Ozone pollution during the COVID-19 lockdown in the spring 2020 over Europe analysed from satellite observations, in situ measurements and models

. We present a comprehensive study integrating satellite observations of ozone pollution, in situ measurements and 20 chemistry transport model simulations for quantifying the role of anthropogenic emission reductions during the COVID-19 lockdown in spring 2020 over Europe. Satellite observations are derived from the IASI+GOME2 multispectral synergism, which provides particularly enhanced sensitivity to near-surface ozone pollution. These observations are first analysed in terms of differences between the average on 1-15 April 2020, when the strictest lockdown restrictions took place, and the same period in 2019. They show clear enhancements of near-surface ozone in Central Europe and Northern Italy, and some other 25 hotspots, which are typically characterized by VOC-limited chemical regimes. An overall reduction of ozone is observed 25 km for IASI at nadir and ground pixels of 80 km × 40 km for GOME-2). IASI+GOME2 jointly fits co-located IR and UV spectra for retrieving a single vertical profile of ozone for each pixel. The horizontal resolution corresponds to that of IASI, using for each pixel the UV measurements from the closest GOME-2 pixel (without averaging). The present work uses daily IASI+GOME2 multispectral observations of ozone available for clear sky and low cloudiness 175 (pixel cloud fractions below 30%) and derived from the average of the retrievals using measurements from all available MetOp satellites (A, B and C for 2020 and A and B for 2019). The version of the algorithms used here is described and validated at global scale against ozone-sondes by Cuesta et al. (2018) (with a correlation of 0.85, a mean bias of -3% and a precision of 16%). The capacity to observe near-surface ozone with IASI+GOME2 has been shown by a good agreement against surface in situ measurements of ozone for two major ozone outbreaks across East Asia and at daily scale (a correlation of 0.69, a weak 180 mean bias of -5%, a precision of 20% and similar standard deviations for both datasets). Single-band approaches as those from IASI only were not able to observe such near-surface variability. 18 whole continent (seen in the free troposphere by ozone sonde measurements, Steinbrecht et al., 2021) and clearly put in 540 evidence over the ocean (Atlantic, North and Mediterranean seas). We compare these observation-based estimations with model-only estimates derived by changing emissions according to the reductions in anthropogenic activities estimated for the lockdown period. For this, we use two different model configurations, with distinct emissions in standard and COVID-19 conditions and meteorological fields. These two model-based estimates provide similar regional patterns of enhancements and reductions of surface ozone, but clearly different magnitudes (by a 545 factor 3 to 6). This is mainly explained by marked differences in vertical mixing within the atmospheric boundary layer, according to the use of two different meteorological models, as well as the year for which the emission inventories were estimated (one simulation uses an inventory from 2010 and the other from 2017) and different assumptions in variations of anthropogenic emissions during the lockdown. As compared to measurements, both configurations underestimate the amplitude of the positive and negative anomalies. However, the configuration that shows amplitudes closer to those observed 550 are the ones whose abundances of ozone precursors (e.g., NO 2 ) in absolute terms (in standard conditions and changes during the lockdown over homogenous hotspots) are closer to those observed by in situ sensors. The differences between the simulations using two different model setups, and with respect to observations highlight the complexity to simulate the effect of the changes in ozone concentrations due to changes in anthropogenic emissions, as occurred during the lockdown.


Introduction
During boreal springtime of 2020, worldwide measures for curbing the spread of the COVID-19 virus have led to 55 unprecedented and abrupt lockdowns in transportation (road, airplanes, and ships) and industry. These strong limitations drastically reduced the emissions of anthropogenic pollutants, inducing significant changes in atmospheric composition and air quality from local to worldwide scales, and particularly in regions such as China e.g. (Le et al., 2020) and Europe e.g.
We will compare this synergetic "observational & model" estimate with the one derived from models only, based on the difference of ozone simulations with inventories COVID and STD and the meteorological conditions of 2020, as follows (2)

155
In the current work, we mainly focus on the period from 1 to 15 April 2020 (and use 1-15 April 2019 as standard period) which corresponds to the 15-day period most perturbed emissions by the lockdown according to the COVID emission reduction factors from CAMS (https://atmosphere.copernicus.eu/covid-data-download). This work reports for example a reduction of -73 % of gasoline road transport for the 5 most populated countries in Europe (Germany, France, Great Britain, Italy and Spain, that are also the largest contributors to the total European emission decreases), whereas it is -64 % and -66 % for the 15 days 160 before and after (and then -54 % and -38 % for the two following fortnights). We also show some of the results for the whole month of April (2020 and 2019) to directly compare our estimates with those from previous works (Ordóñez et al., 2020;Souri et al., 2021) provided as monthly averages.
The following paragraphs describe briefly the datasets used by this approach.

IASI+GOME2 multispectral satellite observation of lowermost tropospheric ozone 165
The IASI+GOME2 satellite approach is designed for observing lowermost tropospheric ozone through the multispectral synergism of thermal infrared (IR) atmospheric radiances observed by IASI (Infrared Atmospheric Sounding Interferometer, Clerbaux et al., 2009) and ultraviolet (UV) earth reflectances measured by GOME-2 (Global Ozone Monitoring Experiment-2, EUMETSAT, 2006), according to the detailed description provided by Cuesta et al., (2013). Both instruments are onboard the MetOp satellite series, and they both offer global coverage every day (for MetOp-A, B and C respectively around 09:30, 170 09:00 and 10:00 local time) with a relatively fine ground resolution (12 km-diameter pixels spaced by 25 km for IASI at nadir and ground pixels of 80 km × 40 km for GOME-2). IASI+GOME2 jointly fits co-located IR and UV spectra for retrieving a single vertical profile of ozone for each pixel. The horizontal resolution corresponds to that of IASI, using for each pixel the UV measurements from the closest GOME-2 pixel (without averaging).
The present work uses daily IASI+GOME2 multispectral observations of ozone available for clear sky and low cloudiness 175 (pixel cloud fractions below 30%) and derived from the average of the retrievals using measurements from all available MetOp satellites (A, B and C for 2020 and A and B for 2019). The version of the algorithms used here is described and validated at global scale against ozone-sondes by Cuesta et al. (2018) (with a correlation of 0.85, a mean bias of -3% and a precision of 16%). The capacity to observe near-surface ozone with IASI+GOME2 has been shown by a good agreement against surface in situ measurements of ozone for two major ozone outbreaks across East Asia and at daily scale (a correlation of 0.69, a weak 180 mean bias of -5%, a precision of 20% and similar standard deviations for both datasets). Single-band approaches as those from IASI only were not able to observe such near-surface variability.
The IASI+GOME2 satellite product include vertical profiles of ozone, partial columns, averaging kernels (representing sensitivity of the retrieval to the true atmospheric state), error estimations and quality flags. Since 2017, global scale IASI+GOME2 retrievals are routinely produced by the French data centre AERIS and they are publicly available (see 185 https://www.aeris-data.fr and https://iasi.aeris-data.fr).

Surface in situ measurements
In this work, we use in situ surface measurements of O3 and NO2 from the European Air Quality e-Reporting (https://www.eea.europa.eu/data-and-maps/data/aqereporting-8). We only consider background stations of all categories (urban, suburban, and rural). We analyze here daily and morning averages of measured surface concentrations and compared 190 them with IASI+GOME2 ozone data and CHIMERE model simulations. A comparison shown in Table 2 and commented in section 3.1 suggests that a slight better agreement is found between daily averages of surface data (as compared to morning averages) and IASI+GOME2. This is likely linked to the fact that IASI+GOME2 retrievals show a relative maximum of sensitivity at about 2.2 km of altitude over land in mid-latitudes (see Cuesta et al., 2013). At this altitude and during the overpass time of the MetOp satellites around 09:30 local time, the IASI+GOME2 approach likely measures ozone 195 concentrations at the residual atmospheric boundary layer. We expect that the variability of these last ones is better represented by daily averages than morning surface concentrations, that have not yet been mixed vertically within the whole boundary layer. Therefore, most comparisons between IASI+GOME2 retrievals and surface concentrations consider daily averages of these last ones.

CHIMERE chemistry-transport model 200
CHIMERE is a state-of-the-art chemistry transport model widely used for studying atmospheric composition and forecast mainly at regional scale e.g. (Vautard et al., 2000;Honoré et al., 2008;Rouïl et al., 2009;Menut and Bessagnet 2010;Menut et al., 2015a;Marécal et al., 2015). It has been compared to numerous measurements, including meteorological variables and atmospheric chemical species (Menut et al. 2000;2005, 2015bBessagnet et al., 2016;Vivanco et al, 2017). Regularly, new versions of the model are proposed for a community of users (Menut et al., 2013;Maillet et al., 2017). Whereas ozone and 205 nitrogen oxides are modeled as explicit species, other species such as particulate matter or VOCs are composed by several ensembles of families of species. Biogenic emissions are estimated with the MEGAN online model (Guenther et al., 2006), mineral dust with the scheme from Alfaro and Gomes (2001) and Menut et al., (2005b) and sea-salt with that from Monahan (1986).
In the present study, we use two different setups of CHIMERE with different anthropogenic emission inventories and 210 meteorological conditions. This provides an estimation of the uncertainties of the ozone pollution simulations during the lockdown conditions, when varying these two key inputs. We also compare the two sets of simulations with respect to satellite and in situ observations, for identifying in which case there is a better match between observational and simulation datasets.  , 2015). The original inventory is considered for STD simulations in 2019 and 2020, whereas for the COVID emissions in 2020 a relative reduction for each activity sector (road transport, residential, industry, aviation, and shipping) is applied. The magnitude of the 220 decrease is estimated for the European domain based on the average for the five European most populated countries (Germany, France, Great Britain, Italy, and Spain) of the CAMS COVID daily emission reduction factors for each sector, averaged over the periods 1-15 April 2020 and 16-30 April 2020. Even though some differences occur in emission variability from country to country, with Spain, Italy and France showing the strongest changes, while Great Britain and mostly Germany weaker ones, we assume here a spatially homogeneous variation of emission factors over the whole domain. This corresponds to -73% (-225 66%), -16% (-13%), -44% (-44%), +5% (+5%), -91% (-90%) and -19% (-18%) for emissions from road transport, power, industry, residential, aviation and shipping sectors, for 1-15 (16-30) April 2020. According to CAMS, the COVID-19 restrictions led to a heterogeneous impact across different pollutants from industrial and residential sectors (https://atmosphere.copernicus.eu/emissions-changes-due-lockdown-measures-during-first-wave-covid-19-europe). For the industry sector, a smaller reduction is observed for non-methanic volatile organic compounds (NMVOC), NH3 and SO2 (mostly 230 emitted from food/beverage and chemistry industries, less affected by lockdown measures) with respect to other pollutants.
For the residential sector, only those pollutants mainly related to wood combustion processes (i.e., PM10, PM2.5, NH3, NMVOC, CO, biogenic CO2 and CH4) experienced a slight increase, while NOx, SOx, (and fossil fuel related CO2) showed a modest decrease. For that reason, CHIMERE C1 emissions for the COVID scenario consider the partitioning for industry between NMVOCs, NH3 and SOx, and that for the residential sector between NOx and SOx, which are used within CAMS revised 235 inventory.
The C1 model configuration uses meteorological fields from the BOLAM e.g. (Buzzi et al., 1994) hydrostatic meteorological model, whose boundary conditions are provided by the GLOBO e.g. (Malguzzi et al, 2011) hydrostatic general circulation model. Both models are developed at CNR-ISAC (https://www.isac.cnr.it/dinamica/projects/forecasts/). Initial conditions of BOLAM and GLOBO are taken each day at 0000 UTC from analyses of the NCEP/GFS model (Kalnay et al., 1996).

Configuration 2 (C2) of CHIMERE
The second configuration (C2) of CHIMERE mainly corresponds to that used by Menut et al., (2020 for the COVID simulation, the "driving" dataset of the Apple company activity database was used. These data were used to apply a daily emissions reduction factor for each European country. Second, the emissions of the industrial and off-road sectors were establish using the same reduction factor but divided by two. Energy, agricultural and waste sectors were not changed.
Finally, residential emissions are increased in order to account for the fact that people stay at home during the lockdown, but by a fourth of the factor obtained from the Apple "driving" dataset. Discussions on the differences between the COVID-250

Results
For better understanding the information provided by observations and simulations, the current multi-data analysis is presented in several steps. First, we focus on the total changes of ozone pollution between 2020 and 2019 (subsection 3.1), that are directly observed by in situ sensors and from space, and which we compare them with the corresponding amount simulated by models. Then, we analyze the model-derived changes between these two years, but only associated with meteorological 260 conditions (subsection 3.2). Finally, we compare changes of ozone pollution only linked with the pandemic lockdown conditions, estimated from models and observations (subsection 3.3). The originality of these results resides in the use of in situ and satellite observational estimates of the changes in ozone pollution associated the lockdown conditions

&
, that are adjusted for avoiding meteorology effects. This adjustment is derived from the model using business as usual emissions and thus avoiding the ambiguities associated to the estimation of anthropogenic emissions during the lockdown period. 265 Additional comparisons of the two model setups (C1 and C2) are also provided for better understanding the differences between these simulations and their agreement with respect to observations (section 4). These comparisons allow the identification of key elements that influence the accuracy of the model and the configuration that better match the observations.

3.1
Changes in ozone pollution in 2020 with respect to 2019

Satellite and in situ surface measurements 270
The first step of our study consists of analyzing all available datasets in terms of the changes in ozone pollution over Europe between the pandemic lockdown period in 2020 (focused here on April) with respect to the same period during the previous year (this difference is hereafter called O3 2020-2019 ). Figure 1 shows a comparison between the two observational datasets averaged over the periods 1-15 and 1-30 April. The first 15 days of the month show the most pronounced changes between the surface measurements coincident in time and space with satellite data. A good agreement is shown in terms of regional ozone patterns between IASI+GOME2 satellite data and in situ surface measurements. Both datasets clearly show similar structures of positive and negative anomalies. Ozone enhancements in 1-15 April 2020 with respect to 2019 are seen both by satellite and surface data over Northern and Eastern France, Western and Southern Germany, Northern Italy, and Southwest England In quantitative terms, scatterplot comparisons of co-located IASI+GOME2 and in-situ data in time and space are presented in  (Fig. 2a), the correlation coefficient between these datasets is 0.55 while the standard deviation of both datasets is practically the same and the root mean squared (RMS) difference is 11.8 ppb. As the variability of the ozone anomalies is smaller for the monthly average ( Fig.  295 2b), the correlation coefficient is moderately lower (0.46), and the RMS difference is reduced (10 ppb). There is an average difference or bias (of 8.6 ppb for 1-15 April) for the average ozone concentration changes, with lower values for the satellite retrievals than those for in situ data. This bias may partially come from the differences in the altitude of the measurements: in situ data are surface measurements whereas IASI+GOME2 measurements are lowermost tropospheric ozone columns. The satellite retrieval of this partial column is typically most sensitive around 2.2 km above sea level over land (quantified over 300 Europe by Cuesta et al., 2013). Therefore, the negative bias for IASI+GOME2 with respect to surface concentrations may suggest that ozone concentrations at atmospheric layers roughly ∼2 km above surface level decreased more than at the surface with respect to the same periods in 2019. This could be explained by several reasons such as a larger sensitivity to emission changes at the surface than at higher altitudes, differences in sampling time and also by the fact that satellite measurements sample air masses of both the boundary layer and free troposphere, where ozone concentrations may have had different 305 variations between 2019 and 2020. Moreover, surface ozone concentrations are directly affected by titration with NO; its impact on ozone columns up to 3 km is expected to be lower. Since the degree of freedom for the LMT partial columns is generally lower than 1 (typically 0.35 over land in Europe), IASI+GOME2 retrievals could also have some influence of the a priori concentrations. for CHIMERE C2 (with correlation coefficients of e.g., 0.67 and 0.44 for C1 and C2, respectively, in 1-15 April). However, both configurations of the models clearly underestimate the variability of the ozone changes in spring 2020 with respect to the previous year, as compared to in situ measurements (larger by more than a factor ∼2). Whereas O3 2020-2019 observed at the surface range from roughly -12 to +12 ppb in terms of daily averages, simulated values only spread from -6 to +6 ppb. The range of values of IASI+GOME2 has the same amplitude as compared to in situ data but shifted towards smaller values (-20 350 ppb to +5 ppb). Similar features are remarked for the average over the whole month of April and 1-15 April, with larger enhancements for this last one.
Generally, simulations (both C1 and C2) show total anomalies in 2020 (with respect to 2019) that are only partially associated with the regions of typical "NOx-limited" and "VOC-limited" chemical regimes. On the other hand, observations from both in situ sensors and satellite do show clearer similarities between the typical regimes (in the regions where both datasets are 355 available) and ozone anomalies during the pandemic lockdown. This might be linked to a large influence of meteorological conditions in the models, as compared to lockdown induced emission changes. Still another reason for these differences could be that boundary conditions for CHIMERE simulations (both C1 and C2) do not account for the changes in anthropogenic emissions linked with the pandemic in 2020 (for 2019 and 2020, the same climatology is used for C1 and the same emission inventories for C2). This would lead to a positive bias in simulated differences if part of the observed decrease at the free 360 troposphere (by Steinbrecht et al., 2020) affected the surface. Overall, the differences between the two model setups, and also with respect to observations highlight the complexity to simulate the effect of the changes in ozone concentrations due to changes in anthropogenic emissions, as occurred during the lockdown. In the following subsections 3.2 and 3.3, we use CHIMERE C1 simulations to derive the adjustments for observations as they show a better correlation to surface observations than those from CHIMERE C2 (Fig. 5). Further discussions on the causes of the differences between C1 and C2 simulations 365 are provided in section 4.  Fig. 11 of this paper) and north of 44°N for Ordóñez et al., 2020 (examining the differences between the lower panels of Fig. 1 of this paper). In quantitative terms, our estimate shown in Fig. 6c agrees well with that for Ordóñez et al. (2020) over France, Belgium, Germany, and Italy. Ordóñez et al., (2020) derives 5, 10 to 13 %, 10 to 12% and 3% for these 4 countries, according to the differences between observed and meteorology-adjusted changes reported in Table 1  In the period 1-15 April, our estimates show that meteorological conditions induce an enhancement MDA8 surface ozone of 385 14 % over Central Europe (Fig. 6a), which is ∼6 % larger than the average over the whole month. This suggests that during the first 15 days of the month the largest reduction of anthropogenic emissions (according to CAMS) is concomitant with a larger European ozone production north of 44°N associated with meteorological conditions.

Impact of COVID19 lockdown of spring 2020 in ozone pollution
Figures 6b and 6d show net changes of ozone pollution in terms of surface MDA8 concentrations only associated with the 390 pandemic lockdown, which are derived from in situ measurements after withdrawing the influence of meteorological conditions (using Eq. 1). We remark that ozone enhancement over a band from Benelux to Northern Italy (green rectangle in Fig. 3a) is clearly less pronounced, but it is still significant when excluding meteorological effects. Over this area, we estimate a MDA8 surface ozone enhancement related to the lockdown during the month of April of ∼3 % (Table 3) and a maximum value over Northeastern Italy of ∼20 %. Moreover, ozone changes derived from adjusted surface observations are negative 395 anomalies over the center of France and part of center of Germany (Fig. 6b, d), while total observed changes were positive in the 1-15 April period (Fig. 3a).
An overall agreement is found in the positive and negative patterns between Fig. 6d and those depicted by Ordóñez et al., (2020) and Souri et al., (2021). In the 3 estimates excluding meteorological effects, a net enhancement of surface ozone is seen across a region extending from Southern England, Benelux, Northeastern France/Southwestern Germany, and Northern Italy. 400 Our estimate of the ozone enhancements over these regions is rather consistent with that estimated by Ordóñez et al. (2020) and Souri et al. (2021), ranging from ∼3 to ∼8 % (according to Figures 1 and 11 respectively of these papers). On the other hand, a net reduction of MDA8 ozone concentrations is derived over Southwestern and Northeastern Europe of -7 and -8 %, respectively (see Table 4). For these regions, Ordóñez et al. (2020) roughly estimate a reduction -2 % to -7 % whereas these values are near zero for the model-derived values from Souri et al. (2021). 405 In the period 1-15 April, the net ozone enhancement associated with the lockdown is clearly larger over Central Europe (∼9 % larger and in the band from Benelux to Northern Italy ∼10 % larger) than in the average over the whole month. It also extends further east and south (see Figs. 6b and 6d, Tables 3 and 4). The greater net increase of ozone pollution during the first https://doi.org/10.5194/acp-2021-785 Preprint. Discussion started: 7 October 2021 c Author(s) 2021. CC BY 4.0 License.
by the revised inventories of CAMS (remarked in section 2). In addition, the net reductions of ozone over Northeastern Europe 410 show slightly less pronounced values in this period (-4 %) than over the whole month (-8 %). Figure 7 presents the changes of daily average ozone concentrations associated with the lockdown derived from IASI+GOME2 satellite retrievals, in situ measurements and the CHIMERE model (C1 and C2 setups). We notice that the spatial patterns of the surface ozone anomalies are qualitatively similar for daily averages (Fig. 7b) and MDA8 (Fig. 6b), while their amplitude are roughly a factor 2 smaller for daily averages (see Table 2). The satellite-derived estimate of lowermost tropospheric ozone 415 changes during 1-15 April 2020 (Fig. 7a) is clearly consistent with that derived from surface in situ measurements (Fig. 7b), both adjusted for avoiding the effects of meteorological conditions. The location of the positive and negative anomalies of ozone satellite retrievals is like those of total changes between 2020 and 2019 (Fig. 1a). Although, as for surface concentrations, the satellite-derived amplitude of the ozone enhancements linked to the lockdown over Central Europe is clearly less pronounced than the total changes. Clear lockdown-derived enhancements of LMT ozone retrievals are seen over Model-only estimations of the impact of the pandemic lockdown on ozone pollution (Fig. 7c-d) show both qualitative similarities and differences with respect to those using observations. Similar patterns of ozone enhancements up to ∼5 ppb are remarked over Eastern England, Northern France, the Benelux and Northern Italy for observation-derived methods and 430 CHIMERE C1 (Fig. 7c). This last one suggests a net ozone enhancement over the part of the Mediterranean Sea located south of France, which is associated with the reduction of shipping activities in this area and is also clearly observed by IASI+GOME2. On the other hand, the model with the C1 setup simulates enhancements of ozone pollution over Western England and the North Sea, where observations suggest reductions. Over Germany and Poland, observations suggest the predominance of ozone reductions while CHIMERE C1 indicates a moderate ozone enhancement (1-2 ppb). We can also 435 notice that C1 simulations overestimate the ozone enhancements over Germany and England and underestimates the ones over Northern Italy, as compared to the two observational datasets. This can be partly attributed to the assumed homogenization of the lockdown conditions for these countries, whereas actual restrictions in Germany and England were less strict than in Italy.
In qualitative terms and over land, the horizontal distribution of positive and negative anomalies associated with the lockdown is quite similar for both CHIMERE setups (C1 and C2), but the absolute values are quite different (Fig. 7c and 7d). The 440 amplitudes of the anomalies are approximately a factor 3 to 6 larger for C1 than those for C2. Ozone enhancements for C1 reach ∼6 ppb over Benelux and the Po Valley, while they are of about ∼1 ppb for C2. The net ozone reduction over the between 2020 and 2019 and between urbanized/rural regions (see Fig. 8 and Table 5). In 2019, NO2 concentrations are larger for C1 than C2, by a factor ∼3 over a large heterogenous area as the Benelux to Northern Italy band and only by ∼50 % for a megacity as Paris. In both cases, these simulated concentrations are lower than those measured by in situ measurements (∼20 % larger than values for C1, see Table 5). For all in situ stations, the mean biases (root-mean-squared differences) of coincident NO2 concentrations simulated in 2019 are quite significant (around -5 (6) ppb for C1 and -6 (8) ppb for C2) and similar for 480 both model setups.
We also find larger changes of NO2 abundances in absolute values between 2020 and 2019 for C1 than for C2, both for the band from Benelux to Northern Italy and Paris (by a factor ∼4 and ∼5 respectively). This likely explains most of the differences in the simulated changes of O3 associated with the lockdown conditions in 2020. The larger O3 precursors emissions changes during the lockdown considered for C1 clearly manifest as larger changes in the abundance of O3, as compared to C2. 485 When comparing with in situ measurements of NO2 2020-2019 , the measured reduction over a horizontally homogenous hotspot as Paris clearly matches C1, while it is underestimated by C2 (by a factor ∼5). Inversely, we remark an agreement for C2 for averages over a heterogenous large area (as the Benelux to Northern Italy band) while it is overestimated in absolute values by C1 (by a factor ∼4). Therefore, the agreement between simulated and measured NO2 2020-2019 clearly depends on the criteria and the horizontal homogeneity of the abundancies over the area considered in the comparison. 490 These large differences in the simulated NO2 concentrations are partly linked to the inventories used in each setup. The configuration C1 is based on emission inventories estimated for 2010 (by HTAPv2.2, on monthly basis) and C2 on an inventory calculated for 2017 (by EMEP and with a seasonal modulation). A sustained negative trend for NO2 concentrations observed over Europe between 2010 and 2017 (e.g., Pazminio et al., 2021) suggests a positive bias for the inventory used for C1 as compared to C2. This trend is ∼30 % in France, slightly higher in Italy or Belgium and smaller for other countries as Germany 495 and Poland (∼13%, see EMEP database, https://www.ceip.at/webdab-emission-database/emissions-as-used-in-emep-models).
Other factors significantly affecting simulated concentrations of ozone and its precursors are clearly linked to the meteorological fields used by the model. Indeed, vertical mixing withing the atmospheric boundary layer can largely modify the concentrations of any atmospheric constituent at surface level. We assess its role by comparing total atmospheric columns of NO2 integrated in the vertical for C1 and C2 (Figure 9), which are not directly affected by vertical mixing. We notice that 500 both total columns in 2019 and differences between the two years ( Fig. 9) are much closer between C1 and C2 than surface concentrations (Fig. 8). Total columns of NO2 in 2019 are only ∼50 % larger for C1 than for C2, both in large regions and megacities (such as the Benelux to Northern Italy band and Paris). This magnitude of difference can be explained by the differences in the years of the emission inventories and remaining differences in the estimations by HTAP and EMEP. We notice as well that the change of abundance between the two years NO2 2020-2019 is a factor 2 larger for C1 and C2, which can 505 stem from how emissions are modified to account for the pandemic lockdown in each of the setups.
An additional comparison between the meteorological fields used in C1 and C2 is shown in Figure 10 northwest of France to Italy, associated with anticyclonic conditions prevailing in 2020 (Fig. 10a-b). The enhancement of surface pressure in 2020 over the Atlantic and France is larger for C2 than for C1. Surface temperature changes between 2020 510 and 2019 are also similar for C1 and C2, except over France where meteorological conditions considered in C1 suggest a clear temperature enhancement in 2020 with respect to the previous year, while those for C2 only show a slight increase and only located in southwestern part of this country. These regional differences in meteorological conditions over France may partly explain (if one accepts a positive temperature-ozone relationship, as often observed during summer) why C1 simulates a clear ozone enhancement all over this country in 2020 with respect to 2019 and C2 shows a very limited change in ozone abundance 515 in this case (see Fig. 4).

Conclusions
We have presented a comprehensive analysis using in situ and satellite measurements of ozone as well as chemistry-transport An additional original aspect in the present analysis is the adjustment of both in situ and satellite observations for accounting for the main effects of the changes in meteorological conditions between the two years. This adjustment is derived from the chemistry-transport model CHIMERE using the meteorological conditions of each year but the same standard (business as usual) anthropogenic emissions. This method relays on the model accuracy in standard conditions and it neglects possible 530 feedbacks between meteorological conditions and photochemical regimes. The influence of biogenic emissions is accounted for by the method, as these sources are derived according to the meteorological conditions. The adjustment can be directly applied to both in situ and satellite data, with good consistency with respect to other independent approaches used for the same kind of adjustment of surface in situ data (Ordóñez et al., 2020) or using integrated process rates derived from a chemical transport model (Souri et al., 2021). 535 Based on meteorology-adjusted satellite and in situ observations, the present work provides a comprehensive estimate of change in ozone pollution associated with the pandemic lockdown. It shows a significant enhancement of ozone in the VOClimited regions of central and northern Europe and the Po Valley, pointed out previously by models and in situ surface data (e.g. Menut et al., 2020;Ordóñez et al., 2020;Souri et al., 2021)  We compare these observation-based estimations with model-only estimates derived by changing emissions according to the reductions in anthropogenic activities estimated for the lockdown period. For this, we use two different model configurations, with distinct emissions in standard and COVID-19 conditions and meteorological fields. These two model-based estimates provide similar regional patterns of enhancements and reductions of surface ozone, but clearly different magnitudes (by a 545 factor 3 to 6). This is mainly explained by marked differences in vertical mixing within the atmospheric boundary layer, according to the use of two different meteorological models, as well as the year for which the emission inventories were estimated (one simulation uses an inventory from 2010 and the other from 2017) and different assumptions in variations of anthropogenic emissions during the lockdown. As compared to measurements, both configurations underestimate the amplitude of the positive and negative anomalies. However, the configuration that shows amplitudes closer to those observed 550 are the ones whose abundances of ozone precursors (e.g., NO2) in absolute terms (in standard conditions and changes during the lockdown over homogenous hotspots) are closer to those observed by in situ sensors. The differences between the simulations using two different model setups, and with respect to observations highlight the complexity to simulate the effect of the changes in ozone concentrations due to changes in anthropogenic emissions, as occurred during the lockdown.  in Europe (rectangles in Fig. 3a), derived from surface in situ measurements. Total, meteorological and lockdown-associated changes are considered in mixing ratio units (ppb), for daily averages and maximum daily 8-hour running averages (also percentage in italics). Values after ± indicate one standard deviation.  Table 5. Comparison of nitrogen dioxide concentrations in the period 1-15 April 2019 and the changes between this period in 725 2020 with respect to 2019, for 2 target regions in Europe (rectangles in Fig. 3a), derived from surface in situ measurements and model simulations with the configurations C1 and C2. Surface concentrations and vertically integrated total columns are compared. Values after ± indicate one standard deviation. For lack of representativity, model standard deviation within the Paris sector is not given.  Therefore, the temporal and spatial sampling of both datasets is the same.    of (a) and (d) idem of (c) but for CHIMERE with the configuration C2.