Emergency Response Measures to Alleviate a Severe Haze Pollution Event in Northern China during December 2015: Assessment of Effectiveness

Using the WRF-Chem model, we simulated the surface PM2.5 concentrations on the North China Plain (NCP) during a severe winter haze episode (December 6–10, 2015) with the goal of assessing the effectiveness of the implemented emergency response measures (ERMs) in alleviating the pollution. We estimated that the ERMs decreased the anthropogenic pollutant emissions, with the exception of NH3, by 8–48% during this event. Inputting these reduced emission estimates, our simulations reproduced the observed PM2.5 concentrations and compositions. Stagnant regional meteorological conditions increased the lifetime of the PM2.5 in the NCP boundary layer from 1 day during the clean period to 5 days during the haze episode. Additionally, local emissions accounted for approximately only 20% of the surface PM2.5 in Beijing but more than 62% over the rest of the NCP. We found that the ERMs achieved a modest reduction in the mean surface PM2.5 concentrations during the event, decreasing them by 7% and 4% in Beijing and across the rest of the NCP, respectively. The limited effect was due to the duration of the ERMs being much shorter than the lifetime of the PM2.5, which prevented the concentrations of the latter from fully reflecting the reduction in emissions. We conclude that anthropogenic emissions on the NCP during severe winter haze episodes must be reduced by a much larger percentage to substantially abate the PM2.5 concentrations.


INTRODUCTION
The region of the North China Plain (hereafter referred to as the "NCP", as shown in Fig. 1(b)), including the Beijing-Tianjin-Hebei (BTH) area and the surrounding provinces of Shandong and Henan, has been experiencing severe winter haze pollution events of hourly PM 2.5 concentrations exceeding 150 µg m -3 in recent years (Dang and Liao, 2019), which pose threats to public health (e.g., Chen et al., 2013). These severe wintertime PM 2.5 pollution events are typically associated with the accumulation of PM 2.5 and its precursors under stagnant weather conditions. Such stagnant conditions prevent the horizontal and vertical ventilation of pollutants and are often associated with high humidity near the surface, which in turn promotes secondary PM 2.5 production (Jeong and Park, 2013;Tie et al., 2017;Leung et al., 2018;. The NCP is blocked by mountains to the north and to the west, which also contributes to the accumulation of PM 2.5 and its precursors under calm or southerly wind conditions . These severe winter haze events are terminated by the passage of a cold front, which ventilates the region with strong wind and clean air from the north . In September 2013, the State Council of China promulgated the Air Pollution Prevention and Control Action Plan (hereafter referred to as the "Action Plan"; State Council of the People's Republic of China, 2013), which outlined the policies on reducing nationwide anthropogenic emissions, as well as set specific improvement targets for the annual mean PM 2.5 concentrations for key areas, including Beijing and the BTH area, by 2017. Between 2013 and 2018, the annual mean PM 2.5 concentrations for most Chinese cities have dropped (Ministry of Environmental Protection of the People's Republic of China, 2014China, , 2015China, , 2016China, , 2017; Ministry of Ecology and Environment of the People's Republic of China, 2018China, , 2019, likely in part due to the nationwide emission reduction. The annual mean PM 2.5 concentrations in Beijing and in BTH in 2017 were reduced to 58 µg m -3 and 64 µg m -3 (Ministry of Ecology and Environment of the People's Republic of China, 2018), respectively, both meeting the improvement targets set by the Action Plan. However, analyses of observations showed that the frequency and intensity of wintertime haze events over the NCP have not shown significant decline since 2013 Dang and Liao, 2019).
To alleviate the severe haze events, the Action Plan mandated that local environmental protection bureaus and meteorological bureaus join forces to establish a protocol for the issuance and execution of emergency response measures (ERMs). For the NCP area, the Monitoring and Warning Scheme for Heavy Pollution Weather in Beijing, Tianjin, Hebei, and Its Surrounding Areas was issued in September 2013 to combat severe haze events (Ministry of Environmental Protection of the People's Republic of China, 2013). This scheme categorized the severity of local air pollution into four levels, based on the duration for which the forecasted hourly air quality index (AQI) exceeds 200 (equivalent to hourly PM 2.5 exceeding 150 µg m -3 ). When such severe PM 2.5 pollution events are forecasted, the provincial and municipal governments are to issue emergency alerts at least 24 hours in advance. Each alert level corresponds to a set of ERMs to reduce anthropogenic pollutant emissions (Table S1): (1) Blue alert: hourly PM 2.5 concentrations are forecasted to exceed 150 µg m -3 and to persist for 24 hours. (2) Yellow alert: hourly PM 2.5 concentrations are forecasted to exceed 150 µg m -3 and to persist for 48 hours. (3) Orange alert: hourly PM 2.5 concentrations are forecasted to exceed 150 µg m -3 and to persist for 72 hours. (4) Red alert: hourly PM 2.5 concentrations are forecasted to exceed 150 µg m -3 and to persist for more than 72 hours. On December 5, 2015, at 11:00 UTC (19:00 local time), the Beijing municipal government issued an orange alert for severe haze pollution, with the corresponding ERMs to be initiated at 16:00 UTC on December 6. The city of Tianjin and most cities in Hebei, Henan, and Shandong Provinces issued various alert levels for severe haze pollution, with corresponding ERMs to be initiated between 16:00 UTC on December 5 and 16:00 UTC on December 7 (Fig. S1). The city of Beijing later upgraded its alert level to the first-ever red alert at 10:00 UTC on December 7, with the corresponding stricter ERMs to be initiated at 23:00 UTC on December 7. Tianjin and Hebei also updated their alert levels, with stricter ERMs to be initiated between 23:00 UTC on December 7 and 16:00 UTC on December 8 (Fig. S1). The issuance of the red alert led to the enforcement of stricter emission reduction measures, including the emergency shutdown of more industrial plants and further restrictions of vehicle numbers on the road (Table S1). This was the first time the Chinese government evoked restrictive ERMs based on air quality forecasts to mitigate severe pollution events. Though the alert levels issued in the NCP varied, the ERM-initiation times in > 90% of the cities were within 16 hours of the ERM-initiation times in Beijing. We referred to the emission reductions in two stages according to the ERMinitiation times in Beijing (Fig. S1). Stage I was between 16:00 UTC on December 6 and 23:00 UTC on December 7, during which the orange alert was in effect in Beijing and Henan, and the yellow alert was in effect in Tianjin, Hebei, and Shandong. Stage II was between 23:00 UTC on December 7 and 04:00 UTC on December 10, during which the red alert was in effect in Beijing, the orange alert was in effect in Tianjin, Hebei, and Henan, and the yellow alert was in effect in Shandong.
The Chinese government had previously administered pre-planned emission controls before and during several important events that took place in Beijing to improve the local air quality (Table S2). Such events included the Sino-African summit in November 2006 (Wang et al., 2007;Cheng et al., 2008), the Beijing Olympic Games in August 2008 Schleicher et al., 2012), the Asia-Pacific Economic Cooperation (APEC) summit in November 2014 (Li et al., 2015;Guo et al., 2016a;Zhang et al., 2016;, and the Chinese Military Parade in September 2015 Huang et al., 2018). The emission control actions taken included temporary closures of factories, restrictions on the numbers and types of on-road motor vehicles, and restrictions on construction. The control actions were enforced in Beijing and its surrounding cities and provinces, including Tianjin, Hebei, Shandong, Shanxi, Henan and the Inner Mongolia Autonomous Region. The resulting reductions in primary pollutant emissions were estimated to be 40-70% in Beijing and 30-70% in the surrounding cities and provinces (Wang et al., 2007;. Previous studies estimated that these emission control actions may have led to dramatic declines in SO 2 , NO x , and PM 2.5 concentrations in Beijing and its surrounding areas by 30-70% during the APEC summit Zhang et al., 2016) and by 50% during the Chinese Military Parade . However, these pre-planned emission controls were enforced over much larger spatial areas, often lasted for weeks, and were in most cases stricter than the ERMs implemented from December 6 to 10, 2015. More importantly, those previous events all took place either in summer or in fall, when the regional meteorological conditions were more favorable to the dispersion or removal of pollutants, aiding the efficacy of the emission control actions Huang et al., 2018).
In comparison, the short-term ERMs administered during severe wintertime haze events were likely to be less effective, because the stagnant meteorological conditions impeded pollutant dispersion, the controls were less strict, and the enforcement was over a smaller domain and for a shorter period of time. A few studies have attempted to quantify the impacts of emission reduction measures on the PM 2.5 concentrations in the NCP during haze events or haze seasons, but these studies did not distinguish the impacts of China-wide long-term emission controls and the short-term ERMs. Wu et al. (2017) simulated the effects of a hypothetical 30% reduction of monthly anthropogenic emissions in the BTH area during the entire January 2012 and found a 20% decline in the local monthly mean PM 2.5 concentration.  estimated that the overall reduction of anthropogenic emissions since the year 2014 over the NCP led to a 9% decline in the monthly mean PM 2.5 concentration over the NCP in December 2015 relative to December 2014. Chen et al. (2019) simulated the PM 2.5 concentrations in Beijing during four seasonal haze episodes (November 2016versus November 2017, and March 2013versus March 2018. They estimated that the overall reduction of anthropogenic emissions since 2013 led to 33% and 16% decreases of the simulated monthly mean PM 2.5 concentrations in Beijing in November 2017 and March 2018, relative to November 2016 and March 2013, respectively. However, it is likely that the decline in monthly mean PM 2.5 concentrations inferred by these three previous studies were mainly due to the long-term emission control in China and not specifically to the ERMs. Wang et al. (2020) hypothetically simulated the effects of emergency emission reduction on PM 2.5 concentrations during a 5-day severe haze event over the Yangtze River Delta area. They found that a short-term, 20-90% reduction of power, transportation, and industry emissions, comparable to the strictest ERMs, would only lead to a 16% reduction in PM 2.5 concentration. To the best of our knowledge, the effectiveness of the ERMs administered over the NCP area during severe haze events has not been explicitly evaluated.
It is also important to quantitatively assess the relative contributions of local and regional pollutant emissions to the local PM 2.5 concentrations during severe haze events, in order to help municipal policy-makers formulate effective emergency responses.  previously found that 50% of the PM 2.5 in Beijing during January 2013 was due to emissions within Beijing, while the remaining 50% were due to emissions from nearby provinces in northern China. However, in view of the large emissions reductions in all provinces and cities in northern China since 2013, that fraction may have changed significantly and should be reevaluated.
In this study, we explicitly simulated surface PM 2.5 concentrations over the NCP from December 2 to 10, 2015, both with and without the emission reductions associated with the ERMs, to assess the effectiveness of the ERMs on alleviating severe haze pollution. We also quantified the relative contributions of local and regional emissions to PM 2.5 concentrations in Beijing and over the rest of the NCP during this severe haze event. Finally, we analytically interpret the effectiveness/ineffectiveness of the ERMs during this haze event, in order to better inform future emergency response strategies.

WRF-Chem Model
We used the WRF-Chem regional air quality model version 3.6.1 (Grell et al., 2005) to simulate surface PM 2.5 over the NCP from November 26 to December 10, 2015. The first 6 days spun up the model; results for December 2-10, 2015, were analyzed. Fig. 1 shows the two nested domains in our simulations with horizontal resolutions of 81 km and 27 km, respectively. The 27-km horizontal resolution in the inner domain was consistent with the 0.25° resolution of the anthropogenic emission inventory (Section 2.2). The model consisted of 30 vertical layers extending from the surface to 50 hPa, with 7 layers in the bottom 1 km. Meteorological initial and boundary conditions into WRF-Chem were from the NCEP FNL Operational Global Analysis data (Kalnay et al., 1996) and updated every 6 hours. Chemical initial and boundary conditions were from a MOZART global model simulation (Emmons et al., 2010), except we reduced the boundary conditions of dust concentrations by 50% following Georgiou et al. (2018). We nudged the temperature, humidity, and wind in WRF-Chem with hourly surface meteorological measurements and twice-daily rawinsonde profiles over China (Guo et al., 2016b) using four-dimensional data assimilation to reduce errors in the simulated meteorology (Gilliam et al., 2012).
Gas-phase chemistry in WRF-Chem was simulated using the SAPRC-99 mechanism (Carter, 2000), updated to include the photochemistry of dicarbonyls . PM 2.5 in our model included primary elemental carbon aerosol (EC), primary organic aerosol (POA), secondary inorganic aerosol (sulfate, nitrate, and ammonium), secondary organic aerosol (SOA), anthropogenic and natural dust, and sea salt. Aerosol microphysics and gas-particle partitioning were simulated using the MOSAIC module (Zaveri et al., 2008). SOA productions from anthropogenic and biogenic volatile organic precursors were simulated with the VBS module (Lane et al., 2008). We also included the aqueous uptake of glyoxal and methylglyoxal as a source of SOA Li et al., 2013). We optimized the deposition velocity of fog droplets in WRF-Chem to 0.3 cm s -1 to be consistent with the observed fog droplet radius of 2-3 µm over the NCP Seinfeld and Pandis, 2006).
A number of studies have proposed fast, heterogeneous sulfate production pathways, potentially catalyzed by nitrogen dioxide or metal ions, during severe haze events over the NCP Cheng et al., 2016;Wang et al., 2016;Zhao et al., 2017;Hung et al., 2018;Shao et al., 2019). The exact pathway responsible for the high sulfate concentrations during the severe wintertime haze events in the NCP is still uncertain. Nevertheless, we added in our WRF-Chem simulation the sulfate production via the heterogeneous uptake of SO 2 by aqueous aerosols as parameterized by Wang et al. (2016), on the account that this parameterization led to a good simulation of the observed sulfate concentration during our study period (Section 3.1.2). The production rate of sulfate by the heterogeneous oxidation of SO 2 on particles was parameterized as: 2-]/dt| hetero was the rate of heterogeneous production of sulfate (ppmv s -1 ); γ was the effective uptake coefficient of SO 2 on aqueous aerosol surfaces; v̅ was the mean molecular speed of SO 2 (m s -1 ); S c was the aerosol surface area density (m 2 m -3 dry air); [SO 2 ] was the gaseous SO 2 concentrations (ppmv). We used the values γ = 1.6 × 10 -5 for relative humidity (RH) < 41%, γ = 2.1 × 10 -5 for RH between 41% and 56%, and γ = 4.5 × 10 -5 for RH > 56%, as observationally constrained by Wang et al. (2016).

Emissions
Chinese monthly anthropogenic emissions of PM 2.5 precursors and primary PM 2.5 were from the Multi-resolution Emission Inventory for China (MEIC), developed by Tsinghua University (http://meicmodel.org;  for the year 2016 at a horizontal resolution of 0.25°. Anthropogenic source activities included power generation, industries, residential activities, on-road transportation, and agriculture. Activity levels were based on provincial statistics for the year 2016. Anthropogenic emissions for the rest of Asia were from the MIX inventory , developed for the year 2010 at a horizontal resolution of 0.25°. We applied diurnal variations to transportation and residential emissions . Emissions from power generation were injected into the second vertical layer in the model, approximately 100 m above the surface. Daily biomass burning emissions were taken from the Fire INventory from NCAR (FINN; version 1.5; Wiedinmyer et al., 2011) for the study period. Dust emissions were calculated online using the Goddard Global Ozone Chemistry Aerosol Radiation and Transport (GOCART) dust scheme (Ginoux et al., 2001) with the Air Force Weather Agency modifications (GOCART-AFWA; LeGrand et al., 2019). Biogenic emissions were calculated online using the MEGAN module (version 2.04; Guenther et al., 2006).

Emission Reductions Associated with ERMs
We systematically surveyed the government announcements and news reports released from December 6 to 10, 2015, to determine the issued alert levels, the ERMs administered for each of the cities and provinces in the NCP, and the start/end times of ERM implementation (Table S1 and Fig. S1). The emission reductions associated with the ERMs for the industrial, residential, and transportation sectors were estimated based on the guidelines and activity advisories in the Emergency Plan for Heavy Air Pollution issued by each city or province. Table 1 summarized the percentages of sectoral emission reductions as a result of the ERMs for Beijing, Tianjin, Hebei, Henan, and Shandong, respectively, from December 6 to 10, 2015. Table 2 shows the calculated short-term reductions in anthropogenic pollutant emissions for each of the cities and provinces in the NCP from December 6 to 10, 2015, using the percentages in Table 1. As shown in Table 2, anthropogenic emissions were reduced most dramatically in Beijing, due to the enforcement of the strictest ERMs. Anthropogenic emissions of VOCs, SO 2 , and NO x in Beijing during Stage II were reduced by 48%, 34%, and 38%, respectively. Anthropogenic emissions of primary PM 2.5 constituents, such as EC, POA, and anthropogenic dust, were also significantly reduced in Beijing due to enforced restrictions on residential emissions and constructions. Emissions of indicates that the alert level was not issued. b '0' indicates that no emission reductions associated with the sector at the issued alert level. most anthropogenic pollutants in Tianjin, Hebei, and Henan were significantly reduced by 11-27%, but the emission reductions in Shandong were generally less than 10%. Anthropogenic NH 3 emissions were reduced by only 1-9% across the NCP, as NH 3 was mostly emitted by agricultural activities and minimally affected by the ERMs.
Several previous studies have assessed the emission reductions associated with pre-planned emission control actions in Beijing using inverse modeling of observations. For example, satellite-based NO 2 observations indicated that the prohibition of heavy-duty diesel vehicles and the 50% reduction of vehicle numbers on city roads throughout the NCP reduced the NO x emission by 30-47% in the NCP during the Beijing Olympic Games , the APEC summit (Huang et al., 2015), and the Chinese Military Parade . Similarly, satellite-based SO 2 observations indicated that the reduction of industrial production, the temporary closure of power plants, and the prohibition of heavy-duty diesel vehicles on roads reduced the SO 2 emissions in Beijing by 34-46% during the APEC summit and the Chinese Military Parade Zhang et al., 2017). These control actions were similar to the ERMs implemented in Beijing during our haze event, and we also estimated a 38% reduction of NO x emissions and 34% reduction of SO 2 emission in Beijing.

Measurements of Meteorological Conditions and Air Pollutants
Hourly air pollutant concentrations were measured at 239 surface sites across the NCP, operated by the China National Environmental Monitoring Centre (http://www.cnemc.cn). At each site, PM 2.5 mass concentrations were measured using either the micro-oscillating balance method or the βabsorption method (Ministry of Environmental Protection of the People's Republic of China, 2012).
Hourly PM 2.5 concentrations were also continuously measured at the Chinese Research Academy of Environmental Sciences in Beijing (CRAES; 40.04°N, 116.42°E) from December 2 to 10, 2015. The CRAES site is located 8 m above ground in a mixed residential/commercial area with no strong point sources nearby . Approximately 10 km away, daily 24-h filter samples of PM 2.5 were collected at the Peking University Urban Atmosphere Environment Monitoring Station (PKUERS; 39.99°N, 116.31°E), located on the roof of an academic building (Tang et al., 2018). Teflon filter samples were analyzed for inorganic ions using the Dionex ICS-2500 and ICS-2000. Quartz filter samples were analyzed for EC and organic carbon (OC) using a thermal-optical instrument (Model-4; Sunset Laboratory) following the NIOSH protocol (Tang et al., 2018).

Sensitivity Simulations
We conducted six sensitivity simulations to examine the impacts of the ERMs on air quality in the NCP for December 2-10, 2015. Table 3 summarizes the setup of the sensitivity experiments. The ERM experiment was the control experiment, reflecting our best knowledge of the actual event. We used the simulated differences between the NOERM and the ERM experiments to evaluate the effects of the ERMs on PM 2.5 concentrations over the NCP. The simulated differences between the ERM and ERM_NOBJ experiments and the simulated differences between the NOERM and the NOERM_NOBJ experiments indicated the impacts of emissions from Beijing with and without the ERMs, respectively. The simulated differences between the ERM and ERM_NOOTH experiments and the simulated differences between the NOBJ and the NOERM_NOOTH experiments indicated the impacts of emissions from other provinces/cities in the NCP with and without the ERMs, respectively. Fig. 2 shows the observed synoptic meteorological conditions from December 2 to 10, 2015. Between December 2 and 4 (hereafter referred to as the "clean period"), the NCP was affected by a mid-latitude cyclone moving across northeastern China to northern Japan. The passage of the cold front and the intrusion of cold, clean air masses ventilated the boundary layer over the NCP with northwesterly winds, resulting in the lower PM 2.5 concentrations (< 60 µg m -3 ) during this period. During Stage I (from 16:00 UTC on December 6 until 23:00 UTC on December 7), the NCP was under the influence of a stagnant high-pressure system, which led to weak southwesterly winds, compressed boundary layer, and consequently the accumulation of PM 2.5 over the NCP. During Stage II (from 23:00 UTC on December 7 until 04:00 UTC on December 10), the conditions over the NCP continued to be stagnant with high relative humidity, which was not only unfavorable to pollutant dispersion but also conducive to secondary PM 2.5 production. Finally, on December 10, a weak cold front passed through the NCP and ventilated the boundary layer, terminating the severe haze event. We verified that the model correctly simulated the progression of the meteorological conditions associated with the haze event from December 2 to 10, 2015. Fig. S2 shows that our model correctly reproduced the observed changes in the planetary boundary layer heights (PBLHs) around the NCP from > 1 km during the clean period to 0.1-0.8 km during the polluted period. The simulated surface temperatures, relative humidity, wind speeds, and wind directions also agreed well with observations at the Beijing Capital International Airport (BCIA; 40.08°N, 116.59°E; Fig. S3) and at other surface sites in the NCP (not shown).

Simulation of the Severe Haze Pollution Event Meteorological Conditions during the Severe Haze Pollution Event
Observed and Simulated Surface PM 2.5 over the NCP Fig. 3 shows the observed hourly surface PM 2.5 concentrations averaged over sites in Beijing, Tianjin, Hebei, Henan, and Shandong, respectively. From December 2 to 4, the observed PM 2.5 concentrations across the NCP were generally less than 80 µg m -3 . Observed PM 2.5 concentration began to rise on December 5 and continued to do so through Stages I and II, peaking at approximately 300 µg m -3 on December 9 in most cities and provinces in the NCP. The exception was Shandong Province, where the observed PM 2.5 peaked at approximately 200 µg m -3 on December 10. Fig. 3 also shows the simulated surface PM 2.5 concentrations at the five cities and provinces in the NCP using the reduced anthropogenic emissions associated with the ERMs. Our simulation captured the temporal variation of PM 2.5 between December 2 and 10 in Beijing, Tianjin, Hebei, Henan, and Shandong. The correlation coefficients (R) between the observed and simulated hourly PM 2.5 concentrations ranged from 0.80 to 0.95 for the five cities and provinces. The model showed good skill in reproducing the observed PM 2.5 concentrations in Beijing, Hebei, and Henan. The model overestimated the PM 2.5 concentrations during the polluted period in Tianjin and Shandong by 40% and 50%, respectively, likely due to an overestimation of local anthropogenic emissions. Fig. 4(a) compares the spatial distribution of the simulated surface PM 2.5 concentrations in the ERM experiment during Stages I and II against those observed. The spatial distribution of the simulated PM 2.5 concentrations was consistent with that of the observations, with high PM 2.5 concentrations exceeding 200 µg m -3 extending from Henan Province northward to Beijing. In Fig. 4(a), the simulation showed a The error bar represents one standard deviation. The site numbers in each province/city (N), the correlation coefficients (R) between the observed concentrations and those simulated in the ERM experiment, simulated concentration differences between the two experiments, as well as the simulated concentration difference percentages between the two experiments are shown inset. The clean period, Stage I, and Stage II are shaded in blue, orange, and red, respectively. hot spot in Tianjin that was not in the observation, likely indicating an overestimation of industrial emissions in Tianjin. Fig. 5 further compares the simulated PM 2.5 concentrations and compositions from the ERM experiment against measurements at the CRAES and PKUERS sites in Beijing from December 2 to 10, 2015. The simulated PM 2.5 concentrations in the ERM experiment agreed well with the measurements at the CRAES site ( Fig. 5(a)). The model captured the relatively low PM 2.5 concentrations from December 2 to 5, the rise of PM 2.5 concentrations from December 5 to 10, as well as the sharp decline of PM 2.5 concentrations on December 10. The simulated peak PM 2.5 concentration of 308 µg m -3 was also in good agreement with the observation (317 µg m -3 ). Fig. 5(c) showed the observed average daily PM 2.5 compositions at the PKUERS site in Beijing for December 6-10. OA (including primary and secondary OA) was the largest chemical component in PM 2.5 measured at PKUERS, constituting 33% of the total PM 2.5 mass. This was followed by nitrate, sulfate, and ammonium, contributing 18%, 15%, and 11% of the total PM 2.5 mass, respectively. At PKUERS, 19% of the observed total PM 2.5 mass was unidentified, most likely constituted of dust. Fig. 5(d) shows our simulated PM 2.5 compositions in the ERM experiment, which were in good agreement with the measurements. The model also indicated OA as the largest chemical component, followed by nitrate, sulfate, and ammonium. Our simulation indicated that 58% of the simulated PM 2.5 was of secondary origin, consistent with previous measurement studies which found that more than half of the PM 2.5 in Beijing during severe wintertime haze events was secondary (Huang et al., 2014).
To better understand the cause of the severe haze event, we examined the lifetime of PM 2.5 (τ) in the boundary layer of the NCP. For a species in a well-defined reservoir, the mass balance equation is (Jacob, 1999): where m is the mass of the species in the reservoir, dm/dt is the rate of mass change, and ∑sources and ∑sinks are the total sources and total sinks of the species, respectively. The residence time of the species in the reservoir is defined as τ ≡ m/∑sinks (Jacob, 1999). In the case of PM 2.5 in the boundary layer of the NCP: where m was the total PM 2.5 mass in the atmospheric boundary layer of the NCP. The denominator in Eq. (3) was the sum of rates of PM 2.5 mass removal from the NCP boundary layer, which included removal by horizontal ventilation (L h-out ), vertical ventilation (L v-out ; including large-scale vertical advection and convective transport), dry deposition (L drydep ), wet scavenging (L wetscav ), and chemical loss (L chem ). We diagnosed the individual terms in Eq. (3) using the NOERM simulation for the clean and polluted periods, respectively (Table S3). There was no rainfall over the NCP during both periods, so L wetscav was 0. The chemical loss of PM 2.5 (L chem ) has been shown to be two orders of magnitude smaller than the removal rates of ventilation and deposition, especially in winter (Guth et al., 2018;Huang et al., 2019). The simulated mass of PM 2.5 in the boundary layer of the NCP averaged during the clean and polluted periods were 7.3 × 10 16 µg and 1.0 × 10 17 µg, respectively. During the clean period, horizontal ventilation by the northwesterly wind was strong throughout the NCP. By applying the numbers in Table S3 into Eq. (3), we estimated a lifetime of 0.9 days for PM 2.5 in the boundary layer of the NCP. In comparison, the horizontal ventilation of PM 2.5 was much weaker under the stagnant conditions during the polluted period. As such, the lifetime of PM 2.5 in the boundary layer of the NCP during the polluted period (Stages I and II) dramatically lengthened to 4.8 days. This led to significant accumulation of PM 2.5 in the boundary layer of the NCP. Figs. 4(b) and 4(c) show the differences in simulated surface PM 2.5 concentrations between the NOERM and the ERM experiments during Stages I and II, which represent the effects of the ERMs on surface PM 2.5 concentrations. During Stage I (Fig. 4(b)), most areas across the NCP saw only a 2-8% decline in surface PM 2.5 as a result of the ERMs. During Stage II, the decline of PM 2.5 exceeded 8% and was statistically significant (relative to the hourly variability) around Beijing, Tianjin, and the industrial areas of southern Hebei and northern Henan. The decrease of PM 2.5 was much smaller over Shandong, where the reduction in anthropogenic pollutant emissions associated with ERMs was less than 10% (Table 2). Overall, the effects of the ERMs were modest, reducing the mean PM 2.5 over the NCP by 2.5% during Stage I and by 4.2% during Stage II. Fig. 5(a) compares the simulated PM 2.5 concentrations in the NOERM and ERM experiments at the CRAES site in Beijing. The mean abatements in the simulated PM 2.5 concentrations at this site due to the ERMs were 3% and 10% during Stage I and Stage II, respectively. The largest simulated abatement of 11% occurred on December 9, reflecting the cumulative effect of emission reductions up to that time. Fig. 5(b) shows the reduction of individual PM 2.5 components at the CRAES site. The reduction in secondary aerosols accounted for over 53% of the total PM 2.5 decrease during the polluted period, including most importantly SOA (-10 µg m -3 ). Primary OA and EC were reduced by 8 µg m -3 and 5 µg m -3 , respectively. Although there was a significant reduction of anthropogenic dust emissions due to the ERMs, the resulting change in PM 2.5 concentrations attributable to dust was only 7% due to its small share in the total PM 2.5 during the haze event. Figs. 5(c)-5(e) compare the simulated PM 2.5 composition in the ERM and NOERM experiments against the observations at the PKUERS site. Although the implementation of ERMs resulted in varying amounts of concentration decrease in different chemical species (Fig. 5(b)), the overall composition in the ERM and NOERM experiments remained similar, as the total PM 2.5 concentration reduction was small.

Effects of the ERMs on Surface PM2.5 Concentrations over the NCP
We also calculated the population-weighted PM 2.5 (PPM 2.5 ) concentrations from our simulations to evaluate the effect of ERMs on population health. Fig. 6 compares the simulated daily PM 2.5 concentrations and PPM 2.5 concentrations from the NOERM and the ERM experiments for each of the five administrative areas for December 6-10. The mean reductions of PM 2.5 (PPM 2.5 ) attributable to the ERMs over Beijing, Tianjin, Hebei, Henan and Shandong during the polluted period (Stages I and II) were 7% (11%), 7% (7%), 5% (6%), 4% (4%), and 3% (3%), respectively. The largest decrease of daily PM 2.5 was 9% in Beijing and Tianjin. The largest decrease of daily PPM 2.5 of 14% was also in Beijing, indicating that the benefit of the ERMs in reducing public PM 2.5 exposure was greatest in the most densely populated locations.
Overall, the reductions in PM 2.5 and PPM 2.5 as a result of the ERMs were surprisingly modest in the NCP, considering that the emissions of PM 2.5 and its precursors were significantly reduced by 8-48% due to the implementation of the ERMs (Table 2). In the sections below, we examined the causes for the ineffectiveness of the ERM in reducing PM 2.5 concentrations over the NCP during this severe winter haze event.

Spatial Source Attribution of PM2.5 in the NCP during Clean and Polluted Periods
We used sensitivity simulations (Table 3) to quantify the relative contributions of local versus regional emissions to PM 2.5 concentrations in Beijing and in the rest of the NCP (Tianjin, Hebei, Henan and Shandong) during the severe haze event. Using model simulations driven by Chinese anthropogenic emissions for the year 2010,  previously found that 50% of the monthly mean PM 2.5 concentration in Beijing in January 2013 was due to emissions within Beijing. However, this number should be re-evaluated, as there have been substantial changes in the annual anthropogenic pollutant emissions in Beijing and its surrounding areas between the 2010 inventory (Zhang et al., 2009;Lei et al., 2011;Lu et al., 2011) used by  and the 2016 inventory  used in our study (Fig. S4). Fig. 7(a) shows the percentages of the PM 2.5 in Beijing attributable to emissions from Beijing, from the rest of the NCP and from outside the NCP, respectively. During the clean period, 60% of the simulated mean PM 2.5 concentration in Beijing was due to emissions from outside the NCP. This was mostly dust transported to Beijing by northwesterly winds. Pollutants emitted within Beijing and pollutants emitted from the rest of the NCP contributed only 25% and 15% of the PM 2.5 in Beijing, respectively, during the clean period. During polluted periods and based on the NOERM simulation, pollutants emitted within Beijing contributed only 18-22% of the PM 2.5 in Beijing during Stages I and II, much lower than the 50% inferred by  for January 2013. In comparison, pollutants emitted from the rest of the NCP and from outside the NCP contributed 49-58% and 23-29% of the PM 2.5 in Beijing, respectively. However, Han and Zhang (2017) got similar results in July 2015 that Beijing local emissions contributed 20-30% to the PM 2.5 concentrations in Beijing and the surrounding regions contributed the majority. The large contribution of pollutants emitted from outside Beijing to the PM 2.5 in Beijing during the severe haze event reflected the regional stagnant weather, which enabled the accumulation of pollutants over the entire NCP.
Due to the implementation of ERMs, the simulated mean PM 2.5 concentrations in Beijing decreased by 4% and 9% relative to the NOERM case during Stages I and II, respectively. During Stage I, the decrease was due to the combined effects of ERMs implemented in Beijing (2%) and in the rest of the NCP (2%). During Stage II, the decrease was mostly due to the combined effects of ERMs implemented in Beijing (5%) and the rest of the NCP (3%). A remaining 1% decrease of the PM 2.5 concentration in Beijing during State II was due to small nonlinearity in the PM 2.5 production. We will return to this point later. Fig. 7(b) shows the simulated attribution of the mean PM 2.5 concentration in the rest of the NCP. Under both clean and polluted conditions, more than 62% of the mean PM 2.5 concentrations in the rest of the NCP was from local emissions, Fig. 7. Simulated contributions of local versus regional emissions to PM 2.5 concentrations in (a) Beijing and (b) the rest of the NCP during the clean period and during the polluted period (Stages I and II). Concentrations and percent contributions from Beijing (red), from the other municipalities within the NCP (dark grey), and from outside the NCP (light grey) are shown inset.
while 34-37% was attributed to emissions from outside the NCP. Contributions from Beijing were small throughout the study period. The simulated mean concentrations of PM 2.5 in the rest of the NCP decreased by 3% and 5% in the ERM case relative to the NOERM case during Stages I and II, respectively. All of the simulated abatements in PM 2.5 were attributed to emission reductions in the rest of the NCP.
Our sensitivity simulations used the "emission zero-out" approach to quantify the percentage contribution of a particular source area to the PM 2.5 concentration in a receptor area. For example, we compared the simulated Beijing PM 2.5 concentrations between the ERM and ERM_NOOTH experiments to quantify the contribution of emissions from the rest of the NCP to the PM 2.5 concentrations in Beijing. One concern for this approach is that secondary PM 2.5 production is nonlinear, and that zeroing out the emissions from a source area may cause nonlinear chemical responses to the simulated PM 2.5 concentrations in the receptor area. Another potential source of nonlinearity involves the feedback between PM 2.5 and meteorology, but that was not an issue in our case because our WRF-Chem simulations were nudged with meteorological observations. We found however, that the overall nonlinear chemical responses were small in this particular pollution event. Using the emission zero-out approach, we estimated that the pollutants from Beijing and from the rest of the NCP contributed 29.7 µg m -3 and 89.4 µg m -3 of PM 2.5 in Beijing under the NOERM scenario during Stage I (Fig. 7(a)). This meant that the PM 2.5 contributed from outside the NCP was 36.1 µg m -3 under the NOERM scenario during Stage I. Under the ERM scenario, we similarly calculated a 36.3 µg m -3 contribution from outside the NCP. This indicated that, during this particular pollution event, there was no strong nonlinearity in PM 2.5 production, otherwise the perturbation in emissions by the ERM enforcement would have led to differences in the calculated contribution from outside the NCP, where no control measures were implemented. We further compared the sensitivity simulations for Beijing and for the NCP during Stages I and II. Overall, the chemical nonlinearity accounted for, at most, a 1% decrease in the PM 2.5 concentration in Beijing during Stage II under the ERM scenario. This small nonlinearity did not change our overall source attribution.

Cause of the Ineffectiveness of the ERMs during the Severe Haze Event of December 6-10, 2015
We analytically examined the cause of the ineffectiveness of the ERMs in abating surface PM 2.5 concentrations over the NCP from December 6 to 10, 2015. Without the ERMs implemented, the daily source of PM 2.5 contributed by source area j to a specific receptor area i during the polluted period was S NOERM,i,j (µg m -3 day -1 ), which included contributions both primary and secondary from source area j. The total daily PM 2.5 source for a receptor area i (S NOERM,i ) was the sum of S NOERM,i,j over all values of j. In our case, j = 1-3, indicating contributions from Beijing (j = 1), from the rest of the NCP (j = 2), and from all other regions outside the NCP (j = 3), respectively: The individual S NOERM,i,j can also be expressed as: where g i,j was the fractional contribution of pollutants from source area j to surface PM 2.5 in receptor area i. The values g i,j for j = 1, 3 were given by the source attribution in Section 5 (Fig. 7). For the receptor area Beijing, the values of g i,j for j = 1, 3 were 0.19, 0.58, and 0.23 during Stage I and 0.22, 0.49, and 0.29 during Stage II. Similarly, for the receptor area of the rest of the NCP, the values of g i,j for j = 1, 3 were 0.01, 0.65, and 0.34 during Stage I and 0.01, 0.62, and 0.37 during Stage II. For a well-defined receptor area i, S NOERM,i was also equal to the sum of the inflow of PM 2.5 from other source areas, the local primary PM 2.5 emissions, and the local chemical production of secondary PM 2.5 : S NOERM,i = S NOERM,i | PM 2.5 inflow + S NOERM,i | local primary PM 2.5 emis + S NOERM,i | local secondary PM 2.5 prod (4) The first two terms on the right-hand side in Eq. (4) can be directly calculated from our NOERM simulation. The third term involves the efficiencies at which precursors were converted to secondary PM 2.5 . We estimate this third term using the simulated local conversion ratios of individual anthropogenic precursors to their respective secondary aerosol products. For example, we used the simulated local sulfur oxidation ratio (SOR) ≡ [SO 4 2-(a)]/([SO 4 2-(a)] + [SO 2 (g)]) to estimate the local conversion efficiency of SO 2 to sulfate (Fang et al., 2017). Similarly, the local nitrogen oxidation ratio (NOR), the local ammonia conversion ratio (NHR), and the anthropogenic VOC oxidation ratio (VOR) were calculated from our NOERM simulation (Fang et al., 2017). The definitions and values of these ratios are shown in Table S4. Our simulated SOR, NOR, and NHR values were in the range of values previously reported in wintertime observation studies over the NCP (Liu et al., 2020). Using the model emissions and the values in Table S4, we estimated the third term on the right-hand side of Eq. (4) as: Using the NOERM simulated results to calculate the values in Eqs. (4) and (5), we estimated S NOERM,i to be 68 µg m -3 day -1 in Beijing and 76 µg m -3 day -1 for the rest of the NCP during the polluted period.
We assumed that the transport pathways and transport efficiencies of PM 2.5 and the production efficiencies of secondary PM 2.5 from precursors during the polluted period were similar with and without the implementation of ERMs for a given period, i.e., we assumed that the values of g i,j and the conversion ratios in Table S4 were similar with and without the implementation of ERMs for a given period. This is a reasonable assumption given that the simulated impacts of the ERMs on PM 2.5 concentrations were small across the NCP. Also, our sensitivity simulations indeed showed the source attributions (g i,j ) to be very similar with and without the ERMs (Fig. 7). Then, the fractional abatement of the total PM 2.5 source contributed from source area j to any other receptor area, α j , was: where S ERM,i,j was the daily source of PM 2.5 contributed by source area j to a specific receptor area i during the polluted period with the ERMs implemented. E NOERM,j and E ERM,j were the emission rates of precursors from source area j. Table 2 shows the emission reduction percentages of primary PM 2.5 and precursors to be in the range of 4% and 15% during Stage I. We estimated the overall α j by weighting the precursor emission reduction percentages with the simulated composition of PM 2.5 (Fig. 5(e)). In this way, we estimated α j to be 12% and 34% for Beijing during Stages I and II, respectively. Using the same method, the values of α j were estimated to be 11% and 14% for the rest of NCP during Stages I and II, respectively. Finally, if the ERMs were implemented, the total daily source of PM 2.5 in area i during the polluted period was S ERM,i : For region i and from an initial PM 2.5 concentration c i (t = t 0 ) in the boundary layer, if ERMs were not implemented, the PM 2.5 concentration (c NOERM,i ) at time t = t 0 + Δt during the polluted period was (Jacob, 1999 where τ i is the lifetime of PM 2.5 in region i, which was 4.8 days throughout the NCP during the polluted period. If ERMs were implemented, the PM 2.5 concentration (c ERM,i ) at where c NOERM,i (t 0 + Δt) and c ERM,i (t 0 + Δt) were described analytically by Eqs. (8) and (9), respectively. Using Eqs. (8), (9), and (10), we can analytically compute the fractional abatements of PM 2.5 concentrations (f i ) in Beijing and in the rest of the NCP as a result of the enforced ERMs, as functions of days since the initiation of the ERM enforcement (Δt). The results are shown in Fig. 8. The fractional abatements of PM 2.5 gradually become more evident over time. The rates of changes of f i were determined not only by the emission reductions, but also by the lifetimes of PM 2.5 , as indicated in Eqs. (8)-(10). In Beijing, the mean analytical fractional abatements of PM 2.5 during Stages I and II were 4% and 9%, respectively ( Fig. 8(a)). In the rest of the NCP, the mean analytical fractional abatement of PM 2.5 during Stages I and II were 3% and 6%, respectively ( Fig. 8(b)). These analytically calculated mean fractional abatements of PM 2.5 were consistent with our simulated results in Section 3.3.
As calculated in Section 3.1.2, the lifetime of PM 2.5 in the boundary layer of the entire NCP (τ i ) was 4.8 days during the polluted periods. Hypothetically, if the lifetime of PM 2.5 was very short, or if the ERMs were enforced for a duration (Δt) much longer than τ i , then Δt/τ i → ∞ and exp(-Δt/τ i ) would approach 0 in Eqs. (8) and (9). Under this hypothetical situation, the fractional abatement (f i ) of PM 2.5 in area i associated with the implementation of the NCP-wide ERMs would approach a maximum, f max,i :  By applying the values α j (estimated by weighting the emission reductions in Table 2, as described above) and g i,j (from Fig. 7), one can calculate the values of f max,i . For Beijing, f max,i were 9% and 15% during Stages I and II, respectively. For the rest of the NCP, f max,i were 8% and 9% during Stages I and II, respectively.
These results explained the ineffectiveness of the ERMs in alleviating surface PM 2.5 pollution from December 6 to 10, 2015. If, hypothetically, the stagnant weather and the same ERM enforcement for Stage II had continued, then the fractional abatements of PM 2.5 would have eventually approached f max,i after approximately 10 days (2τ). This is shown in Fig. 8 by the f i curves eventually approaching the f max,i values, which were 15% and 9% for Beijing and the rest of the NCP, respectively. In reality, however, the passage of a cold front ventilated the NCP boundary layer on December 10, 2015, and terminated the severe haze event. Thus, there was insufficient time for the PM 2.5 concentrations to fully reflect the reduction in emissions, i.e., the fractional abatements of PM 2.5 did not have enough time to reach their maximum potential values, f max,i .

CONCLUSIONS
Between December 6 and December 10 in 2015, the cities and provinces of the NCP, for the first time, executed ERMs to reduce anthropogenic emissions, with the goal of alleviating a severe PM 2.5 episode. To evaluate the effectiveness of these measures, we first surveyed provincial/municipal documents and news reports to assess the emission reductions due to the ERMs in each of the NCP's five cities and provinces. Then, we simulated the PM 2.5 concentrations in these areas during the clean period and with and without the reductions during the polluted period. The results indicated that the measures decreased the anthropogenic PM 2.5 and its precursors, excepting NH 3 , across the NCP by 8-48%. Our model simulations incorporating the reduced emission estimates were able to reproduce the observed PM 2.5 concentrations and compositions throughout the event.
We found the effects of the ERMs on the PM 2.5 concentrations to be surprisingly modest. During the episode, the overall mean decrease resulting from the measures was 7% in Beijing and 4% across the rest of the NCP; furthermore, local emissions accounted for only 18-22% of the surface PM 2.5 in Beijing, whereas emissions originating elsewhere on the NCP and even farther away contributed 49-58% and 23-29%, respectively. Excluding Beijing, 62% of the mean surface concentration across the NCP was due to pollutants emitted locally, with the remainder (34-37%) attributed to pollutants generated outside the NCP.
The ineffectiveness of the ERMs in mitigating the high PM 2.5 concentrations can be explained by the increased lifetime of this pollutant (5 days) in the boundary layer of the NCP-a consequence of stagnant weather conditions that impeded horizontal and vertical ventilation-which, because of the ERMs' shorter duration, prevented the concentrations from fully reflecting the reduction in emissions. Thus, this episode stands in strong contrast to previous occasions, e.g., the Sino-African Summit, the Beijing Olympic Games, the APEC summit, and the 2015 Chinese Military Parade, when temporary emission controls in BTH and its surrounding areas significantly improved the air quality in Beijing (e.g., Cheng et al., 2008;Liu et al., 2015;Zhang et al., 2016;. However, the control measures were implemented for longer periods of time during those events. More importantly, the efficacy of the actions was aided by the effective ventilation of the boundary layer air in Beijing (e.g., Zhang et al., 2016).
Our results illustrate the challenge in reducing surface PM 2.5 concentrations via ERMs during severe winter haze events, when the stagnant weather conditions impede the effectiveness of temporary measures. Anthropogenic emissions on the NCP under these circumstances must be reduced by a much larger percentage to substantially abate the PM 2.5 concentrations, although these reductions would be economically costly. Concurrently issuing a stronger advisory for citizens to reduce outdoor activity and improve indoor air quality (e.g., by using air purifiers to filter PM 2.5 ) is also necessary in order to reduce public exposure during these pollution episodes.