UK Ammonia Emissions Estimated With Satellite Observations and GEOS-Chem

Agricultural emissions of ammonia (NH 3 ) impact air quality, human health, and the vitality of aquatic and terrestrial ecosystems. In the UK, there are few direct policies regulating anthropogenic NH 3 emissions and development of sustainable mitigation measures necessitates reliable emissions estimates. Here, we use observations of column densities of NH 3 from two space-based sensors (IASI and CrIS) with the GEOS-Chem model to derive top-down NH 3 emissions for the UK at fine spatial ( ∼ and time (monthly) scales. We focus on March-September when there is adequate spectral signal to reliably retrieve NH 3 . We estimate total emissions of 272 Gg from IASI and 389 Gg from CrIS. Bottom-up emissions are 27% less than IASI and 49% less than CrIS. There are also differences in seasonality. Top-down and bottom-up emissions agree on a spring April peak due to fertilizer and manure application, but there is also a comparable summer July peak in the top-down emissions that is not in the bottom-up emissions and appears to be associated with dairy cattle farming. We estimate relative errors in the top-down emissions of 11%–36% for IASI and 9%–27% for CrIS, dominated by column density retrieval errors. The bottom-up versus top-down emissions discrepancies estimated in this work impact model predictions of the environmental damage caused by NH 3 emissions and warrant further investigation. large reductions in other precursors of such pollutants detrimental to our health. Here, we use a model and observations of ammonia from two space-based sensors to estimate emissions in the UK where there are no direct policies regulating agricultural sources of ammonia. The satellite-derived emissions, limited to March-September when conditions are ideal for viewing ammonia from space, total 272 kilotonnes from an instrument that passes overhead in the morning and 389 kilotonnes from an instrument with a midday overpass. Though the emissions estimates differ for the two instruments, both exhibit a spring (April) peak due to fertilizer and manure use and summer (July) peak likely associated with dairy cattle farming. The summer peak is missing in bottom-up emissions and total March-September emissions from these inventories are also 27%–49% less than those derived with satellites. Further research is needed to address these discrepancies, as such inventories are widely used for developing policies and assessing environmental damage caused by ammonia. emissions estimated with the UK National Atmospheric Emission Inventory (NAEI) annual emissions and GE-OS-Chem monthly scaling factors are 27% less than IASI-derived emissions and 49% less than CrIS-derived emissions. This is supported by a 38%–42% underestimate in surface NH 3 concentrations from GEOS-Chem driven with the NAEI. We infer UK top-down annual anthropogenic NH 3 emissions of 383–431 Gg from IASI and 559–642 Gg from CrIS compared to 276 Gg from the NAEI. Seasonality in the top-down emissions confirms the well-known spring April peak from fertilizer and manure use, but there is also a summer July peak coincident with intensive dairy farming that is absent in the bottom-up emissions.

Estimates of the contribution of NH 3 emissions to PM 2.5 and mobilization of nitrogen in aquatic and terrestrial ecosystems, assessment of attainment of emissions ceilings commitments and targets, and decisions on effective mitigation measures demand accurate estimates of NH 3 emissions. The NAEI of annual total and mapped UK NH 3 emissions is published each year. These are obtained at high spatial resolution (1 km) with a model that uses climatological environmental factors and incorporates detailed information about farming activities that contribute to NH 3 emissions. The ability to validate the inventory is challenging, as there are no long-term measurements of NH 3 fluxes. There is a network of very reliable measurements of rural 24-h mean surface concentrations of NH 3 that cover the full latitudinal extent of the UK from Cornwall in the south to Shetland in the north (Tang et al., 2018), but there are large monitoring gaps in-between. Individual sites are also unlikely to be representative of inventory grid cells for an emission source with large spatial variability. Satellite observations of NH 3 retrieved from infrared spectral measurements offer complete coverage of the UK and routine daily measurements in the absence of clouds and under good retrieval conditions. Satellites observe NH 3 molecules throughout the atmospheric column, but the majority are within the planetary boundary layer and most of the variability in the column is typically due to NH 3 at or near the surface (Clarisse et al., 2010;Nowak et al., 2010;Schiferl et al., 2016;.
Retrieval of NH 3 from space-based instruments was first described by Beer et al. (2008) for the Tropospheric Emission Spectrometer (TES) instrument. Satellite NH 3 retrieval products have since undergone substantial retrieval development (Clarisse et al., 2009;Shephard et al., 2011Shephard et al., , 2020, intercomparisons , and validation against ground-based observations of total atmospheric column densities and surface concentrations of NH 3 (Dammers et al., 2016(Dammers et al., , 2017. These products have also seen extensive use in characterizing NH 3 emissions. This includes detecting global and regional NH 3 emission hotspots Clarisse et al., 2019;Dammers et al., 2019;Shephard et al., 2020;Van Damme et al., 2018), constraining NH 3 emissions from biomass burning (Adams et al., 2019;), evaluating regional emission inventories (Chen et al., 2021;Fortems-Cheiney et al., 2020), identifying underestimated or missing NH 3 sources in widely used global and regional emission inventories and models (Heald et al., 2012;Hickman et al., 2018;Van Damme, Kruit, et al., 2014), and determining long-term local and regional trends and variability in NH 3 (Hickman et al., 2020;Van Damme, Eriman, et al., 2015;. Here, we use satellite observations of NH 3 and the GEOS-Chem chemical transport model (CTM) to derive top-down NH 3 emissions for the UK and evaluate the NAEI inventory and current understanding of seasonality in emissions as represented in GEOS-Chem. This includes the use of surface observations from the UK monitoring network to evaluate the model driven with the NAEI to corroborate findings from the satellite observations.

Space-Based Observations of Column Densities of NH 3
Satellite observations of NH 3 retrieved in the infrared portion of the light spectrum rely on the spectral signal that depends on the atmospheric state, such as abundance and vertical distribution of NH 3 and thermal contrast between the surface of the Earth and the overlying atmosphere (Clarisse et al., 2010;Shephard et al., 2011). Two prominent products are available from contemporary space-based instruments that pass overhead in the morning (the Infrared Atmospheric Sounding Interferometer or IASI) and midday (the Cross-track Infrared Sounder or CrIS). These products use distinct retrieval approaches, offering two independent data sets to assess the potential to use satellite observations to constrain the magnitude and seasonality of UK NH 3 emissions.

Infrared Atmospheric Sounding Interferometer NH 3
The IASI instrument onboard the Metop-A satellite was launched into low-Earth polar sun synchronous orbit in October 2006. The instrument has two overpass times in the morning (09h30 local solar time or LST) and at night (21h30 LST), providing global coverage twice a day. The elliptical IASI pixels range in ground pixel resolution from 12 × 12 km at nadir (directly below the instrument) to about 20 × 39 km at the edges of the 2,200-km-wide swath (Clarisse et al., 2011). The data product we use is the Level 2 cloud-free reanalysis product of total column NH 3 (version 3R-ERA5) . The retrieval uses machine learning, specifically a neural network trained relationship between column NH 3 and a so-called hyperspectral range index or HRI, where the HRI is a measure of the relative enhancement in the spectral signature due to NH 3 (Van Damme, Clarisse, et al., , 2017. The data product includes reported retrieval errors estimated by perturbing individual input parameters in the neural network framework . Products resulting from the neural network retrieval approach have been validated against global and regional networks of ground-based NH 3 observations of surface concentrations and column densities (Dammers et al., 2016;Guo et al., 2021;. In general, IASI NH 3 reproduces the temporal variability in surface concentrations of NH 3 , but exhibits a low bias (Dammers et al., 2017;. We use daytime (09h30 LST) IASI NH 3 for 2008-2018 to obtain multiyear monthly means. This dampens influence of interannual variability and ensures consistency with NAEI NH 3 emissions that are estimated with 30-years mean meteorology (Ricardo, 2019a). We grid the data to finer spatial resolution (0.1° × 0.1°; ∼10 km) than the native resolution of the instrument using the tessellation oversampling technique described in Zhu et al. (2017) and Sun et al. (2018). This takes advantage of the spatial variability in coverage of individual orbits and the long data record from IASI to reduce noise and smooth out spatial gradients in the gridded product (Sun et al., 2018). Briefly, tessellation involves weighting individual IASI pixels by the area of overlap with the target grid and also includes error weighting using the reported retrieval error. In our application of the tessellation gridding technique, we approximate the area of IASI pixels as a quadrilateral polygon, where the corners of each polygon are estimated as the distance midway between the centers of neighboring IASI pixels. Retrieval of NH 3 over the UK is challenging, due to persistent clouds and relatively cool conditions. Extreme retrievals, identified as absolute columns >5 × 10 17 molecules cm −2 , are removed. We also exclude IASI NH 3 columns retrieved on July 26-27, 2018, coincident with the summer 2018 heat wave (McCarthy et al., 2019). Record high temperatures (>30°C) lead to UK IASI NH 3 column densities 4-times greater (∼4 × 10 16 molecules cm −2 ) than the UK July multiyear mean (∼1 × 10 16 molecules cm −2 ). Including these days increases the July multiyear mean by 11% and reduces its representativeness as a climatological mean for comparison to the NAEI. A similarly large influence of heat waves on IASI NH 3 columns was reported for the summer 2010 heat wave over mainland Europe . After using oversampling to grid the data to 0.1° × 0.1°, gridded multiyear means with large relative error (>50%) are removed. This leads to loss of the majority of IASI NH 3 columns in October-February, so only March-September multiyear means are considered. Additional filtering is applied to the gridded multiyear monthly means to remove extreme values identified as columns <−1 × 10 16 and >1 × 10 17 molecules cm −2 . These only account for <0.1% of the March-September data, but affect spatial consistency between IASI and CrIS. Figure 1 shows the gridded March-September multiyear monthly mean IASI NH 3 columns. The number of observations in each grid ranges from 11 to 128. Values over Scotland are very low (typically <2 × 10 15 molecules cm −2 ) due to weak signal, lower agricultural activity than the rest of the UK, and greater distance from sources in mainland Europe. The range in IASI NH 3 over the rest of the country of 4-8 × 10 15 molecules cm −2 is much less than the NH 3 hotspots in other parts of the world. Columns over global hotspots such as North China, West Africa, the Po Valley (Italy), and the Indo-Gangetic Plain (India) exceed 2 × 10 16 molecules cm −2 Dammers et al., 2019;Van Damme, Clarisse, et al., 2014b. These are associated with industrial and agricultural activity in India and China, and intense seasonal open burning of biomass and relatively low abundance of acidic aerosols in West Africa and northern India. Warm temperatures in these regions also increase NH 3 emissions, suppress partitioning of NH 3 to aerosols, and enhance the spectral signal.

Cross-Track Infrared Sounder NH 3
The first CrIS sensor launched into low-Earth polar sun synchronous orbit in October 2011 is onboard the NOAA Suomi-NPP satellite. Like IASI, CrIS observes the Earth twice daily, though in the early afternoon (13h30 LST) and after midnight (01h30 LST) (Goldberg et al., 2013). It has the same swath width as IASI and similar ground pixel resolution (14 km circular pixels at nadir). The fast physical retrieval (CFPR) approach used to retrieve NH 3 columns is described in detail in Shephard and Cady-Pereira (2015) and Shephard et al. (2020). Briefly, it is based on conventional optimal estimation that involves minimizing the difference between observed and calculated outgoing spectral radiances with a priori vertical profiles of NH 3 (Rodgers, 2000). CFPR uses three prior NH 3 profiles representing polluted, moderately polluted, and remote conditions (Shephard et al., 2020) that are selected based on the ammonia spectral signal. This is different to standard optimal estimation that uses prior information that is independent of the observations and imposes spatial and temporal information. The CFPR retrieval generates averaging kernels that quantify the vertical sensitivity of the retrieval. These typically peak between 900 and 750 hPa (∼1-2.5 km altitude) (Dammers et al., 2017;. We use the Level 2 CrIS NH 3 CFPR version 1.6 product for 2013-2018. The predecessor product (version 1.5) exhibited a positive bias for NH 3 < 1 × 10 16 molecules cm −2 , as values were only retrieved over scenes exceeding the instrument detection limit of ∼2 × 10 15 molecules cm −2 (Dammers et al., 2017; Shephard & Cady-Pereira, 2015). This approach filtered out cloud-free scenes below the instrument detection limit and indirectly removed cloudy scenes when the NH 3 signal below clouds could not be detected. In version 1.6, clouds are explicitly identified with information from the space-based Visible Infrared Imaging Radiometer Suite (VIIRS) (White et al., 2021). We use daytime cloud-free CrIS observations with quality flag ≥4 (Shephard et al., 2020) and thermal contrast >0 K, where thermal contrast is the difference between the reported temperatures at the surface and the lowest atmospheric layer. We identify and correct for a positive trend in the CrIS baseline that appears to be erroneous, as it is not apparent in the IASI data. We do this by estimating a statistically significant (P-value = 0.03) increase in monthly mean background NH 3 columns over Scotland ( Figure S1) of 2.21 × 10 13 molecules cm −2 per month (amounting to 1.6 × 10 15 molecules cm −2 over the whole record) and subtract this from individual CrIS NH 3 column retrievals. We grid the corrected data to 0.1° × 0.1° using the same tessellation code used for IASI, but without error weighting. The individual total column errors include measurement and representative errors and cover a much narrower range (5%-55%; Shephard et al., 2020) than those for IASI (5% to >100%). As a result, higher relative weighting would be applied to low column densities, leading to anomalously low gridded values in the CrIS multiyear means. For consistency with IASI, and because of weak spectral signal in autumn and winter, we only consider CrIS retrievals in March-September. Figure 2 shows the gridded March-September CrIS NH 3 multiyear monthly mean columns. As with IASI, we filter for extreme values in the multiyear means (column densities <−1 × 10 16 molecules cm −2 and >1 × 10 17 molecules cm −2 ), removing <0.1% of the gridded data. Observations during the July 2018 heatwave only increase the July multiyear mean by 1.6%, but for consistency with IASI these days are also removed. The number of CrIS retrievals in each grid ranges from 11 to 96. The CrIS multiyear means are roughly double those for IASI (Figures 1 and S2), in part because CrIS passes overhead at midday when higher ambient temperatures lead to greater volatilization of NH 3 . Differences in vertical sensitivity and distinct retrieval approaches likely also contribute. The difference is particularly large in September when background NH 3 is 5.3 × 10 15 molecules cm −2 more in CrIS than IASI, obtained as the intercept from regressing CrIS against IASI. The spatial correlation between CrIS and IASI multiyear means is R < 0.5 in most months (March, June-September), R = 0.53 in May, and R = 0.55 in April. If extreme values in the gridded products are retained, the spatial correlation degrades to R = 0.42 in April and R = 0.29 in May.

The GEOS-Chem Chemical Transport Model
We use the GEOS-Chem CTM version 12.1.0 (https://doi.org/10.5281/zenodo.1553349) to derive UK NH 3 emissions from IASI and CrIS. The model is driven with NASA GEOS-FP assimilated meteorology from the Global Modeling and Assimilation Office (GMAO). Model simulations are conducted on a horizontal grid at 0.25° × 0.3125° (∼25 km latitude × ∼31 km longitude) nested over western Europe (32.75°-61.25°N, 15°W-40°E). The model extends over 47 vertical layers from the Earth's surface to 0.01 hPa. Dynamic (3-hourly) boundary conditions are from a global GEOS-Chem simulation at 4° × 5°.
Anthropogenic emissions over the UK, including from agriculture, are updated in GEOS-Chem to include gridded emissions from the NAEI for 2016 (Ricardo, 2018a). These are annual totals on a 1 × 1 km grid available at https://naei.beis.gov.uk/data/map-uk-das (last accessed August 2019). The agricultural NH 3 emissions incorporated in the NAEI are calculated at coarser resolution (5 km) than the NAEI with the nitrogen balance models of Webb and Misselbrook (2004) for livestock sources and Misselbrook et al. (2006) for fertilizer sources. These models are driven with 30-years mean meteorology for 1981-2010, so the NH 3 emissions represent a climatological mean (Ricardo, 2019a). Other anthropogenic NH 3 emissions in the NAEI are typically calculated as the product of emission and activity factors representative of the year of interest and mapped to the 1 km NAEI emissions grid (Ricardo, 2018b). Mainland Europe anthropogenic emissions for 2016 are updated with the gridded (0.1° × 0.1°) product provided by the European Monitoring and Evaluation Programme (EMEP) (http://www.ceip.at/new_emep-grid/01_grid_data; last accessed September 2019. Now at https://www.ceip.at/the-emep-grid/gridded-emissions).
Temporal variability of annual NAEI and EMEP NH 3 emissions is represented in GEOS-Chem with gridded monthly scaling factors and spatially uniform diurnal scaling factors. Monthly scaling factors are from the Generation of European Emission Data for Episodes (GENEMIS) project detailed in Friedrich (2000). These lead to peak NH 3 emissions in April. Hourly scaling factors are from Zhu et al. (2015) calculated using information about the dependence of NH 3 on aerodynamic resistance, surface temperature, and Henry's law. As a result of these, 30% of NH 3 is emitted at midday (noon-2pm LST) coincident with the CrIS overpass and 20% in the morning (9am-noon LST) coincident with the IASI overpass. Natural NH 3 sources are from inventories already in GEOS-Chem. These include natural emissions from soils and the ocean from the Global Emissions InitiAtive (GEIA) inventory (Bouwman et al., 1997) and inland and coastal seabird emissions from the Riddick et al. (2012) inventory. We halve the GEIA inventory emissions, as in Paulot et al. (2014), informed by a 50% overestimate identified by Simpson et al. (1999). NH 3 is a semivolatile acid buffer that neutralizes acidic sulfate and nitrate aerosols, so its abundance depends on the abundance of these acidic aerosols. Sulfate forms from oxidation of SO 2 and nitrates from aerosol uptake of nitric acid formed from oxidation of NO x . The version of the NAEI we use includes outdated mapping of the location of ships and no vertical or temporal information for aircraft emissions. To address these issues, we separate ship and aircraft emissions from other sources in the lumped "Other Transport and Mobile Machinery" category of the NAEI emissions inventory and replace ship emissions with updated estimates that use geospatial information from the automatic identification system (Ricardo, 2017). We convert the NAEI aircraft emissions to monthly estimates and distribute these vertically up to 1 km (the altitude limit of the NAEI emissions) by deriving vertical and temporal scaling factors from the global Aviation Emissions Inventory version 2.0 (AEIv2) used in GEOS-Chem (Stettler et al., 2011). Above 1 km, the AEIv2 emissions are used. The existing temporal scaling factors in GEOS-Chem that are applied to NAEI SO 2 and NO x emissions lead to peak emissions in winter, due to an increase in energy demand. SO 2 is emitted in the model as 95% SO 2 and 5% sulfate, using sulfate-to-SO 2 emission ratios for Europe reported by Chin et al. (2000). NAEI emissions are gridded to a uniform 0.1° × 0.1° grid for input to the Harmonized Emissions Component (HEMCO) processing package version 2.1.010 (Keller et al., 2014) that maps all emissions to the model grid and applies relevant scaling factors.
The model includes detailed coupled gas-phase and aerosol-phase chemistry. Sulfate aerosols are formed in the model from oxidation of SO 2 in the gas phase by OH and in the aqueous phase in clouds by ozone and hydrogen peroxide (Park et al., 2004). Partitioning of NH 3 between the gas and acidic aerosol phase is determined dynamically with the thermodynamic equilibrium model ISORROPIA-II (Fountoukis & Nenes, 2007). Wet and dry deposition, terminal sinks of NH 3 , are represented with a standard resistances-in-series scheme for dry deposition (Wesely, 1989) and, for wet deposition, includes scavenging in and below clouds (Amos et al., 2012).
We use network site measurements of trace gases and aerosols to evaluate model accuracy at reproducing surface concentrations of NH 3 , SO 2 , and sulfate. These include two rural sites (Auchencorth Moss in Scotland, Chilbolton Observatory in southern England) that form part of the EMEP network and the mostly rural UK Eutrophying and Acidifying Atmospheric Pollutants (UKEAP) network. The two EMEP sites include hourly measurements from Monitor for AeRosols and Gases in Air (MARGA) instruments (Stieger et al., 2017;ten Brink et al., 2007;Twigg et al., 2015;Walker et al., 2019). The UKEAP network includes monthly mean measurements from low-cost denuder filter sampling packs (Tang et al., 2018). In 2016, there were 30 sites for SO 2 and sulfate and 51 for NH 3 . The MARGA data are from the EMEP Chemical Coordinating Centre EBAS database (http://ebas.nilu.no/; last accessed February 2020) (Tørseth et al., 2012) and the UKEAP data are from the UK-AIR data archive (https://uk-air.defra.gov.uk/data/data-availability; last accessed November 2020).
To ensure consistency between the model and observations, the model is sampled from the lowest to the top model layer during the satellite overpass times of 08-11 LST for use with IASI and 12-15 LST for use with CrIS, and as monthly 24-h means in the lowest model layer for comparison to the surface observations. The model is sampled in March-September 2016 following a 2-months spin-up for chemical initialization. Figure 3 shows the spatial distribution of annual UK NH 3 emissions for 2016 from the NAEI. Table 1 gives the breakdown by sector. Annual emissions for 2016 total 298 Gg, mostly (84%) from agriculture. Natural emissions of 21.6 Gg (7% of the total) are consistent with annual total natural emissions in GEOS-Chem of 21.8 Gg. According to GEOS-Chem, these include soils, vegetation and the ocean (together 18.7 Gg), and seabirds (3.10 Gg). NAEI anthropogenic NH 3 emissions total 276 Gg, 21 Gg less than the UNECE Gothenburg protocol emissions ceiling of 297 Gg (UNECE, 2019). The NAEI version we implement in GEOS-Chem and evaluate against top-down estimates was released in 2018. Two NAEI versions have been released since.

UK Bottom-Up Emissions of NH 3
Reported differences in NH 3 emissions across these versions for consistent years is minor, just 1-3% (Ricardo, 2019b(Ricardo, , 2020. The spatial patterns in Figure 3 coincide with farming activities that dominate NH 3 emissions according to the modeling study by Hellsten et al. (2008). They used the same Webb and Misselbrook (2004) (Hellsten et al., 2008), declined by 19% in the UK, though the relative proportion of urea-based fertilizer has increased (Ricardo, 2020). Regional changes in nitrogen-based fertilizers range from a 3% increase in Scotland to a 37% decrease in Northern Ireland (AIC, 2020).
Inversion of column densities of NH 3 to estimate top-down surface emissions can be complicated by dependence of NH 3 abundance on acidic sulfate aerosols formed from oxidation of SO 2 and acidic nitrate formed from uptake of nitric acid from NO x sources. UK SO 2 emissions are dominated by large industrial and energy sector point sources, ships, domestic and industrial combustion, and traffic (Ricardo, 2018b).
UK NO x emissions are dominated by transport, energy generation, and manufacturing (Ricardo, 2018b). We find particularly large discrepancies between monthly mean March-September 2016 observed (EMEP and UKEAP) SO 2 concentrations and those from the model driven with the NAEI ( Figure S3). The model normalized mean bias (NMB) is >600% for modeled SO 2 > 2 μg m −3 at sites influenced by point sources in Yorkshire and 205% for modeled SO 2 < 2 μg m −3 . Modeled sulfate is also greater than the observations (NMB of 17%) ( Figure S3). This would enhance partitioning of NH 3 to acidic aerosols to form ammonium, leading to a positive bias in the relative amount of NH x (NH 3 + ammonium) present as ammonium.
Positive model biases in both SO 2 and sulfate ( Figure S3) suggest an overestimate in NAEI SO 2 emissions that have implications for UK compliance with commitments to emissions ceilings and reductions. There are many factors other than emissions that could contribute to model biases. These include, but are not limited to, misrepresentation of the height at which SO 2 is emitted from tall stacks, a reported positive bias in mainland Europe SO 2 emissions (Luo et al., 2020), and uncertainties in dry (Fowler et al., , 2007 and wet (Luo et al., 2019) deposition. We conducted sensitivity simulations to assess the contribution of these uncertainties to modeled SO 2 and sulfate. Details of these simulations and the effect on SO 2 and sulfate concentrations are in the accompanying Supporting Information S1. The factor we find to have the largest influence relative to the model bias is wet deposition. The more efficient wet deposition scheme of Luo et al. (2019) leads to an 11% decrease in sulfate concentrations.
Errors in NAEI SO 2 emissions could be due to uncertainties in emissions from domestic and industrial biomass combustion. The third of six generating units at the 3.9 GW generating capacity Drax power station in Yorkshire transitioned from burning coal to biomass in 2016 (Simet, 2017). SO 2 emissions from biomass combustion depend on fuel sulfur content and combustion efficiency. Reported emission factors range widely from 1 to 110 mg SO 2 MJ −1 (Boersma et al., 2008;Paulrud et al., 2006;EMEP, 2019) and so offer limited constraints. To reduce the influence of a possible bias in SO 2 emissions on GEOS-Chem simulation of abundance of sulfate and NH 3 , we decrease land-based gridded (0.1° × 0.1°) NAEI SO 2 emissions by a factor of 3 for grids dominated by point sources (identified as grids with SO 2 emissions >10 g m −2 a −1 ) and by a factor of 1.3 for all other land-based grids. This reduces the original NAEI SO 2 emissions over land by 49% from 164 Gg to 84.1 Gg. With shipping, the updated annual NAEI SO 2 emissions for the domain shown in Figure 3 total 94.5 Gg. The March-September modeled sulfate NMB changes from +17% ( Figure S3) to −8.8%. We use the scaled SO 2 emissions in all subsequent simulations.

Top-Down NH 3 Emissions and Comparison to Bottom-Up Estimates
We calculate gridded satellite-derived 24-h monthly mean top-down NH 3 emissions (E sat ) as follows: where Ω sat is satellite observations of NH 3 multiyear monthly mean columns from IASI ( Figure 1 This suggests that larger Ω sat for CrIS than IASI is not just due to differences in midday and morning environmental conditions. The mass-balance approach that we use in Equation 1 to infer emissions can be susceptible to spatial misattribution of emissions due to displacement of NH 3 from the source. The global mean lifetime of NH 3 is ∼15 h (Hauglustaine et al., 2014), ranging from ∼2 h near large sources  to ∼36 h far from emission sources (Van Damme et al., 2018). The displacement length, the horizontal distance for the target compound to decay to ∼63% of the original concentration at the emission source, provides a measure of the spatial smearing or localization error of the satellite-derived emissions (Marais et al., 2012;Palmer et al., 2003). We estimate a smearing length for satellite-derived NH 3 emissions over the UK of 10-12 km for calm conditions (wind speeds of 5-6 km h −1 ) typical of the UK in summer ( Figure A1f.3 of BEIS (2016)) and a short NH 3 lifetime typical of large sources (2 h). At slightly windier conditions (7 km h −1 ) and over regions with lower emissions and a longer NH 3 lifetime (15 h), the displacement length increases to 105 km.
Maps of the resultant top-down monthly NH 3 emissions are shown in Figure 4 for IASI and Figure 5 for CrIS. Qualitatively, both estimates exhibit spatial patterns similar to the NAEI (Figure 3). This includes relatively low emissions along the Welsh border, and peak emissions in Northern Ireland, the northern portion of the English side of the Welsh border, and in Norfolk in the east. Emissions for retained grid squares total 271.5 Gg for IASI, whereas these are 43% more from CrIS (389.4 Gg). CrIS monthly emissions are 20%-38% more than IASI for March-July. This is similar in magnitude to the 25%-50% low bias in IASI columns, though for an earlier IASI product (Dammers et al., 2017;Whitburn et al., 2016a). The percentage difference increases to 57% for August and >100% for September. The large difference in September is due to 5.4 × 10 15 molecules cm −2 greater background NH 3 in CrIS, even after correcting for the baseline trend (Section 2.2, Figure S1). CrIS emissions excluding September are 33% more than IASI. Differences in sampling periods ( for IASI, 2013 for CrIS) only have a small effect on satellite-derived emissions, but lead to data gaps over Scotland and Northern England. IASI-derived emissions obtained for 2013-2018 are only 6% more (288.3 Gg) than those in Figure 4.
For comparison of monthly top-down and bottom-up emissions, we estimate monthly bottom-up emissions as the product of the annual NAEI emissions in Figure 3 and GEOS-Chem seasonality. The latter we obtain as ratios of GEOS-Chem monthly to annual 24-h NH 3 emissions interpolated onto the 0.1° × 0.1° grid. Figure 6 shows the resultant monthly bottom-up NH 3 emissions for April and July. The other months   are in the supplementary ( Figure S4). The bottom-up emissions peak in April (∼14% of the annual total) coincident with fertilizer application (Hellsten et al., 2007;Paulot et al., 2014). The gridded difference between top-down and bottom-up emissions is also shown in Figure 6 for April and July and in Figure S4 for the other months. Locations where bottom-up emissions exceed those from the top-down approach (red grids) mostly occur where emissions are low. The largest difference is in July when top-down emissions are 30 Gg more (IASI) and 46 Gg more (CrIS) than the bottom-up emissions. Pronounced regional differences include lower bottom-up values in eastern England, particularly in April, where fertilizer use and pigs and poultry farming are dominant sources, as well as in western England and Northern Ireland, particularly in July, where dairy cattle farming dominates (Hellsten et al., 2008). The spatial correlation between top-down and bottom-up gridded emissions in general ranges from R = 0.5 to R = 0.7, except for IASI in September (R = 0.34) when dynamic range in emissions is low.
The bottom-up emissions for March-September total 198.7 Gg. This is 27% less than IASI and 49% less than CrIS. It is unlikely that the relatively low bottom-up emissions are due to the time period (1981-2020) of the 30-years meteorology used to determine agricultural NH 3 emissions for the NAEI. We find that 2-m temperature from the NASA long-term consistent relanalysis product, Modern-Era Retrospective analysis for Research and Applications Version 2 (MERRA-2), is similar over the UK for 1981-2010 (282.750 K) and 1991-2020 (282.957 K). Bottom-up emissions in March-September are 67% of the annual total, similar to ∼60% for the monthly bottom-up NH 3 emissions estimated by Hellsten et al. (2007). If we use this relative contribution (60-67%) to scale IASI and CrIS to annual totals, this suggests annual NH 3 emissions of 405-453 Gg according to IASI and 581-664 Gg according to CrIS. Subtracting the UK annual natural NH 3 emissions of ∼22 Gg (Section 3) yields top-down annual anthropogenic NH 3 emissions of 383-431 Gg according to IASI and 559-642 Gg according to CrIS. Both top-down estimates exceed annual total anthropogenic emissions from the NAEI of 276 Gg (Section 3) and the Gothenburg protocol emissions ceiling of 297 Gg.
Figure 7 compares regional seasonality in UK NH 3 emissions from bottom-up and top-down estimates as the percent change in emissions in each month relative to those in June. Regional seasonality in the top-down emissions is very similar in March-August in all regions except Northern Ireland. The mismatch between IASI and CrIS in September is due to the at least 2-times greater CrIS than IASI columns in that month (Section 2). The July peak in emissions in Northern Ireland is more pronounced in IASI than CrIS. This is also apparent in the seasonality in the column densities ( Figure S5). Northern Ireland has experienced dramatic changes in agricultural activity that includes increases in livestock numbers of 45% for pigs and 42% for table chickens and a decline in nitrogen fertilizer use of 37% from 2000 to 2016 (DEFRA, 2020b). We find though that the top-down emissions estimates are relatively insensitive to differences in temporal coverage of the two sensors ( for IASI, 2013 for CrIS). All emission estimates exhibit a spring peak in April due to intensive fertilizer and manure application in March-April (Hellsten et al., 2007). Paulot et al. (2014) also identified this April peak in NH 3 emissions inferred from ammonium wet deposition measurements, though a recent study questions the utility of these measurements for constraining NH 3 emissions (Tan et al., 2020). A second summer peak in the top-down emissions in July that is not present in the bottom-up emissions could be due to the timing of manure spreading, dairy farming practices, or enhanced volatilization and suppressed dry deposition due to warm summer temperatures (Hellsten et al., 2007;Sutton et al., 1994). Spatial consistency between the July top-down emissions (Figures 4 and 5) and locations dominated by emissions from dairy cattle (Hellsten et al., 2008) suggests that it is due to dairy farming, but this requires further investigation.
In Figure 8, we compare March-September 2016 mean modeled and observed surface concentrations of NH 3 to determine if the model driven with NAEI NH 3 emissions and prior assumptions of NH 3 seasonality and diurnal variability corroborates the results obtained with the satellite observations. Monthly means from model grids coincident with the surface sites are reasonably spatially consistent with the surface observations (R = 0.54) and the model is 38.3% less than the observations. This is midway between the NAEI comparison to the top-down emissions of 27% less than IASI and 49% less than CrIS. There are also low-cost passive sampler measurements of NH 3 concentrations at 39 rural sites, but these have relatively low precision, are not as extensively distributed as the observations in Figure 8, and are only reliable (within ±10% of reference measurements) at NH 3 ≥ 2 μg m −3 (Martin et al., 2019;Sutton et al., 2001). Even so, the model is similarly biased low (by 41.5%) compared to these measurements (not shown).

Error Analysis of the Top-Down Emissions
The reported relative error for NAEI NH 3 emissions is 31% (Ricardo, 2018b). Quantifiable random errors that contribute to total March-September satellite-derived emissions include uncertainties in retrieval of NH 3 , and in the modeled relationship between NH 3 emissions and column densities (Equation 1). For the latter, we test sensitivity to modeled sulfate aerosol and nitric acid abundances and prior assumptions of the spatial and temporal variability of NH 3 emissions. IASI NH 3 retrieval errors for columns ≥2 × 10 15 molecules cm −2 range from 0.7% to 34%. Retrieval errors larger than 34% do occur, but are in locations with very low emissions. The CrIS NH 3 column errors across all grids range from 0.2% to 25%. Errors due to uncertainties in the magnitude and variability in SO 2 and NO x emissions that affect abundance of sulfate and nitrate aerosols and hence the abundance and vertical distribution of NH 3 are small compared to column density retrieval errors. We estimate the error contribution of these as the change in top-down emissions due to a perturbation in SO 2 emissions for sulfate and NO x emissions for nitric acid. The percent change in top-down emissions from a 50% decrease in SO 2 emissions is 4-5%. A 50% increase in NO x emissions increases nitric acid by 14%, aerosol nitrate by 11%, and satellite-derived NH 3 emissions by 8-9%. The limited sensitivity to sulfate and nitrate in the UK is because NH 3 is in excess due to the success of emission controls targeting SO 2 and NO x sources and absence of these for NH 3 sources. This would not occur in regions and times with large unregulated SO 2 and NO x sources. We find that (E/Ω) model used to convert satellite observations of column densities to emissions (Equation 1) is relatively insensitive to perturbations in NH 3 emissions, so is relatively unaffected by errors in the spatial and temporal variability of NH 3 emissions in GEOS-Chem. 10.1029/2021JD035237 13 of 18 A 50% increase in NH 3 emissions only causes a small (3-4%) decrease in satellite-derived NH 3 emissions. The total relative error from adding these individual errors in quadrature is 11%-36% for IASI and 9%-27% for CrIS and is dominated by errors in retrieval of the columns. Total emissions for March-September are 198.7 ± 61.6 Gg for the bottom-up emissions and up to 271.5 ± 97.7 Gg for IASI and 389.4 ± 105.1 Gg for CrIS.
There are also known systematic biases in the satellite observations. Some studies reported that IASI NH 3 column densities are biased low by 25%-50% compared to ground-based measurements (Dammers et al., 2017;. However, these comparisons were for earlier versions of the IASI NH 3 product. The version used here is consistent with columns derived with aircraft observations (Guo et al., 2021), though Guo et al. (2021) caution that their comparison is limited in time (summer) and location (Colorado, USA) and sensitive to errors in column estimates from integrating aircraft measurements. There are no observations of the vertical distribution of NH 3 over the UK. The CrIS column amounts display a gradual increase with time ( Figure S1) that we correct for in this work, though further work is required to determine the cause. Biases in the satellite-derived emissions due to differences in overpass times of the two instruments is mitigated by sampling modeled columns (Ω model in Equation 1) during the satellite overpass.
Both satellite products preferentially sample clear-sky conditions. The bias that this imparts on the topdown emissions estimates is challenging to quantify. The modeled emissions and columns used to derive top-down emissions ((E/Ω) model in Equation 1) are sampled under all-sky conditions, though there would likely be compensating effects of sampling clear-sky conditions on (E/Ω) model . Warmer temperatures and absence of clouds increase Ω by suppressing the amount of NH 3 that partitions to the aqueous phase (Stelson & Seinfeld, 1982;Walters et al., 2018), but E also increases in response to warmer temperatures (Sutton et al., 2013). Preferentially sampling clear-sky conditions likely has the largest impact on Ω sat . We find that the effect is greatest in July when boundary-layer clear-sky air temperatures, according to GEOS-Chem, are warmer than all-sky scenes by 5.6°C during the morning overpass and 5.3°C during the afternoon overpass. According to Sutton et al. (2013), 5°C warmer temperatures increase NH 3 emissions by 42%. Clear-sky temperatures are only 1.6°C-1.7°C warmer in the preceding month (June), so the greater clear-sky temperature in July may in part account for the discrepancies between observed and modeled NH 3 emissions in that month ( Figure 6) and the steep increase in July columns and emissions relative to June (Figures 7 and S5). A challenge though of using GEOS-Chem to diagnose sensitivity of air temperature to cloud cover is that the model is inferior to the satellite observations at resolving clouds, due to its coarser spatial resolution (25-31 km), and only 3%-12% of daily overpass model data are retained in each month after filtering for cloudy scenes (GEOS-FP cloud fractions >0.1). NH 3 emissions in GEOS-Chem also do not include changes in farming practices in response to shifts in meteorology.
Total top-down March-September emissions are 272 Gg from IASI and 389 Gg from CrIS. Bottom-up emissions estimated with the UK National Atmospheric Emission Inventory (NAEI) annual emissions and GE-OS-Chem monthly scaling factors are 27% less than IASI-derived emissions and 49% less than CrIS-derived emissions. This is supported by a 38%-42% underestimate in surface NH 3 concentrations from GEOS-Chem driven with the NAEI. We infer UK top-down annual anthropogenic NH 3 emissions of 383-431 Gg from IASI and 559-642 Gg from CrIS compared to 276 Gg from the NAEI. Seasonality in the top-down emissions confirms the well-known spring April peak from fertilizer and manure use, but there is also a summer July peak coincident with intensive dairy farming that is absent in the bottom-up emissions.
The relative error in the top-down emissions, mostly due to NH 3 column retrieval errors, is 11-36% for IASI and 9-27% for CrIS and is similar to the error reported for the NAEI (31%). The top-down emission estimates are relatively insensitive to model uncertainties in SO 2 , NO x , and NH 3 emissions, as NH 3 is in excess and the relationship between modeled NH 3 columns and emissions is near-linear.
Our study demonstrates the tremendous potential to use satellite observations to derive NH 3 emissions and assess bottom-up emissions under particularly challenging observing conditions (cloudy, cool) in the UK. This is critical for assessing reliability of inventories used to inform policies and mitigation strategies. The discrepancy between bottom-up and top-down emissions identified here warrants further investigation of both approaches.

Data Availability Statement
The top-down and bottom-up emissions estimated in this work are publicly available from the UCL Data Repository (https://doi.org/10.5522/04/14566635). The CrIS CFPR NH3 data are created by Environment and Climate Change Canada and hosted by the Meteorological Service of Canada (MSC) Datamart at https://hpfx.collab.science.gc.ca/∼mas001/satellite_ext/cris/snpp/nh3/v1_6_1/. Login is required and can be requested from data developer MWS (mark.shephard@canada.ca). The IASI NH 3 data are publicly available from the IASI data catalogue (https://iasi.aeris-data.fr/nh3/).