High surface ozone concentrations,
which usually occur when photochemical ozone production takes place, pose a
great risk to human health and vegetation. Air quality models are often used
by policy makers as tools for the development of ozone mitigation strategies.
However, the modeled ozone production is often not or not enough evaluated in
many ozone modeling studies. The focus of this work is to evaluate the
modeled ozone production in Europe indirectly, with the use of the
ozone–temperature correlation for the summer of 2010 and to analyze its
sensitivity to precursor emissions and meteorology by using the regional air
quality model, the Comprehensive Air Quality Model with Extensions
(CAMx). The results show that the model significantly
underestimates the observed high afternoon surface ozone mixing ratios
Surface ozone (O
The peak values of surface ozone concentrations usually occur in the summer afternoon hours when the temperature reaches its diurnal maximum and the incoming solar radiation is still ample. Since the very high ozone concentrations increase the risk for damage to human health, as it happened, for example, during the European heat wave in 2003 (Filleul et al., 2006), the understanding of ozone formation and reduction of risks is of primary interest. In order to better understand the role of drivers for ozone production and to introduce successful ozone mitigation strategies by means of CTMs, a consistent and careful model evaluation and data interpretation are required.
The evaluation of modeled ozone production by just comparing modeled ozone concentrations with measurements may be misleading, as an agreement between modeled and observed ozone concentrations might just be the result of compensating errors. On the other hand, it is known that surface ozone has a high positive correlation with temperature (Sillman and Samson, 1995; Pusede et al., 2015). As a result, temperature has been used in several studies (Neftel et al., 2002; Baertsch-Ritter et al., 2004; Bloomer et al., 2009) as a surrogate to indirectly assess surface ozone production via the ozone–temperature correlation. However, so far, the use of the ozone–temperature correlation was only applied locally for individual stations and not at a greater regional scale. In this study, we adopted alternative methods to assess the ozone concentrations, to unmask compensating errors and to evaluate the modeled ozone production in Europe. Furthermore, by applying sensitivity tests, we characterized the response of modeled ozone production to its two main drivers: emissions and meteorology.
The paper is organized as follows. In Sect. 2, the data and modeling methods are introduced; results are given in Sect. 3 beginning with model evaluation and then continuing with the evaluation of afternoon ozone mixing ratios, ozone production and its response to changes in model input such as emissions, meteorological parameters, initial and boundary conditions. Finally, conclusions are summarized in Sect. 4.
The European model domain and its subregions: the Iberian Peninsula (IP), the Mediterranean (MD), Po Valley (PV), eastern Europe (EA), mid-Europe (ME), Benelux (BX), the British Isles (BI) and Scandinavia (SC). Grey dots indicate the rural background European Air Quality Database v7 (AirBase) stations of the hourly ozone measurements.
In this study, we used the regional air quality model, the Comprehensive Air Quality Model with Extensions
(CAMx, version 6.30;
The gas-phase mechanism used in this study was CB6r2 (Carbon Bond mechanism,
version 6, revision 2; Hildebrandt Ruiz and Yarwood, 2013). We simulated the
particle concentrations using CAMx's fine/coarse options. CAMx uses the
ISORROPIA (Nenes et al., 1998, 1999) model for inorganic thermodynamics and
gas–aerosol partitioning. We calculated the organic aerosol concentrations
using the SOAP model (Strader et al., 1999). The calculation of dry
deposition was based on the algorithms of Zhang et al. (2003). The initial
and boundary conditions for the chemical species were obtained from the
Model of Ozone and Related Chemical Tracers (MOZART) global model data for
2010 with a time resolution of 6 h (Horowitz et al., 2003). These data were
then interpolated to the size and resolution of our grid using the CAMx
preprocessor MOZART2CAMx (RAMBOLL-ENVIRON, 2016). The photochemistry in CAMx
is performed in two steps. First, clear-sky photolysis rates are calculated
externally by the Tropospheric Ultraviolet and Visible (TUV) radiation model
(NCAR, 2011) and then used as input into CAMx, where they are internally
adjusted every hour for clouds, aerosols, pressure and temperature (Emery et
al., 2010). In addition, for more accurate radiative transfer calculations,
the eight-stream discrete ordinates scheme was used (Stamnes et al., 1988).
Total Ozone Mapping Spectrometer (TOMS) data obtained by the National
Aeronautics and Space Administration
(
The meteorological parameters required as input for the air quality
simulations were generated by the Weather Research and Forecasting model
(WRF, version 3.7.1; Skamarock et al., 2008). The model domain and horizontal
resolution were identical to those used for CAMx model (see Sect. 2.1) while
there were 31 vertical layers up to 100 hPa, of which 14 were selected for
the CAMx runs for computational efficiency. The terrain and land use data
were taken from 10 arcmin data available
from the United States Geological Survey (USGS). The selected key physical
options for WRF parameterization are summarized in Table S1. Initial and
boundary conditions for WRF were generated using 6 h European Centre for
Medium-Range Weather Forecasts (ECMWF) reanalysis global data of
0.72
The WRF output was preprocessed with the WRFCAMx algorithm (Ramboll Environ,
2016) before being used by CAMx. The WRFCAMx preprocessor interpolates the
meteorological variables from the WRF domain to the CAMx domain (in our case,
only a vertical selection of the aforementioned 14 layers was done).
Furthermore, it calculates vertical diffusivity (
Total NMVOC (non-methane volatile organic
compounds)
We used the TNO-MACC-III European anthropogenic emission inventory for 2010
provided by the Netherlands Organization for Applied Scientific Research
(TNO). The TNO-MACC-III is an extension of the TNO-MACC-II emission inventory
(Kuenen et al., 2014) with some updates which are described in Kuik et
al. (2016). It contains annual emission data for 10 Selected
Nomenclature for Air Pollution (SNAP) categories per grid cell (Table S2). The TNO
emission domain covers the same geographical space as our domain (Sect. 2.1)
but with a higher horizontal resolution
(0.125
The biogenic emissions (isoprene, monoterpenes, sesquiterpenes, soil NO) were
calculated according to the method described by Andreani-Aksoyoglu and
Keller (1995) using temperature, shortwave solar radiation and USGS land use
data from the WRF output and the GlobCover 2005–2006 inventory
(
Meteorological observations from European stations with 3 h time intervals
were obtained from the British Atmospheric Data Centre (BADC) using the UK
Met Office Integrated Data Archive System (MIDAS) Land Surface Stations
database (Meteorological Office, 2013). Even though the UK stations have
hourly observations, for the sake of a more homogeneous and consistent model
performance evaluation for the whole European domain, the 3 h interval was
used for the UK stations as well. The extracted meteorological parameters
were dew point and air temperature at 2 m (
There are no direct measurements of the PBLH, but it can be estimated with
different methods by using sounding data. Such data were extracted from the
University of Wyoming database
(
The observational data for the surface air pollutant concentrations
(
Definition of statistical metrics for model performance evaluation.
For comparison with surface observations, the values in the lowest model
layer were interpolated (bilinear interpolation) to each station's
coordinates, while for the evaluation of vertical profiles, the
nearest-neighbor method was used for horizontal interpolation together with
linear vertical interpolation to 14 constant heights
above the ground. The statistical
metrics that were used for the meteorological and air quality model
performance evaluation are summarized in Table 1. The statistical metrics for
the wind direction were calculated only for wind speeds higher than
1.5 m s
Performance criteria and goals for model results (from Emery et al., 2001; EPA, 2007; Boylan and Russel, 2006).
Description of sensitivity tests.
In addition to the aforementioned traditional evaluation methods, we used other, less common, approaches for the evaluation of modeled ozone in our study. We applied these non-traditional methods in the afternoon hours (12:00–18:00 UTC; only 12:00, 15:00 and 18:00 UTC for the meteorology) when the ozone production and mixing ratios often reach their maximum. For the evaluation of ozone mixing ratios, we divided the observed values into mixing ratio bins of 10 parts per billion by volume (thereafter ppb) between 20 and 70 ppb, plus one bin incorporating all the values equal to or higher than 70 ppb. For each observed ozone mixing ratio bin, we calculated the mean bias (as defined in Table 1) between the respective model values and observations. This approach shows and quantifies more clearly the model's prediction for each respective observed value set, avoiding compensation of errors on the temporal scale. This greatly improves the interpretation of the model's prediction, especially if it is to be compared with other models or sensitivity tests.
The evaluation of ozone production was performed indirectly, with the use of
its correlation with temperature as discussed in Sect. 1. We made use of the
ozone–temperature correlations as described in the following three
approaches:
We selected eight surface stations (see Table S4 for details), which have
measurements of both temperature and ozone, and performed regression analysis
(using a scatterplot) between afternoon mean ozone mixing ratios and the
respective afternoon mean temperature for each station. Since we used
different measurement networks for the air quality and meteorology, the
characterization of a station as common in both networks was based on the
very small difference ( In order to evaluate the model results using all stations with ozone data
(in the first step, we could use only eight stations which had both ozone and
temperature measurements), we applied an additional method. We compared the
observed ozone–temperature correlation with the correlation between observed
ozone and modeled temperature. This was done to assess and confirm (together
with the meteorological model evaluation in Sect. 3.1) that the modeled
temperature was a good surrogate for the observed temperature in the
ozone–temperature correlation. In this way, we could apply this method to
all stations and evaluate the ozone production in the whole European domain.
It is difficult, however, to interpret the results when the evaluation is
performed for each station separately when the number of stations is large.
We displayed therefore all the calculated slopes of the ozone–temperature
linear fit for both observations and model in a single scatterplot. In
this way, the illustration and interpretation of the modeled ozone production
evaluation for whole domain became simpler. In addition, for more consistent
results, two filters were applied in the method above: (i) we only included
days with afternoon mean temperature higher than or equal to 15 In order to have a more rigorous model evaluation of the ozone production
without the influence of day-to-day variation and local meteorological
conditions, we also applied a binned data analysis in the ozone–temperature
correlation as also used by Bloomer et al. (2009). We divided the modeled
temperature into four bins with 5
In order to characterize the sensitivity of the modeled ozone production to its main drivers, various emission and meteorological sensitivity tests were performed (see Table 3). These tests were based on the emission uncertainties that were discussed in Sect. 2.3 as well as the meteorological uncertainties of this study such as temperature and wind speed underestimation and overestimation of low wind speed, which are quite common in modeling studies (Solazzo et al., 2013, 2017; Im et al., 2015; Bessagnet et al., 2016).
Model performance evaluation for the meteorological parameters in summer (JJA) 2010.
Diurnal profiles of surface O
Model performance evaluation for the daily mean concentrations of
the chemical species in summer (JJA) 2010. The units for MB, MGE and RMSE are
in ppb for the gas species and in
The meteorological model results show a good agreement with the surface
observations for 1051 stations (Table 4) and meet the performance criteria
(Table 2) suggested by Emery et al. (2001). Only the mean gross error (MGE;
see Table 1 for definitions) for the wind direction is slightly off by
5
Diurnal profiles of the surface NO
The overall model performance for the daily mean concentrations of the air
pollutants in summer (JJA) 2010 (Table 5) was reasonably good. The
statistical evaluation results for most chemical species were in line with
those reported for various models and parameterizations for summer periods in
Europe (Bessagnet et al., 2004, 2016; Solazzo et al., 2012a, b; Nopmongcol et
al., 2012; Giordano et al., 2015). Model performance goals and criteria for
O
Scatterplots of modeled vs. observed surface afternoon
(12:00–18:00 UTC) mean O
The diurnal profiles of O
Scatterplots of modeled vs. observed surface afternoon
(12:00–18:00 UTC) mean NO
Scatterplots of surface afternoon (12:00–18:00 UTC) mean O
Mean bias of the afternoon (12:00–18:00 UTC) surface O
In order to investigate the afternoon O
As the afternoon ozone mixing ratios are strongly related to ozone
production, we made use of the ozone–temperature correlation (as discussed
in Sect. 2.5) to examine the modeled ozone production performance. The
regression between surface afternoon mean ozone mixing ratio and temperature
for eight stations is shown in Fig. 7. Three cases are shown: (i) observed ozone
mixing ratios against observed temperature, (ii) observed ozone mixing ratios
against modeled temperature and (iii) modeled ozone mixing ratios against
modeled temperature. For all cases, a strong linear correlation of ozone with
temperature with an upward trend is evident, except for the Nice (FR) station
where ozone stays constant with increasing temperature. A comparison of the
ozone–temperature correlation for the first two cases (black and red colors)
shows that the modeled temperature can be used consistently as a surrogate
for the observed one and can therefore be paired with the observed ozone
mixing ratios. For the third case (blue color), the upward trend of the
ozone–temperature correlation is less steep compared to the other two cases.
This is mainly due to the underestimation of the high ozone mixing ratio
values (
In general, the use of daily means and diurnal profiles for the model performance evaluation may conceal hidden biases as shown above. Especially for a chemical species like ozone, which is greatly influenced by both the meteorology and its complex non-linear chemistry, a model evaluation should be carried out for hourly values to increase the evaluation's consistency but also to better examine and understand the physical and chemical processes leading to the modeled values. Regarding the ozone production, the use of the afternoon ozone–temperature correlation indicated an underestimation of the model, but it was limited to eight stations only. In the next sections, we employ the rest of the methods discussed in Sect. 2.5 on all stations to better evaluate both qualitatively and quantitatively the model afternoon ozone mixing ratio and production, and apply various sensitivity tests to investigate the sources of error.
Figure 8 shows the mean bias in the modeled afternoon ozone mixing
ratios as a function of measured ozone mixing ratio bins (as discussed in
Sect. 2.5) for the base case as well as for four emission scenarios described
in Table 3. The trend of the model bias for the base case is very similar to
the one in Fig. 5: in all regions, afternoon ozone mixing ratios higher than
or equal to 50 ppb are underestimated (3–17 ppb) and this underestimation
increases with the mixing ratio. In the PV region, which has the
largest number of measurement data in the highest mixing ratio bin (
Scatterplots of the modeled vs. observed surface afternoon
(12:00–18:00 UTC) mean O
The model's response to increased VOC emissions
(1.5–2 VOC scenario; Table 3) is relatively weak for most of the regions
except for the MD, PV and BX regions (Fig. 8) with the largest effect of
A larger impact on the ozone mixing ratios
(negative bias improved by
The combined increase of both NO
In general, the PV region exhibits the highest sensitivity to emissions due
to its location, and the model prediction for ozone is generally improved
with the increased NO
Afternoon (12:00–18:00 UTC) surface O
We analyzed the slopes of the regression
lines from the ozone–temperature correlations using the second approach, as
described in Sect. 2.5. The modeled slopes are displayed as a function of
observed slopes for each region and for each emission scenario in Fig. 9. For
the base case, the model underestimates the ozone–temperature slope by about
a factor of 2 or more for most stations in all regions apart from the BI and
SC regions (light blue and purple colors, respectively), where the most
stations are close to the 1 : 1 line. The underestimation of the slopes is
more evident for the IP and MD regions (yellow and pink colors, respectively).
For the MD region, despite the underestimation of the high ozone mixing
ratios, the model also overestimates the low ozone mixing ratios more
significantly than for other regions (see Fig. 8) and this will consequently
influence the trend of the ozone–temperature regression. Increasing the VOC
emissions (1.5–2 VOC scenario) does not change the picture compared to the
base case with the exception of improvement in the BX region (red color),
which is consistent with the results shown in Fig. 8. On the other hand, the
scenarios with increased NO
Mean bias of the afternoon (12:00–18:00 UTC) surface O
The ozone–temperature correlation was also investigated from a different
perspective smoothing out the station-to-station variation by making use of
the third approach discussed in Sect. 2.5. The results are shown in Fig. 10
where the 15
Meteorology affects ozone mixing ratios not only directly (horizontal
advection, vertical diffusion, photolysis rates, etc.) but also indirectly by
influencing the concentrations of its precursors and its chemistry.
Therefore, we performed some tests to explore the impact of key
meteorological parameters like temperature and wind speed. The PBLH is
another meteorological parameter that can have a strong influence on ozone
mixing ratios, but its impact is very complex and can have opposite effects.
Increased vertical mixing dilutes the ozone precursors inhibiting ozone
production, but it also reduces the NO
A temperature increase of 4
Afternoon (12:00–18:00 UTC) surface O
The processes that are mainly influenced by a change in wind speed are
advection, horizontal diffusion and dry deposition. As a result of this
complex effect, the impact of the wind speed on ozone mixing ratios (blue and
purple colors) is less systematic than that of the temperature with its
correlation with the region's air pollution. More specifically, in the less
polluted SC region, the effect of the wind speed reduction is
lower in the highest bins (
A closer look to the influence of meteorology on the ozone production is
shown in Fig. 12, with the use of the ozone–temperature correlation. Since
the impact of meteorology on ozone mixing ratios was thoroughly examined in
Fig. 11, the focus of Fig. 12 is more on the effect of meteorology on the
correlation of ozone with temperature. It has to be noted that the ozone
mixing ratio for the
Overall, the meteorological scenarios that were tested did not improve the
modeled ozone performance as consistently as some of the emission scenarios.
The behavior of the modeled wind speed biases, i.e., overestimation of the
lowest wind speed and underestimation of the rest (Fig. S5), can explain to
some degree the overestimation of low ozone mixing ratios and the
underestimation of the high ones, but not entirely since the tested wind
speed uncertainties are higher than the real ones which are indicated by the
model performance evaluation (Table 4, Figs. S5, S7, S10). The temperature
sensitivity test had a smaller impact than the one of the wind speed, and
also the tested change (
Many studies (Katragkou et al., 2010; Solazzo et al., 2013; Giordano et al.,
2015; Im et al., 2015) have reported a strong influence of the boundary
conditions on ozone mixing ratios, but their impact is less significant near
the surface and inside the PBL, as well as in summer compared to other
seasons (winter or autumn), due to more dominant near-surface effects (e.g.,
photochemistry, emissions, transport, dry deposition). Furthermore, Katragkou
et al. (2010) showed that the impact of increased O
In this work, we used alternative methods to evaluate the modeled surface
afternoon ozone mixing ratios and production more consistently in the whole
European domain for the summer of 2010 using the regional air quality model
CAMx. The results were analyzed in eight European regions. The separation of
the observed surface ozone mixing ratios in bins helps to unmask the hidden
model bias and identify the significant underestimation of high mixing ratios
and overestimation of the low ones. Since the high surface ozone mixing
ratios are more related to photochemical ozone production, an evaluation of
the modeled ozone production was carried out using the ozone–temperature
correlation. The use of the modeled temperature as a surrogate for the
observed one (after the validation of this hypothesis) allowed us to perform
the modeled ozone production evaluation for most of the stations in the whole
European domain. As an additional, alternative approach to the
ozone–temperature correlation, we divided the modeled temperature into bins
and paired it to the respective observed and modeled surface ozone mixing
ratios. The results indicated that the modeled surface ozone mixing ratios
have a less steep increase with temperature than the observed ones. The
modeled ozone–temperature regression slope (ppb
Increasing just the VOC emissions by a factor of 1.5 and 2 for the
anthropogenic and biogenic emissions, respectively, resulted in a small
increase of surface ozone mixing ratios (1–2 ppb) across all observed ozone
mixing ratio bins for most of the regions except the MD, PV and BX regions where the impact was higher (2–4 ppb).
On the contrary, the doubling of only the NO
Both sensitivity tests with increased temperatures by 4
The results obtained in this study indicate that the uncertainties in
emissions (especially the too-low traffic NO
All data are available upon request from the corresponding authors. References to the repositories of the observational data used have been also provided in Sect. 2.
The supplement related to this article is available online at:
The authors declare that they have no conflict of interest.
We would like to thank the following agencies for preparing the datasets used in this study: TNO for the anthropogenic emission inventory; the European Environmental Agency (EEA) for the air quality data; the European Centre for Medium-Range Weather Forecasts (ECMWF) and British Atmospheric Data Centre (BADC) for the meteorological data; the United States Geological Survey (USGS) for the land use data; the World Ozone and Ultraviolet Data Centre (WOUDC) and its data-contributing agencies for the ozonesonde profiles; the University of Wyoming for the radiosonde profiles; the National Aeronautics and Space Administration (NASA) and its data-contributing agencies (NCAR, UCAR) for the TOMS and MODIS data, the global air quality model data and the TUV model. Calculations of meteorological data were performed at the Swiss National Supercomputing Centre (CSCS). Our thanks extend to Ramboll Environ, and especially Cristopher Emery, for their continuous support of the CAMx model. Finally, we would like to thank the two anonymous referees and the co-editor Tim Butler for their constructive comments that helped to improve our manuscript. This work was financially supported by the Swiss Federal Office of Environment (FOEN). Edited by: Tim Butler Reviewed by: two anonymous referees