Interactive comment on “ Hygroscopic growth effect on aerosol light scattering in the urban area of Beijing : a long-term measurement by a wide-range and high-resolution humidified nephelometer system

Abstract. Hygroscopicity is an important feature of ambient aerosols, which is very crucial to the study of light extinction, radiation force, and formation mechanism. The light scattering hygroscopic growth factor (f(RH)) is an important parameter which is usually measured by the humidified nephelometer system and could better describe the aerosol hygroscopicity under wide particle size range and continuous relative humidity (RH). The f(RH) can be applied to the establishment of a parameterization scheme for light extinction, the calculation of hygroscopicity parameter (κ), and also the estimation of aerosol liquid water content (ALWC). However, the humidified nephelometer system in the previous studies could only observe the f(RH) below 90 % due to the larger error of the sensor under high RH (> 90 %). Furthermore, the f(RH) observations in North China Plain needs to be greatly strengthened both in the temporal resolution and the observation duration. In view of this, an improved high-resolution humidified nephelometer system was established to observe the f(RH) of PM2.5 for a wide RH range between 30 %–96 % in the urban area of Beijing over three seasons (winter, summer, and autumn) in 2017. It was found that the f(80 %) at 525 nm of PM2.5 was evidently higher under the polluted conditions and highly correlated with the fractions of all the water-soluble ions. A two-parameter fit equation was selected to fit the observed f(RH) data. For each season, the fitting curve under the very clean condition was lower than that of other conditions. And the f(RH) points of polluted conditions were more concentrated with higher fitting R2 for summer and autumn data. The hygroscopicity of aerosol under higher RH was probably enhanced when compared with the data in the previous study conducted in NCP. In summer, the fitting f(RH) showed a significant dependence on wavelength for each pollution condition. However, there was an opposite performance in the f(RH) curves of different wavelengths for the very clean condition in winter. It was showed that The simulation showed that the maximum uncertainty of f(RH) was less than 10 %.



Introduction
Atmospheric aerosols influence the atmospheric visibility and earth-atmosphere radiation budget directly by scattering and absorbing solar radiation (Charlson et al., 1992;Schwartz et al., 1995;DeBell et al., 2006), and indirectly by modifying microphysical properties and lifetime of clouds (Twomey, 1974;Albrecht, 1989;Rosenfeld, 2000).The particle size, single scattering albedo (SSA), asymmetry parameter (AP), and refractive index are the key parameters in estimating the aerosol radiative forcing; and these parameters are strongly dependent on the relative humidity (RH).Water uptake can result in enlarged particle sizes, SSA, and AP, change the scattering phase functions, and decreased the refractive index (Cheng et al., 2008).
Moreover, numerous studies demonstrate that reaction in aerosol liquid water is an important pathway of secondary aerosol formation, and thus plays a significant role in the overall aerosol chemical composition (McMurry and Wilson, 1983;Ravishankara 1997;Kolb et al., 2010;Jang et al., 2002;Liu et al., 2012;Cheng et al., 2016).
Furthermore, water uptake is also important in the remote-sensing measurements or satellite retrievals of aerosol optical properties (Wang and Martin, 2007;Zieger et al., 2012).
Particle size and chemical composition are the main factors affecting the moisture absorption ability of aerosol particles (Köhler, 1936;Hinds, 2011;Petters and Kreidenweis, 2007;Seinfeld and Pandis, 2016).Particle size mainly affects the extent of the Kelvin effect, while the chemical composition is the most crucial factor.Some simulations demonstrated that the changes of aerosol size distribution could only lead to slight variations in hygroscopic growth factor if the chemical compositions were fixed (Fierz-Schmidhauser et al., 2010;Liu 2015;Kuang et al., 2017).
Generally, aerosol hygroscopicity could be described by the diameter growth factor (g(RH)), light scattering hygroscopic growth factor (f(RH)), and hygroscopicity parameter (κ).Size-resolved g(RH) could be directly measured by the Humidified Tandem Differential Mobility Analyzer (HTDMA) (Liu et al. 1978;Swietlicki et al., 2008).Due to the limitation of the HTDMA itself, only the particles with the dry diameter under 350nm are usually observed.The f(RH) is calculated as the ratio of the scattering coefficient at a certain RH to the corresponding dry scattering coefficient (reference RH<40%), which can be measured by a dual nephelometer system (Covert et al., 1972;Rood et al., 1985).In recent years, the κ-Köhler theory has been widely used to describe the hygroscopic properties of aerosols (Petters and Kreidenweis, 2007), in which all of the chemical composition dependent variables were merged into a hygroscopicity parameter (κ).It facilitates the intercomparison of particle hygroscopicity obtained by different equipment or different relative humidity.It is applicable to both the single-component and the multicomponent aerosol particles.
Knowing the κ, dry diameter, and the relative humidity (RH), the g(RH) and liquid water content (LWC) of aerosol particles could be calculated by using the κ-Köhler equation.
The f(RH) synthetically reflects the impact of water uptake on aerosol particle size, morphology, refractive index, etc., and consequently, on the scattering coefficient.parameterization scheme for light extinction.The parameterization scheme could be built based on chemical compositions (DeBell et al., 2006;Pitchord et al., 2007), mass concentrations (Chen et al., 2015), or volume concentrations (Chen et al., 2012(Chen et al., , 2014)).
No matter which scheme is used, the effect of relative humidity on extinction or scattering coefficient must be taken into account.Secondly, the f(RH) could be used to calculate the hygroscopicity parameter (κ) combining with the measurement of particle number concentration size distributions (PNSD) (Chen et al., 2014;Kuang et al., 2017).This κ does not target specific particle size but represents the overall hygroscopicity of the observed aerosol.Thirdly, the f(RH) results derived from a three-wavelength nephelometer system have also been used for the calculation of aerosol liquid water content (ALWC) (Kuang et al., 2018) and number concentrations of cloud condensation nuclei (Tao et al., 2018).The humidified nephelometer system was firstly used by Pilat and Charlson (1966) to measure the effect of humidity on the light scattering and the size of NaCl particles.Covert et al. (1972) then used a similar system to study the hygroscopic and/or deliquescence effects which were dependent upon relative humidity for pure particles.Rood et al. (1985) improved the humidified nephelometers, which could control the RH more precisely.In the past two decades, Carrico et al. (2000), Day et al. (2000), Koloutsou-Vakakis et al. (2001), Fierz-Schmidhauser et al. (2010), Liu et al. (2016), etc. further improved the humidified nephelometers to accurately measure the f(RH) and study the light scattering enhancement characteristics of aerosol.While all the humidified nephelometer system in the previous studies could only observe the f(RH) below 90% RH because there would be a ±5% error in the RH measurement in the optical chamber of nephelometer when the RH was humidified over 90%.However, for a hygroscopic particle, its particle size or scattering cross section will increase sharply when the relative humidity exceeds 90%.Accordingly, the results above 90% RH would be quite important for f(RH) curve fitting and light scattering calculating.
As we all know, the North China Plain (NCP) is the most severe area of air pollution in China.The aerosol hygroscopicity is an important basis for the studies of atmospheric visibility, radiative forcing, and aerosol secondary formation in this area.
The HTDMA had been used in some campaigns in NCP (Massling et al., 2009;Meier et al., 2009;Liu et al., 2011;Wang et al., 2018).As mentioned above, the g(RH) observations using HTDMA had been basically limited to 350 nm, and could not be set at multiple RH points.In comparison, the overall hygroscopicity of ambient aerosol at several continuous RH points could be obtained by the f(RH) observation, which would also be used to further calculate the hygroscopicity parameter and ALWC.In recent years, the f(RH) observation has been carried out in only a few studies in NCP (Yan et al., 2009;Pan et al., 2009;Chen et al., 2014;Kuang et al., 2017).In these studies, either the temporal resolution of f(RH) was low, or the observation period was short.Moreover, the f(RH) was limited to 90% RH in all these studies.At present, some numerical pollution prediction systems have been established in the region of NCP.However, the prediction of visibility is short of good means and methods.
Thus, an accurate, easy to use, and seasonal representative visibility parameterization scheme is very critical and urgent.Furthermore, owing to the significant changes in the pollution sources, the physical and chemical characteristics of aerosols in NCP have also changed evidently in recent years.Above all, the research on aerosol hygroscopicity is still not enough for NCP.Further study is required, especially in the area of f(RH).In this study, an improved high-resolution humidified nephelometer system was used to observe the f(RH) of PM2.5 for a wide RH range between 30%-96% in the urban area of Beijing over three seasons (winter, summer, and autumn).The main objectives of this article are to characterize the variations of f(RH) and other optical parameters under different seasons and different pollution levels, and set up the optimal expressions of f(RH).

Observation site
The measurement campaign was performed at the Institute of Urban Meteorological in the Haidian district (about 36m above the ground), which located in the northwest urban area of Beijing, outside the third-ring road (39°56'N, 116°17'E).The sampling site was located next to a high-density residential area, which has no significant emissions from industrial in the surrounding neighborhood.Therefore, the observation data could represent the air quality levels of the typical urban area of Beijing.The observations were conducted in three seasons, 12 nd Jan. to 14 th Feb. for winter, 6 th Jul.
to 21 st Aug. for summer, and 30 th Sep. to 13 th Nov. for autumn.

Instruments
The f(RH) was measured by a dual-nephelometer system, one nephelometer for the aerosol scattering coefficient under dry condition, and another nephelometer for the humidified aerosol.The air flow first passed through a PM2.5 inlet, and then was dried by two tandem Nafion dryers (MD-700), which could reduce the RH of air flow lower than 30%.The dried air was separated into two paths, one stream went directly into a nephelometer, another stream was humidified by passing through a Gore-Tex tube, which was set in a stainless steel tube.The interlayer between these two tubes is circulating water headed by the water bath.The minutely scattering coefficients of dry and humidified PM2.5 under three wavelengths (450, 525, and 635nm) were synchronously measured by these two nephelometers (Aurora 3000).In most studies, there was only one water bath been used for humidifying.After a humidifying process, it is necessary to wait for the water in the water bath to cool down enough for the next process.Differently, in this study, two water baths were used.When one water bath was heating up the water for humidifying, another water bath was cooling down the water itself.After a humidifying process, the water bath with cool water would be switched into the humidification pipeline.The use of two water baths could ensure that the effective data of f(RH) is more than twice that of using only one water bath.The temperature of water in the water bath was controlled by an automatic system to ensure the humidifying effect.
Two combined RH and temperature sensors (Vaisala HMP110) were set at the inlet and outlet of the wet nephelometer, respectively.The vapor pressures were calculated by the sensor data, and the average value was considered as the vapor pressure in the optical chamber.Thus, the humidified RH in the chamber could be calculated through the derived vapor pressure and the temperature measured by the sensor in the chamber.As mentioned above, the RH could not be accurately measured by a sensor when it is above 90%.In order to accurately obtain higher relative humidity, the optical chamber of wet nephelometer was deliberately cooled with the temperature lower than that at the inlet and outlet.So the humidified RH could be higher than 95% in the chamber when the RH at the inlet and outlet were lower than 90%.This method makes it possible to observe the f(RH) under high RH.To avoid the vapor condensation and particle activation, the upper limit of humidified RH in the optical chamber of wet nephelometer was set to 97%, which made the effective data of f(RH) could reach RH of 96%.Each humidifying process lasted about 50 minutes, and all the minutely average data were automatically recorded by the control system.During the observation periods, these two nephelometers were calibrated every ten days.Since two Vaisala sensors and two nephelometers were all newly purchased and the relative error of vapor pressures was always less than 1%, the sensors were not calibrated during the observation.
Other than the f(RH) measurement, the six-minute average PM2.5 mass concentrations were also measured by a continuous dichotomous ambient air monitor (TEOM 1405DF).The sample filter and sample conditioner filter of this monitor were replaced every 15 days or when the dust loading exceeded 70%.The five-minute average absorption coefficient of PM2.5 was monitored by a multiangle absorption photometer (MAAP 5012).The quartz-fiber filter could be automatically changed when the light transmission was less than 20%.In addition, the hourly water-soluble ions (SO₄²⁻, NO₃⁻, Cl⁻, NH₄⁺, Na⁺, K⁺, Mg²⁺, and Ca²⁺) of PM2.5 and trace gases (HCl, HNO3, HNO2, SO2, and NH3) were measured by an online analyzer (MARGA).In addition, the aerosol number concentration distribution (SMPS3938+APS3321), and sizeresolved chemical compositions (MOUDI 122) were also synchronously measured during these three observation periods.
In this paper, we mainly focus on the discussions of f(RH).In the near future, the aerosol hygroscopicity would be comprehensively evaluated by making a use of all the results from the above-mentioned observations and would be published in the following papers.
In order to facilitate comparison, we transform the absorption coefficient of 670nm into that of 525nm according to the assumption that absorption is inversely proportional to wavelength (Bond and Bergstrom, 2006;Liu et al., 2018).Thus, the SSA at 525nm (SSA525nm) was the proportion of the scattering coefficient to the sum of scattering coefficient and absorption coefficient.
When the air is very clean, the relative change and fluctuation of the particle concentration in the ambient air would be more intense.Furerthmore, the airflow into two nephelometers could not be completely synchronized during the observation.The Scattering Ångström exponent is generally regarded as an indicator of particle size.
In winter, lower Å450-635 was observed in less polluted conditions.As depicted in Figure 2, the clean conditions in winter occurred mainly in the case of northwest wind with relatively higher wind speed, which led to a greater proportion of larger particles such as crustal dust in the air.Conversely, the Å450-635 in polluted condition was lowest in summer.Compared with winter, the relative humidity in summer was much higher, especially under the condition of pollution, which could make particle collision and coagulation easier to occur Guo et al (2014).
According to the ZSR (Zdanovskii-Stokes-Robinson) assumption (Zdanovskii, 1948;Stokes and Robinson, 1966), the κ value of a multicomponent particle is equal to the volume weighted average of each component.As depicted in Figure 1, the f(80%) at 525nm is highly correlated with the fractions of all the water-soluble ions to PM2.5.
Owing to the proportion of hygroscopic components increased, PM2.5 had higher f(80%) in the polluted conditions.The standard deviations (SD) of f(80%) also indicate that the changes of fractions of hygroscopic components were relatively small when the PM2.5 over 35μg m -3 in summer and autumn.The wind roses obviously reveal the differences in the hygroscopicity and chemical compositions of PM2.5 from different directions.some other studies of the North China Plain (Table 2).Overall, the diurnal variation of f( 80) is not obvious, and the average f( 80) at 12 to 16 pm was slightly higher (Figure 3).

Parameterization schemes of f(RH)
To better describe the dependence of f(RH) on RH, many different empirical expressions have been applied in previous studies to fit the f(RH) measurements.Kotchenruther et al. (1999) proposed that different fitting equations should be used according to the observed curve structure.For monotonic curves in which f(RH) varies smoothly with RH, they proposed the use of equation reported by Kasten (1969) and its variants.The most commonly used equation is the one-parameter fit equation (Hänel, 1980;Gassó et al., 2000;Brock et al. 2016 ) and (Kotchenruther and Hobbs, 1998;Carrico et al., 2003;Zieger et al., 2011;Chen et al., 2014).For deliquescent curves, Kotchenruther et al. (1999) introduced a more complex equation, and more detailed information and fitting equations could be found in Titos et al. (2016).
In this work, four commonly used empirical parameterization schemes were chosen to describe the monotonic curves of f(RH) variation:   3, we find that the f(RH) values with RH above 80% from fitting curves for clean and polluted conditions in Chen's study are evidently lower than those from the respective curves for very clean and polluted conditions in this work.It indicates that the scattering enhancement due to moisture uptake or hygroscopicity of aerosol under high RH is probably higher than before.
In addition, the averaged f(RH) at 450nm, 525nm, and 635nm was also fitted separately for each season.It is clear that the f(RH) showed a dependence on wavelength, especially in summer.The averaged f(RH) increased with increasing wavelength.Similar results were also obtained by Zhang et al. (2015) at Lin'an, China and Zieger et al. ( 2014) at a regional continental research site in Melpitz, Germany.
However, we found that when under very clean conditions in winter and autumn, the mean value of f(80) at 450nm was higher than that at 525nm and 635nm (Fig. S4), and it is more obvious in winter.Through further curve fitting of f(RH) at different wavelengths for three seasons, it is also found that the f(RH) curve of 450nm was evidently higher than that of 525nm and 635nm only under very clean conditions in winter.Our previous work showed that the sulfates, nitrates, and ammonium (SNA) were abundant in aerodynamic diameter of 0.18~1.0μmon clean days in winter with the mass median diameters (MMD) of SNA at about 0.45μm.However, the SNA was mainly concentrated in 0.32~1.8μmwith evidently higher MMDs on polluted days or in other seasons (Zhao et al., 2017;Su et al., 2018).The high fraction of SNA in particles below 500 nm might be responsible for the higher f(RH) at 450nm when under very clean conditions in winter.

Uncertainty analysis for f(RH) measurements
As mentioned above, the humidified RH in the wet nephelometer was calculated through the derived average vapor pressure and the temperature measured by the sensor in the chamber.According to the differences in vapor pressure values from the sensors at the inlet and outlet, a relative error of 0.5% could be calculated for the vapor pressure data.And the mean absolute error of the temperature measurement in the nephelometer was 0.2℃.Then, a Monte Carlo simulation was utilized to estimate the uncertainty of calculated humidified RH values.New values of vapor pressure and temperature were simulated by adding the uncertainties of 0.5% and 0.2℃ to the observation data (following a normal distribution).And new humidified RH could be calculated using the simulated data, and this procedure was repeated 1000 times for each humidified RH value.Then the average standard error of humidified RH in the wet nephelometer could be calculated to be 0.85%.
Next, the Monte Carlo simulation was used again.The RH was assumed to range from 20% to 90% with steps of 1, and assuming that γ ranges from 0 to 1 with steps of 0.01.The chosen interval for γ covers the particle types from non-hygroscopic aerosol particles to very hygroscopic particles.Therefore, more than 7000 conditions were simulated, and each condition corresponded to one set of RH and γ.For each condition, the dry scattering coefficients with a wide range of 1 to 1000 Mm -1 were selected 5000 times as random numbers to present different atmospheric situations and aerosol loads.
The wet scattering coefficients were calculated associated with the previously selected dry scattering coefficient and the f(RH) calculated by Eq. (2).A random error (following a normal distribution with a standard variation of 0.85%) was also added to the RH and the parameter a was set 1.0 when calculating the f(RH).According to the manual of nephelometer Aurora 3000, the standard error of aerosol scattering coefficient is 2.5%.Then we simulated the dry and wet scattering coefficients by assuming that they both had an uncertainty of 2.5% (following a normal distribution).
Thus, the simulated f(RH) can be calculated again.
The mean and relative standard deviation of simulated f(RH) were calculated for each RH and γ and are shown in Fig. 7.For aerosols in this work (γ ~ 0.35), f(RH) errors were below 4% with RH lower than 85% while reached 9.7% when RH= 96%, which can be regarded as a conservative estimation.And other unpredictable factors contributing to f(RH) uncertainty have not been considered in this approach.

Figure 7 4. Conclusions
A wide-range (30%-96%) and high-resolution humidified nephelometer system was developed and a measurement campaign was conducted to study the f(RH) for three seasons in 2017.The f(RH) at higher RH had firstly been monitored and reported.
Based on the overview of the optical properties and f(RH) of PM2.5, we found that the higher SSA525 values generally occurred under the southerly wind components with higher PM2.5 concentrations.The f(80%) at 525nm of PM2.5 was evidently higher under the polluted conditions and highly correlated with the fractions of all the water-soluble ions.The average f(80) at 12 to 16 pm was slightly higher than that of other periods from the average diurnal variations.
By comparing the fitting results and curves of four different empirical parameterization schemes, it was found that one of the two-parameter fit equations can better fit the observed f(RH) data.The fitting curves could be widely applied to the studies of atmospheric visibility, radiative force, or liquid water content due to aerosol moisture absorption.For summer and autumn, the f(RH) points of polluted conditions were more concentrated near the fitting curves in the scatter plots with higher fitting R 2 .
And the fitting curve under the very clean condition was lower than that of other conditions for each season.Compared with the fitting curves in the previous study, the hygroscopicity of aerosol under higher RH has probably been enhanced.In summer, the fitting f(RH) showed a significant dependence on wavelength and increased with increasing wavelength for each pollution condition.Nevertheless, the f(RH) curve of 450nm was evidently higher than that of 525nm and 635nm under the very clean condition in winter, due to the higher fraction of SNA in particles below 500 nm.
The uncertainties of f(RH) were simulated by considering all the predictable uncertainties or errors during the measurement, which was below 10% for RH up to 96%.
Fig.1shows an overview of the hourly averaged light scattering coefficients (σsca,525nm), absorption coefficients (σap,630nm), single scattering albedo (SSA630nm), and scattering Ångström exponent (Å450-635) as well as f(RH) at RH=80% (f(80%)) for PM2.5.The average values of optical parameters and f(80%) in different seasons and under different PM2.5 pollution levels are listed in Table1.The PM2.5 pollution was heaviest in the winter observation period and lightest in summer.The scattering coefficient and absorption coefficient also show the same trends.Single scattering albedo is one of the most important parameters in estimating of the direct aerosol radiative forcing.The SSA525 increased with the aggravation of PM2.5 pollution in all three seasons, indicating that the components with strong scattering ability, such as secondary ions, increased significantly during the pollution process.The wind rose of SSA525 in Figure2also indicates that the higher SSA525 values generally occurred under the southerly wind condition which was often accompanied by higher PM2.5 concentrations.
f(RH)=a (1-RH/100) -γ(RH/100) Chen et al. the magnitude of the scattering enhancement, which is not affected by the RH.The comparison of the fitting results for different expressions is shown in figureS1to S3.According to the fitting results, R 2 , and the comparison between fitting f(80%) values and measured ones, we finally choose the Eq.(2) to describe the scattering enhancement due to monotonic hygroscopic growth.

Figure 4 .
Figure 4. Fitting f(RH) curves under different pollution levels in winter.

Figure 5 .
Figure 5. Fitting f(RH) curves under different pollution levels in summer.

Figure 6 .
Figure 6.Fitting f(RH) curves under different pollution levels in autumn.

Figure 7 .
Figure 7. Simulated f(RH) and its error (color scale) as a function of RH and the hygroscopic parameter γ.

Table 2 .
Comparisons of average f(80%) in different campaigns of NCP area.