An original aspect of this work is the analysis of the diurnal evolution of total column observations of ammonia derived from OASIS. This remote sensing observatory is located in Créteil (OASIS; 48.79∘ N, 2.44∘ E; 56 m a.s.l., above sea level), in the southeastern suburbs of Paris, on the rooftop of the Université Paris-Est Créteil (UPEC; Chelin et al., 2014). It is an urban site mainly affected by background levels of pollution (Fig. 1).

Measurements from other sites over the Paris region are also used in the current study (Fig. 1). Meteorological and detailed atmospheric composition data at the surface level are measured at the “Site Instrumental de Recherche par Télédétection Atmosphérique” supersite near Palaiseau (SIRTA; 48.72∘ N, 2.20∘ E; http://sirta.ipsl.fr, last access: 29 January 2019), located about 19 km southwest from OASIS and southwest of Paris, which is often used for monitoring background air quality conditions in the Paris region (Haeffelin et al., 2005). We use radiosounding measurements of temperature, pressure and humidity profiles from the Trappes station that is about 31 km west of Créteil (15 km away from the SIRTA supersite) and operated by Météo-France. Additional surface measurements of PM2.5 and PM10 (particle matter with aerodynamic diameters, respectively, less than 2.5 and 10 µm) are provided by the Airparif network dedicated to monitoring air quality in the Paris region (https://www.airparif.asso.fr/, last access: 17 January 2019) from the stations of Vitry-sur-Seine, Bobigny and Gennevilliers. In the paper, time series of measurements are presented in terms of hourly median values, except when stated otherwise.

Since 2009, the OASIS observatory regularly records high spectral measurements of solar radiation absorbed and scattered by atmospheric constituents, under clear-sky conditions (Chelin et al., 2014). It uses a medium-spectral-resolution Fourier transform spectrometer manufactured by Bruker Optics (the Vertex 80 model) with a spectral resolution of 0.06 cm-1 (corresponding to a maximum optical path difference of 12 cm). OASIS is routinely used for monitoring air pollutants, such as tropospheric ozone (O3) and carbon monoxide (CO), with good accuracy and high sensitivity to near-surface concentrations (Viatte et al., 2011). This system is particularly suited for air quality monitoring in megacities, given its compactness and moderate cost, and it can play a key role in validating current (e.g., IASI) and future satellite observations (e.g., Infrared Atmospheric Sounder Interferometer Next Generation – IASI-NG – and the infrared sounder aboard the Meteosat Third Generation mission – MTG/IRS).

The observatory is covered by an automatized cupola (Sirius 3.5 “School Model” observatory, 3.25 m high and 3.5 m in diameter), in which the aperture rotates to track the solar position. The altitude–azimutal solar tracker of OASIS uses bare gold-coated mirrors (A547N model from Bruker Optics). Infrared solar radiation spectra are recorded by a DTGS (deuterated triglycine sulfate) detector using a potassium bromide (KBr) beam splitter to cover the large spectral region from 700 to 11 000 cm-1 (0.9–14.3 µm) with no optical filter. The acquisition system is set to average over 30 scans at maximum spectral resolution in order to increase the signal-to-noise ratio of the measurements. This averaging procedure results in an effective temporal resolution of 10 min, that allows for measuring the diurnal variability of relatively short-lifetime species such as NH3. Absolute calibration of spectra measured by OASIS is done every month with a reference internal source of radiation.

Ammonia concentrations integrated over the total atmospheric column are retrieved with the PROFFIT 9.6 code developed by the Karlsruhe Institute of Technology (Germany; Hase et al., 2004), adapted for the medium spectral resolution. As detailed by Tournadre et al. (2020), two spectral microwindows within the ν2 vibrational band of NH3 are used: 926.3–933.9 and 962.5–970 cm-1. The main interfering species in this spectral range are water vapor (H2O), carbon dioxide and O3, whose abundances are taken from the Whole Atmosphere Community Climate Model (WACCM version 6: Chang et al., 2008) and jointly adjusted with that of NH3. We also use climatological concentrations for minor interfering gases (i.e., nitric acid, HNO3; sulfur hexafluoride, SF6; ethane, C2H4; and chlorofluorocarbons – e.g., CFC-12) that may essentially impact the baseline of the spectra. The spectral signatures of absorption of infrared radiation by ammonia are clearly seen in individual spectra measured by OASIS, such as those recorded during a pollution event in March 2012 (as compared to the atlas from Meier et al., 2004; see Fig. 2). Atmospheric columns of ammonia derived from the 9-year database of OASIS range from 0.0005×1016 to 9×1016 molec. cm-2 (molecules per square centimeter), and their retrieval error is estimated to be 20 %–35 % (Tournadre et al., 2020), dominated by the systematic errors that are the combination of uncertainties in the spectroscopic parameters of ammonia and the interfering species (the dominating term), radiometric noise, instrumental parameters, and forward model uncertainties. The magnitudes of these errors are comparable to those estimated by Dammers et al. (2015) for a high-resolution ground-based station at Bremen (Germany). OASIS retrievals of NH3 total columns show a good agreement with co-located observations derived from IASI (the ANNI-NH3-v2.2R version; Van Damme et al., 2017): a linear correlation coefficient of ∼0.8 and a small mean difference of ∼0.08×1016 molec. cm-2, with OASIS-derived concentrations slightly larger (Tournadre et al., 2020). This last aspect could be associated with an enhanced sensitivity to larger concentrations of NH3 near the surface for OASIS as compared to the satellite retrieval which is most sensitive to higher atmospheric layers.

In the present analysis, we use in situ gaseous ammonia measurements at surface level carried out with an AiRRmonia instrument (Mechatronics Instruments, the Netherlands) at the SIRTA observatory (Haeffelin et al., 2005). The principle of this instrument, described in Cowen et al. (2004), is essentially based on conductimetric detection of ammonia that is first absorbed via a gas-permeable membrane and dissolved in water (i.e., in the form of ammonium ions, forming acidic solution). Several intercomparison exercises have shown that this procedure provides more accurate NH3 measurements (Norman et al., 2009; von Bobrutzki et al., 2010). The AiRRmonia was regularly calibrated with 0 and 500 ppb ammonium solution.

Concomitant measurements of the major chemical composition of submicron aerosols were performed with an aerosol chemical speciation monitor (ACSM; Aerodyne Research Inc., Billerica, MA, USA; Ng et al., 2011), providing concentrations of particulate organic matter (OM), nitrate (NO3-), sulfate (SO42-), ammonium (NH4+) and chloride (Cl-), every 30 min. Submicron particles are sampled at 3 L min-1, subsampled at 0.85 L min-1 and focused through an aerodynamic lens for PM1 (particle matter with aerodynamic diameter smaller than 1 µm). Non-refractory particles are then flash-vaporized on a 600 ∘C heated plate, fragmented by electronic impact at 70 keV, and eventually separated and detected by a quadrupole. Calibrations were performed by injecting known concentrations of ammonium nitrate and ammonium sulfate particles with an aerodynamic diameter of 300 nm. Details on the operational conditions of the AiRRmonia and the ACSM instruments at SIRTA are provided by Petit et al. (2015).

As observed by Petit et al. (2015), we expect the daily evolution of ammonia over the Paris region to be closely linked to the gas-to-particle conversion between ammonia (gas) and ammonium nitrate particles. This is a reversible conversion for which the equilibrium is closely linked to the abundance of precursors (NH3 and HNO3) and meteorological conditions, such as temperature and relative humidity (Seinfeld and Pandis, 2016). The conditions needed for volatilization of NH3 from NH4NO3 are given by the relationship of relative humidity and deliquescence relative humidity (DRH), which depends on temperature, whereby volatilization is favored when RH is much lower than DRH. In order to estimate the balance between DRH and RH, we consider the following equation, as suggested by Seinfeld and Pandis (2016): 1DRH(T)=DRH(298)exp⁡ΔHSRA1T-1298-Bln⁡T298-C(T-298), where DRH(298) is the deliquescence relative humidity of NH4NO3 at 298 K, which corresponds to 61.8 %. ΔHS is the enthalpy of solution for NH4NO3 at 298 K which is 25.69 kJ mol-1, R is the universal gas constant and T is the temperature in kelvin. A, B and C are factors for the solubility of common aerosol salts in water as a function of temperature provided by Seinfeld and Pandis (2016) (i.e., 4.3, -3.6×10-2 and 7.9×10-5, respectively). Moreover, partitioning between ammonia and ammonium nitrate is also influenced by the pH of the ambient particles (e.g., Weber et al., 2016; Guo et al., 2018). When pH drops below an approximate critical value of 3 (slightly higher in warm and slightly lower in cold seasons), the NH3 reduction leads to evaporation of NH4NO3, while this is not expected to happen for moderately acidic to neutral conditions (Guo et al., 2018). In addition, it is worth noting that in the present study we use the above expression and the currently available data for a qualitative interpretation of diurnal variations of ammonia and ammonium. However, additional dedicated measurements throughout the atmospheric column are needed in order to perform a quantitative analysis (see more details in the conclusion section).

For characterizing the pollution event during March 2012, we use a suite of satellite and model datasets concerning both the pollutant distributions at regional and continental scales and meteorological conditions. Aerosol optical depth (AOD) data derived from satellite and ground-based measurements are used for analyzing the spatial and temporal evolution of total particle abundance integrated over the atmospheric column. The horizontal distribution of AOD over western Europe is described using MODIS (Moderate Resolution Imaging Spectroradiometer; Remer et al., 2005) data aboard the Terra (MOD04L2) satellite with overpasses at 10:30 LT (from the NASA Worldview website https://worldview.earthdata.nasa.gov/, last access; 27 February 2019; Levy and Hsu, 2015). The MODIS images have a horizontal resolution of 3 km at nadir.

The horizontal distribution of air pollutants at the European scale is studied with CHIMERE chemistry–transport model simulations of PM2.5 provided by the ESMERALDA (EtudeS Multi RégionALes De l'Atmosphère; Cortinovis et al., 2006) project (http://www.esmeralda-web.fr/accueil/index.php, last access: 28 February 2019). The version 2008b of CHIMERE is run hourly and averaged at a daily timescale, with a horizontal resolution of 15 km × 15 km and nine vertical levels between 20 m to 5 km. Meteorological inputs for CHIMERE come from MM5 simulations (Dudhia, 1993), using Final (FNL) analysis data from National Centers for Environmental Prediction (NCEP) as boundary conditions. Chemical reactions are simulated using the MELCHIOR2 mechanisms scheme and the ISORROPIA model (Nenes et al., 1998). This last one has been used to produce tables that are inserted in the CHIMERE model for calculating the thermodynamic equilibrium of the species. Ammonia, nitrate and sulfate are simulated in aqueous, gaseous and particulate phases in the model.

Meteorological conditions are analyzed from in situ measurements and numerical model simulations. We use sea-level pressure, wind and potential temperature fields from the ERA-Interim (ERAI; Simmons et al., 2007) reanalysis of the European Centre for Medium-Range Weather Forecasts (ECMWF) that are provided by the Institut Pierre-Simon Laplace Mésocentre (https://mesocentre.ipsl.fr, last access: 6 March 2019). These simulations have a 0.75×0.75∘ horizontal resolution and 37 pressure levels.

Meteorological conditions at the surface over the Paris region are analyzed by in situ measurements of wind speed and direction performed at the SIRTA site (Haeffelin et al., 2005). Local temperature and relative humidity were measured at Créteil with a LOG 110-EXF sensor, with an accuracy in temperature of ±0.5 ∘C and in relative humidity of ±3 %.

Vertical profiles of temperature and relative humidity from the surface up to 25 km of altitude and with about 10 m vertical resolution are measured by radiosoundings launched around noon and midnight at the Trappes site (southwest suburbs of Paris).

The diurnal evolution of particle pollution over the Paris region is studied in terms of surface measurements of PM2.5 and PM10 from several Airparif sites and AOD measured by ground-based sun photometers (version 3 of level 2.0 data) at the Paris and SIRTA sites from AERONET (AErosol RObotic NETwork; Holben et al., 2001, https://aeronet.gsfc.nasa.gov/, last access: 9 June 2019). We use the distinction between AOD from a fine (e.g., smoke or smog) and coarse (e.g., sea-salt or dust) modes at 500 nm, derived from the wavelength dependence of the AOD (O'Neill et al., 2003; Giles et al., 2019). Errors in AOD data correspond to approximately 0.02 (Giles et al., 2019).

Additionally, we use ground-based lidar measurements from the SIRTA site for describing the vertical distribution of particles over the Paris region, which is used as an indicator of the vertical distribution of air pollutants and the vertical structure of the atmospheric boundary layer. This is done with vertical profiles of attenuated backscatter profiles, measured by an elastic backscatter lidar (the Leosphere ALS model) at 355 nm. The mixing boundary layer height is visually identified as the lowest marked discontinuity of the lidar profiles during daytime hours (from 06:00 to 18:00 UTC).

During late March 2012, the prevailing atmospheric conditions over western Europe are driven by an anticyclonic high-pressure system centered over Great Britain and the North Sea (55∘ N, 0∘ E) on 26 March and moving westwards in the following days (see Fig. 3a, d and g). Following the anticyclonic circulation associated with this system, northeasterly winds blow from the Benelux region (Belgium, the Netherlands and Luxembourg) to northern France. As expected for an anticyclonic period, relatively low wind speeds occur at its core, located over central Europe (from southern France to eastern Germany), which are accompanied by high insulation and low cloudiness (not shown). According to MODIS satellite observations (Fig. 3b, e and h) and CHIMERE simulations (Fig. 3c, f and i), an aerosol plume with moderate AOD (0.2 to 0.3) and moderately large concentrations of PM2.5 at the surface (20 to 30 µg m-3) is formed on 26 March over Benelux and extends across the English Channel. Meanwhile, aerosol baseline levels are observed over northern France (AOD ∼ 0.1 and 10–15 µg m-3 for surface PM2.5). After 27 March, the aerosol plume reaches northern France and southern England. On 28 March, a clear enhancement of the aerosol load over Benelux and northern France is observed both in terms of AOD (up to 0.4) and modeled surface PM2.5 concentrations (up to 50 µg m-3). These high aerosol loads over northern France remain until 30 March (not shown). Both the horizontal extent of the aerosol plume and wind directions suggest that these highly polluted air masses originate over Benelux as well as western Germany and are transported southwestwards, clearly reaching the Paris region after 28 March (also remarked for this pollution event by Fortems-Cheiney et al., 2016).

Over the Paris region, particle concentrations at the surface are moderately high during 26–27 March (PM2.5 concentrations up to 40 µg m-3; see surface measurements of PM levels from several stations of the Paris region in Fig. 4). As polluted air masses are advected from Benelux and western Germany during 28–29 March, PM2.5 levels are clearly enhanced (up to 80 µg m-3), as also seen in daily averaged simulations. Figure 4 also shows the largest peaks of surface PM2.5 concentrations occurring every day during the morning and secondary high values in the late evening.

Very similar temporal evolution patterns of surface particle concentrations are observed over the whole Paris region and during the entire period (26–30 March), both in absolute and relative terms. Figure 4 illustrates this horizontally homogeneous distribution of surface PM as the chosen stations are located at the southeastern, southwestern, northeastern and northwestern suburbs of Paris (Fig. 1). The same peaks and troughs of surface PM are seen for all these locations. Particularly, we also remark that PM1 at SIRTA also shows the same temporal evolution as other stations in the Paris region but with levels roughly ∼30 % below those of PM2.5 during 26–28 March and similar concentrations afterwards (for both PM1 and PM2.5). The 30 % difference between PM1 and PM2.5 observed in the present case could be linked with aging (and/or long-range transport) which has been remarked for measurements in 2015 by Petit et al. (2017). In the Paris region, PM1 generally represents 90 % of PM2.5 (Petit et al., 2017), particularly when PM1 is larger than 20 µg m-3 (although for lower levels, PM1 may represent around 50 % of PM2.5; Petit, 2014). Occasionally, some background levels of PM might not be accounted for in PM1 that are measured as PM2.5 (Petit, 2014). Moreover, comparisons made by Petit (2014) show a very similar statistical distribution for both PM1 at SIRTA and PM2.5 at the urban background stations in Paris suburbs mentioned in Fig. 4. For the future, it should be very interesting to have co-located PM1 and PM2.5 chemical composition measurements. The clear similarity of these measurements at four different locations of the Paris suburbs suggests that we may also expect a consistent evolution of pollution levels at the Créteil site (OASIS observatory), whose observations are also used later in this section for analyzing the evolution of the atmospheric ammonia concentrations during the event.

The time series of surface PM levels suggest the occurrence of two distinct pollution regimes within the period of 26–30 March. Indeed, while daily mean PM2.5 values during 26–27 March remain under the air quality 24 h guideline of WHO (World Health Organization, PM2.5 of 25 µg m-3, except for one station on 1 single day), this PM2.5 threshold is exceeded for all stations during 28–30 March. Hereafter, these two regimes are named period 1 or P1 (26–27 March) and period 2 or P2 (28–30 March). These two different atmospheric conditions are also pointed out by Petit et al. (2015) by analyzing this particular pollution episode using surface measurements at SIRTA. A statistical comparison of the similarity of surface PM measurements from different sites over the Paris region is shown in Table 1 (for periods 1 and 2). For the first period (26–27 March), the time series of hourly PM2.5 measurements performed at three different locations show a moderate correlation between each other (R2 of 0.63 to 0.67), suggesting a similar evolution but with some horizontal heterogeneity over the Paris region. During the second period (28–30 March), correlations between PM measurements are clearly higher (R2 of 0.86 to 0.91) and therefore indicate a more horizontally homogeneous PM distribution over the Paris region. Levels of surface PM10 for the same stations and periods show similar behaviors (Table 1). This different behavior between P1 and P2 is likely linked to the origin of the pollution event, being rather local for P1 and dominant advection of air pollution from Benelux during P2, as remarked in the regional analysis of AOD, PM and wind regimes of Sect. 3.1. Additionally, we note that comparisons of PM2.5 from three stations to PM1 at SIRTA show moderate correlations during P1 as the other measurements but lower ones during P2 (although peaks and troughs are clearly coincident).

During the P1 and P2, ammonia concentrations over the Paris region are observed both at surface level (using in situ analyzer at SIRTA) and integrated over the total atmospheric column (using OASIS at Créteil, Fig. 5). Total atmospheric columns of NH3 show a very marked and clear diurnal evolution: lower column amounts of ammonia in the morning that rise almost monotonically during the day until reaching a maximum in the afternoon. Both on 26 and 27 March, stable ammonia concentrations around 2×1016 molec. cm-2 remain until noon and then increase only in the afternoon. Early morning total columns of NH3 during 28 and 29 March are lower (respectively, 1.4×1016 and 0.6×1016 molec. cm-2) than for the previous days and show a steady enhancement from the early morning to the afternoon. The highest total column of ammonia is measured on 28 March (4.6×1016 molec. cm-2). On 30 March, the diurnal evolution of NH3 total columns is more similar to the first 2 measurement days (26–27 March). Steady total columns around 1.5×1016 molec. cm-2 during the first 1.5 h of the morning are followed by a decrease down to 0.85×1016 molec. cm-2 around 11:00 UTC and afterwards an increase up to 2.5×1016 molec. cm-2, which is lower than those observed during the 4 previous days. This clear enhancement of ammonia total atmospheric columns during the day measured by OASIS is found to be typical of springtime polluted periods as already analyzed by Tournade et al. (2020) but not shown here (e.g., in March 2014 and March 2016). For all these years, the NH3 maximum in the afternoon is above 2×1016 molec. cm-2 (Tournadre et al., 2020).

Meanwhile, surface measurements at SIRTA show relatively high overall levels of ammonia: from 2 to 10 µg m-3, which is higher than Paris urban background levels of 1–3 µg m-3 shown by Petetin et al. (2016), but for the May 2010 to February 2011 period. On each of the days of the event (26–30 March), morning daily maxima (up to 6–9 µg m-3) and smaller evening peaks (around 5 µg m-3, Fig. 5) are clearly depicted. Although both surface (Fig. 5b) and integrated total column (Fig. 5a) ammonia measurements show large concentrations, their daily evolution patterns are clearly different. While total column values increase steadily during the day until reaching a peak in the late afternoon, surface ammonia moderately fluctuates during the day. These differences may be associated with atmospheric processes or interactions with the surface that modify ammonia concentrations differently as a function of altitude. This may be the case for vertical dilution of atmospheric constituents within the atmospheric boundary layer or the vertical variability of gas–particle partitioning related to relative humidity, temperature and particle pH. These aspects are investigated in detail in the following paragraphs.

Vertical variations of atmospheric ammonia concentrations may potentially be associated with temperature and relative humidity vertical profiles. As mentioned in Sect. 2.3, dry conditions lead to volatilization of ammonia from ammonium particles, whereas humidity levels beyond the deliquescence point favor the inverse process (Seinfeld and Pandis, 2016). During the pollution event during 26–30 March, temperature shows the usual steady decrease with altitude from the surface up to 2.5 km (see the median temperature profile measured by radiosoundings launched at Trappes during 26–30 March, Fig. 6a). Relative humidity varies greatly at the lowest few kilometers of the atmosphere, typically increasing with altitude within the mixing boundary layer (up to 1 to 1.5 km a.s.l., for the present case; see Fig. 6b). During 26–27 and 29 March, relative humidity increases from 25 % at the surface up to 35 %–40 % around 900 m a.s.l. and drops above 1000 m a.s.l. down to 10 %–20 %. On 28 March, relative humidity is roughly 15 % higher than on the mentioned days up to 800 m a.s.l., above which it decreases down to 48 % and then rises up to 60 % at 1600 m a.s.l., dropping down to 30 % at 2500 m a.s.l. In all these cases, relative humidity up to 2500 m a.s.l. remains below the deliquescence point (DRH) as shown in Fig. 6b, thus favoring the formation of NH3 by volatilization of ammonium particles. This is also confirmed by relative humidity time series at different altitudes (200, 500, and 1000 m a.s.l.), reconstructed from all radiosounding measurements over the entire event (launched from Trappes both at midday and midnight, Fig. 6c). This supports a hypothesis of an increase in ammonia amounts due to volatilization of ammonium nitrate at higher altitudes. Relative humidity always remains below the DRH (grey band), except for one single measurement at 1000 m on 30 March at noon. We also remark that during the whole period relative humidity does not vary much vertically below 1000 m (except on 30 March) and that the most humid conditions are found at midnights from 28 to 30 March. No contrasting conditions between midday and midnight are either found for the vertical shape of relative humidity. We do not clearly point out any particular link or concomitant temporal variation of relative humidity every 12 h at different altitudes (Fig. 6c) and ammonia measurements (Fig. 5).

An additional analysis was performed with the ISORROPIA II box model (Fountoukis and Nenes, 2007) to investigate the role of temperature and relative humidity in the partitioning of ammonium nitrate. The forward calculation used measurements of the SIRTA site for NH4+, NO3- and NH3 on 28 March 2012, representing the highest concentrations on the studied period, as well as the meteorological parameters. HNO3 concentrations were set constant from values in Petetin et al. (2016) for the same period of the year. As expected, results indicate that partitioning of ammonia into the particulate phase is favored with the decrease of temperature. This temperature decrease is correlated to an increase of relative humidity (while values remain below DRH). Therefore, in equilibrium conditions (e.g., in absence of ammonia and ammonium advection), ammonia likely decreases at higher altitudes. Nevertheless, it should be noted that pH and aerosol chemical composition also impact ammonium nitrate partitioning. PM1 was found to be neutralized during the period study (Fig. S1 in the Supplement); therefore, we expect a limited influence of ambient particle pH. However, our full understanding is limited by the lack of HNO3 in situ and column measurements.

As a conclusion, the decrease in T and the increase of RH within the boundary layer height of 1–1.5 km with respect to ground shift the equilibrium to the aerosol phase. This does not explain the observed daytime column NH3 maximum, which was not observed at the surface. This suggests that T and RH may not be the only drivers regarding the vertical variability in NH3 concentrations. Other possible drivers are analyzed in the following sections.

A joint analysis of the temporal evolution of ammonia and ammonium particles provides further evidence of the role of particle–gas conversion on the evolution of ammonia concentrations. As previously mentioned, volatilization leads to concomitant increases of ammonia concentrations and decreases of NH4NO3 particles (the most abundant ammonium particles observed during this event; Petit et al., 2015). Complementary, the formation of ammonium nitrate particles may be accompanied by a relative reduction of the abundance of its precursors (if they are not in excess) and thus ammonia and HNO3. The following two subsections analyze these processes for periods 1 and 2.

In this subsection, we use vertical profiles of aerosol distributions measured by a backscatter lidar at SIRTA in order to analyze the link between air pollutant concentrations at the surface, their vertical profile and total column integrated amounts. Aerosol vertical distribution is used here as a proxy for air pollution, since no measurements of the diurnal evolution of the vertical profiles of gaseous pollutants such as ammonia or specific particle components such as ammonium or nitrate are available (only possible through very specific field deployments such as airborne in situ instrumentation or with a diode laser spectrometer aboard weather or tethered balloons with open cavity, for avoiding the problems of the sticky nature for ammonia). We depict the average diurnal evolution over P1 and P2, in terms of lidar measurements, sun-photometer-derived AODs and surface PM2.5 (see Fig. 9a–d). Moreover, we extract the time series of lidar attenuated backscatter at 150 m of altitude (the lowest level at which calibrated lidar measurements are available) for depicting the hourly evolution of near-surface air pollutant content (Fig. 9e and f, blue curves), and also we use attenuated backscatter integrated (indicated as IAB – integrated attenuated backscatter) over the altitude range from 150 m to 2.5 km for analyzing the corresponding variability of total column amounts (note that no aerosol layers are observed above 2.5 km, Fig. 9e and f, red curves).

During P1 (27–28 March), baseline aerosol load conditions prevail over the Paris region, with an overall low AOD of both fine (∼0.1) and coarse (∼0.04) particle fractions (see fine mode AOD at SIRTA in Fig. 9a). This is also shown by lidar measurements, showing attenuated backscatter below 2.5 km of altitude (where particles are located), ranging from 2.5 Mm-1 sr-1 during the day up to a nighttime maximum of 4.3 Mm-1 sr-1 near the surface (Fig. 9c and e). According to the evolution of both attenuated backscatter at 150 m and surface PM2.5 concentrations, near-surface aerosol loads display a relative maximum from 07:00 to 10:00 UTC (thus during the morning peak of road traffic at the Paris megacity; Fig. 9a and e). This is followed by a progressive reduction of near-surface particle amounts from 10:00 until 13:00 UTC, as the atmospheric mixing boundary layer grows from a depth of ∼400 m at 10:00 UTC up to ∼1400 m at 13:00 UTC (shown as a dashed magenta line Fig. 9c). A late afternoon aerosol load increase from 16:00 until 19:00 UTC is also depicted by both integrated amounts (particularly backscatter) and near the surface (a small relative maximum), which corresponds to the time of the evening peak of road traffic. The attenuated backscatter data integrated from 0.15 to 2.5 km show two additional distinct peaks at 07:00 and 19:30 UTC of about 5.2 Mm-1 sr-1, which are associated with enhancements of aerosol content from 700 m up to 1500 m of altitude (probably due to horizontal advection, Fig. 9c).

As previously remarked, P2 is characterized by a strong increase in particle load due to transboundary transport of pollution from Benelux and western Germany (Fig. 3c, f and i). This is reflected by large enhancements of the AOD, surface PM2.5 and lidar backscatter (Fig. 9b and f) as compared to P1. An increase of a factor of 3 of the fine mode fraction of the AOD (up to ∼0.29 and ∼0.32 in Paris and SIRTA, respectively) is observed, while the AOD coarse fraction remains stable (∼0.04 and ∼0.03, respectively, in Paris and SIRTA, not shown in the figures). Integrated attenuated backscatter and surface PM2.5 during P2 are about a factor of ∼2 greater than in the 2 previous days (P1). After a reduction from 07:00 to 11:00 UTC (also observed for ammonia total columns on 30 March, Fig. 5a), integrated attenuated backscatter shows a steady hourly enhancement from 12:00 until 19:00 UTC, when it displays a clear evening peak. This steady increase is also measured in terms of AOD but in this case during all daytime (from 07:00 to 17:00 UTC). This sustained increase during the daytime for vertically integrated amounts of particles is similar to that observed for the vertically integrated amount of ammonia measured by OASIS (Fig. 7b). Meanwhile, the near-surface evolution of particle content shown by both attenuated backscatter at 150 m of altitude and surface PM2.5 is clearly different from that of the total amount of particles (respectively, blue and red curves in Fig. 9b and f). Near-surface aerosol amounts depict both the morning (clearly marked for surface PM2.5) and evening relative maxima, that may be associated with road traffic. The reduction of lidar backscatter at 150 m at 07:30 UTC is likely associated with downward mixing of cleaner air (with fewer aerosols) from the residual layer (Fig. 9d at 07:30 UTC above 200 m) above the mixing boundary layer. The afternoon minor reduction of particles coincides with a slight increase in wind speed, leading to vertical and horizontal mixing. An enhancement of particle amounts near the surface is only seen late in the afternoon (around 17:00–18:00 UTC). Surface ammonia concentrations show morning and evening peaks, with this last one only late in the afternoon. This confirms the consistency of the differences between near-surface and total amounts of particle concentrations with those between surface and total column ammonia.

The lidar profile time series reveals the link between surface and vertically integrated amounts of aerosols. It clearly depicts the typical diurnal cycle of the atmospheric boundary layer, with a growth of the mixing boundary layer from ∼150 m at 06:00 UTC to ∼1500 m at 14:00 UTC (see magenta dashed line in Fig. 9d, likely associated with turbulence generated by sunlight surface heating). Turbulence-associated vertical dilution within the mixing boundary layer is most likely a major reason for the clear reduction of near-surface concentrations of particles between 06:00 and 14:00 UTC (and we expect the same behavior for surface NH3 also mixed within the boundary layer). This reduction is not remarked for vertically integrated amounts (AOD, attenuated backscatter or total column NH3) since a change in the vertical distribution does not affect the total atmospheric amount. We assume that relatively large vertically integrated amounts of particles between 00:00 and 07:00 UTC in P2 (compared to P1) are likely linked to particle formation. This is consistent with the relative humidity conditions (maximum of RH of 95 % above DRH of 67 %, Fig. 8d) and also the low NH3 total columns measured by OASIS in the early morning (07:00 UTC) during P2 (see Figs. 7b and 5a, showing this last one early morning total columns of NH3 during 28–29 March smaller than the previous days).

Moreover, it is worth mentioning that although similarly affected by vertical mixing, the evolution patterns of integrated amounts of ammonia and particles are not necessarily linked to the same phenomena. Indeed, the sustained daytime or afternoon enhancement of particle total atmospheric amounts (AOD and integrated attenuated backscatter) during P2 is more likely associated with the advection of larger amounts of particle pollution during the afternoon or particle formation processes but not to volatilization (which is a sink of particles). However, the afternoon enhancement of ammonia may be reinforced by volatilization during the afternoon drier conditions (see Fig. 8d) and likely also horizontal advection of polluted air masses.

An additional analysis that highlights the major role of vertical mixing for comparing vertically integrated and surface measurements of ammonia is shown in Fig. 10. For both periods, we compare the daily evolution of total column of NH3 retrieved by OASIS with that of surface measurements of ammonia multiplied by the atmospheric mixing boundary layer derived from lidar measurements (magenta lines in Fig. 9c and d). This last one corresponds to the vertically integrated amount of ammonia over the mixing layer for the case of a vertically homogeneous distribution of this gas. We clearly remark a very similar diurnal evolution of these two quantities in relative terms, for both periods. This confirms the good consistency between these two independent measurements of ammonia (total column and surface data). Differences in absolute terms (between 0.5 to 1×106 molec. cm-2) likely come from the evolution of the vertical profile of ammonia, changes with respect to the vertically homogeneous distribution, and also the variability of ammonia abundance in the residual boundary layer and the free troposphere above the mixing layer.

Finally, local or regional emission sources could be a possible explanation for the observed NH3 enhancements during the afternoon. Agricultural NH3 emissions are weak over the greater Paris area, but they are strong over the adjacent Picardie and Ardenne-Champagne regions, located 30 to 150 km upwind of the Créteil measurement site for the northeasterly wind conditions during P1 and P2 (Fortems-Cheiney et al., 2020, Fig. 4c based on detailed emission modeling). The diurnal NH3 emission variation is strongly temperature dependent as shown, among others, by Hamaoui-Laguel et al. (2012) from simulations with a mechanistic emission model (Volt'Air; Génermont et al., 2018). Advection of these emissions to the Créteil site could be rapid enough to explain the observed afternoon NH3 column increase, given the surface wind speed of 3 to 5 m s-1 and probably faster winds at altitude.