ANALYSIS OF SUMMER O3 IN THE MADRID AIR BASIN WITH THE LOTOS-EUROS CHEMICAL TRANSPORT MODEL

Tropospheric O3 remains a major air-quality issue in the Mediterranean region. The combination of large anthropogenic emissions of precursors, transboundary contributions, a warm and dry aestival climate and topographical features results in severe cases of photochemical pollution. Chemical transport models (CTMs) are essential tools for studying O3 dynamics and for assessing mitigation measures but they need to be evaluated specifically for each air basin. In this study, we present an optimisation of the LOTOS5 EUROS CTM for the Madrid air basin. Five configurations using different meteorological datasets (from the European Centre for Medium Weather Forecast (ECMWF) and Weather Research and Forecasting (WRF)), horizontal resolution and number of vertical levels were compared for July 2016. LOTOS-EUROS responded satisfactorily in the five configurations reproducing observations of surface O3 with notable correlation and reduced bias and errors. However, the best-fit simulations for surface O3 were obtained by increasing spatial 10 resolution and using a large number of vertical levels to reproduce vertical transport phenomena and the formation of reservoir layers. Using the optimal configuration obtained in the evaluation, three characteristic


Introduction
Ozone (O 3 ) is formed in the troposphere by the interaction of gaseous precursors like nitrogen oxides (NO x ) 10 and volatile organic compounds (VOCs) in the presence of sunlight. Much attention has been given to this secondary air pollutant in the last decades due to the variety of negative effects on health, ecosystems, crops, climate and materials associated with it (see review by Monks et al. (2015) and references therein).
The oxidative effect of O 3 generates inflammation of airways. Increases in morbidity and mortality and chronic alterations of the cardiovascular and cerebrovascular systems have also been associated with expo- 15 sure to O 3 (WHO, 2006(WHO, , 2013a. Tropospheric O 3 is also harmful for vegetation, generating leaf symptoms, reduced growth, senescence, defoliation and reducing crop productivity (Paoletti, 2006;WGE, 2013).
Damage to construction materials like plastics, surface coatings and rubber due to O 3 has been documented (Lee et al., 1996;Screpanti and Marco, 2009). Moreover, O 3 in the troposphere acts as a greenhouse gas with positive global radiative forcing (IPCC, 2013). 20 It is estimated that 98% of the urban population in Europe in 2016 were exposed to excessive concentrations of tropospheric O 3 according to the World Health Organization (WHO) guidelines values, a steady proportion since 2000 (EEA, 2018). However, it is the Mediterranean basin where the most acute episodes are registered (Millán et al., 1997(Millán et al., , 2000Sicard et al., 2013;Querol et al., 2016). In the Iberian Peninsula (IP), located at the Western Mediterranean Basin, the intense solar radiation, high temperatures and lack of 25 precipitation in spring and summer, associated with persistent anticyclonic conditions, favour the formation of O 3 in the area and the accumulation in rural and suburban regions (Escudero et al., 2014Querol et al., 2016;Massagué et al., 2019). The emissions of precursors from anthropogenic sources in the Mediterranean basin and the surrounding regions are considerable, especially in some densely populated areas. In addition to that, the amount of biogenic VOCs emitted in southern Europe is considerably higher than in 5 central and northern Europe (Seco et al., 2011). Moreover, during the frequent biomass burning episodes in summer, air-quality problems associated with tropospheric O 3 are aggravated (Tressol et al., 2008). In particular, the complex orography of the IP with mountain ranges running parallel to the coast intersected by river basins that penetrate towards the inner continental areas and elevated plateaus in the centre of the peninsula, help air masses to recirculate and age under the influence of sea and mountain breezes that develop when 10 synoptic circulation is inhibited by the presence of the Azores high (Millán et al., 2000;Gangoiti et al., 2001;Valverde et al., 2016;Querol et al., 2017). Previous studies also suggest that local strategies designed to meet NO 2 ambient air-quality standards may have caused an increase of urban O 3 that, in turn, causes an increase in oxidative capacity of Madrid's atmosphere by increasing OH and NO 3 radicals (Saiz-López et al., 2017).
In recent years, several comprehensive summer campaigns with intensive measurements of surface and 15 vertical profiles of O 3 concentrations and its precursors have been undertaken near the two main conurbations in Spain: Barcelona (2015Barcelona ( , 2017Barcelona ( and 2018 and Madrid (2016) (Querol et al., 2017Reche et al., 2018;Carnerero et al., 2018). The main objective of these campaigns was to interpret the phenomenology of high O 3 and ultrafine particles' episodes in Spain.
Another essential objective of retrieving data from these intensive campaigns is related to the valida- 20 tion and optimisation of chemical transport models (CTMs). These models constitute an essential tool for analysing O 3 behaviour with high spatial and time resolution, providing air-quality forecasts and supporting the design of policies. This includes the study the NO x -VOC sensitivity (Sillman, 1999;Sillman and West, 2009) essential for proposing and evaluating potential mitigation measures. Regional CTMs have been used to investigate O 3 pollution in Spain in several studies. Most of these studies aimed to describe short-term 25 (rarely exceeding 5 days) pollution events (Toll and Baldasano, 2000;Jiménez et al., 2005;San José et al., 2005;Jiménez et al., 2006;Carvalho et al., 2006;Valverde et al., 2016;Pay et al., 2018) and, in some cases, to discuss the effectiveness of potential mitigation strategies (Palacios et al., 2002;Soret et al., 2014). Despite these efforts, work is still needed to evaluate the impact of changes in the vertical configuration of CTMs, especially in the Mediterranean region where the atmospheric dynamics in summer is characterised by complex recirculation processes with effective vertical exchange (Millán et al., 1997(Millán et al., , 2000Gangoiti et al., 2001;Borge et al., 2010;Querol et al., 2017. The lack of an appropriate representation of the vertical variability of O 3 has been recognised as one of the shortcomings of the CTMs and in consequence a major challenge in the future development of the models (Hess and Zbinden, 2013;Monks et al., 2015). Moreover, 5 it is strongly recommended to combine modelling with observations because this will bring knowledge from both sources together and permit adequate evaluation procedures of the model outputs (Canepa and Builtjes, 2017).
Air quality model results vary at different resolutions especially due to the resolution of emissions and the description of the driving meteorology (?). Some authors have found that the impact of higher horizon-10 tal resolutions in O 3 simulations is more sensitive to the resolution of emissions than to meteorology (?).
Moreover, finer resolution result in less dilution of emissions but also in differences have been found in the O 3 -NO x interaction (??).
In the Iberian Peninsula, the use of fine grids (in the order of 1-5 x 1-5 km) has been found beneficial in the context of complex terrains where mesoscale processes acquire importance for interpreting production 15 and transport of O 3 (Toll and Baldasano, 2000;?). In coastal areas, with complex topography, high resolution simulations have been generally employed with good results (Carvalho et al., 2006;Jiménez et al., 2006;?). Moreover fine grids have been recommended for describing O 3 variability especially in urban and industrial areas (Jiménez et al., 2006;?). In general, the use of finer resolution simulations in the Iberian Peninsula generally imply benefits in the O 3 description such as improvement in correlation and reduction 20 in bias and errors (Jiménez et al., 2006). Less importance has been given to the vertical resolution mostly because the vertical O 3 profile evaluation of CTM is difficult due to the lack of experimental vertical O 3 data. In complex domains in the Iberian Peninsula the models may not reproduce O 3 concentrations due to a poor representation of mesoscale flows and layering and accumulation of pollutants (?). In general, it has been demonstrated that incrementing vertical resolution would help to resolve meteorological phenomena 25 (Carvalho et al., 2006) and would also offer a more realistic vertical exchange between the boundary layer and the free troposphere (Jiménez et al., 2006).
Making use of the results on the O 3 episodes phenomenology from the aforementioned field campaign in Madrid in July 2016, we were able to assess and optimise LOTOS-EUROS CTM v2.0 (Manders et al., 2017) for simulating O 3 in this region. Five configurations combining different meteorological input data and vertical structures were employed after identifying these two aspects as key factors for the capability of the model for reproducing O 3 levels. We simulated the entire month of July 2016 in accordance with the experimental campaign with a spin-up period of 24 h. The aim of this comparison was to elucidate the optimal configurations for operating with LOTOS-EUROS in the region but also to identify relevant factors 5 to set up other CTMs used in this region.
Moreover, employing the optimal configuration of the modelling system, we discuss the phenomenology of tropospheric O 3 in the Madrid air basin (MAB) for the study period. This was done by analysing simulated fields of meteorological variables and pollutants with special emphasis on the vertical variability to test the importance of the up-down transport of O 3 in the region.

Study area
The Madrid Metropolitan Area (MMA) is a densely populated area with more than 5 million inhabitants.
According to Salvador et al. (2015) and Borge et al. (2014), the main sources of pollutants in the region are road traffic, residential heating (which maximize their emissions in winter), a busy airport and minor 15 contributions from industry.
The MMA is located in the centre of the MAB and lies on an elevated plateau (∼700 m above sea level (a.s.l.)) in the middle of the IP (Figure 1). The climate in the area is continental Mediterranean with warm and dry summers and cold and also dry winters. The main orographic features surrounding the basin are, around 120 km to the south of the MMA, the Toledo Mountains (altitudes up to 1600 m a.s.l.) with an E-20 W axis and the Guadarrama range (maximum heights of 2400 m a.s.l.) which runs diagonally from SW to NE, 50 km to the west and north of the MMA. The Guadarrama range is part of the Central System that extends until the Ebro valley and, together with the western flank of the Iberian range delimits a channel to the NE along the Henares valley ( Figure 1). As a result of this configuration, the circulation in the Madrid basin shows a dominant SW-NE direction (Plaza et al., 1997). Under low-gradient synoptic conditions, the 25 combination of the strong convective conditions and the blocking effect of the mountain ranges induces an important vertical development of the boundary layer and mesoscale recirculation. During the night, north-easterly winds prevail over the basin and, after dawn, the eastern slopes of the Guadarrama range are progressively warmed up causing a clockwise turning of wind to an E and S during the day finalising with an SW component in the late afternoon. The drainage flows at night-time re-establish the north-easterlies.
These events are commonly referred as recirculation (REC) episodes. The presence of the Azores high or low pressure systems over the Atlantic in front of the Iberian Peninsula generate advection of Atlantic air masses 5 from the north (we will refer to these as Northern advective or NAD events) or from the south (Southern advective or SAD events).

The LOTOS-EUROS model
The 3D CTM LOTOS-EUROS v2.0 and its previous versions have been extensively used in the past for air-quality studies, including NO x (Schaap et al., 2013;Vlemmix et al., 2015), SO 2 ( Barbu et al., 2009) and 10 particulate matter (PM) (Schaap et al., 2004;Manders et al., 2009;Timmermans et al., 2017). In particular, tropospheric O 3 has been the scientific target in different studies carried out with successive versions of LOTOS-EUROS. It has been employed in health-related studies (van Zelm et al., 2008) and, more recently, Beltman et al. (2013) applied LOTOS-EUROS to simulate the response of tropospheric O 3 in Europe to a 5% shift from crop-and grassland into poplar plantations used for biomass production, while Hendriks 15 et al. (2016) tested the response to a decarbonisation scenario in the continent. Although LOTOS-EUROS has been generally employed in a continental domain (mainly in Europe), a sensitivity study to regional changes in emissions in three areas of Europe (Poland, the Po valley and Flanders) was also performed by Thunis et al. (2015). In addition, LOTOS-EUROS has also been used in a number of intercomparison studies with other CTMs for the simulation of O 3 (Hass et al., 1997;van Loon et al., 2007;Cuvelier et al., 20 2007;Vautard et al., 2007;Solazzo et al., 2012;Im et al., 2015) showing a satisfactory performance. Finally, regarding air-quality predictions, LOTOS-EUROS participates in the CAMS (Copernicus Atmosphere Monitoring Service) ensemble (Curier et al., 2012), which offers operational forecasts for NO 2 , O 3 and PM.

Model experimental design
A detailed description of the 2.0 version of LOTOS-EUROS can be found in its reference guide (Manders 25 et al., 2016) where all technical issues (processes, schemes, etc.) are described and referenced (accessible at www.lotos-euros.nl). In this section, we provide a brief description focusing on the most relevant aspects for this study.
Initial sensitivity studies were performed with the base configuration (configured similar to the operational forecasts that are part of the CAMS regional ensemble as presented in Marécal et al. (2015)) to test the response of the model to changes in the deposition velocity of O 3 because night-time dry deposition has 5 been suggested as a factor that could strongly influence the ability of CTMs to simulate tropospheric O 3 (Stevenson et al., 2006;Monks et al., 2015). The standard dry deposition velocity, calculated by the resistance approach (Manders et al., 2016) by a factor of 1.25 and 0.75. The results (not shown here) reflected a minimal effect of this parameter on O 3 concentrations in the chosen domain and period so deepening in this direction was discarded. 10 As shown in Table 1, two major aspects were modified in the set of five configurations: the meteorological input data and the vertical structure of the model. We fed LOTOS-EUROS with operational data from the reanalysis of the ECMWF model (Flemming et al., 2009) retrieved with a spatial resolution of 7 x 7 km 2 .
A second meteorological gridded dataset was obtained with the WRF model (Skamarock et al., 2008) with a resolution of 1 x 1 km 2 over a square domain of approximately 220 km of side centred on the city of 15 Madrid ( Figure 1). Data from WRF simulations with similar configurations have been previously used to drive air-quality simulations over the IP and, in particular, in the Madrid area (Borge et al., 2008(Borge et al., , 2014. In this case, the WRF model was run on a three -nested-domain configuration as shown in Figure S1. Additional information about the WRF model configuration is provided in Table S1. For the vertical structure, we compared the standard five-level mixed-layer configuration (Manders et al.,20 2017) with a hybrid-layer multilevel scheme. This version uses the lowest 70 layers of the 137 hybrid sigmapressure layers used by ECMWF for the operational meteorological forecasts in 2016. In such a vertical co-ordinate system, model layers are defined by pressure boundaries that follow surface pressure at lower altitudes but slowly evolve into fixed pressure levels in the stratosphere (Eckermann, 2009).
Finally, the MACC III emission inventory (reference year 2011) has been used for all set-ups and initial 25 and boundary concentrations were taken from global simulations produced by and used in CAMS services, as described in Marécal et al. (2015). These include concentrations of the most important trace gases and aerosols.

Monitoring data
Hourly O 3 data for the simulation period (July 2016) were collected from 35 air-quality monitoring sites (17 urban, 9 suburban and 9 rural) located in the MAB (Table 2 and Figure 1) and its boundary region. Traffic stations were in general discarded for the model evaluation due to their limited spatial representativity although four traffic sites were also employed. In spite of belonging to different air-quality networks ( Several previous studies used modelling techniques to analyse intense short-term O 3 episodes (Palacios et al., 2002;San José et al., 2005), and, more specifically, to evaluate the impact of specific environmental policies in the Madrid region (Soret et al., 2014) or the influence of sectoral emissions (Borge et al., All five configurations presented good correlations with observations with an average r of 0.752 although 25 lower values were obtained for the five-layer schemes (0.695 ± 0.077 and 0.745 ± 0.044) than for the configurations using the hybrid-layer scheme (0.750 ± 0.062 -0.801 ± 0.034). Among these multiplelayer-scheme simulations, the one showing the best r is ECMWF_70 which was the configuration with the coarser spatial resolution for LOTOS-EUROS among the three. Therefore, regarding the degree of correlation for multilayer configurations, increasing horizontal resolution to very fine grid sizes in the photochemical model does not improve results (providing that no changes are implemented in the emission inventory). 5 This is known in meteorological modelling as the Double Penalty issue (Mass et al., 2002) and occurs when evaluating simulations using point observations. The high resolution runs may be penalized twice, for not capturing the occurrence of the event and also for not predicting the right location of the event while a low resolution simulation can only fail predicting the event. (0.079 ± 0.059) and ECMWF_HR_70 (0.088 ± 0.060) were substantially lower than those of the other three configurations (0.260 ± 0.095 -0.117 ± 0.095). Clear improvements were observed using finer spatial resolution in LOTOS-EUROS (either WRF simulations or ECMWF_HR_70) while ECMWF_5 and, especially, ECMWF_70 presented systematic but moderate overestimations ( Figure S3). In consequence, the best configurations for adjusting the model bias were WRF_70 and ECMWF_HR_70. 5 A major reason for the overestimation detected for ECMWF runs with coarser spatial resolution was associated with an excessive O 3 formation in the noon hours of the day in situations of low wind speed as shown in Figure 3. This plot shows the correlation between the model bias and the modelled wind speed in the location of El Retiro (see location in Table 2) in Madrid for the five runs. In the plots corresponding to ECMWF_5 and ECMWF_70 runs we observe systematic positive bias especially in the period 14-20 UTC 10 when the formation is strong although it only spiked with low wind speed. This feature was not so marked in the three remaining configurations and, in particular, in the two WRF runs the bias values were randomly distributed around zero. ECMWF_HR_70 run showed a subtler systematic overestimation during daytime but the correlation with low wind speeds was not observed in this case. Analysing the night-time period (0-6 UTC) we detect that the systematic overestimation was only present in the ECMWF_70 execution. (WRF_70 and ECMWF_HR_70 runs) and clearly smoothed. The observations reflected that the timing of the increase was better represented in the mixed-layer scheme runs although the increase was excessively abrupt. The occurrence of this steep increase in the concentrations in the executions performed with the mixed-layer scheme coincided with the first steps of the boundary layer development. In the mixed-layer scheme, a rise of the boundary layer leads immediately to complete mixing of NO x emitted at the surface over the (increased) boundary layer, and thus limits titration of ozone; in the 70-layer schemes, however, the mixing over the boundary layer seems to take place more gradually. A more extensive validation including other tracers than the chemically active ozone should provide insight into which scheme performs better 5 under which conditions, and preferably lead to better characterization of the vertical diffusion.

O 3 vertical profiles
The evaluation of CTMs in the vertical direction has always been a difficult task due to the small number of high-resolution vertical observations to compare with. In this work, data from O 3 free soundings launched from Madrid airport at 12 UTC every seven days on 6, 13, 20 and 27 July 2016 were used for this purpose. 10 Comparisons between the vertical profiles of modelled O 3 with the five different set-ups and the observations are presented in Figure 5. The corresponding profiles of wind direction and speed (modelled data are taken from the input meteorological datasets) can be found in the supplementary information ( Figure S4).
As suggested by , the enrichment of O 3 in the lower troposphere during episodes without ventilation is high as a consequence of the intense photochemical formation and the development of con-  (Figures 5b and 5c).
The event of 27 July was characteristic of an accumulation scenario with high concentrations near the 20 surface while on 13 July (a typical venting event), O 3 in the lower levels was moderate and increased with altitude as described above. These two profiles were correctly simulated by most runs. However, the modelled profiles for the 6 and 20 July showed overestimation in the lower levels with respect to observations and for the 20 July case, all the high-resolution simulations overestimated the observed values. In Figure S4, we can also check how the input meteorological data used to feed the simulations (ECMWF and WRF fields) 25 closely reproduced the wind profiles obtained during the soundings of those four days both on speed and direction. Because of the complexity of the vertical mechanisms, further research should be conducted to investigate the causes of this mismatch in some events. From the qualitative perspective, the first obvious conclusion was that a larger number of vertical levels in the model considerably improved the capability for capturing the vertical gradients of O 3 concentrations with the exception of the lowest level on 20 July. However, even in the simplest vertical scheme (five layers), the model was able to reproduce the general vertical trends. A particular meteorological scenario was present during July 13. We can observe two O 3 layers centred around 3000 and 4300 m a.s.l . The two multilayered 5 WRF runs captured these two layers at, approximately, the correct altitudes although the 3000 m layer was not as marked as in the observations. The ECMWF runs also presented these two features but displaced in altitude by around 200-300 m with respect to observations.
In all the WRF_5 runs and on 27 July WRF_70 simulations a steep drop of surface concentrations was noticed. This is probably associated with the emission model configuration, the fine spatial resolution of the runs (3 x 3 km 2 ) and the vertical mixing in these set-ups. The O 3 soundings were released from Adolfo Suárez Barajas Airport, which is one of the major airports in Europe with more than 53 million passengers and 470 tonnes of goods transported in 2017 (http://www.aena.es/) and significant NO x emissions. The  Table 3 where generally satisfactory values can be observed with poorer results especially for July 20. The

15
FB showed a majority of positive values indicating overestimation although, in most cases, it was moderate (the range of averages for the five configurations was -0.5 to 0.11). The NMSE data in Table 3 support this conclusion because the average errors were small (0.023-0.042).
Summarising, the configuration that presented the best overall performance among the five tested in the previous sections was WRF_70, so it was employed for interpreting the variability of O 3 in the MAB during 20 July 2016.

Interpretation of O 3 in the MAB in July 2016
According to the dominant circulation over the MAB, three different episodes were distinguished and, with the aid of the model outputs, the basic features of the three events were described. Figure S2 shows the location of the selected monitoring stations used to characterise the behaviour of surface O 3 in the different 25 sectors of the MAB.

Recirculation events (REC)
These events correspond to the pattern sketched by Plaza et al. (1997) in which wind direction turns clockwise during the day aided by the effect of the blocking effect of the Guadarrama range while Querol et al.
(2018) described that the mixing layer growth at midday was reduced favouring vertical recirculation at the eastern slopes of the Guadarrama range (see section 2.1). In July 2016 four REC periods were identified: 5 1-6, 8-11, 15-17 and 25-28. To illustrate the main features of REC episodes, the period 15-17 July will be used as an example (Figures 6 and 7). A complete pattern of simulated fields of O 3 , wind and relative humidity (RH) for July 2016 can be consulted in Figure S5.
Surface wind speeds registered during REC episodes were weak (Figures 6 and S5) and the change in direction associated with recirculation is observed. However, despite the local circulation, air masses remain 10 inside the basin during REC days aided by a relatively thin mixing layer at 12 UTC ( Figure S4).
A stable band of high RH centred at around 4000 m is observed in Figures 6 and S5 which can be associated with the evapotranspiration caused by the intense heating registered during these events. The presence of a high-altitude trough located to the west of the IP during the 3-6 July REC period, induced moist south-westerlies at altitude resulting in the development of convective clouds in the evenings ( Figure S5).  Figure 7 shows how the strong convection during REC events injected ground-level pollutants at high altitudes during the late afternoon and the evening reaching up to 3500 m a.s.l. as illustrated in the NO 2 plots. When the 10 18 night-time stable boundary layer forms after sunset, air masses with high O 3 that originated near the surface during the previous day were decoupled and remained in the residual layer at altitudes ranging between 2000 and 4000 m a.s.l. forming reservoir layers (00 and 06 UTC cross-sections in Figure 7) which can fumigate the following day. These reservoir layers can also be observed as a relatively thin band at an altitude of 2000-4000 m a.s.l. during every night of the REC period (Figures 6 and S5). See Figure 1 to consult the the latitudinal and longitudinal cuts.

Northern advective events (NAD)
We will refer to NAD events as those during which the dominant situation consisted of the advection of air masses coming from the north over the MAB. During July 2016, the following two periods matched that description: 12-14 and 22-24.
During NAD periods, surface wind is channelled following the NE-SW axis parallel to the Guadarrama 5 range resulting in prevailing north-easterlies in the lowest tropospheric layers while in the upper levels the dominant component is NW (Figures 8, 9 and S5) often associated with the passage of cold fronts from the Atlantic. Winds are generally stronger than in REC events, which implies a renovation of air masses and lower temperatures.
Humidity during NAD events is conditioned by the arrival of air masses off the Atlantic, which are gen-  presented in Figure 8. Moreover, the formation of reservoir layers during NAD episodes was less common due to the lower convection, relative to that of REC cases presented before.
It is also remarkable in Figure 9 that above 3000-3500 m a.s.l. O 3 concentrations were very high (in the order of 100 ppb according to the model). This was associated with the stratospheric intrusion of very dry air described above. LOTOS-EUROS reproduced this stratospheric intrusion that was detected from data 5 obtained with free and tethered O 3 soundings for the same period during a field campaign . The actual impact of this stratospheric intrusion on surface levels remains unclear. In the referenced paper, the authors estimate a possible but limited impact of the intrusion on surface levels assuming that the boundary layer could exceed the 3000 m a.s.l. during the day. As shown in Figure 9, LOTOS-EUROS predicts that the maximum altitude of the boundary layer according to LOTOS-EUROS reached its maximum 10 21 values of 2500-2700 ma.s.l. limited by the wind ventilation so, probably, the impact on the surface should be low (if any) in this case. See Figure 1 to consult the the latitudinal and longitudinal cuts.

Southern advective events (SAD)
Southern transport implies the arrival of warm air masses, sometimes, coming from northern Africa (maximum temperatures at El Retiro during SAD events in the study period varied between 35.1 and 38.1 • C while 5 during NAD periods they ranged from 27.1 to 35.1 • C). In July 2016, two SAD periods were observed: 18-( Figures 10-12).
SAD events are characterised by constant southerly winds at the surface and at altitude as shown in the simulated fields of U and V (Figures 10 and S5). Because the southern coast of the IP is a densely populated area (especially in summer due to the strong touristic pressure) and anthropogenic emissions of 5 O 3 precursors are high, including industrial emissions around the cities of Huelva, Seville and Algeciras, regional contribution of external O 3 may acquire importance at the basin during SAD events.
High RH values in the middle troposphere are observed during SAD periods (Figures 10 and S5) where relevant increments were registered in the period 19-20 July. Although southerly winds are often associated with rain in the MAB, only small amounts of precipitation were collected during this period. concentrations during SAD periods are slightly lower than during REC events (with specific exceptions like 19 July) but higher than in NAD episodes. In the stations located in the basin hourly concentrations rarely 25 exceeded 180 µg/m 3 during SAD days (twice in July 2016 in the seven SAD days) but concentrations above 120 µg/m 3 were more frequent (1083 in total in all the stations in the study area or 155 per day) especially in the stations like ATA and CAM located on the NE of the basin (see 19 July in Figure 10). This proportion of records above 120 µg/m 3 is higher than during NAD events (average of 56 per day ) and lower than the rate registered during REC events (203 per day).
Vertical cross-sections of O 3 and NO 2 on two consecutive days (18 and 19 July) from a SAD period have been used to illustrate the different behaviours observed (Figures 11 and 12). The intense accumulation of NO 2 observed in the 00 and 06 UTC plots points out that on these days, ventilation was not as effective as 5 in NAD events. As a consequence, the O 3 daily cycles showed a considerable drop associated with titration in the morning rush hour unlike on NAD days and closer to the situation of REC episodes ( Figure 10).
Likewise, for NAD events, vertical mixing is limited as shown in the NO 2 vertical cross-sections of 18 and 19 July preventing the formation of reservoir layers during SAD events. The higher O 3 registered on 19 July seems to be related to the fact that a deeper boundary layer (maximum heights above 4000 m a.s.l. on 18 July, 3200 m a.s.l. on 19 July) allowed larger dilution, lowering surface concentrations. See Figure 1 to consult the the latitudinal and longitudinal cuts. Figure 13a presents a comparison between observed (data from ) and simulated (WRF_70 configuration) BLH at 12 UTC. We can observe that the model tends to overestimate the BLH at midday 5 although the general trends are captured. In particular, the gradual decrease in the 12 UTC BLH from 11 to 14 July allowed O 3 to accumulate smoothly in the basin, which was described in the aforementioned work, is also observed in the simulated data. The overestimation is slight on most days although larger differences are observed in certain periods (5-10 and 29-31 July).  describe lower midday BLH in O 3 accumulation episodes (equivalent to REC events described here) than in venting episodes (NAD or SAD). The simulations with LOTOS-EUROS confirm that finding as observed in Figure 13b. The mixing layer is deeper on 13 July (NAD) than on 16 July (REC) 5 from 0 to 16 UTC which includes the period of most effective photochemical formation of O 3 . This allows a more effective formation of reservoir layers during REC events that fumigate to the surface as the diurnal 26 convective circulation develops. After 16 UTC the BLH during the REC event grows higher than during the NAD day due to the larger convection in the second scenario.

4 Conclusions
Evaluation of a CTM is a basic tool for the analysis and forecasting of photochemical processes that give rise to high concentrations of tropospheric O 3 that frequently occur in the Mediterranean in summer. A preliminary requirement for the application of CTM for policy decisions is that they could reproduce adequately the processes and mechanisms identified by the field campaigns and reasonably reproduce the observations 5 of the monitoring stations, especially during acute O 3 episodes.
In this work, we present the results obtained from a simulation exercise (July 2016) performed with the LOTOS-EUROS CTM over the MAB, representative of summer conditions. Five configurations with different combinations of spatial resolution (25 x 25 and 3 x 3 km 2 ), input meteorological data (ECMWF 7 x 7 km 2 for the IP and WRF, 1 x 1 km 2 for the MAB) and vertical structures (mixed-layer scheme with five 10 altitude levels and hybrid-layer scheme with 70 altitude levels) for model evaluation and optimisation.
The main objective of the paper is to provide a phenomenological interpretation of O 3 events in the area after performing a detailed evaluation of the best configuration of the model for the specific area and period.
Regarding the specific question of the reasonable number of vertical levels in the model configuration, it is dependent on the objective of the study. In this study the environmental analysis was the main objective and 15 it was logical and feasible from the perspective of CPU time to employ a considerable number of vertical levels because it allowed a better representation of the vertical variability of O 3 . In other studies such as air quality forecasting or long term analyses in which CPU time may be large, the reasonable number of levels can be less.
Our results show that the LOTOS-EUROS model performs in a satisfactory manner in the five set-ups. 20 However, regarding surface O 3 , it is clear that the model benefits from finer spatial resolutions in the horizontal and also from the use of multilayered vertical schemes. As a result, WRF_70 and ECMWF_HR_70 were the optimal configurations.
Using multilayered 70 level set-ups, LOTOS-EUROS was able to reproduce the vertical gradients of O 3 in the Madrid basin, although in some cases the model presented an overestimation in the lower levels with 25 respect to observations. In most cases, the model was also able to reproduce features like fine O 3 layers.
The performance of LOTOS-EUROS was partly successful, differentiating the vertical structure of O 3 under distinct meteorological conditions so further research is needed to improve CTMs' performances in this particular aspect with, for example, comparisons with data from O 3 soundings under different meteorological scenarios.
Therefore, the modelling system is suitable to be employed for the interpretation of O 3 variability in the region. In light of the present study, we suggest using vertical schemes of CTMs with a sufficient number of levels for capturing O 3 variability in the simulations of summer episodes in the Mediterranean region. 5 Employing the WRF_70 configuration of LOTOS-EUROS which has shown the best performance simulating surface and vertical concentrations of O 3 in the MAB, we interpreted the variability of O 3 in the region.
Three episode types have been identified regarding the dominating circulation. Two of them are associated with advection, either from the north (NAD) or from the south (SAD), while the third is associated with local/regional recirculation of air masses (REC  (Sillman, 1995), are key parameters for facing model-based NO x -VOC sensitivity studies and the assessment of emission inventories. Some of these parameters (especially NO x ) should also be incorporated in 10 vertical measurements.
In future, similar simulations to the one presented in this study should be performed in the different air basins in the IP where O 3 exceedances have been recorded . CTMs should be configured specifically for each region or air basin to assure the best performance by capturing the influence of topography and local circulations. For such studies, we highlight the importance of conducting experimental 15 campaigns that can support the necessary model evaluation.
Finally, it should be noted that when running such fine resolutions for real applications it is also important to work on the emission datasets (out of the scope of this work). Increasing the detail in emission inventory (mainly based on a bottom-up approach) could improve the performance of CTMs when assessing sensitivities or emission scenarios. Moreover, improving time resolution in the emission models can be beneficial Competing interests. The authors declare that they have no conflict of interest.
Acknowledgements. This work was funded by the Ministry of Economy, Industry and Competitiveness and FEDER funds through the project HOUSE (CGL2016-78594-R), the Ministry of Agriculture, Fishing, Food and Environment, the Madrid City Council, the Madrid Regional Government and by the Department of Research, Innovation and University of the Aragón Regional Government and the European Social Fund (project E23_17D). The study was also partially