Atmospheric ammonia (NH₃) over the Paris megacity: 9 years of total column observations from ground-based infrared remote sensing

In this paper, we present the first multiyear time series of atmospheric ammonia (NH₃) ground-based measurements in the Paris region (Créteil, 48.79^∘ N, 2.44^∘ E, France) retrieved with the midresolution “Observations of the Atmosphere by Solar absorption Infrared Spectroscopy” (OASIS) ground-based Fourier transform infrared solar observatory. Located in an urban region, OASIS has previously been used for monitoring air quality (tropospheric ozone and carbon monoxide) thanks to its specific column sensitivity across the whole troposphere down to the atmospheric boundary layer. A total of 4920 measurements of atmospheric total columns of ammonia have been obtained from 2009 to 2017, with uncertainties ranging from 20 % to 35 %, and have been compared with NH₃ concentrations derived from the Infrared Atmospheric Sounding Interferometer (IASI). OASIS ground-based measurements show significant interannual and seasonal variabilities of atmospheric ammonia. NH₃ total columns over the Paris megacity (12 million people) vary seasonally by 2 orders of magnitude from approximately 0.1×10¹⁶ molec. cm⁻² in winter to 10×10¹⁶ molec. cm⁻² for spring peaks, probably due to springtime spreading of fertilizers on surrounding croplands.

1 Introduction

Ammonia (NH₃) is a reactive and volatile chemical species present in the atmosphere as a trace gas. As the main alkaline atmospheric molecule (Behera et al., 2013), it plays a major role in the formation of secondary fine particulate matter of ammonium salts and thus in particle air pollution (Seinfeld and Pandis, 2006). NH₃ is heterogeneously distributed in the atmosphere, depending on local and regional sources (including agriculture, urban traffic, particle-to-gas conversion, sewage, and industrial activities) and sinks (wet and dry deposition and gas-to-particle conversion; Galloway et al., 2003; Behera et al., 2013; Chang et al., 2016; Van Damme et al., 2018).

In France, atmospheric ammonia (NH₃) is mainly emitted (more than 94 %) by agricultural activities (Ringuet et al., 2016; Genermont et al., 2018) via livestock waste and urine, manure spreading, and the use of synthetic nitrogen fertilizers. Other minor sources exist, including the combustion of biomass and fossil fuels and industrial activity. The source related to motor traffic has increased significantly in recent years due to the increasing use of catalytic (or noncatalytic) NO_x reduction systems on light and heavy-duty vehicles. These devices use an injection of urea or ammonia and can give rise to NH₃ emissions (Chang et al., 2016). The urban area corresponding to the Paris city and its suburbs is the second-largest megacity in Europe, with more than 12 million people. Ammonia emissions in the Paris region reached nearly 11 kt in 2014 of which 93 % is attributed to agricultural activities, with a negligible contribution from livestock (DRIEE, 2017). The Paris megacity is located in the administrative region Île-de-France, which includes 49 % of agricultural land, with particularly strong activity in the eastern half of the territory. Île-de-France agriculture is mainly dedicated to crops, in particular cereals (with a dominance of wheat), which occupy 67 % of the cultivated area. This sector is described as an agricultural belt, with 59 % in the Seine-et-Marne department. Urban traffic in the Paris region is estimated to be responsible for 5 % of ammonia emissions and industry for 2 % (DRIEE, 2017).

Atmospheric ammonia concentrations vary to a large degree during the day, since the atmospheric lifetime of NH₃ is rather short, on the order of hours to a few days (Galloway et al., 2003; Lutsch et al., 2016; Van Damme et al., 2018; Dammers et al., 2019). Wet and dry depositions dominate the atmospheric sink of this inorganic compound. NH₃ is also a gaseous precursor of fine particulate matter (particles of an aerodynamic diameter less than 2.5 µm – PM_2.5). As the main alkaline molecule in the atmosphere, it reacts rapidly with sulfuric (H₂SO₄) and nitric (HNO₃) acids in the atmosphere to form ammonium sulfate or ammonium nitrate (Behera et al., 2013). These ammonium salts may represent more than 50 % of the PM_2.5 fraction during peaks of spring air pollution (Bressi et al., 2013; Petit et al., 2014, 2015), namely periods of fertilizer application by agricultural activities. Concentrations of ammonium nitrate in Île-de-France in the 2009–2010 period can be simulated by the CHIMERE chemistry transport model with only a small bias (< 20 %; Petetin et al., 2016). However, comparisons of the precursors' concentrations measured during the FRANCIPOL (Baudic et al., 2016) campaign (May 2010 to February 2011) show an overestimation of nitric acid and an underestimation of ammonia by the CHIMERE model (Petetin et al., 2016). These differences appear to stem in part from the uncertainties associated with ammonia emissions.

Monitoring the atmospheric concentrations of NH₃ is essential for improving air quality models and for quantifying the fluxes characterizing the nitrogen cycle. The measurement of atmospheric ammonia concentrations is not subject to regulation (in the sense of the European directives) and is really challenging for several reasons: (i) high temporal and spatial dependence of ambient concentration levels (Allen et al. al., 1988; Sutton et al., 1998); (ii) rapid conversion of NH₃ between gaseous, particulate, and aqueous phases (Warneck, 1988); and (iii) measurement artefacts due to NH₃ reactivity with sampling and measurement systems – ${\mathrm{NH}}_{\mathrm{4}}^{+}-{\mathrm{NH}}_{\mathrm{3}}$ conversion, adsorption on surfaces, etc. (Sutton et al., 2008; Von Bobrutzki et al., 2010). This is why there are only a few measuring stations for which atmospheric ammonia concentrations are measured with a time resolution finer than a month. Most of the existing observation networks are designed to evaluate long-term trends and are therefore based on sampling techniques accumulating during several weeks (passive samplers, denuders, or liquid bubblers) and offline analysis in the laboratory.

Other in situ methods enable us to estimate surface atmospheric concentrations of NH₃ (e.g., by cavity ring-down spectroscopy (CRDS) techniques; Sun et al., 2015) or even in the lower atmosphere (airborne campaigns; Leen et al., 2013; Shephard et al., 2015; Sun et al., 2015). Hourly measurements are very rare at present, are mainly available during intensive short-term measurement campaigns, and are limited to a small number of measurement sites of several established networks such as the European Monitoring and Evaluation Programme (EMEP; EMEP/EEA, 2016) used by the French Centre interprofessionnel technique d'études de la pollution atmosphérique (CITEPA; CITEPA, 2015) or the Dutch National Air Quality Monitoring Network (Landelijk Meetnet Luchtkwaliteit – LML) from the Netherlands (Volten et al., 2012).

An innovative and very promising alternative for monitoring atmospheric ammonia is infrared remote sensing. This approach exploits the ammonia absorption spectral signatures of thermal infrared radiation measured by Fourier transform infrared spectroscopy (FTIR) instruments. These methods are free from sampling problems and are noticeably less influenced by local sources than in situ observations. A few observations were first reported from space using an advanced infrared (IR) sounder (Beer et al., 2008) which enables retrievals of atmospheric NH₃. Current or very recent space-based NH₃ data are available from the Infrared Atmospheric Sounding Interferometer sounder (IASI, on board the Metop satellites; Clerbaux et al., 2009; Clarisse et al., 2009), Tropospheric Emission Spectrometer (TES, embedded on the Aura satellite; Shephard et al., 2015; TES's mission ended in January 2018 after nearly a 14-year career of discovery), Cross-track Infrared Sounder (CrIS, embedded in the Suomi National Polar-orbiting Partnership satellite; Shephard and Cady-Pereira, 2015; Dammers et al., 2017), Thermal And Near-infrared Sensor for Carbon Observation–Fourier Transform Spectrometer (TANSO–FTS) on the Greenhouse Gases Observing Satellite (GOSAT; Someya et al., 2020), and Atmospheric InfraRed Sounder (AIRS; Warner et al., 2016; here, a grating spectroscopy instrument installed on the Aqua satellite). These measurements allow for the retrieval of total columns of NH₃ as vertically integrated concentrations between the Earth's surface and the top of the atmosphere. In particular, for the IASI instrument observation processing algorithms have evolved considerably in recent years (Van Damme et al., 2014, 2015; Whitburn et al., 2016) into a data product that we hereafter call NH₃–IASI. These currently available satellite data also feature indicators of the quality of the space-based observations and retrievals and have been validated (Dammers et al., 2016). Even though the observations from space remain complex due to the weak absorption of this species, the availability of a global distribution of NH₃ twice a day (e.g., over Europe) is an important achievement.

In addition, the high-resolution FTIR solar stations of the Network for the Detection of Atmospheric Composition Change (NDACC) measure, from the ground, the total ammonia columns but are also capable of providing information on its vertical distribution (Dammers et al., 2015). Spectroscopic measurements from the ground have been used for decades in the validation of satellite measurements for different trace species (Griesfeller et al., 2006; Strong et al., 2008; Buchholz et al., 2017) and also recently for ammonia (Dammers et al., 2016, 2017). The diurnal variability is observed directly from the ground, with a temporal resolution of a few minutes. The high temporal resolution of these ground-based solar measurements can help us to understand the nature of ammonia sources, especially in urban environments where there is a crucial lack of observations of this short-lived and spatially highly disperse pollutant. A better knowledge of NH₃ emissions and associated atmospheric processes thus requires extended observational networks able to assess high temporal and spatial variabilities of the atmospheric content of ammonia. Such networks still need to be built – especially in France where NH₃ measurements are extremely rare.

In this study, we present the first multiyear time series of atmospheric NH₃ ground-based measurements over a European megacity (Paris) retrieved with the moderate-cost midresolution Observations of the Atmosphere by Solar absorption Infrared Spectroscopy (OASIS) FTIR solar observatory located in the Paris suburbs (Créteil, France). Given its good sensitivity to surface pollutant concentrations, it has been used previously for monitoring urban pollution (tropospheric ozone and carbon monoxide; Viatte et al., 2011; Chelin et al., 2014). A total of 4920 measurements of NH₃ total columns have been obtained between 2009 and 2017, with uncertainties ranging from 20 % to 35 % (1 standard deviation), using the retrieval code PROFFIT (Hase et al., 2004) adapted for our medium spectral resolution and based on the NDACC stations' methodology. The data are compared with NH₃ concentrations from IASI measurements for verifying consistency with the NH₃–OASIS retrievals.

The paper is organized as follows. First, we present in detail the NH₃–OASIS data by describing the site, the retrieval strategy, and uncertainties (Sect. 2). The atmospheric NH₃ time series focusing on seasonal variability is provided in Sect. 3.1, while Sect. 3.2 gives a comparison between the NH₃–OASIS columns and the NH₃–IASI satellite data. Finally, in Sect. 3.3, a few tests are performed using an NH₃ profile-retrieval scheme applied to OASIS spectra with a Tikhonov–Phillips regularization to show the vertical sensitivity to NH₃ provided by OASIS measurements.

Figure 1 Location of the FTIR–OASIS observatory providing NH₃ total columns in the Paris region (France). One can see that the Paris megacity is very urbanized close to the city center and is surrounded by a large rural belt with 49 % agriculture surfaces (Direction Régionale de l'Agriculture et de la Forêt, DRIAAF, http://driaaf.ile-de-france.agriculture.gouv.fr/IMG/pdf/RA_R11_cle4ce9b1.pdf, last access: 11 July 2020).

2 Ground-based FTIR NH₃ data: description and characterization

2.1 The FTIR–OASIS observatory

The “Observations of the Atmosphere by Solar Infrared Spectroscopy” (OASIS; 48.79^∘ N, 2.44^∘ E; 56 m above sea level) observatory is located in the Paris region, which is a European megacity (12 million inhabitants) surrounded by a large rural region and relatively flat terrain (Fig. 1). It has routinely performed solar absorption measurements since 2009 under clear-sky conditions, using a midspectral resolution spectrometer (Bruker Vertex 80, with a spectral resolution of 0.06 cm⁻¹ and maximum optical path difference of 12 cm). In order to carry out air quality research, we have assessed the capability of a medium-resolution FTIR solar absorption spectrometer to monitor pollutants, especially O₃ and CO. We have demonstrated that OASIS is able to continuously monitor tropospheric ozone over Créteil with good accuracy and sufficient information content (Viatte et al., 2011). A 5-year analysis of ozone in the lower troposphere and carbon monoxide has also been made (Chelin et al., 2014). Given the moderate cost and compactness of OASIS, deployment of analogous systems nearby or in large megacities might be useful in support of satellite and air quality studies in other regions of the world.

The observatory comprises an automatized cupola (Sirius 3.5 “School Model” observatory, 3.25 m high and 3.5 m in diameter) in which the upper part (a dome equipped with a mobile aperture) rotates to track the Sun. The alt-azimuthal solar tracker in OASIS is the A547N model manufactured by Bruker Optics, using bare, gold-coated mirrors which are less sensitive to corrosion and pollution than the original Al mirrors. Infrared solar absorption spectra are nominally recorded on a deuterated triglycine sulfate (DTGS) detector using a potassium bromide (KBr) beam splitter, in order to cover the spectral region from 700 to 11 000 cm⁻¹ (0.9 to 14.3 µm) without any optical filter, so that column abundances of many different atmospheric trace gases can be retrieved simultaneously.

Figure 2 Measured spectrum for both spectral windows obtained with the Bruker Vertex 80 at Créteil on 21 March 2012, with individual contributions of the main absorbing species represented from the atlas by Meier et al. (2004) in the first (a) and second (b) spectral windows. Green arrows point out the strong ammonia signatures.

Figure 3 Observed (black line) and calculated spectrum (red line) in radiance unit (nW(cm² sr cm⁻¹)⁻¹) for both spectral windows measured with the Bruker Vertex 80 in Créteil on 21 March 2012, with the fitting residuals (blue line). Green arrows indicate locations of the strong ammonia features in the residual.

2.2 Ground-based NH₃ retrievals

The ammonia spectral signatures used in this study are observed in the 10.6 µm spectral region and belong to the ν₂ vibrational band (Dhib et al., 2007). To achieve a sufficiently high signal-to-noise ratio, each spectrum is produced by coadding 30 scans at the highest spectral resolution, resulting in one interferogram recorded over a period of approximately 10 min. Each coadded interferogram is Fourier transformed to obtain a spectrum without further numerical apodization (i.e., unapodized/boxcar apodization). Carbonyl sulfide (OCS) cell measurements are regularly performed to verify the alignment of the instrument (Chelin et al., 2014). Ammonia absorption lines from the ν₂ vibrational band are also used for satellite-based estimates (Clarisse et al., 2009; Whitburn et al., 2016) and in the first retrievals of ammonia from high-resolution ground-based NDACC–FTIR stations (Dammers et al., 2015). Atmospheric transmission spectra from two spectral microwindows are used in this work (926.3–933.9 and 962.5–970 cm⁻¹). They are slightly larger than those for the Bremen and Lauder NDACC stations (Dammers et al., 2015) because of the coarser spectral resolution of OASIS atmospheric spectra (0.06 cm⁻¹). The choice of using large microwindows aims at retrieving NH₃ abundances from the intensity contrast between the target gas signature and the surrounding continuum. We account for this last one by fitting an empirical background polynomial function with respect to wavelength with 4 degrees of freedom. The main interfering species in these windows are H₂O, CO₂, and O₃ which are simultaneously retrieved together with NH₃. Minor interfering species are HNO₃, SF₆, C₂H₄, and CFC-12. Figure 2 shows an example of a measured spectrum on 21 March 2012, with contributions of all main species, here calculated using the spectral atlas by Meier at al. (2004). The strong spectral signatures of ammonia (pointed out by green arrows in Figs. 2 and 3) are seen in both microwindows even when observed with the medium spectral resolution of the Vertex 80 spectrometer. Note that the NH₃ concentrations of this day were particularly higher than on average, resulting in very strong NH₃ features.

The retrievals are performed using the PROFFIT 9.6 code (Hase et al., 2004), widely used by the NDACC community to retrieve trace gases from high-resolution FTIR measurements, but adapted for the medium resolution. The software is capable of including spectral channeling in the fitting process and for the estimation of error budgets. Channeling is caused by the presence of optical resonators such as filters or windows in the measurement beam. Here, the ability of the code to handle spectral channeling in the fit is of relevance because we intend to quantify a minor absorber, and some channeling was detected in the measured spectra. For this purpose, the channeling frequency is determined from a Fourier transform of the residual of an auxiliary fit. A neighboring spectral section containing few absorption lines and being wide enough to encompass many cycles of the channeling signal is selected for this purpose. For the chosen frequency, a sine and a cosine amplitude are subsequently included in the fit of the NH₃ target windows. Daily temperature and pressure profiles for the meteorological variables are obtained from the Goddard Space Flight Center National Centers for Environmental Prediction (NCEP). For radiative transfer calculations, profiles at 44 altitude levels, from 50 m up to 70 km, are set. Spectroscopic data are taken from the HITRAN 2008 database (which does not include ad hoc changes on line parameters added in HITRAN 2012; Rothman et al., 2009), except for CO₂ lines from the HITRAN 2012 database (Rothman et al., 2013). As for satellite retrievals, and in consistency with the moderate spectral resolution of OASIS, we derive total columns of NH₃ from each radiance spectrum. A scaling factor of a climatological vertical profile of NH₃ is adjusted in order to minimize the difference between measured and simulated spectra, so the degree of freedom for the ammonia retrievals is 1. As the radiance values are rather small below 1000 cm⁻¹, a quality criterion was introduced to select only spectra with a signal-to-noise ratio higher than 30, as measured between 960 and 990 cm⁻¹, covering lines of a weak CO₂ band. Figure 3 shows a measured spectrum (black line), the corresponding simulated spectrum (red line), and the difference between observation and simulation (blue line) in both microwindows. The fits are excellent, with a standard deviation of 2 % in both microwindows.

Table 1 Random and systematic errors according to NH₃ total columns. The total errors, combining the systematic and random errors, vary from 20 % up to 35 %. Note that 9.1×10¹⁶ molec. cm² is the maximum ammonia column measured over Paris during the 2009–2017 period (which occurred exactly in March 2012).

2.3 A priori profiles and uncertainty

A climatological a priori profile of NH₃ that assumes vertically homogenous NH₃ concentrations (0.1 ppb) in the troposphere is taken from the Michelson Interferometer for Passive Atmospheric Sounding (MIPAS) project (Remedios et al., 2007). The a priori profiles of the interfering species (H₂O, CO₂, and O₃) are taken from the Whole Atmosphere Community Climate Model (WACCM version 6; Chang et al., 2008). We consider the a posteriori errors calculated by PROFFIT 9.6. This calculation is based on the error estimation method by Rodgers (2000). For the uncertainty in the NH₃ line parameters, we assume values as stated in the HITRAN 2008 database (Rothman et al., 2009). We assume a conservative value of 20 % for the integrated intensities and of 10 % for the pressure-broadening coefficients. The total errors are dominated by the combination of uncertainties in the spectroscopic parameters (including also the interfering species), the noise in the spectra, and the hypotheses on the retrieval (an a priori profile of NH₃; forward model uncertainties). They are comparable to those estimated by Dammers et al. (2015) for a high-resolution ground-based station at Bremen (Germany). Complementary to these, according to a review paper on NH₃ spectroscopic parameters (Down et al., 2013), the uncertainty of 20 % on line intensities is probably a worst-case estimate. Table 1 summarizes the results of total errors according to the different ammonia total columns, which vary from 20 % to 35 %. Another a priori was tested with higher concentrations of NH₃ in the atmospheric boundary layer (with fixed concentrations up to 900 m, then decreasing NH₃ concentrations until 4 km). Using this a priori profile reduces the mean squared difference between measured and simulated spectra by about 60 %. However, both retrievals with homogenous and sloped a priori profiles show rather similar results, with the same relative evolution in time and differences in absolute terms in the order of magnitude of the total column retrieval error (the use of the sloped a priori profile increases the retrieved NH₃ abundances by 20 % with respect to those using a homogenous a priori).

Table 2 Statistics of the retrieved NH₃–OASIS measurements depending on the season (MAM – March, April, and May; JJA – June, July, and August; SON – September, October, and November; DJF – December, January, and February) and for the whole calendar year. (No. refers to number of spectra, mean, median, and standard deviation.) Total columns are given in 1×10¹⁶ molec. cm⁻².

3 Results

3.1 Seasonal variability of NH₃ ground-based data

Figure 4 shows the multiyear time series of NH₃ total columns derived from FTIR ground-based OASIS measurements retrieved from all of the 4920 available spectra in 2009–2017 resulting from 234 measurement days. Table 2 gives a summary of statistics of the retrieved NH₃ measurements versus the four seasons. Individual measurements with an overall mean total column of 0.84×10¹⁶ molec. NH₃ cm⁻² and a standard deviation of 0.86×10¹⁶ molec. NH₃ cm⁻² indicate a large variability in the observations. They highlight peak abundances in spring (March, April, and May), more precisely in March. The amplitude of the spring peaks varies throughout the years, with maxima in March 2012, reaching about 9×10¹⁶ molec. NH₃ cm⁻². The occurrence of the highest NH₃ concentrations in March is particularly noticeable as all measurements are above 2×10¹⁶ molec. NH₃ cm⁻², which corresponds to the mean of data plus 1 standard deviation over the springtime period (March, April, and May). They are measured for this calendar month for almost every year (2011 – 2 d, 2012 – 11 d, 2014 – 6 d, 2015 – 4 d, 2016 – 4 d, and 2017 – 1 d). March 2012 is therefore a month that is particularly polluted in terms of atmospheric NH₃. In addition to many days with more than 2×10¹⁶ molec. NH₃ cm⁻², a peak reaching the quadruple of that threshold on 21 March 2012 represents the maximum retrieved ammonia total column.

Figure 4 First multiyear time series of NH₃ total columns derived from OASIS measurements over Paris and retrieved from 4920 infrared atmospheric transmission spectra measured during 234 d between 2009 and 2017.

The seasonal variability is analyzed in Fig. 5 in terms of monthly averages over the 2009–2017 period. The mean NH₃ column in March is 1.65×10¹⁶ molec. NH₃ cm⁻², which is 2 times higher than the overall mean total column over the 4920 measurements in 9 years.

As for many other regions, NH₃ seasonality is well marked with high values that might be connected with the timing of agricultural manure spreading. As shown by both approaches of Ramanantenasoa et al. (2018) for 2005–2006, mineral fertilizers are mainly used in the Île-de-France region because they are major arable crop (especially cereals) farming areas. They account for 59 % of NH₃ emissions according to the CADASTRE_NH₃ framework based on the process-based Volt'Air model (Garcia et al., 2011, 2012). These fertilizers are mostly spread during springtime. In Bremen, NH₃ atmospheric total columns have a similar seasonal cycle with highest levels during spring. The maximum values occur around April, which is consistent with temporal emission patterns for manure application reported for this region (Dammers et al., 2015). Note that, for example, the highest levels in Bremen were observed during springtime with total columns reaching up to 9.3×10¹⁶ molec. NH₃ cm⁻², which is close to the maximum peak observed by OASIS in March 2012.

Figure 5 Average annual cycle of monthly NH₃ total columns (molec. cm⁻²), observed by OASIS over Paris, averaged over the 2009–2017 period. The red crosses represent the number of OASIS data per calendar month on a log scale (right axis).

One can observe also in Fig. 4 significant concentrations in June (2017), July (2010, 2012, 2013, 2014, 2015, and 2017) and in August 2016, with another threshold of 1.5×10¹⁶ molec. NH₃ cm⁻², which corresponds roughly to the mean of data plus 1 standard deviation over the summer time period (June, July, and August). This shows a second seasonal peak for the summer period. On the contrary, during winter months (December, January, and February) the NH₃ total columns have a pronounced minimum (mean total column of 0.12×10¹⁶ molec. NH₃ cm⁻² and a standard deviation of 0.12×10¹⁶ molec. NH₃ cm⁻²). It is worth noting that fewer observations are available during this season due to frequent overcast conditions.

This evolution during early winter and, to a lesser extent, summer are also observed in the Bremen measurements, emphasizing some similarities between both the Paris and Bremen station environments. Such seasonal behavior is found not only in Europe but also in other megacities such as Seoul (South Korea), where surface ammonia concentrations exhibit higher values during spring and in the warm season (summer), while dropping to a minimum in the cold season (winter; Phan et al., 2013). One possible explanation is the increased volatility of ammonia in agricultural soils and ammonium particles (in the atmosphere in aqueous or solid phase) under warm meteorological conditions.

3.2 Comparison between NH₃–OASIS and NH₃–IASI

The validation of the satellite products of IASI (Van Damme et al., 2014) is limited by the scarcity of a long-term series of atmospheric ammonia measurements. A first attempt to validate NH₃–IASI measurements was made with correlative data from surface in situ and airplane-based measurements (Van Damme et al., 2015). They confirmed consistency between the NH₃–IASI data set and the available in situ observations and showed promising results for validation by using independent airborne data from the California Research at the Nexus of Air Quality and Climate Change (CalNex) campaign. Nevertheless, that study was limited by the availability of independent measurements and suffered from representativeness issues when comparing satellite observations to surface concentration measurements. Recently, Dammers et al. (2016) reported a first step in the validation of NH₃–IASI products by comparing ammonia columns with high-resolution FTIR measurements from several NDACC stations around the world. They concluded that IASI reflects similar pollution levels and seasonal cycles as shown by FTIR observations, and the best correlation (R=0.83 and a slope of 0.60) was obtained with the NDACC Bremen station. Bremen is located in the northwest of Germany, which is characterized by intensive agriculture. It is most suitable for comparisons with IASI given the very high atmospheric concentrations of NH₃ observed there. Compared to this work over Bremen, our analysis is the first comparison of ground-based NH₃ measurements performed in a megacity (Paris) with NH₃–IASI and covers 9 years of data. In this study, we consider NH₃ total columns retrieved from the IASI-A instrument (aboard the Metop-A platform) observations along with those retrieved from the IASI-B instrument (aboard the Metop-B platform) observations, with IASI-B data being available only for the period from 8 March 2013. Average and median errors of these satellite measurements used in this study are 89 % and 60 %, respectively, which are coherent with uncertainties for most of the NH₃–IASI data listed in Van Damme et al. (2014, 2017) because of small absorption features by ammonia observed with the relatively coarse spectral resolution of IASI as compared to ground-based instruments. The sensitivity of IASI retrievals to surface concentrations of NH₃ is also limited, and the time of IASI overpasses (early morning and early evening) is not coincident with the typical early afternoon peaks of NH₃ concentrations.

Figure 6 Correlations between the FTIR–OASIS and NH₃–IASI total columns fulfilling the following temporal and spatial coincidence criteria: a time difference smaller than 30 min and a 15 km radius centered on the ground-based FTIR station. The broken line is the straight equation with slope 1.

For the comparison, only morning overpass (AM) satellite observations (i.e., 09:30 local time) are considered here as they are generally more sensitive to NH₃ owing to more favorable thermal contrast in the daytime (Van Damme et al., 2014) and better temporal coincidence (OASIS observations are only carried out during daytime). The satellite data have a circular footprint of 12 km diameter at the nadir and an ellipsoid-shaped footprint of up to 20 km×39 km at the outermost angles (Clarisse et al., 2009). We use raw observations from 1 January 2009 to 31 December 2016 from the most recent version of NH₃–IASI retrievals (the near-real-time neural network retrieval version 2 with reanalyzed meteorological inputs called ANNI-NH3-v2.2R; Van Damme et al., 2017).

Note that the lifetime of atmospheric NH₃ is rather short, on the order of hours not being uncommon (Galloway et al., 2003; Lutsch et al., 2016; Van Damme et al., 2018), up to a few days, due to efficient deposition and fast conversion to particulate matter. Thus, NH₃ concentrations vary strongly as a function of emission strengths and meteorological conditions (such as temperature, precipitation, wind, and vertical mixing in the atmospheric boundary layer).

Figure 7 Description of the 15 km rings centered on OASIS observatory that were used to select different IASI pixels for the correlation with FTIR data.

Figure 8 Correlation coefficients R (blue lines, left axis) between IASI and ground-based FTIR observations with temporal sampling difference smaller than 30 min, as a function of d_min to OASIS observatory, considered for the calculation of the 15 km wide rings. The data (orange lines, right axis) show the total number of IASI–OASIS coincident observations.

Figure 9 Regression slope (blue lines, left axis) between IASI and ground-based FTIR observations with temporal sampling difference smaller than 30 min, as a function of d_min to OASIS observatory, considered for the calculation of the 15 km wide rings. The data (orange lines, right axis) show the total number of IASI–OASIS coincident observations.

In order to minimize differences associated with the temporal variability of NH₃, we only consider NH₃ measurements from OASIS performed within ±30 min with respect to the IASI morning overpass. The spatial coincidence criterion is 15 km between the center of IASI pixels and OASIS. Tests with wider coincidence criteria (50 km and ±90 min) do not show significant differences (similar correlations are obtained despite a greater number of coincidences). We also exclude NH₃–IASI data with relative errors higher than 100 %, except for those with absolute errors lower than 0.5×10¹⁶ molec. cm⁻² and retrievals with no physical sense (negative values). The majority of IASI retrievals corresponding to weak NH₃ abundances are screened out by the relative error criterion; thus the exclusion of negative values has a negligible effect in the comparisons.

Figure 6 shows a scatterplot comparison between NH₃–OASIS and NH₃–IASI data. We obtain a very good correlation of R=0.79 and a slope of 0.73 for a total of 52 coincidences. These results are similar to those from a comparison between a ground-based FTIR at Bremen and NH₃–IASI lookup table version R=0.83 and a slope of 0.60 for a total of 53 coincidences (Dammers et al., 2016), and NH₃–IASI neural network version R=0.67 and a slope of 0.96 for 802 coincidences from several ground-based FTIR stations (Dammers et al., 2017). Here, for Paris, we used stricter spatiotemporal collocation criteria for the comparison as we have more data than the Bremen study. For this correlation, we calculated the absolute differences (AD) between satellite (y axis) and FTIR–OASIS NH₃ total columns (x axis), which are defined here as follows:

$$\begin{array}{}\text{(1)}& \mathrm{AD}=({\mathrm{NH}}_{\mathrm{3}}-\mathrm{IASI})\mathrm{column}-\left({\mathrm{NH}}_{\mathrm{3}}-\mathrm{OASIS}\right)\mathrm{column}.\end{array}$$

The average of the absolute differences is $-\mathrm{0.08}\times {\mathrm{10}}^{\mathrm{16}}$ molec. cm⁻², with a root mean square error (RMSE) equal to 0.49×10¹⁶ molec. cm⁻² and a standard deviation of error (STDE) equal to 0.48×10¹⁶ molec. cm⁻². The mean absolute difference is a factor of 10 smaller than the average values of the NH₃ abundances (0.78×10¹⁶ and 0.70×10¹⁶ molec. cm⁻², respectively, for OASIS and IASI) and also smaller than OASIS total retrieval errors (20 % to 35 %). This reveals a very good consistency between NH₃–IASI and NH₃–OASIS, with this last one being analyzed during 9 years (2009–2017) in Sect. 3.1.

We further investigate the representativeness of the OASIS site for the Île-de-France region by comparing NH₃–OASIS data with that from IASI at different distances from the ground-based site. We use 15 km wide rings (Fig. 7) centered on the OASIS observatory and increase the minimal distance (d_min) of the rings from OASIS observatory with a 1 km step. The 15 km width of the rings is chosen to minimize the impact of ammonia spatial variability and to consider a sufficient number of coincidences for statistics. The minimal distance (d_min) varies from 0 up to 400 km. Figures 8 and 9 show, respectively, the correlation and regression slope as a function of d_min using a maximum allowed sampling time difference of 30 min. The numbers on the right axis of each of the figures show the number of coincident observations used in the comparison. An increasing d_min shows a decreasing correlation (blue lines) and a changing slope (increasing with distance up to 120 km, then decreasing). One can distinguish three different regimes in the plot of Fig. 8. For d_min between 0 and 12 km, the highest correlations are seen with R varying from 0.84 down to 0.57, then between 12 and 120 km R is around 0.6 despite lots of noise, and then we get decreasing correlations for d_min between 120 and 400 km. This comparison underlines that measurements from OASIS observatory provide information about atmospheric NH₃ variability on a regional scale, up to 120 km away from the site, but might also be affected by more local processes and/or emissions.

For d_min between 0 and 12 km regression slopes vary between 0.73 and 0.81, revealing an underestimation of observed columns by IASI compared to OASIS (Fig. 9) as already mentioned by Dammers et al. (2016) when evaluating IASI with some high-resolution NDACC stations. This underestimation may be explained by IASI's lower sensitivity to surface ammonia concentrations due to the coarse spectral resolution and weak thermal contrast between the surface and the lower troposphere and to the spatial heterogeneity of ammonia within the IASI footprint. The regression slope increases until 120 km for d_min and decreases beyond. This might be linked to the fact that the main surrounding agricultural regions (e.g., Picardie and Champagne) are located up to about 120 km away from OASIS, and therefore NH₃ sources (Clarisse et al., 2019; Van Damme et al., 2018), and these sources are more heterogenous beyond this distance.

3.3 Vertical distribution of sensitivity of the NH₃–OASIS approach

As mentioned in Sect. 2.2, the NH₃–OASIS data set presented in Figs. 4, 8, and 9 is derived from a scaling factor retrieval scheme whose state vector has only one scalar value associated with the NH₃ abundance. Therefore, this approach does not provide an averaging kernel matrix as optimal estimation or Tikhonov schemes do but only a single value of degrees of freedom (DOF) without any information on the vertical distribution of the retrieval sensitivity. In order to estimate the vertical sensitivity to NH₃ provided by OASIS measurements, we have performed a few tests using an NH₃ profile-retrieval scheme applied to OASIS spectra with a Tikhonov–Phillips regularization (as similarly implemented for ozone profiles by Viatte et al., 2011). Figure 10 presents examples of averaging kernel diagonals for NH₃ profile retrievals based on OASIS spectra measured on 13 March 2014 at different times of the day and thus different solar zenith angles (SZAs). We remark that OASIS measurements may provide information on the abundance of NH₃ located around 500 m, with maximum sensitivity for smaller solar zenith angles corresponding to thicker air masses (occurring in the early morning or late afternoon). These OASIS averaging kernel diagonals peak at similar altitudes as those estimated by Dammers et al. (2017) for a high spectral resolution Fourier thermal infrared spectrometer at the Pasadena site (peaking around 940 hPa, thus approximately at 600 m above sea level). These altitudes are typically located within the atmospheric boundary layer during springtime and summer at midlatitudes where most of the atmospheric NH₃ column variability is expected to occur. Additional tests (not shown) using different spectroscopic databases (HITRAN 2008 and HITRAN 2012) change the estimation of the sensitivity of the OASIS retrieval very little.

Figure 10 Typical NH₃–OASIS averaging kernel diagonal elements illustrated on the 13 March 2014 at different times of the day, and solar zenith angles (SZAs in degrees) using spectroscopic data from the HITRAN 2008 (HIT08) database (except for CO₂ lines).

4 Conclusions and perspectives

Ground-based infrared remote sensing is undoubtedly a promising and powerful spectroscopic technique to retrieve ammonia columns, even when using FTIR instruments with a moderate spectral resolution such as OASIS. Recording one spectrum over a period of approximately 10 min, the OASIS instrument can provide over 40 spectra per day over the Paris region and allows for the observations of the diurnal cycle of ammonia. In this paper, we presented the first multiyear time series (2009–2017) of atmospheric NH₃ total column measurements from ground-based infrared remote sensing over the Paris megacity. NH₃ total columns vary seasonally by 2 orders of magnitude, approximately from 0.1×10¹⁶ molec. cm⁻² in winter to 10×10¹⁶ molec. cm⁻² for spring peaks. Error estimations show random errors of about 10 % and systematic errors less than 25 % for individual observations, mainly due to the combination of uncertainties in the spectroscopic parameters (including also the interfering species).

This study is based on the 9-year NH₃–OASIS time series in order to analyze the seasonal variability of atmospheric NH₃ in the Paris region. The predominance of NH₃ peaks occurring in March is particularly noticeable: all measurements are greater than 2×10¹⁶ molec. of NH₃ cm⁻², which corresponds to the average values of the data set plus 1 standard deviation over the springtime period (March, April, and May) and are coincident with manure-spreading time periods (Ramanantenasoa et al., 2018). Mineral fertilizers are mainly applied in the Île-de-France region because they are major arable crop (especially cereals) farming areas, which may volatilize and can generate high ammonia concentrations under sunny conditions. This study also finds high summer values for NH₃–OASIS total columns greater than 1.5×10¹⁶ molec. cm⁻², which corresponds roughly to the average value of the data set plus 1 standard deviation over the June, July, and August time period. This is likely linked to an increase in the volatility of ammonia under warm meteorological conditions.

Satellite remote sensing instruments such as the Infrared Atmospheric Sounding Interferometer (IASI) on board the Metop platforms provide global distributions of atmospheric NH₃ relying on valuable information (Van Damme et al., 2018). We have compared NH₃–IASI data with our measurements from OASIS over the 2009–2016 time period. We show a very good correlation of 0.79 and a mean bias of $-\mathrm{0.08}\times {\mathrm{10}}^{\mathrm{16}}$ molec. cm⁻² between the two data sets, which represents 10 % of the average of the values in the range of 0–1.5×10¹⁶ molec. cm⁻². Besides the space instrument's lower sensitivity to the surface ammonia concentrations, spatial heterogeneity of ammonia within the IASI footprint can lead to lower values. Furthermore, OASIS observatory measurements of NH₃ are representative of the variability at a 100 km regional scale.

This study also illustrates that ground-based stations like OASIS may play a key role in validating current (e.g., IASI) and future satellite observations (e.g., Infrared Atmospheric Sounder Interferometer coupled with the New Generation, IASI–NG, and the Meteosat Third Generation coupled with the InfraRed Sounder, MTG–IRS) of ammonia in the infrared for a better understanding of the space and time variability of this major source of nitrogen species in the troposphere.

Since ammonia is a major precursor of PM_2.5 over Europe, as shown by, e.g., Fortems-Cheiney et al. (2016) during a European spring haze episode, we expect a link between high ammonia concentrations and inorganic salts such as ammonium nitrate. That period during late winter 2012 (documented by Petit et al., 2014) was probably the most polluted month of March of the last 10 years in the Paris region (Petit et al., 2017), with the highest NH₃–OASIS total columns in the period 2009–2017 over the Paris region. The link between ammonia concentrations and the formation and volatilization of fine particles such as ammonium salts is beyond the scope of this paper and will be discussed in a future study on the diurnal analysis of total and surface ammonia measurements from the Paris region during a high spring pollution event. Moreover, future versions of the NH₃–OASIS approach will be based on vertical profile retrieval using a Tikhonov–Phillips-type regularization. Tests of different a priori NH₃ profiles will also be performed for reducing the spectral residuals between measured and simulated spectra, while providing accurate retrievals of NH₃ abundances. First tests with a different a priori profile for NH₃ show a significant reduction of the residuals between the radiance spectra measured by OASIS and those simulated by PROFFIT.