Evaluation of the LOTOS-EUROS NO 2 simulations using ground-based measurements and S5P/TROPOMI observations over Greece

. The evaluation of chemical transport models, CTMs, is essential for the assessment of their performance regarding the physical and chemical parameterizations used. While regional CTMs have been widely used and evaluated over Europe, their validation over Greece is limited. In this study, we investigate the performance of the LOTOS-EUROS v2.2.001 regional 15 chemical transport model in simulating nitrogen dioxide, NO 2 , over Greece from June to December 2018. In-situ NO 2 measurements obtained from the National Air Pollution Monitoring Network are compared with surface simulations over the two major cities of Greece, Athens and Thessaloniki. The model reproduces well the spatial variability of the measured NO 2 with a spatial correlation coefficient of 0.85 for the period between June and December 2018. About half of the 14 air quality monitoring stations show a good temporal correlation to the simulations, higher than 0.6, during daytime (12-15 p.m. local 20 time), while the corresponding biases are negative. Most stations show stronger negative biases during winter than in summer. Furthermore, the simulated tropospheric NO 2 columns are evaluated against ground-based MAX-DOAS NO 2 measurements and space-borne Sentinel 5-Precursor TROPOMI tropospheric NO 2 observations in July and December 2018. LOTOS-EUROS captures better the NO 2 temporal variability in December (0.61 and 0.81) than in July (0.50 and 0.21) when compared to the corresponding measurements of the MAX-DOAS instruments in Thessaloniki and the rural azimuth viewing direction in 25 Athens respectively. The urban azimuth viewing direction in Athens region however shows a better correlation in July than in December (0.41 and 0.19, respectively). LOTOS-EUROS NO 2 columns over Athens and Thessaloniki agree well with the TROPOMI observations showing higher spatial correlation in July (0.95 and 0.82, respectively) than in December (0.82 and 0.66, respectively) while the relative temporal correlations are higher during winter. Overall, the comparison of the simulations with the TROPOMI observations shows a model underestimation in summer and an overestimation in winter both in Athens 30 and Thessaloniki. Updated emissions for the simulations and model improvements when extreme values of boundary layer height are encountered are further suggested. of background column missing in the model simulations. This might be related to an underestimation of the free tropospheric column by the model due to missing lightning emissions. Secondly, the profiles of LOTOS-EUROS peak more strongly near the surface, and this leads to a smaller model-simulated TROPOMI value and a 455 subsequent strong difference in the free troposphere. The slope for July is 0.40 and the offset 0.22×10 15 molec . cm -2 as seen in is equal to 0.52 and the mean bias is -10% from June to December period. LOTOS-EUROS follows nicely the hourly variability of the measurements during daytime (12 to 15 p.m. 550 local time) and the temporal correlations between the simulations and the measurements range between 0.43 and 0.72 (excluding the traffic station “Piraeus” due to the impact of high traffic emissions close to it), while overall underestimates the measurements over the 14 stations in Greece. During night-time (0 to 3 a.m. local time) the model overestimates the surface NO 2 when the measurements are low suggesting a low boundary layer height assumption. A slight dependency on the season was found since the model underestimates stronger the NO 2 values during winter showing an average relative bias of -15% 555 and a spatial correlation coefficient of 0.78, while in summer the average relative bias is -1% and the spatial correlation coefficient reaches 0.86. The mean temporal correlation coefficient in both seasons is similar (about 0.50).

for the simulations of the atmospheric components were defined as a coarsening of the meteorological model levels and are spanning the troposphere from the surface to a top around 175 hPa (about 12 km). The anthropogenic emission inventory used is the CAMS-REG (CAMS Regional European emissions) version 2 for the year 2015 at 0.1°×0.05° (Granier et al., 2019;115 Kuenen et al., 2014). Biogenic emissions (isoprene) are calculated online using the meteorology and a detailed land use and tree-species database. Soil NO emissions are taken from a parametrization depending on soil type and soil temperature (Novak and Pierce, 1993) while NOx production from lightening is not included in the model. The aggregated total NO emissions from anthropogenic and biogenic sources used from June to December 2018 are shown in Figure S2 next to the biogenic NO emissions for the same period. Βiogenic NO emissions constitute 11% of the total NO emissions in the area as seen in Figure  120 S2, which shows Greece, south Albania, south North Macedonia, south Bulgaria and west Turkey, while the 97% of the NOx emissions is emitted as NO and the rest as NO2 in the model.  (Gery et al., 1989), while the aerosol chemistry is represented by the ISORROPIA II (Fountoukis and Nenes, 2007). More details on the chemistry module of LOTOS-EUROS can be found in Manders et al. (2017). For the biomass burning emissions and wildfires, the Global Fire 130 Assimilation System (GFAS), that assimilates fire radiative power (FRP) observations from satellite-based sensors (Kaiser et al., 2012), is used in the LOTOS-EUROS simulations. In the Greek domain and during the period of study some artificial fires were detected but since no big wildfires have been recorded in the area, these GFAS inventory was not taken into account in the nested simulation.

2.2Ground-based measurements 135
In order to validate the NO2 simulations derived from LOTOS-EUROS over Greece we compare model-derived surface concentrations with in-situ air quality measurements performed in the regions of Athens and Thessaloniki. Furthermore, the simulated NO2 tropospheric columns over the Greek domains are compared against MAX-DOAS NO2 columns, also situated in Athens and Thessaloniki.

In-situ NO2 measurements 140
Hourly in-situ measurements of NO2 concentrations over the Greek domain were obtained from the National Air Pollution Monitoring Network (http://www.ypeka.gr/) from June to December 2018. The aforementioned data are routinely reported to the European Environmental Agency Air Quality database. A chemiluminescence method is used for the measurement of nitrogen dioxide concentrations at the stations. The stations in Athens are operated by the Department of Air Quality while the rest of the Greek stations are operated by regional administrations. Hourly NO2 measurements are available for Athens and 145 Thessaloniki. The stations used for the evaluation of the model over Greece, were selected carefully in order to be well distributed and to be representative of the local emission sources. The locations of the stations are given in Figure 2 for Thessaloniki (top) and Athens (bottom). The marked colours over the stations refer to the average NO2 measured at each station between June and December 2018. We should acknowledge possible representativity errors when comparing the measurements from urban traffic stations with the mean value of a model grid cell (0.1°x0.05°) (Blond et al., 2007). For this 150 reason, stations characterized as urban traffic stations, localised close to busy traffic roads of the city and showing very large values, are excluded from the validation. As a result, out of a total of 24 stations reporting to the repository, 5 and 9 stations for the region of Thessaloniki and Athens, respectively are retained for the model evaluation. Measurements from these stations https://doi.org/10.5194/acp-2020-987 Preprint. Discussion started: 30 October 2020 c Author(s) 2020. CC BY 4.0 License.
were used in the past to investigate the NOx trends in Athens (Mavroidis and Ilia, 2012) to assess the impact of the economic crisis in Greece after 2008 (Vrekoussis et al. 2013), and to reveal how the surface NO2 concentrations are reflected on the 155 OMI/Aura retrieval (Zyrichidou et al., 2013).

MAX-DOAS measurements
In this study, tropospheric NO2 columnar measurements from Multi Axis Differential Optical Absorption Spectroscopy (MAX-DOAS) systems located in Thessaloniki and Athens are compared with LOTOS-EUROS simulated columns for July and December 2018. For the region of Thessaloniki, Phaethon, a miniature spectrometer ground-based MAX-DOAS system, is 160 used. The system was developed in 2006 at the Laboratory of Atmospheric Physics (LAP) in Thessaloniki, Greece (Kouremeti et al., 2008). This system operates regularly on the roof of the Physics Department of the Aristotle University campus which is located in the centre of Thessaloniki (Drosoglou et al., 2017). At the same location, an air quality measuring station, labelled AUTH in Figure 2, is in operation by the Region on Central Macedonia. For this study, we used MAX-DOAS observations at 15° elevation angle in order to avoid uncertainties introduced due to aerosols at lower elevation angles (Sinreich et al., 2005) 165 and at two azimuth angles: 220° and 255° designated in Figure 2 by the purple lines 1 and 2, respectively. In these viewing directions the MAX-DOAS system probes air over the centre of the city and the gulf of Thessaloniki, an area which is usually supplied by air from the western part and the industrial area of the city, i.e. directly from the urban environment. The average tropospheric columns from the two azimuth angles was calculated since both directions fall into the same grid pixel of the model simulations. The retrievals are based on geometrically approximated Air Mass Factor (AMF) (Wagner et al., 2010). 170 The tropospheric NO2 derived from MAX-DOAS instruments positioned at three different locations around Thessaloniki and the OMI/Aura satellite were compared during a 6-month campaign showing good agreement over the rural and the suburban areas (Drosoglou et al., 2017).
In Figure 2 (bottom) the location of the MAX-DOAS instrument used for the comparisons of the NO2 tropospheric column in Athens is marked. The MAX-DOAS instrument is installed at northeast of Athens, at Penteli mountain (527 m above sea 175 level), and belongs to the BREDOM network (Bremian DOAS network for atmospheric measurements), https://www.iup.unibremen.de/doas/ (Gratsea et al., 2016). Two azimuthal viewing angles are selected in this case as well, at 120° and at 232.5°, and are represented by the purple lines in Figure 2. The first one, marked with "R" is characterized as a rural unobstructed direction, while the other one is named "U" and views towards an urban direction ( Figure 2). The azimuthal viewing angles selected are representative for urban and rural air quality environment conditions in Athens. The tropospheric NO2 vertical 180 columns were derived, as in the case of Thessaloniki, using the geometric approximation.

S5P/TROPOMI NO2 observations 185
The Sentinel-5 Precursor, S5P, satellite was launched on October 13 th , 2017, carrying the TROPOspheric Monitoring Instrument, TROPOMI (Veefkind et al., 2012). The satellite flies in a near-polar, sun-synchronous orbit in an altitude of 824 km with an equatorial crossing at 13:30 local solar time (LST). TROPOMI is a passive, nadir-viewing spectrometer measuring wavelengths between the ultraviolet and the shortwave infrared. The swath width of TROPOMI in the Earth's surface is approximately 2600 km and the ground pixel of the instrument at nadir is 7×3.5km 2 , 5.5×3.5km 2 since August 2019, 190 achieving near global coverage in one day. The wavelength range used for the NO2 column retrieval algorithm is between 405 and 465 nm, while detailed information on the algorithm and the data can be found in the TROPOMI NO2 Algorithm Theoretical Basis Document (van Geffen et al., 2019(van Geffen et al., , 2020. The data are constantly validated by the Mission Performance Center Validation Data Analysis Facility 1 , VDAF, as well as several TROPOMI NO2 validation papers that have been recently submitted (Judd et al., 2020;Verhoelst et al., 2020). Compared with globally deployed ground-based remote sensing  DOAS instruments, the TROPOMI tropospheric NO2 shows on average 30% lower levels (S5P MPC Routine Operations Consolidated Validation Report, 2020). The validation work also shows high correlations with the independent observations and a nearly linear scaling of the error with the tropospheric column amount. Part of the bias is attributed to retrieval inputs, in particular cloud pressure retrievals and the albedo used. But a second part is attributed to the a-priori, which is available globally at a resolution of 1x1 degree. This is not enough to resolve the NO2 profiles near point sources and over cities. As a 200 result the tropospheric column is often underestimated at the hotspots (and somewhat underestimated in rural regions). Note, however that because the averaging kernels are used in our case, the comparison with LOTOS-EUROS is not influenced by the retrieval a-priori (Eskes and Boersma, 2003). Therefore, we expect a TROPOMI low bias of the order of 10-20% to remain, influencing the comparisons.
In this study we used the reprocessed daily data, RPRO, version 01.02.02 of TROPOMI for July and the offline data, OFFL, 205 for December 2018, which can be obtained via the Copernicus Open Data Access Hub (https://s5phub.copernicus.eu/). The data are filtered with a quality assurance value qa_value>0.75, ensuring mostly cloud-free observations, and gridded onto the LOTOS-EUROS grid at 0.1°×0.05°. The TROPOMI tropospheric vertical column is compared with the simulated NO2 column derived from the LOTOS-EUROS after the averaging kernel of the TROPOMI vertical columns is applied to the simulations in order to ensure a consistent comparison between the modelled and measured columns (Eskes and Boersma, 2003). 210

LOTOS-EUROS and in-situ measurements
In the following, we compare the hourly modeled surface NO2 concentrations from the lowest layer of the simulations with hourly in-situ measurements. Table 1 and Table 2  December are only taken into account, the summer period that includes July and August, the day period that includes only the daytime data between 12 p.m. and 15 p.m. local time during the whole period, and finally the night period that refers only to 220 the hours between 0 a.m. and 3 a.m. local time during the whole period. The distinct periods are selected in order to study the seasonal (summer and winter) and diurnal (day and night) performance of the model when different parameters may affect the simulations. For instance the atmospheric mixing and the parameterization of boundary layer pollutants are mostly affected by the diurnal cycle (Nester and Fiedler, 1992).
In the region of Thessaloniki (Table 1) the correlation coefficients calculated over June to December range from r = 0.49 to r 225 = 0.58. Overall, the mean correlation in summer is slightly lower (0.49) than in winter (0.55). During daytime the correlation generally increases by 13-21% and decreases by 27-34% during night in comparison to the whole period except for the urban industrial station "Sindos" where it decreases by about 20% during daytime and it slightly increases at night. In the case of the two urban background stations, "Malakopi" and "AUTH" the correlations are very good (r=0.69 and 0.63, respectively) during daytime. 230 In the region of Athens (Table 2) the same calculations are performed for the 9 selected stations. The average correlation over June to December is ~0.5 with two suburban background stations ("Ag. Paraskevi" and "Liosia") having the lowest values (r=0.39 and r=0.34 respectively) and the urban background/traffic stations ("N. Smurni" and "Marousi") the highest ones (r=0.62). In this case, a clear seasonal pattern in the model's performance, as is the case for Thessaloniki, was not found. On the other hand, when limiting the analysis to the day time period, the majority of the correlations are improved compared to 235 the June-December period and increased by about 5-55% and 70% for the suburban background station "Liosia". The urban background station "Nea Smurni" and the urban traffic station near the port of Piraeus exhibit decreased correlations during day in comparison with the June to December period (-29% and -55% respectively). Finally, the correlations worsen during nighttime (decreased by about 9-56% compared to the whole period) in Athens except for the "Piraeus" urban traffic station, which is located in the city port (Table 2. The correlation at the suburban industrial station "Geoponiki" during daytime reaches 240 0.72, compared to 0.44 for the nightime period.  In Figure 3 the mean NO2 simulations for each period chosen are compared against the NO2 measurements for all 14 stations and in Table 3 the respective statistics are given. Each period is marked with a different color: winter in blue, summer in orange, daytime in red, nighttime in purple and June-to-December in green. Overall, and irrespective of the temporal choice, the simulations are found to underestimate the in-situ measurements, as shown by the linear regression slopes, colored in tandem to the datasets. The model behaves similarly in winter (0.36 and 14.61 μg.m -3 the slope and the offset of the regression 255 line respectively) and summer (0.43 and 10.96 μg.m -3 the slope and the offset respectively), while the spatial variability is better reproduced in summer when the spatial correlation coefficient is 0.86. The difference between day and night comparisons, shown in Table 1 and Table 2 is evident whereas the strong model underestimation in daytime (0.14 and 5.55 μg.m -3 the slope and the offset respectively) merits further analysis. During the night period the model overestimates the measurements of low NO2 and underestimates the higher concentrations (0.63 and 9.85 μg.m -3 the slope and the offset 260 respectively). The spatial correlation is higher during day (0.80) than in night (0.70).

Figure 3 Scatterplot of the simulated and observed NO2 concentrations for the 14 stations used for the validation at the different periods. Winter is the November and December months (blue line), summer July and August (orange line), June to December includes measurements for the whole period of study (green line), day represents the daily hours (between 12 to 15 p.m. local time) during the whole period (red line) and night the night hours (between 0 and 3 a.m. local time) during the whole period (purple line).
265 Table 3 The coefficients of the scatterplot in Figure 3 between the averaged NO2 simulations and the in-situ measurements for the 14 stations used for the validation at the different periods; June to December, winter, summer, daytime and night-time. To investigate the performance of LOTOS-EUROS simulations in greater detail we compared the hourly NO2 concentrations at the urban background representative stations "Malakopi" in Thessaloniki and "Peristeri" in Athens. Time series for these 270 stations are shown in Figure 4. For both stations the LOTOS-EUROS simulations follow the measurements satisfactorily.

Period
During some periods the NO2 simulations are low compared to the measurements and this in most cases coincides with the daytime underestimation of the model, as shown at Figure 3 as well. This occurs, for instance, at the "Malakopi" station (upper panel) during the period of 1 June to 15 July and at the "Peristeri" station in early June. On the other hand, there are some days for which much higher NO2 levels are simulated than observed, mainly during nighttime, as seen for example during July at 275 the "Peristeri" station (lower panel).

280
The relative biases between the simulated and measured values of the each periods are shown in the box and whisker plot in Figure S3. The period over June to December (green box) shows a range of biases between -40% and 23% showing an underestimation of the measurements in most cases and a median bias of -10%. The relative biases in winter (blue box) range between -50% and 8% showing a clear underestimation of the measurements, while Ag. Paraskevi has a high positive bias of 285 54%. In summer though, the median relative bias is -2% and the model both underestimates and overestimates the measurements almost equally. The strong underestimations of the measured NO2 during day hours is depicted by the high negative biases at the daytime period and the absence of positive biases (red box). The station "Piraeus" shows a very low bias (-78%) during daytime which can be explained by the very high pollutant levels emitted near the station due to traffic and shipping. The night simulations show very high overestimation at the suburban industrial and background stations of "Elefsina" 290 and "Ag Paraskevi" respectively.
From this comparison we found that LOTOS-EUROS NO2 surface simulations are biased compared to the in-situ measurements over the two major cities of Greece over June to December 2018 showing an underestimation of the measurements with a mean relative bias of -13%, a median relative bias of -10%, a high spatial correlation coefficient equal to 0.85 and an average temporal correlation of 0.52. The separate evaluation during distinct periods of time shows that the 295 model underestimates the NO2 surface concentrations mostly during daytime (12 to 15 pm local time) and overestimates the low concentrations during the night-time (0 to 3 am local time). The daytime underprediction could be partly due to representation issues related to the location of the stations, which lie near urban city centres and industrial areas, that cannot be well resolved by the model at 0.10°x0.05°  as in the case of "Piraeus" station near the port in Athens.
Further, the daytime boundary layer height provided is likely too high, thus resulting in low NO2 surface concentrations 300 (Huijnen et al. 2010). Moreover, the chemistry of a model is indirectly affected by the photolysis rate and the meteorology (as an example the solar radiation is affected by the cloud coverage which in our case hourly cloud coverage data are obtained from ECMWF). However, the mismatch between the simulations and the measurements is found to be more significant during night-time when the model strongly overestimates in some cases the surface observations, the spatial correlation coefficient is lower (0.80 during day and 0.70 during night) and the temporal correlation coefficient is much lower than in daytime as well. 305 This could be due to possible flaws in the representation of the boundary layer and can be explained by a very low boundary layer height adoption during the night and a small vertical mixing as well (Bessagnet et al., 2016). The boundary layer height in LOTOS-EUROS is taken from the ECMWF operational weather analysis data and is based on the bulk Richardson number following the conclusions of the Seidel et al. (2012) review. Lampe (2009) further showed that during night the urban heat island can cause a larger boundary layer height and a stronger mixing that leads to the decrease of surface pollutants levels. 310 We find that the model shows a slight dependency on the season underestimating the NO2 during winter at most stations (average relative bias -15%) while in summer the average relative bias is -1% but with a larger range of the biases.  (Table 4). In 325 both seasons the bias is negative, higher in December, about -33% (-4.11×10 15 molec.cm -2 ), and lower in July, about -8% (-0.42×10 15 molec.cm -2 ) ( Table 4). The daily mean correlation in December is 22% higher than in July (0.61 and 0.50 respectively). The model shows similar characteristics for NO2 columns and surface concentrations when compared with MAX-DOAS and ground observations respectively with a negative bias in AUTH for both summer (-8.44% and -20.8% respectively) and winter (-32.75% and -39.2% respectively). 330  and December (right) are shown in Figure 6. Linear regression lines and equations are given along with the plots at the top.

345
Comparisons of the average diurnal cycles for July and December are shown in Figure 7. Overall, LOTOS-EUROS reproduces very well the diurnal cycle in July with the highest values between 6 and 7 UTC in the morning. On the other hand, the modelled NO2 levels in December are about 32% lower than the MAX-DOAS columns. As expected, the NO2 measurements are higher in winter than in summer because of the higher emissions in winter and the strong photochemical loss of NO2 in summer (Boersma et al., 2009). The MAX-DOAS columns show a small peak at 10 UTC while LOTOS-EUROS shows a 350 quite constant diurnal cycle. The lower model values in December could be partly explained by the fact that the NOx modelled lifetime may be too short, due to underestimated NOx emissions or because of a low boundary layer. Other parameters that play a pivotal role in the measurements and the simulations are related to meteorology, such as temperature and cloud coverage.
According to Schaub et al. (2007), high temperatures and longer days result in shorter NOx lifetime compared to lower temperatures and less hours of daylight, while they also showed that a cloud fraction of 0.2 results in a longer NOx lifetime 355 than a cloud fraction of 0.1, as a result of the decreased amount of solar radiation caused by higher cloud fraction. Therefore, uncertainties in the meteorological input data (cloudiness and temperature) in the model may induce uncertainties in the photochemical conversion and lifetime of NOx.
The same procedure followed for the case of Thessaloniki is also followed for the MAX-DOAS in Athens; the two distinct azimuthal angles have been selected; the azimuthal viewing angle towards the urban area (U) and the azimuthal viewing angle 360 towards rural area (R). Since the MAX-DOAS instrument in Athens is located in a mountainous area around 500 m above sea level and in order to succeed consistency in the comparison between the measurements and the simulations, we integrated the modelled NO2 columns above the model altitude of about 424 m. Figure 8 shows the time series of the tropospheric NO2 vertical column density from the MAX-DOAS in Athens and the simulated NO2 tropospheric columns from LOTOS-EUROS (above 424 m) at the corresponding model grid cells for July (left) and December (right) between 6 a.m. and 13 p.m. in the 365 https://doi.org/10.5194/acp-2020-987 Preprint. Discussion started: 30 October 2020 c Author(s) 2020. CC BY 4.0 License. urban direction, while for the rural direction is shown in Figure S4. The model underestimates slightly the measurements for both periods at the urban direction, similar to Thessaloniki, and stronger at the rural direction.

Figure 8: Time series of LOTOS-EUROS (red) and MAX-DOAS (green) NO2 columns over Athens for July (left) and December (right) 2018 in urban direction.
This is further confirmed by the statistics in Table 5 for the urban and rural directions. The average NO2 tropospheric columns 370 measured for July are 4.44±3.11 and 1.87±1.94 (10 15 molec.cm -2 ) while the simulated columns are 4.34±3.77 and 1.19±1.30 (10 15 molec.cm -2 ) for the urban and rural directions respectively. The model slightly underestimates the NO2 columns in July at the urban direction with a relative bias of -2.23% while at the rural direction indicates a stronger underestimation and the bias is -36.48%. In December the biases show similar characteristics as in July at both the urban and the rural directions and underestimate the measurements (-14.48% and -26.78% respectively). Similar biases were seen for the comparison of surface 375 simulations with ground based observations that are available in the same model cell pixel as the urban direction MAX-DOAS measurement (-2.91% for the summer period and -17.16% for the winter period). The daily mean correlation in July between the integrated columns above 424 m and the corresponding observations at the urban and rural directions are 0.41 and 0.21 respectively, while the correlations found with the full profile simulations are higher and equal to 0.56 and 0.42 for the two directions showing that the full profile follows better the variability of the observations, as seen in Figure 9 and Figure S5. The 380 daily mean correlation between the measurements and the partial columns above 424 m is low in comparison to the full profile of LOTOS-EUROS columns in December as well (0.19 and 0.41, respectively) at the urban direction while for the rural direction increases and is slightly higher for the integrated simulation above 424 m (0.81) compared to the full profile (0.79).
The number of available observations is higher in winter than in summer in the case of Athens and for the urban direction is 155 for summer and 181 for winter and for the rural direction 90 and 194 measurements were used respectively. 385    Figure S6 shows the average diurnal cycle for the rural direction. It is further confirmed here that the full profile column of LOTOS-EUROS captures better the daily variability of the measurements 395 compared to the partial column ( Figure 10). On the other hand the observed diurnal cycle is highly overestimated by the LOTOS-EUROS full profile while it is much similar to the partial column diurnal line. In this version of the model the mixing is more directly determined by the boundary layer height obtained from ECMWF data and possible uncertainties induced by the boundary layer height could explain the differences between the variability of the LOTOS-EUROS partial column and full profile with the measurements over Athens above the height of 500 m. We plotted the relative biases between the simulated 400 partial column and the measurements at the urban direction in July against the boundary layer heights used. We can see that when the height of the boundary layer is relatively low (between 0 and 500 m) the model highly underestimates the measurements ( Figure S7). The height of the boundary layer remains below 500 m mostly early in the morning, as seen in Figure 10, and the stronger difference between the partial column and the measurements is also observed at the same time.

Figure 9: Scatter plots between daily LOTOS-EUROS integrated column above 424 m (blue) and full profile (magenta) with MAX-DOAS NO2 columns in Athens for July (left) and December 2018 (right) for the urban direction. The linear regression equation and
This could point at an underestimated boundary layer height before 7 a.m and a subsequent mixing in the model mostly in the 405 lower heights and lower concentrations above the 500 m. At the rural direction the model's partial column underestimates the observed diurnal amplitude. The model underestimates the columns in both directions and that could be explained as well by underestimated emissions in the model or by missing transported pollution by neighbouring regions to the rural area. boundary layer assumption appears to play a pivotal role in the case of Athens, where the NO2 columns are measured above the first 500 m, and contributes to high mixing of pollutants below the 500 m early in the morning and a subsequent underestimation of NO2 at higher altitudes. For both urban areas, in Thessaloniki and Athens, the model underestimates slightly the measurements in July while the underestimation is higher during the winter month, as in the case of the surface 420 observations, which could point at underestimated NOx emissions and too short NOx modelled lifetime. Further sources of model uncertainties include the meteorology used for the simulations, and in particular from temperature and cloudiness. In addition, the MAX-DOAS tropospheric columns in both cities have been derived using the geometric approximation without taking into account the actual NO2 profile, introducing therefore, additional uncertainty. Finally, the one azimuthal directional observation in Athens compared with a grid cell of the model may not be representative of the relatively large grid pixel of the 425 model simulation, underestimating a possible horizontal plume from industrial areas i.e. from chimneys. Vlemmix et al., (2015) found that MAX-DOAS low daily averaged NO2 columns are overestimated by LOTOS-EUROS while higher columns are underestimated, well in agreement with the results of this study at the rural directions.

Comparison with Sentinel/5P TROPOMI vertical columns
Sentinel/5P TROPOMI data are gridded onto LOTOS-EUROS grid with the same spatial resolution (0.1°×0.05°). The 430 TROPOMI averaging kernels provided by the satellite product, that express the sensitivity of the instrument to the NO2 abundance within the distinct layers of atmospheric column, are applied to the model profiles in order to allow a consistent comparison between the modeled and observed columns and to eliminate any possible errors in the TM5-MP a priori profile shapes (Eskes and Boersma, 2003). The averaging kernels are applied directly by the LOTOS-EUROS model. Monthly averaged tropospheric NO2 columns for the TROPOMI observations and LOTOS-EUROS simulations are given in Figure 11  435 and Figure 12 for July and December, respectively over the inner area of Greece and the two sub-regions of Athens and Thessaloniki. The right columns at both figures show the absolute difference between the TROPOMI and LOTOS-EUROS columns. The regions to be analysed on more details later are marked with black rectangles over the maps in Figure 11 and are named Greece, Athens and Thessaloniki as seen at the maps of the inner area, Attica Basin and Thessaloniki respectively.
Over Greece, LOTOS-EUROS captures generally well the observed NO2 column abundances in July and December (upper 440 panels of Figure 11 and Figure 12, respectively), such as the densely populated area of Athens and the lignite-burning power plants at the northwest of Greece, in the area of Ptolemaida. We can further note that the TROPOMI background NO2 columns are higher than LOTOS-EUROS and this is mostly noticeable during July when the mean bias is strongly negative, at -0.59×10 15 molec.cm -2 (-45%) and the spatial correlation coefficient is lower than in December (0.78 and 0.87 respectively) ( Table 6). The high background levels of TROPOMI can be easily distinguished in the difference plot (right column) where 445 the purple color covering the entire region is about -0.50×10 15 molec.cm -2 . Comparison of the TROPOMI NO2 data with Pandora measurements in Helsinki showed that TROPOMI slightly overestimates the NO2 columns when they are relatively low, and underestimates the high columns (Ialongo et al., 2020). This is further shown by Dimitropoulou et al. (2020) who validated TROPOMI NO2 tropospheric columns with MAX-DOAS measurements over an urban area in Belgium and confirmed that TROPOMI underestimates the measurements by about 40-50% at urban sites and argued the need of more 450 appropriate a priori profiles in the TROPOMI algorithm retrieval. However, since in our case the averaging kernel is applied there is no NO retrieval profile-related bias influencing the comparisons. Therefore, the large background difference between the two datasets may be as well a result of background column missing in the model simulations. This might be related to an underestimation of the free tropospheric column by the model due to missing lightning emissions. Secondly, the profiles of LOTOS-EUROS peak more strongly near the surface, and this leads to a smaller model-simulated TROPOMI value and a 455 subsequent strong difference in the free troposphere. The slope for July is 0.40 and the offset 0.22×10 15 molec.cm -2 as seen in https://doi.org/10.5194/acp-2020-987 Preprint. Discussion started: 30 October 2020 c Author(s) 2020. CC BY 4.0 License. Table 6. Moreover, in December the LOTOS-EUROS NO2 columns are slightly higher over the high polluted areas such as the northwest of Greece and the temporal correlation between TROPOMI and LOTOS-EUROS is lower than in summer (0.32 and 0.44 respectively) unlike the spatial correlation that is higher in winter, and the bias is very low and slightly positive at 0.09×10 15 molec.cm -2 (5.9%) showing a small model overestimation. The slope in this case is equal to 0.30 while the offset is 460 1.15×10 15 molec.cm -2 (Table 6).

465
Statistics and maps over the sub-regions of the two largest and most populated regions in Greece, i.e. Athens and Thessaloniki, are given along with the ones in Greece to draw further conclusions. TROPOMI shows higher NO2 columns than LOTOS-EUROS in July at the sub-region of Athens ( Figure 11, middle panels). The spatial correlation when we study alone the polluted region of Athens is very high and equal to 0.95 while the temporal correlation is 0.48 and the bias quite low and still negative, about -17.9% (-0.48×10 15 molec.cm -2, ) as seen in Table 6. In December, the NO2 columns of LOTOS-EUROS are higher 470 mostly at the southern part of the sub-region of Athens comparing to the TROPOMI (Figure 12, middle panels), while the spatial correlation between them is lower than in summer (0.82) and the temporal correlation much higher than in summer

Figure 12: TROPOMI [left column] and LOTOS EUROS [middle column] NO2 tropospheric vertical columns over the region of Greece [upper row] and the sub-regions of Athens [middle row] and Thessaloniki [lower row] for December and their absolute differences (right column).
Thessaloniki shows a similar behaviour as Athens (lower panels of Figure 11 and Figure 12), with higher NO2 columns 480 observed by TROPOMI in July and lower in December in comparison with LOTOS-EUROS simulations. In July, their spatial correlation is higher than in December (0.82 and 0.66 respectively) while the temporal correlation is lower than in the winter month (0.30 and 0.58 respectively)( Table 6). LOTOS-EUROS underestimates the NO2 column during July (the bias is -0.52×10 15 molec.cm -2 ) and overestimates slightly in December (the bias is 0.15×10 15 molec.cm -2 ) as well in Thessaloniki.
The LOTOS-EUROS results over Athens in Figure 11 and Figure 12 clearly show a peak at the Isthmus of Corinth, the narrow 485 land bridge which connects the Peloponnese peninsula with the rest of the mainland of Greece, near the city of Corinth. Corinth has an important port (mostly cargo), while vessels navigate through the canal, and it is also an industrial area home to the largest oil refining company in Greece. In December (Figure 12, middle panels) LOTOS-EUROS simulates high NO2 columns (mean value ⁓5×10 15 molec.cm 2 ) near the Isthmus of Corinth, which are not supported by the TROPOMI observations, pointing to a possible overestimation of the NOx emissions in the area. Possible NOx reductions in the area should be studied when 490 emission inventories for year 2018 will be released. December are shown in Figure 13. For the region of Greece (upper panel) in July (left) we can see that the model reports low and nearly invariant NO2 columns while TROPOMI shows slightly larger daily variability, while in December (right) they both show similar daily patterns. Of course there are days on which the simulations agree well with the observations (18/12-21/12 and 29/12-31/12) while on others, they deviate (25/12 and 26/12). For Athens (middle panel) and Thessaloniki (lower panel) we can see that the simulated columns are variable; in July the CTM underestimates on average the satellite observed 500 NO2 columns for both regions, while in December, they follow closely the high values of TROPOMI which explains the higher temporal correlations in winter.

505
To conclude, LOTOS-EUROS, in general, underestimates the NO2 columns during July over Greece compared to the TROPOMI observations while it overestimates the NO2 levels during December. The emission inventory used as input for the simulations is outdated by a few years and this fact may impact on the NO2 simulations. Verhoelst et al., 2020 reported that TROPOMI shows an underestimation of NO2 tropospheric columns of -23 to -37% in clean to slightly polluted conditions that corroborates previous findings such as those of Ialongo et al. (2020) and could partly explain the model overestimation in 510 December at the two urban centres. However the simulations are well in line with the observations vis-a-vis the spatial distribution of the NO2 values, and mostly over the region of Athens. For both Thessaloniki and Athens in summer the spatial correlation is higher than in winter while the temporal correlation is lower. The model underestimation of the background values in summer compared to satellite retrievals according to Huijnen et al. (2010) can be a result of an underestimation of NOx lifetime in the model as well as an under-prediction of the transport processes in the free troposphere. NOx emissions 515 from lightening also are not included in the model and as a result some NO2 is missing in the free troposphere. While it is generally assumed that underestimated or missing soil-NO emissions usually form an important factor of uncertainty in the models note that the background values of TROPOMI remain high even over the sea, indicating that natural emissions might not be the cause of the high background values. Additionally, the simulations are considered at the TROPOMI overpasses over the Greek domain, about 12 p.m., a time when LOTOS-EUROS has already demonstrated low NO2 surface concentrations 520 when compared with in-situ measurements. In summer the bias between the surface NO2 measured by the in-situ stations in the region of Thessaloniki is -16.2% while in Athens is -3.8% and the biases between the TROPOMI and LOTOS-EUROS columns for the same regions are -26.9% and -17.9% in July respectively (Table 7). However, in winter the model underestimates as well the surface observations in Thessaloniki and Athens (-33.1% and-15.9% respectively) while overestimates the TROPOMI tropospheric columns (3.6% and 16.8% respectively). For the same grid cells that we analyzed 525 the MAX-DOAS observations in Thessaloniki and Athens we extracted respectively the TROPOMI observations for July and December and in-situ measurements when available for the summer and winter periods and we show the models biases (Table   7). For the grid cell in Thessaloniki (AUTH) we found that the model underestimates all kind of observations in July with a higher negative bias found for the surface simulations (-20.8%), while in winter LOTOS-EUROS highly underestimates both the surface observations and the tropospheric column measured by MAX-DOAS (-39.2% and -32.8% respectively) and slightly 530 overestimates the TROPOMI observation (8%). In the urban grid cell in Athens we found the same characteristics as in AUTH, where in summer the model underestimates the surface measurements and columnar observations while in winter as well only the TROPOMI observations are overestimated (8.6%). The MAX-DOAS and the in-situ measurements are underestimated only by 2%-3% while TROPOMI shows a higher negative underestimation (-12.6%) in July. The rural grid cell measurements of NO2 tropospheric columns are underestimated by the model when compared to MAX-DOAS and TROPOMI observations 535 both in summer and in winter.
https://doi.org/10.5194/acp-2020-987 Preprint. Discussion started: 30 October 2020 c Author(s) 2020. CC BY 4.0 License. Table 7 Summarized table of relative biases between LOTOS-EUROS NO2 surface simulations and in-situ measurements and between simulated and observed vertical tropospheric columns. The "grid cell" in the "Region" column refers to biases of NO2 values found at the same grid cell, while "mean area" refers to the biases calculated for the total area in summer and winter. The positive biases are shown in bold. In-situ -3.8% -15.9%

Conclusions
In this work, we evaluate NO2 simulations over Greece from the LOTOS-EUROS regional CTM using in-situ surface concentrations from 14 air quality stations during June and December 2018. Further we compare LOTOS-EUROS columns against MAX-DOAS and Sentinel/5P TROPOMI tropospheric columns in July and December 2018. The model setup is based 545 on the anthropogenic emission inventory TNO-CAMS v2.2 for the year 2015, ECMWF meteorological data and CAMS nearreal time initial and boundary conditions. We conclude the following: Overall, the spatial correlation between the modelled and the measured surface NO2 values ranges between 0.70 and 0.86 depending on the time period. The mean temporal correlation coefficient is equal to 0.52 and the mean bias is -10% from June to December period. LOTOS-EUROS follows nicely the hourly variability of the measurements during daytime (12 to 15 p.m. 550 local time) and the temporal correlations between the simulations and the measurements range between 0.43 and 0.72 (excluding the traffic station "Piraeus" due to the impact of high traffic emissions close to it), while overall underestimates the measurements over the 14 stations in Greece. During night-time (0 to 3 a.m. local time) the model overestimates the surface NO2 when the measurements are low suggesting a low boundary layer height assumption. A slight dependency on the season was found since the model underestimates stronger the NO2 values during winter showing an average relative bias of -15% 555 and a spatial correlation coefficient of 0.78, while in summer the average relative bias is -1% and the spatial correlation coefficient reaches 0.86. The mean temporal correlation coefficient in both seasons is similar (about 0.50).
Very good agreement in the diurnal cycle variability of the LOTOS-EUROS NO2 columns compared to those of the MAX-DOAS instrument in Thessaloniki is found for July, as well as a small underestimation (relative bias of -8.44%) of the observations. The underestimation of the measurements is higher during December (-32.8%). The model underestimates as 560 well the NO2 columns over the urban region of Athens negligibly in July (-2.2%) and slightly stronger in December (-14.5%), while follows much better the diurnal cycle when the model's full profile is considered compared to the partial column above the 424 m, pointing to uncertainties in the boundary layer height which could depend on parameters that affect urban environments too and are not taken into account (e.g. the urban heat island). The measurements at the rural direction are https://doi.org/10.5194/acp-2020-987 Preprint. Discussion started: 30 October 2020 c Author(s) 2020. CC BY 4.0 License. underestimated in both summer and winter showing possible underestimation of pollution transfer from neighboring regions, 565 while in winter the correlation is pretty high and equal to 0.81.
The model reproduces very well the spatial variability of TROPOMI NO2 columns over Greece capturing the locations of low and high NO2 columns. The spatial correlation between the simulations over Athens and the TROPOMI observations is 0.95 in July and 0.82 in December while the levels of NO2 are underestimated and overestimated respectively in summer and winter by ⁓18%. The same characteristics are observed over the city of Thessaloniki as well, with higher spatial correlation in summer 570 (0.82) and negative relative bias (-26.9%) and lower spatial correlation in winter (0.66) and a negligible positive bias (3.6%).
Higher background values of NO2 are observed in TROPOMI product mainly during summertime possibly due to an underestimation of the free tropospheric column (missing lightening emissions in the simulations, under-prediction of the transport processes in the free troposphere) and the model-simulated TROPOMI column (higher concentrations of LOTOS-EUROS near the surface). 575 The underestimation of the MAX-DOAS columns in summer and winter at the urban area of Thessaloniki is consistent with the negative biases of the surface observations and could point at underestimated emissions or under predicted lifetime of NO2 mostly during the winter season. Same characteristics are found for the urban region in Athens, where MAX-DOAS views the troposphere above the first about 500 m, suggesting lower NO2 concentrations at higher altitudes as well. The relative biases between TROPOMI and the modelled columns for both areas of Athens and Thessaloniki are negative and higher than in the 580 case of surface measurements possibly due to missing emissions at higher altitudes (i.e. lightening). However, in winter the surface measurements are underestimated by the model and the vertical TROPOMI columns overestimated possibly due to the already known TROPOMI NO2 underestimation over slightly polluted areas. Further studies on emission inversions using the data assimilation package of LOTOS-EUROS should be conducted in a next step to account for uncertainties due to outdated emission inventories. Further, the model simulations found to rely on possible uncertainties on the meteorological input data 585 and improvements when the height of boundary layer is very low or very high are suggested.