure 6a) suggested that the local sources rather than the long-range transport of air pollutants caused the event. The analysis was also supported by the spatial heterogeneity of the PM_2.5/CO during the blue shadow period. For example, the PM_2.5/CO corresponding to the maximum concentration of the PM_2.5 reached 149 µg m⁻³/mg m⁻³ at the Tiantan site and occurred at 19:00 on 12 January (Figure 3a). One hour later, the maximum PM_2.5/CO of 120 µg m⁻³/mg m⁻³ occurred at the Wanliu site (Figure 3d). Two hours later, the maximum PM_2.5/CO simultaneously occurred at the Aoti (128 µg m⁻³/mg m⁻³) and Shunyixincheng (104 µg m⁻³/mg m⁻³), although there was 38 km between the two sites. Although the long-range transport from the south, e.g., Langfang, to Beijing cannot be completely excluded, the maximum value of the PM_2.5 in Langfang was only 438 µg m⁻³ and occurred at 13:00 on 12 January. The PM_2.5/CO corresponding to the maximum concentration of the PM_2.5 in Langfang was 117 µg m⁻³/mg m⁻³, which was almost the same as the corresponding maximum value of 120 µg m⁻³/mg m⁻³ at the Wanliu site. However, the maximum value was 24% smaller than that observed at the Tiantan site in Beijing.

4.2. Hypothesis for In-Fresh-Stack-Plume Non-Precipitation-Cloud Processing of Aerosols

Figure 1a shows that the in-fresh-stack-plume non-precipitation-cloud processing of aerosols includes several processes: (1) the rapid cooling of water vapor immediately after the plume leaves the stack, leading to the formation of water droplets similar to a non-precipitation cloud (Figure 1d); (2) highly concentrated air pollutants would experience non-precipitation-cloud processing of aerosols, including the fast chemical conversion of air pollutants in droplets, and the evaporation of droplets to generate aerosols with largely decreasing liquid water content afterward [46,47,48,49,50,51,52,53]; (3) freshly processed aerosols would experience further evaporation, with the plumes to be diluted, if the aerosols include semi-volatile species such as NH₄NO₃ and small molecular weight organics [42]. The Hoppel effect is the direct result of the non-precipitation-cloud processing of secondary aerosols [50], leading to an accumulation mode (100–200 nm) of particle number concentration and a trough at 50–100 nm in the particle number concentration spectra (Figure 1b). No measurements in the stack plumes during the extreme events were available in this study. The non-precipitation-cloud processing of aerosols in the stack plumes can clearly be seen during the observations shown in Figure 1b, the data of which were measured by an unmanned aerial vehicle equipped with a particle sizer flying through the in-fresh-stack-plume non-precipitation cloud (Figure 1c–d). In the ambient air near the ground level during the unmanned aerial vehicle observational period, the Aitken mode particles dominated the particle number size spectra with the GMMD of Aitken mode particles at 50–60 nm (Figure 1b). In the stack plumes, the accumulation mode particles dominated the particle number size spectra with a trough in 50–100 nm. More details on the unique measurement can be found in Supporting Information.

The non-precipitation-cloud processing rate of secondary aerosols, for example, sulfate aerosols, is commonly assumed to be approximately two orders of magnitude faster than that in ambient air without the processing of aerosols [52]. The in-fresh-stack-plume non-precipitation-cloud processing of aerosols should have an even higher formation rate of secondary aerosols than in ambient air because of the high concentrations of air pollutants in plumes. For example, the lower limit values of SO₂ were 200, 100, and 50 mg·m⁻³ for coal burning, oil burning, and gas-fired boilers, respectively, based on the emission standards of air pollutants in China (GB13271-2014). The three limit values were 200, 200, and 150 mg·m⁻³ for NO_x. The limit values of SO₂ and NO_x in the plumes were approximately three to four orders of magnitude higher than their concentrations in the ambient air in northern China [8,15,18,42]. Thus, assuming that the production rates of particulate chemicals via in-fresh-stack-plume non-precipitation clouds are three to four orders of magnitude higher than those of ambient air, it can be roughly estimated that 1 min of processing would generate aerosols equal to that generated via ambient FSPM continuing for tens of hours to several days, particularly considering the role of NO₂ in chemical reactions [16,17,22]. The lifetime of non-precipitation cloud droplets in fresh stack plumes seems critical to determine the amount of processed aerosols, but the processing needs further evaluation to be included in air quality models.

Based on our observations, the cloud droplets from community house-heating stacks normally survive 20–30 s under wind speeds of 2–4 m s⁻¹ (Supplementary Materials Video S1). The lifetime of the non-precipitation cloud droplets depends on the amount of water vapor released from the stacks and the ambient conditions. High ambient RH, low ambient temperature, and stagnant weather conditions would prolong the lifetime of in-fresh-stack-plume cloud droplets from dozens of seconds to five minutes (Supplementary Materials Videos S2–4, the lifetime was counted manually and from the plume emitted from the outlet to the disappearance), generate extremely high concentrations of PM_2.5 within minutes, and largely increase the PM_2.5/CO ratios through fast conversion of air pollutants therein. However, the lifetime was only approximately 15 s at high wind speed and low RH (Supplementary Materials Video S5).

5. Implications

In the literature, all novel hypotheses to be proposed are based on two key issues: (1) the fast conversion of chemicals was required to generate high concentrations of the PM_2.5 during the extreme events, (2) but no natural cloud was present in the planetary boundary layer and thereby no rapid generation of the PM_2.5 through the cloud processing of aerosols was expected [1,3,5,6,7,16,17,18,19,20,21,22]. The fast conversion was thereby conventionally called the “missing formation mechanism”. The non-precipitation-cloud processing of aerosols, however, always occurs in fresh combustion plumes during the extreme events. In addition, the non-precipitation-cloud processing of aerosols can be largely enhanced under high ambient RH, low ambient temperature, and stagnant weather conditions with the prolonged lifetime of water droplets in fresh plumes. To prevent extreme PM_2.5 pollution events in the future, environment-friendly measures are needed to reduce the water vapor released from stacks in China.