The NO2 tropospheric columns in the Île-de-France region are measured by two ground-based SAOZ instruments (Pommereau and Goutail, 1988) that are part of French research infrastructure of ACTRIS (Aerosols, Clouds and Trace gases Research Infrastructure). The first one was installed in 2005 at the observation platform of QualAir (http://qualair.aero.jussieu.fr/, last access: January 2021) at the Sorbonne University in Paris (urban station) and the second one has been operational at the LATMOS (Laboratoire Atmosphères, Observations Spatiales) laboratory in Guyancourt (southwestern suburban station) since 2011. SAOZ is a UV-Visible spectrometer primarily designed for monitoring the stratospheric ozone and NO2 during twilight observations in the frame of the NDACC (Network for the Detection of Atmospheric Composition Change; see Hendrick et al., 2011, for a description of retrieval). The long-term data series of SAOZ instruments were compared with data from most satellite missions to validate or monitor their performance. For example, SAOZ instruments participated in the validation of the latest satellite mission (Sentinel-5 Precursor) launched on October 2017 for the measurements of ozone (Garane et al., 2019) and stratospheric NO2 (Verhoelst et al., 2021) columns.

During the day, SAOZ observations are sensitive to increased tropospheric NO2 amounts in polluted regions (Tack et al., 2015). Every ∼ 2 min, the sunlight backscattered by the atmosphere in the zenith direction of SAOZ is acquired, and the DOAS (differential optical absorption spectroscopy) method (Platt and Stutz, 2008) is applied in the NO2 absorptions bands to obtain the respective slant column densities. The stratospheric NO2 columns are removed from slant columns to retrieve the tropospheric NO2 for solar zenith angles (SZAs) lower than 80∘ (see Dieudonné et al., 2013, for a detailed description of the SAOZ tropospheric NO2 retrieval). The SAOZ data set of tropospheric NO2 measurements at Paris was used in different studies to relate the NO2 concentrations at the surface with the integrated NO2 column in the boundary layer (Dieudonné et al., 2013) to interpret ozone measurements (Klein et al., 2017) and the seasonal cycle of the ozone gradient (Ancellet et al., 2020).

SAOZ tropospheric NO2 columns are available at the SAOZ web page (http://saoz.obs.uvsq.fr/SAOZ_tropo_Paris.html, last access: 1 January 2021 and http://saoz.obs.uvsq.fr/SAOZ_tropo_Guyancourt.html, last access: 1 January 2021). These data were averaged daily between 06:00 and 18:00 UT and between 11:00 and 14:00 UT for comparison with satellite observations.

Tropospheric NO2 columns retrieved by TROPOspheric Monitoring Instrument (TROPOMI) aboard Sentinel 5 Precursor (S5P) satellite (Veefkind et al., 2012) launched in October 2017 were also used to discriminate air masses above SAOZ instruments benefiting from the high spatial resolution of this instrument (3.5×7 km2 and 3.5×5.5 km2 since August 2019). TROPOMI is a passive-sensing hyperspectral nadir-viewing imager, aboard a near-polar sun synchronous orbit satellite at an altitude of 817 km, with an overpass at 13:30 local time and practically daily global coverage.

Retrieval applied on TROPOMI data allows the distinction between tropospheric, stratospheric and total NO2 columns. The algorithm was adapted from the DOMINO/TEMIS (Dutch OMI NO2/Tropospheric Emission Monitoring Internet Service) approach for the ozone monitoring instrument (OMI; Boersma et al., 2007, 2011), based on the DOAS method to obtain slant column densities (SCDs) of NO2 that are assimilated to the TM5-MP chemical transport model (CTM) to separate the SCD. The CTM runs using 0–12 h forecast meteorological data from the European Centre for Medium-Range Weather Forecasts (ECMWF) corresponding to the offline product. Finally, each slant column is converted to vertical column using the precalculated air mass factor (AMF) look-up tables. A detailed description can be found at the TROPOMI web page (http://www.tropomi.eu/data-products/nitrogen-dioxide, last access: January 2021).

Van Geffen et al. (2020) analyzed the uncertainties of the SCD of TROPOMI and compared them to OMI-QA4ECV data (Boersma et al., 2018). They show a very good agreement over a remote Pacific Ocean sector, with a correlation of 0.99, but values with 5 % higher than the OMI-QA4ECV ones. Verhoelst et al. (2021) compared NO2 total, tropospheric and stratospheric columns with the data of ground-based instruments of Pandora, multi-axis differential optical absorption spectroscopy (MAX-DOAS) and zenith-scattered light DOAS (ZSL-DOAS or SAOZ) distributed around the world. Observations from MAX-DOAS were used for tropospheric comparisons since they are sensitive to absorbers in the lowest few kilometers of the atmosphere (Hönninger et al., 2004). A negative bias from 23 % to 37 % is observed in the cases of clean to slightly polluted conditions. In the case of highly polluted areas, the bias can reach 51 %.

TROPOMI tropospheric NO2 columns have been widely used to estimate the reduction in NO2 amounts linked to the lockdown in 2020, which was implemented in different countries to prevent the spread of COVID-19 (e.g., Bauwens et al., 2020; Ding et al., 2020; Liu et al., 2020; Biswal et al., 2021; Koukouli et al., 2021).

In their validation paper against consolidated ground-based data, Verhoelst et al. (2021) used TROPOMI's tropospheric columns of NO2 with a quality assurance (QA) value higher than 0.75 to remove cloudy scenes presenting cloud radiance fraction higher than 0.5, snow- or ice-covered scenes and problems in the retrieval. In our study, we have decided to use a less restrictive threshold of 0.5 in order to enhance the number of days and to avoid biassing the results towards clear-day conditions. This resulted in doubling the number of data taken into account. The monthly mean NO2 tropospheric columns of TROPOMI present a similar seasonal evolution within 2σ for both QA values (not shown).

TROPOMI tropospheric NO2 columns are available at the Copernicus web page (https://s5phub.copernicus.eu, last access: March 2021).

TROPOMI tropospheric NO2 measurements in 2020 were widely used to show a decrease in NO2 amounts in different countries, which was attributed to policies restricting human activities by comparing the lockdown and pre-lockdown period or the same period in 2019 (e.g., Ding et al., 2020; Prunet et al., 2020; Siddiqui et al., 2020; Koukouli et al., 2021). SAOZ measurements between 11:00 and 14:00 UT were averaged to match overpass time of TROPOMI above the stations. TROPOMI data were previously filtered for the QA >0.5 (see Sect. 2.1.2) and a radius of 5 km around SAOZ stations. Figure 5 shows the evolution of the monthly mean and two standard errors (2σ) of tropospheric NO2 columns above Paris and Guyancourt stations since January 2019, as observed by SAOZ and TROPOMI (left panels). The standard error corresponds to the standard deviation of the mean divided by the root number of considered days. Similar intermonthly evolution is observed by both instruments, with a generally good agreement within ±2σ and a correlation of 0.80 at Paris and 0.70 at Guyancourt. TROPOMI presents generally lower NO2 values than SAOZ but within the 2σ uncertainty level. This is not the case in May 2020 (month 17 in Fig. 5) during which TROPOMI NO2 amounts are significantly larger at the 2σ level than at SAOZ. Monthly mean values present a seasonal variation, reaching values above 10 Pmolec. cm-2 in winter in Paris, while they vary between 4 and 7 Pmolec. cm-2 in Guyancourt. The first months of 2020 present lower values compared to 2019, mostly due to weather conditions, while the March–May NO2 decrease (month 15–17) is coincident with the lockdown period. A histogram of the differences between TROPOMI and SAOZ is also shown in Fig. 5 (right panels). A mean and median difference of -0.2 and +0.12 Pmolec. cm-2, respectively, is obtained at the Paris station and of -0.6 and -0.7 Pmolec. cm-2, respectively, at Guyancourt. It corresponds to a median relative difference of 2 % at the Paris station and -22 % at Guyancourt. The dispersion of the difference represented by the half of the 68 % interpercentile (IP68/2) is 2.9 and 1.6 Pmolec. cm-2, respectively, at Paris and Guyancourt.

TROPOMI and SAOZ data selected for days with t>30 min were averaged between 11:00 and 14:00 UT for the period of the 2020 lockdown in France (17 March to 10 May), and median values were computed from the SAOZ and TROPOMI data for the 2011–2020 annual range (Fig. 6). TROPOMI NO2 decrease in 2020 compared to 2019 is 35±12 % for Paris and 22±27 % for Guyancourt. Bauwens et al. (2020) found a decrease of 28 % during the first 21 d of lockdown over a 50 km region, centered over Paris, using TROPOMI and OMI data compared to same period in 2019. A larger tropospheric NO2 decrease of about 47 % is found from SAOZ observations between 2019 and 2020 at both studied stations (see Fig. 6). Prunet et al. (2020) found an even larger decrease in NO2 values, varying from 52 % to 86 %, during the lockdown in a 120 km region around Paris using yearly 2019–2020 TROPOMI data and the city-scale NO2 plume mass method.

It should be noted that the SAOZ data sets have shown a long-term negative trend since 2011. Font et al. (2019) have used in situ data to study the impact of policy initiatives in different megacities. They have shown a mean NO2 decrease in roadside (background) sites of -2.9 (1.7) % yr-1 in Île-de-France for the 2010–2016 period, linked to the introduction of the Euro V regulations for heavy-duty vehicles in October 2009; other policies were implemented thereafter (e.g., Euro VI regulations in 2014). The trend of tropospheric NO2 amounts needs to be considered to better quantify the effects of lockdown on air pollution, which cannot rely on the comparison with a single reference year as was done in many other studies (e.g., Bauwens et al., 2020; Prunet et al., 2020).

To better account for traffic-related pollution events in the daily averaged NO2 columns, the full daytime data of tropospheric NO2 measurements by SAOZ (SZA <80∘) of the corresponding day were considered. The median value of the daily columns with t>30 min was computed for each year during periods 2 and 3 above Paris and Guyancourt. Periods 1 and 4 were not considered since only 1 d with t>30 min was observed above the stations during these periods in 2020. Period 3 was restricted to 11 May–15 July (period 3') to avoid the effect of NO2 seasonal variations in the final median value. A robust regression fit (reweighted bisquare function to reduce weight of outliers far ∼ 5 times from the median) was applied to period 2 and 3' to compute the trend for the 2011–2019 period. We will focus only on the period of lockdown since important NO2 interannual variability in the period 3' does not present a 2σ significant slope value neither at Paris, nor at Guyancourt. Only the lockdown period presents a significant negative slope of -1.51±0.48 (1σ) Pmolec. cm-2 yr-1 at Paris and -1.42±0.14 (1σ) Pmolec. cm-2 yr-1 at Guyancourt, as shown in Fig. 7. These values correspond to a negative trend of -5.86±1.92 % yr-1 at Paris and -6.79±0.66 % yr-1 at Guyancourt relative to 2011. Previous studies have presented similar values over western Europe. Zhou et al. (2012) found significant negative trends in the 2004–2009 period, varying from -4 % yr-1 to -8 % yr-1, using OMI tropospheric NO2 columns. Curier et al. (2014) computed the trend from the synergistic use of OMI NO2 tropospheric columns and the chemistry transport model LOTOS-EUROS, finding significant negative trends of 5 % yr-1–6 % yr-1. The year 2020 presents the lowest values of NO2 at both stations (5.4 Pmolec. cm-2 at Paris and 4.4 Pmolec. cm-2 at Guyancourt) that are significantly different, at 1σ, from previous years (Fig. 7). The median value in 2020 is lower than the extrapolated value, using the computed 2011–2019 trend, by 55.6±15.7 % at Paris and 45.6±11.8 % at Guyancourt. If the tropospheric median column of NO2 in 2019 had been used as a reference for comparison, slightly higher declines would have been obtained within ±1σ:56.7±9.1 % and 52.6±14.5 % at Paris and Guyancourt, respectively. Choosing other reference years would obviously yield different results, e.g., a slightly lower value at Paris (55±10.7 %) and an even higher value at Guyancourt (58.9±12.5 %) when using the year 2018 as a reference (Fig. 7). Moreover, choosing earlier years as a reference would pose the problem of NO2 variability factors associated with both the lockdown and the long-term NO2 reductions. This confirms the advantage of our method that calculates the reference from a decadal database and corrects for the long-term trend. It should be noted that the data filtering procedure based on meteorological conditions (wind speed and direction) significantly changes the result of the NO2 reduction estimate in Guyancourt, making it statistically insignificant (9.7±41.6 %) if filtering is not applied; at the same time, the estimate for Paris has not changed much (58.3±20.9 %). Table 3 presents a summary of the NO2 reductions in 2020, using different data sets described previously in the text. This indicates that the results at the Paris site located in the center of the agglomeration are not dependent, in 2020, on meteorological conditions. On the contrary, for the Guyancourt site at the edge of the agglomeration, selecting the days when the site is impacted by emissions within the agglomeration is crucial.

The annual median NO2 concentration at Airparif stations, since 2011 (Table 1), were computed from daily available hourly data during the lockdown period, filtered for the wind speed and direction as it has been done for the tropospheric NO2 column (t>30 min). Figure 8 presents the interannual variability in the NO2 concentration at the five Airparif stations. In addition, the calculated robust fit for the decadal evolution at each station is shown. The background or urban stations (Paris 7 and 13) present similar interannual variability, with higher values at Paris's seventh district. The station of Quai de Célestins, in close proximity to local traffic, shows much higher values which are significantly different from those at other urban sites. The suburban station of Versailles presents similar values to Paris's 13th district at ±1σ. The observation station located at the Eiffel Tower at 300 m height near Paris's seventh district station shows the lowest values.

The five Airparif stations present negative trends from -3 to -1.3 µg m-3 yr-1, equivalent to -4.6 % yr-1 to 2.4 % yr-1 (Table 4). Font et al. (2019) found a similar negative trend, varying from -3.4 % yr-1 to -2.4 % yr-1, for roadside stations in Paris for the 2010–2016 period. These trends appear to be less negative than those obtained from column measurements. Possible reasons for this are an increase in the NO2 to NOx emission ratio and a limitation of the available amount of O3 for the NO to NO2 conversion. Both factors affect the surface concentration than the boundary layer column more strongly, which could lead then to the different trend estimates.

Incomplete NO to NO2 conversion is, for example, suggested by NO2 and ozone concentrations of the same order of magnitude at Paris's urban background sites (Fig. 38 in Airparif, 2020). In such a situation, the NO2 trends are both impacted by the NOx emission and ozone trends. Figure 38 in Airparif (2020) cited above shows a strongly increasing ozone average urban background over Paris, e.g., 35 to 43 µg m-3, respectively, for the 2007–2009 and 2017–2019 periods. This positive ozone trend buffers, to some extent, the negative NOx emission trend.

However, while this reasoning would qualitatively explain the differences in trends between column and in situ measurements, it fails to explain differences in trends between different in situ sites in the sense that larger NOx values would lead to smaller negative trends. This is not observed; on the contrary, the NO2 trend is more negative at base of the Eiffel Tower than at altitude when NOx becomes lower. Thus, the exact explanation of the differences in trends at different sites and heights still needs more investigation. In 2020, significant decreases, compared to the extrapolated value using the above-calculated linear trends, are observed at all stations and reach similar median values, which are slightly higher for the traffic station and slightly lower for Eiffel Tower observation station. The relative values of NO2 reductions are shown in Table 4. Comparable values at 1σ are observed for traffic and urban stations in Paris, with lower values at Paris's 13th district, where the standard error is higher. Nevertheless, the reduction in NO2 concentrations observed in absolute values is more important at traffic stations (such as CELES – Quai de Célestins) compared to the urban station (such as Paris's seventh district). The observation station installed at the Eiffel Tower at 300 m height presents a 53 % reduction that is identical to the station at Paris's seventh district, which is located at the base of the tower. The suburban station of Versailles presents the lowest reduction of 28.5 %, which is significantly different to other stations at 1σ, except for Paris's 13th district. It should be noted that both stations show an almost twice as large standard deviation of 14 %. The reasons for these lower values are not clear. It can be speculated that, at this suburban site, the relative contribution of residential heating to NOx sources is stronger than at Paris sites, and probably, these sources increased during the lockdown period due to the presence of people in their homes (Menut et al., 2020).

Collivignarelli et al. (2021) compared the NO2 concentration observed by the traffic and urban stations of Airparif during the lockdown in 2020 to the same period in previous years (2017–2019). They found a decrease of 15 % for the urban stations and 33 % for traffic stations. However, when considering similar meteorological conditions with respect to rainfall, temperature and wind speed, the authors found a reduction of 51.5 % corresponding to traffic stations and approximately 45 % for background ones, similar to values obtained in this study.