Visibility Degradation and Its Contributors at an Urban Site in Korea

In order to provide a better knowledge of visibility degradation during the PM2.5 event day (episodic high PM2.5 level, hereafter called as “event day”), the relationship between visibility and the chemical species of PM2.5 measured in Gwangju, Korea was estimated. Moreover, a visibility forecasting model was constructed by a statistical approach. The diurnal variation of visibility and PM2.5 concentration on the event day indicated that as the concentration of PM2.5 increased, more light was absorbed and scattered, resulting in visibility deterioration. The averaged visibility during the event day was 7.9 km, which was almost three times lower than that observed during a non-event day. Although the hygroscopic growth of aerosol was not considered in this study, it has been proved that NH4NO3 and organics dominantly contributed to the light scattering during the PM2.5 event day in Gwangju, Korea. The visibility determined in this study had also a negative correlation with PM10, nitrate, relative humidity, EC, OC, and sulfate. Meanwhile, visibility was positively linked with wind speed and temperature. The results of interrelationship and a multiple regression model suggest that among the meteorological variables, temperature was the main variable that influenced visibility.


INTRODUCTION
Fine particles, especially those with diameter 2.5 micrometers or less (generally called as PM 2.5 ), are believed to be the main cause of reduced visibility.When PM 2.5 in the atmosphere scatters light, haze is caused and reduces the distance people can see and obscures the color and clarity of the objects.Visibility of less than 100 m is usually reported as zero.In this condition, roads and airports may be closed.Due to repeatedly poor visibility at many airports around the world, sometimes flights are delayed or cancelled.
Visibility degradation is mainly caused by the absorption and scattering of light by particles and gases.The field observation by Kim et al. (2008) suggested that the light scattering by particles accounted for 52.8 to 81.3% of the worsening of visibility at the urban site of Gwangju, Korea.Among various PM 2.5 components, major four kinds are the most effective at reducing visibility.They are sulfates, nitrates, organic carbon (OC), and elemental carbon (EC, also called as BC (Black Carbon)) (Sloane et al., 1991).
The contribution of aerosol components to the lightscattering varies according to the region.Wang et al. (2015) reported that the apportionment contributions from organic, sulfate, ammonium nitrate and ammonium chloride were 54%, 24%, 12%, and 10%, respectively in Beijing, China during a wintertime intense haze episode.They also mentioned that the organic components contributed greatly to the light extinction (about 45% in Shenzhen, China (Yao et al., 2010) and 9-50% in the Eastern USA (Watson, 2002)) by quoting other research papers.
It is significantly meaningful to study visibility based on the chemical compositions of aerosol obtained through a real-time chemical analysis.However, the real-time or quasi-real-time monitored chemical compositions of PM 2.5 have seldom been applied.
In the present study, the relationship between visibility and the simultaneously measured chemical composition of PM 2.5 was estimated.Moreover, the simplified visibility forecasting model was constructed by a statistical approach based on the routinely monitored PM 2.5 and meteorological parameters.If visibility can be easily forecasted through the routinely monitored data, we can protect ourselves from various risks of our health and our daily lives.

1 Description of the Monitoring Site
Gwangju Metropolitan City, the sixth largest city in South Korea, is located in between a mountainous area in the east and a plains area of the southwestern part of the Korean Peninsula.Gwangju has a population of 1.49 million people over an area of 501.2 km 2 .The annual average temperature is 15°C and the highest average is 33.1°C in August.
As a major source of air pollution, about 636,743 motor vehicles including 92,396 heavy duty are registered in Gwangju (The car registration number by year, 2016).The main industries of Gwangju are automobile, home appliances, high-tech components, semicondu ctor, and cultural contents.
Monitoring was carried out on the roof of a three-story building (13.5 m) of the Honam area air pollution inten-sive monitoring site (35.23N; 126.85E).The surroundings of this monitoring site are education and research facilities, and residential and commercial areas with some minor point sources such as electronic manufactories and precision machinery industries.Six-lane roads, usually with light traffic, are located 50 m all around the monitoring site.

2 Instruments and Measurements
The hourly data of the ionic species in PM 2.5 were directly obtained using the URG-9000D Ambient Ion Monitor (AIM, URG Co.).AIM has an additional ion detector to allow for time-resolved direct measurements of both cations and anions.PM 2.5 is drawn into the AIM Monitor through a PM 2.5 sharp-cut cyclone and then the sample is drawn through a Liquid Diffusion Denuder.PM 2.5 laden air stream next enters the Aerosol Super-Saturation Chamber to enhance particle growth.The enlarged particles in the An Inertial Particle Separator are then injected into the Ion Chromatograph.Ion Chromatography (IC) (ICS-2000, Dionex Co.) was used to determine the concentrations of various ions present in PM 2.5 .Before running the collected PM samples, the IC system was calibrated using each standard solution.By comparing the data obtained from the PM samples to those obtai ned from the known standard solutions, the ionic species of PM samples could be identified and quantitated.Samples were analyzed for ammonium, nitrate, and sulfate.Detection limits of ammonium, nitrate, and sulfate are 0.339, 0.020, and 0.023 μg/m 3 , respectively.
The comparison study between the filter-based laboratory IC technique and AIM method conducted by Beccaceci et al. (2015) showed an overall good correlation (R 2 >0.83) for ammonium, nitrate, and sulfate.
The concentration of OC and EC in PM 2.5 collected on the quartz filters for 45 minutes with a flow rate of 8 lpm was determined using the thermal-optical transmittance (TOT) method.TOT method was designed for the analysis of the carbonaceous fraction of particulate diesel exhaust based on the National Institute of Occupational Safety and Health method 5040 (NIOSH 5040) (Karanasiou et al., 2015).In order to estimate the carbon of the blank filter contaminated during handling and transport, the laboratory blank filters were also analyzed in the same way as samples.Carbon dioxide produced in the analytical procedure of OC and EC was mea sured by the Non-Dispersive Infra-Red (NDIR) detectors that are the industry standard method of measuring the concen-tration of carbon oxides (CO and CO 2 ).The limit of detections (LODs) of OC and EC were calculated as 0.65 and 0.02 μg/m 3 , respectively.Karanasiou et al. (2015) gave a full detail of TOT method.
In order to measure a highly variable light-scattering coefficient, the three-wavelength (i.e., 450 nm (blue), 550 nm (green), and 700 nm (red)) Integrating Nephelometer with backscatter shutter (Model 3563, TSI Scientific Inc., USA) was applied.This unique instrument is one of well-known excellent devices for the measurement of short-term measurements of the light-scattering coefficient of ambient particles.Sensitivity to light-scattering coefficients is as low as 2.0 × 10 -7 per meter (60second averaging time) (Anderson and Ogren, 1998).Integrating Nephelometers measure the angular integral of light scattering that yields the quantity called the scattering coefficient, which is used in the Beer-Lambert Law to calculate total light extinction (Bodhaine et al., 1991).
Light-absorption coefficient of particles was measured by the Aethalometer (Magee Scientific Inc., USA).The aethalometer is the earliest and most common method used for a real-time readout of the concentration of BC in an air stream.Aethalometers are now constructed to perform their optical analyses simultaneously at multiple wavelengths, typically spanning the range from 370 nm (near-ultraviolet) to 950 nm (near-infrared).The Aethalometer collects the sample on a quartz fiber filter tape, and performs a continuous optical analysis, while the sample is collecting (Hansen et al., 1982).
All monitoring and measurement instruments were intensively operated from May 26 to June 15, 2016.

3 Determination of Visibility
Visibility is generally described as the maximum horizontal distance at which a target with a sky background can be visually observed by our eyes.As the early theory of visibility, Koschmieder (1924) developed below formula (i.e., Koschmieder's formula).
where σ ext is extinction coefficient and it is the total attenuation of visible radiation due to scattering and absorption by gas molecules, aerosols, and other components (e.g.fog and cloud droplets) in the atmosphere.
The σ ext may be divided into a scattering coefficient, σ scat and an absorption coefficient, σ abs : The scattering and absorption coefficients may therefore be further subdivided into Therefore, σ ext is ultimately composed as follows.
where the subscripts of s, a, g, and p denote scattering, absorption, due to gas molecules, and due to PM, respectively.
The σ sg , i.e., scattering due to gas molecules is also called 'Rayleigh scattering' .It was calculated by Rayleigh scattering efficiency 0.12 × 10 -4 /m (Malm et al., 1996).The σ ag was determined by 3.3 × NO 2 concentration (ppm) × 10 -4 (Malm et al., 1994).The σ sp was directly measured by a Nephelometer.The σ ap could be calculated by multiplying the BC concentration (ng/m 3 ) measured by Aethalometer by a factor of 8.28 m 2 /g.The conversion factor of 8.28 m 2 /g used in this study was applied in Beijing previously by Yan et al. (2008) and He et al. (2009).

4 Concept of Multivariate Analysis for Visibility
Forecast A multiple regression analysis was applied to the visibility forecasting.It can be performed based on the relationship between multiple explanatory variables and a response variable by fitting a linear equation as follows: where Y i , C 0-4 , x 1-4 , and e are response variable (visibility), constants, explanatory variables (PM 2.5 , temperature, wind speed, and relative humidity), and random error, respectively.

1 Diurnal Variation of Visibility and PM 2.5
Photo 1 shows the surrounding from the ground station of this study on a non-event day (a clean day with the PM 2.5 concentration of 21 μg/m 3 ) (top) and the event day when PM 2.5 reached 63 μg/m 3 (bottom).As shown in the photo, it is possible to find out the clear difference between two scenic views.
The diurnal variations of visibility on the event day and a non-event day are drawn in Fig. 1.Visibility shows obvious diurnal variations on both the event day and a non-event day.The diurnal variations will be different depending on the situations of the day, e.g., human activ-ities, the local meteorological elements, and PM 2.5 (domestic and an inflow of from abroad).On a nonevent day, it ranged from 12.9 to 39.7 km with an average 22.9 km.The highest visibility appeared at around 5 p.m. and then rapidly decreased until around 3 a.m.local time.Meanwhile, during the event day, it showed a gentle time-variation with a significantly reduced value ranged 5.1-13.4km with an average 7.9 km.The peak value occurred in the early afternoon between 1 p.m. and 2 p.m. local time and then it decreased until the morning rush hour of the next day.Fig. 2 shows the time dependent variation of the visibility with the measured PM 2.5 during the event day.The environmental standard of PM 2.5 in Korea is also displayed in the figure.The visibility was determined by the ε sca measured by a Nephelometer.
During the event day, the daily average mass concentration of PM 2.5 ranged from 42 to 83 μg/m 3 with an average concentration of 63 μg/m 3 .The time serial PM 2.5 was severely fluctuated throughout the whole event day.Unlike non-event day, the rush-hour peak of PM 2.5 was not found.This result indicates that, added to the domestic factors including the local sources, the inflow of PM 2.5 from abroad had a significant influence on the di urnal variation of PM 2.5 on event day.This unusual hourly variation of PM 2.5 at Gwangju on the event day has also been introduced in a previous study (Park et al., 2013).Meanwhile, a symmetrical temporal variation with the concentration of PM 2.5 indicates that visibility has a strong inverse proportion PM 2.5 .The opposite diurnal patterns between PM 2.5 and visibility has also been measured in an urban area of Northeast China by Zhao et al. (2017).
Photo 1.Comparison of visual characteristics from the ground station of this study on a non-event day (PM 2.5 : 21 μg/m 3 ) (top) and the event day (bottom) when PM 2.5 reached 63 μg/m 3 .Fig. 1.Diurnal variation of visibility on the event day and a non-event day.

2 ε sca Neph. and Three Kinds of Particles
To evaluate the contribution of each particle type to the ε sca measured by a Nephelometer (ε sca Neph.), the scatter plots of ε sca Neph.and the theoretically reconstructed mass concentration of major three particle types (i.e., (NH 4 ) 2 SO 4 , NH 4 NO 3 , and organics) were drawn Fig. 3 (on the event day) and Fig. 4 (on a non-event day).Sulfate was assumed to be fully neutralized (i.e., (NH 4 ) 2 SO 4 ).
The mass concentration (m) of (NH 4 ) 2 SO 4 , NH 4 NO 3 , and organics was theoretically calculated.The m of NH 4 NO 3 , and (NH 4 ) 2 SO 4 described as following equations was calculated from the IC results of ammonium, nitrate, and sulfate ions under assumption that the combined forms of NH 3 with nitrate and sulfate were NH 4 NO 3 and (NH 4 ) 2 SO 4 , respectively.The mass of organics, m Organics , was determined by multiplying the amount of OC concentration (μg/m 3 ) by 1.6 (Cabada et al., 2004;Countess et al., 1980).
As shown in Fig. 3, the high correlations (r values varied from 0.74 to 0.91) between the ε sca Neph.and three particle types appeared on the event day.The high correlation coefficients, especially, with NH 4 NO 3 and organics, indicate that the nitrate and organic compounds dominantly contributed to the light scattering in Gwangju during PM 2.5 event day.The similar results have been obtained in the studies carried out in the urban sites of Taiwan (Kuo et al., 2013) and China (Yu et al., 2019).These studies suggested that the visibility degradation due to nitrate was much higher than that due to sulfate.However, these results are a little bit different from that of Detroit's summertime atmosphere reported by Wolff et al. (1982).They suggested that sulfate was the most efficient light-scattering species per unit mass of dry weight.
According to the study conducted by Tao et al. (2007) in Guangzhou, China, the correlation coefficients of sulfate, nitrate, and OC between visibility were -0.66, -0.64, and -0.63, respectively.Jung et al. (2009) also reported that (NH 4 ) 2 SO 4 , NH 4 NO 3 , and organics contributed 42.2%, 24.9%, and 9.0% of light extinction coefficient (b ext ) in Beijing during the maximum polluted period in 2006.
Although there were slight differences in contribution rates in various studies reported previously, there is no doubt that the major secondary particles contribute greatly to visibility deterioration.
Meanwhile, as shown in Fig. 4, the ε sca Neph.has a strong correlation (r = 0.94) with only NH 4 NO 3 in the nonevent day.However, a related study carried out by Yu et al. (2019) in an urban-industrial site in China during the recent wintertime suggested that sulfate was the largest contributor for the clean period.The results in Fig. 4 also indicate that there was no apparent correlation between OM and the ε sca Neph.(r = 0.47) on a non-event day.

3 Interrelationships among Particle
Components, Meteorological factors, and Visibility The interrelationships among components (visibility, sulfate, nitrate, OC, EC, temperature (Temp.),wind speed (W.S.), relative humidity (Rh), and PM 2.5 ) on the event day were analyzed (Fig. 5).The visibility on the event day had negative correlations with PM 2.5 , PM 10 , nitrate, relative humidity, EC, OC, and sulfate.Especially, PM 2.5 and PM 10 showed fairly strong negative correlation coefficients of -0.76 and -0.73 with visibility.Meanwhile, visibility had a slightly high correlation coefficient with temperature (r = 0.65) and wind speed (r = 0.59), while it has a negative correlation with relative humidity (r = -0.58).This result is remarkably consistent with that of Xue et al. (2015).Xue et al. (2015) reported that the visibility of Shanghai reached about 25 km at a time when temperature and wind speed were high, while visibility decreased to 16 km under the weather type of low wind speed and temperature, and high relative humidity.High temperature can promote the dispersion of the air pollutants, as a result, cause higher visibility.High wind leads to unstable meteorological conditions which accelerate the dispersion of pollutants, and then contribute to improved visibility.High relative humidity can induce the increasing of PM concentration because aerosol hygroscopic incr eases significantly.Finally, the increasing of PM, especially PM 2.5 , can induce the aerosol scattering capability and the visibility decreasing.

4 Visibility Forecasting Model Based on the
Routinely Collected Measurement Data Ann et al. (2000) reported that visibility showed strong seasonal fluctuations in Seoul, Korea with the worst in spring.According to their study, visibility was less than 10 km in Seoul on most days of spring.Therefore, if it is possible to easily predict the visibility even in spring, it will be very helpful in our daily lives.
In this study, a simple visibility forecasting model was constructed by a statistical approach.The advantage of our model is that it does not require special observations using professional measuring devices to obtain the data needed to build the model.Our statistical model is based on the routinely monitored PM 2.5 and meteorological data from the ambient air monitoring stations widely distributed throughout the country.

4. 1 Multicollinearity Test among Explanatory
Variables Fig. 6 shows the statistical procedure for the multi-step regression model with step-wise induction of variables.PM 2.5 , temperature, wind speed, and relative humidity data were selected as the explanatory variables.These data can be taken in routinely from the Honam area air pollution intensive monitoring site in Gwangju Metropolitan City.
At first, in order to check the multicollinearity among explanatory variables, a multicollinearity index known as the variance inflation factor (VIF) was calculated by the following equation (Hossain et al., 2010): where n is the number of predictor variables and R 2 i is the square of the multiple correlation coefficient of the i th variable with the remaining (n-1) variables.
If VIF ranges from 0 to 5, there is not the multicollinearity problem (Hossain et al., 2010).The calculated VIFs among the combination of four explanatory variables are listed in Table 1 and they varied from 1.00 to 1.03.It can therefore be said that there is no correlation among four selected explanatory variables.

4. 2 The Built Model Formula
The coefficients for four kinds explanatory variables determined by a multi-step regression model are also listed in Table 1.The constructed visibility forecasting model was as follow: As expected, the constructed model suggests that PM 2.5 is the most important contributor to worsening visibility.Among the three potential meteorological variables, temperature is the most influential factor in determining the visibility reduction.Our model also indicates that wind speed, along with temperature, contributes to the improvement of visibility.This may seem reasonable because the wind plays a major role in spreading pollutants.Meanwhile, relative humidity, as shown already in Fig. 5, has a role to play in reducing visibility.
The impact of meteorology on visibility was also revealed in the field study conducted in Shanghai, China by Xue et al. (2015).They reported that visibility was reduced to 16 km under the weather condition of low wind speed and temperature, and high relative humi dity.

4. 3 Test for Residual Normality
A residual plot check is a key part of next steps after model-building.Here, the residual means the values for the difference between the observed data and the predicted data.Examining residual normality suggests that the assumptions were reasonable in the process of building the model and that the model selection was appropriate.In general, the overall pattern of the residuals should be normally distributed.
As shown in Fig. 7, the bell-shaped frequency of regression standardized residuals with a normal density function on the histogram indicates that the model built in this study is very reasonable.

4. 4 Model Validation
To validate the built model in this study, the predic-ted visibilities by model were compared with those of actually observed at the Gwangju Regional Meteorological Administrations (GRMA).The GRMA (35.17N; 126.89E) is about 9 km away from the Honam area air pollution intensive monitoring site.Fig. 8 shows the scatter plot of the observed and predicted visibilities with a dotted regression line.A very high correlation (r 2 = 0.892) between the observed and predicted visibilities has been proved.It can therefore be said that the model designed in this study can predict visibility with accuracy, and predicted results can be useful in our daily lives.

CONCLUSIONS
The results of this study verified that a high PM 2.5 mass concentration is one of the important factors in the deterioration of visibility once again.The visibility   during PM 2.5 episode was also closely linked to particle kinds, specially the fine secondary particles.Through the correlation analysis between the measured ε sca by a Nephelometer and the theoretically reconstructed mass concentration of three major types of particles, it can be concluded that NH 4 NO 3 and organics dominantly contributed to light scattering during the PM 2.5 event day in Gwangju.Although a large amount of measured data must be used to build a good regression prediction model, the model built in this study used the monitored data over a limited period of spring.Therefore, it is not reasonable to predict the visibility of the whole season.However, it is expected that our results will contribute to establishing measures for our daily life by predicting visibility during the springtime when episodically high PM 2.5 concentrations are often obser ved.

Fig. 2 .
Fig. 2. Diurnal variation of visibility and PM 2.5 concentration on the event day.The South Korea's PM 2.5 criteria is the daily mean value observed every hour.

Fig. 7 .
Fig. 7. Approximately normal distribution of standardized residuals produced by a model for a calibration process.

Table 1 .
The coefficients for four kinds variables by multi step regression model.