- Email: [email protected]
Identifying risk sources of air contamination by polycyclic aromatic hydrocarbons
Accepted Manuscript Identifying risk sources of air contamination by polycyclic aromatic hydrocarbons
Jiri Huzlik, Frantisek Bozek, Adam Pawelczyk, Roman Licbinsky, Magdalena Naplavova, Michael Pondelicek PII:
S0045-6535(17)30673-2
DOI:
10.1016/j.chemosphere.2017.04.131
Reference:
CHEM 19193
To appear in:
Chemosphere
Received Date:
10 February 2017
Revised Date:
17 March 2017
Accepted Date:
26 April 2017
Please cite this article as: Jiri Huzlik, Frantisek Bozek, Adam Pawelczyk, Roman Licbinsky, Magdalena Naplavova, Michael Pondelicek, Identifying risk sources of air contamination by polycyclic aromatic hydrocarbons, Chemosphere (2017), doi: 10.1016/j.chemosphere.2017.04.131
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT
Identifying risk sources of air contamination by polycyclic aromatic hydrocarbons
1 2 3 4
Jiri Huzlik a, Frantisek Bozek b*, Adam Pawelczyk c, Roman Licbinsky d, Magdalena Naplavova e
5
Michael Pondelicek f
6 7
a
Transport Research Centre, 33a Lisenska, 636 00 Brno, Czech Republic, telephone: +420 541 641 374, e-mail: [email protected]
8 9
b
University of Defence, [email protected]
10 11
c
Wroclaw University of Technology, Faculty of Chemistry, 4/6 Norwida St., 50-373 Wrocław, Poland, e-mail: [email protected]
12 13
d
Transport Research Centre, 33a Lisenska, 636 00 Brno, Czech Republic, e-mail: [email protected]
14 15
e
University of Defence, 65 [email protected]
16 17
f
The College of Regional Development, 68 Zalanskeho, 163 00 Praha 17 – Repy, Czech Republic, e-mail: [email protected]
18
*
corresponding author
65
Kounicova,
Kounicova,
662
662
10
10
Brno,
Brno,
Czech
Czech
Republic,
Republic,
e-mail:
e-mail:
19 20
Highlights
21
PAH concentrations were measured in air and exhaust gases.
22
Marker BghiPe/BaP was used to identify air pollution sources in relation to time.
23
Various statistical methods were used precisely to evaluate the BghiPe/BaP ratio.
24
In warmer periods, transport is exclusively the source of PAH air pollution.
25
In colder periods, local heating stoves contribute to the presence of PAHs in air.
26
Abstract
27
This article is directed to determining concentrations of polycyclic aromatic hydrocarbons (PAHs),
28
which are sorbed to solid particles in the air. Pollution sources were identified on the basis of the
29
ratio of benzo[ghi]perylene (BghiPe) to benzo[a]pyrene (BaP). Because various important
30
information is lost by determining the simple ratio of concentrations, least squares linear regression
31
(classic ordinary least squares regression), reduced major axis, orthogonal regression, and Kendall–
1
ACCEPTED MANUSCRIPT
32
Theil robust diagnostics were utilized for identification. Statistical evaluation using all
33
aforementioned methods demonstrated different ratios of the monitored PAHs in the intervals
34
examined during warmer and colder periods. Analogous outputs were provided by comparing
35
gradients of the emission factors acquired from the measured concentrations of BghiPe and BaP in
36
motor vehicle exhaust gases. Based on these outputs, it was possible plausibly to state that the
37
influence of burning organic fuels in heating stoves is prevalent in colder periods whereas in warmer
38
periods transport was the exclusive source because other sources of PAH emissions were not found
39
in the examined locations.
40
Keywords:
41
air pollution, benzo[a]pyrene, benzo[ghi]perylene, pollution sources, polycyclic aromatic
42
hydrocarbons, regression, transport.
43
1
44
Effectively executed protection of a population consists in the first phase of identifying the sources
45
of hazards followed by qualitative and semi-quantitative estimation or quantitative calculation of
46
those risks resulting from the identified sources of hazards, delimiting critical risks, and finally
47
proposing measures for their mitigation (Božek and Urban, 2008). Serious sources of hazards
48
currently include air pollution, to which contamination by natural sources (forest fires, volcanic
49
eruptions) contribute in part but which in recent decades is mainly due to anthropogenic sources
50
(Villar-Vidal et al., 2016). Important anthropogenic sources of pollution include in particular
51
industry, energy, transport, agricultural production, communal waste incinerators, oil spills, but also
52
to a considerable degree domestic heating stoves. There are, therefore, a number of pollutants in the
53
air which present a considerable health risk to human populations and a threat to ecosystems,
54
especially in the vicinity of busy roads and cities with high concentrations of people and industry
55
(Bozek et al., 2011; Liu et al., 2015).
56
A priority condition for successfully minimizing risks from contaminated air in heavily polluted
57
locations is thus identification of the relevant sources of such pollution. This can be carried out in
58
various ways.
Introduction
2
ACCEPTED MANUSCRIPT
59
2
60
Basic approaches to deciding about origins of emission sources essentially comprise the following:
61
a)
62 63
Analysis of the current state
morphological and dimensional determination of a complex mixture of captured solid particles (Goldstein, 1992; Leal et al., 2014),
b)
64
examination of the physical properties and chemical composition of solid particles (Leal et al., 2014; Winifred et al., 2016),
65
c)
receptor modelling (Dutton, 2010),
66
d)
measurement of characteristic markers consisting in the chemical composition of selected
67
contaminants bound to dust particles.
68
Use of the ratios of various polycyclic aromatic hydrocarbons (PAHs) is the most commonly
69
encountered method in identifying emission sources on the basis of aerosol chemical composition.
70
That is because PAHs are omnipresent, form a broad group of compounds, and many of them
71
demonstrate toxic, mutagenic, teratogenic, carcinogenic, and embryotoxic properties (Haritash and
72
Kaushik, 2009). For the given purpose, benzo[a]pyrene (BaP) seems to be an especially suitable
73
PAH due to its being stable while also demonstrating a relatively consistent and high contribution to
74
PAHs’ carcinogenic activity (Hailwood et al., 2001). The ratios of its concentration to
75
benzo[ghi]perylene (BghiPe) (Gustafsson and Gschwend, 1997) and to benzo[e]pyrene (Yang, 2006)
76
are frequently used as markers for identifying sources of pollution. The ratios of
77
indeno[1,2,3-cd]pyrene (IPy) to the sum of concentrations of IPy and BghiPe (Yang, 2006),
78
fluoranthene to pyrene, and phenanthrene to anthracene (Kocourek et al., 2003) are also applied.
79
Broad overviews of organic markers, including PAHs, which may be used for identifying sources of
80
atmospheric aerosols were completed by Genualdi (2008) and Křůmal et al. (2012).
81
Mere monitoring of pollutant concentrations ratios in the atmosphere can, however, lead to losing a
82
good deal of valuable information in relation to specifying pollution sources. For example,
83
Chuesaard et al. (2014) demonstrated through a more detailed examination fluctuations in the
84
BghiPe/BaP ratio within the interval 1.07 to 3.29 as a function of air humidity. They explained
85
decrease in the BghiPe/BaP ratio in dry periods by a contribution of emissions from biomass
86
combustion. As an indicator for assessing the contribution of biomass combustion they proposed the
87
ratio of 9-nitroanthracene and 1-nitropyrene. In parallel, dependencies of organic markers on
3
ACCEPTED MANUSCRIPT
88
variables other than air humidity can be found in the literature as can discrepancies in presenting
89
their values for a given pollution source (Genualdi, 2008).
90
The application of statistical methods in processing measured data is one possibility for resolving
91
existing ambiguities, expanding knowledge in this area, and increasing the precision of markers of
92
specific sources of air contamination under various conditions. In the present article, some statistical
93
procedures which can be applied effectively also for more exact characterization of other markers of
94
this type are presented for evaluating relationships between BaP and BghiPe concentrations in the
95
vicinity of busy urban roads.
96
3
97
BaP and BghiPe concentrations were measured in the air of two urban locations. BaP and BghiPe
98
emission factors were determined in exhaust gases of four passenger motor vehicles. The acquired
99
results became the basis for identifying air pollution sources using selected statistical methods.
Methods and devices applied
100
3.1
101
Pollution sources were identified by comparing the regression lines of BghiPe versus BaP.
102
According to our previous findings, using only the ratio of their concentrations leads to a loss of
103
certain important information. We used the linear regression module of the statistical program
104
package QC.Expert to evaluate the results of classic ordinary least squares (OLS) regressions
105
(Kupka, 2013). The fact that BaP concentrations display a smaller error of determination than do
106
BghiPe concentrations was an argument for selecting the model in the form of Equation (1), where
107
cBghiPe and cBaP represent the concentrations of BghiPe and BaP, respectively, and b1 and b0 represent
108
the coefficients of the linear term and absolute term, respectively.
109
Statistical methods
c BghiPe b0 b1 c BaP
(1)
110
The model’s reliability and regression parameter estimates were verified using the multiple
111
correlation coefficient R, coefficient of determination R2, predicted correlation coefficient Rp, mean
112
squared prediction error MSPE, and the Akaike information criterion. The legitimacy of using the
113
linear model was tested using the Fisher–Snedecor significance test, Scott multicollinearity criterion,
114
Cook–Weisberg heteroscedasticity test, Jarque–Bera normality test, Wald and Durbin–Watson
115
autocorrelation tests, and the residual sign test (Meloun and Militký, 2004). 4
ACCEPTED MANUSCRIPT
116
The robust method by Kendall and Theil, which is not sensitive to outlying values, was subsequently
117
applied. The Kendall–Theil Robust Line software is a Visual Basic program that may be used with
118
the Microsoft Windows operating system to calculate parameters for robust, nonparametric estimates
119
of linear regression coefficients between two continuous variables (Granato, 2006). The slope of the
120
line was calculated as the median of all possible pairwise slopes between points. The intercept was
121
calculated so that the line runs through the median of input data. The confidence interval of the slope
122
was calculated, as well.
123
Because it could not be expected for the OLS regression to display an error close to zero, the reduced
124
major axis (RMA) method was used to verify the results’ reliability (Turner et al., 2009; Friedman et
125
al., 2013) using RMA software (Bohonak, 2005). The method consists in concurrent minimization of
126
both the y and x axes residuals. The classical linear model (RMA L) and a model based on the
127
jackknife method (RMA JK) were applied in the study. In parallel, the orthogonal regression (OR)
128
method was included in the program Number Cruncher Statistical System (NCSS) based on
129
minimization of the perpendicular distance of regression points from the regression line (Hintze,
130
2007).
131
3.2
132
Air samples were collected over a period of 24 h using a medium-volume Leckel MVS6 sampler
133
and, similarly, the procedure presented in our previous publication (Bozek et al., 2016) was used to
134
determine the PAHs.
135
To simulate urban traffic we selected a standard test according to the Economic Commission for
136
Europe (ECE), Urban (Part One) Driving Cycle (ECE 15), Type I test (UN, 20015). It consisted of
137
four elementary urban cycles, included 15 phases, and lasted for 195 seconds, as shown in Fig. 1.
138
The measurements were taken with the engine heated to operating temperature and were repeated in
139
three immediate successions. The entire process during which the exhaust gases were collected to
140
determine PAHs is subsequently referred to as the standard driving cycle (SDC). To simulate cold
141
starts, only one test was performed under the same conditions, subsequently referred to as SDC-CS.
142
Collection was performed using an apparatus connected directly to the vehicle’s exhaust pipe, as
143
described by Huzlik et al. (2011).
144
Sample collection and analysis were performed according to standards COSMT (2006) and COSMT
145
(2007) while accepting the fact that measurements in a car’s exhaust pipe cannot be performed
Measurement of emissions
5
ACCEPTED MANUSCRIPT
146
isokinetically. Measurements on the TUV-ÚVMV apparatus were performed on a roller
147
dynamometer (SCHENCK 364/GS56), which simulated inertial masses and driving resistance as if
148
the vehicle was moving on a road.
149
4
150
With the objective of identifying sources of air pollution using PAHs, we evaluated concentrations of
151
BghiPe and BaP in the air. In order to compare the results obtained in this way, we used ratios of
152
measured BghiPe and BaP emission factors for passenger motor vehicles.
153
4.1
154
Results and discussion
Determination of benzo[a]pyrene and benzo[ghi]perylene concentrations in the atmosphere
155
Concentrations of PAHs bound to PM2.5 solid particles in the air were monitored in the period from
156
16 March 2007 to 2 March 2008 during eight weekly collection campaigns at locations L1 and L2 in
157
Brno, Czech Republic. Traffic intensity was TI1 = 3.6 104 vehicles per day at location L1 and
158
TI2 = 8 103 vehicles per day at location L2. To specify the pollution sources, we selected BghiPe
159
and BaP concentrations because their measured concentrations ranged at relatively high levels. At
160
each location, a total of 56 pairs of concentration values for these substances were determined. These
161
are presented in Table 1 together with the weekly median maximum tmedMax and minimum tmedMin
162
air temperatures.
163
Medians were used in order to eliminate the impact of the obliqueness of the statistical distribution
164
of data during the weekly sampling campaigns.
165
4.2
166
Determination of benzo[a]pyrene and benzo[ghi]perylene emission factors in exhaust gases
167
Emission factors EfBghiPe and EfBaP were measured in the period from June 2005 to July 2006 in the
168
driving modes “urban cycle” (SDC) and “cold start” (SDC-CS). Additive-free petrol and petrol with
169
the addition of ethanol and ethyl-tert-butyl ether as well as summer diesel fuel with and without fatty
170
acid methyl ester (FAME) from rapeseed oil as additive were used. The results are presented in
171
Table 2.
6
ACCEPTED MANUSCRIPT
172
4.3
Identification of air pollution sources using linear regression
173
In the first phase, we performed calculations of linear regression between BghiPe and BaP using the
174
OLS method and determined parameters b0 and b1 separately for each entire location. The
175
calculations did not include cases where at least one of the measured concentrations was under the
176
detection limit of LOD 0.05 ng m−3. However, the data exhibited a heteroscedasticity which could
177
not be removed even by omitting outlying data or introducing weighted regression. The residuals
178
analysis indicated that both systems act differently in the warmer period of the year W, when tmedMin
179
5°C (mid-April to early October), and in the colder period C, when tmedMin 5°C (mid-November
180
to early March). The data from each location was therefore divided into two groups according to
181
tmedMin and regression was calculated separately. The predicted residuals were predominantly
182
negative in the warmer period for both locations but predominantly positive in the colder period.
183
Regression diagnostics between BghiPe and BaP concentrations for the data set from the warmer
184
period at location L1 demonstrated that the residuals exhibit heteroscedasticity (p = 1.91 10−2).
185
Measured concentrations at points 2 and 7 stated in Table 1 are probably outliers. They were
186
therefore excluded from further data analysis and the calculations were made again. The results are
187
presented in Table 3 in the column marked L1W. This procedure eliminated the heteroscedasticity
188
from the data. Other regression statistics were satisfactory, with the exception of the data’s positive
189
autocorrelation.
190
Calculations for location L1 and the colder period were performed similarly. Based on diagnostic
191
graphs, point 54 was excluded as an outlier (see Table 1) and the calculations were repeated to
192
eliminate the corresponding concentrations. The results thus obtained are reported in Table 3 in the
193
column marked L1C. Following this correction, all tests fulfilled the regression criteria.
194
After excluding outlying points at location L2, regression was tested identically as for location L1.
195
The results indicated exclusion of the point denoted as 58 in Table 1 for the warmer period and
196
points 106 and 107 for the colder period. The results obtained after eliminating outlying
197
concentrations fulfilled the regression criteria for all tests and are shown in Table 3 in column L2W
198
for the warmer period and L2C for the colder period.
199
In order to compare all four regression lines, common regression of the sum of all tested sets was
200
calculated. The results are presented in Table 3 under column L1L2WC. In this case, however, the
7
ACCEPTED MANUSCRIPT
201
data exhibited heteroscedasticity (p = 2.8 10−5), the residuals did not have normal distribution
202
(p = 5.73 10−5), and the sign test demonstrated an increasing trend (p = 2.59 10−4).
203
Heteroscedasticity in the results remained even after eliminating outlying points.
204
Linear model parameters were subsequently calculated. The values for the warmer and colder
205
periods are presented in Table 3 under columns W and C, respectively. Only a significant positive
206
autocorrelation was demonstrated in the data for the warmer period, whereas for the colder period all
207
tests fulfilled the regression criteria.
208
In order to judge whether pollution came from the same source, it was necessary to verify the match
209
of the regression lines, the number of which was represented for the individual comparisons by the
210
symbol M. For this purpose it was first necessary to verify whether residual variances for all
211
regression lines were identical by performing Bartlett’s test of heteroscedasticity. The test was
212
processed for M independent estimates of variances with j = nj-M degrees of freedom, where nj is
213
the number of elements of the j-th set in the respective column, as is apparent from Tables 3 and 4.
214
The null hypothesis was then defined as H 0 : 2 2j ; j 1, M j , M N , all residual variances
215
being
216
H 1 : 2 2j ; j 1, M j , M N , stating that at least one of the variances is different from the
217
others. As testing criterion B we used Equation (2), in which the total degrees of freedom V is
218
calculated according to Equation (3), the summarized estimate of variance C2 according to Equation
219
(4), and parameter D according to Equation (5):
220
equal
indicating
B
its
validity,
the
alternative
hypothesis
as
M V ln { C2 } ( j ln { C2 }) j 1
(2)
D V
221
M
j 1
222
and
C2
M
( j 1
j
j
2j V 1 )
(3)
(4)
8
ACCEPTED MANUSCRIPT
M
D 1
223 224
( j 1
1 j
V 1 ) (5)
3 ( M 1)
Values of n are presented in Table 3 and were calculated using Equation (6), where j N.
n
225
M
n j !
(6)
j
226
Given the validity of hypothesis H0, the distribution of B is asymptotically Pearson’s 2 with M − 1
227
degrees of freedom and H0 is considered to be accepted at the level of significance if
228
B < 12- M 1 . The estimate of variance 2 is the combined estimate of variance 𝜎2𝑐 . Bartlett’s test
229
is sensitive to deviations of the residuals from normality. Groups of compared regression lines for
230
which the assumption that they have a common combined variance 𝜎2𝑐 was rejected are marked with
231
an asterisk in Table 4.
232
The first part of comparing the matching of the lines was to test for homogeneity of intersects
233
j k (intercepts). The null hypothesis H 0 :b0 b0 ; j , k 1, M j , k , M N was formulated, stating that
234
j k intersects are identical, and the alternative hypothesis H 1 : b0 b0 ; j , k 1, M j , k , M N ,
235
j k stating that at least one of the intersects differs from the others. Symbols b0 , b0 represent absolute
236
members of the j-th and k-th regression lines, respectively. The testing equation E of line intersects
237
j was expressed as Equation (7), where w0 denotes the weight of absolute members of the intersect of
238
c the j-th regression line calculated according to Equation (8), b0 is the combined estimate of absolute
239
members of the compared regression lines calculated using Equation (9), and RSCj is the residual
240
sum of squares of the j-th line. M
241
E
(n 2 M ) w0j (b0j b0c ) 2 j 1
M
( M 1) RSC j
(7)
j 1
9
ACCEPTED MANUSCRIPT
nj
w0j
242
n j ( xi , j x j ) 2 i 1 nj
(x
i, j
i 1
M
b0c
243
(8)
(w
j 0
j 1
nj
)
2
b0j )
w j 1
(9)
j 0
244
If the null hypothesis H0 is valid, then the testing statistic E has a Fischer–Snedecor distribution with
245
1 = M − 1 and 2 = n – 2 M degrees of freedom. If E < F1 ( M 1, n 2 M ) , then it can be
246
c stated that all lines exhibit the same intersect with the value of the absolute term b0 at the level of
247
significance .
248
The test of homogeneity of intersects with acceptance of Bartlett’s test of homoscedasticity
249
demonstrated a match of regression lines intersects of the functions BghiPe and BaP for the warmer
250
(L1W-L2W) and colder (L1C-L2C) periods at the level of significance = 0.05, including summary
251
data for the examined period at both monitored locations (L1-L2).
252
The matching of lines was then compared using a test of the homogeneity of slopes. The null
253
j k hypothesis H 0 : b1 b1 ; j , k 1, M j , k , M N was defined, stating that slopes of regression
254
lines are identical, and the alternative hypothesis H 1 : b1j b1k ; j , k 1, M j , k , M N , stating
255
that at least one slope is different from the others. The symbols b1j , b1k denote the slopes of the j-th
256
and k-th regression lines, respectively. As testing criterion G we used Equation (10), where w1j is the
257
weight of the slope of the j-th regression line calculated according to Equation (11), b1c is the
258
combined estimate of slopes calculated as a weighted combination of the estimates of individual
259
slopes b1j using Equation (12), and the other symbols have the same meanings as in Equation (7).
10
ACCEPTED MANUSCRIPT
M
G
260
(n 2 M ) w1j (b1j b1c ) 2 j 1
M
( M 1) RSC j
(10)
j 1
w j 1
261
nj
(x i 1
i, j
M
b1c
262
(w
j 1
j 1
nj
x j )2
b1j )
w j 1
(11)
(12)
j 0
263
Assuming the validity of hypothesis H0, the testing characteristic G has a Fischer–Snedecor
264
distribution with 1 = M − 1 and 2 = n − 2 M degrees of freedom. If G < F1 ( M 1, n 2 M ) , all
265
regression lines have an identical slope b1c with an estimate defined by Equation (12) at the level of
266
significance . Hypothesis H0 on matching slopes for combinations of lines L1W-L2W-L1C-L2C and
267
L2W-L2C was rejected.
268
All lines of BghiPe and BaP dependency that were found to be identical by the test of intersects of
269
the regression model (1) can also be considered identical from the aspect of slope equality. Although
270
the data from Table 6 seem to suggest that slopes of lines for the warmer and colder periods at
271
location L1 and lines of all data for the warmer and colder periods could be considered identical, this
272
cannot be regarded as conclusive due to the heteroscedasticity of the residuals, as evidenced in Table
273
4.
274
The final step in comparing the matching of lines was the general matching test, which is a
275
combination of homogeneity tests for intersects and slopes. This consists in comparing the residual
276
sum of squares RSCk obtained after interlaying all M data groups with one common line with
277
k estimates of parameters b0 , b1k and the sum of residuals calculated separately for each group. The
278
null hypothesis H0 was defined, as expressed by Equation (13) and stating that all lines are identical,
279
and the alternative hypothesis H1, as expressed by Equation (14), stating that at least one line is
280
different from the others.
11
ACCEPTED MANUSCRIPT
281
H 0 : b0j b0k b1j b1k j , k 1; M j , k , M N
(13)
282
H 1 : b0j b0k b1j b1k b0j b0k b1j b1k b0j b0k b1j b1k j , k 1; M j , k , M N
(14)
M
283
The testing criterion R was calculated according to Equation (15), where RSCc RSC j : j 1
R
284
RSCk RSCc
(n 2 M ) 2 RSCc ( M 1)
(15)
285
If H0 is valid, then the testing criterion R demonstrates a Fischer–Snedecor distribution with
286
1 = 2 (M − 1) and 2 = n − 2 M degrees of freedom. If R < F1 [2 ( M 1), n 2 M ] , then it
287
can be stated that at the level of significance all regression lines are identical with the common
288
k estimate of intersect b0 and slope b1k . The results of the calculations are presented in Table 4.
289
Asterisks mark combinations of regression lines for which hypothesis H0 in regard to their matching
290
was rejected.
291
The overall match test and the acceptance of results of Bartlett’s test of homoscedasticity indicate
292
that only pairs of lines for warmer (L1W-L2W) and colder (L1C-L2C) periods at both examined
293
locations can be considered identical. For other groups of lines, it is not possible to consider the
294
results to be conclusive.
295
4.4
296
The results of classic OLS linear regression were verified using the robust method according to
297
Kendall and Theil without eliminating outlying points. The ratios of the measured emission factors
298
were assessed using the same method. The results of the Kendall–Theil regression are presented
299
graphically in Fig. 1, which displays the 95% confidence intervals for the slopes. Confidence
300
intervals in the warmer and colder periods do not overlap and therefore differ significantly. Robust
301
diagnostics therefore provide analogous results for the values of regression parameters as does
302
classical least squares regression.
303
The analysis of regression line slopes for the motor vehicle emission factors and for the measured
304
PAHs concentrations in the air of the examined locations clearly indicates higher content of BghiPe
Robust diagnostics
12
ACCEPTED MANUSCRIPT
305
as compared to BaP in the warmer period. It can therefore be logically deduced that transport is the
306
predominant source of air pollution in the summer period. Certain differences in the slope values
307
calculated from the concentrations of BghiPe and BaP in the air and in exhaust gases can be
308
explained by the disparate composition of commercially available fuels, especially in relation to
309
additive contents. In contrast, the concentration of BaP is dominant in the colder period. Given that
310
no sources of PAHs other than transport and local heating stoves were identified in the vicinity of
311
locations L1 and L2, and because BaP is dominant in relation to BghiPe for fossil fuel combustion
312
(Křůmal et al., 2012), it is reasonable to deduce that in the winter period local heating stoves
313
contribute to air pollution in addition to transportation.
314 315
As with the Kendal–Theil regression, the RMA and orthogonal regression (OR) methods also
316
confirmed the difference in sources of PAHs in the warmer and colder periods, as is evident from
317
Table 5. It may also be stated that the slope values for emission factors in cold starts are lower than
318
those in standard cycles, which signals an increased proportion of BaP. It can therefore be reasonably
319
presumed that in the colder period the increased content of BaP versus BghiPe is caused not only by
320
the combination of sources of local combustion and operation of motor vehicles under colder
321
temperatures but apparently also by the higher number of cold starts.
322
5
323
Linear regression using the method of least squares demonstrated a statistically significant match of
324
slopes of dependencies of concentrations of BghiPe on BaP in the warmer period (campaigns 1–5)
325
for both monitored locations L1 and L2 as well as for the colder period (campaigns 6–8). The match
326
of summary regression lines for the entire year for locations L1 and L2 was not statistically
327
significant. Comparing other combinations of locations and periods using this method was shown to
328
be incorrect due to breaching the assumption of homoscedasticity of residuals. The RMA and OR
329
methods as well as the Kendal–Theil regression, which are not sensitive to outlying values,
330
confirmed the difference in dependencies of concentrations of BghiPe on BaP in the warmer and
331
colder periods. Comparing confidence intervals of regression slopes using robust methods confirmed
332
the statistical significance of the difference between slopes of the regression summary line for the
333
warmer period with the corresponding slope for the colder period.
Conclusions
13
ACCEPTED MANUSCRIPT
334
This means that the ratio of BghiPe/BaP concentrations characterized by the slope of the regression
335
line is not different for the measured locations but it is different for the different periods of the year.
336
The comparison of slopes of BghiPe and BaP emission factors obtained by measuring the
337
concentrations of both PAHs in exhaust gases of motor vehicles led to the same results. For cold
338
starts, as opposed to SDC, a higher ratio of BaP was determined relative to BghiPe in exhaust gases.
339
In view of the fact that no sources of PAHs other than transportation and local heating stoves were
340
found in the monitored locations, we may conclude that in the warmer period the only source of
341
PAHs is transportation. In the colder period, fossil fuel combustion in combination with the
342
operation of motor vehicles at lower temperatures also contributes to the content of PAHs in the air.
343
6
344
This article was produced with the financial support of the Ministry of Education, Youth and Sports
345
within the programme of long-term conceptual development of research institutions on the research
346
infrastructure and within the National Sustainability Programme I, project of the Transport R&D
347
Centre (LO1610), on the research infrastructure acquired from the Operation Programme Research
348
and Development for Innovations (CZ.1.05/2.1.00/03.0064).
349
References
350
Bohonak, A. J. 2004. RMA: Software for Reduced Major Axis Regression. Verion 1.17. [online].
351 352 353 354 355 356 357
Acknowledgement
[2016-07-29]. URL:. Bozek F. et al. 2011. Emissions of Selected Pollutants While Applying of Specific Additive EnviroxTM. WSEAS Transactions on Environment and Development, 7, (8), 233-243. Bozek, F. et al. 2016. Polycyclic Aromatic Hydrocarbon Adsorption on Selected Solid Particulate Matter Fractions. Atmospheric Environment, 126, 128-135. Božek, F., Urban, R. 2008. Management rizika. 1. vyd. Brno: Univerzita obrany, 2008. 145 s. ISBN 978-80-7231-259-7.
358
Chuesaard, T. et al. 2014, A. Influence of Biomass Burning on the Levels of Atmospheric Polycyclic
359
Aromatic Hydrocarbons and Their Nitro Derivatives in Chiang Mai, Thailand. Aerosol and Air
360
Quality Research, 14, (4), 1247-1257.
14
ACCEPTED MANUSCRIPT
361
Czech Office for Standards, Metrology and Testing (COSMT). 2006. ČSN EN 1948-1. Stationary
362
Source Emissions - Determination of Mass Concentration of PCDDs/PCDFs and Dioxin-like
363
PCBs - Part 1: Sampling of PCDDs/PCDFs. Prague: COSMT.
364
Czech Office for Standards, Metrology and Testing (COSMT). 2007. ČSN EN 1948-3. Stationary
365
Source Emissions - Determination of Mass Concentration of PCDDs/PCDFs and Dioxin-Like
366
PCBs - Part 3: Identification and Quantification of PCDDs/PCDFs. Prague: COSMT.
367
Dutton, S. J. et al. 2010. Source Apportionment Using Positive Matrix Factorization on Daily
368
Measurements of Inorganic and Organic Speciated PM2.5. Atmospheric Environment, 44, (23),
369
2731-2741.
370 371 372 373
Friedman, J., Bohonak, A. J., Levine, R. A. 2013. When are Two Pieces Better than One: Fitting and Testing OLS and RMA Regression. Environmetrics, 24, 306–31. Genualdi, S. 2008. Semi-Volatile Organic Compounds as Molecular Markers for Atmospheric and Ecosystem Transport. [Dissertation]. Corvallis: Oregon State University.
374
Goldstein, J. I. et al. 1992. Scanning Electron Microscopy and X-Ray Microanalysis. A Text for
375
Biologist, Material Scientist, and Geologist. 2nd Ed. New York, London: Plenum Press. 823 pp.
376
ISBN-13:978-1-4612-7653-1.
377
Granato, G. E. 2006. Kendall-Theil Robust Line, KTRLine, Version 1.0. A Visual Basic Program for
378
Calculating and Graphing Robust Nonparametric Estimates of Linear-Regression Coefficients
379
between Two Continuous Variables: Techniques and Methods of the U. S. Geological Survey,
380
Book
381
.
4,
Chap.
A7,
31
pp.
[online].
[2015-12-27].
URL:
382
Gustafsson, O., Gschwend, P. M. Soot as a Strong Partition Medium for Polycyclic Aromatic
383
Hydrocarbons in Aquatic Systems. In: Eganhouse, R. P. Molecular Markers in Environmental
384
Geochemistry,
385
eISBN 9780841216358.
Washington:
American
Chemical
Society,
Chapter
24,
365–381.
386
Hailwood, M . et al. 2001. Ambient Air Pollution by Polycyclic Aromatic Hydrocarbons. [Position
387
Paper]. Prepared by the Working Group On Polycyclic Aromatic Hydrocarbons.
388
ISBN 92-894-2057-X.
[online].
[2015-12-25].
URL:
15
ACCEPTED MANUSCRIPT
389
`
390
(Anexes).
391 392 393 394 395 396
Haritash, A. K., Kaushik, C. P. 2009.
Biodegradation Aspects of Polycyclic Aromatic
Hydrocarbons: A review. Journal of Hazardous Materials, 169, 1-15. Hintze, J. 2007. NCSS Statistical Software. NCSS, LLC. Kaysville, Utah. [online]. [2011-07-31]. URL:. Huzlik, J., Licbinsky, R., Durcanska, D. 2011. Polychlorinated Dibenzodioxins and Dibenzofurans Emissions from Transportation. Komunikácie, 13, (9), 41-47.
397
Kocourek, V., Hajšlová, J., Tomaniová, M., 2003. Přehled imisní zátěže agrárního ekosystému
398
vybranými prioritními organickými polutanty. Praha: Výzkumný ústav rostlinné výroby,
399
[online]. [2015-12-27]. URL:.
400 401
Křůmal, K., Mikuška, P. Večeřa, Z. 2012. Využití organických markerů pro identifikaci zdrojů atmosférických aerosolů. Chemické listy, 106, 95-103.
402 403
Kupka K., 2013. QC Expert User’s Manual - TriloByte Statistical Software, [Czech Version]. Pardubice: TriloByte, Ltd. 266 pp.
404
Leal, R. R., Martinez, M. V., Campas, M. C. 2014. Morphology and Chemical Composition of PM10
405
by SEM-EDS. Microscopy and Microanylysis, 20 (S3), 1982-1983.
406
Liu, Y. et al. 2015. Particulate matter, gaseous and particulate polycyclic aromatic hydrocarbons
407
(PAHs) in an urban traffic tunnel of China: Emission from on-road vehicles and gas-particle
408
partitioning. Chemosphere, 134, 52-59.
409 410 411 412
Meloun, M., Militký, J. (2004). Statistická analýza experimentálních dat. 1st Ed. Praha: Academia. 953 pp. Turner, J. L. et al. 2009. Parallelization and Optimization of Genetic Analyses in Isolation by Distance Web Service. BMC Genetics, 10, (1): 28. doi:10.1186/1471-2156-10-28.
413
United Nation (UN). 2015. Agreement Concerning the Adoption of Uniform Technical Prescriptions
414
for Wheeled Vehicles, Equipment and Parts which Can be Fitted and/or be Used on Wheeled
415
Vehicles and the Conditions for Reciprocal Recognition of Approvals Granted on the Basis
416
of these Prescriptions. UN Regulation No. 83. Uniform Provisions Concerning the Approval
16
ACCEPTED MANUSCRIPT
417
of Vehicles with regard to the Emission of Pollutants according to Engine Fuel Requirements.
418
E/ECE/324/Revision 1/Addendum 82/Revision 5. Geneva: UN, 16-20.
419 420
Villar-Vidal, M. et al. 2014. Air Polycyclic Aromatic Hydrocarbons (PAHs) associated with PM2.5 in a North Cantabric coast urban environment. Chemosphere, 99, 233-238.
421
Winifred, U. A. et al. 2016. Study of Surface Morphology, Elemental Composition and Sources of
422
Airborne Fine Particulate Matter in Agbara Industrial Estate, Nigeria. International Journal of
423
Applied Environmental Sciences, 11, (4), 881-890.
424
Yang, H. H. et al. 2006. Source Identification and Size Distribution of Atmospheric Polycyclic
425
Aromatic Hydrocarbons during Rice Straw Burning Period. Atmospheric Environment, 40, (7),
426
1266-1274.
17
ACCEPTED MANUSCRIPT List of Figures – 1
Fig. 1. Slopes of lines Sf with confidence intervals at the level of significance = 0.05 obtained by Kendall–Theil regression. L1W = location 1, warmer period; L2W = location 2, warmer period; L1C = location 1, colder period; L2C = location 2, colder period; W = warmer period; C = colder period; SDC-cars = standard driving cycle, slope for all emission factors; SDC-SI = standard driving cycle, spark ignition; SDC-DF = standard driving cycle, diesel fuels; SDC-CS = standard driving cycle, cold starts.
ACCEPTED MANUSCRIPT List of Tables - 5 Table 1 Concentrations of BghiPe and BaP at locations L1 and L2, including weekly median maximum and minimum air temperatures.
Date
2007-04-16 2007-04-17 2007-04-18 2007-04-19 2007-04-20 2007-04-21 2007-04-22 2007-05-28 2007-05-29 2007-05-30 2007-05-31 2007-06-01 2007-06-02 2007-06-03 2007-07-09 2007-07-10 2007-07-11 2007-07-12 2007-07-13 2007-07-14 2007-07-15 2007-08-21 2007-08-22 2007-08-23 2007-08-24 2007-08-25 2007-08-26 2007-08-27 2007-10-01 2007-10-02 2007-10-03 2007-10-04 2007-10-05 2007-10-06 2007-10-07
Campaign no.
1
Median weekly air temperature tmedMax [°C]
17.09
tmed Min [°C]
3.17
2
19.56
10.60
3
24.33
11.30
4
28.16
16.30
5
17.67
9.17
Locality L1 Point no. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35
Locality L2
PAHs [ng m-3] BaP BghiPe 0.77 0.94 0.23 0.45 0.29 0.71 0.75 0.24 0.23 0.30 0.35 0.21 0.21 0.27 >LOD 0.25 0.28 0.34 0.27 >LOD >LOD >LOD 0.12 >LOD >LOD >LOD 0.17 >LOD 0.16 0.37 0.22 0.22 0.24 0.21 0.27
1.75 1.26 1.09 1.29 1.31 1.53 2.81 0.68 0.76 1.07 1.46 1.36 0.93 1.03 0.20 0.59 0.32 0.45 0.85 0.59 0.67 0.13 0.14 >LOD 0.28 0.11 0.16 0.11 0.08 0.73 0.90 0.86 0.54 0.99 1.16
Point no. 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91
PAHs [ng m-3] BaP BghiPe 1.62 0.20 0.36 0.80 0.59 0.77 0.52 >LOD >LOD 0.16 0.25 0.16 >LOD >LOD >LOD 0.18 0.20 0.28 0.21 0.16 0.15 0.18 >LOD >LOD >LOD >LOD >LOD >LOD >LOD 0.14 0.25 0.24 >LOD 0.37 0.29
2.67 1.24 0.68 1.75 1.00 1.71 1.51 0.77 0.21 0.69 0.57 0.66 0.64 0.29 0.18 0.22 0.26 0.65 0.36 0.40 0.51 0.17 0.24 0.07 >LOD >LOD >LOD 0.11 0.33 0.64 0.77 1.07 0.52 1.30 1.06
ACCEPTED MANUSCRIPT
Date
2007-11-19 2007-11-20 2007-11-21 2007-11-22 2007-11-23 2007-11-24 2007-11-25 2008-01-14 2008-01-15 2008-01-16 2008-01-17 2008-01-18 2008-01-19 2008-01-20 2008-02-25 2008-02-26 2008-02-27 2008-02-28 2008-02-29 2008-03-01 2008-03-02
Campaign no.
Median weekly air temperature tmedMax [°C]
tmed Min [°C]
6
5.96
1.46
7
7.20
2.43
8
12.19
1.04
LOD = limit of detection.
Locality L1 Point no. 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56
Locality L2
PAHs [ng m-3] BaP BghiPe 0.51 0.63 0.90 1.42 3.07 3.13 0.74 >LOD 0.21 0.23 0.63 0.83 0.14 >LOD 4.26 2.63 1.58 2.06 2.88 0.47 0.51
1.00 1.53 2.54 2.53 4.44 6.44 4.56 2.21 0.75 1.72 3.77 3.38 3.09 2.19 4.65 6.98 1.04 2.10 8.18 2.66 2.91
Point no. 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112
PAHs [ng m-3] BaP BghiPe 1.41 1.00 1.73 2.36 6.12 1.87 0.37 0.79 0.15 0.73 0.48 1.94 0.31 >LOD 3.82 2.67 1.21 2.14 1.24 >LOD 0.21
2.76 2.20 2.60 4.05 6.20 5.29 1.78 2.38 1.66 1.88 4.40 5.11 3.10 1.04 3.07 6.40 2.15 4.62 3.15 2.29 1.69
ACCEPTED MANUSCRIPT Table 2 Measured values of EfBghiPe and EfBaP in passenger cars. Date 2005-06-28 2005-06-28 2005-06-29 2005-06-29 2005-06-29 2005-10-11 2005-10-11 2005-10-12 2005-10-12 2005-11-23 2005-10-11 2005-11-23 2005-11-24 2006-03-08 2006-03-08 2006-03-08 2006-03-09 2006-03-09 2006-03-09 2006-05-31 2006-05-31 2006-05-31 31.5.2006 2006-06-01 2006-06-01
Vehicle SKODA Felicia 1.3/50 kW SKODA Felicia 1.3/50 kW SKODA Felicia 1.3/50 kW SKODA Octavia SDI/55 kW SKODA Octavia SDI/55 kW SKODA Fabia 1.4/44 kW SKODA Fabia 1.4/44 kW SKODA Fabia 1.4/44 kW SKODA Octavia SDI/55 kW SKODA Fabia 1.4/44 kW SKODA Octavia SDI/55 kW Ford Focus Flexifuel Ford Focus Flexifuel Ford Focus Flexifuel Ford Focus Flexifuel SKODA Fabia 1.4/44 kW SKODA Octavia SDI/55 kW SKODA Octavia SDI/55 kW SKODA Octavia SDI/55 kW SKODA Fabia 1.4/44 kW SKODA Fabia 1.4/44 kW SKODA Fabia 1.4/44 kW SKODA Fabia 1.4/44 kW SKODA Octavia SDI/55 kW SKODA Octavia SDI/55 kW
Fuel Natural 95 Natural 91 with 5% EtOH Natural 91 with 15% ETBE DF summer DF summer with 5% FAME Natural 95 Natural 91 with 5% EtOH Natural 91 with 15% ETBE DF summer with 31% FAME Natural 95 DF, summer Natural 95 with 85% EtOH Natural 95 with 85% EtOH Natural 95 with 85% EtOH Natural 95 with 85% EtOH Natural 95 DF summer DF summer with 5% FAME DF summer with 31% FAME Natural 95 Natural 95 Natural 91 with 5% EtOH Natural 91 with 15% ETBE DF summer with 5% FAME DF summer with 31% FAME
Driving mode
EfBaP
[µg km-1]
EfBghiPe
[µg km-1]
SDC SDC
0.0299 0.0129
0.0399 0.0326
SDC
0.0159
0.0295
SDC
0.0475
0.1084
SDC
0.0378
0.0962
SDC SDC
0.0186 0.0071
0.0202 0.0103
SDC
0.0134
0.019
SDC
0.0487
0.0719
SDC-CS SDC-CS
4.6033 0.134
4.9951 0.1584
SDC
0.0074
0.0119
SDC-CS
4.6914
7.9835
SDC-CS
0.3398
0.6441
SDC
0.0033
0.0038
SDC-CS SDC-CS
0.2581 1.7053
0.8296 2.1926
SDC
0.0597
0.134
SDC
0.0585
0.1035
SDC-CS SDC SDC
0.0889 0.0069 0.006
0.2296 0.0153 0.0112
SDC
0.0047
0.0079
SDC
0.0305
0.056
SDC
0.0256
0.0633
Natural 95, Natural 91 = additive-free petrols; EtOH = ethanol; ETBE = ethyl-tert-butyl ether; DF = diesel fuel; FAME = fatty acid methyl ester from rapeseed oil.
ACCEPTED MANUSCRIPT Table 3 Summary of regression model testing results. Parameter
L1W
L2W
L1C
L2C
L1
L2
W
C
L1L2WC
Absolute term of regression: b0
0.314 0.283
1.890
2.157
0.858
0.969
0.343
2.031
0.793
Slope of regression lines: b1
1.923 1.614
0.921
0.763
1.315
1.154
1.633
0.833
1.326
Number of points in regression: n Number of regression parameters: Q Degrees of freedom: v Residual variance: 2
25
21
18
17
43
38
46
35
81
2
2
2
2
2
2
2
2
2
23
19
16
15
41
36
44
33
79
0.120 0.0718
Residuals sum of squares: RSC 2.749 1.365 Weights for calculating common 47.90 27.28 section: w0 Weights for calculating common 2.736 5.361 slope: w1
2.022
0.974
1.150
0.936
32.36
14.61
47.14
33.69
9.96
9.62
26.49
23.61
56.74
65.65
59.48
71.01
0.0985 4.336
1.435
1.112
47.34
122.30
17.21
16.15
67.85
8.10
122.39
130.49
ACCEPTED MANUSCRIPT Table 4 Summaries of results of Bartlett’s test for mutual comparisons of variances and of F-test for mutual comparisons of regression lines. L1W L2W
Compared lines
L1C L2C
Number of points on compared lines: n
81
Number of regression lines compared: M Total degrees of freedom: V
L1W
L2W
L1W
L1C
L1
W
L1C
L2C
L2W
L2C
L2
C
43
𝑘
Bartlett’s criterion: B Critical value of Bartlett’s criterion: (𝑀 ‒ 1) Line match testing characteristic: R Critical test value: R < F0.95(M − 1, n − 2 M)
2 𝜒0.95
46
35
81
81
4
2
2
2
2
2
2
73
39
34
42
31
77
77
1.030
1.024
1.032
1.013
1.013
Parameter: D 1.024 1.027 Combined residual sum of squares of the 51.082 35.105 compared lines: RSCc Residual sum of squares of all groups of 122.296 47.144 compared data: RSCk 2 0.700 0.900 Combined estimate of variance: 𝜎 𝑐 Combined estimate of absolute regression 0.793 0.858 𝑘 members: 𝑏0 Combined estimate of regression slopes: 𝑏1
38
15.977
4.114 46.969
33.690
4.336 47.340 122.296 122.296
0.470
0.0979 1.515
80.833 51.676
1.050
0.671
0.969
0.343
2.031
0.793
0.793
1.154
1.633
0.833
1.326
1.326
1.326
1.315
60.53*
32.62*
24.03*
1.282
1.943
0.399
58.56*
9.49
5.99
5.99
5.99
5.99
5.99
5.99
5.654*
1.672
4.712* 0.283
0.0306
4.937* 13.153*
2.226
3.238
3.276
3.305
3.115
3.220
3.115
ACCEPTED MANUSCRIPT Table 5 Comparison of model line slopes (1) calculated by various statistical methods. Dataset-Statistical method
b1 value
b1 lower bound
b1 upper bound
b0 value
b0 lower bound
b0 upper bound
W-OR
2.348
1.419
2.996
0.145
0.002
0.318
W-RMA JK
2.045
1.747
2.902
0.222
0.063
0.306
W-RMA L
2.129
1.834
2.424
0.210
0.101
0.319
C-OR
1.441
0.598
2.068
1.247
0.531
2.184
C-RMA JK
1.271
0.962
1.909
1.488
0.846
1.890
C-RMA L
1.323
1.008
1.639
1.438
1.008
1.639
SDC-OR
1.927
1.620
2.238
0.000
SDC-RMA JK
1.961
1.650
2.430
-0.004
-0.015
-0.001
SDC-RMA L
1.961
1.938
1.984
-0.004
-0.011
0.003
SDC-CS-OR
1.451
1.162
1.786
0.000
SDC-CS-RMA L
1.416
1.021
1.812
0.053
-0.902
1.008
SDC-CS-RMA JK
1.507
1.011
1.727
-0.035
-0.266
0.382
W = warmer period; C = colder period; OR = orthogonal regression; RMA JK = reduced major axis jackknife method; RMA L = reduced major axis linear model.
ACCEPTED MANUSCRIPT
Table 6 Summary of F-test results for mutual comparison of slopes. L1W L2W
Slopes compared
L1C L2C 𝑐
L1W
L2W
L1W
L1C
L1
W
L1C
L2C
L2W
L2C
L2
C
Combined estimate of regression slopes: 𝑏1
0.891
0.968 0.827
1.718 0.836 1.227 0.883
Testing characteristic of line slopes: G
3.261*
2.909 7.637* 1.770 0.506 0.801 2.285
Critical test value: G < F0.95(M − 1, n − 2 M)
2.730
4.091 4.130
4.073 4.160 3.965 3.965
Asterisks indicate groups of regression lines for which the hypothesis on matching of slopes was rejected.
Jiri Huzlik, Frantisek Bozek, Adam Pawelczyk, Roman Licbinsky, Magdalena Naplavova, Michael Pondelicek PII:
S0045-6535(17)30673-2
DOI:
10.1016/j.chemosphere.2017.04.131
Reference:
CHEM 19193
To appear in:
Chemosphere
Received Date:
10 February 2017
Revised Date:
17 March 2017
Accepted Date:
26 April 2017
Please cite this article as: Jiri Huzlik, Frantisek Bozek, Adam Pawelczyk, Roman Licbinsky, Magdalena Naplavova, Michael Pondelicek, Identifying risk sources of air contamination by polycyclic aromatic hydrocarbons, Chemosphere (2017), doi: 10.1016/j.chemosphere.2017.04.131
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT
Identifying risk sources of air contamination by polycyclic aromatic hydrocarbons
1 2 3 4
Jiri Huzlik a, Frantisek Bozek b*, Adam Pawelczyk c, Roman Licbinsky d, Magdalena Naplavova e
5
Michael Pondelicek f
6 7
a
Transport Research Centre, 33a Lisenska, 636 00 Brno, Czech Republic, telephone: +420 541 641 374, e-mail: [email protected]
8 9
b
University of Defence, [email protected]
10 11
c
Wroclaw University of Technology, Faculty of Chemistry, 4/6 Norwida St., 50-373 Wrocław, Poland, e-mail: [email protected]
12 13
d
Transport Research Centre, 33a Lisenska, 636 00 Brno, Czech Republic, e-mail: [email protected]
14 15
e
University of Defence, 65 [email protected]
16 17
f
The College of Regional Development, 68 Zalanskeho, 163 00 Praha 17 – Repy, Czech Republic, e-mail: [email protected]
18
*
corresponding author
65
Kounicova,
Kounicova,
662
662
10
10
Brno,
Brno,
Czech
Czech
Republic,
Republic,
e-mail:
e-mail:
19 20
Highlights
21
PAH concentrations were measured in air and exhaust gases.
22
Marker BghiPe/BaP was used to identify air pollution sources in relation to time.
23
Various statistical methods were used precisely to evaluate the BghiPe/BaP ratio.
24
In warmer periods, transport is exclusively the source of PAH air pollution.
25
In colder periods, local heating stoves contribute to the presence of PAHs in air.
26
Abstract
27
This article is directed to determining concentrations of polycyclic aromatic hydrocarbons (PAHs),
28
which are sorbed to solid particles in the air. Pollution sources were identified on the basis of the
29
ratio of benzo[ghi]perylene (BghiPe) to benzo[a]pyrene (BaP). Because various important
30
information is lost by determining the simple ratio of concentrations, least squares linear regression
31
(classic ordinary least squares regression), reduced major axis, orthogonal regression, and Kendall–
1
ACCEPTED MANUSCRIPT
32
Theil robust diagnostics were utilized for identification. Statistical evaluation using all
33
aforementioned methods demonstrated different ratios of the monitored PAHs in the intervals
34
examined during warmer and colder periods. Analogous outputs were provided by comparing
35
gradients of the emission factors acquired from the measured concentrations of BghiPe and BaP in
36
motor vehicle exhaust gases. Based on these outputs, it was possible plausibly to state that the
37
influence of burning organic fuels in heating stoves is prevalent in colder periods whereas in warmer
38
periods transport was the exclusive source because other sources of PAH emissions were not found
39
in the examined locations.
40
Keywords:
41
air pollution, benzo[a]pyrene, benzo[ghi]perylene, pollution sources, polycyclic aromatic
42
hydrocarbons, regression, transport.
43
1
44
Effectively executed protection of a population consists in the first phase of identifying the sources
45
of hazards followed by qualitative and semi-quantitative estimation or quantitative calculation of
46
those risks resulting from the identified sources of hazards, delimiting critical risks, and finally
47
proposing measures for their mitigation (Božek and Urban, 2008). Serious sources of hazards
48
currently include air pollution, to which contamination by natural sources (forest fires, volcanic
49
eruptions) contribute in part but which in recent decades is mainly due to anthropogenic sources
50
(Villar-Vidal et al., 2016). Important anthropogenic sources of pollution include in particular
51
industry, energy, transport, agricultural production, communal waste incinerators, oil spills, but also
52
to a considerable degree domestic heating stoves. There are, therefore, a number of pollutants in the
53
air which present a considerable health risk to human populations and a threat to ecosystems,
54
especially in the vicinity of busy roads and cities with high concentrations of people and industry
55
(Bozek et al., 2011; Liu et al., 2015).
56
A priority condition for successfully minimizing risks from contaminated air in heavily polluted
57
locations is thus identification of the relevant sources of such pollution. This can be carried out in
58
various ways.
Introduction
2
ACCEPTED MANUSCRIPT
59
2
60
Basic approaches to deciding about origins of emission sources essentially comprise the following:
61
a)
62 63
Analysis of the current state
morphological and dimensional determination of a complex mixture of captured solid particles (Goldstein, 1992; Leal et al., 2014),
b)
64
examination of the physical properties and chemical composition of solid particles (Leal et al., 2014; Winifred et al., 2016),
65
c)
receptor modelling (Dutton, 2010),
66
d)
measurement of characteristic markers consisting in the chemical composition of selected
67
contaminants bound to dust particles.
68
Use of the ratios of various polycyclic aromatic hydrocarbons (PAHs) is the most commonly
69
encountered method in identifying emission sources on the basis of aerosol chemical composition.
70
That is because PAHs are omnipresent, form a broad group of compounds, and many of them
71
demonstrate toxic, mutagenic, teratogenic, carcinogenic, and embryotoxic properties (Haritash and
72
Kaushik, 2009). For the given purpose, benzo[a]pyrene (BaP) seems to be an especially suitable
73
PAH due to its being stable while also demonstrating a relatively consistent and high contribution to
74
PAHs’ carcinogenic activity (Hailwood et al., 2001). The ratios of its concentration to
75
benzo[ghi]perylene (BghiPe) (Gustafsson and Gschwend, 1997) and to benzo[e]pyrene (Yang, 2006)
76
are frequently used as markers for identifying sources of pollution. The ratios of
77
indeno[1,2,3-cd]pyrene (IPy) to the sum of concentrations of IPy and BghiPe (Yang, 2006),
78
fluoranthene to pyrene, and phenanthrene to anthracene (Kocourek et al., 2003) are also applied.
79
Broad overviews of organic markers, including PAHs, which may be used for identifying sources of
80
atmospheric aerosols were completed by Genualdi (2008) and Křůmal et al. (2012).
81
Mere monitoring of pollutant concentrations ratios in the atmosphere can, however, lead to losing a
82
good deal of valuable information in relation to specifying pollution sources. For example,
83
Chuesaard et al. (2014) demonstrated through a more detailed examination fluctuations in the
84
BghiPe/BaP ratio within the interval 1.07 to 3.29 as a function of air humidity. They explained
85
decrease in the BghiPe/BaP ratio in dry periods by a contribution of emissions from biomass
86
combustion. As an indicator for assessing the contribution of biomass combustion they proposed the
87
ratio of 9-nitroanthracene and 1-nitropyrene. In parallel, dependencies of organic markers on
3
ACCEPTED MANUSCRIPT
88
variables other than air humidity can be found in the literature as can discrepancies in presenting
89
their values for a given pollution source (Genualdi, 2008).
90
The application of statistical methods in processing measured data is one possibility for resolving
91
existing ambiguities, expanding knowledge in this area, and increasing the precision of markers of
92
specific sources of air contamination under various conditions. In the present article, some statistical
93
procedures which can be applied effectively also for more exact characterization of other markers of
94
this type are presented for evaluating relationships between BaP and BghiPe concentrations in the
95
vicinity of busy urban roads.
96
3
97
BaP and BghiPe concentrations were measured in the air of two urban locations. BaP and BghiPe
98
emission factors were determined in exhaust gases of four passenger motor vehicles. The acquired
99
results became the basis for identifying air pollution sources using selected statistical methods.
Methods and devices applied
100
3.1
101
Pollution sources were identified by comparing the regression lines of BghiPe versus BaP.
102
According to our previous findings, using only the ratio of their concentrations leads to a loss of
103
certain important information. We used the linear regression module of the statistical program
104
package QC.Expert to evaluate the results of classic ordinary least squares (OLS) regressions
105
(Kupka, 2013). The fact that BaP concentrations display a smaller error of determination than do
106
BghiPe concentrations was an argument for selecting the model in the form of Equation (1), where
107
cBghiPe and cBaP represent the concentrations of BghiPe and BaP, respectively, and b1 and b0 represent
108
the coefficients of the linear term and absolute term, respectively.
109
Statistical methods
c BghiPe b0 b1 c BaP
(1)
110
The model’s reliability and regression parameter estimates were verified using the multiple
111
correlation coefficient R, coefficient of determination R2, predicted correlation coefficient Rp, mean
112
squared prediction error MSPE, and the Akaike information criterion. The legitimacy of using the
113
linear model was tested using the Fisher–Snedecor significance test, Scott multicollinearity criterion,
114
Cook–Weisberg heteroscedasticity test, Jarque–Bera normality test, Wald and Durbin–Watson
115
autocorrelation tests, and the residual sign test (Meloun and Militký, 2004). 4
ACCEPTED MANUSCRIPT
116
The robust method by Kendall and Theil, which is not sensitive to outlying values, was subsequently
117
applied. The Kendall–Theil Robust Line software is a Visual Basic program that may be used with
118
the Microsoft Windows operating system to calculate parameters for robust, nonparametric estimates
119
of linear regression coefficients between two continuous variables (Granato, 2006). The slope of the
120
line was calculated as the median of all possible pairwise slopes between points. The intercept was
121
calculated so that the line runs through the median of input data. The confidence interval of the slope
122
was calculated, as well.
123
Because it could not be expected for the OLS regression to display an error close to zero, the reduced
124
major axis (RMA) method was used to verify the results’ reliability (Turner et al., 2009; Friedman et
125
al., 2013) using RMA software (Bohonak, 2005). The method consists in concurrent minimization of
126
both the y and x axes residuals. The classical linear model (RMA L) and a model based on the
127
jackknife method (RMA JK) were applied in the study. In parallel, the orthogonal regression (OR)
128
method was included in the program Number Cruncher Statistical System (NCSS) based on
129
minimization of the perpendicular distance of regression points from the regression line (Hintze,
130
2007).
131
3.2
132
Air samples were collected over a period of 24 h using a medium-volume Leckel MVS6 sampler
133
and, similarly, the procedure presented in our previous publication (Bozek et al., 2016) was used to
134
determine the PAHs.
135
To simulate urban traffic we selected a standard test according to the Economic Commission for
136
Europe (ECE), Urban (Part One) Driving Cycle (ECE 15), Type I test (UN, 20015). It consisted of
137
four elementary urban cycles, included 15 phases, and lasted for 195 seconds, as shown in Fig. 1.
138
The measurements were taken with the engine heated to operating temperature and were repeated in
139
three immediate successions. The entire process during which the exhaust gases were collected to
140
determine PAHs is subsequently referred to as the standard driving cycle (SDC). To simulate cold
141
starts, only one test was performed under the same conditions, subsequently referred to as SDC-CS.
142
Collection was performed using an apparatus connected directly to the vehicle’s exhaust pipe, as
143
described by Huzlik et al. (2011).
144
Sample collection and analysis were performed according to standards COSMT (2006) and COSMT
145
(2007) while accepting the fact that measurements in a car’s exhaust pipe cannot be performed
Measurement of emissions
5
ACCEPTED MANUSCRIPT
146
isokinetically. Measurements on the TUV-ÚVMV apparatus were performed on a roller
147
dynamometer (SCHENCK 364/GS56), which simulated inertial masses and driving resistance as if
148
the vehicle was moving on a road.
149
4
150
With the objective of identifying sources of air pollution using PAHs, we evaluated concentrations of
151
BghiPe and BaP in the air. In order to compare the results obtained in this way, we used ratios of
152
measured BghiPe and BaP emission factors for passenger motor vehicles.
153
4.1
154
Results and discussion
Determination of benzo[a]pyrene and benzo[ghi]perylene concentrations in the atmosphere
155
Concentrations of PAHs bound to PM2.5 solid particles in the air were monitored in the period from
156
16 March 2007 to 2 March 2008 during eight weekly collection campaigns at locations L1 and L2 in
157
Brno, Czech Republic. Traffic intensity was TI1 = 3.6 104 vehicles per day at location L1 and
158
TI2 = 8 103 vehicles per day at location L2. To specify the pollution sources, we selected BghiPe
159
and BaP concentrations because their measured concentrations ranged at relatively high levels. At
160
each location, a total of 56 pairs of concentration values for these substances were determined. These
161
are presented in Table 1 together with the weekly median maximum tmedMax and minimum tmedMin
162
air temperatures.
163
Medians were used in order to eliminate the impact of the obliqueness of the statistical distribution
164
of data during the weekly sampling campaigns.
165
4.2
166
Determination of benzo[a]pyrene and benzo[ghi]perylene emission factors in exhaust gases
167
Emission factors EfBghiPe and EfBaP were measured in the period from June 2005 to July 2006 in the
168
driving modes “urban cycle” (SDC) and “cold start” (SDC-CS). Additive-free petrol and petrol with
169
the addition of ethanol and ethyl-tert-butyl ether as well as summer diesel fuel with and without fatty
170
acid methyl ester (FAME) from rapeseed oil as additive were used. The results are presented in
171
Table 2.
6
ACCEPTED MANUSCRIPT
172
4.3
Identification of air pollution sources using linear regression
173
In the first phase, we performed calculations of linear regression between BghiPe and BaP using the
174
OLS method and determined parameters b0 and b1 separately for each entire location. The
175
calculations did not include cases where at least one of the measured concentrations was under the
176
detection limit of LOD 0.05 ng m−3. However, the data exhibited a heteroscedasticity which could
177
not be removed even by omitting outlying data or introducing weighted regression. The residuals
178
analysis indicated that both systems act differently in the warmer period of the year W, when tmedMin
179
5°C (mid-April to early October), and in the colder period C, when tmedMin 5°C (mid-November
180
to early March). The data from each location was therefore divided into two groups according to
181
tmedMin and regression was calculated separately. The predicted residuals were predominantly
182
negative in the warmer period for both locations but predominantly positive in the colder period.
183
Regression diagnostics between BghiPe and BaP concentrations for the data set from the warmer
184
period at location L1 demonstrated that the residuals exhibit heteroscedasticity (p = 1.91 10−2).
185
Measured concentrations at points 2 and 7 stated in Table 1 are probably outliers. They were
186
therefore excluded from further data analysis and the calculations were made again. The results are
187
presented in Table 3 in the column marked L1W. This procedure eliminated the heteroscedasticity
188
from the data. Other regression statistics were satisfactory, with the exception of the data’s positive
189
autocorrelation.
190
Calculations for location L1 and the colder period were performed similarly. Based on diagnostic
191
graphs, point 54 was excluded as an outlier (see Table 1) and the calculations were repeated to
192
eliminate the corresponding concentrations. The results thus obtained are reported in Table 3 in the
193
column marked L1C. Following this correction, all tests fulfilled the regression criteria.
194
After excluding outlying points at location L2, regression was tested identically as for location L1.
195
The results indicated exclusion of the point denoted as 58 in Table 1 for the warmer period and
196
points 106 and 107 for the colder period. The results obtained after eliminating outlying
197
concentrations fulfilled the regression criteria for all tests and are shown in Table 3 in column L2W
198
for the warmer period and L2C for the colder period.
199
In order to compare all four regression lines, common regression of the sum of all tested sets was
200
calculated. The results are presented in Table 3 under column L1L2WC. In this case, however, the
7
ACCEPTED MANUSCRIPT
201
data exhibited heteroscedasticity (p = 2.8 10−5), the residuals did not have normal distribution
202
(p = 5.73 10−5), and the sign test demonstrated an increasing trend (p = 2.59 10−4).
203
Heteroscedasticity in the results remained even after eliminating outlying points.
204
Linear model parameters were subsequently calculated. The values for the warmer and colder
205
periods are presented in Table 3 under columns W and C, respectively. Only a significant positive
206
autocorrelation was demonstrated in the data for the warmer period, whereas for the colder period all
207
tests fulfilled the regression criteria.
208
In order to judge whether pollution came from the same source, it was necessary to verify the match
209
of the regression lines, the number of which was represented for the individual comparisons by the
210
symbol M. For this purpose it was first necessary to verify whether residual variances for all
211
regression lines were identical by performing Bartlett’s test of heteroscedasticity. The test was
212
processed for M independent estimates of variances with j = nj-M degrees of freedom, where nj is
213
the number of elements of the j-th set in the respective column, as is apparent from Tables 3 and 4.
214
The null hypothesis was then defined as H 0 : 2 2j ; j 1, M j , M N , all residual variances
215
being
216
H 1 : 2 2j ; j 1, M j , M N , stating that at least one of the variances is different from the
217
others. As testing criterion B we used Equation (2), in which the total degrees of freedom V is
218
calculated according to Equation (3), the summarized estimate of variance C2 according to Equation
219
(4), and parameter D according to Equation (5):
220
equal
indicating
B
its
validity,
the
alternative
hypothesis
as
M V ln { C2 } ( j ln { C2 }) j 1
(2)
D V
221
M
j 1
222
and
C2
M
( j 1
j
j
2j V 1 )
(3)
(4)
8
ACCEPTED MANUSCRIPT
M
D 1
223 224
( j 1
1 j
V 1 ) (5)
3 ( M 1)
Values of n are presented in Table 3 and were calculated using Equation (6), where j N.
n
225
M
n j !
(6)
j
226
Given the validity of hypothesis H0, the distribution of B is asymptotically Pearson’s 2 with M − 1
227
degrees of freedom and H0 is considered to be accepted at the level of significance if
228
B < 12- M 1 . The estimate of variance 2 is the combined estimate of variance 𝜎2𝑐 . Bartlett’s test
229
is sensitive to deviations of the residuals from normality. Groups of compared regression lines for
230
which the assumption that they have a common combined variance 𝜎2𝑐 was rejected are marked with
231
an asterisk in Table 4.
232
The first part of comparing the matching of the lines was to test for homogeneity of intersects
233
j k (intercepts). The null hypothesis H 0 :b0 b0 ; j , k 1, M j , k , M N was formulated, stating that
234
j k intersects are identical, and the alternative hypothesis H 1 : b0 b0 ; j , k 1, M j , k , M N ,
235
j k stating that at least one of the intersects differs from the others. Symbols b0 , b0 represent absolute
236
members of the j-th and k-th regression lines, respectively. The testing equation E of line intersects
237
j was expressed as Equation (7), where w0 denotes the weight of absolute members of the intersect of
238
c the j-th regression line calculated according to Equation (8), b0 is the combined estimate of absolute
239
members of the compared regression lines calculated using Equation (9), and RSCj is the residual
240
sum of squares of the j-th line. M
241
E
(n 2 M ) w0j (b0j b0c ) 2 j 1
M
( M 1) RSC j
(7)
j 1
9
ACCEPTED MANUSCRIPT
nj
w0j
242
n j ( xi , j x j ) 2 i 1 nj
(x
i, j
i 1
M
b0c
243
(8)
(w
j 0
j 1
nj
)
2
b0j )
w j 1
(9)
j 0
244
If the null hypothesis H0 is valid, then the testing statistic E has a Fischer–Snedecor distribution with
245
1 = M − 1 and 2 = n – 2 M degrees of freedom. If E < F1 ( M 1, n 2 M ) , then it can be
246
c stated that all lines exhibit the same intersect with the value of the absolute term b0 at the level of
247
significance .
248
The test of homogeneity of intersects with acceptance of Bartlett’s test of homoscedasticity
249
demonstrated a match of regression lines intersects of the functions BghiPe and BaP for the warmer
250
(L1W-L2W) and colder (L1C-L2C) periods at the level of significance = 0.05, including summary
251
data for the examined period at both monitored locations (L1-L2).
252
The matching of lines was then compared using a test of the homogeneity of slopes. The null
253
j k hypothesis H 0 : b1 b1 ; j , k 1, M j , k , M N was defined, stating that slopes of regression
254
lines are identical, and the alternative hypothesis H 1 : b1j b1k ; j , k 1, M j , k , M N , stating
255
that at least one slope is different from the others. The symbols b1j , b1k denote the slopes of the j-th
256
and k-th regression lines, respectively. As testing criterion G we used Equation (10), where w1j is the
257
weight of the slope of the j-th regression line calculated according to Equation (11), b1c is the
258
combined estimate of slopes calculated as a weighted combination of the estimates of individual
259
slopes b1j using Equation (12), and the other symbols have the same meanings as in Equation (7).
10
ACCEPTED MANUSCRIPT
M
G
260
(n 2 M ) w1j (b1j b1c ) 2 j 1
M
( M 1) RSC j
(10)
j 1
w j 1
261
nj
(x i 1
i, j
M
b1c
262
(w
j 1
j 1
nj
x j )2
b1j )
w j 1
(11)
(12)
j 0
263
Assuming the validity of hypothesis H0, the testing characteristic G has a Fischer–Snedecor
264
distribution with 1 = M − 1 and 2 = n − 2 M degrees of freedom. If G < F1 ( M 1, n 2 M ) , all
265
regression lines have an identical slope b1c with an estimate defined by Equation (12) at the level of
266
significance . Hypothesis H0 on matching slopes for combinations of lines L1W-L2W-L1C-L2C and
267
L2W-L2C was rejected.
268
All lines of BghiPe and BaP dependency that were found to be identical by the test of intersects of
269
the regression model (1) can also be considered identical from the aspect of slope equality. Although
270
the data from Table 6 seem to suggest that slopes of lines for the warmer and colder periods at
271
location L1 and lines of all data for the warmer and colder periods could be considered identical, this
272
cannot be regarded as conclusive due to the heteroscedasticity of the residuals, as evidenced in Table
273
4.
274
The final step in comparing the matching of lines was the general matching test, which is a
275
combination of homogeneity tests for intersects and slopes. This consists in comparing the residual
276
sum of squares RSCk obtained after interlaying all M data groups with one common line with
277
k estimates of parameters b0 , b1k and the sum of residuals calculated separately for each group. The
278
null hypothesis H0 was defined, as expressed by Equation (13) and stating that all lines are identical,
279
and the alternative hypothesis H1, as expressed by Equation (14), stating that at least one line is
280
different from the others.
11
ACCEPTED MANUSCRIPT
281
H 0 : b0j b0k b1j b1k j , k 1; M j , k , M N
(13)
282
H 1 : b0j b0k b1j b1k b0j b0k b1j b1k b0j b0k b1j b1k j , k 1; M j , k , M N
(14)
M
283
The testing criterion R was calculated according to Equation (15), where RSCc RSC j : j 1
R
284
RSCk RSCc
(n 2 M ) 2 RSCc ( M 1)
(15)
285
If H0 is valid, then the testing criterion R demonstrates a Fischer–Snedecor distribution with
286
1 = 2 (M − 1) and 2 = n − 2 M degrees of freedom. If R < F1 [2 ( M 1), n 2 M ] , then it
287
can be stated that at the level of significance all regression lines are identical with the common
288
k estimate of intersect b0 and slope b1k . The results of the calculations are presented in Table 4.
289
Asterisks mark combinations of regression lines for which hypothesis H0 in regard to their matching
290
was rejected.
291
The overall match test and the acceptance of results of Bartlett’s test of homoscedasticity indicate
292
that only pairs of lines for warmer (L1W-L2W) and colder (L1C-L2C) periods at both examined
293
locations can be considered identical. For other groups of lines, it is not possible to consider the
294
results to be conclusive.
295
4.4
296
The results of classic OLS linear regression were verified using the robust method according to
297
Kendall and Theil without eliminating outlying points. The ratios of the measured emission factors
298
were assessed using the same method. The results of the Kendall–Theil regression are presented
299
graphically in Fig. 1, which displays the 95% confidence intervals for the slopes. Confidence
300
intervals in the warmer and colder periods do not overlap and therefore differ significantly. Robust
301
diagnostics therefore provide analogous results for the values of regression parameters as does
302
classical least squares regression.
303
The analysis of regression line slopes for the motor vehicle emission factors and for the measured
304
PAHs concentrations in the air of the examined locations clearly indicates higher content of BghiPe
Robust diagnostics
12
ACCEPTED MANUSCRIPT
305
as compared to BaP in the warmer period. It can therefore be logically deduced that transport is the
306
predominant source of air pollution in the summer period. Certain differences in the slope values
307
calculated from the concentrations of BghiPe and BaP in the air and in exhaust gases can be
308
explained by the disparate composition of commercially available fuels, especially in relation to
309
additive contents. In contrast, the concentration of BaP is dominant in the colder period. Given that
310
no sources of PAHs other than transport and local heating stoves were identified in the vicinity of
311
locations L1 and L2, and because BaP is dominant in relation to BghiPe for fossil fuel combustion
312
(Křůmal et al., 2012), it is reasonable to deduce that in the winter period local heating stoves
313
contribute to air pollution in addition to transportation.
314 315
As with the Kendal–Theil regression, the RMA and orthogonal regression (OR) methods also
316
confirmed the difference in sources of PAHs in the warmer and colder periods, as is evident from
317
Table 5. It may also be stated that the slope values for emission factors in cold starts are lower than
318
those in standard cycles, which signals an increased proportion of BaP. It can therefore be reasonably
319
presumed that in the colder period the increased content of BaP versus BghiPe is caused not only by
320
the combination of sources of local combustion and operation of motor vehicles under colder
321
temperatures but apparently also by the higher number of cold starts.
322
5
323
Linear regression using the method of least squares demonstrated a statistically significant match of
324
slopes of dependencies of concentrations of BghiPe on BaP in the warmer period (campaigns 1–5)
325
for both monitored locations L1 and L2 as well as for the colder period (campaigns 6–8). The match
326
of summary regression lines for the entire year for locations L1 and L2 was not statistically
327
significant. Comparing other combinations of locations and periods using this method was shown to
328
be incorrect due to breaching the assumption of homoscedasticity of residuals. The RMA and OR
329
methods as well as the Kendal–Theil regression, which are not sensitive to outlying values,
330
confirmed the difference in dependencies of concentrations of BghiPe on BaP in the warmer and
331
colder periods. Comparing confidence intervals of regression slopes using robust methods confirmed
332
the statistical significance of the difference between slopes of the regression summary line for the
333
warmer period with the corresponding slope for the colder period.
Conclusions
13
ACCEPTED MANUSCRIPT
334
This means that the ratio of BghiPe/BaP concentrations characterized by the slope of the regression
335
line is not different for the measured locations but it is different for the different periods of the year.
336
The comparison of slopes of BghiPe and BaP emission factors obtained by measuring the
337
concentrations of both PAHs in exhaust gases of motor vehicles led to the same results. For cold
338
starts, as opposed to SDC, a higher ratio of BaP was determined relative to BghiPe in exhaust gases.
339
In view of the fact that no sources of PAHs other than transportation and local heating stoves were
340
found in the monitored locations, we may conclude that in the warmer period the only source of
341
PAHs is transportation. In the colder period, fossil fuel combustion in combination with the
342
operation of motor vehicles at lower temperatures also contributes to the content of PAHs in the air.
343
6
344
This article was produced with the financial support of the Ministry of Education, Youth and Sports
345
within the programme of long-term conceptual development of research institutions on the research
346
infrastructure and within the National Sustainability Programme I, project of the Transport R&D
347
Centre (LO1610), on the research infrastructure acquired from the Operation Programme Research
348
and Development for Innovations (CZ.1.05/2.1.00/03.0064).
349
References
350
Bohonak, A. J. 2004. RMA: Software for Reduced Major Axis Regression. Verion 1.17. [online].
351 352 353 354 355 356 357
Acknowledgement
[2016-07-29]. URL:
358
Chuesaard, T. et al. 2014, A. Influence of Biomass Burning on the Levels of Atmospheric Polycyclic
359
Aromatic Hydrocarbons and Their Nitro Derivatives in Chiang Mai, Thailand. Aerosol and Air
360
Quality Research, 14, (4), 1247-1257.
14
ACCEPTED MANUSCRIPT
361
Czech Office for Standards, Metrology and Testing (COSMT). 2006. ČSN EN 1948-1. Stationary
362
Source Emissions - Determination of Mass Concentration of PCDDs/PCDFs and Dioxin-like
363
PCBs - Part 1: Sampling of PCDDs/PCDFs. Prague: COSMT.
364
Czech Office for Standards, Metrology and Testing (COSMT). 2007. ČSN EN 1948-3. Stationary
365
Source Emissions - Determination of Mass Concentration of PCDDs/PCDFs and Dioxin-Like
366
PCBs - Part 3: Identification and Quantification of PCDDs/PCDFs. Prague: COSMT.
367
Dutton, S. J. et al. 2010. Source Apportionment Using Positive Matrix Factorization on Daily
368
Measurements of Inorganic and Organic Speciated PM2.5. Atmospheric Environment, 44, (23),
369
2731-2741.
370 371 372 373
Friedman, J., Bohonak, A. J., Levine, R. A. 2013. When are Two Pieces Better than One: Fitting and Testing OLS and RMA Regression. Environmetrics, 24, 306–31. Genualdi, S. 2008. Semi-Volatile Organic Compounds as Molecular Markers for Atmospheric and Ecosystem Transport. [Dissertation]. Corvallis: Oregon State University.
374
Goldstein, J. I. et al. 1992. Scanning Electron Microscopy and X-Ray Microanalysis. A Text for
375
Biologist, Material Scientist, and Geologist. 2nd Ed. New York, London: Plenum Press. 823 pp.
376
ISBN-13:978-1-4612-7653-1.
377
Granato, G. E. 2006. Kendall-Theil Robust Line, KTRLine, Version 1.0. A Visual Basic Program for
378
Calculating and Graphing Robust Nonparametric Estimates of Linear-Regression Coefficients
379
between Two Continuous Variables: Techniques and Methods of the U. S. Geological Survey,
380
Book
381
4,
Chap.
A7,
31
pp.
[online].
[2015-12-27].
URL:
382
Gustafsson, O., Gschwend, P. M. Soot as a Strong Partition Medium for Polycyclic Aromatic
383
Hydrocarbons in Aquatic Systems. In: Eganhouse, R. P. Molecular Markers in Environmental
384
Geochemistry,
385
eISBN 9780841216358.
Washington:
American
Chemical
Society,
Chapter
24,
365–381.
386
Hailwood, M . et al. 2001. Ambient Air Pollution by Polycyclic Aromatic Hydrocarbons. [Position
387
Paper]. Prepared by the Working Group On Polycyclic Aromatic Hydrocarbons.
388
ISBN 92-894-2057-X.
[online].
[2015-12-25].
URL:
15
ACCEPTED MANUSCRIPT
389
390
391 392 393 394 395 396
Haritash, A. K., Kaushik, C. P. 2009.
Biodegradation Aspects of Polycyclic Aromatic
Hydrocarbons: A review. Journal of Hazardous Materials, 169, 1-15. Hintze, J. 2007. NCSS Statistical Software. NCSS, LLC. Kaysville, Utah. [online]. [2011-07-31]. URL:
397
Kocourek, V., Hajšlová, J., Tomaniová, M., 2003. Přehled imisní zátěže agrárního ekosystému
398
vybranými prioritními organickými polutanty. Praha: Výzkumný ústav rostlinné výroby,
399
[online]. [2015-12-27]. URL:
400 401
Křůmal, K., Mikuška, P. Večeřa, Z. 2012. Využití organických markerů pro identifikaci zdrojů atmosférických aerosolů. Chemické listy, 106, 95-103.
402 403
Kupka K., 2013. QC Expert User’s Manual - TriloByte Statistical Software, [Czech Version]. Pardubice: TriloByte, Ltd. 266 pp.
404
Leal, R. R., Martinez, M. V., Campas, M. C. 2014. Morphology and Chemical Composition of PM10
405
by SEM-EDS. Microscopy and Microanylysis, 20 (S3), 1982-1983.
406
Liu, Y. et al. 2015. Particulate matter, gaseous and particulate polycyclic aromatic hydrocarbons
407
(PAHs) in an urban traffic tunnel of China: Emission from on-road vehicles and gas-particle
408
partitioning. Chemosphere, 134, 52-59.
409 410 411 412
Meloun, M., Militký, J. (2004). Statistická analýza experimentálních dat. 1st Ed. Praha: Academia. 953 pp. Turner, J. L. et al. 2009. Parallelization and Optimization of Genetic Analyses in Isolation by Distance Web Service. BMC Genetics, 10, (1): 28. doi:10.1186/1471-2156-10-28.
413
United Nation (UN). 2015. Agreement Concerning the Adoption of Uniform Technical Prescriptions
414
for Wheeled Vehicles, Equipment and Parts which Can be Fitted and/or be Used on Wheeled
415
Vehicles and the Conditions for Reciprocal Recognition of Approvals Granted on the Basis
416
of these Prescriptions. UN Regulation No. 83. Uniform Provisions Concerning the Approval
16
ACCEPTED MANUSCRIPT
417
of Vehicles with regard to the Emission of Pollutants according to Engine Fuel Requirements.
418
E/ECE/324/Revision 1/Addendum 82/Revision 5. Geneva: UN, 16-20.
419 420
Villar-Vidal, M. et al. 2014. Air Polycyclic Aromatic Hydrocarbons (PAHs) associated with PM2.5 in a North Cantabric coast urban environment. Chemosphere, 99, 233-238.
421
Winifred, U. A. et al. 2016. Study of Surface Morphology, Elemental Composition and Sources of
422
Airborne Fine Particulate Matter in Agbara Industrial Estate, Nigeria. International Journal of
423
Applied Environmental Sciences, 11, (4), 881-890.
424
Yang, H. H. et al. 2006. Source Identification and Size Distribution of Atmospheric Polycyclic
425
Aromatic Hydrocarbons during Rice Straw Burning Period. Atmospheric Environment, 40, (7),
426
1266-1274.
17
ACCEPTED MANUSCRIPT List of Figures – 1
Fig. 1. Slopes of lines Sf with confidence intervals at the level of significance = 0.05 obtained by Kendall–Theil regression. L1W = location 1, warmer period; L2W = location 2, warmer period; L1C = location 1, colder period; L2C = location 2, colder period; W = warmer period; C = colder period; SDC-cars = standard driving cycle, slope for all emission factors; SDC-SI = standard driving cycle, spark ignition; SDC-DF = standard driving cycle, diesel fuels; SDC-CS = standard driving cycle, cold starts.
ACCEPTED MANUSCRIPT List of Tables - 5 Table 1 Concentrations of BghiPe and BaP at locations L1 and L2, including weekly median maximum and minimum air temperatures.
Date
2007-04-16 2007-04-17 2007-04-18 2007-04-19 2007-04-20 2007-04-21 2007-04-22 2007-05-28 2007-05-29 2007-05-30 2007-05-31 2007-06-01 2007-06-02 2007-06-03 2007-07-09 2007-07-10 2007-07-11 2007-07-12 2007-07-13 2007-07-14 2007-07-15 2007-08-21 2007-08-22 2007-08-23 2007-08-24 2007-08-25 2007-08-26 2007-08-27 2007-10-01 2007-10-02 2007-10-03 2007-10-04 2007-10-05 2007-10-06 2007-10-07
Campaign no.
1
Median weekly air temperature tmedMax [°C]
17.09
tmed Min [°C]
3.17
2
19.56
10.60
3
24.33
11.30
4
28.16
16.30
5
17.67
9.17
Locality L1 Point no. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35
Locality L2
PAHs [ng m-3] BaP BghiPe 0.77 0.94 0.23 0.45 0.29 0.71 0.75 0.24 0.23 0.30 0.35 0.21 0.21 0.27 >LOD 0.25 0.28 0.34 0.27 >LOD >LOD >LOD 0.12 >LOD >LOD >LOD 0.17 >LOD 0.16 0.37 0.22 0.22 0.24 0.21 0.27
1.75 1.26 1.09 1.29 1.31 1.53 2.81 0.68 0.76 1.07 1.46 1.36 0.93 1.03 0.20 0.59 0.32 0.45 0.85 0.59 0.67 0.13 0.14 >LOD 0.28 0.11 0.16 0.11 0.08 0.73 0.90 0.86 0.54 0.99 1.16
Point no. 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91
PAHs [ng m-3] BaP BghiPe 1.62 0.20 0.36 0.80 0.59 0.77 0.52 >LOD >LOD 0.16 0.25 0.16 >LOD >LOD >LOD 0.18 0.20 0.28 0.21 0.16 0.15 0.18 >LOD >LOD >LOD >LOD >LOD >LOD >LOD 0.14 0.25 0.24 >LOD 0.37 0.29
2.67 1.24 0.68 1.75 1.00 1.71 1.51 0.77 0.21 0.69 0.57 0.66 0.64 0.29 0.18 0.22 0.26 0.65 0.36 0.40 0.51 0.17 0.24 0.07 >LOD >LOD >LOD 0.11 0.33 0.64 0.77 1.07 0.52 1.30 1.06
ACCEPTED MANUSCRIPT
Date
2007-11-19 2007-11-20 2007-11-21 2007-11-22 2007-11-23 2007-11-24 2007-11-25 2008-01-14 2008-01-15 2008-01-16 2008-01-17 2008-01-18 2008-01-19 2008-01-20 2008-02-25 2008-02-26 2008-02-27 2008-02-28 2008-02-29 2008-03-01 2008-03-02
Campaign no.
Median weekly air temperature tmedMax [°C]
tmed Min [°C]
6
5.96
1.46
7
7.20
2.43
8
12.19
1.04
LOD = limit of detection.
Locality L1 Point no. 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56
Locality L2
PAHs [ng m-3] BaP BghiPe 0.51 0.63 0.90 1.42 3.07 3.13 0.74 >LOD 0.21 0.23 0.63 0.83 0.14 >LOD 4.26 2.63 1.58 2.06 2.88 0.47 0.51
1.00 1.53 2.54 2.53 4.44 6.44 4.56 2.21 0.75 1.72 3.77 3.38 3.09 2.19 4.65 6.98 1.04 2.10 8.18 2.66 2.91
Point no. 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112
PAHs [ng m-3] BaP BghiPe 1.41 1.00 1.73 2.36 6.12 1.87 0.37 0.79 0.15 0.73 0.48 1.94 0.31 >LOD 3.82 2.67 1.21 2.14 1.24 >LOD 0.21
2.76 2.20 2.60 4.05 6.20 5.29 1.78 2.38 1.66 1.88 4.40 5.11 3.10 1.04 3.07 6.40 2.15 4.62 3.15 2.29 1.69
ACCEPTED MANUSCRIPT Table 2 Measured values of EfBghiPe and EfBaP in passenger cars. Date 2005-06-28 2005-06-28 2005-06-29 2005-06-29 2005-06-29 2005-10-11 2005-10-11 2005-10-12 2005-10-12 2005-11-23 2005-10-11 2005-11-23 2005-11-24 2006-03-08 2006-03-08 2006-03-08 2006-03-09 2006-03-09 2006-03-09 2006-05-31 2006-05-31 2006-05-31 31.5.2006 2006-06-01 2006-06-01
Vehicle SKODA Felicia 1.3/50 kW SKODA Felicia 1.3/50 kW SKODA Felicia 1.3/50 kW SKODA Octavia SDI/55 kW SKODA Octavia SDI/55 kW SKODA Fabia 1.4/44 kW SKODA Fabia 1.4/44 kW SKODA Fabia 1.4/44 kW SKODA Octavia SDI/55 kW SKODA Fabia 1.4/44 kW SKODA Octavia SDI/55 kW Ford Focus Flexifuel Ford Focus Flexifuel Ford Focus Flexifuel Ford Focus Flexifuel SKODA Fabia 1.4/44 kW SKODA Octavia SDI/55 kW SKODA Octavia SDI/55 kW SKODA Octavia SDI/55 kW SKODA Fabia 1.4/44 kW SKODA Fabia 1.4/44 kW SKODA Fabia 1.4/44 kW SKODA Fabia 1.4/44 kW SKODA Octavia SDI/55 kW SKODA Octavia SDI/55 kW
Fuel Natural 95 Natural 91 with 5% EtOH Natural 91 with 15% ETBE DF summer DF summer with 5% FAME Natural 95 Natural 91 with 5% EtOH Natural 91 with 15% ETBE DF summer with 31% FAME Natural 95 DF, summer Natural 95 with 85% EtOH Natural 95 with 85% EtOH Natural 95 with 85% EtOH Natural 95 with 85% EtOH Natural 95 DF summer DF summer with 5% FAME DF summer with 31% FAME Natural 95 Natural 95 Natural 91 with 5% EtOH Natural 91 with 15% ETBE DF summer with 5% FAME DF summer with 31% FAME
Driving mode
EfBaP
[µg km-1]
EfBghiPe
[µg km-1]
SDC SDC
0.0299 0.0129
0.0399 0.0326
SDC
0.0159
0.0295
SDC
0.0475
0.1084
SDC
0.0378
0.0962
SDC SDC
0.0186 0.0071
0.0202 0.0103
SDC
0.0134
0.019
SDC
0.0487
0.0719
SDC-CS SDC-CS
4.6033 0.134
4.9951 0.1584
SDC
0.0074
0.0119
SDC-CS
4.6914
7.9835
SDC-CS
0.3398
0.6441
SDC
0.0033
0.0038
SDC-CS SDC-CS
0.2581 1.7053
0.8296 2.1926
SDC
0.0597
0.134
SDC
0.0585
0.1035
SDC-CS SDC SDC
0.0889 0.0069 0.006
0.2296 0.0153 0.0112
SDC
0.0047
0.0079
SDC
0.0305
0.056
SDC
0.0256
0.0633
Natural 95, Natural 91 = additive-free petrols; EtOH = ethanol; ETBE = ethyl-tert-butyl ether; DF = diesel fuel; FAME = fatty acid methyl ester from rapeseed oil.
ACCEPTED MANUSCRIPT Table 3 Summary of regression model testing results. Parameter
L1W
L2W
L1C
L2C
L1
L2
W
C
L1L2WC
Absolute term of regression: b0
0.314 0.283
1.890
2.157
0.858
0.969
0.343
2.031
0.793
Slope of regression lines: b1
1.923 1.614
0.921
0.763
1.315
1.154
1.633
0.833
1.326
Number of points in regression: n Number of regression parameters: Q Degrees of freedom: v Residual variance: 2
25
21
18
17
43
38
46
35
81
2
2
2
2
2
2
2
2
2
23
19
16
15
41
36
44
33
79
0.120 0.0718
Residuals sum of squares: RSC 2.749 1.365 Weights for calculating common 47.90 27.28 section: w0 Weights for calculating common 2.736 5.361 slope: w1
2.022
0.974
1.150
0.936
32.36
14.61
47.14
33.69
9.96
9.62
26.49
23.61
56.74
65.65
59.48
71.01
0.0985 4.336
1.435
1.112
47.34
122.30
17.21
16.15
67.85
8.10
122.39
130.49
ACCEPTED MANUSCRIPT Table 4 Summaries of results of Bartlett’s test for mutual comparisons of variances and of F-test for mutual comparisons of regression lines. L1W L2W
Compared lines
L1C L2C
Number of points on compared lines: n
81
Number of regression lines compared: M Total degrees of freedom: V
L1W
L2W
L1W
L1C
L1
W
L1C
L2C
L2W
L2C
L2
C
43
𝑘
Bartlett’s criterion: B Critical value of Bartlett’s criterion: (𝑀 ‒ 1) Line match testing characteristic: R Critical test value: R < F0.95(M − 1, n − 2 M)
2 𝜒0.95
46
35
81
81
4
2
2
2
2
2
2
73
39
34
42
31
77
77
1.030
1.024
1.032
1.013
1.013
Parameter: D 1.024 1.027 Combined residual sum of squares of the 51.082 35.105 compared lines: RSCc Residual sum of squares of all groups of 122.296 47.144 compared data: RSCk 2 0.700 0.900 Combined estimate of variance: 𝜎 𝑐 Combined estimate of absolute regression 0.793 0.858 𝑘 members: 𝑏0 Combined estimate of regression slopes: 𝑏1
38
15.977
4.114 46.969
33.690
4.336 47.340 122.296 122.296
0.470
0.0979 1.515
80.833 51.676
1.050
0.671
0.969
0.343
2.031
0.793
0.793
1.154
1.633
0.833
1.326
1.326
1.326
1.315
60.53*
32.62*
24.03*
1.282
1.943
0.399
58.56*
9.49
5.99
5.99
5.99
5.99
5.99
5.99
5.654*
1.672
4.712* 0.283
0.0306
4.937* 13.153*
2.226
3.238
3.276
3.305
3.115
3.220
3.115
ACCEPTED MANUSCRIPT Table 5 Comparison of model line slopes (1) calculated by various statistical methods. Dataset-Statistical method
b1 value
b1 lower bound
b1 upper bound
b0 value
b0 lower bound
b0 upper bound
W-OR
2.348
1.419
2.996
0.145
0.002
0.318
W-RMA JK
2.045
1.747
2.902
0.222
0.063
0.306
W-RMA L
2.129
1.834
2.424
0.210
0.101
0.319
C-OR
1.441
0.598
2.068
1.247
0.531
2.184
C-RMA JK
1.271
0.962
1.909
1.488
0.846
1.890
C-RMA L
1.323
1.008
1.639
1.438
1.008
1.639
SDC-OR
1.927
1.620
2.238
0.000
SDC-RMA JK
1.961
1.650
2.430
-0.004
-0.015
-0.001
SDC-RMA L
1.961
1.938
1.984
-0.004
-0.011
0.003
SDC-CS-OR
1.451
1.162
1.786
0.000
SDC-CS-RMA L
1.416
1.021
1.812
0.053
-0.902
1.008
SDC-CS-RMA JK
1.507
1.011
1.727
-0.035
-0.266
0.382
W = warmer period; C = colder period; OR = orthogonal regression; RMA JK = reduced major axis jackknife method; RMA L = reduced major axis linear model.
ACCEPTED MANUSCRIPT
Table 6 Summary of F-test results for mutual comparison of slopes. L1W L2W
Slopes compared
L1C L2C 𝑐
L1W
L2W
L1W
L1C
L1
W
L1C
L2C
L2W
L2C
L2
C
Combined estimate of regression slopes: 𝑏1
0.891
0.968 0.827
1.718 0.836 1.227 0.883
Testing characteristic of line slopes: G
3.261*
2.909 7.637* 1.770 0.506 0.801 2.285
Critical test value: G < F0.95(M − 1, n − 2 M)
2.730
4.091 4.130
4.073 4.160 3.965 3.965
Asterisks indicate groups of regression lines for which the hypothesis on matching of slopes was rejected.