Introduction

Ambient PM2.5 (fine particulate matter with an aerodynamic diameter ≤2.5 µm) air pollution is one of the leading risk factors for premature mortalities worldwide, according to the Global Burden of Disease (GBD) study1, responsible for millions of deaths and lost years of healthy life annually in recent years1,2,3. Long-term policies for PM2.5 mitigation have been implemented in many countries and have effectively reduced PM2.5 concentrations4,5,6,7. However, large numbers of people are still exposed to harmful PM2.5 levels even in places with relatively clean ambient air such as Europe7; 59% of European stations exceed the World Health Organization (WHO) guideline for the PM2.5 annual mean (10 μg·m−3) in 20198. Recent epidemiological studies also demonstrated that PM2.5 air pollution can affect human health at very low levels9,10. The WHO thus released an updated guideline value for PM2.5 annual mean concentrations (5 μg·m−3)3, which was exceeded at 97% of European monitoring stations in 20198. This poses a tremendous challenge for cleaning up European air as much more stringent mitigation measures will be needed to achieve such an ambitious goal. The WHO also suggested interim targets to be considered, although recognizing that interim targets are insufficient to remove adverse health impacts.

Excess reactive nitrogen (Nr), including nitrogen oxides (NOx = NO + NO2), ammonia (NH3), nitrate (NO3−), and ammonium (NH4+) are recognized environmental threats to ecosystems, deteriorating the quality of air, soil, and water11,12. Anthropogenic Nr sources have dramatically increased since 196013, exacerbating the global nitrogen cycle and consequent damaging effects on human health and ecosystems13,14. Capping anthropogenic Nr emissions (mainly NOx and NH3) is a high priority for environmental protection15,16. In particular, Nr controls benefit PM2.5 mitigation because both NOx and NH3 are precursors of secondary inorganic aerosols (SIAs, including sulfate, nitrate, and ammonium) components in PM2.5, apart from sulfur dioxide (SO2)14,17,18. SIAs strongly contribute to the PM2.5 mass concentrations in Europe19,20,21, contributing above 50% of total annual PM2.5 mass concentrations in parts of Europe, i.e., Germany, the Netherlands, and Belgium22,23,24. Atmospheric abundance of NH3 and NOx gases determine the formation of SIAs, and effectiveness of PM2.5 mitigation from Nr controls25,26,27. NH3 preferably reacts with sulfuric acid (H2SO4, produced by the oxidation of SO2) to form ammonium sulfate aerosol, and with more NH3 available, further reacts with nitric acid (HNO3, produced by the oxidation of NOx) to form ammonium nitrate aerosol.

The revised Gothenburg Protocol has set national Nr emission ceilings for 2020, i.e., 42% NOx emission reductions and 6% NH3 emission reductions in 2020 relative to 2005 for the European Union (EU)28 and other participating countries. The National Emissions Ceiling Directive further establishes national Nr emission reduction targets in 2030, i.e., 63% NOx emission reductions and 19% NH3 emission reductions in 2030 relative to 2005 in the EU29. All existing national targets show more ambitious controls for NOx than NH3. Most countries have not prioritized limiting NH3 emissions in part due to uncertainties in NH3 sources and concerns about its control effectiveness for PM2.5 mitigation, in addition to food security concerns17, with agriculture being the dominant source of NH3. However, recent studies found agricultural (mainly NH3) emissions make the largest relative contribution to PM2.5 mortality in Europe among all sources2,30,31. Gu et al. 18 also found that the cost of 50% NH3 emissions abatement is much less than that of NOx emissions globally. However, the priority for NOx or NH3 emission reductions to meet the updated WHO guideline and zero pollution action plan32 in Europe remains uncertain.

In this study, we quantify the contribution and efficiency of Nr emission reductions for PM2.5 mitigation in Europe for 2015 and derive the optimal pathway for Nr emission controls. We use recent European emission estimates, a regional air quality model, the newly developed exposure mortality model, Nr control scenarios, and emission control costs to systematically analyse the impact of Nr emission controls on PM2.5 air pollution (Methods). We demonstrate that Nr emission controls can reduce PM2.5 concentrations, PM2.5-related health impacts, and help achieve the WHO guideline in Europe. The optimal pathway targeting PM2.5 abatement changes towards prioritizing NH3 measures after considering control costs, indicating NH3 emission reductions are the most cost-effective way to combat European PM2.5 air pollution.

Results and discussion

The contribution of Nr emissions on PM2.5 air pollution

Our ECLIPSE inventory derived from the GAINS (Greenhouse gas and Air pollution Interactions and Synergies) model (Methods) estimates the total anthropogenic NO2 and NH3 emissions over Europe in 2015 to be 3.7 Tg N and 4.4 Tg N, respectively, which are comparable to other emission inventories (Supplementary Table 1). Considering monthly time factors, NH3 emissions tend to peak during the warm season (April-September), while NOx emissions peak during the cold season in Europe. Such seasonality appears stronger in our estimates than in the EDGAR and EMEP inventories (Supplementary Fig. 1). Both NH3 and NOx emissions are higher in the western part of Europe than in the east. This ECLIPSE emissions inventory is used as an input to the Weather Research and Forecasting model coupled with Chemistry (WRF-Chem) regional air quality model to assess the impacts of Nr emission reductions on PM2.5 air pollution in Europe. A series of WRF-Chem simulations are conducted over Europe for the representative months (January, April, July, and October) in 2015 (Methods). The baseline simulation in Europe, after improving simulated organic carbon (OC) and dust by matching observations of PM2.5 components (Supplementary Figs 2 and 3), well captures measured surface PM2.5 concentrations with the correlation coefficients (R) > 0.59 and mean bias (MB) < −6% (Supplementary Fig. 3). The magnitudes and variations of the observed SIAs concentrations are generally captured by the baseline simulation, except in summer, when the model underestimates nighttime nitrate volatility and overestimates nitrate concentrations. The simulated surface annual NH3 concentrations are also in good agreement with measurements in Europe with R = 0.92 and MB of within −3% (Supplementary Fig. 4).

The contribution of anthropogenic Nr emissions to PM2.5 air pollution can be calculated as the difference between the baseline simulation and a sensitivity simulation with anthropogenic Nr emissions set to zero (Methods). Figure 1 shows that the reduction in regional annual mean PM2.5 concentrations when phasing out anthropogenic Nr emissions is 2.3 ± 1.2 μg·m−3 (mean ± standard deviation) in Europe for 2015. The response to such emission controls for PM2.5 concentrations is stronger in Western Europe than in Eastern Europe, with the largest effects occurring in the Netherlands, Belgium and northern Germany. For evaluation, we separate Europe into Western Europe and Eastern Europe along country borders, guided by the spatial difference of PM2.5 changes from Nr emission controls (the thick black line in Fig. 1). We further apply a metric of N-share18 to quantify the contribution of Nr compounds to total PM2.5 concentrations, which is defined as the relative change in model simulated PM2.5 concentrations with vs. without anthropogenic Nr emissions. The N-share caused by anthropogenic Nr emissions contributes about 29% (range, 17–31%) to PM2.5 pollution in Western Europe and 12% (8.7–16%) in Eastern Europe for 2015, exceeding 50% in some parts of Western Europe. The N-shares of NH3 emissions are larger than those of NOx emissions and close to the N-shares of total Nr emissions because the NH3 reductions curtail both contributions of NOx and SO2 to SIAs formation, which is in agreement with results of Gu et al.18.

Fig. 1: Contribution of reactive nitrogen (Nr) emissions to PM2.5 air pollution over Europe in 2015.
figure 1

a–c Changes of PM2.5 concentrations induced by phasing out anthropogenic Nr (NOx + NH3) emissions (a), NOx emissions (b), and NH3 emissions (c) respectively. d–f N-shares of PM2.5 air pollution associated with Nr (d), NOx (e), and NH3 (f) emissions.

Full size image

Nr abatement would help Europe to achieve the limit set in the updated WHO guidelines for PM2.5 concentrations and substantially mitigate PM2.5-related health burdens. In 2015, only 14% of Western Europe met the PM2.5 annual mean <5 μg·m−3 (the updated WHO guideline value) and all of Eastern Europe exceeded this guideline level. Figure 2 shows that phasing out Nr emissions prompt an additional 28% of Western Europe to achieve the guideline value for annual mean PM2.5. NH3 emission controls render twice as much area in Western Europe meeting the guideline value compared to similar strengths of NOx emission controls. However, annual mean PM2.5 concentrations in Eastern Europe cannot reach this guideline value with Nr abatement alone and need to reduce emissions of other PM2.5 precursors. In addition, Western Europe and Eastern Europe have 18% and 38% of all days in 2015 exceeding the guideline value for the daily average PM2.5, and Nr abatement cuts the daily exceedance by 41% and 16% respectively. The Global Exposure Mortality Model (GEMM) is then applied to assess PM2.5-related chronic health impacts (Methods). We further find setting anthropogenic Nr emissions to zero could avoid 99,000 (95% confidence interval: 92,000-106,000) PM2.5-related premature deaths in Europe in 2015, decreasing the annual PM2.5-related mortality by 29% and 6% in Western Europe and Eastern Europe, respectively (Supplementary Fig. 5).

Fig. 2: Impacts of reactive nitrogen (Nr) emission controls on the World Health Organization (WHO) air quality guideline (AQG) achievement.
figure 2

a Changes in regional mean daily WHO AQG (15 μg·m-3) PM2.5 level exceedance when European Nr (solid lines), NOx (long dash lines), and NH3 (short dash lines) emissions are gradually decreased in 2015. b–d the reduction in Nr (b), NOx (c), NH3 (d) emissions required to meet the WHO AQG for annual mean PM2.5 (5 μg·m−3). Blue, red, and black lines in (a) represent changes of regional mean daily PM2.5 exceedance in Western Europe, Eastern Europe, and all Europe, respectively. Gray areas in (b), (c), and (d) represent locations where even 100% Nr emission controls cannot lead to achievement.

Full size image

The efficiency of Nr emission reductions in Europe

The analyses above illustrate a larger contribution of NH3 emissions to PM2.5 concentrations, and that phasing out NH3 emissions leads to larger areas and more days meeting the updated WHO guideline value for PM2.5 air pollution compared to phasing out NOx. We further investigate the effectiveness of Nr emission controls under different reduction levels in sensitivity simulations with NH3 and NOx emission reductions of 0%, 30%, 60%, 80%, and 100% over Europe in 2015 (Methods). We should note such reduction strengths extend to very drastic changes of atmospheric conditions that, while currently not seeming realistic, guide the way towards conditions compatible with the WHO guideline values and provide the information needed to devise efficient abatement.

Figure 3 shows changes in regional mean PM2.5 concentrations and related premature deaths in Western Europe and Eastern Europe as anthropogenic Nr emissions are gradually reduced. In Western Europe, the PM2.5 concentrations decline non-linearly following NH3 emission reductions, resulting in modest PM2.5 changes with limited NH3 emission reductions, which is similar to the response found in China26. The regional mean PM2.5 concentrations in Western Europe for 2015 would decrease by 0.40 ± 0.15/1.03 ± 0.41/2.51 ± 1.06 μg·m-3 with 30%/60%/100% NH3 emission reductions in Europe. Only a deep NH3 abatement (up to about 80%) would yield larger total PM2.5 decreases in Western Europe than the same level of NOx abatement. We note a difference to previous studies33,34 that expect higher efficiency for NH3 already at a much lower level of abatement, which we understand to be the result of a change in the chemical regime since these earlier studies were performed. While PM2.5 decreases in Eastern Europe associated with NH3 emission reductions tend to be more linear than those in Western Europe, the responses are similar to NOx emission reductions. Notably, during summer, both regions exhibit a more linear relationship between PM2.5 concentrations and NH3 emission reductions, primarily due to the greater availability of HNO3 compared to other seasons (Supplementary Fig. 6). Meanwhile, we find stronger non-linear responses on PM2.5-related premature deaths in both regions due to the non-linear relationship between health risk and PM2.5 exposure; their values can be decreased by 8.7 (8.0–9.3)/24 (22–26)/72 (67–78) thousands PM2.5-related premature deaths in Western Europe for 2015 when NH3 emissions are reduced by 30%/60%/100%.

Fig. 3: Effectiveness of reactive nitrogen (Nr) emission reductions in Europe in abating regional mean surface PM2.5 air pollution in 2015.
figure 3

a, b Reductions in Western Europe (a) and Eastern Europe (b) annual mean PM2.5 concentrations (black lines), and PM2.5-related premature deaths (red lines) when European Nr (solid lines), NOx (long dash lines), and NH3 (short dash lines) emissions are gradually abated in 2015. c, d changes in Western Europe (c) and Eastern Europe (d) annual mean instant efficiency (refers the instant response of PM2.5 in percentage to 1% reduction in Nr emissions under each Nr emission scenario) associated with Nr (purple bars), NOx (green bars), and NH3 (red bars) emission controls. The baseline simulated regional annual mean PM2.5 concentrations (denoted as “C”), and PM2.5-related premature deaths (denoted as “Deaths”) are shown in (a) and (b). Shading in (a) and (b) represent values (means ± one spatial standard deviation) of PM2.5 concentrations or PM2.5-related premature deaths. Vertical bars in (c) and (d) represent values (means ± one spatial standard deviation) of instant efficiency.

Full size image

We quantify the effectiveness for PM2.5 reductions by calculating the instant efficiency of Nr emission controls (β, Methods). βNr/βNOx/βNH3 estimates the instant response of total PM2.5 mass in percent for each 1% mass reduction in Nr/NOx/NH3 emissions. As shown in Fig. 3, the regional mean βNH3 increases rapidly while βNOx slowly decreases as the level of emission reduction rises. Under the 2015 emission condition, the βNH3 efficiencies in Western Europe increase from 0.12 ± 0.05%/% in the base condition to 0.48 ± 0.33%/% (a factor of 4 higher) when NH3 emissions are reduced by 80%, while the βNOx efficiencies decrease from 0.25 ± 0.13%/% to 0.14 ± 0.07%/%. The efficiencies in Eastern Europe are less than half of those in Western Europe due to the higher PM2.5 concentrations and lower N-share (Fig. 1), and are less sensitive to Nr emission changes.

We then apply the chemical regime metric of G ratio25 to explain the changes in the instant efficiency associated with NH3 emission controls. The G ratio denotes the ratio between free ammonia (NH3 and NH4+) and total nitrate (HNO3 + NO3−) after neutralization of H2SO4 (note, all the terms are expressed on a molar basis, Methods). The 2015 mean G ratio is found to be almost always above 1 across Europe (4.5 ± 2.4 in Western Europe and 2.8 ± 1.7 in Eastern Europe), indicating a HNO3-limited chemical regime and causing the SIAs formation to be more sensitive to small changes in NOx emissions than those in NH3 emissions (Supplementary Fig. 7). This phenomenon is particularly evident in April (with a G ratio of 6.3 ± 3.2 in Western Europe and 4.2 ± 2.0 in Eastern Europe) due to the high ammonia emissions occurring in Europe during this month. In contrast, the G ratio is close to 1 in most of Eastern Europe for January, July, and October, when NH3 controls are slightly more efficient than NOx controls (Supplementary Fig. 6). The G ratio decreases as we gradually reduce NH3 emissions and Europe shifts to the NH3-limited chemical regime, leading to NH3 abatement becoming increasingly effective (Supplementary Fig. 8). When we gradually reduce NOx emissions, Europe remains in the HNO3-limited chemical regime but the βNOx decreases due to decreases in oxidants (Supplementary Figs. 8 and 9). Changes in βNr depend on both βNOx changes and shifts in the chemical regime. This results in a slow decrease of βNr in Western Europe and a trend of first decreasing and then increasing in Eastern Europe with deeper emission reductions.

A tipping point for the Nr controls can be identified where the PM2.5 response from NH3 emission reductions outweighs that from NOx emission reductions, i.e., by interpolating βNH3−βNOx or G−1 to reach zero among a series of NH3 and NOx abatement sensitivity tests (Methods). We find that the βNH3−βNOx tends to be positive as Nr emissions are reduced, and it has larger changes in the places with more excessive NH3 (Supplementary Fig. 10). Figure 4 shows the tipping point of Nr emission reductions for instant efficiencies are 36% ± 16% and 18% ± 22% in Western Europe and Eastern Europe, respectively. It indicates small mitigation for NH3 or NOx can decrease the same PM2.5 concentrations in Western Europe after around 36% emission reductions in 2015. However, the G ratios are still above 1 under these abatement scenarios, and the tipping point for G ratio = 1 needs a deeper NH3 abatement (73% ± 16% in Western Europe and 46% ± 24% in Eastern Europe). The discrepancies between metrics of the chemical regime for SIAs formation and the effectiveness for PM2.5 decreases are also found in Thunis et al.35.

Fig. 4: Tipping point of reactive nitrogen (Nr) emission control effectiveness targeting PM2.5 abatement.
figure 4

a, c Tipping point for surface PM2.5 response from NH3 emission reductions outweighs that from NOx emission reductions identified by βNH3 = βNOx (a) and the NH3 saturation ratio G = 1 (G, the ratio between free ammonia (NH3 and NH4+) and total nitrate (HNO3 + NO3−) after neutralization of H2SO4) (c). b, d Vertical profiles of Western Europe (red) and Eastern Europe (blue) mean tipping point for βNH3 = βNOx (b) and G = 1 (d). Gray areas in (a), and (c) represent that 100% Nr emission controls here cannot achieve the tipping point. Shading in (b) and (d) represent values (means ± one spatial standard deviation) of tipping point.

Full size image

We find the first explanation for the discrepancy between the tipping point from β efficiency and that from the G ratio would be the different unit of Nr abatement, the former indicating the same PM2.5 decreases with per 1% mass reduction of Nr emissions and the latter indicating per unit mole reduction. This discrepancy is reduced by 36–41% when we transfer the unit of β from mass-based to molar-based. The residual discrepancy can be explained by their different definitions and applicable targets. The G ratio is based on a homogeneous air parcel at any specific moment and it loses extreme values when applied to the regional air quality model. As shown in Fig. 4, the tipping point for the G ratio falls rapidly with height while β changes steadily. Therefore, the decrease of surface PM2.5 depends on the chemical regime of each specific grid cell and time interval. This demonstrates that the instant efficiency is more suitable for evaluating Nr emission controls for PM2.5 mitigation while the chemical regime only provides a rough direction.

The optimal pathway for Nr abatement in Europe

Here, we develop and apply a diagnostic diagram for the effectiveness of PM2.5 abatement to find the optimal pathway of Nr emission controls in Europe (Methods). It visualizes the regional mean PM2.5 reductions as isopleths and the combined instant efficiency for PM2.5 abatement (the gradient) as arrows. As shown in Fig. 5, the gradient in Western Europe for the 2015 base condition tends to shift towards NOx, which indicates that NOx emission controls would initially be most effective. By contrast, for Eastern Europe NH3 and NOx emission controls have the similar effects in the early stage. Following the direction of gradients, we find that the optimal pathway of Nr emission controls in Western Europe entails always stronger reductions in NOx than NH3 emissions so that the regional mean PM2.5 concentrations decline the fastest. This pathway approaches to reductions of ~100% NOx emissions and 40% NH3 emissions inducing PM2.5 decreases by 2.2–2.4 μg·m−3, and further NH3 emission reductions will lead to an additional 0.4 μg·m−3 decrease. In contrast, the optimal pathway of Nr emission controls in Eastern Europe shall have a deeper NH3 abatement where ~100% NH3 and 60% NOx emission reductions result in PM2.5 decreases by 1.7–1.9 μg·m-3.

Fig. 5: Diagnostic diagram to identify the optimal pathway of reactive nitrogen (Nr) emission controls towards effective PM2.5 abatement and minimal control costs.
figure 5

a, d the diagnostic diagram for effectiveness of regional annual mean PM2.5 abatement to find the optimal pathway (purple line) of Nr emission controls in Western Europe (a) and Eastern Europe (d). b, e Control costs for Nr emissions according to the NH3 (x-axis) and NOx (y-axis) emission changes from 0 to 50% in Western Europe (b) and Eastern Europe (e). c, f the diagnostic diagram for the ratio of control costs and PM2.5 abatement to find the optimal pathway (red line) of Nr emission controls in Western Europe (c) and Eastern Europe (f). Black, white circles, and black arrows in (a) and (d) show 13 sets of simulated regional mean PM2.5 concentrations, interpolated PM2.5 concentrations, and their gradients in the diagnostic diagram. Black circles in (b) and (e) show control costs from five sets of feasible scenarios in the GAINS model. Black circles and arrows in (c) and (f) show the ratio of control costs and PM2.5 abatement at each 10% control level and their gradients in the diagnostic diagram.

Full size image

In addition to PM2.5 abatement, cost considerations are also essential information for policy-making. Here we note that the optimal pathway of Nr emission controls changes uniformly in favor of NH3 emission reductions when we consider control costs. We quantify Nr emission abatement technologies and related costs (including investment costs, fixed and operating costs) according to the GAINS model36,37,38. Figure 5 shows the annual total costs for Nr abatement in Western Europe and Eastern Europe according to integration and interpolation among five sets of feasible emission control scenarios at the national level reported by Amann et al.39. Here, the costs refer to the extra annual costs incurred by the individual abatement measures to reduce Nr emissions. The abatement measures for NOx emissions stem mainly from the power and industrial sectors, while those for NH3 are mainly from agricultural livestock farming and fertilizer use. We find NH3 emission controls are always cheaper than NOx emission controls and Western Europe has higher costs to control Nr emissions than Eastern Europe due to the reduction policies already in place and higher levels of Nr emissions. Currently feasible emission abatement technologies in the GAINS model can reduce NOx and NH3 emissions by 16% and 29% in Western Europe, annually costing 3.7 and 0.8 billion euros, respectively; reduce NOx and NH3 emissions by 32% and 31% in Eastern Europe, annually costing 1.9 and 0.2 billion euros, respectively.

We further update the diagnostic diagram of cost-effectiveness using the ratio of control costs and PM2.5 abatement, which denotes the annual costs per unit PM2.5 decreases (Fig. 5). Due to the limited availability of cost data for high abatement levels, we merely extrapolate to 50% emission reductions. In Western Europe, controlling 10% (30%) NH3 emissions from the 2015 Base would decrease regional PM2.5 by 0.11 (0.40) μg m−3 and require implementation costs of 0.08 (1.1) billion euros. In comparison, controlling 10% (30%) NOx emissions there would decrease regional PM2.5 by 0.23 (0.67) μg m−3 with implementation costs of 1.1 (12.7) billion euros. Similar cost-effective NH3 emission controls can be seen in Eastern Europe. The optimal pathway of cost-effective Nr emission controls follows the lowest isopleths for the cost/PM2.5 abatement ratio, inferring NH3 emission reductions only in both Western Europe and Eastern Europe. The much lower cost and increasing efficiency of NH3 emission control diminish the need for NOx emission control.

Nr emission controls thus can help Europe towards achieving the updated WHO guideline value, reducing PM2.5 air pollution by 12–29% and PM2.5-related mortality by 6–29% in Europe in 2015. Eastern Europe (G ratios of 1–5) represents an area slightly in excess of NH3 where modest NOx or NH3 emission reductions abate similar PM2.5 concentrations. In contrast, Western Europe (G ratios of 2–7) represents a highly NH3-excessive area where PM2.5 abatement does not become effective until NH3 reduction reaches above 40%. When considering control costs, the optimal pathway for halving Nr emissions clearly points towards NH3 management. Policy challenges specific to NH3 abatement have been described recently by Gu et al. 40. Identifying the optimal pathway for Nr emission reductions combining the effectiveness of PM2.5 abatement and emission control costs may also be important in many other regions over the globe, such as China, India, and the United States, where nitrogen pollution has continued to grow in recent years41,42,43. These regions are also facing heavy loads of PM2.5 air pollution and challenges to meet the updated WHO guideline value7,44. In addition to the benefits for PM2.5 mitigation and human health, Nr emission controls can also help reduce nitrogen deposition and surface ozone (Supplementary Fig. 9), which should be integrally considered in environmental strategies in future work.

In closing, we note that some uncertainties are associated with our analyses. First, several emission inventories based on different sectoral categories and collected by different institutions provide a considerable range of Nr emission estimates in Europe. While the estimates we use are within the range (Supplementary Table 1), the net results will be affected by the choice of Nr input. Second, the availability of NH3 may significantly elevate the aerosol water content and alkalinity and then enhance the production of SIAs and SOA45,46. The contribution of NH3 emissions on PM2.5 air pollution would thus present a lower estimate only, as here, the alkalinity is limited to the heterogeneous production of SIAs, and the SOA formation is parameterised as a multiple of OC concentrations. Furthermore, our study has not included the bidirectional exchange of NH3, i.e., simultaneous fluxes from and deposition to agricultural areas, which is highly uncertain and may alter the effectiveness of PM2.5 mitigation47,48.

Methods

Observations of air pollutants and meteorology

Daily and hourly surface chemical measurements in Europe in 2015 are obtained from the air quality database of the European Environment Agency (EEA). We use 964 background stations for PM2.5, which are then spatially aggregated into 565 grid cells to get a more representative evaluation of the model results. In addition, we use 27 stations for BC, 30 stations for OC, 34 stations for SIAs (Supplementary Table 2), and 21 stations for NH3. Of these, 26 stations containing all PM2.5 components are used to improve model SOA. Meteorological observations at 2072 stations in Europe in 2015 are collected from the National Climatic Data Center (NCDC), which consists of hourly 10-m wind speed (WS10), 10-m wind direction (WD10), 2-m air temperature (T2), and 2-m relative humidity (RH2). Evaluations of baseline simulated meteorology with these observations generally show good agreement (Supplementary Table 3).

The WRF-Chem model

Model configuration

The Weather Research and Forecasting model coupled with Chemistry (WRF-Chem) version 4.0.3 is applied to simulate the meteorology and PM2.5 concentrations. The model domain includes most European countries and their surrounding regions using 150 (east-west) × 100 (south-north) grid cells at a 27-km spatial resolution. We divide the vertical atmosphere into 38 levels, with a first layer height of 10 meters above ground and a top pressure of 5000 Pa. The initial and lateral meteorological boundary conditions are based on hourly datasets from the European Centre for Medium-Range Weather Forecasts (ECMWF) Integrated Forecasting System (IFS) with a spatial resolution of 0.25°49. We nudge every 2-day meteorological fields with ERA5 reanalysis data to keep actual atmospheric conditions. The chemical initial and boundary conditions are driven by the CAM-Chem model output at 0.9° × 1.25° horizontal resolution50. Chemical processes are assessed using the gas-phase Carbon-Bond Mechanism Z mechanism (CBMZ)51 and the four-bin sectional Model for Simulating Aerosol Interactions and Chemistry (MOSAIC) aerosol scheme with dry diameters of 0.039–0.156, 0.156–0.625, 0.625–2.5, and 2.5–10.0 µm52. The SIAs formation is described in the CBMZ-MOSAIC through precursor gas oxidation (the gas-phase oxidation of SO2/NOx, the aqueous-phase oxidation of SO2/NOx in clouds, and the hydrolysis of dinitrogen pentoxide) to form H2SO4/HNO3 and subsequent neutralization/condensation by/with NH3. The thermodynamics and phase equilibrium of SIAs are simulated by the Multicomponent Taylor Expansion Method (MTEM) and a computationally efficient Multicomponent Equilibrium Solver for Aerosols (MESA) in the thermodynamic module of MOSAIC. The gas-particle equilibrium of semi-volatile components (e.g. ammonium nitrate) is determined by the Adaptive Step Time-Split Euler Method in the gas-particle partitioning module of MOSAIC52. We add the heterogeneous sulfate formation reactions on particle surfaces based on Chen et al.53 to improve SIAs simulation. The Rapid Radiative Transfer Model for GCMs (RRTGM) scheme is used to parameterize shortwave and longwave radiation transfer54. Other physical parameterizations are the same as those used by Liu et al.26.

Model improvement

Our model will underestimate PM2.5 concentrations as the chemical mechanisms do not consider online secondary organic aerosol (SOA) formation due to its high uncertainty. Here we use a multiple of the OC concentrations to make up the SOA component by comparing simulated PM2.5 components to the observations (Supplementary Fig. 2). SOA concentrations are characterized as three times that of OC in summer and two times in other months55. This SOA assumption has little effect on our results because SIA and SOA chemistry are decoupled in the model. In addition, to account for OC underestimates, we increase the OC concentrations in Eastern and Central European countries by two and five times, respectively.

Model emissions

Anthropogenic emissions in 2015 use the monthly estimates from the Evaluating the Climate and Air Quality Impacts of Short-Lived Pollutants (ECLIPSE) Project at 0.1° × 0.1° spatial resolution for Europe, deriving from the GAINS (Greenhouse gas and Air pollution Interactions and Synergies) model and the Emissions Database for Global Atmospheric Research (EDGARv5.0) inventory at 0.1° × 0.1° spatial resolution for regions outside Europe (https://edgar.jrc.ec.europa.eu/dataset_ap50). We further use sector-specific diurnal weighting profiles for the anthropogenic emissions from power, industry, residential, transportation, and agriculture sectors as model hourly emission inputs (Supplementary Table 4). ECLIPSE estimates of anthropogenic NO2 and NH3 emissions over Europe in 2015 are 3.7 Tg N and 4.4 Tg N, respectively, which are consistent with current anthropogenic emission inventories (Supplementary Table 1). NOx emissions have high values in the Netherlands, Belgium, and other European urban areas. While the high values of NH3 emissions are in the Netherlands, northern Germany, western France, and northern Italy (Supplementary Fig. 1). Biomass burning emissions adopt the Fire Inventory from the NCAR56. Biogenic emissions are estimated online using the Model of Emissions of Gases and Aerosols from Nature (MEGAN)57, except for the soil NOx emissions that are from the GEOS-Chem model (http://geoschemdata.wustl.edu/ExtData/HEMCO/OFFLINE_SOILNOX/).

Model sensitivity simulations

We conduct a Base simulation in 2015 and a series of sensitivity simulations to examine the impacts of Nr emission reductions on PM2.5 air quality in Europe. First, the baseline simulation (denoted as Base) incorporates the 2015 emissions described above that have been assessed using observations. Second, a group of sensitivity simulations (denoted as S1RN, N = 30, 60, 80, and 100) reduces anthropogenic Nr emissions (both NOx and NH3 emissions) over Europe by 30%, 60%, 80%, and 100%, respectively. The differences in PM2.5 concentrations between Base and S1RN can estimate the effects of Nr emission reductions. Third, a group of sensitivity simulations (denoted as S2RN, N = 30, 60, 80, and 100), similar to S1RN, but only reduces NOx emissions. Fourth, another group of sensitivity simulations (denoted as S3RN, N = 30, 60, 80, and 100), similar to S1RN, but only reduces NH3 emissions. The comparison of PM2.5 changes between S2RN and S3RN then quantifies the respective effectiveness of NOx and NH3 abatement. For all simulations, typical months for the four seasons (January, April, July, and October) after a 3-day spin-up for initialization are simulated to represent yearly results due to limited computing resources.

Health-impact of Nr emission on PM2.5

We assess the PM2.5-related chronic health impacts through the Global Exposure Mortality Model (GEMM)58. It develops a PM2.5-mortality hazard ratio function according to cohort studies of worldwide outdoor air pollution, and has been widely used in recent studies59,60. This concentration-response function-based method focuses on total PM2.5 mass without assessing individual PM2.5 components for which evidence is limited9, corresponding with the GBD study1. The total health burden of long-term PM2.5 exposure is attributed to noncommunicable diseases and lower respiratory infections (NCD and LRI). PM2.5-related premature deaths (ΔMort) for adults (≥25 years) with age groups in 5-year intervals from 25 to greater than 85 are calculated by the following formulas:

$$R{R}_{i}(c)=\exp (\theta \times \,\log (\frac{z}{\alpha+1})/(1+\exp (-\frac{z - \mu }{\upsilon }))),\, z=\,{\max }(0,\, c - 2.4)$$
(1)
$$\varDelta Mort=\mathop{\sum }\limits_{i=1}^{12}Bas{e}_{i}\times Po{p}_{i}\times \frac{1}{R{R}_{i}}$$
(2)

where RRi is the relative risk of NCD or LRI for age group i (i = 1, 2,…,12), which means the contribution of PM2.5 pollution to the baseline mortality rate; c is ambient annual PM2.5 concentrations, exp is the natural exponential function, and θ, α, μ and υ are parameters that determine the shape of relative risk in GEMM and are specified for each age group. The PM2.5 threshold is 2.4 μg·m-3, below which no impact occurs. Basei is the baseline mortality rate of NCD or LRI for age group i, obtained from the Global Burden of Disease Study of 2015, and Popi is the gridded population within age group i, derived from the Gridded Population of the World version 4.11 (GPWv4) dataset. We further use the Monte Carlo method to provide the 95% confidence interval (CI) of deaths through ten thousand-time estimates.

Indicators of Nr control for PM2.5 mitigation

Here, we apply a mass-based indicator of instant efficiency to quantify the effectiveness of Nr controls in reducing total PM2.5 concentrations, and then use a molar-based indicator of G ratio to explain changes in the effectiveness of PM2.5 mitigation. We calculate the efficiency of Nr emission controls based on the sensitivity simulations following the definition in Liu’s (2021)26. The instant efficiency is defined as:

$${{{\upbeta }}}_{{LN}}=\frac{\varDelta {[{{{{{{\rm{PM}}}}}}}_{2.5}]}_{{LN}} - \varDelta {[{{{{{{\rm{PM}}}}}}}_{2.5}]}_{{L}({N}{+}{1})}}{{[{{{{{{\rm{PM}}}}}}}_{2.5}]}_{base}}/\frac{\varDelta {[{Emi}]}_{{LN}}-\varDelta {[{Emi}]}_{{L}({N}{+}{1})}}{{[{Emi}]}_{base}},\, {N}=1$$
(3)
$${{{\upbeta }}}_{{LN}}=\frac{\varDelta {[{{{{{{\rm{PM}}}}}}}_{2.5}]}_{{L}({N}{-}{1})} - \varDelta {[{{{{{{\rm{PM}}}}}}}_{2.5}]}_{L({N}{+}{1})}}{{[{{{{{{\rm{PM}}}}}}}_{2.5}]}_{base}}/\frac{\varDelta {[{Emi}]}_{{L}({N}{-}{1})}-\varDelta {[{Emi}]}_{{L}({N}{+}{1})}}{{[{Emi}]}_{base}},\, {N}=2,\,3,\,\ldots,\,10$$
(4)
$${{{\upbeta }}}_{{LN}}=\frac{\varDelta {[{{{{{{\rm{PM}}}}}}}_{2.5}]}_{{L}({N}{-}{1})}-\varDelta {[{{{{{{\rm{PM}}}}}}}_{2.5}]}_{LN}}{{[{{{{{{\rm{PM}}}}}}}_{2.5}]}_{base}}/\frac{\varDelta {[{Emi}]}_{{L}({N}{-}{1})}-\varDelta {[{Emi}]}_{{LN}}}{{[{Emi}]}_{base}},\,{N}=11$$
(5)

where [Emi]base is baseline Nr emissions (in unit of Tg N), [PM2.5]base is baseline simulated total PM2.5 mass concentrations (in unit of μg·m−3), Δ[Emi]LN is the mass of Nr emissions reductions for level LN (L1 = 0%, L2 = 10%, L3 = 20%, …, L10 = 90%, and L11 = 100%) relative to the baseline (in unit of Tg N), and Δ[PM2.5]LN is associated total PM2.5 mass decreases (in unit of μg·m-3), denoting the instant response of total PM2.5 mass in percentage to 1% mass reduction in Nr emissions under each Nr emission scenario. PM2.5 concentrations on each 10% Nr (NOx or NH3) emission scenario are calculated through a shape-preserving piecewise cubic spline interpolation among the 13 sets of simulated PM2.5 concentrations. We define the tipping point of NH3 emission reductions as βNH3−βNOx = 0.

G ratio is applied to diagnose the availability of NH3 in the air and to analyze the chemical regime of SIAs formation25. We calculate the G ratio using the equations below.

$$G=\frac{[{{{{{{\rm{NH}}}}}}}_{3}]+[{{{{{\rm{NH}}}}}}_{4}^{+}]-2\times [{{{{{{\rm{SO}}}}}}}_{4}^{2-}]}{[{{{{{{\rm{HNO}}}}}}}_{3}]+[{{{{{\rm{NO}}}}}}_{3}^{-}]}$$
(6)

where [NH4+], [NH3], [H2SO4],[HNO3], and [NO3−] are the molar concentrations (in unit of µmol·m−3). Values of the G ratio below 1 mean that ammonia is insufficient to neutralize all H2SO4 and total nitrate, indicating an NH3-limited chemical regime where SIAs formation is limited by the availability of NH335. In contrast, values of the G ratio above 1 mean that ammonia is sufficient to neutralize total nitrate, which characterizes a HNO3-limited chemical regime where SIAs formation is limited by the availability of HNO3. The G ratio of 1 means that changes (in mol units) in NH3 and NOx emissions can result in similar PM2.5 changes. We note, that such diagnoses concerning SIAs formation to NH3 and to HNO3 availability are invalid for high temperature and low relative humidity as little ammonium nitrate aerosol is present25.

Diagnostic diagram for Nr control pathways

The diagnostic diagram is developed for evaluating Nr control pathways targeting PM2.5 abatement (Fig. 5). We first calculate PM2.5 concentrations on each 10% Nr (NOx or NH3) emission scenario (the white circle symbols in the diagnostic diagram) through a shape-preserving piecewise cubic spline interpolation among the 13 sets of simulated regional mean PM2.5 concentrations (the black circle symbols in the diagnostic diagram) to obtain PM2.5 isopleths. The isopleth gradients (the blue arrows in the diagnostic diagram) are then calculated as Eq. 7 to represent the effectiveness of PM2.5 reductions.

$$\nabla [{{{{{{\rm{PM}}}}}}}_{2.5}]=\frac{\partial [{{{{{{\rm{PM}}}}}}}_{2.5}]}{\partial x}\times \vec{{{{{{\bf{i}}}}}}}+\frac{\partial [{{{{{{\rm{PM}}}}}}}_{2.5}]}{\partial y}\times \vec{{{{{{\bf{j}}}}}}}={{{\upbeta }}}_{{{{{{\rm{NH3}}}}}}}\times \vec{{{{{{\bf{i}}}}}}}+{{{\upbeta }}}_{{{{{{\rm{NOx}}}}}}}\times \vec{{{{{{\bf{j}}}}}}}$$
(7)

Here x and y axes represent the strengths of the NH3 and NOx emission controls, respectively. i and j are unit vectors in the x-direction and y-direction, respectively. ∂PM2.5/∂x is the gradient in the x-direction showing instant efficiency from NH3 controls. ∂PM2.5/∂y is the gradient in the y-direction showing instant efficiency from NOx controls. The effectiveness diagnostic diagram can illustrate the total PM2.5 declines by isopleths, the combined instant efficiency on PM2.5 abatement by the size of arrows, and the relative efficiency between NH3 and NOx by the direction of arrows. Therefore, the optimal pathway for effective PM2.5 declines can be found by following the isopleth gradients.