Abstract

This paper presents a technique used to empirically classify operationally defined “haze” days in South Korea from 2000 to 2007 into long-range transported (LRT) and stagnant (STG) haze cases. A total of 547 haze days were classified into these two cases by tracking consecutive 6-day synoptic weather charts and air trajectories. The meteorological features associated with long-range transport of haze were identified by contrasting the values of 35 candidate meteorological parameters corresponding to the two types of haze cases. A suitable subset of synoptic variables was then chosen to diagnose the unique meteorological features of each case. The statistical test showed that geostrophic wind speed, vorticity, vorticity advection at a geopotential height of 850 hPa, and the vertical stability index of the lower atmosphere were indicated as highly effective parameters for distinguishing between the LRT and STG cases. The classification accuracies showed 93.2%, 87.8%, 85.4%, and 84.4% for these four variables, respectively. The STG case was well characterized by negative vorticity, with stable atmospheric stability conditions and weak geostrophic wind speed, that is, ~2.8 m s−1 at a geopotential height of 850 hPa, whereas the LRT case had relatively strong geostrophic wind speed, >6 m s−1. For both cases, the location of the anticyclone played an important role in haze occurrence, directly and indirectly. A high pressure system led to stable STG haze with weak ventilation, resulting from upper-atmospheric subsidence. The LRT case was associated with a strong anticyclone that prevailed over southwestern China, maintaining the pressure gradient force that generated the westerly wind that was persistently conducive to downwind long-range transport of haze.

1. Introduction

Atmospheric aerosols—in particular, small particles having diameters between 0.1 and 1 μm—interact strongly with short wavelengths of light, and at high atmospheric concentrations they lead to low-visibility conditions identified as haze; haze can also adversely affect human health (Bowman and Johnston 2005; Park et al. 2006; In et al. 2007). Aeolian dust, volcanic ash, and smoke from wildfires and agricultural burning can all contribute to atmospheric haze. In many regions, however, haze mainly consists of particles of anthropogenic origin, derived from industrial, home-heating and -cooking, and transportation sources.

Haze phenomena in a particular region may be caused by the trapping of direct particle emissions and/or local secondary generation of particulate matter through photochemical reactions of air pollutants. Asian mineral dust frequently originates directly from the source regions, and oxidation of sulfur dioxide to sulfate aerosol is one of the most important processes in eastern Asia in the formation of secondary aerosol particles. The haze can be transported from its original source regions over long distances in the windward direction to the receptor site, however. For example, haze generated over eastern Asia can be transported to the western Pacific Ocean region (Yu et al. 2008; United Nations Environment Programme 2002). Frequent and thick veils of haze have occurred in eastern China and have been observed through satellite imagery. As the frequency and intensity of such haze events increase, the impact on air quality at downwind receptors due to the long-range transport of haze is expected to accelerate rapidly (Sun et al. 2006). The latitude and longitude of the Korean Peninsula place it in prevailing westerlies and thus downwind of many Asian pollution sources. Previous studies have reported that haze phenomena in South Korea are influenced not only by the local emission but also by the long-range transport of pollutants from China (Lee et al. 2001; Davis and Jixiang 2000). Haze phenomena occur over South Korea in all seasons equally, whereas the long-range transport of Asian dust occurs mainly in the springtime, especially in April (Chun and Lim 2004).

The results from many analyses of haze phenomena, including the analysis of physical characteristics as well as measurements of the chemical components, have been previously reported (Levin et al. 2005; Sun et al. 2006; Wang et al. 2006). There are relatively few haze studies, however, that are about meteorological processes in particular and to what extent the long-range transported haze over both the source and receptor areas can contribute to degraded air quality. The goals of this study were to investigate the meteorological features of two haze-generation-mechanism classifications—stagnant (STG) haze and long-range transported (LRT) haze observed over Korean urban areas—and to develop an in-depth understanding of the representative meteorological characteristics of both haze types. We also characterized LRT haze cases arising from two different sources by tracking the upwind pathways of haze-laden air masses.

2. Method and data sources

Since we sought to identify the meteorological features of LRT haze by contrasting two individual groups of haze cases, the LRT and the STG cases, accurate haze classification was essential to reach a robust conclusion. Numerous approaches toward weather classification have been reported, including empirical or qualitative approaches through the direct visual inspection of weather maps (Heidorn and Yap 1986; Sanchez et al. 1990); semiquantitative approaches, which apply both statistical and qualitative methods (Scott and Diab 2000; Kallos et al. 1993); and fully quantitative approaches, which employ automated cluster analysis techniques such as nonhierarchical (K mean), hierarchical clustering, or combined two-stage linkage methods (Eder et al. 1994; Davis and Kalkstein 1990). All of these approaches are related to mean air pollution spatial patterns in the region of interest but are not connected to the nonstagnant air masses during transport processes over the region of interest. Thus, we undertook a new semiempirical method of tracking the moving air mass to distinguish between LRT and STG haze cases. We used meteorological data observed over 8 yr (2000–07) in the South Korean cities of Seoul and Busan, as indicated in Fig. 1, to identify days that were likely to be affected by anthropogenic pollution. Weather codes and hourly observed meteorological observation data provided by the Korea Meteorological Administration included precipitation, relative humidity, visibility, and cloud cover. In this dataset, days that were unequivocally influenced by Asian dust were denoted by a separate category and thus removed from further consideration in this work. We classified the days, where possible, as dominated by “haze,” “mist,” or “fog.” Although these terms actually describe a continuum of aerosol-related phenomena associated with increasing amounts of particle-phase water, in this work we conformed to the operational definitions for these conditions used by the Korea Meteorological Administration: haze refers to conditions with daily mean relative humidity < 75% and daily mean visibility between 1 and 10 km, mist refers to conditions with daily mean relative humidity ≥ 75% and daily mean visibility between 1 and 10 km, and fog refers to days with daily mean relative humidity ≥ 75% and daily mean visibility < 1 km. Note that these criteria differ from those in World Meteorological Organization (2008). Using hourly observations and classifying days on the basis of more than 12 h meeting these criteria resulted in less than a 10% change in the number of hazy days, and thus we retained the use of daily means for the classification. For the days without haze, mist, or fog, a “clear” category was defined as a group of cloud-free days, and the remaining days with precipitation were classified into a “rain” category. Others are classified into an “unclassified” category. Twelve percent of the days in Seoul and 1% of the days in Busan were classified into the unclassified category. A total of 547 haze days were recorded in the two cities over our selected time period (Table 1). For each atmospheric classification, we also derived the statistics for available gas- and particulate-phase pollutant measurements (Table 1). These hourly measured ambient air quality data were collected by and made available from the Korean Ministry of Environment.

Fig. 1.

Locations of meteorological observation sites for the selection of haze days. The inner rectangle is the area used for semiempirical classification of haze days into STG and LRT cases.

Fig. 1.

Locations of meteorological observation sites for the selection of haze days. The inner rectangle is the area used for semiempirical classification of haze days into STG and LRT cases.

Table 1.

Statistical classification of the atmospheric conditions in Seoul and Busan for the study period of 2000–07. The values in parentheses indicate standard deviations.

Statistical classification of the atmospheric conditions in Seoul and Busan for the study period of 2000–07. The values in parentheses indicate standard deviations.
Statistical classification of the atmospheric conditions in Seoul and Busan for the study period of 2000–07. The values in parentheses indicate standard deviations.

Next, we further classified the 547 haze days into two types (STG and LRT hazes) by tracking the moving air masses with computed, 6-day trajectories (3 days backward from the selected day, and 3 days forward) using the National Oceanic and Atmospheric Administration/Air Resources Laboratory Hybrid Single-Particle Lagrangian Integrated Trajectory, version 4 (HYSPLIT-4), model (Draxler and Rolph 2012; Draxler 1999). All trajectories were started at 500 m in Seoul at 1200 LST (Fig. 2). Input data for HYSPLIT-4 utilized final-analysis (FNL) meteorological data that are 6-hourly archived National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis data with 2.5° × 2.5° resolution. The haze-type classification was done as follows. 1) Since the synoptic systems that develop over China usually pass over the Korean Peninsula within about 3 days (Park et al. 2002; Shim and Park 2004; Choi et al. 2009), any cases in which both 3-day backward and forward trajectories remained within the inner box (Fig. 2a) during the six consecutive days was designated as an STG case. The inner box has a typical synoptic horizontal scale of 2000 km (area of 4 × 106 km2), covering the entire Korean Peninsula as indicated in Fig. 2. 2) If the air mass was initially transported from the outer region and then passed completely through the inner rectangular box during the six days, the result was designated as an LRT case (Fig. 2b). 3) Other cases belonging to neither these STG nor LRT subsets were defined as unclassifiable haze (Fig. 2c).

Fig. 2.

Schematic representation of the haze classification using 6-day trajectories (see text for full description) for (a) STG cases, (b) LRT cases, and (c) unclassified cases.

Fig. 2.

Schematic representation of the haze classification using 6-day trajectories (see text for full description) for (a) STG cases, (b) LRT cases, and (c) unclassified cases.

Third, 35 synoptic meteorological parameters downloaded from the NCEP–NCAR reanalysis database were diagnosed to determine a suitable subset of these variables that could be used to further characterize the STG and LRT haze cases. We employed the meteorological parameters used for air pollution potential diagnosis in previous studies (Dobbins 1979, 137–139) and supplemented them with additional variables that were helpful in distinguishing between the two haze cases (Davis and Gay 1993; Sanchez et al. 1990). The meteorological variables tested in the current study are shown in Table 3, which is described in more detail below. Most are self-explanatory, but here we describe a few of the less commonly used variables. Vorticity [ζ = (∂υ/∂x) − (∂u/∂y)], vorticity advection [V · ζ = u(∂ζ/∂x) + υ(∂ζ/∂y)], and geostrophic wind Vg are used. Geostrophic wind speeds were calculated by interpolating the value of the geostrophic wind components Vg(ug, υg) over 37.5°N, 127.5°E, a location that is approximately centered in South Korea, using four nearby grid points of reanalysis data. Geostrophic wind direction was calculated using tan−1(ug/υg). The vertical stability index of the lower atmosphere S describes the temperature lapse rate between the 1000- and 850-hPa geopotential-height fields and is defined as T2 minus T1, where T1 and T2 are the temperatures at 1000 and 850 hPa, respectively. Index ED, defined as 100(T2 − TM)/(H850h), is a stability index between the surface and 850 hPa at 0000 UTC. Here, TM is minimum temperature at the surface, H850 is geopotential height at 850 hPa, and h is the height of the observation site. The humidity index EDI between the surface and 850 hPa (=T2 − Td , where Td is dewpoint temperature at the surface), also known as “isoin,” is an indicator of potential precipitation (Davis and Gay 1993). The Montgomery dry potential M, moist static energy Ms, and effective humidity RHeff are respectively defined as M = cpT + gH, Ms = cpT + gH + Lq, and RHeff = 0.3(RHd + 0.7RHd−1 + 0.72RHd−2), where cp represents the specific heat at constant pressure, g is gravitational acceleration, H is geopotential height, L is latent heat, q is mixing ratio, RH is relative humidity, and the subscript di stands for the ith day prior to the observed day (Lee and Park 1997; Won et al. 2010).

We then employed discriminant function analysis (DFA) in an effort to determine which variables accurately discriminate between LRT and STG cases. First, we performed a Student’s t test on each of the 35 candidate meteorological variables to determine whether the values of that variable were statistically significantly different between the LRT and STG cases. The null hypothesis was defined as no difference between the STG and LRT cases, and the significance of the separation between STG and LRT cases was diagnosed by a two-tailed test. We selected those variables for further consideration that had significance p values < 0.05 from the Student’s t test. Next, we performed DFA using the Statistical Product and Service Solutions, Inc., (SPSS) 20.0 proprietary software program for Microsoft, Inc., Windows to determine the classification accuracies (%) for the selected variables. DFA has been used to determine which variable(s) are the best predictors from a set of continuous predictors in discriminating data into two or more groups (Cooley and Lohnes 1971; Dunteman 1984; Ghiaus 2005). In this work, the variables determined to have relatively higher classification accuracies from DFA were highlighted for interpretation of LRT and STG conditions. Last, we composited all of the 850-hPa geopotential-height weather charts computed from the NCEP–NCAR reanalysis data, for each of the retained STG and LRT cases, and diagnosed the meteorological features of each case.

3. Results

a. Statistical characteristics of haze occurrences

Table 1 summarizes the days falling into the six weather conditions considered: haze, mist, fog, rain, Asian dust, and clear weather, as observed in Seoul and Busan for the period of 2000–07. Because mist was observed very often with precipitation, we subdivided mist into (clear) mist and mist + precipitation cases. Table 1 showed that the most frequent cases for Seoul were clear days (41%), haze (16%), precipitation (15%), and mist + precipitation (12%). Air quality data showed that the concentration levels of particulate matter with diameter less than 10 μm (PM10) was highest, with a daily-average value of 125.7 (±53.3) μg m−3, in the Asian dust case. Haze days had the second-highest PM10 concentrations, with a daily-average value of 97.0 (±32.3) μg m−3. The concentrations of all of the gas-phase pollutants [nitrogen dioxide (NO2), sulfur dioxide (SO2), and daily maximum ozone (O3)] ranked highest in the haze cases, however. This finding suggests that the highest levels of gaseous pollutants were associated with the formation of haze, likely through photochemical reactions of both local emissions (from the Korean Peninsula) as well as emissions transported to the Korean Peninsula from other locations. The frequencies of cases for Busan were clear days (55%), haze (5%), precipitation (23%), and mist + precipitation (10%). Days classified as “rain” occur more frequently in Busan, a coastal city, than in Seoul (Table 1), and improved visibility due to frequent washout of pollution results in a lower frequency of haze cases.

Table 2 summarizes the semiempirical transport pattern classification of the hazy days only. The results show that relatively higher air pollution emission in Seoul shows no significant difference in the occurrence frequency of the STG case (67.7%) relative to that of Busan (61.8%). STG cases observed in each city (Seoul or Busan) occur at an overall frequency of 63.8%, showing the preponderance of STG haze cases over the LRT cases, which occur only 27.2% of the time among the 547 total cases for the period of 2000–07 (Table 2).

Table 2.

Semiempirically classified occurrence frequencies of haze phenomena in the Korean Peninsula for the study period of 2000–07.

Semiempirically classified occurrence frequencies of haze phenomena in the Korean Peninsula for the study period of 2000–07.
Semiempirically classified occurrence frequencies of haze phenomena in the Korean Peninsula for the study period of 2000–07.

b. Meteorological features associated with haze phenomena

Table 3 shows the averaged, daily-mean synoptic meteorological parameters calculated for both the STG and LRT cases, and the p values from the Student’s t test for each of the 35 variables. Variables obtained from two different haze groups were compared using the two-tailed Student’s t test to test if they are significantly different from each other at the 95% significance level. The null hypothesis was first defined as no difference between two groups, and the t test was applied to test the hypothesis.

Table 3.

Meteorological parameters tested in the classification of South Korean haze phenomena. STG and LRT refer to the haze classifications discussed in the text. The p value is the probability of obtaining a test statistic as determined by a two-tailed t test for the significance of the separation between STG and LRT cases. Asterisks represent variables with p values < 0.05.

Meteorological parameters tested in the classification of South Korean haze phenomena. STG and LRT refer to the haze classifications discussed in the text. The p value is the probability of obtaining a test statistic as determined by a two-tailed t test for the significance of the separation between STG and LRT cases. Asterisks represent variables with p values < 0.05.
Meteorological parameters tested in the classification of South Korean haze phenomena. STG and LRT refer to the haze classifications discussed in the text. The p value is the probability of obtaining a test statistic as determined by a two-tailed t test for the significance of the separation between STG and LRT cases. Asterisks represent variables with p values < 0.05.

The results of the statistical tests suggest further emphasis on 17 variables, having p values < 0.05 (starred in Table 3). In comparison with two groups, the STG cases had significantly higher values of surface pressure and 850- and 500-hPa geopotential heights, significantly weaker pressure gradients, and more stable atmospheric conditions relative to the LRT cases. In contrast, the LRT cases had relatively stronger upper-atmosphere wind speeds, and all of the humidity indices, such as surface relative humidity, surface effective humidity, and EDI were found to be drier than the STG cases. These findings reflect that trajectory speed and, to some extent, direction were used to initially differentiate between the LRT and STG cases, and thus high pressure and stable conditions would favor trajectories remaining within the indicated boxed region and the corresponding classification as STG.

Table 4 provides univariate discriminant function (unstandardized coefficients and constants), centroids, demarking points, and classification accuracy of each discriminant function score for the selected variables with p < 0.05. In Table 4, each centroid is the mean discriminant score for each of the STD and LRT cases. Accuracy for distinguishing between STD and LRT cases can be estimated by multiplying the value of each variable with its corresponding unstandardized coefficient and then adding the constant. If the resulting discriminant score is lower than the given demarking point, then the day is considered to be an STG case, whereas a larger score indicates an LRT case. If, for example, a particular daily mean geostrophic wind speed |Vg|850 was 4.5 m s−1, the discriminant score y is calculated as, y = (4.5 × 0.999) + (−3.982) = 0.5135. The result, 0.5135, is above the demarking point, indicating that this day should be classified as an LRT case. It is obvious that when the discriminant score is close to the demarking point the probability of correctly classifying an individual case is lower. The last column in Table 4 gives the accuracy with which each case could be put into the STG or LRT subsets using the 17 retained variables.

Table 4.

Univariate discriminant function and demarking points for those variables that had p values < 0.05.

Univariate discriminant function and demarking points for those variables that had p values < 0.05.
Univariate discriminant function and demarking points for those variables that had p values < 0.05.

From Table 4, we see that the most suitable variables for separation of the LRT and STG cases were meteorological dynamic parameters: geostrophic wind speed |Vg|850, vorticity ζ850, and vorticity advection V · ζ850 at a geopotential height of 850 hPa, with the estimated classification accuracies of 93.2%, 87.8%, and 84.4%. It is not surprising that geostrophic wind speed is the most important variable for discriminating between the two types of hazy day because of the semiempirical method of classification. In any case of STG, the total 6-day trajectory should remain within the inner box in Fig. 2a. Stability index S also shows high classification accuracy: 85.4%. Other parameters having classification accuracies above 70% were surface dewpoint temperature Tσ-00, surface effective relative humidity RHeff, surface minimum temperature Tmin, surface pressure Ps00, ED00, and EDI00, which are mostly thermodynamic or humidity parameter groups (Table 3).

Table 5 shows the identified discriminant function score y by combining all of the variables with p < 0.05. The discriminant function score is expressed as a linear combination of predictors

 
formula

where the ai are the weighting coefficients for the ith variable. The above multivariate discriminant function is interpreted by means of a standardized coefficient for each variable. Given a standardized beta coefficient for each variable, the larger the standardized coefficient is, the greater is the contribution of the respective variable to the discrimination between two groups. Another way to determine which variables define a particular discriminant function is to look at the factor structure. The factor structure coefficients are the correlations between the variables and the discriminant functions and are accepted as being at a significant level for a factor structure coefficient of >0.3.

Table 5.

Multivariate discriminant function for those variables that had p values < 0.05.

Multivariate discriminant function for those variables that had p values < 0.05.
Multivariate discriminant function for those variables that had p values < 0.05.

Table 5 provides the results of multivariational discriminant function analysis including unstandardized and standardized coefficients, a classification function (linear regression coefficients), and factor structure coefficients of each discriminant function score for the selected variables with p < 0.05. Table 5 also demonstrates that the most highly contributing variables to the discriminant function score include |Vg|850, S, and ζ850 at a geopotential height of 850 hPa for separation of the LRT and STG cases, with the standardized coefficients of 0.80, 0.54, and 0.47, respectively, suggesting these efficient three synoptic variables for classification as noticed in Table 4. The factor structure coefficient for these three variables is mostly greater than 0.3, the level of significance grade discussed earlier.

In the STG cases, this is mainly due to stagnation over the Korean Peninsula associated with high pressure systems and the corresponding stable atmospheric profiles. In the LRT cases, however, this is mainly due to strong continental outflow, generally behind the intensified trough. As explained below, this westerly or northwesterly continental flow can be maintained in general by the pressure gradient force exerted by the prevailing anticyclone over southwestern China. The resulting moving trough was associated with a surface low pressure system accompanied by a weak cold front extending eastward through the Korean Peninsula in the LRT cases.

Figure 3 shows scatter diagrams of the three synoptic meteorological variables having the highest classification accuracies (vorticity, stability, and geostrophic wind speeds at the 850 hPa level) on the y axis against daily mean PM10 concentrations on the x axis for all haze events observed in both Seoul and Busan. STG, LRT, and unclassified cases are indicated by different symbols in each plot. Note from Fig. 3 that there is no significant correlation between PM10 and these parameters but that meteorological values are distinct in signs and magnitudes between the STG and LRT cases, as expected from the prior statistical analyses. The STG cases had generally negative vorticities, with an average of −8.3 × 10−6 s−1, whereas the LRT cases had positive vorticities averaging +3.7 × 10−6 s−1 (Fig. 3a). The lower-atmosphere S for the STG cases is positive (stable) with weak geostrophic wind speeds of 2.8 m s−1 on average; for the LRT cases, S is negative (unstable) with strong geostrophic wind speeds ranging from 4.5 to 7.7 m s−1 (Figs. 3b,c). In all of these plots, similar PM10 concentrations could be found in either type of case.

Fig. 3.

Scatter diagram for daily mean concentrations of PM10 and the synoptic meteorological parameters of (a) vorticity, (b) stability, and (c) geostrophic wind speed at 850 hPa for haze events over South Korea.

Fig. 3.

Scatter diagram for daily mean concentrations of PM10 and the synoptic meteorological parameters of (a) vorticity, (b) stability, and (c) geostrophic wind speed at 850 hPa for haze events over South Korea.

Figure 4 shows the 850-hPa geopotential-height composite synoptic map for the STG and LRT cases. The STG cases show a prevailing high pressure system over the Korean Peninsula (Fig. 4b), whereas the LRT cases do not show noticeable pressure characteristics over South Korea but do show high pressure prevailing over southwestern China (Fig. 4a). As described earlier, because the synoptic systems that develop over China usually pass over the Korean Peninsula in about 3 days, we calculated the vorticity and vorticity advection on the 850-hPa geopotential level over the anticyclone center (of the composite synoptic map) located in southwestern China for the preceding 72 h (3 days). The results in Fig. 5 showed distinctive negative vorticity and relatively weaker but positive vorticity advection over southwestern China for the LRT case. This vorticity (Fig. 5a) and vorticity advection (Fig. 5b) in the LRT case showed a contrast to that over South Korea for the LRT case indicated in Fig. 3a, which showed positive vorticity and its negative advection over the Korean Peninsula in the LRT case. This finding indicates that the anticyclone located over mainland China was able to maintain a strong pressure gradient and generated relatively strong westerly winds for the LRT case. This anticyclone prevailing in southwestern China led to frequent onset of westerlies or northwesterlies resulting from the associated pressure gradient, which could advect polluted air masses very well to the Korean Peninsula.

Fig. 4.

The 850-hPa geopotential-height composite synoptic map for the (a) LRT and (b) STG cases.

Fig. 4.

The 850-hPa geopotential-height composite synoptic map for the (a) LRT and (b) STG cases.

Fig. 5.

Scatter diagram of daily mean PM10 concentrations for haze events over South Korea (38°N, 127°E) on the x axis and synoptic meteorological parameters of (a) vorticity and (b) vorticity advection at 850 hPa over China (37°N, 100°E) for the preceding 3 days on the y axis.

Fig. 5.

Scatter diagram of daily mean PM10 concentrations for haze events over South Korea (38°N, 127°E) on the x axis and synoptic meteorological parameters of (a) vorticity and (b) vorticity advection at 850 hPa over China (37°N, 100°E) for the preceding 3 days on the y axis.

Because this synoptic situation was associated with hazy days in our study, we inferred that the pollutants emitted from industrial areas of China enhanced the pollution levels in air masses that were transported toward the Korean Peninsula. Although local pollutants were certainly emitted into these moving LRT air masses as they passed over the Korean Peninsula, because of their speed (relative to the air masses in the STG cases) we do not expect the local emissions to be the dominant source of the haze.

The above-discussed synoptic meteorological conditions can be summarized as follows: When haze phenomena occur over the Korean Peninsula, an anticyclone was involved either directly or indirectly. The STG haze that occurred over South Korea was favored by weak ventilation effects and slower transport processes of pollutants under the prevailing high pressure system. The STG cases had negative vorticity with weak geostrophic wind speeds of less than 3 m s−1 in the lower atmosphere (850 hPa) and were well characterized by stable atmospheric conditions with a sinking upper atmosphere. In contrast, the LRT haze, which was influenced by long-range transport effects, was characterized by an anticyclone over southwestern China, with mostly positive vorticity and relatively strong geostrophic wind speeds of 4.5–7.7 m s−1 over the Korean Peninsula. The anticyclone that prevailed for the LRT cases favored the maintenance of a strong pressure gradient, and the resultant westerlies were able to transport pollutants emitted from central (or northern) China toward the Korean Peninsula.

4. Conclusions

To investigate the characteristics of the physical meteorological conditions associated with haze observed in South Korea, we semiempirically classified the observed haze days into two types, STG haze and LRT haze, using trajectory location and speed through a region centered on the Korean Peninsula. The results of the classification study showed that STG cases, with an occurrence frequency of 63.8%, were dominant over the LRT cases, with a 27.2% occurrence frequency in this study, among the 547 cases for the period from 2000 to 2007. This result indicates that the haze phenomena that occurred over South Korea during this period had a greater contribution from the emissions in South Korea itself than from the LRT pollutants originating from inland China.

We further conducted a statistical study to show that the semiempirical classification scheme could be supported by a more objective classification that is based on observed meteorological parameters. The Student’s t test and DFA analysis yielded the best four meteorological discriminant predictors for discriminating between the STG and LRT cases: vorticity, vorticity advection, geostrophic wind speed at a geopotential height of 850 hPa, and the vertical stability index for the lower atmosphere. The 850-hPa geopotential level indicated that, when STG cases occurred, the Korean Peninsula witnessed the effects of negative vorticity with significantly lower wind speeds and more stable atmospheric conditions than for LRT cases. In contrast, the LRT cases were characterized by positive vorticity and relatively strong wind speeds over the 850-hPa level. In addition, a strong high pressure system over southwestern China was a common feature of LRT cases. This high pressure system location led to frequent onset of westerlies or northwesterlies resulting from the associated pressure gradient, which advected air masses to the Korean Peninsula.

This study has helped to elucidate the characteristics of the physical conditions of LRT haze phenomena over South Korea. Further work is necessary to model quantitative source–receptor relationships in the South Korean atmospheric environment.

Acknowledgments

This subject is supported by the Korea Ministry of Environment as the “Climate Change Correspondence R&D Program” (2012001310002). The authors thank S. Kreidenweis for assistance with the manuscript.

REFERENCES

REFERENCES
Bowman
,
D. M. J. S.
, and
F. H.
Johnston
,
2005
:
Wildfire smoke, fire management, and human health
.
EcoHealth
,
2
,
76
80
.
Choi
,
E.-M.
,
S.-H.
Kim
,
T. M.
Holsen
, and
S.-M.
Yi
,
2009
:
Total gaseous concentrations in mercury in Seoul, Korea: Local sources compared to long-range transport from China and Japan
.
Environ. Pollut.
,
157
,
816
822
.
Chun
,
Y. S.
, and
J. Y.
Lim
,
2004
:
The recent characteristics of Asian dust and haze event in Seoul, Korea
.
Meteor. Atmos. Phys.
,
87
,
143
152
.
Cooley
,
W. W.
, and
P. R.
Lohnes
,
1971
: Multivariate Data Analysis. John Wiley and Sons, 364 pp.
Davis
,
B. L.
, and
G.
Jixiang
,
2000
:
Airborne particulate study in five cities of China
.
Atmos. Environ.
,
34
,
2703
2711
.
Davis
,
R. E.
, and
L. S.
Kalkstein
,
1990
:
Using a spatial climatological classification to assess changes in atmospheric pollution concentrations
.
Phys. Geogr.
,
11
,
320
342
.
Davis
,
R. E.
, and
D. A.
Gay
,
1993
:
A synoptic climatological analysis of air quality in the Grand Canyon National Park
.
Atmos. Environ.
,
5
,
713
727
.
Dobbins
,
R. A.
,
1979
: Atmospheric Motion and Air Pollution: An Introduction for Students of Engineering and Science. John Wiley and Sons, 344 pp.
Draxler
,
R. R.
,
1999
: HYSPLIT_4 user’s guide. NOAA Tech. Memo. ERL ARL-230, 35 pp.
Draxler
,
R. R.
, and
G. D.
Rolph
, cited
2012
: HYSPLIT—Hybrid Single-Particle Lagrangian Integrated Trajectory Model. NOAA Air Resources Laboratory. [Available online at http://ready.arl.noaa.gov/HYSPLIT.php.]
Dunteman
,
G. H.
,
1984
: Introduction to Multivariate Analysis. Sage, 232 pp.
Eder
,
B. K.
,
J. M.
Davis
, and
P.
Bloomfield
,
1994
:
An automated classification scheme designed to better elucidate the dependence of ozone on meteorology
.
J. Appl. Meteor.
,
33
,
1182
1199
.
Ghiaus
,
C.
,
2005
:
Linear fuzzy-discriminant analysis applied to forecast ozone concentration classes in sea-breeze regime
.
Atmos. Environ.
,
39
,
4691
4702
.
Heidorn
,
K. C.
, and
D.
Yap
,
1986
:
A synoptic climatology for surface ozone concentrations in southern Ontario
.
Atmos. Environ.
,
20
,
695
703
.
In
,
H.-J.
,
D. W.
Byun
,
R. J.
Park
,
N.-K.
Moon
,
S.
Kim
, and
S.
Zhong
,
2007
:
Impact of transboundary transport carbonaceous aerosols on the regional air quality in the United States: A case study of the South American wildland fire of May 1998
.
J. Geophys. Res.
,
112
,
D07201
,
doi:10.1029/2006JD007544
.
Kallos
,
G.
,
P.
Kassomenkos
, and
R. A.
Pielke
,
1993
:
Synoptic and mesoscale weather conditions during air pollution episodes in Athens, Greece
.
Bound.-Layer Meteor.
,
62
,
163
184
.
Lee
,
B.
, and
S. U.
Park
,
1997
:
Method for forecasting air pollution potential using synoptic meteorological conditions
.
Asia-Pac. J. Atmos. Sci.
,
33
,
641
656
.
Lee
,
J. H.
,
Y. P.
Kim
,
K. C.
Moon
,
H. K.
Kim
, and
C. B.
Lee
,
2001
:
Fine particle measurements at two background sites in Korea between 1996 and 1997
.
Atmos. Environ.
,
35
,
635
643
.
Levin
,
Z.
,
H.
Gershon
, and
E.
Ganor
,
2005
:
Vertical distribution of physical and chemical properties of haze particles in the Dead Sea valley
.
Atmos. Environ.
,
39
,
4937
4945
.
Park
,
R. J.
,
D. J.
Jacob
,
N.
Kumar
, and
R. M.
Yantosca
,
2006
:
Regional visibility statistics in the United States: Natural and transboundary pollution influences, and implications for the regional haze rule
.
Atmos. Environ.
,
40
,
5405
5423
.
Park
,
S.-U.
,
Y.-H.
Lee
, and
E.-H.
Lee
,
2002
:
Estimation of nitrogen dry deposition in South Korea
.
Atmos. Environ.
,
36
,
4951
4964
.
Sanchez
,
M. L.
,
D.
Pascual
,
C.
Ramos
, and
I.
Perez
,
1990
:
Forecasting particulate pollutant concentrations in a city from meteorological variables and regional weather patterns
.
Atmos. Environ.
,
6
,
1509
1519
.
Scott
,
G. M.
, and
R. D.
Diab
,
2000
:
Forecasting air pollution potentials: A synoptic climatological approach
.
J. Air Waste Manage. Assoc.
,
50
,
1831
1842
.
Shim
,
J.-M.
, and
S.-U.
Park
,
2004
:
Acidic loadings in South Korean ecosystems by long-range transport and local emissions
.
Atmos. Environ.
,
38
,
5623
5636
.
Sun
,
Y.
,
G.
Zhuang
,
A.
Tang
,
Y.
Wang
, and
Z.
An
,
2006
:
Chemical characteristics of PM2.5 and PM10 in haze-fog episodes in Beijing
.
Environ. Sci. Technol.
,
40
,
3148
3155
.
United Nations Environment Programme
,
2002
: The Asian brown cloud: Climate and other environmental impacts. Center for Clouds, Chemistry and Climate Rep. UNEP/DEWA/RS.02-3, 53 pp. [Available online at http://www.rrcap.ait.asia/issues/air/impactstudy/index.cfm.]
Wang
,
Y.
,
G.
Zhuang
,
Y.
Sun
, and
Z.
An
,
2006
:
The variation of characteristics and formation mechanisms of aerosols in dust, haze, and clear days in Beijing
.
Atmos. Environ.
,
40
,
6579
6591
.
Won
,
M. S.
,
S. Y.
Lee
,
M. B.
Lee
, and
S
. Ohga
,
2010
:
Development and application of a forest fire danger rating system in South Korea
.
J. Fac. Agric., Kyushu Univ.
,
55
,
221
229
.
World Meteorological Organization
,
2008
: Reports on the progress in addressing the work plan of the expert team. WMO Rep. Doc.3.1(2), 14 pp. [Available online at http://www.wmo.int/pages/prog/www/IMOP/meetings/Surface/ET-STMT-2/Doc.%203.1(2).pdf.]
Yu
,
H.
,
L. A.
Remer
,
M.
Chin
,
H.
Bian
,
R. G.
Kleidman
, and
T.
Diehl
,
2008
:
A satellite-based assessment of transpacific transport of pollution aerosol
.
J. Geophys. Res.
,
113
,
D14S12
,
doi:10.1029/2007JD009349
.