Vapor isotopic evidence for the worsening of winter air quality by anthropogenic combustion-derived water.

Anthropogenic combustion-derived water (CDW) may accumulate in an airshed due to stagnant air, which may further enhance the formation of secondary aerosols and worsen air quality. Here we collected three-winter-season, hourly resolution, water-vapor stable H and O isotope compositions together with atmospheric physical and chemical data from the city of Xi'an, located in the Guanzhong Basin (GZB) in northwestern China, to elucidate the role of CDW in particulate pollution. Based on our experimentally determined water vapor isotope composition of the CDW for individual and weighted fuels in the basin, we found that CDW constitutes 6.2% of the atmospheric moisture on average and its fraction is positively correlated with [PM2.5] (concentration of particulate matter with an aerodynamic diameter less than 2.5 μm) as well as relative humidity during the periods of rising [PM2.5]. Our modeling results showed that CDW added additional average 4.6 μg m-3 PM2.5 during severely polluted conditions in the GZB, which corresponded to an average 5.1% of local anthropogenic [PM2.5] (average at ∼91.0 μg m-3). Our result is consistent with the proposed positive feedback between the relative humidity and a moisture sensitive air-pollution condition, alerting to the nontrivial role of CDW when considering change of energy structure such as a massive coal-to-gas switch in household heating in winter.

An estimated 3 million people are killed each year owing to outdoor air pollution (1). Overall mean mortality rate increases ∼1.2% with each 10 μg m−3 increase in [PM2.5] (concentration of particulate matter with an aerodynamic diameter less than 2.5 μm) (2). Considering a close relationship between air pollution and energy structure (3, 4), countries facing severe air pollution have been adjusting their energy structure to improve air quality. In the past several years, China has invested heavily in reducing air pollution in major cities (5), and there has been a significant decrease in annual [PM2.5] since 2013 (6). Despite many drastic efforts, haze events, correlated with high [PM2.5], still occur frequently, especially in cities on the North China Plain (7, 8). In the heavily polluted Beijing-Tianjin-Hebei region, a series of regulatory policies has been implemented (5, 9), including using natural gas instead of coal (10, 11). Since 2015, a large-scale project of coal-to-gas switch has been deployed in urban and rural areas in China (12, 13).

It has been proposed that in northern China, severe haze is the synergetic effect of the interactions between anthropogenic emissions and atmospheric processes (7). Among the many causes of haze events, atmospheric water vapor or specifically relative humidity (RH) enhances the rate of heterogeneous oxidation of SO2 and NOX and in turn exerts a positive feedback on the rising of [PM2.5] (14–17). Water vapor in the planetary boundary layer (PBL) comes mostly from the oceans via evaporation and transport or from continental water via evapotranspiration (18, 19). Combustion-derived water (CDW), a source of water vapor coming from fossil fuel or biomass burning, is negligible in the global atmospheric water budget. However, Gorski et al. (20) reported up to 13% of the water vapor in PBL was from CDW during certain days in Salt Lake City, Utah, located in an enclosed basin in northern America. This is a significant water contribution and should be verified independently in similar urban environments. The increased CDW fraction in air moisture could simply be a passive result of multiday accumulation in a polluted/stagnant PBL. Or, the CDW added to an airshed during polluted days could accelerate the formation of secondary aerosols, further stabilize the PBL, and reinforce CDW accumulation (15, 20). Pinning the exact role of CDW is important to energy policy in mitigating air pollution in China and other developing nations. However, the nonlinearity of atmospheric processes renders any firm conclusion hard to come by. Here we report the results of a 3-y, multiparameter sampling campaign in combination with atmospheric chemical model to examine the role of CDW.

Results and Discussion

CDW Isotopic Compositions and Fraction in Total Air Moisture.

To examine the impact of CDW on air pollution, we must first quantify the fraction of CDW in an airshed. CDW has a set of characteristic δD and δ18O values that are locally specific and can be estimated based on energy inventory in a region. Here we conducted a winter-season atmospheric physical and chemical observational campaign from the year 2016–2018 in Xi’an, a densely populated city in the Guanzhong Basin (GZB) in northwestern China (). The GZB is surrounded by the Loess Plateau to the north and Qinling Mountains to the south with a narrow opening to the east, a topography resulting in frequently stagnant air and heavy haze events in winter. Near-surface atmospheric water vapor δD and δ18O values were continuously measured and compiled with rainy days excluded. End-member δD and δ18O of fossil fuels such as coal, natural gas, and gasoline were individually determined experimentally (), and a weighted end-member isotope composition for CDW was obtained by using energy inventory in Xi’an Municipal Bureau of Statistics (21). A statistical analysis of the interrelationships among the fraction of CDW in total air moisture, [PM2.5] range, RH, SO2 concentrations ([SO2]) range, and NO2 concentrations ([NO2]) range were conducted. A 19-d heavy haze event from Dec. 27, 2015 to Jan. 15, 2016 was then modeled using the Weather Research and Forecasting (WRF)-Chem model to examine the impact of SO2, NO2, and the added CDW on PM2.5 formation.

Time-series data revealed that atmospheric RH was higher than nonhaze periods during all of the 40 observed heavy haze events when the [PM2.5] were higher than 75 μg m−3 for more than 24 h (Fig. 1). The more positive δ18O values (up to −15‰) for water vapor occurred during the periods of heavy haze events (). Overall, the water vapor d-excessvap (defined as d-excessvap= δ2Hvap -8*δ18Ovap) in Xi’an became less positive with a minimum value of 3‰ at higher [PM2.5] (Fig. 1 and ). By measuring the δD and δ18O of CDW for individual fuels of different exhaust types, we obtained a weighted δD and δ18O values for CDW in Xi’an (Fig. 2) at −134.4‰ and 9.0‰, respectively ().

Fig. 1.

Time-series variations in Xi’an during 2016–2018 heating seasons: [PM2.5] (A); relative humidity (B); humidity (C); δ18Ovap (vapor δ18O) (D); d-excessvap (vapor d-excess) (E); calculated fractions for CDW (F).

Fig. 2.

Surface air vapor isotopic composition (δ2Hvap and δ18Ovap) and different fossil fuel CDW vapor isotopic composition in Xi’an. BGL represents the background line of vapor isotopic composition calculated according to a Gaussian mixing model. The circles of three different colors represent different ranges of [PM2.5]. The solid green triangle is Xi’an’s weighted isotope composition of CDW calculated according to energy inventory. The dashed lines represent CDW fraction in total moisture. Uncertainties of the data are marked or smaller than the symbol sizes.

We consider atmospheric water vapor as the sum of natural water (NW) and CDW. Recent studies have demonstrated that a high d-excessvap is generally observed in airshed with good dispersion condition (20, 22, 23) and low RH (24, 25). Thus, the δD and δ18O of NW from such atmospheric conditions define a background line in δD-δ18O space for those of polluted days to compare to. This background line was determined using an expectation-maximization algorithm for Gaussian mixture models which is known to be effective in separating individual components from a multicomponent mixture (26) (). To calculate the CDW fraction, we use the two-endmember mixing equation (20):where dvap is the observed water vapor d-excess (point C in ), dCDW is Xi’an’s weighted CDW d-excess (point B in ), dNW is d-excess of NW (point A in , the intersection of the extended BC line with the background line), and X is the fraction of CDW (the length ratio of line AC over line AB, ). For those points to the left side of the background line, we consider their CDW fraction as zero. The position of the background line determines the CDW fraction. There are uncertainties, including condensation due to adiabatic lifting (27), kinetic effects due to evaporation (28), and mixing of air masses (29), that are associated with the determination of the background line. We compared three independent methods: 1) using an optimized smooth function on observed time-series d-excess data (20), 2) using measured d-excess data prior to the rising of air pollutants (23, 27), and 3) using an expectation-maximization algorithm for Gaussian mixture models in this study. We found that the Gaussian mixture models gave CDW fractions higher than those in Gorski et al. (20), but nearly identical to those in Fiorella et al. (23, 27) (). The average contribution of CDW is 6.2% in Xi’an, and the highest fraction is found to be 16.2% of the total surface air moisture.

The Xi’an water vapor isotope data display a similar δD-δ18O pattern as that observed in Salt Lake City (20). This may be attributed to the similar enclosed topography of the two sites, where CDW is kept under often stagnant meteorological conditions. Occurrences of winter haze in Xi’an provide us an opportunity to explore the relationship between CDW fraction and its potential effect on secondary PM2.5 formation by using water vapor isotopes.

Relations among [SO2], [NO2], RH, CDW, and [PM2.5].

We plotted the ranges of [SO2] and [NO2] with those of RH and [PM2.5] for the three heating seasons (Fig. 3 and ). The data show that when the [SO2] are less than 15 μg m−3 and RH is lower than 60%, the [PM2.5] remains below 35 μg m−3. This is also the case as long as the [NO2] are less than 20 μg m−3. Severe haze events with [PM2.5] higher than 150 μg m−3 occur only when RH is higher than 40% while in the same time [SO2] are higher than 30 μg m−3 or [NO2] higher than 60 μg m−3. The higher the RH is, the higher is the CDW fraction in total moisture during periods of [PM2.5] rising (). Overall, the CDW fraction increases from an average of 5.5% to 6.5% when the RH increases from 20–40% to 60–80%, respectively, which also means that in absolute quantity, the increase of CDW is significant in air with high RH (Fig. 3).

Fig. 3.

The [PM2.5] at different RH and SO2 (concentration) ranges (A) and at different RH and NO2 (concentration) ranges (B); the bar graphs in the top panel represent percentage contributions of NW and CDW. The blue color shades represent the SO2 and NO2 ranges from 0 to above 40 μg m−3 and from 0 to above 80 μg m−3, respectively. The hollow squares inside the vertically elongated boxes are the average [PM2.5]; the center line within a box is the median of the [PM2.5] dataset; the upper and lower edges of the box are the 25% and 75% of the [PM2.5] dataset; the ends of the lines extending from the interquartile range (IQR) represent the extreme values within 1.5× the IQR; the cross and the minus signs represent outliers that are at a greater distance from the median than 1.5× the IQR (see for statistical test).

The rapid increase of [PM2.5] is controlled by many factors, such as the seasonally enhanced emissions of primary pollutants from residential heating in winter (e.g., coal combustion, natural gas burning, and biomass burning) (30), the fast growth of secondary aerosols (31, 32), and unfavorable meteorological conditions (15, 33). For the formation of secondary aerosols, previous studies have noticed the enhanced reaction rates at high RH level (32, 34–36). Here, considering the contribution of CDW to RH, we used the WRF-Chem model to evaluate CDW’s role in PM2.5 formation during a persistent and heavy haze episode in the GZB ().

Modeling.

The WRF-Chem model performed well in simulating particulate pollution in China (17, 37, 38), and reasonably reproduced the observed temporal variations and spatial distributions of [PM2.5], [O3], [NO2], and [SO2] in the GZB and Xi’an city () during the simulation period in the present study. The results show that the addition of CDW generally increases [PM2.5] and can be up to 16.6 μg m−3 or an additional 8% in the GZB (Fig. 4 ), and 20.0 μg m−3 or 10% in Xi’an airshed (). On average, CDW promoted additional PM2.5 formation by 2.8% (equivalent to 4.6 μg m−3) in the GZB and 3.4% (equivalent to 6.0 μg m−3) in Xi’an and its surrounding areas during the 19-d period of heavy particulate pollution when the average [PM2.5] were at 164.1 μg m−3 in the GZB and 177.3 μg m−3 in Xi’an. Approximately 86% of the enhanced PM2.5 from CDW is contributed by secondary aerosols, in which the contributions from secondary organic and inorganic aerosols are 44% and 42%, respectively. The rest of the PM2.5 enhancement is mainly contributed by the aerosol-radiation feedback induced by an increase in aerosol liquid water.

Fig. 4.

Effects of CDW on near-surface [PM2.5] in the GZB. (A and B) Diurnal profiles of average [PM2.5] increase in GZB from 00:00 (Beijing time, BJT) Dec. 27, 2015 to 00:00 BJT Jan. 15, 2016 in μg/m3 or in %; (C and D) Spatial distribution of average [PM2.5] increase during the simulation period in μg m3 or in %. The outer black line outlines the location of GZB, and the inner black line outlines the urban area of Xi’an.

The severe and persistent particulate pollution in the GZB could be attributed to a synergy of massive anthropogenic emissions and unfavorable synoptic conditions related to the topography (7, 39–41). We further conducted a sensitivity simulation considering only the local anthropogenic emissions in the GZB in the emission inventory (mainly including agriculture, industry, power, residential, and transportation sources) used in the WRF-Chem model to evaluate the contribution of CDW to the PM2.5 level from local anthropogenic emissions. On average the CDW-enhanced [PM2.5] accounts for about 5.1% of the average [PM2.5] (91.0 μg m−3) and up to 18.2% from the local anthropogenic emissions in the GZB. Considering that [PM2.5] control target for the fall and winter of 2019 (a 6-mo period) in Xi’an was a reduction of 2 μg m−3, this additional increase of [PM2.5] would likely counteract many of the implemented [PM2.5] reduction efforts.

To quantify the effectiveness of curtailing the formation of secondary aerosols by reducing NO2 or SO2 emissions or CDW, we have conducted a sensitivity test using a typical polluted day as our base case ([SO2] = 47.9, [NO2] = 70.8, [PM2.5] = 164.1, all in μg m−3, and RH = 58.9%) (). When both RH and [SO2] are reduced to 53.1% and 10.3 μg m−3, respectively, the [PM2.5] decreases by 8.0 μg m−3. When both RH and [NO2] are dropped to 53.1% and 19.7 μg m−3, the [PM2.5] increases, however, by 0.9 μg m−3 comparing with the RH reduction-only scenario. When both RH, [SO2], and [NO2] are dropped to 53.1%, 8.9 μg m−3, and 19.6 μg m−3, respectively, the [PM2.5] decreases by 20.2 μg m−3. Note that the model predicts a 4.6 μg m−3 reduction of [PM2.5] in the GZB if we subtract from the model the CDW contribution which is on average ∼10% of the overall RH during the 19-d period in early 2016 (). This level of PM2.5 reduction by eliminating CDW is not overwhelming but still significant when comparing to the effort that would be required to achieve the same reduction by reducing [SO2] and/or [NO2].

The positive feedback mechanism between RH and PM2.5 growth rate has been proposed previously (15, 42). Briefly, the RH increases and induces the formation of secondary particulate matter through aqueous reactions in a shallow PBL condition. This in turn enhances the formation of particulate matter and results in solar dimming, which then limits the PBL height. Wu et al. (17) estimated that the average contribution of the water vapor (including natural and anthropogenic factors) to the near-surface increased [PM2.5] is 17.5%, indicating its importance. Using water vapor isotope data, we have further differentiated the PM2.5 contribution between anthropogenic factor and natural water vapor effect. Our results suggest that CDW promotes additional PM2.5 formation, being consistent with the previously proposed positive-feedback hypothesis. A long-term observational dataset in combination with modeling is required to further quantify CDW’s positive-feedback effect on secondary aerosol formation.

Policy Implications.

The Chinese government issued a series of counterpollution measures in which the “coal-to-gas” campaign was widely implemented in household heating in the North China Plain, aiming at improving the air quality (43, 44). The natural gas combustion can produce 3× more water vapor than coal burning. This number is based on mean calorific capacities of coal and natural gas and therefore the quantity needed to generate 1 megawatt-hour equivalent energy (). As demonstrated by our study, CDW in urban PBL could promote additional formation of secondary aerosols and therefore exacerbate air quality. Thus, while the enhancing effect of CDW on PM2.5 formation needs to be further quantified in additional urban areas, caution is warranted in implementing the coal-to-gas switch for household in North China Plain and other areas.

Materials and Methods

A Cavity Ring-Down Spectroscopy analyzer (model Picarro-L2130i, Picarro, Inc.) was used for δD and δ18O measurement. A recommended dual mode (45) was followed. The humidity effect was corrected by a humidity-isotope calibration response function (46). The data were calibrated to the VSMOW-GISP scale with six known standards runs at regular intervals (47, 48). Large chambers up to the size of a shipping container were used to experimentally determine the δD and δ18O of CDW generated by industry-scaled natural gas and coal burning. Those from diverse car exhausts were measured via a pipe on a transmission device at Xi’an Automobile Testing Center. Weighted isotopic compositions were calculated using energy inventory obtained from Xi’an Municipal Bureau of Statistics (21). To obtain a set of background of d-excessvap values, we followed Steen-Larsen et al. (28) and performed a cluster analysis using an expectation-maximization algorithm for Gaussian mixture models (26). The results displayed a distinct cluster with a mean d-excessvap of 19.7‰ and an SD (σ) of 4.38‰, and a second cluster centered around 28‰ with σ = 8.35‰. We arbitrarily defined a high d-excessvap value being +3σ higher than the mean d-excessvap value of the main cluster, i.e., at 32.8‰. A background δ18Ovap −δ2Hvap line with d-excessvap greater than 32.8‰ was constructed by linear regression.

Hourly concentration data of atmospheric gases, such as SO2, NO2, CO, O3, and PM2.5, and the meteorological data including temperature, RH, and amount of precipitation in Xi’an were downloaded from government websites of https://www.aqistudy.cn/ and http://www.weather.com.cn/, respectively. In determining the impact of [SO2], [NO2], and RH on [PM2.5], we compared the significant differences of [PM2.5] with the [SO2] and [NO2] categories at different RH ranges by using one-way ANOVA tests at P = 0.01 level ().

The WRF-Chem model modified by Li et al. (49–52) was used to simulate a 19-d persistent and heavy haze episode from late December 2015 to January 2016. The WRF-Chem model adopts one grid with a horizontal resolution of 6 km centered at 34.25°N and 109.0°E (), and 35 sigma vertical levels with a stretched vertical grid with spacing ranging from 30 m near the surface, to 500 m at 2.5 km and 1 km above 14 km, and the grid cells used for the domain are 150 × 150. The physical parametrizations employed in the model are listed in . The contribution of CDW to the [PM2.5] was quantitatively determined by comparing the model-simulated 19-d case (which broadly fits the observation) with a corresponding case in which the CDW input was arbitrarily set at zero in the GZB.