Introduction
Photochemical grid models (PGMs), such as the Comprehensive Air quality
Model with Extensions (CAMx; Ramboll Environ, 2014) and the Community
Multiscale Air Quality (CMAQ) model (Byun and Schere, 2006), are widely used
in air quality management to assess the effectiveness of potential control
strategies for secondary pollutants such as O3 and fine particulate matter (PM2.5). Source
apportionment (SA) analysis is an important component of this process to
identify source sectors and/or regions that are contributing to O3 and
PM2.5. The traditional approach to source attribution analysis is
brute-force or zero-out sensitivity analysis in which the emissions
from a given source sector are removed to quantify the contribution of that
sector. This approach is expensive and impractical for assessing the
contributions of a large number of source categories. Furthermore, it suffers from the
limitation that the sum of zero-out impacts over all sources will not equal
the total concentration (Koo et al., 2009). Tagged species methods, such as
the Ozone and Particulate Source Apportionment Technology (OSAT/PSAT) in
CAMx (Dunker et al., 2002; Yarwood et al., 2007), can efficiently track
contributions from many source sectors and/or regions and provide source
contributions that sum to the total concentration. These methods are
increasingly being used to help understand complex air quality issues (e.g.,
Wagstrom et al., 2008; Burr and Zhang, 2011; Baker and Kelly, 2014; Collet et
al., 2014; Wang et al., 2009; Li et al., 2012; Skyllakou et al., 2014).
Many source attribution studies in Europe have used receptor models and back
trajectory analysis for inert pollutants, i.e., primary particulate matter
(e.g., Querol et al., 2001, 2004, 2009; Belis et al., 2011, 2013; Viana et
al., 2014). Source attribution studies in Europe for secondary pollutants,
such as ozone and secondary PM2.5, have used PGMs with the zero-out
approach. Skyllakou et al. (2014) used the CAMx PSAT approach to
distinguish the contributions of local and regional sources to fine PM in
Paris. Reis et al. (2000) studied the impact of road traffic emissions on
regional ozone levels in Europe by zeroing out traffic emissions. Derwent et
al. (2005) used a similar approach to determine the contribution of shipping
emissions to ozone and acid deposition in Europe. Sartelet et al. (2012)
estimated the contributions of biogenic and anthropogenic emissions to
O3 and PM concentrations in Europe and North America by zeroing out one
source category at a time. Under the TRANSPHORM (Transport related Air
Pollution and Health impacts – Integrated Methodologies for Assessing
Particulate Matter) program, the source contributions of transport emissions
(road transport, shipping, aviation) to O3 and PM contributions in
Europe were assessed using WRF-CMAQ and the zero-out approach (TRANSPHORM,
2014). Derwent et al. (2008) conducted sensitivity studies with a global
chemical transport model to understand the effects of long-range transport
from North America and Asia to surface ozone in Europe. Aksoyuglu et al. (2016) estimated the contribution of ship emissions to the concentration
and deposition of air pollutants in Europe using CAMx with a zero-out
approach.
All of the PGM source attribution studies above for Europe investigated the
contributions of a limited number of source categories. Examples of European
source attribution studies addressing a larger number of source categories
include EMEP (2009), Brandt et al. (2013), and Tagaris et al. (2015). The
European Monitoring and Evaluation Programme (EMEP) study used 15 %
sector-by-sector emission reductions in the Unified EMEP model to determine
contributions from different emission sectors to depositions and air
concentrations in countries within the EMEP domain for 2006 (EMEP, 2009).
Brandt et al. (2013) used the EVA (Economic Valuation of Air Pollution) model system, which includes the Danish
Eulerian Hemispheric Model (DEHM), with a tagging approach to evaluate the
contributions from the 10 Standard Nomenclature for Air Pollution (SNAP)
source sectors to air-pollution-related health costs in Europe and Denmark
for the 2000 base year. More recently, Tagaris et al. (2015) used the
zero-out approach with CMAQ to calculate the contributions of emissions from
the 10 SNAP source sectors to air quality over Europe for a summer month
(July 2006).
The study described in this paper differs from and complements the previous
source attribution studies. We take advantage of the source apportionment
tools (OSAT and PSAT) in CAMx to calculate the contributions of the 10 SNAP
sectors, biogenic emissions, dust emissions, and sources outside the
modeling domain (boundary conditions) to a large number of cities in Europe.
The CAMx modeling database for 2010, developed as part of the Air Quality
Model Evaluation International Initiative (AQMEII), is used for the analysis.
AQMEII is a joint regional air quality model evaluation effort between the
North American and European modeling communities to improve the
understanding of uncertainties in model predictions of ozone and PM2.5
and to use this knowledge to guide model improvements (Rao et al., 2011). In
Phase 1 of the AQMEII, a large number of offline photochemical air quality
models were applied and evaluated in Europe and North America for the year
2006 using consistent inputs to the extent possible (Rao et al., 2011;
Galmarini et al., 2012). The second phase of AQMEII was dedicated to the
evaluation of online coupled chemistry–meteorology models over both
continents, with a primary focus on the year 2010 (Galmarini et al., 2015).
In Phase 3, the focus is on diagnostic evaluation through sensitivity
studies on model inputs and spectral decomposition of model errors
(Galmarini et al., 2016; Solazzo and Galmarini, 2016; Solazzo et al., 2017),
as well as on collaboration with the Task Force on Hemispheric Transport of
Air Pollution (TF-HTAP) on coordinating global–hemispheric–regional modeling
experiments to characterize long-range transport. As part of Phase 3, a CAMx
modeling database has been developed and evaluated for Europe (Solazzo et
al., 2017), and this database was used in the study described in this paper.
Model setup
Model configuration, domain, and inputs
Solazzo et al. (2017) developed a 2010 CAMx database for Europe and applied
CAMx version 6.1 (Ramboll Environ, 2014) over the European Union (EU). The
simulations used the Carbon Bond 2005 (CB05) gas phase chemistry (Yarwood et
al., 2005) and the coarse–fine (CF) aerosol scheme. CAMx was applied for the
whole year for a domain covering Europe and a portion of Africa. The domain
(see Fig. 1) is defined in a Lambert conic conformal projection that
includes 270 by 225 grid cells with a 23 km horizontal grid resolution.
CAMx modeling domain with 270 by 225 grid cells
at 23 km horizontal grid resolution. The figure also shows the 16 cities
considered for the source attribution analysis.
Input meteorological data were generated using WRF-Chem 3.4.1, the coupled
chemistry version of the Weather Research and Forecasting (WRF) model
(Skamarock et al., 2008), driven by the European Centre for Medium-Range
Weather Forecasts (ECMWF) analysis fields. WRF-Chem was used rather than WRF
to obtain emission estimates for wind-blown dust. Analysis nudging for wind
speed, temperature, and water vapor mixing ratio was employed both within and
above the boundary layer, with a nudging coefficient of 0.0003 s-1.
The WRF vertical grid with 33 layers extends from the surface to a fixed
pressure of 50 hPa (about 20 km), with a surface layer depth of 24 m. The
WRF-CAMx pre-processor (version 4.2; Ramboll Environ, 2014) was used to
create CAMx input files, collapsing the 33 layers used by WRF to 14 layers in
CAMx, but keeping layers up to 230 m above ground level identical to the WRF
layers.
Anthropogenic emissions for the calendar year 2009 were derived from the
TNO-MACC_II emission inventory (Kuenen et al., 2014; Pouliot
et al., 2015), resolved by the SNAP sector (see Table 1). The primary data
sources were national emission inventories developed by European countries
in accordance with guidance provided by the European Environment Agency
(EEA, 2013). SNAP sector 34 combines industrial combustion (sector 3)
with industrial processes (sector 4) to mitigate inconsistent
classification of sources to sector 3 or 4, as discussed by Kuenen et al. (2014). Supplement Sect. A provides a summary of SO2, NOx, and
PM2.5 emissions from the nine SNAP sectors for the summer and winter
months and presents spatial maps of total surface emissions and surface
emissions for some sectors.
SNAP sector classification of anthropogenic emissions.
Sector number
Description
1
Energy industries (e.g., power generation and refineries)
2
Non-industrial (residential) combustion
34
Industry∗
5
Extraction and distribution of fossil fuels
6
Solvent and other product use
7
Road transport (includes exhaust, evaporative, tire–brake–road wear)
8
Non-road transport (includes rail, aircraft, shipping, construction equipment)
9
Waste treatment
10
Agriculture
∗ Sector 34 combines industrial combustion (SNAP 3) with
industrial processes (SNAP 4) to mitigate inconsistent classification of
sources to sector 3 or 4 (see Kuenen et al., 2014).
For the road transport sector (SNAP 7), it should be noted that the provided
emission inventory does not include information on the composition of the
vehicle fleet in different cities in Europe because the emission inventory
was made available to the AQME participants with source contributions
grouped according the SNAP classification but without any additional
information about the car fleet or other proxies introduced in emission
computation. However, the MACC-II emission inventory (Kuenen et al., 2014)
that was used for this study was constructed by using the reported emission
national totals by sector. Therefore, for each country a representative car
fleet was used and the differences in fleet composition among different
countries are implicitly accounted for in the provided emission inventory.
The non-road transport sector (SNAP 8) includes a variety of emission
sources, such as off-road transport (shipping, railways, aviation, inland
waterways) and machinery (agriculture, forestry, industry, military,
airports).
Biogenic volatile organic compound (VOC) emissions were estimated by applying the Model of Emissions of
Gases and Aerosols from Nature (MEGAN; Guenther et al., 2012) v2.04. Sea
salt emissions were estimated using published algorithms (de Leeuw et al.,
2000; Gong, 2003). Dust emissions were based on the GOCART (Ginoux et al.,
2001, 2004) model implemented in WRF-Chem (Zhao et al., 2010). Chemical
boundary conditions were derived from the Monitoring Atmospheric Composition
and Climate (MACC) project using the Composition–Integrated Forecast System
(C-IFS) model (Flemming et al., 2015). The MACC data were available at
3-hourly time intervals with horizontal resolution of 1.125∘ × 1.125∘.
Variables were provided as 3-D fields in pressure hybrid vertical coordinates
and included gas phase species, namely CO, O3, NO, NO2, PAN,
HNO3, CH2O (formaldehyde), SO2, H2O2,
C2H6 (ethane), CH3COCH3 (acetone), CH3OH
(methanol), C3H8 (propane), C2H5OH (ethanol),
C2H4 (ethene), PAR (paraffins), ALD2 (aldehydes), OLE (olefins),
C5H8 (isoprene), CHOOH (formic acid), CH3OOH
(methyl peroxide), ONIT (organic nitrates), and aerosol species (dust,
sulfate, hydrophobic and hydrophilic organic matter, and hydrophobic and
hydrophilic black carbon, BC). Mineral dust aerosols were provided in three
different size bins, ranging from 0.03 to 20 µm. More information on
MACC data as well as their evaluation against a set of ground-based
measurements can be found in Inness et al. (2013) and Giordano et al. (2015).
Model performance evaluation summary
Solazzo et al. (2017) conducted a detailed model performance evaluation of
CAMx for 2010 in the framework of the AQMEII Phase 3 project. Here we
present a brief summary of model performance using a set of ground-based
observations belonging to the AirBase network (http://www.eea.europa.eu/data-and-maps/data/airbase-the-european-air-quality-database-8)
and covering most of the computational domain. Only background stations are
considered in the analysis. Furthermore, analysis was carried out for the
whole set, including urban, suburban, and rural sites (all background, AB),
as well as for rural background sites only (RB). Note that the rural
background sites are the most appropriate for model performance evaluation
because the coarse resolution (23 km) used in the simulation cannot
reproduce local-scale processes that can take place, particularly within the
urban areas, and that can also influence the observed values at background
sites. However, since urban areas were the main focus of the source
attribution analysis presented here, the evaluation was performed over
background stations, not just the rural background sites. The comparison of
the model performance for these two different sets of sites provides a
quantitative evaluation of the possible degradation of the CAMx results when
evaluated at urban sites, whose spatial representativeness is not always
adequate to the model resolution, even for background sites. As noted by
Terrenoire et al. (2015), model performance for RB sites is better than for
urban background (UB) sites, even when using a fine resolution of 8 km.
The model performance was evaluated over the whole year and based on daily
mean concentrations. Only stations that had more than 75 % of data
availability on a yearly basis have been included in the comparisons. The
number of available stations range according to the chemical species. The
highest availability was noted for NO2, with more than 2500 stations.
Ozone and SO2 were available at more than 1500 sites, while for
PM10 more than 2300 sites were available. PM2.5 observations were
available in about 700 sites all over Europe, with about 300 sites
corresponding to RB stations. Model performance was also evaluated at city
level for selected cities for the winter (January to March) and summer (July
to September) seasons of 2010 for consistency with the source attribution
analyses described in this paper.
Annual model performance evaluation for entire domain
CAMx model performance metrics at all (AB) and rural (RB)
background AirBase sites. Metrics are computed for daily mean concentrations
for the calendar year 2010.
SO2 [ppb]
NO2 [ppb]
O3 [ppb]
OX [ppb]
PM 10 [µg m-3]
PM 2.5 [µg m-3]
Parameter
AB
RB
AB
RB
AB
RB
AB
RB
AB
RB
AB
RB
# observations
550 113
90 446
954 709
141 241
646 965
144 139
561 059
108 438
842 896
115 022
267 121
36 378
Observed mean
2.3
1.6
14.0
6.9
26.3
29.2
36.7
34.6
27.8
21.7
17.5
14.5
Modeled mean
1.2
0.9
6.1
4.9
31.8
31.6
36.8
35.8
22.6
21.8
14.0
13.9
Observed SD
4.6
2.2
10.1
6.1
11.3
11.1
10.3
10.1
22.0
17.0
15.9
13.5
Modeled SD
1.4
1.0
4.8
4.2
12.0
12.2
10.6
10.7
13.8
13.5
9.2
9.2
Mean bias
-1.1
-0.6
-7.9
-2.0
5.6
2.4
0.0
1.1
-5.3
0.1
-3.4
-0.6
NMB (%)
-47.9
-40.9
-56.3
-29.4
21.1
8.2
0.0
3.3
-18.9
0.4
-19.7
-4.2
Mean error
1.7
1.1
8.4
3.3
8.6
7.4
6.8
6.6
14.4
11.9
8.3
7.1
NME (%)
72.1
69.2
60.0
47.8
32.7
25.2
18.5
19.1
51.7
55.0
47.3
49.3
FB (%)
-45.4
-37.6
-73.1
-29.3
19.7
6.4
-0.3
2.9
-15.4
3.5
-13.7
4.2
FE (%)
81.9
76.5
81.6
55.6
33.5
28.2
19.5
20.0
53.1
52.6
49.8
50.1
Correlation
0.24
0.32
0.52
0.59
0.66
0.69
0.64
0.67
0.28
0.30
0.48
0.52
RMSE
4.6
2.2
11.7
5.4
11.1
9.5
8.9
8.5
23.1
18.3
14.5
11.8
IoA
0.3
0.5
0.6
0.7
0.8
0.8
0.8
0.8
0.5
0.5
0.6
0.7
Table 2 provides a domain-wide summary of model performance for the AB and
RB sets of stations. The statistical performance measures used in the
evaluation are defined in Supplement Sect. B (Sect. B1). Correlation
refers to the Pearson correlation coefficient (r) and expresses the
temporal correlation between the observed and computed daily mean
concentrations. The yearly mean of the observed SO2 concentration is
2.3 ppb, while the modeled value is 1.2, corresponding to a 48 % low
bias. Similar results are noted at RB sites (normalized mean bias, NMB = -41 %). NO2
yearly mean concentrations are clearly underestimated when all background
sites are considered. However, when the analysis is limited only to RB
sites, which are more suitable for comparisons with a model run using a 23 km horizontal grid resolution, there is a noticeable improvement in model
performance. The NMB improves from -56.3 to -29.4 % and the
normalized mean error (NME)
decreases from 60 to 47.8 %. As a consequence, the RMSE drops from
11.7 to 5.4 ppb, and the daily correlation grows from 0.52 to 0.60. The
underestimation of NO2 concentrations may be because the grid
resolution is too coarse to resolve many of the monitoring locations, or
alternatively it may indicate that NOx emissions are underestimated in this
inventory.
Annual mean ozone concentrations at AB sites are overestimated (NMB =
21.1 %), while the standard deviation (SD) of the yearly series is
correctly reproduced (standard deviation of 11.3 ppb observed versus 12 ppb
modeled). Similar performance for SD is observed at RB sites, together with
a clear improvement in terms of the yearly mean, as pointed out by the
decrease in NMB and NME. These results suggest that the ozone bias at AB
sites is partially due to overestimation at urban and suburban sites, where
the horizontal grid resolution is insufficient to resolve ozone suppression
at monitors of nearby sources of NOx. This hypothesis is confirmed by
analysis of the Ox concentration (Ox = O3+NO2), which removes
the local effect of NOx titration. Ox concentrations at both AB and RB sites
are very well reproduced in terms of both mean and SD. Also, the temporal
variation of Ox concentrations is well reproduced, as shown by the
correlation value (0.64 and 0.67 at AB and RB sites, respectively).
PM10 concentrations are underestimated at AB sites (NMB = -19 %),
but the bias for RB sites is small (NMB = 0.4 % and fractional bias (FB) = 3.5 %).
Conversely, the NME (and fractional error, FE) remains high for both sets of stations. In
particular, the NME increases from 51.7 % at AB sites to 55 % at RB
sites. The temporal correlations are also low (< 0.3) in both cases.
The overall performance suggests that CAMx correctly captured the yearly
mean burden of aerosol but not its temporal evolution. This could be due to
compensating errors among different sources that could be either
underestimated or overestimated. The correlations for PM2.5 are better
(correlation = 0.48 and 0.52 at AB and RB sites, respectively), although
there is more underestimation bias for PM2.5 than for PM10. These
results suggest that coarse PM mass is likely overestimated and its temporal
evolution is not correctly reproduced by CAMx. Note that a large fraction of
the coarse PM can be attributed to dust and/or sea salt sources and there
are large uncertainties in estimating emissions from these sources.
Supplement Sect. B provides additional details and discussion on the
spatial and temporal annual performance of the model. Below, we present
model performance results for specific cities during the winter and summer.
Seasonal model performance evaluation for selected cities
Model performance was also evaluated at those cities selected for source
apportionment analysis (see Sect. 2.3 for the selected cities) that had
available background stations. For each city, all AirBase background
stations belonging to an area of about 50 × 50 km2 placed around the
city were selected. For all cities, at least two sites were included when
available. The analysis was carried out over two quarters: January–March and
July–September 2010. The two quarters cover the months of February and
August that were selected for SA analysis. Ozone was evaluated only for the
summer season, while PM2.5 was evaluated for both periods. Model performance evaluation (MPE) was
based on the same indicators used for the performance analysis of the annual
results (see Sect. B1 in Supplement B).
Summary of CAMx model performance evaluated at background AirBase
sites belonging to the selected cities. Statistics are computed for daily
mean O3 concentrations over the summer season
(1 July–30 September).
City
#
Obs.
Model
Obs.
Model
Mean
NMB
Mean
NME
FB
FE
Corr.
RMSE
IoA
obs.
mean
mean
SD
SD
bias
(%)
error
(%)
(%)
(%)
Amsterdam
327
18.7
19.4
7.6
6.5
0.7
3.9
4.0
21.3
6.1
23.0
0.75
5.1
0.86
Budapest
341
27.2
38.3
8.3
10.4
11.1
40.9
11.4
42.1
35.0
36.2
0.70
13.4
0.62
Helsinki
265
27.0
28.7
10.3
7.8
1.7
6.3
5.3
19.4
10.2
21.4
0.78
6.6
0.86
Oslo
178
23.3
23.7
7.4
6.7
0.4
1.5
5.0
21.4
2.5
23.4
0.63
6.1
0.79
Athens
816
39.9
39.3
13.1
5.5
-0.6
-1.4
9.6
24.2
3.8
26.3
0.42
11.9
0.52
Barcelona
1769
29.6
34.8
8.0
6.9
5.2
17.5
7.7
26.1
18.0
25.5
0.44
9.5
0.60
Berlin
735
28.3
31.7
12.8
10.9
3.4
12.1
5.7
20.0
14.9
22.4
0.88
6.9
0.91
Copenhagen
178
20.2
24.6
7.3
6.6
4.4
21.7
5.3
26.1
21.8
26.1
0.77
6.4
0.80
Lisbon
179
34.4
35.4
10.0
6.8
1.0
2.9
5.0
14.7
5.3
15.1
0.79
6.3
0.85
London
428
17.4
20.1
7.0
5.7
2.7
15.3
4.9
28.1
16.8
29.2
0.55
6.7
0.72
Paris
1517
25.4
25.7
8.0
6.6
0.3
1.2
3.7
14.6
3.0
15.5
0.81
4.7
0.89
Warsaw
451
24.6
30.3
9.6
9.5
5.8
23.5
6.6
26.7
23.8
26.3
0.80
8.3
0.82
The performance evaluation results for summer ozone are summarized in Table 3 and Figs. 2 and 3. CAMx reproduced the ozone summer mean fairly well,
though it showed a slight and systematic overestimation. FB was lower than
20 % at all sites, except for two cities in eastern Europe, Warsaw
(24 %) and Budapest (35 %). Temporal correlations ranged between 0.6 and
0.8 for all sites, with the exception of two Mediterranean sites, Barcelona
and Athens, where the correlation dropped to 0.4. As shown in Fig. 3, the
degradation in correlation in Barcelona is due to an overestimation taking
place in July and September, while in August ozone concentrations were
correctly captured. In contrast, the worsening in model performance at
Athens is probably due to the higher variability, in both space and time,
shown by observed data, which is not captured by CAMx.
Evaluation of CAMx model performance for
O3 at selected cities for 1 July to 30 September
2010. (a) Comparison of the observed (black) and modeled (red) seasonal
mean concentrations. Bars show the corresponding observed (grey) and modeled
(orange) standard deviations. (b) The seasonal fractional bias
(orange), fractional error (red), and correlation (green) computed for each
city.
Time series of the box and whisker plots for the distribution of
the observed (black and grey) and computed (red and orange) daily concentrations of
O3 at background AirBase sites in the selected cities for 1 July–30 September 2010. Bars show the interquartile range and lines show the
median values. Values for the 25th, 50th, 75th, and 95th quantiles are also
reported for each city.
The analysis of CAMx time series of ozone concentrations illustrates the
systematic overestimation in Budapest for all percentiles, while in Warsaw
the overestimation is primarily associated with the median and the third
quartile. At all other sites, CAMx is able to capture both the seasonal
trend, slightly decreasing from July to September, as well as the
development of most of the short-term episodes (e.g., during the first and
second half of July in Lisbon, Paris, Berlin, Amsterdam, and London).
Summary of CAMx model performance evaluated at background AirBase
sites belonging to the selected cities. Statistics are computed for daily
mean PM2.5 concentrations over the summer season
(1 July–30 September).
City
#
Obs.
Model
Obs.
Model
Mean
NMB
Mean
NME
FB
FE
Corr.
RMSE
IoA
obs.
mean
mean
SD
SD
bias
(%)
error
(%)
(%)
(%)
Amsterdam
204
11.5
13.6
5.3
4.8
2.1
18.3
4.6
40.3
18.9
37.9
0.38
6.0
0.62
Helsinki
541
9.6
8.9
6.9
6.4
-0.6
-6.8
4.4
46.1
-6.4
47.7
0.51
6.6
0.70
Oslo
533
8.1
7.7
2.9
3.6
-0.4
-4.9
2.8
34.3
-11.0
37.5
0.43
3.5
0.65
Athens
163
25.2
14.0
8.0
6.4
-11.2
-44.4
11.5
45.7
-60.0
61.6
0.57
13.1
0.55
Barcelona
630
13.8
11.1
5.4
5.0
-2.7
-19.6
5.4
39.3
-22.5
41.7
0.22
7.1
0.51
Berlin
537
13.0
8.7
5.0
3.3
-4.3
-33.3
5.3
40.9
-37.9
48.1
0.30
6.7
0.53
Copenhagen
172
10.3
10.4
4.0
4.6
0.1
1.3
3.7
36.2
-2.3
36.5
0.38
4.8
0.62
Lisbon
172
11.3
10.1
5.7
8.6
-1.3
-11.1
5.4
47.8
-11.2
50.8
0.48
7.8
0.66
London
560
10.7
10.5
4.0
4.3
-0.2
-1.9
3.9
36.5
-3.8
35.7
0.24
5.1
0.54
Paris
430
11.0
12.6
4.7
5.0
1.6
14.7
3.9
35.0
15.5
32.6
0.55
4.8
0.72
Warsaw
276
19.9
9.7
9.0
5.3
-10.2
-51.3
11.2
56.3
-67.1
73.1
0.29
13.6
0.49
Stockholm
482
7.2
6.8
4.0
2.9
-0.4
-5.0
3.1
43.5
-1.2
42.0
0.25
4.4
0.50
PM2.5 was evaluated for both summer and winter. During the warm
season, the observed mean concentration ranges between 6 and 14 µg m-3 (see Table 4 and Fig. 4), except in Athens and Warsaw, where the
seasonal values are around 25 and 20 µg m-3, respectively.
PM2.5 mean concentrations were underestimated at most sites, with FB
substantially close to 0 in Copenhagen and London and ranging between
-5 and -20 % in Barcelona, Lisbon, Berlin, Oslo, and Helsinki. As
already mentioned, the worst performance was for Athens and Warsaw. Finally,
PM2.5 concentrations are partially overestimated in Paris and
Amsterdam. FE is generally lower than 40 %, again with the exception of
Athens and Warsaw, proving that, beyond the mean values, the whole
distribution of the daily mean concentrations is also reproduced fairly
well. Conversely, temporal correlations range between 0.2 and 0.6, pointing
out the model difficulty in capturing the exact temporal variability of the
observed values. As shown in Fig. 5, this is probably due to the very
low variability of the observed concentrations over the summer season. At
the Lisbon site, CAMx strongly overestimates an episode, showing a
concentration of about twice the observed one, which is probably due to a
corresponding overestimation of the contribution of the natural sources
(e.g., sea salt).
Evaluation of CAMx model performance for
PM2.5 at selected cities for July 1 to September 30,
2010. (a) Comparison of the observed (black) and modeled (red) seasonal
mean concentrations. Bars show the corresponding observed (grey) and modeled
(orange) standard deviation. (b) The seasonal fractional bias
(orange), fractional error (red), and correlation (green) computed for each
city.
Time series of the box and whisker plots for the distribution of
the observed (black and grey) and computed (red and orange) daily concentrations of
PM2.5 at background AirBase sites in the selected cities for 1 July–30 September 2010. Bars show the interquartile
range and
lines show the median values. Values for the 25th, 50th, 75th, and 95th quantiles
are also reported for each city.
Evaluation of CAMx model performance for PM2.5 at selected
cities from 1 January to 31 March 2010. (a) Comparison of the observed
(black) and modeled (red) seasonal mean concentrations. Bars show the
corresponding observed (grey) and modeled (orange) standard deviation. (b) The seasonal fractional bias (orange), fractional error
(red),
and correlation (green) computed for each city.
Summary of CAMx model performance evaluated at background AirBase
sites belonging to the selected cities. Statistics are computed for daily
mean PM2.5 concentrations over the winter season (1 January–31 March).
City
#
Obs.
Model
Obs.
Model
Mean
NMB
Mean
NME
FB
FE
Corr.
RMSE
IoA
obs.
mean
mean
SD
SD
bias
(%)
error
(%)
(%)
(%)
Amsterdam
260
25.6
23.6
19.6
11.3
-2.0
-7.7
10.3
40.3
3.1
38.6
0.56
16.3
0.67
Budapest
104
28.0
21.9
15.7
10.0
-6.1
-21.8
10.9
38.9
-17.4
41.5
0.38
16.2
0.59
Helsinki
507
11.8
18.1
6.4
8.3
6.3
53.3
8.1
68.7
42.8
55.0
0.34
10.6
0.51
Oslo
532
15.9
18.7
9.6
11.7
2.8
17.8
11.2
70.7
14.9
59.2
-0.14
16.4
0.21
Athens
212
19.3
12.9
11.4
9.4
-6.4
-33.3
9.2
47.5
-36.4
50.3
0.34
13.7
0.56
Barcelona
404
19.3
12.5
9.8
4.8
-6.9
-35.6
8.8
45.3
-36.1
50.2
0.31
11.7
0.51
Berlin
431
34.7
24.1
26.1
11.1
-10.5
-30.4
16.9
48.7
-18.1
48.3
0.45
25.6
0.56
Copenhagen
164
13.9
20.9
7.9
10.6
7.1
51.2
9.7
70.0
36.2
54.2
0.42
12.5
0.57
Lisbon
167
8.6
10.4
4.8
4.8
1.8
21.3
5.0
58.2
22.4
53.2
0.14
6.6
0.45
London
499
18.0
20.6
10.2
9.0
2.6
14.4
7.2
39.8
16.5
37.3
0.58
9.2
0.74
Paris
323
24.3
25.4
14.3
11.8
1.0
4.2
10.4
42.6
10.0
43.1
0.50
13.3
0.68
Warsaw
278
42.4
26.3
21.7
13.2
-16.1
-38.0
18.4
43.4
-43.4
51.7
0.51
24.7
0.60
Stockholm
491
9.6
14.2
6.1
7.8
4.6
48.4
6.8
71.3
34.6
54.8
0.38
9.1
0.56
During winter, CAMx is partially able to capture the spatial
variability of the observed concentrations, which range between 9 µg m-3 (Lisbon) and 42 µg m-3 (Warsaw), as shown in Table 5
and Fig. 6. CAMx clearly underestimates the seasonal mean concentration in
Warsaw (FB = -40 %) and Berlin (FB = -20 %), which show the highest
observed values, as well as in Athens and Barcelona (FB = -36 %).
Discrepancies between modeled and measured values in eastern Europe are
mainly related to the underestimation of several very strong episodes taking
place over the area in January (Fig. 7). However, CAMx performs better in
February (the month used for the SA analysis) and March. CAMx correctly
reproduces the mean concentrations in central–western Europe (London, Paris,
and Amsterdam), while it partially overestimates the observations at
northern European sites. The latter is probably due to an overestimation of
the sea salt contribution.
Time series of the box and whisker plots for the distribution of
the observed (black and grey) and computed (red and orange) daily concentrations of
PM2.5 at background AirBase sites in the selected cities for 1 January–31 March 2010. Bars show the interquartile range and lines
show the median values. Values for the 25th, 50th, 75th, and 95th quantiles are
also reported for each city.
The seasonal analysis performed at city level shows that CAMx is generally
able to capture the spatial and temporal patterns of the pollutant
concentrations across Europe, providing confidence in the different source
contributions estimated at each city, discussed in the following section.
Moreover, in the case of PM2.5, CAMx is also able to correctly capture
the seasonal variations.
Source attribution modeling
The source attribution modeling with CAMx used the OSAT and PSAT tagged
species methods in CAMx version 6.1 (Ramboll Environ, 2014). In addition to
the SNAP sector emission categories, the contributions of biogenic
emissions, dust and sea salt emissions (for PM), and boundary conditions of
model chemical species were explicitly tracked. Secondary organic aerosol
(SOA) was not apportioned by PSAT because of the high computer memory
requirement to track SOA categories on the large CAMx modeling grid. The
total biogenic and anthropogenic SOA were both available from the CAMx CF
aerosol scheme. Note that the PSAT approach apportions contributions to PM
species independently and can thus handle particulate ammonium having
different source contributions (e.g., from agriculture) than particulate
nitrate (e.g., from urban traffic emissions).
The source attribution simulations were conducted for a summer month (August 2010) and a winter month (February 2010). Although a model spin-up period of
1 week (last week of January 2010 for the winter simulation and the last
week of July 2010 for the summer simulation) was used to minimize the
influence of initial conditions, the contributions of initial conditions to
surface ozone and PM concentrations are included in the boundary condition
attribution component in the discussion of results in the following section.
Results
We selected 16 cities, representing the Nordic countries; countries in
western, central, and eastern Europe; and countries along the Mediterranean
coastline, for the source attribution analysis. The contributions of the
various source sectors to ozone and PM2.5 concentrations were
calculated for these cities and are discussed in this section. The
calculations were conducted using horizontal bilinear interpolation over
eight
grid cells around each city location.
Ozone source apportionment – summer
The European standard for ozone is based on the maximum daily 8 h mean
(not to exceed a threshold of 120 µg m-3, about 60 ppb, for 25 days
averaged over 3 years). Accordingly, the source apportionment results for
ozone are presented for the maximum daily 8 h average (referred to as
H1MDA8) for the summer month. Ozone results for the winter month are not
discussed here because H1MDA8 levels at all the selected cities are less
than the threshold and because boundary conditions dominate the ozone levels
in winter, with contributions at the 16 cities ranging from a low of 61 %
to a high of 96 %. The spatial pattern of calculated H1MDA8 ozone
concentrations across the modeling domain is shown in Fig. C1 in
Supplement C. Over most of western and northern Europe, ozone levels are
below 60 ppb. The highest ozone values (about 120 ppb) are predicted near
Moscow, Russia. The 60 ppb level is exceeded in some of the Mediterranean
countries and in parts of central and eastern Europe.
Sectors contributing 5 % or more to summertime H1MDA8
ozone concentrations. Sector contributions in percentages are shown in
parentheses.
City (ppb)
Sector∗ contributions (%)
Barcelona (58)
BC (28)
SNAP 7 (21)
SNAP 8 (18)
Biogenic (15)
SNAP 34 (7)
SNAP 1 (5)
Lisbon (61)
BC (34)
SNAP 7 (20)
Biogenic (19)
SNAP 8 (11)
SNAP 34 (6)
SNAP 1 (6)
Athens (69)
BC (26)
SNAP 7 (24)
SNAP 8 (16)
Biogenic (15)
SNAP 1 (8)
SNAP 34 (6)
Istanbul (73)
BC (26)
Biogenic (24)
SNAP 7 (15)
SNAP 8 (13)
SNAP 34 (9)
SNAP 1 (8)
Minsk (58)
BC (25)
Biogenic (23)
SNAP 7 (19)
SNAP 1 (15)
SNAP 8 (10)
–
Budapest (63)
SNAP 7 (35)
BC (29)
SNAP 1 (11)
Biogenic (10)
SNAP 8 (7)
SNAP 34 (5)
Warsaw (66)
BC (28)
SNAP 7 (24)
SNAP 1 (17)
Biogenic (14)
SNAP 8 (7)
SNAP 34 (7)
Kiev (70)
Biogenic (33)
BC (21)
SNAP 7 (18)
SNAP 8 (10)
SNAP 1 (9)
SNAP 34 (6)
London (41)
BC (56)
SNAP 8 (12)
SNAP 7 (11)
Biogenic (8)
–
–
Paris (44)
BC (59)
SNAP 7 (13)
Biogenic (10)
SNAP 8 (6)
SNAP 6 (6)
–
Amsterdam (51)
BC (29)
Biogenic (21)
SNAP 7 (19)
SNAP 6 (10)
SNAP 8 (8)
SNAP 1 (6)
Berlin (56)
BC (46)
SNAP 7 (17)
SNAP 1 (13)
Biogenic (11)
SNAP 8 (6)
–
Copenhagen (44)
BC (29)
SNAP 7 (23)
SNAP 8 (14)
SNAP 1 (13)
Biogenic (12)
SNAP 34 (5)
Oslo (50)
BC (41)
Biogenic (20)
SNAP 8 (14)
SNAP 7 (12)
–
–
Helsinki (50)
BC (31)
SNAP 8 (21)
SNAP 7 (17)
SNAP 1 (13)
Biogenic (13)
–
Stockholm (57)
SNAP 7 (24)
BC (21)
SNAP 8 (18)
Biogenic (18)
SNAP 1 (12)
–
∗ See Table 1 for anthropogenic (SNAP) sector descriptions.
Sectors contributing 5 % or more to summertime monthly mean
PM2.5 concentrations. Sector contributions in percentages are shown in
parentheses.
City (µg m-3)
Sector∗ contributions (%)
Lisbon (11)
BC (45)
SNAP 8 (18)
SOA (15)
SNAP 34 (6)
SNAP 7 (5)
–
–
Barcelona (12)
BC (40)
SNAP 8 (19)
SOA (11)
SNAP 7 (10)
SNAP 34 (5)
–
–
Athens (16)
BC (38)
SNAP 1 (15)
SNAP 8 (10)
SOA (9)
SNAP 10 (8)
SNAP 7 (7)
SNAP 34 (6)
Istanbul (17)
BC (49)
SNAP 34 (11)
SOA (8)
SNAP 1 (8)
SNAP 10 (7)
SNAP 8 (6)
–
Budapest (10)
SNAP 1 (23)
BC (23)
SNAP 10 (13)
SOA (13)
SNAP 34 (9)
SNAP 7 (8)
SNAP 8 (5)
Warsaw (13)
SNAP 1 (24)
BC (21)
SOA (13)
SNAP 10 (12)
SNAP 7 (10)
SNAP 8 (8)
SNAP 34 (8)
Minsk (13)
BC (27)
SOA (18)
SNAP 10 (14)
SNAP 1 (14)
SNAP 7 (8)
SNAP 8 (7)
SNAP 34 (7)
Kiev (13)
BC (37)
SOA (17)
SNAP 10 (12)
SNAP 1 (9)
SNAP 8 (9)
SNAP 34 (6)
–
Berlin (8)
SOA (19)
SNAP 1 (15)
SNAP 8 (14)
BC (14)
SNAP 7 (12)
SNAP 34 (10)
SNAP 10 (8)
London (10)
SOA (32)
SNAP 8 (23)
SNAP 7 (13)
BC (12)
SNAP 1 (7)
SNAP 34 (5)
–
Paris (11)
SOA (18)
SNAP 8 (16)
SNAP 10 (14)
BC (14)
SNAP 7 (13)
SNAP 34 (8)
SNAP 1 (8)
Amsterdam (13)
SNAP 8 (28)
SOA (23)
SNAP 7 (13)
SNAP 1 (9)
BC (9)
SNAP 34 (6)
SNAP 10 (6)
Oslo (8)
SNAP 8 (25)
SOA (25)
SNAP 2 (11)
BC (10)
SNAP 7 (9)
SNAP 1 (7)
SNAP 34 (6)
Stockholm (8)
SOA (31)
SNAP 8 (15)
BC (15)
SNAP 1 (12)
SNAP 7 (10)
SNAP 34 (7)
–
Helsinki (8)
SOA (31)
BC (15)
SNAP 8 (15)
SNAP 7 (13)
SNAP 1 (10)
SNAP 34 (5)
–
Copenhagen (11)
SNAP 8 (26)
SOA (23)
BC (11)
SNAP 7 (10)
SNAP 1 (10)
SNAP 10 (7)
SNAP 34 (6)
∗ See Table 1 for anthropogenic (SNAP) sector descriptions.
Sectors contributing 5 % or more to wintertime monthly mean
PM2.5 concentrations. Sector contributions in percentages are shown in
parentheses.
City (µg m-3)
Sector∗ contributions (%)
Lisbon (13)
SOA (47)
SNAP 2 (15)
SNAP 8 (13)
SNAP 7 (7)
SNAP 34 (6)
–
–
–
Barcelona (13)
SNAP 8 (21)
SOA (18)
SNAP 7 (18)
SNAP 2 (17)
SNAP 10 (7)
SNAP 1 (7)
SNAP 34 (7)
–
Athens (15)
SNAP 2 (20)
SNAP 8 (17)
SOA (13)
BC (12)
Dust (10)
SNAP 7 (10)
SNAP 1 (9)
–
Istanbul (26)
SNAP 2 (25)
SNAP 7 (11)
BC (11)
SNAP 34 (11)
SNAP 1 (10)
SNAP 8 (10)
SNAP 10 (9)
SOA (6)
Budapest (30)
SNAP 2 (29)
SNAP 7 (18)
SNAP 1 (17)
SNAP 10 (15)
SNAP 8 (7)
SNAP 34 (7)
–
–
Minsk (30)
SNAP 2 (33)
SNAP 10 (16)
SNAP 1 (13)
SNAP 7 (12)
SNAP 8 (10)
SNAP 34 (7)
–
–
Kiev (31)
SNAP 2 (37)
SNAP 10 (12)
SNAP 1 (11)
SNAP 8 (10)
SNAP 7 (10)
SNAP 34 (9)
–
–
Warsaw (38)
SNAP 2 (34)
SNAP 7 (17)
SNAP 10 (16)
SNAP 1 (12)
SNAP 8 (7)
SNAP 34 (6)
–
–
London (21)
SNAP 8 (23)
SOA (23)
SNAP 7 (19)
SNAP 2 (11)
SNAP 10 (7)
SNAP 1 (6)
–
–
Paris (25)
SNAP 2 (30)
SOA (16)
SNAP 7 (16)
SNAP 8 (13)
SNAP 10 (8)
SNAP 1 (6)
SNAP 34 (6)
–
Amsterdam (26)
SNAP 7 (19)
SNAP 8 (18)
SNAP 2 (16)
SNAP 10 (13)
SOA (12)
SNAP 1 (10)
SNAP 34 (7)
–
Berlin (32)
SNAP 2 (24)
SNAP 7 (18)
SNAP 10 (15)
SNAP 1 (12)
SNAP 8 (11)
SNAP 34 (7)
SOA (6)
–
Stockholm (17)
SNAP 7 (22)
SNAP 2 (19)
SNAP 8 (16)
SOA (14)
SNAP 1 (10)
SNAP 10 (7)
SNAP 34 (6)
–
Oslo (19)
SNAP 2 (47)
SNAP 8 (16)
SNAP 7 (11)
SOA (7)
SNAP 1 (6)
SNAP 10 (5)
–
–
Helsinki (21)
SNAP 2 (33)
SNAP 7 (18)
SNAP 8 (14)
SNAP 1 (9)
SOA (9)
SNAP 10 (7)
SNAP 34 (5)
–
Copenhagen (24)
SNAP 2 (20)
SNAP 8 (19)
SNAP 7 (14)
SNAP 10 (12)
SNAP 1 (12)
SOA (11)
SNAP 34 (6)
–
∗ See Table 1 for anthropogenic (SNAP) sector
descriptions.
The source attribution results for summertime H1MDA8 ozone in each city are
reported in Table 6 for contributors of 5 % or more. In the four cities near
the Mediterranean from Lisbon, Portugal, in the west to Istanbul, Turkey, in
the east, H1MDA8 ozone in August 2010 is estimated to be above or very close
to the standard of 60 ppb. Boundary conditions are the largest contributor
to H1MDA8 ozone in all four Mediterranean cities, with contributions ranging
from 26 to 34 % from east (Istanbul, Athens) to west (Barcelona,
Lisbon). Contributions from on-road transport (SNAP 7) are the next largest
(20 to 24 %) at three of the four cities (Lisbon, Barcelona, Athens). At
Istanbul, the second highest contribution (24 %) comes from biogenic
emissions, while on-road transport is the third largest contributor at
15 %. Non-road transport (SNAP 8; 12 to 18 % contribution) and
biogenic emissions (15 to 24 % contribution) are also significant
contributors at all four locations. The other anthropogenic sectors
contributing 5 % or more to summertime ozone in the Mediterranean cities
are industry (SNAP 34; 6 to 8 % contribution) and the energy sector (SNAP
1; 5 to 8 % contribution).
Boundary conditions are again an important contributor in the four cities in
central and eastern Europe, making the largest contributions in Minsk (25 %)
and Warsaw (28 %) and the second largest contributions in Budapest (29 %)
and Kiev (21 %). Road transport is the largest contributor (35 %) in
Budapest, while biogenic emissions are the largest contributor (33 %) in
Kiev. Road transport is the second largest contributor in Warsaw (25 %)
and biogenic emissions are the second largest contributor (23 %) in Minsk.
Biogenic emissions contribute less in Budapest (10 %) and Warsaw (14 %)
than in the other two cities. Other important contributing source sectors in
the central and eastern European cities are the energy sector (9 % in Kiev
to 17 % in Warsaw), the non-road sector (7 to 10 % contribution), and
the industry sector (5 to 7 % contribution).
In all four cities in western Europe (Paris, London, Amsterdam, and Berlin),
H1MDA8 ozone concentrations are below the 60 ppb threshold. Boundary
conditions are the largest contributor (29 to 59 %) to HD1MA8 ozone at
all four cities and contribute more than half the HD1MA8 ozone in London and
Paris and nearly 50 % in Berlin. Road transport is the next largest
contributor in Paris (13 %) and Berlin (17 %), while non-road transport
(12 %) and biogenic emissions (21 %) are the second largest contributors
in London and Amsterdam. For London, the most relevant
non-road transport emissions are likely due to the very intense shipping
activity along the English Channel, resulting in large NOx emissions (see, for
example, Fig. 8 in Kuenen et al., 2014). Road transport contributions in
London and Amsterdam rank third at 11 and 19 %, respectively. Non-road
transport contributes less than 10 % to HD1MA8 ozone in Paris, Amsterdam,
and Berlin. The energy sector is an important contributor (13 %) in
Berlin and contributes 5 to 6 % in London and Amsterdam. The energy
sector contribution in Paris is small (less than 3 %) since France
derives over 75 % of its electricity from nuclear energy. The solvent- and
product-use sector (SNAP 6) contributes 6 and 10 % to summertime ozone
in Paris and Amsterdam, respectively, but its contributions in London and
Berlin are less than 3 %.
Like the selected cities in western Europe, H1MDA8 ozone levels in the
Nordic cities (Oslo, Copenhagen, Stockholm, Helsinki) are below the European
threshold of 60 ppb. Boundary conditions again play an important role for
ozone in these cities and are the largest contributors in three of the four
cities. Road transport is the largest contributor (24 %) in Stockholm,
followed by boundary conditions (21 %). Road transport contributions to
the other three cities range from 12 % in Oslo to 23 % in Copenhagen.
Non-road transport is an important contributing sector (14 to 21 %) and
its contributions in Oslo and Helsinki are higher than on-road transport. As
noted by Kukkonen et al. (2016), emissions from shipping and harbors are an
important non-road transport influence for Oslo and Helsinki. Biogenic
emissions are also important contributors in all four cities, with
contributions ranging from 12 to 20 %. The energy sector contributes 12 to
13 % in Helsinki, Stockholm, and Copenhagen, but its contribution to H1MDA8
ozone in Oslo is slightly less than 5 %.
Figure 8 shows the source apportionment results for the 16 cities across the
distribution of summertime MDA8 ozone values. Results are shown for the
upper and lower quartiles, the median, and the 90th percentile value
for each city. Boundary conditions are clearly the primary contributors
across the MDA8 ozone distribution at a majority of the cities selected for
analysis, and particularly in London and Paris. Except for Budapest, boundary
conditions are the primary contributors at all cities at the low end of the
distribution. Road transport contributions are important in many cities,
particularly Budapest and Athens (across the distribution), Warsaw, Lisbon,
Minsk, Kiev, and Berlin (at the higher end of the distribution). Non-road
transport contributions are important in the Mediterranean cities and in the
Nordic cities, particularly at the higher end of the distribution. Biogenic
emissions are the highest contributors in Kiev at the high end of the
distribution and are also important in the Mediterranean cities and most
of the other selected cities. Contributions from the energy sector
are important in the central and eastern European countries,
particularly Warsaw.
As noted previously, boundary condition contributions also include the
contribution of initial conditions, which are expected to decrease over the
1-week spin-up period and the subsequent month-long simulation period. At
the end of the 1-week spin-up period (i.e., on 1 August), initial
condition contributions to summertime H1MDA8 ozone were estimated for six
cities (Paris, Lisbon, Warsaw, Athens, Oslo, and London) and ranged from
5 % at Lisbon (H1MDA8 ozone of 38 ppb) to 54 % at Oslo (H1MDA8 ozone of
22 ppb).
PM2.5 source apportionment – summer
Source attribution results for the distribution of
summertime MDA8 ozone concentrations for the 16 cities.
The European standard for fine PM is an annual average concentration of 25 µg m-3. Since we obtained source attribution for only 2 months,
our discussion of the PM2.5 source attribution focuses on the summer
and winter monthly average concentrations. The spatial patterns of monthly
mean PM2.5 concentrations for August 2010 across the modeling domain
are shown in Fig. C2 in Supplement C. The highest PM2.5
concentrations are near the southern and southeastern boundaries of the
domain and in the Mediterranean countries. These high concentrations are
likely due to the transport of Saharan dust from northern Africa from the
boundary conditions as well as from the dust emissions within the modeling
domain (which includes part of the Sahara) estimated by the GOCART model in
WRF-Chem. While most of the Saharan dust is coarse, a significant fraction
is fine mode (e.g., Zauli Sajani et al., 2012; Fig. 7 in Pio et
al., 2014). Removing the dust component from the calculated total PM2.5
concentrations reduces the highest concentrations along the southern
boundary by a factor of 2.
Table 7 shows the source attribution results for monthly mean PM2.5
concentrations in August 2010. In the Mediterranean cities of Lisbon,
Barcelona, Athens, and Istanbul, boundary conditions are the largest
contributors to mean August PM2.5 concentrations, with contributions
ranging from 38 to 49 %. Non-road transport and SOA are the second and
third largest contributors in Lisbon and Barcelona. In Athens, the energy
and non-road transport sectors are the second and third largest contributors,
while in Istanbul the industry sector is the second largest contributor and
SOA and the energy sector are the third largest contributors. Road transport
contributions are less than 5 % in Istanbul and less than 10 % in Lisbon
and Athens. The highest on-road transport contribution to the selected
Mediterranean cities is 10 % in Barcelona. The industry sector
contributions in all four Mediterranean cities are 5 % or more, while the SOA
contributions in the four cities are 8 % or more. The agriculture sector
(SNAP 10) contribution to August 2010 mean PM2.5 concentrations is 7 to
8 % in Athens and Istanbul and less than 5 % in the other two cities.
Boundary conditions are important contributors to monthly average PM2.5
concentrations at cities in central and eastern Europe as well, as shown in
Table 7, but the relative BC contributions in these regions are lower than
those in southern Europe. BC is the largest contributor in Minsk and
Kiev, while the energy sector is the largest contributor in Budapest and
Warsaw. The dominant component (> 60 %) of the boundary
condition contribution in Minsk and Kiev is primary fine crustal material.
The energy sector contributions range from 9 % in Kiev to 24 % in
Warsaw. SOA is also an important contributor in all four cities and is the
second largest contributor in Minsk (18 %) and Kiev (17 %). The
agriculture sector also has a large contribution in all four cities (12 % to
14 %), suggesting that ammonia emissions from agricultural activity leads
to formation of particulate nitrate. The industry sector contributes from
6 to 9 % of PM2.5 concentrations in the four cities. Road transport
contributions are 8 % in Budapest and Minsk and 10 % in Warsaw, but less
than 5 % in Kiev. Non-road transport contributions are more than 5 % in
the four cities, but less than 10 %.
Boundary conditions are not large contributors to the August monthly average
PM2.5 concentrations in any of the four western European cities. Boundary
condition contributions range from 9 % in Amsterdam to 14 % in Paris and
Berlin. SOA is the largest contributor in London, Paris, and Berlin, while
non-road transport is the largest contributor (28 %) in Amsterdam.
Non-road transport is an important contributor in the other three cities as
well, with contributions ranging from 14 % in Berlin to 23 % in London.
The energy sector has a 15 % contribution in Berlin, but less than 10 %
in the other three cities. Agriculture has a large contribution (14 %) in
Paris, but lower contributions in Berlin (8 %) and Amsterdam (6 %).
Agriculture contributions to the mean August 2010 PM2.5 concentrations
in London are less than 5 %. Road transport is an important but not
major contributor (12 to 13 %) in any of the four western European cities.
The source attribution results for the four cities in the Nordic countries show
the decreasing influence of boundary conditions in the northern portion of
the modeling domain. Boundary condition contributions are not as large as
for some of the cities to the south and they range from 10 to 15 %. SOA and
non-road transport are the highest contributors in Oslo and contribute about
25 % each. SOA is the largest contributor (about 31 %) in Helsinki and
Stockholm, while non-road transport is the largest contributor in
Copenhagen. Non-road transport is the second highest contributor in
Stockholm. Energy sector emissions contribute from 7 to 12 % to monthly
mean PM2.5 concentrations, while the on-road transport sector
contributes 9 to 13 %. Residential combustion (SNAP 2) contributes 11 %
in Oslo but less than 5 % in the other three Nordic cities.
Source attribution results for the distribution of
summertime daily PM2.5 concentrations for the 16
cities.
Figure 9 shows the source apportionment results for the 16 cities across the
distribution of daily average summertime PM2.5 concentrations. In the
Mediterranean cities, boundary conditions are more important at the high end
of the distribution, suggesting that the higher PM2.5 concentrations in
these cities are often associated with transport from outside of the domain.
Non-road transport is an important contributor across the distribution in
Barcelona and Lisbon, but it is less important in Athens and Istanbul. SOA
contributions are important in all cities, particularly Athens.
Boundary conditions are important contributors in Minsk and Kiev,
particularly at the high end of the distribution and to the 90th
percentile daily average PM2.5 in Warsaw. The energy sector is a large
contributor across the distribution in Warsaw and at the 90th
percentile value in Budapest. In Minsk, the energy sector is more important
at the low end of the distribution. SOA contributions are important at the
high end of the distribution in Minsk and Warsaw.
Boundary condition contributions to summertime PM2.5 are less important
at the selected cities in western Europe. Non-road transport and SOA are the
largest contributors in London and Amsterdam, particularly at the high end
of the distribution. In Berlin, SOA contributions are important across the
PM2.5 distribution, while non-road transport contributions are
important at the 75th and 90th percentile values. The
contributions of boundary conditions are small to negligible and SOA
contributions are high in all four selected Nordic cities. Non-road transport
contributions are important in Copenhagen and Oslo, particularly at the high
ends of the distribution. The energy sector is the highest contributor at
the 90th percentile value in Stockholm, but is less important at the lower
quantiles. Energy sector contributions are small but non-negligible in
Helsinki.
Initial condition contributions are also included in the boundary condition
contributions shown in Table 7 and Fig. 9. At the end of the 1-week
spin-up period on 1 August, estimated initial condition contributions to
24 h average summertime PM2.5 concentrations at six cities ranged
from 6 % at Warsaw (24 h average PM2.5 of 14 µg m-3)
to 17 % at Oslo (24 h average PM2.5 of 13 µg m-3).
These initial condition contributions are expected to be smaller for the
monthly average PM2.5 concentrations shown in Table 7.
PM2.5 source apportionment – winter
Figure C3 in Supplement C shows the spatial distribution of monthly mean
PM2.5 concentrations for February 2010 across the modeling domain. The
highest PM2.5 concentrations are again along the southern boundary of
the modeling domain, but the influence of boundary conditions further inside
the domain is lower than for the summertime PM2.5 concentrations, as
shown in Table 8. High PM2.5 concentrations are also predicted over
Poland and we see from Table 8 that, from the 16 cities selected for the
analysis, the highest PM2.5 concentration (38 µg m-3) is in
Warsaw.
As mentioned above, Table 8 shows that boundary condition contributions to
wintertime PM2.5 concentrations in cities along the Mediterranean
coastline are much lower than summertime contributions, particularly at
cities in the west, such as Lisbon and Barcelona, where BC contributions are
less than 5 %. BC contributions are slightly higher than 10 % in the
eastern Mediterranean cities (Athens and Istanbul). There are some
variabilities in source contributions among the four Mediterranean cities. In
Lisbon, SOA is the single largest contributor, explaining nearly 50 % of
the winter average PM2.5. Residential combustion is the next
largest contributor at 15 %, followed by non-road transport at 13 %.
Non-road transport is the largest contributor (21 %) in Barcelona,
followed by SOA, on-road transport, and residential combustion with
comparable contributions (17 to 18 %). Residential combustion is the
largest contributor in both Athens (20 %) and Istanbul (25 %). Non-road
transport is the next highest contributor in Athens, while on-road transport,
industry, and boundary conditions are the second highest contributors
(11 %) in Istanbul. Energy sector contributions are more important in the
eastern Mediterranean cities (9 to 10 %) than in the western cities (less
than 5 % in Lisbon and 7 % in Barcelona). Road transport contributions
in Lisbon and Athens are 10 % or less. Dust emissions within the modeling
domain contribute 10 % of the PM2.5 in Athens.
At the four selected cities in central and eastern Europe, residential
combustion is the single largest contributor to wintertime PM2.5, with
contributions ranging from 29 to 38 %. Boundary condition contributions
are less than 5 % in all four cities. Road transport and the energy sector
are the second highest contributors in Budapest (17 to 18 %), followed by
agriculture at 15 %. In Kiev, agriculture and the energy sector are the
second highest contributors (11 to 12 %), followed by on-road and
non-road transport at 10 % and industry at 9 %. Agriculture is the
second highest contributor in Minsk (16 %), followed by the energy sector
and on-road transport (12 to 13 %). In Warsaw, the second highest
contributions to wintertime PM2.5 are from on-road transport and
agriculture (16 to 17 %), while the energy sector contributes 12 %.
Non-road and industry contributions in Warsaw are comparable and less than
10 %.
There is significant variability in the source-sector PM2.5
contributions among the cities in western Europe. In London, non-road
transport and SOA are the largest contributors (23 %), followed by on-road
transport (19 %) and residential combustion (11 %). Nearly 90 % of the
SOA concentration in London is from biogenic precursors. Note that the
source attribution simulation conducted in this study only considers source
categories and does not distinguish among source regions. Thus, the SOA
concentration in London could be of local origin or transported. The main
contributors to the biogenic SOA concentrations in London were oxidation
products of monoterpenes (46 %) and sesquiterpenes (11 %) as well as
oligomerization of oxidized compounds (27 %). CAMx includes four pathways for
monoterpene oxidation (oxidation by OH, O3, NO3, or atomic oxygen)
and three pathways (OH, O3, NO3) for sesquiterpenes. The importance of
wintertime SOA in a number of different European countries is discussed in
Aksoyoglu et al. (2011) and Crippa et al. (2013).
The contribution from agriculture and the energy sector to wintertime
PM2.5 in London is about 7 %. In Amsterdam, on-road and
non-transport are the largest contributors (18 to 19 %), residential
combustion ranks second (16 %), and agriculture and SOA contribute 12 to
13 %. The energy sector contributes 10 % of wintertime PM2.5 in
Amsterdam, while the industry sector contributes 7 %. In both Paris and
Berlin, residential combustion is the largest contributor (30 % and
24 %, respectively). However, there are differences in the contributions
of the other source sectors in these two cities. SOA and on-road transport
contributions rank second in Paris at about 16 %, followed by non-road
transport at 13 %, and 6 to 8 % contribution from agriculture and the
energy and industry sectors. In Berlin, on-road transport also ranks second
but the contribution of SOA is only about 6 %. Agriculture (15 %), the
energy sector (12 %), non-road transport (11 %), and industry (7 %) are
also significant contributors to wintertime PM2.5 in Berlin.
Table 8 shows that, for all four cities in the Nordic countries, the
contribution of boundary conditions is less than 5 %. The largest
contributors in Oslo and Helsinki are residential combustion sources (47
and 33 %, respectively). The non-road and on-road transport sectors have
significant contributions as well in these two cities (16 and 11 % in
Oslo, respectively, and 14 and 18 % in Helsinki, respectively). SOA,
the energy sector, and agriculture contribute 5 to 7 % and 7 to 9 % of
the wintertime PM2.5 in Oslo and Helsinki, respectively. Residential
combustion is also the largest contributor in Copenhagen (20 %) but it is
followed closely by non-road transport (19 %). Road transport contributes
14 % of the wintertime PM2.5 in Copenhagen, and agriculture, the
energy sector, and SOA contribute about 11 to 12 %. Industry contributions
in Copenhagen are about 6 %. Road transport is the largest contributor
(22 %) in Stockholm but residential combustion and non-road transport are
significant contributors as well, with contributions of about 19 % and
16 %, respectively. SOA contributes 14 % to wintertime PM2.5 in
Stockholm, while the energy sector contributes about 10 % and agriculture
and industry contribute 6 to 7 %.
Source attribution results for the distribution of
wintertime daily PM2.5 concentrations for the 16
cities.
The source apportionment results for the 16 cities across the distribution
of daily average wintertime PM2.5 concentrations are shown in Fig. 10. The contribution of boundary conditions is negligible to small across
the distribution in all cities. SOA contributions in Lisbon dominate other
sources at the 90th percentile value and are also important at the
50th and 75th percentile values. Residential combustion sources
are important across the entire distribution in Istanbul and are the primary
contributors at the 90th percentile value. The industry and agriculture
sectors are also important contributors to the higher levels of wintertime
PM2.5 in Istanbul. Residential combustion is an also important
contributor in Athens and Barcelona, as is the non-road transport sector.
Residential combustion is the largest contributor across the distribution of
wintertime PM2.5 concentrations in the four selected cities in central and
western Europe. Other important sectors are agriculture and road transport.
Non-road transport contributions are also important in Minsk, Warsaw, and
Kiev, and the energy sector is important in Minsk, Budapest, and Warsaw. SOA
contributions are large in three (London, Paris, Amsterdam) of the four cities in
western Europe, particularly at the 90th percentile levels in London
and Paris. The importance of wintertime SOA in Paris is consistent with the
findings of Crippa et al. (2013). Residential combustion is an important
source in Paris, Amsterdam, and Berlin and has non-negligible contributions
in London as well. The energy sector is an important source in Berlin and,
to a smaller extent, in Amsterdam. The road and non-road transport sectors
are also important contributors in all four cities.
Residential combustion is an important source sector in the four Nordic cities,
particularly Oslo, where contributions from this sector dominate over the
entire distribution of wintertime PM2.5. SOA is the largest contributor
at the 75th percentile level in Stockholm and is also an important
contributor in Copenhagen. Both transport sectors are important in all four
cities, with the non-road transport sector contribution being larger than
the road transport contribution in Copenhagen and Oslo.
Initial condition contribution is also included in the boundary condition
contribution shown in Table 8 and Fig. 10. At the end of the 1-week
spin-up period on February 1, estimated initial condition contributions to
24 h average wintertime PM2.5 concentrations at six cities ranged
from less than 1 % at many cities to 5 % at Oslo. These initial
condition contributions are expected to be negligible for the monthly
average PM2.5 concentrations shown in Table 8.
Discussion
The source attribution analysis results show that long-range transport of
ozone from beyond Europe has a strong influence on summertime ozone in
August 2010 over most of Europe. The background summertime ozone
contribution, simulated by the boundary condition tracer in the OSAT
simulation, is about 26 to 34 % in southern Europe and 20 to 30 % in
central and eastern Europe. The boundary condition contributions in western
Europe are larger, ranging from about 30 to 60 %. In the Nordic cities, BC
contributions range from about 20 % in Stockholm to 40 % in Oslo.
Wintertime ozone levels are below the EU standard and are dominated by boundary
conditions (60 % to over 90 %). The contribution of intercontinental
transport (from North America and, to a smaller extent, from Asia) to ozone
levels in Europe has been studied extensively through data analysis and
modeling (e.g., Parrish et al., 1993; Wild and Akimoto, 2001; Lelieveld et
al., 2002; Li et al., 2002; Naja et al., 2003; Trickl et al., 2003; Derwent
et al., 2004, 2008; Auvray and Bey, 2005; Fehsenfeld et al., 2006; Guerova
et al., 2006; Richards et al., 2013).
Summertime ozone contributions from biogenic emissions range from about
10 to 30 %. At the cities selected for the analysis, the largest
biogenic contribution of 33 % is in Kiev, while the lowest contribution of
8 % is in London. For anthropogenic emission sectors, the combined
transportation sector (on-road and non-road transport) contributions range
from 30 to 40 % in cities along the Mediterranean coastline, cities in
central and eastern Europe, and cities in northern Europe. In western
Europe, the combined transport sector has a contribution of 20 to 30 %.
Contributions from the on-road transport sector are generally higher than
those from the non-road transport sector, except for a few cities. The two
transport sector contributions are comparable (within 3 %) in Barcelona,
Istanbul, London, and Oslo. Non-road transport contributions are slightly
higher than on-road contributions in Oslo and Helsinki. These results for
summertime ozone concentrations are qualitatively consistent with those of
Tagaris et al. (2015), who found that the on-road transport sector was the
largest overall anthropogenic source sector contributing to July 2006 ozone
concentrations in Europe, with non-road transport contributions ranking
second. Pouliot et al. (2015) noted that emissions from on-road transport in
Europe decreased from 2006 to 2009, while emissions from shipping increased.
This explains some of the higher contributions of non-road transport to
ozone concentrations in some cities that were calculated in our study.
The largest contributions of the energy sector were in central and eastern
Europe (9 to 17 %) and in the Nordic cities (5 to 13 %). The
power sector was also identified as a major contributor in Europe in the
study by Brandt et al. (2013). Industry contributions to summertime ozone
were important for the Mediterranean cities and cities in central and
eastern Europe, with contributions ranging from 5 to 9 %.
For summertime ozone, the total contribution from sources that cannot be
controlled within Europe (i.e., the boundary conditions and biogenic
emissions) ranges from 39 to 69 %. The largest uncontrollable
contributions are 69 % in Paris and 64 % in London, where the H1MDA8 city
center ozone concentrations are 44 and 41 ppb, respectively, well below
the 60 ppb threshold. However, lower ozone levels are not necessarily
associated with higher uncontrollable contributions, or vice versa. For
example, the H1MDA8 ozone concentration in Copenhagen is 44 ppb, with
anthropogenic sources contributing nearly 60 %. The highest H1MDA8 ozone
concentrations among the selected cities are predicted in Istanbul (73 ppb)
and Kiev (70 ppb), and the uncontrollable contributions are 50 and
54 %, respectively.
Boundary conditions constitute a large fraction (40 to 50 %) of the August 2010 average PM2.5 concentrations in the Mediterranean cities. The
influence of boundary conditions decreases from southern to northern Europe.
This decreasing south-to-north gradient suggests that the Mediterranean
cities were influenced by long-range transport of dust emissions from
northern
Africa. These results are qualitatively consistent with numerous studies on
the transport of Saharan dust and its contributions to PM levels in the
Mediterranean Basin and other parts of Europe (e.g., Querol et al., 2001,
2004, 2009; Lyamani et al., 2005; Escudero et al., 2005, 2007a, b;
Vanderstraeten et al., 2008; Marconi et al., 2014; Duchi et al., 2016). In
contrast, there is an increasing south-to-north gradient in contributions of
SOA (organic PM2.5 formed in the atmosphere from precursor VOC species)
to summertime PM2.5 levels. Modeled SOA in Europe and North America is
primarily associated with biogenic emissions (e.g., Sartelet et al., 2012).
The contributions of SOA to summer PM range from 8 to 15 % in the
Mediterranean cities to 23 to 31 % in the Nordic cities.
The anthropogenic source-sector contributions to summertime average
PM2.5 vary with region. The important anthropogenic sectors in summer
are the transport sector (both on-road and non-road), the energy sector, the
industry sector, and agriculture. These sectors were also shown to be
important for annual PM2.5 in the EMEP (2009) study. The
contribution of other anthropogenic source sectors to the mean monthly
PM2.5 is generally less than 10 %, with the exception of the solvent-
and product-use sector, which has a contribution of over 10 % in
Amsterdam.
The source attribution results for wintertime PM2.5 are significantly
different from the summertime results. The contributions of boundary
conditions are generally less than 5 %, with the exception of the eastern
Mediterranean cities of Athens and Istanbul, where the BC contributions are
12 and 11 %, respectively. SOA contributions are small (less than 10 %)
to moderate (about 20 %) at most locations, except in Lisbon, where the
SOA contribution is nearly 50 %.
The important anthropogenic sectors for wintertime PM2.5 are
residential combustion, the combined transport sector (on-road and
non-road), the energy sector, and agriculture, again qualitatively
consistent with the EMEP (2009) results. Residential combustion
contributions in winter are much higher than in summer and range from over
10 % in London to nearly 50 % in Oslo. Residential combustion is the
largest contributor in 11 of the 16 cities studied in this work. Higher
winter contributions from this sector are consistent with residential wood
burning for heating in winter (e.g., Denier van der Gon et al., 2015;
Crilley et al., 2015), particularly in northern Europe (e.g., Krecl et al.,
2008). As shown in Supplement Sect. A, primary PM2.5 emissions from
residential combustion are a factor of 10 higher in winter than in summer.
Our model results are subject to limitations in model formulation and input
data. Model performance evaluations presented here and by others, such as
AQMEII Phase 3 contributors (see Solazzo et al., 2017), can suggest where
modeling uncertainties exist and how they can influence source
contributions. Important sources of uncertainty include anthropogenic
emission inventories, biogenic emissions, dust emissions, sea salt
emissions, boundary conditions, meteorology, and model formulation (e.g., SOA
treatment). These uncertainties influence model performance as well as the
source attribution analysis. A detailed uncertainty analysis using
sensitivity studies would provide more insight on the linkage between model
performance and the source attribution analysis. Although such an analysis
was not conducted as part of this study, it is useful to discuss how
uncertainties in inputs and model formulation can introduce uncertainties in
the source attribution results. For example, when differences between
modeled and observed concentrations are mostly driven by meteorology, we may
expect, as a first approximation, that the relative source contributions are
reasonable even though the absolute contributions are not well captured. In
contrast, discrepancies related to emissions, boundary conditions, or model
processes can be expected to bias both the absolute and relative
contributions of specific sources. Uncertainties in boundary conditions, NOx
emissions, and biogenic emissions are important for both O3 and
PM2.5. Uncertainties in SOA formation algorithms and dust emissions are
important for PM2.5. For example, model underestimation for PM2.5
in summer could be due to underestimation of secondary organic aerosol caused by missing emission
categories (e.g., intermediate VOC), biased inventories (e.g.,
uncertain biogenic emissions), and/or biased model SOA schemes, and these
errors would influence the calculated source contributions. Quantifying
source contributions can help assess when uncertainties are influential,
keeping in mind that errors that underestimate impacts from a specific
source may be less obvious than overestimation. The performance evaluation
for summertime ozone showed overestimation at many cities, particularly in
Budapest (fractional bias of 35 %). The source attribution analysis showed
that boundary conditions had a significant contribution to summertime ozone
in many cities, including a large contribution in Budapest (29 %),
suggesting that boundary condition contributions may be overstated, leading
to the overestimation bias.
The study presented here provides useful information on the contribution of
sources that can be controlled (anthropogenic sources within Europe) versus
uncontrollable sources, such as boundary conditions and biogenic
emissions. This information can be used as part of the decision making
process (along with economic, political, and societal considerations) by
policy makers in efforts to improve air quality.