Source attribution of European surface O3 using a tagged O3 mechanism

Tropospheric ozone (O3) is an important air pollutant that affects human health, ecosystems, and climate. The contributions of O3 precursor emissions from different geographical source regions to the O3 concentration can help to quantify the effects of local versus remote transported precursors on the O3 concentration in a certain area. This study presents a “tagging” approach within the WRF-Chem model that attributes O3 concentration in several European receptor regions to nitrogen oxides (NOx) 5 emissions from within and outside of Europe during April-September 2010. We also examine the contribution of these different precursor sources to various O3 metrics and their exceedance events. Firstly, we show that the spatial distributions of simulated monthly mean MDA8 from tagged O3 sources regions and types for late spring, summer and early autumn 2010 varies with season. For summer conditions, O3 production is dominated by national and intra-European sources, while in the late spring and early autumn intercontinental transported O3 is an important contributor to the total O3 concentration. We have also 10 identified shipping activities in the Mediterranean Sea as an important source of O3 for the Mediterranean countries, as well as the main contributor to high modelled MDA8 O3 concentration in the Mediterranean Basin itself. Secondly, to have a better understanding of the origin of MDA8 O3 exceedances, we compare modelled and observed values of MDA8 O3 concentration in the “Po Valley” and “Germany-Benelux” receptor regions, revealing that the contribution from local sources is about 41 % and 38 % of modelled MDA8 O3 during the exceedances days respectively. By examining the relative contributions of remote 15 NOx sources to modelled and observed O3 exceedance events, we determine that model underrepresentation of long-range O3 transport could be contributing to a general underestimation of modelled O3 exceedance events in the Germany-Benelux receptor region. Thirdly, we quantify the impact of local vs. non-local NOx precursors on O3 production for each European receptor region using different O3 metrics. The comparison between mean, MDA8 and 95 percentile O3 metrics accentuate the importance of large contributions from locally-emitted NOx precursors to the high-end of the O3 distribution. When we 20 compare the vegetation and health metrics, we notice that the SOMO35 and AOT40 indexes exhibit rather similar behaviour, while the W126 index accentuates the importance of local emissions. Overall, this study highlights the importance of a tagging approach to quantify the contribution of local and remote sources to the MDA8 O3 concentration during several periods as well to different O3 metrics. Moreover, this method could be applied to assess different mitigation options.


Introduction
Tropospheric ozone (O 3 ) is formed primarily during the oxidation of volatile organic compounds (VOC) in the presence of nitrogen oxides (NO x ) and sunlight. Ground-level O 3 is an important air pollutant that damages human health (Fleming et al., 2018) and vegetation (Mills et al., 2018). It also affects the radiative forcing (e.g. Ramaswamy et al., 2001;Stevenson et al., 2013), and therefore contributes to climate change. Impacts of O 3 on human health are associated with lung disease, chronic 5 disease and death from respiratory ailments. To protect human populations from exposure to high levels of O 3 , the World Health Organization (WHO, 2006(WHO, , 2017 recommended an air quality guideline for ozone in which the maximum daily average 8-h (MDA8) for O 3 should not exceed 100 µg m −3 . The European Environmental Agency (EEA, 2017a) reported that the EU long-term objective target concentration of 120 µg m −3 is often exceeded and that more than 90 % of the urban population of the European Union was exposed to O 3 levels higher than the stricter recommendation set by the WHO. A 2010 report from 10 HTAP (HTAP, 2010) shows that the observed baseline O 3 concentrations (concentrations without the contribution from local anthropogenic emissions) have increased throughout the last several decades since overall global anthropogenic emissions of O 3 precursors have increased. However, a more recent study by Gaudel et al. (2018) has established that the global surface O 3 trends exhibit high variability, and depend on several factors such as season, region, elevation and proximity to fresh ozone precursor emissions. However, since the network capable of monitoring ozone levels is sparse, it is difficult to quantify the O 3 15 changes on a global scale. Satellite-derived O 3 measurements can be used to quantify changing levels of O 3 , but Gaudel et al. (2018) showed that these products are not capable of quantifying significant trends. Surface O 3 pollution due to urbanization and motorization processes are serious challenges for large cities (e.g. Chan and Yao, 2008;Folberth et al., 2015;Li et al., 2017Li et al., , 2019. Paoletti et al. (2014) showed that in Europe and the United States of America, the average O 3 concentration in the cities has increased at a faster rate than those observed in rural areas. Fleming et al. (2018) showed that the 4th highest 20 daily maximum 8-hour O 3 (4MDA8) is more ubiquitous at urban sites than at non-urban sites. This leads to a worsening of general air quality that, ultimately, affects human health and ecosystems (Paoletti et al., 2014;Monks et al., 2015;WHO, 2017;Fleming et al., 2018;Mills et al., 2018). To improve the air quality in certain areas, it is important to know the extent to which different precursors (NO x and VOCs) contribute to tropospheric O 3 concentrations.
Information regarding levels of NO x and VOC emissions and weather conditions enhances our ability to predict the forma- 25 tion of tropospheric O 3 . The continuous development of chemical transport models can lead to a better understanding of the processes that contribute to high-O 3 episodes. Knowing the impacts of NO x and VOC emissions from sources such as surface anthropogenic activities, fires, soil, and the stratosphere on total O 3 production can help authorities develop strategies aimed at reducing the impact of high levels of O 3 on the well-being of both humans and ecosystems. Several approaches have been used to determine the extent to which individual sources contribute to total levels of O 3 . For example, perturbation of different 30 emission categories have allowed scientists to make estimations regarding the contributions of individual sources of O 3 to total O 3 levels (e.g. Fiore et al., 2009).
Tagging techniques have also been used in modelling studies to determine source/receptor relationships and how individual sources of pollutants contribute to total pollution levels at given locations. Pollutants with relatively low chemical reactivities, such as carbon monoxide (CO), can be "tagged" according to their emission sectors or regions for attribution studies (e.g. Pfister et al., 2011). Sudo and Akimoto (2007), and Derwent et al. (2015) used O 3 tracers tagged by their region of formation to show that intercontinental transport of O 3 can occurring from polluted source regions, such as North America and East-Asia, appears to be the most important source of tropospheric O 3 in Europe. Other studies, including those of Wang et al. (2009) and Grewe et al. (2010and Grewe et al. ( , 2012and Grewe et al. ( , 2017 have used tagging methods to identify the contribution of individual sources of O 3 to overall levels. 5 This method is especially useful since it can track emitted NO x species during transport and chemical processing. Moreover, Grewe et al. (2012) showed the impact of the tagging method on mitigation measures, while Dahlmann et al. (2011) examined the contribution of O 3 sources to O 3 radiative forcing. Work by Emmons et al. (2012) and Butler et al. (2018) describe a procedure for tagging O 3 produced from NO x sources through updates to the MOZART chemical mechanism, and Butler et al. (2018) expanded the tagging technique to account for VOC sources. 10 Based on the work of Emmons et al. (2012), Pfister et al. (2013) and Safieddine et al. (2014) were able to use the WRF-Chem regional model to quantify the contribution of inflow (tagged O 3 and odd nitrogen species entering into the regional domain at the lateral boundaries) and of anthropogenic NO x precursors (named NO x in the following) on the surface O 3 levels. Using a slightly different methodology, Gao et al. (2016) have implemented within WRF-Chem framework a tagging method based on Ozone Source Apportionment Technology (OSAT) (Yarwood et al., 1996) incorporated in the Comprehensive Air quality 15 Model with extensions (CAMx).
Much effort has been focused understanding the origin of tropospheric O 3 and the key role played by the intercontinental transport, the contribution of stratospheric O 3 intrusion, and of different emissions sources to tropospheric O 3 concentration in a wide range of receptor regions. To better understand the changes in air pollution levels, it is necessary to know the relationship between levels of an emitted species and its atmospheric concentration. When this information is available, it is possible to 20 quantify the contribution of different emission precursor sources to overall O 3 concentration levels at a particular receptor location. For this purpose, we followed a strategy outlined in Emmons et al. (2012) and Butler et al. (2018) to implement a tagging technique into the regional WRF-Chem model. The model can be used to quantify source contributions to the tropospheric O 3 concentration, by "tagging" NO x emissions, and corresponding products so that they can be traced to the final production of O 3 . 25 When studying the effects of O 3 , the impact of the compound on humans and vegetation is of the utmost importance. Therefore, several exposure indexes have been defined to describe the relationship between O 3 and both human health and agricultural crop yield that are based on hourly averaged data. Musselman et al. (2006), Agathokleous et al. (2018), and Lefohn et al. (2018) review literature describing O 3 metrics. Additionally, a work by Paoletti et al. (2007) has provided a list of common O 3 exposure metrics used to assess risk to human health and vegetation. Here we use some well-known O 3 metrics, 30 such as MDA8, SOMO35, AOT40, and W126. The MDA8 index has been defined as the maximum daily average 8-h (MDA8) O 3 values (ppb) (Lefohn et al., 2018). SOMO35 (WHO, 2001) (EU directive 2008/50/EC, 2008 AOT40 metric is measured throughout daytime periods from May to July (growth season) and has a defined target limit of 18000 µg m −3 h (9000 ppb − hours) and a long term objective of 6000 µg m −3 h (3000 ppb − hours). A standard of 15 ppm − hours has been defined for the seasonal W126 index, which is averaged over three years. These metrics have been used to assess the impact of mitigation strategies (Avnery et al., 2013), the impact of industry on air quality management issues (Vijayaraghavan et al., 2016), and the impact of high O 3 5 levels and temperatures on crops (Tai and Val Martin, 2017).
In this paper, we use a tagged O 3 mechanism in the WRF-Chem model to understand the contribution of emitted O 3 precursors from different geographical source regions and types on the modelled O 3 concentration in several European receptor regions. In Section 2 we discuss the details of implementing this tagging technique, and describe changes made to both the chemical mechanism and WRF-Chem code. Section 2 also describes the WRF-Chem configuration, simulation design, and 10 input data used in the study. An analysis of the WRF-Chem simulation is presented in Section 3, while Section 4 summarizes our findings.
2 Model simulation

Tagging technique
To perform a WRF-Chem model simulation using a tagging approach, several changes must be implemented in the model code 15 to accommodate additional tracers and reactions representing tagged constituents. Butler et al. (2018) describes in detail how the tagging technique was implemented in the Community Earth System Model. The tagging technique used in this study is based on the same approach and uses the same modified version of the MOZART chemical mechanism. Further detail on how the chemical mechanism was extended can be found in Butler et al. (2018).
To use the NO x tagging mechanism, a new chemistry option was added in the namelist.input file: chem − opt=113 and 20 through the code. The coupling of the new chemical scheme with microphysics and radiative processes requires several modifications to the code: 1) The first step is to create a new chemistry option. The package mozart − tag − kpp (chemopt==113) has been added to~/WRFV3/Registry/registry.chem together with new model variables for tagged NO x species (e.g. O3 − X − INI, O3 − X − STR, etc). For this purpose, the pre-processing software described in Butler et al. (2018) was adapted to produce a new chemical mechanism; 2) The new chemistry package is a KPP option. Therefore, we created a new subdirectory in 25~/ WRFV3/chem/KPP/mechanisms/ directory containing the files (*.spc, *.eqn, *.kpp, and *.def) which defined the chemical model species and constants, chemical reactions in KPP format, model description, computer language, precision, and integrator.
The new chemistry option considers a large number of species and reactions; therefore we exceeded hard-coded limits that the KPP chemical preprocessor, version 2.1 (Sandu and Sander, 2006) allows. To overcome these limits, we increased 30 MAX − EQN and MAX − SPECIES in the header file gdata.h, located in~/WRFV3/chem/KPP/kpp/kpp-2.1/src. Further, we updated the subroutines in the~/WRFV3/chem directory to consider the new chemistry package. The modules that we modified are described in the Appendix.
Although WRF-Chem uses the Advanced Research WRF (ARW) dynamic core in this simulation which conserves mass and scalar mass (Grell et al., 2005), the tagged O 3 species are advected independently. Thus, numerical errors associated with the advection scheme led to gradients in the sum of tagged species concentration compared to the "real" concentration; therefore, the relationship between these variables is not conserved. Since the advection scheme fails to reproduce the expected solution (in which the sum of the tagged species concentration at each grid point must be equal to "real" concentration), we solve 5 this by fixing all undershoots and/or overshoots assuming that the sum of tagged species mass is proportional to the "real" concentration. This technique was also applied in Flemming et al. (2015), and Gromov et al. (2010).
Compared to Pfister et al. (2013) and Safieddine et al. (2014) work, the expanded tagging technique used in this study has the advantage that multiple tags can be defined in each model run.  Model (Iacono et al., 2008) for longwave and Goddard shortwave scheme (Chou and Suarez, 1994), the Yonsei University boundary-layer parameterization (Hong et al., 2006), and the Monin-Obukhov scheme for the surface layer (Jiménez et al., 2012). The initial and boundary conditions for meteorological fields are taken from the European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis. Anthropogenic emissions were obtained from the TNO-MACC III emission inventory for Europe (Kuenen et al., 2014). Because the model domain extends beyond the edges of the TNO-MACC III inventory, we 20 used for completion emissions from the HTAP V2 inventory (http://edgar.jrc.ec.europa.eu/htap − v2). Biogenic emissions were computed on-line using the Model of Emissions of Gases and Aerosols from Nature (MEGAN) model (Guenther et al., 2006).

Experimental setup
The biomass burning emissions are based on Fire INventory from NCAR (FINN) .
For this WRF-Chem simulation, the tagged MOZART chemical mechanism for NO x emissions (Butler et al., 2018) is used to represent the gas-phase chemistry. The photolysis rates were computed using the Fast Tropospheric Ultraviolet and Visible 25 (FTUV) Radiation Model (Tie et al., 2003;Li et al., 2005). The dry deposition was calculated following the Wesely (1989) resistance method, while the wet removal scheme for the tagged MOZART chemistry is based on Neu and Prather (2012).
NOx emitted by several source regions and types are tagged and explicitly tracked using additional tagged reactions and tracers. Thus, we follow the contribution to the total ozone concentration from each specific emission source and type, from both within and outside the European model domain. Table 1 summarizes tagged sourceis regions and types that are used in 30 this study. Using a division of source regions within the European model domain, 15 geographical source regions are specified in Table 1 and depicted in Figure 1. A similar division of European regions has been used by Christensen and Christensen (2007) and Otero et al. (2018) to address the main sources of uncertainty in regional climate simulations, as well as during the AQMEII project (i.e. Struzewska et al., 2015). Except for ALP, the source regions within the European domain are identical to receptor regions. Given the complex topography of the ALP source region, we split this region into two receptor regions: the Po Valley region and the high Alps (regions above 1500 m elevation).
To represent the impact of transported O 3 from different regions outside of the domain, we used chemical boundary conditions derived from the extended CAM-Chem version 1.2 global simulations. Butler et. al (in preparation)  The BIO, BMB, LGT, and STR source types are also included in the tagged chemical mechanism used in this simulation, but without including them into the division of source regions; we refer to these sources as "other global source types" from 10 here on. Ozone due to these other global source types can originate both from long-range transport from remote source regions through the lateral model boundaries as well as from precursor emissions within the European model domain.
For each receptor region, we analyse the impact of the NO x emissions coming from different source regions and types to the total O 3 concentration.

Ozone metrics 15
Using different metrics to assess the impact of O 3 , we can determine which precursor sources most highly influence the accumulation of O 3 in different receptor regions, and thus to provide insights into which type of mitigation measures will be useful for a particular geographic area. These metrics include the mean O 3 concentration, the mean of MDA8, the cumulative exposure to mixing ratios above 35 ppb (SOMO35) (Colette et al., 2012), and the 95 th percentile for surface O 3 . Neither the impact of O 3 exposure on trees, plants and ecosystems (W126) (Lapina et al., 2014), nor the AOT40 accumulation metric (the 20 threshold is 40 ppb) were used to assess risk to vegetation from O 3 exposure (UNECE, 2010).
The European Air Quality Directive (EU directive 2008/50/EC, 2008) specifies that O 3 exposure should remain below a target MDA8 O 3 value of 120 µg m −3 , which can be exceeded up to 25 days per calendar year averaged over three years. The modelled daytime AOT40 (during local daylight hours 8 AM -7 PM) was calculated according to Equation (1).
(1) 25 According to European legislation (EU directive 2008/50/EC, 2008), the AOT40 metric is accumulated over the daytime period from May to July (growth season) and it has a defined target limit of 18000 µg m −3 h (9000 ppb − hours) and a long term objective of 6000 µg m −3 h (3000 ppb − hours). W126, however, is described according to U.S. EPA regulations (https://www.gpo.gov/fdsys/pkg/FR-2015-10-26/pdf/2015-26594.pdf). A standard of 15 ppm − hours is defined for the seasonal W126 index, which is an average over a three-year period. For this study, the hourly surface O 3 tagged outputs for April through September are used to calculate the highest 3-month W126 index values (see Eq. 2): According to Lefohn et al. (1988) 3). For this metric, the EU air quality directives do not prescribe a limit or a target values.
The contribution of tagged O 3 is based on formulations of each metric and is calculated from the model output. In the case of the MDA8 and 95 th percentile metrics, we searched for the specific period in which calculated values of total O 3 concentration meet the requirements for the formulation of these metrics. Once this is identified, tagged O 3 concentrations are extracted 10 for the same period which can then be used for further analysis. However, the contribution of concentration of tagged O 3 on cumulative metrics is slightly different, a large proportion of each tagged species is used to determine total O 3 , as illustrated below for AOT40 at a specific time period: Based on their formulation, we grouped metrics into either non-cumulative (mean O 3 , MDA8, and the 95 th percentile) or 15 cumulative (SOMO35, W126, and AOT40) categories. Since the latter metrics have different formulations (including hourly O 3 values above a specific threshold) and do not cover the same periods, to facilitate a more direct comparison between findings from multiple O 3 metrics, an analysis of the relative contribution of different source regions to total O 3 in each receptor region was performed. This was done using averaged values for non-cumulative metrics, and 6-month sums for SOMO35, AOT40, for cumulative metrics useful for evaluating effects on crops (cumulated over May-July period) and a maximum of 3-month 20 sums for every consecutive 3-month period determined using the W126 index.

Results and discussions
Our discussion of the results of the model is focused on the April-September 2010 period. We first briefly evaluate the ability of WRF-Chem to reproduce meteorological parameters using measurements from the Global Weather Observation (GWO) dataset provided by the British Atmospheric Data Center (BADC), and observed O 3 concentrations using the measurements included 25 in AirBase, a European air quality database (EEA, 2017b). We then provided a more detailed analysis of the contribution of different source regions and types to MDA8 values describing total O 3 for the analysed period.

Evaluation of meteorology and chemistry
Since the accurate simulation of meteorological parameters represents a key factor affecting the concentrations of trace gases, we briefly compare the modelled mean sea level pressure (MSLP), 2 m temperature (T2M), 10 m wind speed (WS10M) and direction (WD10M) variables against GWO measurement. Predicted model variables were then evaluated against observations using statistical scores that include normalized mean bias (NMB), and the correlation factor between simulated and measured 5 values (r).
An extensive evaluation of WRF-Chem using the MOZART chemical mechanism to predict long term meteorological data and O 3 levels has been presented previously (Mar et al., 2016). The main differences between the set-up used in this study and the model described by Mar et al. (2016)  Due to the coarse resolution of our domain, the air parcel dynamics associated with the complex topography of mountainous 15 areas was not properly reproduced. Thus, we assessed the ability of the model to reproduce the meteorological variables using only those sites located below 1500 m above sea level. MSLP data were well reproduced over the entire period (NMB = 0 % and r = 0.98). The model predicted T2M values well (r = 0.91), however temperature was underestimated by 3 % (see Table 2).
WS10M was also fairly well reproduced both in terms of spatial and temporal variability (NMB = 8 %, r = 0.63). Yet, WD10M data could not be predicted as well as other meteorological variables (NMB = 13 %, r = 0.47), behaviour could be related to 20 the existence of unresolved topography features (Jimenez and Dudhia, 2012). However, the model performance is similar to Mar et al. (2016) and Tuccella et al. (2012).
We also compared modelled MDA8 O 3 concentrations with observations provided by the publicly-available AirBase dataset.
The relatively coarse resolution of the domain may not be representative of changes in local emissions when the measurements are taken from urban areas; therefore, to aid in the analysis, we used only those stations characterized as rural. As can be seen in 25   Table 3, evaluation of the model over entire period revealed that the model performs quite well with respect to the prediction of concentration and temporal evolution. Mar et al. (2016) reported a mean bias (MB) value of 15.85 µg m −3 and an NMB of 17% for the June-August 2007 period when the MOZART mechanism was used to assess the chemical performances of the model.
For the same time period, we obtained an MB value of -5.92 µg m −3 and an NMB value of -6.3%. Tuccella et al. (2012) reported an annual MB of -1.4 µg m −3 when the RADM2 chemical mechanism was used to simulate a period throughout . Errors of the model may be explained by a wide range of uncertainties related to modelled physical and chemical processes such as grid resolution, vertical and horizontal transport, boundary layer mixing, emission inventory, chemistry and photolysis rates, dry deposition, wet scavenging, etc. It is also possible that uncertainties in measurements contribute to observed errors. Since the focus of this study is on the contribution of different sources of precursors to the total tropospheric O 3 concentration of a particular area, a more thorough analysis of the ability of the model to reproduce the observed meteorological variables is beyond the scope of this paper.  Table S1). O 3 from RST (a 7.5 -15 % contribution) is the main source from overseas. O 3 from shipping NOx  Table S1). Emissions from local sources do not only affect local O 3 mixing ratios, but also impact O 3 levels of bordering countries due to strong horizontal pollution transport. In all receptor regions, local anthropogenic sources have a lower contribution to MDA8 O 3 mixing ratios than the sum of O 3 due to anthropogenic sources in other European source regions and long-range 20 transport of ozone from intercontinental source regions. The contribution of intercontinental transport to the total MDA8 O 3 mixing ratio in Europe is consistent with previously reported results, i.e. Fiore et al. (2009) and Karamchandani et al. (2017), while this study allows us to identify which anthropogenic sources exert a strong influence on MDA8 O 3 predicted in different regions. Using observations, Danielsen (1968), Thouret et al. (2006) showed that the transport of O 3 from the stratosphere also contributes to tropospheric O 3 . Here, stratospheric O 3 contributes up to 7 ppb (12.5 % in SCA) to the total MDA8 O 3 mixing 25 ratio, which is a finding similar to that reported by Derwent et al. (2015). A similar tagged system for predicting O 3 levels, using the CAM-Chem model (Butler et al., 2018), has also shown that stratospheric O 3 significantly contributes to the total tropospheric O 3 mixing ratio. The MOZART chemical mechanism used in this study does not explicitly treat stratospheric chemistry; thus surface stratospheric O 3 could be attributed to the vertical and horizontal transport of stratospheric O 3 and stratospheric tagged precursors species concentrations coming from the CAM-Chem extended model that enters the domain 30 through lateral boundaries.
During June-August 2010, Western Europe was mostly influenced by a high-pressure system centered over the Atlantic (see Fig. S1). In the upper troposphere, a ridge influenced the vertical atmospheric structure, especially over southern Europe.
Therefore, these "usual summer conditions" favoured the intrusion of warm air coming from Africa and the Arabian peninsula and led to a warm and dry climate characterized by subsidence, stability, clear skies and high-intensity solar radiation. Hence, the photochemical formation of O 3 was enhanced, and influenced the stronger contribution of local emissions to the total mixing ratio compared to the previous period examined. Figure 3 depicts the average MDA8 O 3 for June-August 2010. For most regions, we notice that levels of O 3 produced from local sources from June-August compared with April-May were enhanced ( Figure 2). Local sources can contribute to more than 20% of the mean MDA8 O 3 mixing ratio (from 14.6 % in SCA 5 to 35.7 % in the Po Valley, see Table S1). This shows that local sources play a strong role in the formation of O 3 throughout the June-August period, as has been previously shown by Jiménez et al. (2006) and Querol et al. (2018). Compared with late spring, the relative contribution of overseas sources decreased in summer, varying from 10.9 % in the Po Valley receptor region to 44.8 % in the UKI region in the month of July (Figs. 2 and 3; Table S1). We noticed the spread of O 3 produced from European anthropogenic precursors over bordering regions compared with late spring 2010 (Figs. 2 and 3). The increase in The decrease in photochemical activity in September 2010 is reflected in decreases in total O 3 mixing ratios compared with 20 the summer of the same year, as well as in a reduction associated with the local source contribution to the total O 3 mixing ratio ( Fig. 4). Thus, only in IBE, TCA, FRA, Po Valley, the high Alps, and RBU regions was contribution of local sources to total MDA8 O 3 higher than 20 % (Table S1). On the other hand, we noticed an increase in O 3 coming from anthropogenic overseas sources and from lightning in autumn, stressing that seasonal variations exist within the outflow from other continents. There also is variation in the lifetime of O 3 which is shortest during the summer as a result of enhanced photolytic activity. 25 Although we have seen that long-range transport plays a major role in total O 3 mixing ratios, the tagging technique helps to gain more insight into which region of the world dominates these mixing ratios in spring or autumn. In early fall, the Western European receptor regions exhibit a slight increase of 1.6 % in O 3 mixing ratios coming from North America compared with spring, while the contribution of O 3 mixing ratios coming from other overseas sources decreases. This could be linked to the prevailing westerly wind and the synoptic conditions seen during the first period of September, when the Azores High extended 30 far to the east and north (Fig S1). This phenomenon creates conditions that are conducive to the transatlantic transport of

Tagged ozone precursor contributions to exceedances of MDA8 target values -case study
As previously mentioned, the European Air Quality Directive (EU directive 2008/50/EC, 2008) has defined a target value of 120 µg m −3 for the MDA8 O 3 concentration, which can be exceeded up to 25 days per calendar year (over a three-year span). In the following, we refer to values that surpass 120 µg m −3 as exceedances, and values below 120 µg m −3 as nonexceedances. Figure S2 shows the spatial distribution of the number of exceedances observed and calculated throughout the  were, in all cases, taken at the location of the measurement stations, throughout the April-September 2010 period. Figure 6 shows the average conditions during the exceedance of the MDA8 O 3 target value, and also, at times, occurred when the target 5 value was not exceeded. To perform the source attribution for the observed values, we have scaled these values proportionally by the relative concentrations of each tagged O 3 tracer in our model output.
The relative contribution of emissions from different source regions to modelled and to observed MDA8 O 3 values, after being scaled to account for the contribution of modelled sources of O 3 types is generally similar for Po Valley and GEN receptor regions (see Fig. 6). In the Po Valley, we can pinpoint the main remote contributor as being MBS (see Fig. 6), followed by GEN, 10 and FRA, suggesting a dominant westerly and northerly air flow. The recirculation of air masses in the Gulf of Genoa could accentuate the sea breeze and therefore more O 3 coming from NO x associated with shipping activities in the Mediterranean will be transported to the coastal and inland station.
The high Alps receptor region is less influenced by ALP emissions than the Po Valley, and is more influenced by remote sources (see Fig. 6). The increased contribution of O 3 from CEE, ITA and FRA to both exceedance and non-exceedance days in 15 the high Alps receptor region compared with the Po Valley receptor region highlights the impact of the transboundary transport of O 3 and its precursors . Furthermore, the contribution of stratospheric as well as long-range sources was generally 6 % higher in this receptor region than in the Po Valley receptor region.
In GEN, the main remote source regions are FRA and CEE during the exceedance days and FRA and UKI during nonexceedance days (Fig. 6) 6). 25 This kind of analysis can be applied to improve our knowledge of the origin of O 3 precursors and their contribution to MDA8 O 3 health metrics. Hence, by using this tagging technique, policymakers can identify future actions required to control the NO x emissions at local and regional levels.

Tagged ozone precursor contributions to regulatory ozone metrics
In this section, we discuss the contribution of O 3 mixing ratios from diverse emissions sources and types to several metrics  Splitting the non-cumulative metrics into early (April-June) and late (July-September) simulation periods clearly illustrates that the European receptor regions are more prone to be influenced by intercontinental transport during the early period than the late period. The contribution of intercontinentally transported O 3 to mean O 3 values in different receptor regions is higher 10 during the early period and it spans between 22.8 % and 54.3 % of total O 3 . In the late period it accounts for between 16 % and 48.9 % of total O 3 (see Fig. 7 and Table S2). Since O 3 associated with intercontinental transport comes, in this case, solely from boundary conditions, errors in boundary conditions affect the predicted mixing ratio of various chemical species, and, consequently, the contribution of overseas sources of O 3 to levels observed in Europe O 3 (Tang et al., 2007;Giordano et al., 2015;Im et al., 2018). 15 The shorter lifespan of O 3 over remote ocean regions throughout the warm season, combined with synoptic conditions, has led to decreased levels of intercontinentally transported O 3 to Europe. Thus, for most receptor regions, the O 3 coming from Asia and the rest of the world was reduced by more than half when compared with the cold period. The O 3 mixing ratio from the stratosphere is, in general, 2.5 times higher in the cold season than in the warm season which is consistent with the findings of a study by Butler et al. (2018) which showed that the stratospheric O 3 mixing ratio varies with altitude and its lifetime is Even though the SOMO35 and AOT40 metrics are not accumulated over the same period (SOMO35 is accumulated over the entire simulated period, and AOT40 metric is accumulated over the May-July period) and do not use same input data (daily 30 MDA8 O 3 for SOMO35 vs daytime O 3 mixing ratios for AOT40), since they are based on threshold exceedances and are designed to measure exposure to high O 3 levels of humans (SOMO35) and vegetation (AOT40), there is a way to directly compare data from each metric type. As shown in Figure 7 and Table S2, the contribution of different sources of emissions and types as a proportion of total SOMO35 and AOT40 metrics is similar for most of the European receptor regions. Their spatial distribution (not shown) is also comparable, with minimum values over the UK, NW Europe and Scandinavia and maximum 35 13 values over Italy, the Alps, south of Spain, east of Turkey and in the metropolitan area of Moscow, Russia. These results are consistent with previous studies performed by Aksoyoglu et al. (2014), andAnav et al. (2016). The overseas sources contribute similarly to SOMO35 and AOT40 indexes (usually less than 30 %) for most of the receptor regions used in this study. However, in UKI the overseas sources account for 32 % of AOT40 and 38 % of SOMO35, and in SCA they contribute to~22 % of AOT40 and 30 % of SOMO35. This suggests that these metrics are more sensitive with respect to the O 3 mixing ratios from remote 5 sources in areas having a low level of O 3 pollution. In the RBU receptor region, these indicators are sensitive to O 3 coming from biomass burning emissions (20 % of SOMO35 and 24 % of AOT40), whereas for the remaining receptor regions the contribution of natural sources to SOMO35 and AOT40 is similar. Local sources account for a range of~12 % (SCA) to~38 % (GEN) of these metrics. These data highlight the occurrence of increased O 3 production from local sources in comparison with northern European countries as well as large emissions of NO x in the GEN source region. Since the difference between 10 AOT40 and SOMO35 is only a few percentage points, regardless of the receptor region, we were able to conclude that they behave similarly, according to thresholds used to define these metrics.
The tagging method allows a better understanding of the main precursor sources responsible for exceedances of regulatory  (Table S2). Generally, the O 3 mixing ratio and its precursors transported from other anthropogenic European  index compared with the other metrics for most of the downwind receptor regions. This behaviour is related to how these 30 metrics have been defined. Due to its sigmoidal weighted formulation, as discussed in Westenbarger and Frisvold (1995), and Lapina et al. (2014), W126 includes all daytime values rather than O 3 levels above a certain threshold, as is done using SOMO35 and AOT40; therefore lower weighting factors of less than 0.5 are given to low O 3 values and weighting factors above 0.5 are given to O 3 values situated above the inflection point of 67 ppb.
The modelled mean AOT40 and W126 values in the Po Valley receptor region exceeded standards (26368 ppb − hours for AOT40 and 28.9 ppm − hours for W126) during the May-July 2010 period, and, as shown in Fig. 7 and index would require values be expressed in ppb − hours. Further, all metrics showed a similar level of temporal variation in which they peaked in the first half of July. Also, whenever the averaged O 3 mixing ratio was lower than 60 ppb (Fig. 8a), W126 value was lower than AOT40 (Fig. 8d). This way of acting was most probably due to the weighting factor being less than 0.3, and above this mixing ratio W126 tends to be higher than AOT40. This behaviour is closely linked to the definition 10 of these metrics. If the O 3 mixing ratio is less than 40 ppb, W126 has a weighting factor lower than 0.03, while AOT40 has a weighting factor of 0. Above this threshold, AOT40 has a weighting factor of 1, while in the case of W126 only O 3 values higher than 100 ppb have a weighting factor of 1. Due to the way these metrics are defined, predicted O 3 values in each grid cell are accounted for the W126, may not be accounted for the AOT40 index.
In addition, visual analysis of the time series also revealed that when the O 3 mixing ratios from local sources are~20 ppb, 15 these mixing ratios have a higher contribution to W126 than AOT40. To better understand this observation, we have further analysed the relationship between mean O 3 values from ALP sources (O 3 -ALP) and the percent contribution of these O 3 tracers to mean O 3 , W126, and AOT40 metrics. Figure 9 shows scatter plots for O 3 -ALP that relate the contributions of these mixing ratios on mean O 3 , W126, and AOT40. In addition, the linear regressions of Y vs X (Y=a*X+b) using all data sets have been applied. We saw that in general, high mean O 3 -ALP mixing ratios have a higher contribution to W126 than to AOT40; 20 this was also confirmed by the linear regression between O 3 -ALP and W126 that yields a slope of 1.52 compared to a slope of Extending this analysis to all receptor regions, we can explain why the W126 index is more sensitive to O 3 coming from local sources compared with the other cumulative metrics. In addition, W126 accentuates the contribution of BIO and BMB in RBU, TCA and SEE, most likely because the metric includes all daytime values, and not just those above a certain threshold.
Thus, the use of W126 highlights the considerable impacts of BIO and BMB emissions on total O 3 mixing ratios throughout the summer and from burning vegetation that ultimately influence the extent to which O 3 causes damage to vegetation. 30 We have seen that the contribution of NO x to total O 3 varies depending on metrics and regions considered. Hence, the tagging method could help design different emission control strategies in specific source regions depending on which impacts need to be reduced in specific receptor regions.

15
Here, we implemented a new chemical mechanism within the WRF-Chem model to account for source attribution of O 3 from NO x . We investigated the origin of surface O 3 using the "tagging" technique from April-September 2010, as well as the contribution of different sources to O 3 metrics, and their exceedance events. The modification introduced and described in Section 2 as well as the model data can be provided upon request to the corresponding author. LGT lightning STR stratospheric O3