Impact of Air Humidity Fluctuation on the Rise of PM Mass Concentration Based on the High-Resolution Monitoring Data

Hourly particulate matter mass concentrations and meteorological parameters, recorded by the China National Environmental Monitoring Center (CNEMC) and China Observatory, between June 2013 and March 2016 in Beijing, Xi’an, Shanghai and Guangzhou were examined to explore correlations. Characteristics of a rapid (abrupt) rise in PM2.5 mass concentration during early stage of a serious urban PM pollution event were examined and compared with pollution events with a gradual (accumulative) rise in PM2.5 mass concentration. The accumulative rise in air pollution is characterized by a prolonged slow PM2.5 growth rate (3–5 μg m h), and could eventually lead to middle level pollution (ambient PM2.5 mass concentration of about 150 μg m), accompanied with an uncertain temporal variation in SO2, NO2, O3 and CO concentrations. The abrupt rise process is characterized by a short-term high aerosol growth rate (> 10 μg m h), and could eventually form severe air pollution (PM2.5 mass concentration exceeds 250 μg m), with a constant increase in gaseous pollutants concentrations. The average relative humidity (RH) was observed to have a less impact on the rise of PM2.5 concentration, but the fluctuation in RH was found to have a strong correlation with the rise in PM2.5 concentration. Further analysis has indicated that both abrupt and accumulative rise in PM2.5 mass concentration could be due to the RH fluctuation in atmosphere. The RH variation is important for the study of the fine-particle growth and for prediction of PM pollution episodes.


INTRODUCTION
Research on airborne particulate matter (PM) has become one of the most important subject areas for the understanding of the urban air pollution (Chan and Yao, 2008;Kan et al., 2012;Lu et al., 2015).The serious air pollution has been threatening the environmental heath of many mega or medium-sized cities in many countries, and is believed to have contributed to the deaths of more people than AIDS, malaria and breast cancer all over the world (WHO, 2012).The PM pollution has been estimated to have caused 3 to 7 million deaths every year due to their adverse impact on worsening cardiorespiratory diseases (Pope et al., 2002;Hoek et al., 2013;Yang et al., 2013;WHO, 2014).Meanwhile, the seriously high concentrations of PM 2.5 in ambient air is contributing to the ash haze weather in various cities (Friedlander and Marlow, 1977;Du et al., 2012).Various studies have shown that the mass concentration and chemical compositions of aerosols in each city are very different due to the different urban economic developments, traffic conditions, industrial activities and meteorological factors (Zhang et al., 2011;Zhang and Gu, 2013;Huang et al., 2014;Rohde and Muller, 2015), however, the characteristics of the atmospheric conditions on the formation of haze episodes have not been sufficiently studied.
The sudden formation of severe haze has become a common problem in developing countries, especially in China, India, Brazil and other industrial countries, due to rapid urbanization (Reid et al., 1998;Pachauri et al., 2013;Wang et al., 2014a).An interesting feature is the soaring PM 2.5 concentrations that occurred during the early stage of severe hazes (Friedlander and Marlow, 1977;Du et al., 2012;Wang et al., 2014b), which is called the abrupt rise period.Abrupt rise of PM 2.5 haze was also observed during Beijing winter hazes by Zheng et al. (2015), which was presented as dramatic hourly fluctuation with a maximum mass growth rate of up to 351.8 µg m -3 h -1 and PM 2.5 concentrations rose in excess of 400 µg m -3 within 1-3 hours during the episode in January 2013.The abrupt rise of PM 2.5 concentrations could only be observed with high-resolution monitoring data, which is the reason that most previous studies failed to focus on the temporal change in particle mass.What happened during the early stage of haze episodes is the key to understand the formation of high pollutions.
Previous studies have demonstrated that wind speed and atmospheric relative humidity (RH) are the two most important external meteorological factors affecting the mass concentrations of aerosol (Sun et al., 2006;Tian et al., 2014;Zhang et al., 2016).In the stable atmosphere, high atmospheric RH could enhance fine particulate accumulation and therefore formation of haze, and could also induce a rise in concentrations of water soluble inorganic ions (SO 4 2-, NO 3 -, NH 4 + ) in the atmosphere, and reduce the visibility in urban areas (Yan et al., 2009;Ding and Liu, 2014;Gao et al., 2015;Guo et al., 2014;Lin et al., 2015).During haze episodes, high atmospheric RH could accelerate the formation of secondary particles, causing the aggravation of atmospheric pollution in ambient air (Sun et al., 2013).The components and size distribution of atmospheric PM could determine the deliquescence point, growth rate and the light scattering coefficient of fine particles (Wang et al., 2003;Gysel et al., 2007;Cao et al., 2012;Wang et al., 2012).Air humidity is also a significant factor that could determine the movement of particles in troposphere due to the distribution of electrical charges on particles (Wei and Gu, 2015;Zhang et al., 2016).
High RH might not necessary be a direct impact factor on the growth in PM 2.5 concentration.For instance, the abrupt rise in PM 2.5 concentration, rising from 140 to 318 µg m -3 in 5 hours, happened with an average RH as low as 50% in Beijing during January 17 th , 2010 (Zhao et al., 2013).Similar phenomenon was also observed in Los Angeles in 1969 (Husar et al., 1972).On the other hand, the average humidity as high as 85% did not result in further rise in PM 2.5 concentration during this high concentration phase, the PM 2.5 concentration was ranging from 302 to 447 µg m -3 during the October 9 th to 11 th , 2014.
Furthermore, regardless of the average humidity was high or not, stronger fluctuation of relative humidity was recorded before or at the beginning of each abrupt rise event of PM 2.5 concentration (Yang et al., 2011;Zhao et al., 2013;Wang et al., 2014b).For instance, the abrupt rise in PM 2.5 concentration, rising from 50 to 600 µg m -3 in 20 hours, happened with a strong RH fluctuation from 38% to 87% in Shanghai during December 5 th , 2013 (Leng et al., 2016).The purpose of this study is to examine if the RH fluctuation is related to PM 2.5 abrupt rise event.The RH variation in 6 hours (Var-RH6) would be presented by an index to analyse the influence of RH on the abrupt rise in PM concentration.Using the hourly high-resolution atmospheric pollution data, the effect of air humidity on the process of abrupt rise in urban PM concentration was analysed to illustrate the significance of atmospheric water vapour fluctuation.This could provide an important indicator in a warning system for prediction of an onset of a serious haze episode and the abrupt growth of pollution aerosols to form part of a development of amelioration strategy for urban management of atmospheric pollution.

METHODS
China is facing serious ambient air pollutions in world's history (Chan and Yao, 2008;Kan et al., 2012), where high average concentrations of all kinds of gaseous pollutants have been recorded.This is a typical consequence of rapid industrialization of developing countries which is being experienced all over the world.
We studied the ambient air pollution in various cities in China and monitored atmospheric PM concentrations in 4 megacities and areas around: Beijing, Xi'an, Shanghai and Guangzhou.The relative locations of these cities in China are shown in Fig. 1.

Data Sources
The original data was mainly collected from the China National Air Quality Network, which has 945 fastresponding and real-time monitoring sites at 160 cities in China, providing the high precision concentrations of PM 2.5 , PM 10 , SO 2 , NO 2 , O 3 and CO in the ambient atmosphere in these cities.However, most archived observations are not publicly available.To compensate, real-time data was downloaded during a 33 month interval from June 2013 to March 2016.Due to the download restrictions on the official report, two different third-party sources (PM 25 .inand AQICN.org) were used.PM 25 in is a direct mirror of data from the China's national network, while AQICN.orgprovides a large amount of the world's real-time air quality data and included many additional sites in China and surrounding areas.Thus we have 38 stations in Beijing, 14 stations in Xi'an, 13 stations in Shanghai and 17 stations in Guangzhou.Logging interval at these stations was set at 1 hour.According to the Chinese national standard (HJ/T 193-2005), the sampling pedestals of automatic monitoring instruments were placed 3-15 m above the ground or 1m above the top of buildings.In each city the monitoring stations were located in different downtown and suburb areas.The average PM 2.5 concentrations were monitored at these downtown monitoring sites to serve as the regional PM 2.5 data for our study.
The ambient temperature, relative humidity, atmospheric pressure, wind scale and wind direction were recorded together with the concentrations of ambient air pollutants by local official meteorological stations at China Observatory.The logging interval of these data were set at one hour.

Data Quality Control
Automated quality control checks were performed on each air pollutant to remove the repeated values and implausible zeroes, common signs of missing measurements being replaced with a placeholder.Collectively, these quality control checks excluded 7% of the reported values from China.The most common exclusion criterion (54% of exclusions) was for continuous non-zero concentration of pollutants reported for several hours in a row.Due to natural variability, instruments will give the same value in a row occasionally.If the data is the same for more than 5 hours, it is more likely that the instrument has stopped monitoring and the reporting system is simply repeating the last data received.Repeated values once continued every hour for as long as 3 weeks.
In addition, a regional consistency check has been performed for every pollutant where the value at each location was compared to the average value at nearby stations at the same time.The least consistent 0.6% of observations were excluded.In many cases, the observations removed are more than double what one would expect based on the average of their neighbours.Data removed as a result of this test is not necessarily inaccurate, but may reflect a pollution source in the immediate vicinity of the monitoring site that is not representative of conditions in the rest of the city or region.As we wish to focus on largescale patterns, it is desirable to remove such local outliers.Additional steps to compensate for local noise and outliers were built into the interpolation scheme described below.In addition to these quality control steps, data gaps in a station record lasting two hours or less were filled by linear interpolation to help reduce the noise in the reconstruction due to missing stations.
Similar to other scholars' results (Karar et al., 2006;Tai, et al., 2010), PM concentration is inversely correlated with both of the strong wind and precipitation, which are determined by severe convections (the correlation coefficient of strong wind and PM 2.5 concentration is R 2 = -0.86).Occasionally in spring, strong wind often brings dust storms covering the northern cities, causing a sharp rise in the PM 10 concentration and small increase in the PM 2.5 concentration; hence causing serious PM 10 pollution.Because the purpose of this research is focusing on the process of rapid growth of the fine particles (PM 2.5 ), the time of strong wind and precipitation weather were excluded from the analyses of raw data.
The monitoring period in Xi'an was from June 2013 to March 2016 with 19258 hours of total raw data.The monitoring period in other cities was from January 2014 to March 2016 and with 16468 hours of data in Shanghai, and 15025 hours in Guangzhou, and 16334 hours in Beijing.
The effective data moment (t 0 ) was defined as below: When t ∈ (t 0 ± 5h), if wind speed < 3.3 m s -1 , precipitation J = 0, then t 0 is accepted the effective data.
The duration of effective data in this analysis: 12668 hours in Xi'an, 9007 hours in Shanghai, 10297 hours in Guangzhou is and 8275 hours in Beijing.
The variance of six-hours RH was adopted to measure the strength of fluctuations in the atmospheric humidity based on the long-time monitoring in the four cities.The variance is a typical index to represent the fluctuation of data and 6 hours is 1/4 a period of fluctuation, which is easy to compare because this is always stable in different times.So we adopt the natural logarithm of the variance value of six-hours RH, as shown in Eq. ( 1), to measure the strength of the atmospheric humidity fluctuation, the calculation equation is as follows:

Valid Data Analysis-Air pollution Condition of Study Areas
Through the raw data analysis, excluding the invalid data, the annual main air pollutants and meteorological elements of four typical cities are shown in Fig. 2. The analysis of various air gaseous pollutants indicated that the concentration of NO 2 is similar (43.9-47.0µg m -3 ), the concentrations of SO 2 is highest in Xi'an (21.8 µg m -3 ) and the concentration of O 3 is highest in Shanghai (169 µg m -3 ).PM 2.5 pollution is most severe in Beijing (74.3 µg m -3 ) and PM 10 pollution is most severe in Xi'an (136.1 µg m -3 ).PM 2.5 and PM 10 concentrations of inland cities such as Xi'an and Beijing are much higher than coastal cities such as Shanghai and Guangzhou.All these four cities' people experience the unhealthy air condition (PM 2.5 > 35 µg m -3 ) (GB 2012).The northern cities (Beijing and Xi'an) are more polluted than the southern cities (Shanghai and Guangzhou), but certainly not as humid.

Effective Data Analysis-Formation Period of Severe Haze Period
During the monitoring period, a lot of extremely severe PM 2.5 pollution events (PM 2.5 mass concentration > 250 µg m -3 , lasting for a period for over 3h, PM 2.5 is the main pollutant) have been observed which exceeded the highest threshold of Chinese national standard, WHO recommend and US EPA standard (HJ 633-2012;GB 3095-2012;WHO, 2006;EPA, 2014) in four cities.Further analysing the early period of haze in the effective data, the growth rate of PM 2.5 concentration was above 10 µg m -3 h -1 , accompanied with stable atmospheric condition (wind speed < 3.3 m s -1 , planetary boundary layer is lower than 500 m, no precipitation).This phenomenon is defined as "abrupt rise" of PM 2.5 concentration.
Fig. 3(a) shows a typical abrupt rise event in Beijing.The duration of abrupt rise period in Beijing was always within 12 hours and the average concentration growth was above 10 µg m -3 h -1 , leading to a serious PM 2.5 pollution (PM 2.5 mass concentration > 250 µg m -3 ).In contrast, the rise in PM 2.5 concentration is nearly linear during the formation period of normal aerosol pollution (115 µg m -3 < PM 2.5 mass concentration < 200 µg m -3 ).This period is defined as "accumulative rise" of PM 2.5 concentration.
The duration of the accumulative rise period can exceed 24 hours (sometimes in excess of 48 hours) and the average concentration growth was 3-5 µg m -3 h -1 , in which would not cause a serious PM 2.5 pollution.A typical event in Xi'an is shown in Fig. 3(b).Meanwhile, during the accumulative rise period, the gaseous pollutants concentration varied irregularly and the relative humidity fluctuated within a tight range during a 24-hour cycle.However, during the abrupt rise period, the concentration of various gaseous pollutants would rise (except O 3 ) simultaneously with a rise in PM 2.5 concentration, and the relative humidity was shown to exhibit extreme volatility prior to this period.
Fig. 4 illustrates 3 severe haze episodes recorded under the stable atmosphere during October 1 st to 30 th in Beijing.During this period, the increase of PM could be distinguished into two modes, the abrupt rise (yellow rectangle) and accumulative rise (light blue rectangle).Fig. 4 illustrated two modes of PM increase.
Pollution emissions are always localized during these abrupt rise events, especially in these mega cities because most of the largest emissions sources are located in or near urban districts.As shown in the 3 haze episodes of Fig. 4, under the stable atmosphere in this region, although there was little migration of PM in the atmosphere, we observed similar deviation in the curves of the PM 2.5 concentration in three cities close to each other (260 km apart): Beijing (BJ), Shijiazhuang (SJZ) and Langfang (LF), illustrating the character of localization of PM pollution.Therefore, during the abrupt rise event, the migration is not the main cause in the rise in PM 2.5 mass concentration.
Further analysing the effective data, it is worth to mention that before each abrupt rise period, Ln(Var-RH) is larger than 6, referring that the fluctuation of RH is stronger.In the accumulative rise period and severe haze period, Ln(Var-RH) is lower than 6.
Fig. 5 demonstrate the abrupt rise events in Xi'an (a), Guangzhou (b) and Shanghai (c).Under the stable atmosphere, the growth rate of all these three events is more than 10 µg m -3 h -1 , and finally result severe haze (PM 2.5 concentration is above 250 µg m -3 ).Values of Var-RH were also higher before each abrupt rise period.

Effective Data Analysis-Average RH Influence
As discussed in the introduction, RH is an important external meteorological factors affecting the mass concentrations of aerosol.RH do not influence the increase of PM directly.In the effective data, the average growth rate of PM 2.5 concentration under different relatively humidity sectors of our four monitoring cities in the stable atmosphere was calculated to determine the direct effect of air humidity to the rise in PM 2.5 concentration (shown in Fig. 6).The relative humidity was divided into eight sectors namely 0%-24%, 25%-34%, 35%-44%, 45%-54%, 55%-64%, 65%-74%, 75%-84%, 85%-100%.The results of our analyses indicated that: low humidity conditions are not conducive to the growth of PM 2.5 concentrations, when RH < 25%, the concentration of PM 2.5 has generally a downward declining trend.When RH is between 45% and 84%, the concentration of PM 2.5 is generally on the rise and the growth rate is 0.67-1.13µg m -3 h -1 ; when RH > 85%, the growth rate of PM 2.5 concentration is only 0.39 µg m -3 h -1 .In fact, neither too low nor too high relative humidity conditions would be conducive to the rapid rise in the PM 2.5 concentration.However, the peak growth rate within the scope of the entire humidity sections is only 1.13 µg m -3 h -1 , which is obviously not enough to support the abrupt rise process (minimum growth rate is 10 µg m -3 h -1 ).Thus the average humidity is not the most significant condition leading to the abrupt rise of urban PM 2.5 concentration.
In addition, neither of sunlight nor high humidity conditions would be the essential condition that cause the process of abrupt rise in PM 2.5 concentration, as shown in Fig. 3(a1), during the period of 10:00-17:00, 9 th December, Fig. 4. The high-resolution observation result of regional PM 2.5 , atmospheric pressure, RH, variance of RH in 6 hours (Var-RH), wind speed and precipitation from October 1st to 30th in Beijing.
the urban air relative humidity was only 17% to 23%, which is a typical low humidity condition.Events of the abrupt rise process occurring at night, were repeatedly observed, suggesting that the sunshine could also not an essential factor for the abrupt rise in PM 2.5 concentration.

Effective Data Analysis-Fluctuation RH Influence
The fluctuation of RH is the key to analyse the role (or effect) of RH on the occurrence of a sudden heavy haze event.The violent fluctuation of RH was commonly observed prior to every abrupt rise in PM 2.5 concentration, however we are still do not clearly understand the effect in the lab study.The period of RH fluctuation in the atmosphere is usually 24 hours which is usually accompanied with a daily fluctuation in temperature.However, in different cities and during different seasons, the difference in the average atmospheric humidity and the fluctuation range of relative humidity is very large.In fact, the daily fluctuation range of RH is usually 30% to 60%, but in Fig. 3(a), prior to the abrupt rise in PM 2.5 concentration, there was an obvious RH jump, from 85% to 17% in only six hours, and the fluctuation range of RH before and after the abrupt rise period was very stable (within 30%).Similar phenomenon were also shown in Figs. 4 and 5.In contrast, in Fig. 3(b), the process of accumulative rise, relative humidity was kept to 35% to 65% and the cycle was stable.Even during this slow incremental rise process, before a small jump in PM 2.5 concentration, there was a relatively strong fluctuation in the atmospheric RH.Statistical analysis of RH fluctuation shows highly correlation with the increase of PM concentration under the stable atmosphere.When ln(Var-RH) is lower than 2, the fluctuation of RH is weak, PM concentration has no clear trend; When ln(Var-RH) is between 2 and 4, the fluctuation of RH is normal, PM concentration always rise accumulatively (growth rate < 3 µg m -3 h -1 ); When ln(Var-RH) is between 4 and 6, the fluctuation of RH is strong, PM concentration always rise rapidly (growth rate is from 3 to 10 µg m -3 h -1 ); When ln(Var-RH) is above 6, the fluctuation of RH is violent, PM concentration always rise abruptly (growth rate is above 10 µg m -3 h -1 ).Analysis results show that the strong humidity fluctuations would have a strong influence on the growth of PM 2.5 concentration.Actually, the strong RH fluctuation was shown to appear prior to every abrupt rise in PM 2.5 concentrations.In order to evaluate the function of RH fluctuation, the abrupt rise in PM 2.5 concentration is defined as the growth rate over 5 µg m -3 h -1 , with a duration of over 5 hours, which would have a more rapid rise than accumulative rise.Fig. 7(b) indicates that when the fluctuation in RH is strong (ln(Var-RH) > 4), the probability of a abrupt rise in PM 2.5 mass concentration appearing in 10 hours is as high as 89% (average RH > 80%), 73% (average RH < 70%).When the fluctuation in RH is weak (ln(Var-RH) < 2), the probability of an abrupt rise in PM 2.5 mass concentration appearing in 10 hours is as low as 27% (average RH > 80%), 15% (average RH < 70%).
Based on the study results, further research will be focusing on the mechanism of humidity fluctuations that could affect the aggregation and growth in the PM concentration, especially the growth in the secondary pollutants in urban air.In addition, because of the large regional differences in this research, more precision chemical composition would be needed and specific changes in meteorological data would also be required to enable the quantitative analysis of the interaction between the growth of atmospheric aerosols and the fluctuation of air humidity.

CONCLUSIONS
In this study, we collected the long period air pollutant concentration and meteorological data from June 2013 to March 2016 in four typical mega cities.The differences in meteorological conditions and pollution sources between cities would determine the different aerosol pollution, but there are always a direct correlation between the occurrence of serious aerosol pollution and the strong fluctuation of air humidity.Our main research conclusions are as follows: 1.The process of abrupt rise, occurs at the early stage of the severe PM pollution, remain underappreciated.The early stage of PM pollution process can be distinguished into two modes, the accumulative rise and the abrupt rise, due to their respective growth rate and duration.In the process of abrupt rise, gaseous pollutants concentration rise in step with PM 2.5 concentration, determined by the combined effects of physical and chemical process that lead to the extremely severe pollution.The mechanism and the process of accumulative rise, and the subsequent concentrations are very different from the abrupt rise.2. The analysis results indicate that under the stable atmospheric environment, the humidity levels are significant to low visibility, but have little effect on the rise in PM concentration.When RH is in 35%-85%, the growth rate of PM 2.5 concentration was only 0.67-1.13µg m -3 h -1 .Extremely high or low RH could not have an effect on the rise in PM concentration.3.Under the stable atmospheric environment, the fluctuation of air humidity would have a strong correlation with the process of accumulative rise and abrupt rise in PM concentration.The strong fluctuation in RH is an essential condition for the 32 times observed process of abrupt rise in PM concentration.The probability of abrupt rise in PM 2.5 concentration is as high as more than 80% due to the aftereffect of strong RH fluctuation, which is also significant in the process of accumulative rise in PM 2.5 concentration.

Fig. 1 .
Fig. 1.The locations of the four monitoring cities: Beijing, Xi'an, Guangzhou and Shanghai in China.

Fig. 2 .
Fig. 2. The annual average temperature and RH and the concentration of PM and several main air pollutants in Xi'an, Beijing, Shanghai and Guangzhou.

Fig. 3 .
Fig. 3. (a): An abrupt rise event in Beijing and (b): an accumulative rise event in Xi'an.(a1) and (b1) shows the variation in the concentrations of PM 2.5 , PM 10 and RH value, (a2) and (b2) shows the variation in the concentrations of SO 2 and O 3 , (a3) and (b3) shows the concentrations of NO 2 and CO.

Fig. 7
(a) demonstrate the classification of RH fluctuation, the normal fluctuation range is 2-4 (54.8% of the total time), the weak fluctuation range is 0-2 (18.5% of the total time) and the strong fluctuation range is above 4 (26.8% of the total time).

Fig. 5 .
Fig. 5.The high-resolution observation result of PM 2.5 concentration, Var-RH, wind speed and precipitation of several typical abrupt rise events in Xi'an (a), Guangzhou (b) and Shanghai (c).

Fig. 6 .
Fig. 6.Average increment of PM 2.5 concentration per hour in the different RH range.

Fig. 7 .
Fig. 7. (a): The distribution of ln(Var-RH) to show the weak, normal, strong and violent RH fluctuation; (b): the probability of rapid growth in PM 2.5 concentration.