Atmospheric Ammonia in Sweden: Regional Modeling and Assessment of National Emission Reduction Benefits

2.1 Regional chemical transport modeling

The MATCH model (Robertson, Langner and Engardt, 1999) is an established regional-scale chemical transport model (CTM) with various gas and particle chemistry and physics schemes, including photochemistry, oxidation and other gas-phase reactions, secondary aerosol formation from gases, and dry and wet deposition of gases and aerosols. MATCH is used widely in research and operational work, including national air quality modeling in Sweden (e.g. Alpfjord Wylde, Leung and Andersson, 2023; Andersson, Langner and Bergström, 2007; CAMS, 2025; SMHI, 2025a).

Detailed descriptions of the model transport schemes and boundary layer parameterization can be found in previous works (Robertson, Langner and Engardt, 1999; Colette et al., 2025). The gas-phase chemistry follows the EMEP MSC-W EmChem09 scheme (Simpson et al., 2012) with modified isoprene chemistry (Carter, 1996), and reaction rates updated according to EmChem19 (Bergström et al., 2022). Details regarding the gas-phase chemistry mechanism are given in Supplementary Information Section S3.4. Gas-to-particle conversions include the formation of secondary inorganic aerosol (SIA; consisting of ammonium sulfate and nitrate) and secondary organic aerosol (SOA). Particles are treated as bulk aerosol in two size classes, corresponding to fine and coarse PM with diameters smaller and larger than 2.5 μm, respectively. Ammonium nitrate (NH₄NO₃) equilibrium is calculated according to temperature- and relative humidity-dependent equilibrium coefficient (Mozurkewich, 1993), and coarse nitrate formation is described by transfer of gaseous nitric acid (HNO₃) to aerosol-phase NO₃ (Strand and Hov, 1994).

Primary organic aerosol emissions are treated as non-volatile in the model setup used in this study. Secondary organic aerosol formation from oxidation of volatile organic compounds (VOC) is treated using a volatility basis set (VBS) scheme (Bergström et al., 2012). This includes anthropogenic SOA (ASOA) mainly from aromatic hydrocarbons, and biogenic SOA (BSOA) from isoprene, monoterpenes and sesquiterpenes. A simple scheme that does not include atmospheric aging of the SOA species is used (Langner et al., 2020), which is a slightly modified version of the VBS-NPNA scheme (Bergström et al., 2012).

MATCH has a number of different options for the treatment of wet and dry deposition of gases and particles. Here, relatively simple scavenging coefficients are used for gases and particles that are subject to wet deposition, as described in Supplementary Information Section S3.5. Dry deposition of gases by the resistance model, aerosol dry deposition, and stomatal resistance are described following the EMEP MSC-W schemes (Simpson et al., 2012). In this work, the dry deposition description of NH₃ is updated to a bidirectional flux parameterization (Wichink Kruit et al., 2017 and references therein) within the model development project Nordic Nature & Nitrogen to improve simulation of nitrogen in the Nordic countries (Frohn et al., 2025). While the standard resistance model assumes only one-directional flux of NH₃ from the atmosphere to the surface, the bidirectional flux approach considers also the opposite flux to the atmosphere. The opposite flux is based on compensation points, corresponding to concentrations already present in the media, that determine the ability to take up more gas. Here, the main focus is uptake on vegetated surfaces, as described in Supplementary Information Section S3.1.

MATCH is applied for the European domain at a spatial resolution of 0.2° × 0.2° and for a nested domain over Sweden at a resolution of 5 km × 5 km (see e.g. Figure 1 and Figure 2 for the two domains). The European domain is used for model evaluation by large sets of available quality-controlled observation data (Sections 2.2 and 2.3). The high-resolution domain is designed to be used in national air quality modeling for Sweden, and to study the national emission reduction effects. European-scale modeling is performed for selected full years, including 2017 and 2019, which are considered representative for recent trends. This excludes years corresponding to the COVID-19 pandemic and 2018, which was an unusually warm year in Sweden with exceptional heatwaves and forest fires. We focus on 2019, and perform additional model evaluation for 2017 to ensure the robustness of the model results. The Swedish emission reduction assessments are conducted for 2019.

The MATCH simulations use meteorological data from the European Centre for Medium-Range Weather Forecasts Integrated Forecast System (IFS) (ECMWF, 2025) with 50 hybrid vertical levels, reduced to 25 levels in the MATCH model corresponding to a vertical extent of approximately 6–8 km. In the present study, land-use data are based on the CCE CLC/SEI database (LRTAP, 2025); 16 different land-cover classes are included (as in Simpson et al., 2012). For the European-scale simulations, anthropogenic emissions are obtained from the CAMS-REG emission inventory, using version V5.1c and with emission year 2018 applied for all years 2017–2019 (Denier van der Gon, Gauss and Granier, 2023, pp. 8–18; Kuenen et al., 2022). Swedish anthropogenic emissions from the SMED database (SMED, 2025) are used for the Swedish-domain simulations. Emissions are distributed temporally according to the CAMS-REG-TEMPO-v3.1 profiles (Denier van der Gon, Gauss and Granier, 2023, pp. 32–47; Guevara et al., 2021). The emission-sector-dependent split of total volatile organic compounds into the different VOC model species (Table S2) is based on data from CAMS (Kuenen et al., 2022; following a similar methodology as Bergström et al., 2022). Details and further information about the emission setup are given in Supplementary Information Section S3.2.

Biogenic emissions of isoprene and monoterpenes are calculated in the model (using the methodology of Simpson et al., 2012); sesquiterpene emissions are also added based on plant chamber experiments with no observable biotic stress (5% of the monoterpene emissions; following Bergström et al., 2014). Emission of sea salt particulate matter is modelled based on a parameterisation considering wind speed, water salinity and water temperature (Sofiev et al., 2011). Natural aeolian dust emissions are modelled based on the DEAD model (Zender, Bian and Newman, 2003). Gaseous and particulate emissions from biomass fires are taken from the GFAS fire emission database (Kaiser et al., 2012) (Supplementary Information Section S3.3). Emissions of sulfur species from seas and volcanoes (here Etna, Stromboli, Vulcano) are included in a simplified manner; oceanic dimethyl sulfide is treated as oxidized to SO₂ and sulfate before being introduced in the model.

Boundary data for the European-scale domain is obtained from the IFS global forecasts for the following species: NO, NO₂, HNO₃, PAN, O₃, SO₂, CO, CH₄, C₂H₆, C₂H₄, C₅H₈, HCHO, sulfate and desert dust similar to the operational CAMS system (Colette et al., 2025), but in the present study the global boundary data are read with 6-hour intervals. For other model species, the model uses seasonal climatological boundary concentrations. Results of the European-scale simulations are used as boundary data for the nested Swedish-scale simulations for all advected model components; these boundary data are read with 3-hour intervals and interpolated in the model.

2.2 Studied species

To assess the impacts of emission reductions, we examine selected air quality and environmental indices that are expected to be notably affected by ammonia:

• Gas-phase NH₃

• PM_2.5

• Nitrogen deposition (Ndep)

For model evaluation, we also apply available observation data for related nitrogen species:

• Total ammonia and ammonium NH_x = NH₃(g) + NH₄⁺(pm)

• Total nitric acid and particulate nitrate ΣNO₃ = HNO₃(g) + NO₃^–(pm)

Finally, in addition to the species directly related to ammonia, we benchmark the model setup against observation data for other common pollutants, namely:

• O₃

• NO₂

• PM₁₀

We study and evaluate the atmospheric concentration data at hourly or daily time resolution, depending on the component: O₃ and NO₂ components exhibit distinct diurnal cycles that are typically well described by models, and thus hourly resolution is applied. For other gas and aerosol components, daily mean values are used due to both availability of observation data (see Section 2.3) and diurnal variations being less important for model evaluation. For deposition, seasonal accumulated values are studied.

2.3 Observation data

Model results are evaluated against surface observations from atmospheric monitoring sites at hourly or daily temporal resolution, retrieved from the EBAS atmospheric database (EBAS, 2025). For NH₃, we use also additional observations collected from national databases: data collected by Aarhus University for Denmark (Ellermann et al., 2021; stations DK-01, DK-09, DK-10), and the RIVM database for the Netherlands (RIVM, 2025; stations NL10131, NL10444, NL10538, NL10633, NL10738, NL10929).

For some species, here NH₃, PM_2.5 and PM₁₀, observation data are available at both hourly and daily resolutions, typically for different stations. For these species, hourly data are converted to daily means for consistency and for improved spatial data coverage. For calculating evaluation statistics for given periods (e.g. monthly or seasonal), at least 67% (2/3) observation data coverage is required.

For NH_x, data are combined by complementing available NH_x data with additional reported NH₃ and NH₄ data aggregated into the sum component NH_x. This allows including four Danish stations that only report NH₃ and NH₄ separately but are relevant for evaluating model performance in Northern Europe.

2.4 Emission reduction assessments

The effects of Swedish national emission reductions are tested by scaling down the emissions from the given sectors by 20%. The scaling is applied temporally and spatially uniformly to (1) NH₃ emissions from the agricultural sector, and (2) NO_x (as NO₂) emissions from the road transport sector. These tests are performed using the high-resolution domain over Sweden and surrounding areas.

The magnitude of the test reductions is in line with the Swedish commitments for 2030, for which NH₃ and NO_x emissions need to be reduced by 7% and 37%, respectively (EEA, 2024a). For simplicity, the same reduction is applied for both components. For NH₃ the scaling factor is more ambitious, but such goals are relevant for assessing the potential effects and benefits of lower NH₃ emissions, considering the urgent need to reduce the emissions both in Sweden and elsewhere.

3.1 Modeled regional concentrations

Figures 1 and 2 present seasonal means of gas-phase NH₃ simulated by the MATCH model for the European and Swedish domains, respectively. The highest concentrations in Europe, appearing at spring and summer time, are located around major agricultural sources (Figure S18). In the vicinity of Sweden, these include sources in Denmark, the Netherlands and Northern Germany (Figure 1). As expected, total reduced nitrogen species NH_x exhibit similar patterns as NH₃ (to some extent, the modeled total oxidized nitrogen also shows similar distributions; Figure S1). NH₃ concentrations are generally lower in Northern Europe except for Denmark, with the highest Nordic concentrations (excluding Denmark) located in southern parts of Sweden and Finland, and Southern and mid-Norway. In Sweden, elevated NH₃ levels are centered upon the southernmost areas (Figure 2).

Figure 1

Figure 2

3.2 Model–measurement comparisons

Figure 3 shows monthly modeled and observed NH₃ concentrations, normalized mean bias (NMB) and correlation over all stations for the two model domains. Seasonal station-wise NMB and correlation for the European domain are shown in Figure 4; the station-wise statistics are similar for the Swedish-domain simulation. The global monthly trends, here characterized by the median and the 25^th and 75^th percentiles, are qualitatively well represented by the model (Figure 3, panels (a) and (d)). The modeled and observed distributions show a spring peak due to application of fertilizers, followed by elevated levels in summer time and decline towards winter.

Figure 3

Figure 4

The available observations primarily cover Fennoscandia and Netherlands, with a few individual stations in United Kingdom and Eastern Europe (Figure 4). In general, modeled concentrations tend to be lower than measured values, with some overestimation in spring and autumn seasons mainly at a few stations in Denmark and Southern Sweden (Figure 4a). Correlation is generally higher in spring and summer with elevated NH₃ (Figure 4b). The global monthly NMB for all available stations in the European domain varies between –76% and –19% (Figure 3b), with a global annual NMB of –44% (Table S1). For global correlation, the monthly and annual values are 0.59…0.86 (Figure 3c) and 0.65 (Table S1), respectively. The NMB and correlation values can be considered satisfactory compared to previous NH₃ model studies applying a regional CTM, reporting absolute NMB of 48% and correlation of 0.59 for weekly or monthly surface observations in Netherlands and Germany (Ge et al., 2020).

The overall model results are improved with the updated NH₃ scheme, which decreases the positive bias in Denmark and Southern Sweden (Section S2.2; Figure S13). As shown in Figure S2, the new scheme predominantly reduces the predicted NH₃ in the European domain, although some increases occur depending on location and season. The changes are due to the net effects of the updated deposition flux and the introduction of the compensation points (Supplementary Information Section S3.1). The compensation points, that determine the upward flux, peak in spring and summer, as summarized in Supplementary Information Section S1.1. Within Sweden, modeled NH₃ is decreased by up to approximately a factor of 2. The relative changes are similar over the main NH₃ emission areas in Southern and Central Sweden, except for the most intensive national agricultural hotspots that are less affected likely due to elevated compensation points. It can be noted that the modeled NH₃ may involve more uncertainties in clean remote regions with no observations. Compensation points in such environments are less well understood, and ground emissions may occur especially under warm and dry conditions (Walker et al., 2023; Wu et al., 2023).

Evaluation statistics for the other nitrogen species NH_x and ΣNO₃, summarized in Figures S14, S15 and S17, are comparable to those of NH₃, or in some cases better. For NH_x, the global monthly NMB varies between –13% and 72%, but values higher than 25% (40%…72%) only occur in March–May. For ΣNO₃, the bias is within –28%…24%. Global monthly correlation for both species is within 0.60…0.81 (Supplementary Information Section S2.2). Model evaluations for the standard regulated pollutants O₃, NO₂, PM_2.5 and PM₁₀ are presented in Figures S9–S12 and S16, showing performance similar to previous MATCH and other model studies (e.g. Frohn et al., 2022). It can be noted that while PM_2.5 and PM₁₀ involve more prominent systematic bias compared to the other species, this is attributed to PM components other than ammonium and nitrate that are the focus of this work (Supplementary Information Section S2.1). The European-domain simulation for 2017 shows generally similar evaluation results for the studied species in terms of NMB and correlation (Table S1), and their temporal and spatial trends. The main difference to 2019 is that NH₃ doesn’t exhibit a sharp global peak in April (as in panel (a) in Figure 3). Instead, observations and model results for 2017 show only moderately elevated levels or a low maximum, respectively. This can be partly due to a few more observation stations in central Europe being available and included in the statistics. ΣNO₃ involves less global month-wise variation with a rather flat median curve (compared with panel (a) in Figure S15) for both observations and model. Overall, the simulation years show similar model performance for the nitrogen species, supporting the application of the model setup for the ammonia reduction assessments. While multi-year simulations are outside the scope of the present study, long-term trends in PM simulated by the MATCH model have been evaluated in previous work (Tsyro et al., 2022). The previous European-scale simulations for 2000–2010 show decreases in PM following the European emission reductions, in line with observations and with consistent model bias without significant year-to-year variability.

3.3 Air pollution mitigation potential in Sweden by NH₃ reduction

Modeled changes in nitrogen-containing species for the Swedish national emission reduction tests are presented in Figures 5, 6, 7. Decreases in gaseous NH₃ upon 20% reduction of agricultural NH₃ emissions are highest in regions with elevated NH₃ levels and hotspots in Southern Sweden (Figure 5; cf. Figure 2). In these areas, the corresponding relative changes are within ca. –10…–20%, which suggests that such emission reductions can effectively decrease NH₃ levels.

Figure 5

Figure 6

Figure 7

Concentrations of fine particulate matter PM_2.5 and predicted effects of 20% national emission reductions in (1) agricultural NH₃ and (2) road-transport NO_x (as NO₂) are shown in Figure 6. The decreases in seasonal PM_2.5 upon reduction of NH₃ (panel (b)) show similar spatial patterns as decreases in NH₃ (Figure 5), as NH₃ can directly contribute to inorganic PM by nitrate or sulfate formation with NO_x and SO₂ oxidation products, respectively. The largest decreases in PM_2.5 occur during the winter and autumn seasons, when NH₃ concentrations are lower. The impact is generally smaller during spring and summer with elevated NH₃ levels, as PM formation is less limited by ammonia under these seasons.

By contrast, effects of NO_x reduction on PM_2.5 are close to negligible (panel (c)), even if NO₂ levels are decreased especially in Stockholm region (Figure S3). NO_x needs to undergo oxidation to form nitric acid, which finally can contribute to nitrate PM in the presence of NH₃. That is, NO_x reduction effects on secondary inorganic PM are limited by the availability of both oxidants and NH₃, and the contribution of sulfate PM. These results agree well with previous European-scale assessments of season-dependent NH₃ contributions to PM_2.5 (Backes et al., 2016; Clappier et al., 2021), and with European and global estimates of the benefits of NH₃ abatement over NO_x (Clappier et al., 2021; Gu et al., 2021). In Figure 6c, the minor decreases in PM_2.5 upon NO_x reduction apply mainly to areas affected by agricultural NH₃, but their magnitude is not sensitive to the NH₃ levels (Figure 2). NO_x reduction can also increase PM_2.5 through increases in O₃ (and other oxidants) and hence in the oxidative potential of the atmosphere (Gaubert et al., 2021; Jhun et al., 2015). Here, very small increases in PM_2.5 are seen in Northern Sweden due to minor increases in O₃ (Figure S4), mainly affecting secondary organic PM. The O₃ effect may also make inorganic PM reduction less efficient due to increased NO₂ and SO₂ oxidation (Clappier et al., 2021).

Figure 7 shows the impacts of NH₃ and NO_x reductions on total deposited nitrogen mass (Ndep), including dry and wet deposition. Similarly to PM_2.5 (Figure 6), reducing NH₃ decreases Ndep following the NH₃ reduction patterns (Figure 5), while NO_x shows no or negligible effects. Here, also the season-dependent magnitude of the Ndep reduction follows that of the NH₃ reduction, i.e. it shows largest reductions in spring and summer time, as the atmospheric concentration is directly linked to deposition. Here, NO_x effects are limited by the relatively smaller contribution of nitrogen mass to the total emissions compared with NH₃ (as a large fraction of emitted NO_x mass is oxygen). A substantial contribution to the decreases in Ndep upon NH₃ reduction (Figure 7b) comes from reduced dry deposition of gaseous NH₃. Dry deposition, for which NH₃ gas is a major component, constitutes a notable fraction of total nitrogen deposition as shown in Figures S5 and S6. Here, changes in dry deposition of NH₃ contribute significantly to the total Ndep changes (panels (b) in Figure S5 and Figure 7, respectively). Reduced wet deposition has large impacts, especially in the areas with the largest total Ndep (and NH₃) decreases in Southern Sweden in the summer period (Figure 7b).

3.4 Implications

Overall, the model performance for the studied reduced and oxidized nitrogen components NH₃, NH_x = NH₃(g) + NH₄⁺(pm) and ΣNO₃ = HNO₃(g) + NO₃^–(pm) is satisfying, especially in capturing the qualitative seasonal variations. The observed temporal trends in monthly medians and 25^th–75^th percentiles are qualitatively reproduced by the model, and the NMB and correlation statistics are comparable to previous regional model studies (Section 3.2).

Results obtained by the model setup for NH₃ and NO_x emission reduction impacts in Sweden highlight the following implications:

• Reducing agricultural NH₃ emissions is potentially beneficial for both human health and the environment through decreases in airborne fine particulate matter (PM_2.5) and nitrogen input to ecosystems by atmospheric deposition (Ndep), respectively. Although the nitrogen budget of Sweden is affected by emissions in other countries through long-range transport, reducing national NH₃ emissions has distinct impacts on air quality and environmental indicators in Sweden, especially in regions around the emission sources. It can be noted that here the reduced nitrogen input through direct deposition primarily applies to land ecosystems over areas affected by the agricultural sources, with only minor reductions for the Baltic Sea and coastal areas.

• Sweden’s most urgent reduction targets currently include both NH₃ and NO_x. NO_x reductions contribute to improved air quality by decreased NO_x concentrations, but the present assessments suggest very minor or no effects on PM_2.5 and Ndep. The impacts naturally depend on the applied emission scaling factors, here 20% for both species, corresponding to approximately equal reductions in total emitted mass (9.5 ktons NO_x and 9.3 ktons NH₃, respectively). The potential to reduce Ndep depends on the reductions in emitted nitrogen (N) mass, which is here approximately 2.6 times higher for NH₃ (i.e. the relative reduction in NO_x emissions would need to be substantially higher than 20% to achieve similar theoretical potential as NH₃ to reduce the nitrogen inputs). On the other hand, large NO_x reductions will also lead to changes in concentrations of O₃ and other oxidants (OH and nitrate radicals), which may enhance secondary PM formation—and thus lead to increasing effects on PM_2.5—in some regions and seasons. This makes the health and environmental impacts of changes in NO_x emissions more complex.

• The NH₃ reduction impacts on PM_2.5 exhibit a significant seasonal dependence due to the seasonally varying NH₃ concentrations in relation to acidic PM precursors. NH₃ reductions are most beneficial in ammonia-limited conditions, that is, in the winter half-year with low NH₃ concentrations. However, it must be noted that NH₃ reduction is always beneficial for decreasing Ndep, and especially at high NH₃ concentrations. Here, a notable contribution to decreases in total nitrogen deposition is due to reduced direct dry deposition of gaseous NH₃. The minor effects of NO_x reduction on PM_2.5 suggest that PM_2.5 formation in Sweden is primarily not limited by NO_x. In addition to NH₃ reductions, secondary inorganic PM could potentially be mitigated by sulfate reduction through SO_x emissions, especially in the summer season when neither NO_x nor NH₃ reduction show significant effects. On the other hand, Swedish SO_x emissions are already relatively low and the potential for further sulfate PM reductions may be limited.