Mitigation of severe urban haze pollution by a precision air pollution control approach

Severe and persistent haze pollution involving fine particulate matter (PM2.5) concentrations reaching unprecedentedly high levels across many cities in China poses a serious threat to human health. Although mandatory temporary cessation of most urban and surrounding emission sources is an effective, but costly, short-term measure to abate air pollution, development of long-term crisis response measures remains a challenge, especially for curbing severe urban haze events on a regular basis. Here we introduce and evaluate a novel precision air pollution control approach (PAPCA) to mitigate severe urban haze events. The approach involves combining predictions of high PM2.5 concentrations, with a hybrid trajectory-receptor model and a comprehensive 3-D atmospheric model, to pinpoint the origins of emissions leading to such events and to optimize emission controls. Results of the PAPCA application to five severe haze episodes in major urban areas in China suggest that this strategy has the potential to significantly mitigate severe urban haze by decreasing PM2.5 peak concentrations by more than 60% from above 300 μg m−3 to below 100 μg m−3, while requiring ~30% to 70% less emission controls as compared to complete emission reductions. The PAPCA strategy has the potential to tackle effectively severe urban haze pollution events with economic efficiency.

For observations of PM 2.5 chemical composition, Table S1b lists the information about the monitoring stations from which the observational chemical composition of PM 2.5 is used to evaluate the model performance for each study case. For the Beijing case from Oct 27 to Nov 3, 2013, an Aerodyne high resolution time-of-flight aerosol mass spectrometer was used to measure the chemical compositions of PM 2.5 and black carbon was measured by a single-wavelength (670 nm) Thermo multiangle absorption photometer 6 . For the Xian case from Dec 15 to 28, 2013, the sulfate, nitrate, ammonium, and organic aerosols are measured by the Aerodyne High Resolution Time-of-Flight Aerosol Mass Spectrometer (HR-ToF-AMS) with a novel PM 2.5 lens at the Institute of Earth Environment, Chinese Academy of Sciences (34.23_ N, 108.88_ E) in Xi'an, China 1 . For the Shanghai and Hangzhou cases, daily chemical composition of PM 2.5 at Lian, Taiyangshan and Zhengzhou stations was obtained from the Chinese Meteorological Administration (CMA) Atmospheric Watch Network (CAWNET) 2 . Note that the observations of chemical composition of PM 2.5 at Zhengzhou and Gaolanshan stations from CAWNET are also used for the model evaluation for Beijing and Xian cases, respectively, as shown in Table S1b.

48-h air mass back trajectories and their cluster analyses.
To locate possible regional transport pathways of air masses and evaluate the relative contributions by long range transport, 48-h back trajectories starting at the arrival level of 100 m from the monitoring sites were calculated with the NOAA HYSPLIT model (http://ready.arl.noaa.gov/HYSPLIT.php) for each period studied. The back trajectories were calculated eight times per day at starting times of 00:00, 03:00, 6:00, 09:00, 12:00, 15:00,18:00 and 21:00 UTC. To be consistent with the WRF-CMAQ model simulations, the same WRF meteorological fields are used to calculate the back trajectories. The trajectory cluster analysis for each case was performed with the clustering option of Euclidean distance 10,33,36 . Figs S2-S5 show the results of the trajectory cluster analyses and the 48 h back trajectories in different periods on the basis of observed PM 2.5 concentration intervals (6 intervals: the entire period,75 g m -3 ≤PM 2.5 <115 g m -3 , 115 g m -3 ≤PM 2.5 <150 g m -3 , 150 g m -3 ≤PM 2.5 <250 g m -3 , PM 2.5 ≥250 g m -3 and PM 2.5 ≥150 g m -3 ) for the five heavy haze episodes in Beijing, Shanghai, Hangzhou and Xian. To discuss differences in the distributions of backward trajectory clusters arriving at the receptor sites in the vertical direction, the pressure profiles of the trajectory clusters for each case are also shown in Figs. S2a(g)-S5g. Table S8 summarizes the results of mean PM 2.5 concentrations, and percentages of trajectories for each trajectory cluster for five cases.
The corresponding values of pressures and heights for each trajectory cluster at 48-h earlier before arriving at the receptor sites are also summarized in Table S8.
For the Beijing case in the forecast simulations from Jan 24-26, 2017, Fig. S2a shows that three clusters for all data during the entire period were determined by the cluster analysis algorithm: one short distance transport pathway: SW (Southwest), and two long distance transport pathways: NW (Northwest) and W (West). Figs. S2a(d), S2a(e) and S2a(f ) indicate that most of the 48 h back trajectories for the heavy haze periods with the PM 2.5 ≥ 150 g m -3 , which mainly belong to SW clusters, originated from the southwest of Beijing and brought the dirty air masses to Beijing by passing through the industrialized cities such as Baoding, Langfang, and Shijiazhuang. The vertical distributions of the trajectory clusters in  For the Shanghai case, Fig. S3a shows that three clusters for all the data during the entire period were determined by the cluster analysis algorithm: one long distance transport pathway: NW (Northwest), and two short distance transport pathways: NW-S (Northwest-South) and Evaluation of WRF-CMAQ model performance for the five severe haze episodes. In parallel with the hourly observations, concurrent hourly predicted concentrations at the monitoring sites in the city were averaged. The averaged observed and predicted concentrations are compared to evaluate the model performance for each haze episode, as shown in Figs. S7-S10 for PM 2.5 at the related cities for each severe haze episodes. Fig. S6 shows the model simulations for PM 2.5 concentrations with observed data overlaid (circles) at 18:00 (local time), 20:00, 21:00 and 22:00 on December 2, 2013. As can be seen, the model captures the spatial pattern of most of observations reasonably well for this severe haze episode. Figs. S7-S10 show time-series comparisons of mean observed and predicted PM 2.5 concentrations for each city for the five severe urban haze cases. Time-series comparisons of observations and simulations for PM 2.5 , PM 10 , O 3 , SO 2 , NO 2 , and CO in Beijing, Shanghai, Hangzhou and Xian are presented in Figs. S11, S12, S13 and S14a, respectively. Model performance in terms of normalized mean bias (NMB) values for PM 2.5 , O 3 , SO 2 , NO 2 , and CO for each city and each study case is summarized in Tables 2-5.
For the Beijing case in the forecast simulations from Jan 24-26, 2017, Fig. S7a shows that the model captured the temporal variations of mean PM 2.5 concentrations at all related cities very well. The NMB values for PM 2.5 range from -4.5% at Tangshan to -31.2% at Tianjin at all related cities (see Table S2a). The NMB values for SO 2 are within ±30% at all related cities except Beijing, Langfang and Tianjin where the NMB values for SO 2 are 54.7, 62.9 and 65.4%, respectively (see Table S2a). The NMB values for NO 2 are within ±20% at all related cities (see Table S2a). The NMB values for CO are within ±31% at all related cities except Qinhuangdao where the NMB value for CO is -36.2% (see Table S2a). On the other hand, for the Beijing case in the retrospective simulations from Oct 27-Nov 3, 2103, Fig. S7b shows that the model captured the temporal variations of mean PM 2.5 concentrations at all related cities very well except Tangshan city, for which there is consistent overestimation of PM 2.5 . The NMB values for PM 2.5 range from 0.2% at Handan to -19.5% at Qinhuangdao at all related cities except Tangshan where the NMB value for PM 2.5 is 45.8% (see Table S2c).
The NMB values for O 3 are within ±20% at all related cities except Langfang and Qinhuangdao for which the NMB values for O 3 are 42.3% and 42.9%, respectively (see Table   S2c). The NMB values for SO 2 are within ±20% at all related cities except Hengshui and Qinhuangdao where the NMB values for SO 2 are 36.2% and 50.2%, respectively (see Table   S2c). The NMB values for NO 2 are within ±30% at all related cities except Chengde and Qinhuangdao where the NMB values for NO 2 are -49.3% and -37.4%, respectively (see Table   S2c).The NMB values for CO are within ±21% at all related cities except Qinhuangdao where the NMB value for CO is -33.8% (see Table S2c).
For the Shanghai case, Fig. S8 shows that the model captured well the temporal variations of mean PM 2.5 concentrations at all related cities. The NMB values for PM 2.5 range from5.6% at Changzhou to -26.2% at Ningbo (see Table S3). The NMB values for O 3 are within ±40% at all related cities except Huzhou, Jiaxing and Ningbo where the NMB values for O 3 are 54.9%, 49.7% and 45.6%, respectively (see Table S3). The NMB values for SO 2 are within ±40% at all related cities except Nanjing and Suzhou where the NMB values for SO 2 are 51.5% and 56.0%, respectively (see Table S3). The NMB values for NO 2 are within ±33% at all related cities except Huzhou and Ningbo where the NMB values for NO 2 are -41.8% and -38.8%, respectively (see Table S3).The NMB values for CO are within ±35% at all related cities except Nantong and Suzhou where the NMB values for CO are -47.3% and -47.5%, respectively (see Table S3).
For the Hangzhou case, Fig. S9 shows that the model captured the temporal variations of mean PM 2.5 concentrations at all related cities very well. The NMB values for PM 2.5 are within ±25% except for Huaian, Huzhou, and Lianyungang where the NMB values for PM 2.5 are -35.9%, -37.1% and -45.8%, respectively (see Table S4).The NMB values for O 3 are within ±45% at all related cities (see Table S4). The NMB values for SO 2 are within ±35% except for Huaian, Lianyungang and Yangzhou where the NMB values for SO 2 are -48.7%, -48.6% and -46.1%, respectively (see Table S4). The NMB values for NO 2 are within ±35% except for Lianyungang where the NMB value for NO 2 are -44.0% (see Table S4).The NMB values for CO are within ±30% at all related cities except Changzhou and Huaian where the NMB values for CO are -38.4% and -55.5%, respectively (see Table S4).
For the Xian case, Fig. S10 shows that the model captured well the temporal variations of mean PM 2.5 concentrations at all related cities except for Tangchuan and Baoji where the model did not capture the peaks of observed PM 2.5 concentrations. The NMB values for PM 2.5 range from -11.1% at Xian to -37.1% at Tongchuan (see Table S5). The NMB values for O 3 are within ±33% at all related cities except Baoji and Weinan where the NMB values for O 3 are 50.5% and 57.4%, respectively (see Table S5). The NMB values for SO 2 are within ±26% at all related cities except Xianyang and Yanan where the NMB values for SO 2 are 62.5% and -57.6%, respectively (see Table S5). The NMB values for NO 2 are within ±15% at all related cities (see Table S5).The NMB values for CO are within ±34% at all related cities (see Table S5).
Model performances for PM 2.5 chemical composition on the basis of available measurements for the Beijing, Shanghai, Hangzhou and Xian cases in the retrospective simulations are summarized in Tables 6a, 6b, 6c and 6d, respectively. The temporal variations of comparisons of predictions and observations for each PM 2.5 component are shown in Figs. S15-S18.
As  Table S6a). The model simulations underestimate OC at all sites and cases except the Shanghai case at Taiyangshan site where the model simulations slightly overestimate observed OC by 13.3% (see Table   S6b), while the model simulations underestimate NO 3 at all sites and cases except the Beijing case at the Beijing site and Xian case at Xian site where the model simulations overestimate   3d).  Table S9 are the contributions of all, agriculture, industrials, power plants, residential, and transportation sectors to mean PM 2.5 in

Contributions of different emission sectors over the
Beijing. As can be seen, the mean contributions of the all, agriculture, industrials, power plants, residential, and transportation sectors to average PM 2.5 concentrations in Beijing were estimated at 91.5, 5.6, 28.71, 3.0, 40.9, and 5.5%, respectively. The results in Table S9 also show that the transport from outside of the Beijing-Tianjin-Hebei region contributed 8.5% to average PM 2.5 concentrations in Beijing.

Comparison of the results for the Beijing case in 2013 retrospective simulations and in
2017 forecast simulations. Table S9 shows that the mean reduction percentages of PM 2.5 for the Beijing case in 2013 retrospective simulations in Cases 1, 3, and 5 were 33.0%, 20.7%               Table 1.  (2013). The vertical and horizontal lines represent the ranges of the PM 2.5 reduction percentages and the CWT intervals, respectively. The points represent the mean values for each case. The right panels show the results for the cases 1, 3, 5, and 7 in Table 1.