Professional Documents
Culture Documents
Chang 2008
Chang 2008
Abstract
Data collected from the ve air-quality monitoring stations established by the Taiwan Environmental Protection Administration in
Taipei City from 1994 to 2003 are analyzed to assess the temporal variations of air quality. Principal component analysis (PCA) is
adopted to convert the original measuring pollutants into fewer independent components through linear combinations while still
retaining the majority of the variance of the original data set. Two principal components (PCs) are retained together explaining 82.73%
of the total variance. PC1, which represents primary pollutants such as CO, NOx, and SO2, shows an obvious decrease over the last 10
years. PC2, which represents secondary pollutants such as ozone, displays a yearly increase over the time period when a reduction of
primary pollutants is obvious. In order to track down the control measures put forth by the authorities, 47 days of high PM10
concentrations caused by transboundary transport have been eliminated in analyzing the long-term trend of PM10 in Taipei City. The
temporal variations over the past 10 years show that the moderate peak in O3 demonstrates a signicant upward trend even when the
local primary pollutants have been well under control. Monthly variations of PC scores demonstrate that primary pollution is signicant
from January to April, while ozone increases from April to August. The results of the yearly variations of PC scores show that PM10 has
gradually shifted from a strong correlation with PC1 during the early years to become more related to PC2 in recent years. This implies
that after a reduction of primary pollutants, the proportion of secondary aerosols in PM10 may increase. Thus, reducing the precursor
concentrations of secondary aerosols will be an effective way to lower PM10 concentrations.
r 2007 Elsevier Ltd. All rights reserved.
Keywords: Pollutant trend; Primary pollutants; Secondary pollutants; Principal component analysis; PM10
0301-4797/$ - see front matter r 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jenvman.2006.12.029
ARTICLE IN PRESS
628 S.C. Chang, C.-T. Lee / Journal of Environmental Management 86 (2008) 627635
CO (ppm)
and maintained to ensure data quality. Scheduled quality 4
control procedures included daily zero and span checks,
biweekly precision checks, quarterly multiple-point cali- 2
SO2 (ppb)
20
variations on the study and obtain spatial representation,
we calculated the average of the daily pollutant averages 10
obtained at all ve stations for a long-term trend study. In
addition, only daily averages with more than 16 effective 0
hours per day were considered valid in this study. 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
For the study of ozone characteristics, we adopted the Year
spatial average of O3, max. Eder et al. (1993) used O3, max to
analyze the spatial and temporal variability of ozone over 200
NOX
non-urban regions in the eastern US. This concept is based 150
NOX (ppb)
180
will better index the seriousness of air pollution.
120
2.4. Statistical analysis 60
0
Factor analysis is useful in identifying variations of
1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
recurring and independent modes if they exist in the data
Year
set (Vukovich and Sherwell, 2003). In this study, we
performed PCA to determine the variation patterns of ve
200
pollutants, i.e., SO2, CO, NOx, PM10, and O3, from 1994 to O3, max
O3, max (ppb)
linear combinations while still retaining the majority of Fig. 2. Time series of the pollutants adopted in the air-quality analysis of
the variance of the original data set. However, the exact Taipei City from 1994 to 2003 (a) CO, (b) SO2, (c) NOx, (d) PM10 and
number of retained components remains a subject of (e) O3, max.
controversy (Lehman et al., 2004). When conducting the
study of O3, max for non-urban regions in the eastern
United States, Eder et al. (1993) applied the Scree test, total. Vukovich and Sherwell (2003) used PCA to analyze
which looks for a discontinuity in the curve dened by O3, max with meteorological factors for the Baltimore-
plotting the percentage variance explained for each Washington corridor and adopted the 75% variance rule
component in the order extracted, to retain six principal (the explained cumulative variance of retained components
components (PCs) with 64% cumulative variance of the exceeding 75% of the total).
ARTICLE IN PRESS
630 S.C. Chang, C.-T. Lee / Journal of Environmental Management 86 (2008) 627635
3. Results and discussion PC2 loading of O3, max is 0.95, which represents the
contribution of photochemical reactions. The PC1 loading
3.1. Implication of the retained principal components of SO2 is 0.72 implying that most SO2 in Taipei City has a
variance similar to CO and NOx; however, some SO2 might
Table 1 lists the loadings, eigenvalues, and variances be contributed by sources other than local motor vehicles,
explained of the unrotated PCs for the original data set in e.g., those from neighboring Keelung City and even from
this study. The rst two PCs (PC1 and PC2) accumulated transboundary transportation (Chang et al., 2000; Lin,
81.13% variance of total. Variances explained in the third 2001; Lin et al., 2005; Chang and Lee, 2006a). It should be
and fourth PCs (PC3 and PC4) were only at 9.63% and noted that the PC1 loading of PM10 is 0.64, but its PC2
7.91%, respectively, which are much less than 18.84% of loading is slightly lower at 0.52. The PM10 loading in both
PC2. Since only ve variables were included in the analysis, PC1 and PC2 indicate that in addition to local sources of
we retained two PCs for parsimony purpose. We then emissions, other factors such as the photochemical reac-
applied the orthogonal Varimax rotation of the retained tions in producing secondary aerosols (Rodr guez et al.,
PCs to increase their independence as shown in Table 2. 2002), and yellow sand (Lin, 2001; Lin et al., 2005) also
After the Varimax rotation, the cumulative variance cause PM10 to vary signicantly.
explained by these two PCs remained at 81.13% of the
total, but their variances explained become homogeneous,
and each individual component loading is maximized. 3.2. Separation of transboundary transport from the long-
The PC1 loadings of CO and NOx are 0.95 and 0.94, term trend in PM10
respectively. Both loadings are greater than 0.9, indicating
that CO and NOx have strong positive correlations with Recently, Asian sandstorms were determined to have
the PC1 (Henry and Hidy, 1982). According to the Taiwan caused high PM10 concentrations in Taiwan (Lin, 2001; Lin
EPAs emission inventory, CO and NOx in the emissions et al., 2005). In order to track down the control measures
from motor vehicles in Taipei City make up 98% and 77%, put forth by the authorities, the transboundary transport
respectively, of the total emissions (Taiwan EPA, 2005). of PM10 from outside sources must be eliminated from the
Thus, PC1 represents the primary pollutant contributed analysis of the long-term trend or seasonal variation of
from local vehicle emissions. The PC1 loading of O3, max is PM10 in Taipei City.
0.07; therefore, it is not related to the variance of PC1. The Previous results show that a signicant increase of PC2
loading in PM10 may indicate contributions from second-
Table 1 ary aerosols produced in strong photochemical activities.
The component loadings and explained variance from principal compo- Based on our previous study (Chang and Lee, 2006b), the
nent analysis (PCA) of daily mean concentrations of SO2, CO, NOx, O3, max at 60 ppb is an appropriate index for separating low
PM10, and daily maximum concentration of O3 (18,250 records)
from high photochemical activity in Taipei City. By using
PC1 PC2 PC3 PC4 PC5 this criterion, we think that the enhancement of secondary
aerosol in PM10 might be found for the level of O3, max
SO2 0.81 0.03 0.42 0.40 0.02
higher than 60 ppb, while less PM10 is contributed from
CO 0.91 0.28 0.23 0.02 0.19
O3, max 0.46 0.83 0.28 0.10 0.01 secondary aerosol for O3, max less than 60 ppb. Further-
PM10 0.80 0.20 0.30 0.47 0.00 more, it is noted that local pollutants tend to accumulate
NOx 0.87 0.36 0.28 0.02 0.17 under low wind speed; in contrast, the yellow sand is
Eigenvalue 3.11 0.94 0.48 0.40 0.07 transported by Mongolian high-pressure systems with
Variance explained (%) 62.30 18.84 9.63 7.91 1.33
strong winds to Taipei City. Thus, scatter plots of PM10
Cumulative variance (%) 62.30 81.13 90.76 98.67 100.00
and PC1 scores were segregated by ozone concentration
and wind speed (Fig. 3).
When the wind speed is less than 1 m s 1, or when O3, max
Table 2 is higher than 60 ppb and the wind speed is less than
The rst two component loadings resulting from PCA after Varimax 2 m s 1, PM10 PC1 scores are larger than for at 0.64 as
rotation and explained variance for the ve major air pollutants collected shown in Fig. 3(a), (b) and (d), and Table 2. When O3, max
in Taipei City from 1994 to 2003
is less than 60 ppb and the wind speed is higher than
PC1 PC2 1 m s 1, or when O3, max is higher than 60 ppb and the wind
speed is higher than 2 m s 1, there is a weaker correlation
SO2 0.72 0.37
between PM10 and PC1 scores than for the full data set at
CO 0.95 0.13
O3, max 0.07 0.95 0.64 as shown in Fig. 3(c), (e) and (f), and Table 2. The
PM10 0.64 0.52 correlation is the least when O3, max is less than 60 ppb, and
NOx 0.94 0.04 the wind speed is greater than 2 m s 1 as shown in Fig. 3(e).
Eigenvalue 2.73 1.33 As PC1 represents local emission contribution, a smaller
Variance explained (%) 54.60 26.53
correlation coefcient between PM10 and PC1 scores
Cumulative variance (%) 54.60 81.13
indicates that the high PM10 concentrations are affected
ARTICLE IN PRESS
S.C. Chang, C.-T. Lee / Journal of Environmental Management 86 (2008) 627635 631
240
80
0
240
c d
PM10 (g m-3)
r = 0.62 r = 0.73
160
80
0
240
e r = 0.48
WS: > 2 (m sec-1)
f r = 0.58
160
80
0
-4 -2 0 2 4 6 8 -4 -2 0 2 4 6 8
O3, max : <= 60 (ppb) O3, max : > 60 (ppb)
PC 1 scores
Fig. 3. Scatter plots of PM10 associated with PC1 scores at different O3 and WS levels.
Table 4 to be greater than the long-term average. Fig. 4(a) also shows
The rst two component loadings resulted from PCA after Varimax that in 1994 and 1995, PC1 scores were frequently greater
rotation and explained variance for 10 years of the ve major air
than 3 something which did not occur in the most recent 3
pollutants collected in Taipei City after sorting out 47 days of high-PM10
events years, indicating that local primary pollutants have been well
under control (Chang and Lee, 2006a).
PC1 PC2 Fig. 4(b) plots the PC2 scores representing photochemi-
cal pollution. PC2 scores also demonstrate extremes
SO2 0.73 0.36
CO 0.95 0.12 (greater than 3) that appeared in 1994 and 1995. Although
O3, max 0.08 0.96 the extremely high values of PC2 scores did not increase for
PM10 0.70 0.51 the years after 1995, positive PC2 scores show a tendency
NOx 0.94 0.04 of increasing annually that is grossly different from the
Eigenvalue 2.81 1.33
Variance explained (%) 56.15 26.58
aforementioned PC1 scores. The moderate peak in O3
Cumulative variance (%) 56.15 82.73 demonstrates a signicant upward trend even if the local
primary pollutants have been well under control (Chang
and Lee, 2006b). It seems that the current strategy of air
represent the source contributions from local source pollution control should be adjusted in order to control the
activities in Taipei City. The PCA is rerun with Varimax increasing moderate peak in O3.
rotation while retaining the two PCs. The results include the
loadings, eigenvalues, and variances explained of the two 3.4. Monthly variations of local and photochemical
PCs, as shown in Table 4. The cumulative variance pollutions
explained amounts to 82.73% of the total variance.
The results of the PCA after deducting the inuence of The time series of the two PC scores demonstrate
transboundary transport show that the component load- obvious temporal variations, as shown in Fig. 4. If the
ings of 0.95 and 0.94 for NOx and CO, respectively, are still variation in pollution caused by anthropogenic emission
the two highest PC1 loadings. The component loading of sources in a basin is small such as in Taipei City, seasonal
O3, max is 0.08 in the PC1 and 0.96 in PC2. For SO2, the variations of various pollutants will be dominated by
component loading in the PC1 and the PC2 are 0.73 and changes in meteorological conditions. This is an analogous
0.36, respectively. These results are similar to those to the accumulation of pollutants found in Milan due to
obtained without deducting the inuence of transboundary persistent thermal inversions (Marcazzan et al., 2001).
transport. However, the PM10 loading in PC1 increases to Yang (2002) also reported that under the inuence of low
0.70 and slightly decreases to 0.51 in the PC2. This temperatures and fewer rainy days in winter, the PM10
indicates that after deducting the inuence of transbound- concentration was observed to be high in southern Taiwan.
ary transport, PC1 will better represent PM10 pollution In order to show monthly variations, we made the box
from local emissions, while the PC2 still corresponds to plots of daily PC scores in a like month (around 300 values)
PM10 produced from photochemical reactions. in Figs. 5(a) and 6(a) and calculated the median of the PC
scores for a like day (a 10-value average) over the 10-year
3.3. Trend of local and photochemical pollutions period in Figs. 5(b) and 6(b). Using the median instead of
the mean will prevent the dominance of any single value in
As the magnitude of PC scores is the linear combination determining the typical day over the 10-year period (Eder
of PC loadings from various pollutants (Lehman et al., et al., 1993). Fig. 5(a) shows a larger uctuation of PC1
2004), positive PC scores indicate that the pollutant scores from January to April, while the median of PC1
concentration for a day is higher than the mean concentra- scores shows positive scores at the same period, as shown
tion, and vice versa for the negative PC scores. A distance- in Fig. 5(b). After May, most days have negative PC scores.
weighted least-squares smoothing procedure (the inuence This indicates that incidents of air pollution caused by the
of individual points decreases with the horizontal distance local emissions mostly occur from January to April. When
from the respective points on the curve) (Statistica Rel. 6.0, Taipei City sits in the warmer and relatively calm region
Statsoft Incorporated, 2001) is used to smooth out the PC before a cold front, pollutants frequently accumulate as
scores; the results will provide valuable references for was the case in Tel Aviv, Israel (Graham et al., 2004). An
evaluating long-term trends. accumulation of pollutants also occurs with the decreasing
Fig. 4 shows the time series of the two rotated PC scores. northeastern wind during the change of seasons. However,
Fig. 4(a) illustrates the variations of PC1, which represents the period between late January and early February is
local pollution. The positive scores occurred mostly in the exceptional. Chinese New Year holidays always fall within
early 5 years (19941998) and decreased yearly, while the this period, when residents living in Taipei City return to
negative scores increased annually. Since the average and their hometowns, thus causing an empty period of about 6
standard deviation of PC scores are 0 and 1, respectively, for days and leading to an apparent reduction of primary
those PC scores three times greater than the standard pollutants. Hence, PC1 scores show uncommon negative
deviation, therefore, the air pollutant concentration is shown values from the end of late January to early February.
ARTICLE IN PRESS
S.C. Chang, C.-T. Lee / Journal of Environmental Management 86 (2008) 627635 633
8
a
6 PC 1 scores
PC 1 scores
4
-2
-4
6
b PC 2 scores
4
PC 2 scores
-2
-4
1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
Year
Fig. 4. Time series of the daily standardized principal component scores for the two retained principal components. The bold line represents the distance-
weighted least-squares smoothing curve of PCs.
8 6
6
Median 25%-75% Min-Max
4
a Median 25%-75% Min-Max
PC 2 scores
PC 1 scores
4 2
2 0
0 -2
-2 -4
1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12
Month Month
3 2
2 PC 1 scores b PC 2 scores
PC 1 scores
PC 2 scores
1
1
0
0
-1 -1
-2 -2
1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12
Month Month
Fig. 5. Seasonal variations of the rst component (PC1) scores over a 10- Fig. 6. Seasonal variations of the second component (PC2) scores over a
year period (a) box plots of daily distribution in a like month, (b) the time 10-year period (a) box plots of daily distribution in a like month, (b) the
series and the median PC1 scores for a like day from 1994 to 2003. The time series and the median PC2 scores for a like day from 1994 to 2003.
bold line represents the distance-weighted least-squares smoothing curve The bold line represents the distance-weighted least-squares smoothing
of PC1. curve of PC2.
During the summer months; however, the strong south- Fig. 6(a) indicates that there are larger variations of PC2
western wind or thermal convection provides favorable scores from April to August, while the median of PC2
conditions for pollutants to disperse, resulting in relatively scores shows positive scores at the same time period, as
lower pollutant concentrations. When fall begins, the wind shown in Fig. 6(b). After April, the temperature rises, and
speed picks up, and the meteorological conditions become high temperatures favor the formation of ozone. According
favorable for dispersing pollutants such that they have to the statistics published by the Central Weather Bureau
difculty accumulating locally. (CWB http://www.cwb.gov.tw/), the lowest averages of
ARTICLE IN PRESS
634 S.C. Chang, C.-T. Lee / Journal of Environmental Management 86 (2008) 627635
monthly temperatures in Taiwan take place in January and from the two retained PCs amounts to 82.73% of the total
February at 13.3 and 13.6 1C, respectively. The highest variance. PC1, which represents primary pollutants such as
average of monthly temperature can reach 25.9 1C, creating CO, NOx, and SO2, has shown a decreasing trend for the
meteorological conditions favorable to ozone formation. In last 10 years. A reduction of primary pollutant concentra-
the summer, from May to August, both the solar insolation tions found in this study demonstrates that the control
and temperature are high, causing high ozone concentra- measures implemented by the authorities are effective.
tions and positive PC2 scores. It is worth noting that the However, PC2, that represents secondary pollutants, e.g.,
weakening northeastern monsoon in April causes positive ozone, shows an annually increasing trend after the
scores for both the PC1 and the PC2. reduction of local pollutant emissions.
The resolved two PCs are further used for sorting out
cases with PC1 scores lower than 1.5 and O3, max
3.5. Source contribution shifting in PM10
concentrations less than 60 ppb, which were classied as
originating from transboundary transport such as from
Previous analyses indicate that the primary pollutant
Asian yellow sand. After removing these individual cases,
concentration in Taipei City is much improved, but ozone
the results of the long-term tendencies or seasonal patterns
pollution is becoming worse. In order to understand the
will better represent the source contributions from local
variations in the relationships between PM10 concentra-
activities in Taipei City.
tion and other pollutants after the decrease of primary
Air pollution concentrations clearly show monthly
pollutants, this study compiled yearly PC1 and PC2
variations in Taipei City. Most of the PC1 scores are
loadings from 1994 to 2003 in Table 5. From 1994 to
positive from January to April averaged in 10 years
1998, the PC1 loadings of PM10 range from 0.63 to 0.74,
because of poor dispersion resulting in pollutant accumu-
which are similar to the average of 0.70 for the decade. In
lation. The PC2 scores are positive from April to August,
contrast, the PC2 loadings of PM10 are between 0.46 and
indicating high photochemical pollution. After September,
0.64. It is worth noting that starting in 1999, all the yearly
the wind speed increases, and the temperature drops. Both
loadings of the PC1 of PM10 are less than 0.6, and the
the scores of the PC1 and the PC2 become negative,
loadings of the PC2 of PM10 become higher than that in
indicating low pollutant concentrations from local source
PC1. For an urban area, most PM10, CO, and NOx
activities.
emissions originate from motor vehicles. These three
The PM10 concentration was strongly correlated to local
pollutants should have high correlations with one another
pollution during early years, but has gradually changed to
(Smith et al., 2001). The PCA results indicate that the PM10
being more correlated with the variations of ozone
concentration was highly correlated with local pollution in
concentration in recent years. After the reduction of local
the early years, but in recent years, it has become more
primary pollutant concentrations, secondary aerosol is
related to ozone concentration. Thus, the reduction of local
expected to contribute more to PM10 concentration. The
primary emissions will lower the primary PM10. However,
study on ozone formation and aerosol characterization for
increasing ozone concentrations will trigger higher second-
source apportionment hence becomes a need in the near
ary aerosol contribution to PM10.
future.
4. Conclusions
Acknowledgments
This study adopted PCA to analyze the air-quality
monitoring data collected from 1994 to 2003 in Taipei City We are grateful to the Taiwan Environmental Protection
by the Taiwan EPA. The cumulative variance explained Administration (TEPA) for providing the monitoring data
Table 5
The rst two component loadings resulted from PCA after Varimax rotation and explained variance for each year of the ve major air pollutants collected
in Taipei City
1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
PC1 PC2 PC1 PC2 PC1 PC2 PC1 PC2 PC1 PC2 PC1 PC2 PC1 PC2 PC1 PC2 PC1 PC2 PC1 PC2
SO2 0.49 0.65 0.53 0.69 0.77 0.39 0.51 0.71 0.71 0.53 0.57 0.54 0.48 0.65 0.39 0.71 0.42 0.70 0.29 0.76
CO 0.94 0.23 0.95 0.21 0.92 0.23 0.93 0.22 0.94 0.17 0.92 0.26 0.95 0.17 0.94 0.17 0.95 0.23 0.93 0.28
O3, max 0.08 0.93 0.03 0.93 0.18 0.96 0.10 0.92 0.11 0.95 0.13 0.91 0.00 0.91 0.04 0.87 0.08 0.89 0.11 0.80
PM10 0.72 0.54 0.63 0.64 0.74 0.46 0.63 0.61 0.70 0.48 0.55 0.57 0.53 0.65 0.53 0.60 0.50 0.71 0.42 0.72
NOx 0.97 0.06 0.97 0.10 0.95 0.12 0.96 0.17 0.96 0.08 0.96 0.11 0.97 0.10 0.96 0.13 0.96 0.16 0.96 0.13
Eigenvalue 2.59 1.62 2.52 1.81 2.93 1.36 2.46 1.79 2.81 1.46 2.42 1.54 2.35 1.71 2.25 1.67 2.26 1.86 2.06 1.83
Variance explained (%) 51.85 32.47 50.49 36.12 58.58 27.19 49.14 35.80 56.12 29.30 48.50 30.71 47.00 34.23 44.93 33.38 45.28 37.23 41.13 36.65
Cumulative variance (%) 51.85 84.32 50.49 86.61 58.58 85.77 49.14 84.94 56.12 85.42 48.50 79.21 47.00 81.23 44.93 78.31 45.28 82.51 41.13 77.78
ARTICLE IN PRESS
S.C. Chang, C.-T. Lee / Journal of Environmental Management 86 (2008) 627635 635
used in this study. Although, the data were taken from the Ho, K.F., Lee, S.C., Chan, C.K., Yu, J.C., Chow, J.C., Yao, X.H., 2003.
TEPA ofcial monitoring network, the results of this paper Characterization of chemical species in PM2.5 and PM10 aerosols in
Hong Kong. Atmospheric Environment 37, 3139.
were not peer-reviewed by TEPA, and the mention of
Hooper, R.P., Peters, N.E., 1989. Use of multivariate analysis for
instrument trade names does not connote the endorsement determining sources of solutes found in wet atmospheric deposition
of TEPA. We would also like to show our gratitude for the in the United States. Environmental Science and Technology 23,
contributions from two anonymous reviewers for their 12631268.
constructive comments. Hosiokangas, J., Rhuskanen, J., Pekkanen, J., 1999. Effects of soil
dust episodes and mixed fuel sources on source apportionment of
PM10 particles in Kuopio, Finland. Atmospheric Environment 33,
38213829.
References Lehman, J., Swinton, K., Bortnick, S., Hamilton, C., Baldridge, E., Eder,
B., Coxb, B., 2004. Spatio-temporal characterization of tropospheric
Ashbaugh, L.L., Myrup, L.O., Flocchin, R.G., 1984. A principal ozone across the eastern United States. Atmospheric Environment 38,
component analysis of sulfur concentrations in the western United 43574369.
States. Atmospheric Environment 18, 783791. Lin, C.Y., Liu, S.C., Chou, C.K., Huang, S.J., Liu, C.M., Kuo, C.H.,
Buhr, M.P., Trainer, M., Parrish, D.D., Sievers, R.E., Fehsenfeld, F.C., Young, C.Y., 2005. Long-range transport of aerosols and their
1992. Assessment of pollutant emission inventories by principal impact on the air quality of Taiwan. Atmospheric Environment 39,
component analysis of ambient air measurements. Geophysical 60666076.
Research Letters 19, 10091012. Lin, T.H., 2001. Long-range transport of yellow sand to Taiwan in Spring
Buhr, M.P., Parrish, D., Elliot, J., Holloway, J., Carpenter, J., Goldan, P., 2000: observed evidence and simulation. Atmospheric Environment
Kuster, W., Trainer, M., Montzka, S., McKeen, S., Fehsenfeld, F., 35, 58735882.
1995. Evaluation of ozone precursor source types using principal Liu, C.M., Huang, C.Y., Shieh, S.L., Wu, C.C., 1994. Important
component analysis of ambient air measurements in rural Alabama. meteorological parameters for ozone episodes experienced in the
Journal of Geophysical Research 100, 2285322860. Taipei Basin. Atmospheric Environment 28, 159173.
Chang, K.H., Jeng, F.T., Tsai, Y.L., Lin, P.L., 2000. Modeling of long- Marcazzan, G.M., Vaccaro, S., Valli, G., Vecchi, R., 2001. Characteriza-
range transport on Taiwans acid deposition under different weather tion of PM2.5 and PM10 particulate matter in the ambient air of Milan
conditions. Atmospheric Environment 34, 32813295. (Italy). Atmospheric Environment 35, 46394650.
Chang, S.C., Lee, C.T., 2006a. Evaluation of the trend of air quality in Rodr guez, S., Querol, X., Alastuey, A., Mantilla, E., 2002. Origin of high
Taipei, Taiwan from 1994 to 2003. Environmental Monitoring and summer PM10 and TSP concentrations at rural site in Eastern Spain.
Assessment, in press, doi:10.1007/s10661-006-9262-1. Atmospheric Environment 36, 31013112.
Chang, S.C., Lee, C.T., 2006b. Ozone variations through vehicle Smith, S., Stribley, F.T., Milligan, P., Barratt, B., 2001. Factors
emissions reductions based on air quality monitoring data in Taipei inuencing measurements of PM10 during 19951997 in London.
City, Taiwan, from 1994 to 2003. Atmospheric Environment 40, Atmospheric Environment 35, 46514662.
35133526. StatSoft, Inc., 2001. STATISTICA (data analysis software system),
Chou, C.K., Huang, S.H., Chen, T.K., Lin, C.Y., Wang, L.C., 2005. Size- version 6.0 /www.statsoft.comS.
segregated characterization of atmospheric aerosols in Taipei during Taiwan EPA, 2005. Environmental database. /http://edb.epa.gov.tw/S.
Asian outow episodes. Atmospheric Research 75, 89109. Tao, Z., Larson, S.M., Williams, A., Caughey, M., Wuebbles, D.J., 2005.
Eder, B.K., 1989. A principal component analysis of SO24 -precipitation Area, mobile, and point source contributions to ground level ozone: a
concentrations over the eastern United States. Atmospheric Environ- summer simulation across the continental USA. Atmospheric Envir-
ment 23, 27392750. onment 39, 18691877.
Eder, B.K., Davis, J.M., Bloomeld, P., 1993. A characterization van der Wal, J.T., Janssen, L.H.J.M., 2000. Analysis of spatial and
of the spatiotemporal variability of non-urban ozone concentrations temporal variations of PM10 concentrations in the Netherlands using
over the eastern United States. Atmospheric Environment 27A, Kalman ltering. Atmospheric Environment 34, 36753687.
26452668. Vukovich, F.M., 1994. Boundary layer ozone variations in the eastern
Graham, B., Falkovich, A.H., Rudich, Y., Maenhaut, W., Guyon, P., United States and their association with meteorological variations:
Andreae, M.O., 2004. Local and regional contributions to the long-term variations. Journal of Geophysical Research 99 (D3),
atmospheric aerosol over Tel Aviv, Israel: a case study using elemental, 1683916850.
ionic and organic tracers. Atmospheric Environment 38, 15931604. Vukovich, F.M., Sherwell, J., 2003. An examination of the relationship
Henry, R.C., Hidy, G.M., 1979. Multivariate analysis of particulate between certain meteorological parameters and surface ozone varia-
sulfate and other air quality variables by principal componentsPart tions in the Baltimore-Washington corridor. Atmospheric Environ-
I: annual data from Los Angeles and New York. Atmospheric ment 37, 971981.
Environment 13, 15811596. Yang, K.L., 2002. Spatial and seasonal variation of PM10 mass
Henry, R.C., Hidy, G.M., 1982. Multivariate analysis of particulate concentrations in Taiwan. Atmospheric Environment 36, 34033411.
sulfate and other air quality variables by principal componentsPart Yu, T.Y., Chang, L.F., 2001. Delineation of air quality basins utilizing
II: Salt Lake City, Utah, and St. Louis, Missouri. Atmospheric multivariate statistical methods in Taiwan. Atmospheric Environment
Environment 16, 929943. 35, 31553166.