Multielemental analysis and characterization of fine aerosols at several key ACE-Asia sites

Authors


Abstract

[1] PM2.5 and PM10 particle size fractions were measured every Wednesday and Sunday for an 18-month period from 1 January 2001 at three key ACE-Asia sites: Hong Kong, Cheju Island in southern Korea, and Sado Island in Japan. Median 24 hour PM2.5 mass loadings of 29, 16, and 9.1 μg/m3 and coarse mass loadings of 33, 14, and 11 μg/m3 were measured at Hong Kong, Cheju, and Sado Island sites, respectively, during the study period. The corresponding maximum PM2.5 and coarse mass values for the three sites were 109, 81, and 78 μg/m3 and 101, 162, and 253 μg/m3, respectively. Accelerator-based ion beam analysis (IBA) techniques were used to quantify major components as well as significant trace elements. These included total hydrogen, black carbon, F, Na, Al, Si, P, S, Cl, K, Ca, Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Zn, Br, and Pb, with detection limits close to or below 1 ng/m3. The average PM2.5 percentage composition by weight across the three sites was estimated to be around (8.4 ± 4)% black carbon, (7.7 ± 7)% soil, (43 ± 14)% ammonium sulfate, (11 ± 16)% organic matter, (10 ± 12)% salinity, and (0.6 ± 0.3)% trace elements. Soil fingerprints for the east Asian region were generated using oxides of measured Al, Si, K, Ca, Ti, Mn, and Fe concentrations. The coarse fraction was dominated by wind blown soil (23%) and sea salts (48%). [PM10/PM2.5] mass ratios were typically (2.1 ± 0.4) averaged across all sites for the whole year. [PM10/PM2.5] mass ratios for the 21 IBA elements analyzed were also provided. This quantitative data providing both masses and dates over an 18-month period provide useful input for aerosol transport modeling for the east Asia region.

1. Introduction

[2] The scattering and absorption of light by fine airborne particles directly affects Earth's radiative balance. These same particles also have indirect effects by acting as cloud condensation nuclei, influencing the albedo, lifetime, and extent of clouds. The extent of the direct radiative forcing due to anthropogenic particles in the atmosphere was clearly identified and estimated in 1996 by the Intergovernmental Panel on Climate Change [IPCC, 1996] to be between −0.25 and −1.0 Wm−2 and the indirect aerosol contribution at between 0 and −1.5 Wm−2. This was shown to be comparable but opposite in sign to forcing due to well mixed greenhouse gases (+2.1 to +2.8 Wm−2). Since then the IPCC has released another report [IPCC, 2001] confirming the negative radiative forcing effects of atmospheric aerosols. The latter report also showed the level of scientific understanding was significantly lower for aerosols than greenhouse gases.

[3] The global distribution of particles tends to be extremely inhomogeneous resulting in different radiative forcing consequences in different regions of the globe. The International Global Atmospheric Chemistry Project (IGAC), recognizing this, has planned a series of Aerosol Characterization Experiments (ACE) that integrate in situ measurements, satellite observations, and models to reduce the uncertainty in estimates of the climate forcing due to aerosols [Bates et al., 1998]. ACE-Asia is the fourth in this series of experiments and is focused on eastern China, Korea and Japan [Huebert et al., 2003]. The objectives of ACE-Asia study are stated in the ACE-Asia Prospectus of February 2001 (see http://saga.pmel.noaa.gov/aceasia/prospectus/prospectus022601.html). The Asian aerosol source composition is different to that of Europe and North America, much more coal and vegetation are burnt in Asia, adding more absorbing soot and organic aerosol to parts of the Asian and Pacific atmosphere [Chameides et al., 1999]. Aerosols in the ACE-Asia region will likely be a complex mixture of sea-salt, combustion derived ionic, organic and soot particles, mineral dust and biogenic non-sea salt sulfate and organic species [Carmichael et al., 1996; Kim et al., 1998; Wang et al., 2003].

[4] According to Arndt et al., [1998] China produces around 11 Tg S/yr of sulfur emissions but only 6.1 Tg S/yr are deposited in China itself the rest is transported out of the region generally in an easterly direction. This mass transport of sulfur is 2.5 times higher in the winter and spring, than the summer and autumn months. In contrast, in summer when on shore winds predominate, sulfur from Japan is deposited on South Korea. Model calculations by Xiao et al. [1997] suggested that between 20% to 40% of the total sulfur production rate in the Asian Pacific regions in early spring was due to chemical conversion of SO2 to sulfate in the presence of mineral dust and anthropogenic aerosols. This is one of the reasons for measuring sulfur as well as mineral dust present in the fine mass fraction at ACE Asia sites. Chang et al. [1996] and Kim et al. [2003] both showed that windblown soil, was an important feature of Asian dust. The key fine particles of interest in the Asian region are sulfates, black carbon and organics from such human activities as fossil fuel combustion, biomass burning and industrial production in the study region and soil and sea salt from natural sources [Lee et al., 2001; Mori et al., 2002; Huebert et al., 2003; Wang et al., 2003].

[5] This study contributes to these ACE-Asia project goals by further quantifying and characterizing the chemical properties of fine (PM2.5) particles at recognized key ACE-Asia sites in Hong Kong, Korea and Japan over an 18-month period from January 2001. These longer-term ground station studies are an integral part of the ACE-Asia project (see http://saga.pmel.noaa.gov/aceasia/prospectus/prospectus022601.html), providing information on spatial variability of aerosol chemical and physical properties as well as seasonal and longer-term trends in these properties. The quantitative data sets presented here are important for verifying transport models such as the Navy Aerosol Analysis and Prediction System (NAAPS) at the Naval Research Laboratory, Monterey, CA, USA [Benkovitz et al., 1996] which currently uses uncertain emission inventories, and for putting the ACE-Asia chemical results in a broader global context.

2. Site Location and Sampling

[6] Here we report on data from three ACE Asian sites located on Hong Kong Island off the Chinese mainland, Cheju Island off the southern tip of Korea and Sado Island off the west coast of Japan. The location of each of these sites is shown in Figure 1.

Figure 1.

Map of the ACE-Asia region showing the location of the three sampling sites, Hong Kong, Cheju Island, and Sado Island used in this study.

[7] The Hong Kong site was located on Cape D'Aguilar (Hok Tsui) at the southeastern end of Hong Kong Island at 22.22°N, 114.25°E atop a 60 m cliff facing the South China Sea. The population density of Cape D'Aguilar is relatively low and the nearest major urban/industrial town of Chai Wan is 10 km away. A detailed description of the site and its surroundings has been given in the work of Wang et al. [1997, 2001] and Lam et al. [2001]. For this study sampling of aerosols commenced at this site on 3 January 2001.

[8] The Kosan site on Cheju Island was located at the western tip of the island at 33.30°N, 126.15°E on a cliff 70 m above sea level and faced the Chinese mainland across the Yellow Sea. The island is approximately 480 km south of Seoul. Sampling commenced on 30 March 2001. Kim et al. [1998] stated that Cheju Island was one of the cleanest areas in Korea and an ideal site to monitor long-range transport of air pollution from China [Lee et al., 2001].

[9] The Sado island site was located on the north western coast of the island off the west coast of Japan's Honshu Island at 38.20°N, 138.35°E atop a 90 m cliff facing the Korean mainland across the Sea of Japan. The site was approximately 304 km north west of Tokyo. Sampling commenced somewhat later than the other two sites on 16 September 2001.

[10] Two different types of sampling unit were used at each site in this study, the ASP PM2.5 fine sampler [Cohen et al., 1996], based on the U.S. IMPROVE cyclone system [Malm et al., 1994] and the GENT stacked filter unit (SFU) with a coarse (8μm pore size) and a fine (0.4 μm pore size) 47 mm diameter Nuclepore polycarbonate filter [Hopke et al., 1997; Maenhaut et al., 1994]. The coarse Nuclepore filter was Fomblin coated to reduce particle bounce effects. The ASP sampler used a 25 mm stretched Teflon filter and had a nominal flow rate of 22 l/min to collect particles with aerodynamic diameters less than 2.5μm (PM2.5). The GENT sampler used a nominal flow rate of 16 l/min and simultaneously collected a fine (PM2.5) fraction and a coarse fraction (2.5μm to 10μm). The effective collection areas of the Teflon and Nuclepore filters were 2.27 cm2 and 11.95 cm2, respectively.

[11] The ASP PM2.5 cyclone unit was specially designed for use with the multielemental ion beam analysis methods described in detail elsewhere [Cohen, 1993]. The very thin (nominal 230 μg/cm2) smaller diameter Teflon substrate [(CF2)n], was ideally suited to provide good “mass closure” and allowed for a fairly complete interference free aerosol characterization [Cohen et al., 1996; Cohen, 1998; Cohen et al., 2002a]. Average site temperatures, sampling times, flow rates and masses for both sampling systems during the study period for each of the three ACE-Asia sampling sites are given in Table 1.

Table 1. Average Sampling Parameters for the Study Period at Each of the Three ACE-Asia Sampling Sites for Both the ASP and the GENT Samplers
ASP and GENT Sampling ParametersHong Kong Mean ± SDCheju Island Mean ± SDSado Mean ± SD
Tmax, °C38 ± 821 ± 627 ± 8
Tmin, °C20 ± 513 ± 65 ± 5
ASP sampling time, hours20 ± 524 ± 324 ± 3
ASP flow, 1/min19 ± 421 ± 220 ± 2
ASP volume, m323 ± 731 ± 329 ± 4
ASP mass/filter, μg665 ± 389638 ± 417362 ± 310
Teflon thickness, μg/cm2239 ± 55229 ± 61232 ± 54
GENT sampling time, hours17 ± 624 ± 324 ± 3
GENT flow rate, l/min15 ± 317 ± 217 ± 2
GENT volume, m315 ± 424 ± 424 ± 3
GENT fine mass/filter, μg387 ± 208347 ± 193252 ± 126
GENT coarse mass/filter, μg547 ± 348574 ± 592409 ± 686

[12] Filters were weighed to ±1μg before and after exposure in a temperature and humidity controlled laboratory where the temperature was 22°C and the relative humidity less than 50%. Where possible filters were generally collected over a 24-hour period from midnight to midnight every Wednesday and Sunday throughout the year. For optimum analysis conditions it was important to try to keep the mass/unit area of material collected on the filters comparable to, or larger than, the mass/unit area of the filter itself. The average fine particle mass/unit area collected on the Teflon filter was between 160 μg/cm2 and 290 μg/cm2 which was comparable to the average Teflon filter thickness. For the Nuclepore filter, thickness ∼1 mg/cm2 (more than 4 times thicker than the Teflon filter), the collected fine particulate mass/unit area was between 21 μg/cm2 and 32 μg/cm2, significantly lower than on the Teflon filters. This lead to significantly poorer mass detection limits for trace elements measured on Nuclepore filters than equivalent measurements on Teflon filters. For some sites during heavily polluted periods samplers were operated on a two hours on two hours off basis over the 24-hour collection period to minimize reduced flow rate and filter clogging problems.

[13] The larger diameter (47 mm) and thicker (1 mg/cm2) Nuclepore polycarbonate [H14C16O3] filters used in the GENT sampling system tended to produce poorer “mass closure” because total deposited hydrogen could not be measured accurately on a filter which contained large amounts of hydrogen. Hence the GENT system was used mainly to obtain the coarse and fine mass and elemental ratios.

3. Filter Analysis Methods

[14] Accelerator-based ion beam analysis (IBA) methods have been applied throughout this study [Cohen et al., 1996; Cohen, 1998; Cohen et al., 2002a; Maenhaut et al., 1997]. In particular, the three simultaneous IBA techniques of particle induced X-ray emission (PIXE), particle induced gamma ray emission (PIGE) and proton elastic scattering analysis (PESA) have been used. They are ideally suited to the multielemental analysis of fine particulate matter on the appropriate filter media. They cover a wide range of elements in the periodic table from hydrogen to uranium, they have (μg/g) detection limits, they have short analysis times (typically 5 min per sample) and are nondestructive.

[15] The combination of these three IBA methods provided data on the following 21 elements, H, F, Na, Al, Si, P, S, Cl, K, Ca, Ti, V, Cr, Mn, Fe, Co, Cu, Ni, Zn, Br and Pb. Typical minimum detectable limits (MDLs) range down to less than 1 ng/m3 of air sampled and experimental errors were around ±7% to ±15% for most of the elements listed. A comprehensive discussion of MDLs and errors related to these IBA techniques has been presented by Cohen et al. [2002b].

3.1. Black Carbon Estimates

[16] Black carbon (BC) is the light absorbing part of the carbonaceous aerosol while elemental carbon (EC) is define by its thermal behavior and has an operational definition requiring temperatures above 340 °C and as high as 650 °C to volatilize. Petzold and Niessner [1995] describe and compare several different black carbon and elemental carbon (EC) measurement techniques. All measurements reported here are related to black carbon and not elemental carbon.

[17] Black carbon on the Teflon filters was estimated by measuring the transmission of [He/Ne] laser light (wavelength 0.633 μm) through the filters before and after exposure. That is, BC0 (μg/m3) = [100*A/(εV)] ln[I0/I] where A in cm2 was the filter collection area, V in m3 was the total sampled volume, I0 was the pre-exposure and I was the post-exposure laser intensity. Generally for particles produced by combustion processes the aerodynamic diameters are less than a micron and values of the mass absorption coefficient ɛ ranging from 5 to 10 m2/g are appropriate for black carbon estimates [Fuller et al., 1999]. Literature values of the black carbon mass absorption coefficient range from 4 to 20 m2/g with 10 m2/g previously being used as the accepted value for black carbon from diesel smoke. However, Fuller et al. [1999] suggest that this value for diesel smoke maybe as much as 50% too high.

[18] The mass absorption coefficient is a function of the wavelength as well as the absorbing particle diameter and density [Horvath, 1993, 1997]. For absorbing particles with diameters much less than the [He/Ne] laser wavelength, the mass absorption coefficient ε is a slowly varying function of particle diameter and can be considered as constant for particles of similar density. Calibration measurements by Cohen et al. [2000b] on smoke from acetylene flames and candle soot showed ε = 7 m2/g to be a good value for black carbon estimates on fine particles collected on Teflon filters where the particle diameters were less than 0.6 μm. Malm et al. [1994] used a similar laser absorption technique for black carbon estimates and made a small correction for absorption effects due to the soil component. They used the following factor for the black carbon estimates, [BC] = [BC0]−0.11*[Soil], where the [ ] brackets denote the mass/unit volume concentration of that species and BC is the soil corrected black carbon estimate. All black carbon estimates in this paper have been performed using this soil correction.

3.2. Sulfate Estimates

[19] Sulfur was one of the major elemental components measured by IBA techniques in fine particulate matter. Its origin is mainly anthropogenic being produced by burning of fossil fuels by industry, in cars and for power production.

[20] Airborne sulfur rarely occurs as the pure element it is usually produced as SO2 gas which readily converts to sulfate (SO42−) ions under normal atmospheric conditions. The sulfate ions can exist in the atmosphere as sulfuric acid producing acid rain or be partially neutralized to ammonium bisulfate or fully neutralized to ammonium sulfate [Lee et al., 2001]. If we assume that sulfur occurs in the atmosphere in one of these three forms then clearly the sulfur concentrations can be used to estimate the concentration of any of these sulfate species [Malm et al., 1994]. That is for the sulfate ion [SO42−] = 3*[S], for acidic sulfate [H2SO4] = 3.063*[S], for partially neutralized sulfate [NH4HSO4] = 3.594*[S] and for the fully neutralized sulfate [(NH4)2SO4] = 4.125*[S]. Where the brackets denote the concentration of that species.

3.3. S, K, and Ca Non-Sea Salt Estimates

[21] Sulfur, potassium and calcium occur naturally in sea spray. For seawater the generally accepted sulfur, potassium and calcium mass ratios relative to sodium are 0.084, 0.036 and 0.038, respectively [Weast and Astle, 1982]. Studies over seven years of the sulfur, potassium and calcium associated with sea spray in the PM2.5 size fraction at the global base line station at Cape Grim in Tasmania, Australia [Cohen et al., 2000a] found the [S/Na] = 0.092 ± 0.085, [K/Na] = 0.032 ± 0.013 and [Ca/Na] = 0.038 ± 0.025 in good agreement with Weast and Astle [1982]. Cape Grim sea salt (NaCl) accounted for 65% of the total PM2.5 mass so this was an excellent site to measure these elemental ratios. The seawater values of Weast and Astle [1982] have been used to calculate non-sea salt sulfur (nss-S), potassium (nss-K) and calcium (nss-Ca) throughout this work.

3.4. Total Hydrogen and Organic Matter Estimates

[22] Hydrogen has three primary sources, hydrogen in ammonium associated with sulfate and nitrate ions, hydrogen in organic matter and hydrogen associated with water vapor. The ion beam analysis (IBA) technique of proton elastic scattering (PESA) [Cohen, 1993] provides unique information on the total hydrogen content of the PM2.5 fraction collected on Teflon filters. This analysis was performed in vacuum and hence loosely bound water vapor was generally lost, also nitrates were not well held on Teflon filters [Malm et al., 1994]. Hence ammonium sulfate and organic matter were the only two measurable major sources of hydrogen considered here.

[23] Malm et al. [1994] gives the average composition of organic matter in the PM2.5 size fraction as 9% hydrogen, 20% oxygen and 71% carbon. We have tested this assumption using Rutherford backscattering (RBS) methods to measure total carbon and oxygen and PESA to measure total hydrogen [Cohen, 1993; Cohen et al., 1996; Cohen, 1998] and agree with their organic composition to within ±25% for each of these three components. Assuming the average organic particle was 9% hydrogen we can estimate the average organic content in our PM2.5 size fraction from the measured total hydrogen and sulfur through the following expression, [Organics] = {[Htot] − [0.25*S]}/0.09, where the square brackets indicate the concentration (μg/m3) of that species. Note we again assume that the sulfate ion was fully neutralized to ammonium sulfate. This may not always be the case [Malm et al., 1994; Lee et al., 2001].

3.5. Total Salinity Estimates

[24] Salinity was an estimate of the total sea salt content on the filter. It included salts of the seven most anions and cations which make up 99.7% of the salinity found in seawater, namely Na+, Mg+, Ca2+, K+, Cl, SO42− and HCO3 [Weast and Astle, 1982]. Sodium chloride is the most abundant salt in seawater accounting for about 86% of the total salinity. For this work total salinity was estimated by 3.25*[Na] and the sodium chloride estimate, sometimes referred to as sea salt, was estimated by 2.54*[Na]. It was assumed that the majority of the measured sodium mass was associated with sea spray at each of the three sites. This was a reasonable assumption since the sodium associated with crustal material is typically much less than 3% by weight [Mason and Moore, 1982] and sea salt and soil were of comparable mass concentrations in the PM2.5 mass fraction at the ACE Asia sites [Cohen et al., 2002c]. In the coarse fraction (2.5 μm to 10 μm) sea salt was typically 4 to 5 times the soil concentration; hence the effects of sodium in soil on total salinity estimates were minimal.

4. PM2.5 Fine Mass

[25] The 24-hour averaged PM2.5 mass fraction results for the Hong Kong site are plotted in Figure 2 as monthly box and whisker plot. The shaded box represents 25% to 75% of the distribution of all points; the upper and lower whiskers represent 1.5 times the distribution shown by the box. The horizontal bar in the box indicates the median and the plus sign the mean of the distribution for that month. All points lying outside the range defined by the whiskers (extreme events) are plotted as outlier dots. The horizontal date axis is represented as year and month with, for example, 200204 being the month of April 2002.

Figure 2.

Monthly averaged box and whisker plot for the fine mass measurements at the Hong Kong site during the study period.

[26] At the Hong Kong site the PM2.5 cyclone sampled on 170 days between the 3 January 2001 and 30 June 2002. Average 24-hour fine mass values ranged from below 10 μg/m3 to over 100 μg/m3 with a median value of 29 μg/m3.

[27] The median values of Figure 2 show that May to August inclusive (approximate summer period) were relatively low fine mass months and November through to March (approximate winter-spring period) were significantly higher months for the Hong Kong site. The prevailing winds were easterly and southeasterly in the summer period. During the winter season the prevailing winds in Hong Kong are from the north and northeast bringing in industrial pollution from the eastern Chinese mainland north of Hong Kong [Cheng et al., 2000; Lam et al., 2001; Wang et al., 2003]. Clearly there were selected periods of the year when the mass concentrations were particularly high. These periods have been identified in previous studies also and were related to significant international events such as dust emissions from desert regions in China as well as favorable meteorological conditions bringing pollution from eastern China into Hong Kong [Arndt et al., 1998; Benkovitz et al., 1996; Zhang et al., 1997; Mori et al., 2002].

[28] Figure 3 is similar plot to Figure 2 for PM2.5 particle mass at the Cheju Island sampling site. Filters were collected on 121 days between the 30 March 2001 and 30 June 2002.

Figure 3.

Box and whisker plot for monthly averaged fine mass concentration at the Cheju Island site during the study period.

[29] The Cheju Island fine mass data were generally lower than the Hong Kong data but still showed large ranges in the daily data from below 10 μg/m3 to well above 50 μg/m3. The seasonal trends at Cheju Island site were not as strongly developed as at the Hong Kong site. However the spring (March and April) highs as identified by other studies [Carmichael et al., 1996; Kim et al., 1998; Lee et al., 2001] were also present in this study.

[30] At the Sado Island site the PM2.5 cyclone sampled on 80 days between the 16 September 2001 and 30 June 2002. A complete set of twelve months worth of data has not yet been collected for the Sado Island site. Nevertheless, the fine mass values were significantly lower than the other two ACE-Asia sites (see Figure 4), not unexpected for a more remote site even further removed from major fine particle pollution sources on the Chinese mainland. Similar spring time (February to April) fine mass highs were also evident in data for this site. Again other workers have shown high mass concentrations at this time of the year may be attributed to long-range transport of pollution into this region from China [Mori et al., 2002; Kim et al., 2003]. The 24-hour mean, median, standard deviation (SD) and maximum values of the mass data set for the sampling period for each site are summarized in Table 2.

Figure 4.

Box and whisker plot for the monthly averaged fine mass concentration at the Sado Island site during the study period.

Table 2. Summary of the PM2.5 Fine Fraction Mass Measurements for Hong Kong, Cheju Island, and Sado Island for the Study Period to 30 June 2002
PM2.5Hong KongCheju IslandSado Island
Mean, μg/m3332113
Median, μg/m329169.1
Standard deviation, μg/m3241611
Maximum, μg/m31098178
Number of filters exposed17012180

[31] The Hong Kong site had a much broader fine mass distribution extending to significantly higher masses (whiskers over 60 μg/m3) than Cheju or Sado Islands. It also had 12 extreme events above 84 μg/m3 compared with 10 above 53 μg/m3 at Cheju and 7 above 30 μg/m3 at Sado Island. This was not unexpected as Hong Kong was relatively closer to fine particle urban and industrial sources in and around Hong Kong and Kowloon itself, industrial sources on the Chinese eastern mainland coast [Wang et al., 2003] and possible soil sources from desert regions in the northern China [Zhang et al., 1997; Mori et al., 2002].

[32] The 29 PM2.5 mass extreme events for the three sites are summarized in Table 3. These extreme events were significant in that, at all three sites, they represented fine mass fractions well above two standard deviations above the mean for that site. Also they were extremely high levels of PM2.5 for what were relatively isolated sites. The Hong Kong site is farther south than the Cheju Island and Sado Island sites and during the early parts of the year can be affected more by transport from the north of industrial pollution (sulfates) from eastern China [Wang et al., 2003]. Whereas Cheju Island and Sado Island being farther north are more affected by airborne soils transported from desert regions in northern China [Chang et al., 1996; Cheng et al., 2000; Kim et al., 2003]. These extreme PM2.5 mass events which were associated with long-range transport tended to occur in spring months of February, March and April at all three sites. Whether these high fine mass events were dominated by high soil, sulfate or black carbon is discussed further below when the contributions to the PM2.5 mass are presented.

Table 3. Summary of the Extreme PM2.5 Mass Events for the Hong Kong, Cheju Island, and Sado Island Sampling Sites for the Study Period to 30 June 2002
Hong KongCheju IsandSado Island
DateMass, μg/m3DateMass, μg/m3DateMass, μg/m3
28-Feb-019413-Apr-017120-Feb-0233
10-Mar-0110327-Apr-015510-Mar-0246
11-Mar-019424-Oct-016917-Mar-0231
26-Sep-019103-Feb-027624-Mar-0235
11-Nov-019106-Feb-025403-Apr-0278
26-Dec-0110931-Mar-026707-Apr-0231
30-Dec-0110010-Apr-028010-Apr-0237
08-Jan-029317-Apr-0254  
09-Jan-028705-Jun-0254  
25-Jan-029112-Jun-0257  
01-Feb-02109    
10-Feb-0284    

5. Fine Soil Estimates

[33] Soil has previously been identified as a significant component of the Asian aerosol in our study region [Carmichael et al., 1996; Chang et al., 1996; Holmes and Zoller, 1996; Kim et al., 1998; Cheng et al., 2000; Hien et al., 2002; Jaffe et al., 1999]. Kim et al. [2003] found that during the spring of 2001, Asian dust events were the prominent sources of major crustal components in the fine fraction of particulate matter. In order to obtain a reliable estimate of the fine soil mass associated with the total PM2.5 mass measured at each site, Principal Components Analysis (PCA) was used to identify major elements associated with the soil fingerprints for the region [Cohen et al., 1996; Cohen, 1998; Cheng et al., 2000]. This was done for each site separately and for the combined data for all three sites. Table 4 shows the factor loadings from a PCA analysis of 371 PM2.5 Teflon filters for the combined data for all three sites.

Table 4. Principal Components Analysis (PCA) With Varimax Rotation for all PM2.5 Data From Hong Kong, Cheju Island, and Sado Island to 30 June 2002
ElementSoil 1Autos/Industry 2Oil/Diesel 3Sea Salt 4Communality
Al0.9760.0730.1340.0020.975
K0.9690.1820.1280.0370.99
Fe0.9690.1810.1280.0370.99
Si0.9680.1590.0950.0340.973
Ti0.9080.3210.1180.0120.942
Ca0.9060.259−0.0040.1220.902
Mn0.7850.511−0.0030.1140.892
Cu0.7030.3640.3920.0880.788
S0.1860.8990.099−0.0570.856
Pb0.2870.8920.0870.1460.906
Zn0.2920.8750.1370.1720.899
BC0.1880.8480.1760.0850.792
H0.2150.8150.170.0470.742
Br0.2220.7880.180.3540.827
P0.0220.7470.18−0.3460.711
Cr0.4490.693−0.074−0.0350.689
V0.2380.0890.9110.0330.896
Ni0.1110.470.807−0.0160.884
Cl−0.014−0.0790.0630.8960.813
Na0.2230.455−0.0640.6730.714
 
Variance, %54177.806.90total: 86
Eigenvalue10.93.41.61.4 

[34] Factors with two or more elements with factor loadings above 0.5 have been highlighted. Four factors were required to explain 86% of the sample variance. Factor loadings near 1.0 demonstrate that the element has a strong association with that individual factor. Elements with communalities below 0.5 (e.g., Co) were excluded from the analysis.

[35] Factor 1 in Table 4 explained 54% of the variance and included the elements Al, Si, Ti, K, Ca, Mn and Fe, which were recognized as typical soil indicators [Chang et al., 1996; Kim et al., 1998; Cohen et al., 2002c]. Factor 2 explained 17% of the sample variance and was labeled autos/industry as it included high factor loadings for S, Pb, Zn, black carbon and H, typical of car and industry emissions [Cohen et al., 1996; Cohen, 1998]. Factors 3 and 4 explained 7% to 8% of the variance each and contained only two key elements. These were labeled oil/diesel [Cheng et al., 2000] and Sea salt [Cohen et al., 2000a, 2002c], respectively. Also the factor loadings for Na were higher in the sea salt factor 4 than the soil factor 1 demonstrating most of the data set variability for Na was associated with sea salt rather than soil in the fine mass fraction as discussed previously.

[36] The seven elements, Al, Si, K, Ca, Ti, Mn, and Fe were common to the soil factor for each site and for the combined site data shown in Table 4. However, there was a range of other trace elements that were associated with the soil factors at selected sites. For example copper was associated with the soil factors at the Cheju and Sado Island sites (factor loadings > 0.5) but not at the Hong Kong site and hence appeared in the soil factor of the combined data of Table 4. Close inspection of the individual site data for copper showed that this association with the soil factor was due mainly to 11 out of the 121 points at Cheju Island and 6 out of the 81 points at Sado Island having copper concentrations in excess of twice the average values for each of these sites during the study period. Each site will generally have its own unique trace elements associated with local sources, soils and conditions. These trace elements, although important for source fingerprinting, have little influence when estimating the soil mass contributions and the sum of the seven selected elements listed above dominated the total soil mass estimates.

[37] It is important not to include elements in a soil estimate that may have significant contributions from sources other than soil, as inclusion of these will tend to over estimate the soil component. We have attempted to minimize this problem in two ways, firstly by removing known components of elemental concentrations from sources other than soil. For example, by removing the sea salt component from the total potassium and calcium estimates through a knowledge of the [K/Na] and [Ca/Na] ratios in seawater [Weast and Astle, 1982]. Secondly, by plotting each element against silicon, the major soil component measured in this study, we obtain the minimum least squares fitted gradient that excludes excess contributions of this element not associated with silicon. In this way a soil fingerprint based on silicon can be readily constructed.

[38] A good example of this problem was potassium. Potassium has three possible primary sources, smoke from biomass burning, sea salts and windblown soil. This may have been the reason that Malm et al. [1994] excluded K from their soil estimates. If we remove the sea salt component of potassium by considering nss-K then a standard plot of nss-K against silicon will allow us to separate the smoke from the soil source and hence estimate the potassium content of the soil.

[39] This technique is well demonstrated by Figure 5 where the plot for the nss-K at Cheju Island shows two distinct groupings, low potassium associated with silicon in soil and higher potassium concentrations with lower silicon concentrations not associated with soil. This technique allowed us to extract an estimate of the potassium associated with soil alone from the measured total potassium concentration. The gradient of fits to data similar to the plots to Figure 5 at each site allowed the fraction of potassium relative to silicon in the soil fingerprints to be estimated with reasonable confidence.

Figure 5.

Correlation plot for silicon versus non-sea salt potassium at Cheju Island for the study period to 30 June 2002.

[40] The element Mn was included in our soil fingerprint but not in soil estimates of Malm et al. [1994]. Table 4 shows that Mn had significant factor loadings for soil (factor 1) as well as for cars and industry (factor 2). Hence like potassium it too had several possible sources. In similar treatments to that described in Figure 5, a Mn factor in soil was extracted by considering the appropriate minimum gradient correlation plots with Si and ignoring some higher Mn concentrations above about 50 ng/m3 which were not associated with soil.

[41] Finally the seven elements Al, Si, K, Ca, Ti, Mn and Fe in their oxide form were used to define a universal ACE-Asia soil fingerprint. The gradient of least squares fits for silicon to each of the above elements is given in Table 5 together with the standard error (SE) of the fit, the correlation coefficient (R2) of the fit, and the number of points used for each fit. The oxygen content was calculated by assuming each of the seven elements in the soil fingerprint occurred in its most abundant natural oxide state.

Table 5. Gradient, Standard Error (SE), and Correlation Coefficient (R2) of the Correlation Plots Relative to Silicon for Each Soil Element Identified by PCA for All Data From Three ACE-Asia Sites Over the Total Study Period
ElementGradient ± SER2Number of PointsComments
Al0.323 ± 0.0020.98369Al < 1600 ng/m3
K0.190 ± 0.0010.98334K < 1000 ng/m3
Ca0.213 ± 0.0030.92368Ca < 900 ng/m3
Ti0.034 ± 0.00050.93371 
Mn0.014 ± 0.00040.79365Mn < 50 ng/m3
Fe0.317 ± 0.0020.98368Fe < 1600 ng/m3
O2.55 ± 0.0060.99371assumes common oxide form

[42] The techniques used to obtain the gradients of Table 5 enabled us to define a fingerprint for soil at each of the sampling sites and across all three sampling sites. Figure 6 is such a plot. It shows the elemental fractions for soil at Hong Kong, Cheju Island, Sado Island and all three sites together.

Figure 6.

Calculated elemental fractions in soil for the Hong Kong, Cheju Island, Sado Island, and all sites together using the seven elements fingerprint.

[43] When the errors were considered there was very little difference between the major element soil fingerprints at Hong Kong, Cheju Island and Sado Island. This was expected considering the methods used to obtain these fingerprints. The main differences were slightly increased Al and K and decreased Ca at Sado Island and reduced Al and increased Ca and Mn at Hong Kong.

[44] Table 6 gives the fractions of each of the seven elements and their oxides for the universal ACE-Asia soil fingerprint shown in Figure 6. The average Earth's crustal values taken from Mason and Moore [1982] have been included for comparison purposes. The agreement was quite good considering the variability of crustal material and soils around the world. Also the sum of the eight Mason and Moore elements for crustal material accounts for 99.97% of their total crustal mass. Demonstrating that the oxides of these seven elements account for the vast bulk of the crustal soil mass.

Table 6. Mean Elemental Fractions in Soil From All ACE-Asia Sites Obtained From Least Squares Fits to Each Element Relative to Silicon and Assumig Each Element Occurs as Its Common Oxide
ElementFraction ± SDCrustal Materiala
Al0.069 ± 0.0100.086
Si0.215 ± 0.0390.294
K0.041 ± 0.0060.027
Ca0.046 ± 0.0140.039
Ti0.0074 ± 0.0020.0047
Mn0.0030 ± 0.0020.001
Fe0.068 ± 0.0100.053
O0.550 ± 0.0270.495

[45] Figures 7, 8, and 9show monthly box and whisker plots for the fine PM2.5 soil estimates for the Hong Kong, Cheju Island and Sado Island for the study period to 30 June 2002. The main feature of all three plots was the high soil events associated with the months February, March and April. Hong Kong and Cheju Island showed relatively lower soil concentrations during June, July and August than at other times of the year. Not enough data have been collected yet to observe similar summer trends for the Sado Island site.

Figure 7.

PM2.5 soil estimates for Hong Kong during the study period to 30 June 2002.

Figure 8.

PM2.5 soil estimates for Cheju Island during the study period to 30 June 2002.

Figure 9.

PM2.5 soil estimates for Sado Island during the study period to 30 June 2002.

[46] When compared with the total fine mass plots of Figures 24 we see that the median percentage fine soil values at the three sampling sites are very similar all being below about 10% of the fine mass. The Hong Kong site had 14 extreme soil events with 14% or more of the fine mass as soil, the Cheju island site had 11 extreme events with the soil content above 21% and Sado Island 9 such events with soil above 17% during the sampling period to 30 June 2002. High percentage soil events at each site tended to occur in the same periods as the high mass events (see Table 3), that is, mainly in February, March and April of each year. This was consistent with fine soil being transported from the Chinese mainland to Korean and Japanese sites as reported by others [Uematsu et al., 1985; Carmichael et al., 1996; Zhang et al., 1997; Kim et al., 2003].

[47] Table 7 is a summary table showing the mean, median, standard deviation and maximum PM2.5 soil concentrations and percentages at three ACE-Asia sampling sites during the study period to 30 June 2002. The median values were highest in Hong Kong and lowest at Sado Island as expected. However for extreme situations the soil concentrations at Sado Island were not too different to Hong Kong or Cheju Island reaching levels up to 26 μg/m3 and contributing more than 35% to the total PM2.5 mass fraction.

Table 7. Mean, Median, Standard Deviation, and Maximum PM2.5 Soil Concentrations and Percentages at Three ACE-Asia Sampling Sites
PM2.5Hong KongCheju IslandSado Island
Soil, μg/m3
  Mean, μg/m32.62.31.5
  Median, μg/m31.40.920.49
  Standard deviation, μg/m32.93.73.5
  Maximum, μg/m3172226
 
Soil, %
  Mean, %78.77.7
  Median, %5.25.45.8
  Standard deviation, %5.88.36.6
  Maximum, %364335

[48] The ACE-Asia soil fingerprints of Figure 6 showed slight differences in the Ca and Al fractions at different sampling sites. It is expected that there are significant differences in soil compositions across China. China National Environmental Monitoring Centre [1994] gives higher Ca values in the northwestern regions and higher Al values in the southeastern regions with widely ranging (Ca/Al) ratios from the southeast to the northwest. Carmichael et al. [1996] and Chang et al. [1996] both discuss wind blown soils from arid regions in central and eastern China rich in calcium being transported across our current study region.

[49] Figure 10 is a plot of the (nss-Ca/Al) ratio against the percentage of soil in the fine PM2.5 size fraction for each of the three sites. The data split into two distinct groups with no clear differences between the three sites.

Figure 10.

Plot of the nssCa/Al ratio for all three ACE-Asia sites during the study period to 30 June 2002.

[50] The first group was the high percentage soil (Soil > 5%) with the (nss-Ca/Al) ratio less than 2 and the second group was the low percentage soil (Soi l < 5%) with the (Ca/Al) ratio greater 2. The high (nss-Ca/Al) ratio events could be associated with local soils or local Ca sources such as cement production impacting each site. Whereas the high percentage soil events with relatively constant (nss-Ca/Al) ratios around unity were more likely to be due to long-range windblown soils transported from the Chinese mainland [Carmichael et al., 1996; Holmes and Zoller, 1996; Zhang et al., 1997]. Average coarse dust [Al/Ca] ratios of (1.3 ± 1.2) measured by Holmes and Zoller [1996] for Asia dust over the Mauna Loa observatory on Hawaii are also consistent with this argument. Consideration of back trajectories and meteorological data would help resolve this problem and will be considered further in a future publication when we look in detail at source fingerprinting, source apportionment and transport using this data set.

[51] Large variations may occur in the (nss-Ca/Al) composition of Chinese mainland soils measured on the ground at different sites across China but these differences do not appear to be showing up in the fine airborne soil contributions blown thousands of kilometers across the Korean peninsula and into Japan. This was an important finding, consistent with findings of Holmes and Zoller [1996] who found that transported mineral aerosol first lofted and then experiencing mixing and fallout during transport no longer had an elemental fingerprint resembling that of the bulk soil from which it was originally derived.

6. Sulfur and Sulfate

[52] Sulfur was one of the major elemental components measured on all filters during this study. Using a [S/Na] ratio of 0.084 for seawater [Weast and Astle, 1982] we estimated that, for the PM2.5 fraction, at all three ACE-Asia sites in this study, the sea-salt related sulfur component of the total sulfur mass accounted for less than (3 ± 5)% by weight. Hence the bulk of the sulfur was probably anthropogenic in origin.

[53] Figures 11, 12, and 13 are the monthly averaged box and whisker plots for the total PM2.5 sulfur mass at the Hong Kong, Cheju Island and Sado Island sites. The median percentage sulfur mass concentrations and the seasonal variations were quite similar for all three sites. Sulfur values tended to be higher in the winter months. The main difference being that Hong Kong had more extreme events with total sulfur above 6 μg/m3 than the other two sites.

Figure 11.

Monthly averaged box and whisker plot for the fine sulfur measurements at the Hong Kong site during the study period.

Figure 12.

Monthly averaged box and whisker plot for the fine sulfur measurements at the Cheju Island site during the study period.

Figure 13.

Monthly averaged box and whisker plot for the fine sulfur measurements at the Sado Island site during the study period.

[54] The box and whisker plot of Figure 14 shows that the estimated percentage sulfate (SO42−) concentrations present in the fine mass fraction at the Hong Kong, Cheju Island, and Sado Island sites were very consistent, with median values between 25% and 35% of the total fine mass. These values were consistent with average sulfate ion concentrations measured by Kim et al. [1998], ranging from 30% to 50% of the total PM2.5 mass fraction at the Cheju Island site in the summer of 1994. They were also comparable with the annual measured and model calculation of estimates of Arndt et al. [1998] for the average annual sulfate concentrations (2 μg/m3−10 μg/m3) at 16 locations across Japan, Korea and eastern China. Lee et al. [2001] report sulfate concentrations, at selected times in 1996 and 1997 at Cheju Island, between 4 μg/m3 and 6 μg/m3 which represented 21% to 35% of the total PM2.5 mass measured. Again consistent with our continuous longer term estimates at the same site.

Figure 14.

Box and whisker plot of the percentage fine particulate sulfate estimated at the Cheju Island, Hong Kong and Sado Island sites for the study period to 30 June 2002.

[55] There were only four extreme events with sulfate percentages above 60%, three at Cheju Island and one at Hong Kong. These occurred on 4, 22 and 29 July 2001 at Cheju Island and on 2 June 2002 at Hong Kong. The low outlier at the Sado Island site occurred on the 3 April 2002 and has already been identified as a highest fine mass day (78 μg/m3) dominated by soil from mainland China (see Table 3). The high percentage sulfate days in general were not related to high absolute sulfur mass events. For Hong Kong site, for example, the 18 highest sulfur events with S > 6 μg/m3, had an average sulfate fine mass percentage of (28 ± 5)%. For Cheju Island the 15 highest sulfur events with S > 3 μg/m3, had an average sulfate fine mass percentage of (32 ± 11)%. Whereas for Cheju Island the 9 highest sulfur events with S > 2 μg/m3, had an average sulfate fine mass percentage of (27 ± 8)%. These percentage values were consistent with annual average sulfate percentages for each of the sites. Furthermore the highest percentage fine sulfate tended to occur in the summer months of July and August whereas the highest absolute sulfur occurred in different months at Hong Kong, Cheju Island and Sado sites. Showing that these sites were affected at different times by differing meteorological conditions and differing sources of fine sulfur. Similar findings were obtained by Wang et al. [2003] in the Hong Kong region suggesting that many sulfate sources were local sources associated with shipping in the South China Sea and urban and biomass burning. Table 8 gives the mean; median, standard deviation and maximum PM2.5 sulfate estimates and percentages at three ACE-Asia sampling sites for the study period.

Table 8. Mean, Median, Standard Deviation, and Maximum PM2.5 Sulfate Estimates and Percentages at Three ACE-Asia Sampling Sites
PM2.5Hong KongCheju IslandSado Island
Sulfate, μg/m3
  Mean, μg/m39.65.43.3
  Media, μg/m38.74.82.9
  Standard deviation, μg/m36.13.42.2
  Maximum, μg/m3301912
 
Sulfate, %
  Mean, %323030
  Median, %322831
  Standard deviation, %101210
  Maximum, %606948

[56] These estimates show that if sulfate occurs solely as ammonium sulfate in the fully neutralized form it would account for the largest mass fraction in the PM2.5 size range, being consistently 40% to 45% of the total fine mass across all three ACE Asian sites.

7. Black Carbon

[57] Like sulfur, black carbon was also a significant component of the PM2.5 mass fraction at Asian sites. It has its origin in sources such as emissions from motor vehicles, and the combustion of fossil fuels, animal waste, crops and vegetation [Kim et al., 1998]. Figures 15, 16, and 17 show the monthly averaged box and whisker plots for the PM2.5 black carbon measurements at the Hong Kong, Cheju Island and Sado Island sites, respectively during the study period to 30 June 2002.

Figure 15.

Monthly averaged box and whisker plot for the fine black carbon measurements at the Hong Kong site during the study period.

Figure 16.

Monthly averaged box and whisker plot for the fine black carbon measurements at the Cheju Island site during the study period.

Figure 17.

Monthly averaged box and whisker plot for the fine black carbon measurements at the Sado Island site during the study period.

[58] The median black carbon values ranged from about 0.8 μg/m3 at Sado Island to around 1.8 μg/m3 at the Hong Kong site. Peak black carbon values at Hong Kong were 50% higher than at the Cheju Island site and 3 times high than at the Sado Island site. There were 16, 13 and 5 daily events with black carbon above 5 μg/m3, 2.5 μg/m3 and 1.5 μg/m3 at Hong Kong, Cheju Island and Sado Island sites, respectively. At all three sites winter black carbon concentrations were above twice the summer values.

[59] A comparison with the extreme soil events showed that only on one occasion at Cheju Island, on 13 April 2001, and one occasion at Sado Island, on 7 April 2002, did an extreme soil event correspond with an extreme black carbon event at the same site. There were no such occurrences at all for the Hong Kong site. This was strong evidence for the soil and black carbon at each of the ACE Asia sites, during extreme events, having different source locations.

[60] Table 9 summarizes the measured fine black carbon masses and their percentages for the three sampling sites during the study period. The average percentage black carbon at the three sites was similar at around 8% to 9% of the total PM2.5 mass with maximum percentages ranging from 23% to 31% of the fine mass.

Table 9. Mean, Median, Standard Deviation, and Maximum PM2.5 black Carbon Concentrations and Percentages at Three ACE-Asia Sampling Sites During the Study Period
PM2.5Hong KongCheju IslandSado Island
Black Carbon, μg/m3
  Mean, μg/m32.31.50.87
  Median, μg/m31.81.10.77
  Standard deviation, μg/m31.71.20.46
  Maximum, μg/m39.96.62.7
 
Black Carbon, %
  Mean, %8.28.49
  Median, %77.48.3
  Standard deviation, %3.64.54.2
  Maximum, %232531

[61] Kim et al. [1998] measured fine elemental carbon (note this is not the same as black carbon, Petzold and Niessner [1995] at Cheju Island in July to August 1994 at around 1% of the total fine mass and ranging from 0.01 to 0.80 μg/m3. This was less than the present black carbon estimates of (7 ± 3)% ranging from 0.15 to 1.5 μg/m3 for these same summer months in 2001–02 when values were traditionally lower.

[62] Extreme events with a high percentage black carbon do not correspond well with absolutely high black carbon events of Figures 1517. For Cheju Island there were only three extreme event days, the 7 and 14 October 2001 and the 7 November 2001 with black carbon above 20% of the total fine mass. Sado Island had only one such day, 1 May 2002. While Hong Kong had 10 extreme event days with black carbon above 15% of the total fine mass. For Cheju Island the 14 October 2001 was the only high percentage black carbon day with black carbon greater than 2.5 μg/m3. For Hong Kong and Sado Island none of the high percentage extreme events had absolute black carbon masses above 2.5 μg/m3. This showed that extreme absolute black carbon events at all three sites tended to occur when the percentage black carbon was typical of the mean percentage value (around 8% to 9%). Extreme percentage black carbon events tended to occur when the total fine mass was significantly less than the median value for that site. That is, during out study period high percentage black carbon was not associated with absolutely high black carbon and was mainly produced by other components such as sulfate and soil being relatively lower, thus reducing the total fine mass.

[63] The PCA techniques of Table 4 showed that, across all sites, both black carbon and sulfur were strongly associated (factor loadings > 0.84) with the source assigned to automobiles and industry (factor 2). Figure 18 shows that these two elements do indeed correlate fairly well (R2 = 0.50) particularly for black carbon concentrations below about 2 μg/m3. The points to the right of the least squares fitted line represent higher black carbon values not associated with sulfur and were probably associated with smoke sources from biomass burning. Whereas points to the left of the line represented fine particle sources such as fossil fuel burning with relatively higher sulfur and lower black carbon components [Cohen et al., 2002c]. Similar sulfur and black carbon correlations were also observed for the Cheju and Sado Island sites during the study period.

Figure 18.

Black carbon-sulfur correlation plot for all the PM2.5 data from the Hong Kong site for the study period to 30 June 2002.

8. Total Hydrogen and Organic Matter

[64] As discussed previously ammonium ions and organic matter were the two major sources contributing to the total measured hydrogen considered here. Figures 19, 20, and 21 are monthly averaged box and whisker plots for the PM2.5 total hydrogen measurements at the Hong Kong, Cheju Island and Sado Island sites during the study period up to 30 June 2002. Hong Kong showed the strongest seasonal variations being higher in the winter and lower in the summer months. The Cheju and Sado sites showed relatively higher total hydrogen in the months of February, March and April.

Figure 19.

Monthly averaged box and whisker plot for the fine hydrogen measurements at the Hong Kong site during the study period.

Figure 20.

Monthly averaged box and whisker plot for the fine hydrogen measurements at the Cheju Island site during the study period.

Figure 21.

Monthly averaged box and whisker plot for the fine hydrogen measurements at the Sado Island site during the study period.

[65] Table 10 summarizes the total hydrogen mass measurements for the data shown in Figures 1921. Hydrogen was typically 3% to 4% of the total fine mass, but this could rise to 10% to 12% if the sulfate and organic components increased significantly.

Table 10. Mean, Median, Standard Deviation, and Maximum PM2.5 Hydrogen Concentrations and Percentages at Three ACE-Asia Sampling Sites During the Study Period
PM2.5Hong KongCheju IslandSado Island
Hydrogen, μg/m3
  Mean, μg/m31.30.730.41
  Median, μg/m30.960.560.29
  Standard deviation, μg/m310.610.4
  Maximum, μg/m35.72.92.6
 
Hydrogen, %
  Mean, %3.83.73.2
  Median, %3.63.43
  Standard deviation, %1.51.61.2
  Maximum, %9.61210

[66] Comparisons of Figures 19 to 21 with Figures 11 to 13 for sulfur show that the total hydrogen follows the total sulfur content having similar seasonal variations. Also high hydrogen tended to correlate with high sulfur. For example, the extreme events plotted as dots on 4 July 2001 and 1 February 2002 at Hong Kong, 3 June 2001 and 31 March 2002 at Cheju Island and 24 October 2001 and 15 May 2002 at Sado Island were all high sulfur and high hydrogen days at each of the sites.

[67] This hydrogen-sulfur correlation is further demonstrated in Figure 22, where the total hydrogen is plotted against the total sulfur mass for the Hong Kong site for the study period to June 2002. The solid line, [S] = 4*[H], is the line expected for fully neutralized ammonium sulfate. Points to the left of this line (reduced hydrogen) represent partially neutralized sulfate ions or acidic aerosols [Malm et al., 1994; Lee et al., 2001] and points to the right represent the higher hydrogen concentration events not associated with sulfur, that is hydrogen in organic matter. Total hydrogen concentrations below 1 μg/m3 at the Hong Kong site tended to be associated with fully neutralized ammonium sulfate and relatively small organic matter content, whereas for total hydrogen above 1.5 μg/m3, significantly more organic matter was present.

Figure 22.

Correlation plot for PM2.5 total hydrogen versus total sulfur at the Hong Kong site for all data from 1 January 2001 to 30 June 2002. The solid line represents the fully neutralized ammonium sulfate line.

[68] Figure 23 is a box and whisker plot for estimated fine organic matter at the three sites. The median organic matter contents were 2.1 μg/m3, 1.2 μg/m3, and 0.65 μg/m3 at the Hong Kong, Cheju Island, and Sado sites, respectively for the study period. The seasonal trends for estimated organic matter were the same as for total hydrogen.

Figure 23.

Box and whisker plot for the estimate PM2.5 organic matter at the three study sites for the period to 30 June 2002.

[69] There were 8 extreme organic matter events above 20 μg/m3 at Hong Kong, 12 extreme events above 10 μg/m3 at Cheju Island and 10 extreme events above 4 μg/m3 at Sado. 50% of these events at Hong Kong and Cheju Island and 70% of these events at Sado Island had both high sulfur and high organic matter concentrations at the same time. Further comparison with extreme black carbon events showed that 25% of the extreme organic matter events at Hong Kong, and 30% at Cheju and Sado Islands had high organic matter, high black carbon and high sulfur all within days of each other. That is, these three sites were impacted by similar sulfate, black carbon and organic sources on the same days. Strong evidence for significant PM2.5 pollution transported from mainland China affecting all three sites during the study period [Arndt et al., 1998; Mori et al., 2002; Kim et al., 2003].

9. PM2.5 Composition

[70] The IBA analysis techniques provided information on over 20 different chemical species in the PM2.5 mass fraction. Table 11 provides the mean, standard deviation (SD) and minimum and maximum concentrations for species measured and estimated at the Hong Kong, Cheju Island and Sado Island sites for all data collected over the study period to 30 June 2002. Generally elemental and chemical species quoted in previous tables have not been included. Except for the mass and the percentage reconstructed mass (%RCM) all values are quoted in (ng/m3) rounded to no more than three significant digits. The large standard deviations for many values reflect the non-Gaussian nature of the distribution and the large seasonal variations in the data and not the measurement errors. Measurement errors were typically of the order of ±7% to ±15% for most species determined and have been fully quantified elsewhere [Cohen et al., 2002b].

Table 11. PM2.5 Means and Standard Deviations for All Measured and Estimated Species Over the Study Period From 1 January to 30 June 2002 at Each of the ACE-Asia Sites
Average PM2.5 Value, ng/m3Hong KongCheju IslandSado Island
Mean ± SDMin to MaxMean ± SDMin to MaxMean ± SDMin to Max
Na935 ± 8280–3970438 ± 4390–2360309 ± 4560–2330
Al170 ± 2170–1200160 ± 2920–1830113 ± 2930–2270
Si571 ± 65711–3600511 ± 80520–4640333 ± 76211–5670
P50 ± 460–29622 ± 200–10118 ± 150–67
S total3210 ± 2040336–10,1001800 ± 1150343–64001110 ± 740124–3900
S non-sea salt3140 ± 2000197–98901760 ± 1150290–63501090 ± 738124–3870
Cl182 ± 3420–225059 ± 1710–1390173 ± 3140–1730
K total539 ± 57310–2730300 ± 34918–1910153 ± 1857–1090
K non-sea salt505 ± 5580–2640284 ± 3460–1913142 ± 1805–1070
K nonsoil427 ± 4878–2470196 ± 2690–182089 ± 970–465
Ca157 ± 18612–1450114 ± 2033–163071 ± 1222–840
Ca non–sea salt121 ± 1730–137097 ± 1970–154060 ± 1170–817
Ti23 ± 230–11418 ± 260–15411 ± 250–190
V9 ± 90–452 ± 50–391.2 ± 50–43
Cr4 ± 60–383 ± 40–231.1 ± 20–12
Mn11 ± 130–688 ± 100–755 ± 80–58
Fe187 ± 2150–1270173 ± 2828–1870106 ± 2792–2250
Co2 ± 40–342 ± 40–321.1 ± 40–33
Ni4 ± 40–321.2 ± 10–70.8 ± 20–18
Cu6 ± 80–584 ± 80–673 ± 100–89
Zn116 ± 1250–57732 ± 351–21516 ± 180–83
Br12 ± 120–706 ± 40–215 ± 30–15
Pb70 ± 780–38624 ± 270–14810 ± 110–62
BC0 (nosoil)2590 ± 1850363–10,4301670 ± 1330174–70001028 ± 630273–3220
Organic carbon3770 ± 51200–27,8902260 ± 30900–13,6601490 ± 2,3600–15,130
Ammonium sulfate13,260 ± 84001390–41,6607430 ± 47301410–26,4004590 ± 3,050510–16,080
Organics matter5190 ± 73100–39,2903090 ± 44300–19,2401460 ± 3,0900–21,300
Salinity3040 ± 26900–12,9101420 ± 14300–76701530 ± 1,2200–7570
Traces273 ± 2561330–132095 ± 8311–47055 ± 508–248
Mass, μg/m332.7 ± 243–10920.4 ± 164–8012.6 ± 111–78
Percent RCM83 ± 15 81 ± 16 75 ± 12 

[71] Aluminum and silicon were highly correlated (see Table 5), as expected, through their association with soil. The aluminum-silicon ratio for each site was between 0.30 and 0.34, having an average value of (0.32 ± 0.04) consistent with the Mason and Moore [1982] value of 0.29 for crustal material.

[72] The average sea salt component of sulfur was less than 2% for all three sites, demonstrating that most of the sulfur and hence the sulfates in the PM2.5 fraction were anthropogenic in nature. This is consistent with the findings of Kim et al. [1998] and Lee et al. [2001] who measured similar contributions of around 1% to 3% for selected months during 1996 and 1997 at the Cheju Island site.

[73] For fine potassium the average sea salt component was less than 7% of the total potassium mass for all three sites, the difference being made up mainly from potassium associated with windblown soil and smoke from biomass burning. The nonsoil, non-sea salt potassium was an excellent indicator of smoke from biomass burning [Malm et al., 1994; Cohen et al., 1996]. For all sites the average fine potassium associated with smoke was more than the sum of the sea salt and soil components together. The Hong Kong site was more affected by smoke from biomass burning than either the Cheju Island or Sado Island sites.

[74] For fine calcium the bulk of the non-sea salt component was associated with windblown soil as discussed above. The non-sea salt calcium (nss-Ca) was between 77% and 85% of the total calcium for all three sites.

[75] Cobalt was measured by particle induced X-ray emission (PIXE) techniques [Cohen, 1993] as having an average value of 1 to 2 ng/m3 this was generally above the minimum detectable limit. However, there maybe significant adjacent inter-element interferences associated with the measurement of these small amounts of cobalt in the presence of larger amounts of iron using this technique. Hence this estimate of the cobalt concentration should be considered as an absolute upper estimate only.

[76] Chrome, copper, zinc and lead were associated with automobiles and industry (see PCA results of Table 4) and had the highest median values and extreme events at the Hong Kong site. Nickel and vanadium were also highest at the Hong Kong site and were probably associated with diesel and heavy oil burning for power generation and shipping [Cheng et al., 2000; Wang et al., 2003].

[77] The BC0(nosoil) component was the black carbon measurement without the 11% soil correction [Malm et al., 1994]. This correction to the black carbon estimates amounted to (10 ± 23)%, (13 ± 13)% and (11 ± 14)% reduction for the Hong Kong, Cheju and Sado Island sites, respectively.

[78] Organic carbon was estimated by assuming the average organic matter composition was 71% carbon [Malm et al., 1994] as discussed above and was not included in the percentage reconstructed mass (%RCM) as it was already included through the total organic matter estimates.

[79] The total trace element concentration was define as the sum of the concentrations of P, V, Cr, Co, Ni, Cu, Zn, Br and Pb and the percentage trace element masses were generally much less than 1% of the total fine mass. The extensive range of trace elements measured on filters will be most useful in distinguishing source fingerprints and for performing source apportionment studies using data from these sites. This work is planned for a future publication.

[80] The total PM2.5 mass was largest at the Hong Kong site as expected as this was significantly impacted by anthropogenic pollution from local sources in suburban Hong Kong as well as longer-range transport from mainland China to the north [Wang et al., 2001, 2003]. Kim et al. [1998] also observed an average PM2.5 mass for the period from July to August 1994 of (21 ± 4) μg/m3, similar to our 18-month average to 30 June 2002 of (21 ± 16) μg/m3 at Cheju Island.

[81] The%RCM was the percentage reconstructed mass calculated by following the procedure of Malm et al. [1994] and adding together all the measured components and comparing this sum with the measured gravimetric mass. The missing percentage mass was mainly due to water vapor and nitrates which were not measured and are generally not well held on Teflon filters. The high%RCM estimates reflect the relatively good “mass closure” obtained in this study using Teflon filters.

[82] Although the absolute PM2.5 masses were significantly different at the three sites the percentage compositions for the major components were remarkably similar. All three sites, being coastal sites, had similar percentage salinity components. The percentage soil components at Cheju and Sado Islands tended to be higher than at Hong Kong. The average PM2.5 mass loadings at Cheju and Sado Islands were affected more by soil from mainland China than at the Hong Kong site over the study period. The fraction of fine organic matter measured at Sado Island was consistently lower than either Hong Kong or Cheju Island sites.

[83] Loss of chlorine was observed at each of the three sites being (88 ± 22)%, (92 ± 15)% and (72 ± 32)% at the Hong Kong, Cheju Island and Sado Island sites, respectively. The percentage chlorine loss was calculated assuming all the sodium was associated with sea salt (NaCl) and comparing the measured total chlorine with the expected sea salt chlorine from the sodium concentrations. Cheng et al. [2000] also observed 15% to 20% chlorine losses in TSP and PM10 size fractions at the Hong Kong site. This was much lower than that observed here for the PM2.5 size fraction. This difference may be expected for the finer size fractions containing a relatively higher sulfate component.

10. Coarse Component

[84] The GENT stacked filter sampler provided a PM2.5 mass estimate and a coarse mass (2.5 μm to 10 μm) estimate. The PM2.5 mass estimates from the ASP and GENT samplers have been favorably compared elsewhere [Hopke et al., 1997] and will not be repeated here. Figures 24, 25, and 26 are box and whisker plots for the average monthly coarse mass for the Hong Kong, Cheju Island and Sado Island sites, respectively. Median coarse mass loadings of 33 μg/m3, 14 μg/m3 and 11 μg/m3, were measured at Hong Kong, Cheju and Sado Island sites, respectively during the study period. Comparison of Figure 24 and Figure 2 for Hong Kong site shows that coarse and PM2.5 showed similar seasonal trends, being higher in the October to March period than in the July to September period. Similar patterns were also evident at the Cheju Island and Sado Island sites.

Figure 24.

Box and whisker plot for the average monthly coarse mass fraction for the Hong Kong site during the study period to 30 June 2002.

Figure 25.

Box and whisker plot for the average monthly coarse mass fraction for the Cheju Island site during the study period to 30 June 2002.

Figure 26.

Box and whisker plot for the average monthly coarse mass fraction for the Sado Island site during the study period to 30 June 2002.

[85] Unlike the fine fraction, the coarse size fraction was dominated by two noncombustion source contributions, namely wind blown soil and sea spray. Figure 27 is a box and whisker plot of the percentage soil in the coarse size fraction at the three sites. On average, soil accounted for about (13 ± 8)% at the Hong Kong site, (23 ± 13)% at the Cheju Island site, and (15 ± 9)% at the Sado site of the total coarse mass fraction. For some extreme events in spring coarse soil accounted for more than 25% at Hong Kong, 40% at Cheju Island and 25% at Sado Island of the total measured mass.

Figure 27.

Box and whisker plot for percentage coarse soil at Hong Kong, Cheju Island, and Sado Island during the study period to 30 June 2002.

[86] All three ACE-Asia sites were on the coast and hence strongly influenced by sea spray. Figure 28 is a box and whisker plot for the percentage salinity in the coarse fraction. Salinity was estimated by assuming all sodium originated from sea spray as described earlier. For these coastal sites this was not an unreasonable assumption as the only other possible significant source of coarse sodium was from windblown soil. Mason and Moore [1982] give the (Na/Si) = 0.10 for crustal material, this was significantly less than our median measured coarse value for all ACE-Asia sites of (Na/Si) = 6.1. Demonstrating a large sodium source not associated with soil. Furthermore PCA techniques showed that coarse sodium was strongly correlated with coarse chlorine at all three sites (as indeed was the fine mass fraction see Table 4) and plots of sodium versus silicon for the coarse mass fraction showed little or no correlation demonstrating that sea spray at our sites was the dominant sodium source for the coarse mass fraction.

Figure 28.

Box and whisker plot for percentage coarse salinity at Hong Kong, Cheju Island, and Sado Island during the study period to 30 June 2002.

[87] The median salinity was 68% at the Hong Kong site, 47% at the Cheju Island site and 65% at the Sado site of the total coarse mass fraction. Addition of the fine and coarse measured masses gave the PM10 size fraction for each sampling day. Table 12 is a summary of the PM10 mass fraction measurements for the three ACE-Asia sites for the study period to 30 June 2002.

Table 12. Summary of the PM10 Mass Measurements for Hong Kong, Cheju Island, and Sado Island for the Study Period to 30 June 2002
PM10 MassHong KongCheju IslandSado Island
Mean, μg/m3643928
Median, μg/m3622921
Standard deviation, μg/m3323032
Maximum, μg/m3152176276

[88] A comparison of the PM2.5 mass data from Table 2 with the data of Table 12 shows that PM2.5 was a significant percentage of the PM10 size fraction. Figure 29 is a plot of the PM10 mass against the PM2.5 mass for the data collected at all three sites. PM10 and PM2.5 mass fractions were well correlated, with the ratio (PM10/PM2.5) = (2.1 ± 0.4), demonstrating that on average, across all sites, the PM2.5 fraction represented about half of the total PM10 mass.

Figure 29.

Correlation plot of PM10 versus PM2.5 mass fractions for all ACE-Asia sites for the 18-month study period.

[89] Points to the left of the least squares fitted line (reduced PM2.5) represent events having a higher percentage of coarse particles, such as sea salt and soil. While points to the right represent higher fractions of PM2.5 particles originating from combustion sources.

[90] Ion Beam Analysis (IBA) was performed on all filters obtained using GENT stacked filter units as well. This enabled (PM10/PM2.5) ratios to be measured for many of the elements obtained in this study. Figure 30 shows such a plot sorted with increasing ratios from left to right.

Figure 30.

The [PM10/PM2.5] ratio versus element measured for all sites for the 18-month study period.

[91] Elements associated with combustion sources such as black carbon, sulfur, lead, zinc etc clustered near unity, demonstrating the low coarse and high fine fraction. Whereas elements associated with sources having high coarse and low fine components such as soil and sea salt cluster to the right well above unity. The element chlorine had the highest measured (PM10/PM2.5) ratio, again confirming that this element was associated with coarser sea spray sources rather than fine combustion sources such as motor vehicles or biomass burning.

11. Summary

[92] This study provides, for the first time, total mass as well as chemical composition of fine (PM2.5) particles, sampled on the same days, at three key ACE Asia study sites spanning an eighteen month period. The sites were at Hong Kong, Cheju Island off South Korea and Sado Island off the west coast of Japan. Fine particles were measured twice weekly at each of these three sites and more often during extreme pollution events related to particulate matter transported from mainland China. Median 24 hour PM2.5 mass loadings of 29 μg/m3, 16 μg/m3 and 9.1 μg/m3 were measured at Hong Kong, Cheju and Sado Islands during the study period. The corresponding maximum PM2.5 values for the three sites were 109 μg/m3, 81 μg/m3 and 78 μg/m3, respectively.

[93] Ion beam analysis, together with laser absorption techniques, was used to quantify over twenty different chemical species for the PM2.5 size fraction. These included, total hydrogen, black carbon, F, Na, Al, Si, P, S, Cl, K, Ca, Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Zn, Br, and Pb with minimum detection limits close to or below 1 ng/m3. The key species of interest in the ACE Asia project were windblown soils, ammonium sulfate and black carbon emissions all of which were directly measured or estimated from the above suite of elements.

[94] Wind blown soil at the three sampled sites was typically 7% to 9% of the PM2.5 total mass and was represented by the sum of Al (6.9%), Si(22%), K(4.1%), Ca(4.6%), Ti(0.74%), Mn(0.30%) and Fe(6.8%) with oxygen representing the remaining fraction at 55%. All three sites were impacted by significant fine soil events during the study period with PM2.5 soil concentrations up to 26 μg/m3 or 35% of the fine mass. These high soil days were most likely to occur in spring and were due to soil transported thousands of kilometers from the Chinese mainland [Zhang et al., 1997; Mori et al., 2002; Kim et al., 2003].

[95] Sulfate was predominantly anthropogenic (98%) and estimated from the measured sulfur concentrations with median values of 8.7 μg/m3, 4.8 μg/m3 and 2.9 μg/m3 for the Hong Kong, Cheju Island and Sado Island sites, respectively during the study. There were significant seasonal variations, particularly at the Hong Kong site, concentrations being higher in the winter months. The corresponding maximum PM2.5 sulfate values for the three sites were 30 μg/m3, 19 μg/m3 and 12 μg/m3, respectively. The average sulfate composition was fairly consistent across all three sites ranging from 30% to 32% of the total fine mass.

[96] Black carbon was estimated using [He/Ne] laser absorption techniques and assuming a mass absorption coefficient of 7 m2/g for laser light. There were strong seasonal variations being several times larger in the winter than the summer months with median values or 1.8 μg/m3, 1.1 μg/m3 and 0.77 μg/m3 for the Hong Kong, Cheju Island and Sado Island sites, respectively during the study period. This represented a fairly consistent 8% to 9% of the total fine mass across all three sites. The average PM2.5 percentage composition by weight, across the three sites was estimated to be around (8.4 ± 4)% black carbon, (7.7 ± 7)% soil, (43 ± 14)% ammonium sulfate, (11 ± 16)% organic matter, (10 ± 12)% salinity and (0.6 ± 0.3)% trace elements. The remainder was mainly composed of water vapor and nitrates not measured in this experiment.

[97] The PM10 size fraction was approximately equally split between the fine and coarse size fractions with the average [PM10/PM2.5] mass ratio being (2.1 ± 0.4) across all sites for the study period. The coarse fraction, across all three sites, was dominated by sea salts (59 ± 25)% and windblown soil (17 ± 11)% as expected.

[98] This quantitative data on chemical and physical composition of fine particulates over an 18-month period collected on the same days at three different Asian sites fulfils one of the goals of the ACE Asia project and provides useful unique input for aerosol chemical characterization and climatic modeling studies for the east Asia region.

Acknowledgments

[99] We would like to acknowledge the help of local staff at each of the sampling site locations for regular filter changes throughout this 18-month study. We are also indebted to all the 3 MV Van de Graaff accelerator staff at ANSTO for assistance with all the IBA measurements. This research was supported by the ANSTO, Hong Kong Polytechnic University, the Research Grants Council of the Hong Kong Special Administrative Region (project 5063/01E) and the NRL project of Korea Ministry of Science and Technology (project M1-0001-00-0018).

Ancillary