Ozone (O3) has well-documented detrimental impacts on human and ecosystem health, and in the mid to upper free troposphere is a greenhouse gas1,2. Since the mid-twentieth century, it has been known that O3 is formed through complex radical chain-reaction chemistry involving the oxidation of volatile organic compounds (VOCs) in the presence of oxides of nitrogen (NOx) (ref. 3). The chain reaction is initiated through the photolysis of photo-labile compounds such as O3, CH2O and HONO, propagated through the interconversion of peroxyl (HO2 and RO2) and hydroxyl (OH) radicals and terminated either through the self-reaction of peroxyl radicals to form longer-lived species (such as peroxides) or from the reaction of a hydroxyl radical with nitrogen dioxide (NO2) to form nitric acid (HNO3)4,5. From the perspective of reducing O3 pollution, the dominant termination step is often used to identify whether reductions in the emissions of NOx or VOCs should be preferentially employed6. If peroxyl-radical self-reactions dominate (known as the ‘NOx limited’ regime), reductions in NOx emissions are most beneficial, whereas if the NO2 + OH reaction dominates (known as the ‘VOC limited’ regime) reductions in organic emissions are most effective. The existence of these two well-defined O3 production regimes has been used to guide national-, regional- and city-level air-quality policy choices for many decades3.

Since the development of the NOx- versus VOC-limited model, understanding of atmospheric chemistry has advanced, with increased understanding of VOC chemistry7, aerosols8, halogens9 and so on. In this study, we use a chemistry transport model with an advanced representation of atmospheric chemistry (GEOS-Chem10) to diagnose the balance of termination steps. Notably, the model includes the uptake of HO2 onto aerosol surfaces, which has been identified as an important process for chain termination over the past decades8,11,12,13,14,15. We compare the rate of radical termination by peroxyl-radical self-reactions, the OH + NO2 reaction and the uptake of HO2 onto aerosol in July for the Northern Hemisphere in the pre-industrial era (1750), 1970 and 2014 (Methods). Concentrations of key components calculated by the model are shown in Supplementary Figs. 1–6.

Chain termination

Figure 1 shows the surface distribution of the relative chemical fluxes of these termination steps for July 1750, 1970 and 2014 (enlargements over North America, Europe and Asia are in Supplementary Figs. 810).

Fig. 1: Map of the fractional contribution of each radical loss pathway for the years 1750, 1950 and 2014.
figure 1

Mean simulated July solar noon-time fraction of radical termination at the surface that occurs through OH + NO2 (red), peroxyl-radical self-reactions (green) and aerosol uptake of HO2 (blue) for 1750, 1970 and 2014. Enlargements of North America, Europe and Asia are shown in the Supplementary Information.

In the pre-industrial era (1750), peroxyl-radical self-reactions (green) dominate. Biomass burning (forest fires) in Russia and Canada show termination by the uptake of HO2 onto aerosol (blue), whereas fires over Kazakhstan show termination by OH + NO2 (red) (Extended Data Fig. 1 shows a simulation with biomass burning switched off). This reflects the differing balance of NOx relative to particulates emitted by different fires16. Desert dust makes only a limited contribution to the aerosol surface area, leading to some increased contribution of HO2 uptake over Mauritania, Oman and Iran, but overall the impact is minor (Extended Data Fig. 1 shows a simulation with no desert dust emissions). Some HO2 uptake by sea salt occurs over the northern Atlantic/Pacific and over the Arabian sea (Extended Data Fig. 1 shows a simulation with no sea-salt emissions).

In 1970, industrial and transport emissions over North America, Europe and Japan17 have a profound impact on the radical termination processes. Many of the cities in North America, Europe and Japan are red, indicating NO2 + OH as the dominant termination step. However, over Europe, large regions of the North Sea and eastern Germany are magenta (balance between OH + NO2 and aerosol uptake), with much of eastern Europe being cyan (balance between peroxyl-radical self-reaction and aerosol uptake). Over North America, West Coast cities (Los Angeles, San Francisco and Seattle) show NO2 + OH domination, while cities around the Great Lakes and those on the East Coast show large fractions of the radical termination being due to HO2 uptake (magenta, cyan), reflecting the different industrial and transport sources in these regions. Most of Japan is red or yellow, showing limited impact of HO2 uptake. In 1970, China and India remain dominated by peroxyl-radical self-reaction, but there is some increased uptake of HO2 by aerosols over populated regions relative to the pre-industrial era. Extensive areas of the oceans are cyan, indicating large HO2 uptake. Some of this is driven by sea salt (Extended Data Fig. 1), but some is driven by shipping emissions (see Extended Data Fig. 2 without shipping emissions) and some by outflow of aerosol from industrial regions.

By 2014, the balance has changed as North America, Europe and Japan implemented policies to substantially reduce emissions of NOx, VOCs, primary aerosols and aerosol precursors17. The relative importance of both OH + NO2 and HO2 uptake onto aerosols for termination has reduced, reflecting these emissions reductions. Over Asia, however, emissions have increased due to population growth and industrialization, leading to increases in NOx and PM concentrations. Large regions are now coloured magenta (balanced HO2 uptake and OH + NO2) or blue (dominated by HO2 uptake).

This behaviour is seen predominantly within the boundary layer (Extended Data Fig. 3). In regions with more active mixing between the surface and the mid-troposphere, uptake onto aerosols can remain regionally important at 800 hPa, and NO2 + OH can be important over cities. In general, the importance of these regimes decreases through the atmosphere so that at 500 hPa, the atmosphere is dominated by peroxyl-radical self-reactions. At higher altitudes, lightning NOx can lead to an increase in the importance of NO2 + OH as a termination process.

Photochemical regimes

Figure 2 splits the domain based on the local largest termination step (enlarged versions for North America, Europe and Asia are in Supplementary Figs. 1113). At a global scale, most locations are in an NOx-limited regime (green), with some cities in a VOC-limited (red) regime. However, for large regions, the uptake of HO2 onto aerosol (blue) is the largest termination reaction, and thus they do not fall into either of the two classically applied policy-control regimes. We call this third regime ‘aerosol inhibited’ as the presence of aerosols inhibits the formation of O3 on short photochemical timescales. Area- and population-weighted fractions over differing regions for the three regimes are given in Supplementary Tables 1 and 2.

Fig. 2: Map of the photochemical ozone-control regime for the years 1750, 1950 and 2014.
figure 2

Mean simulated July solar noon-time photochemical ozone-control regime for 1750, 1970 and 2014. Enlargements of North America, Europe and Asia are shown in the Supplementary Information.

As with the radical loss plot (Fig. 1), in 1750 the world was dominated by radical self-reactions, and so, except for forest-fire locations, the world was entirely NOx limited. In the 1970s, the large North American cities, northern Europe and southern Japan had become VOC limited. Although not recognized at the time, aerosol-inhibited regimes existed in regions downwind of these VOC-limited environments (eastern seaboard of the United States, Poland, downwind of Japan) and in some remote locations impacted by desert dust and shipping. In 1970, around 30% of the populations of Europe and North America and 71% of the population of Japan lived in VOC-limited regions, but only a small portion (<10%) lived in an aerosol-inhibited regime. Negligible fractions lived in that regime in the rest of the Northern Hemisphere.

By 2014, the success of air-quality strategies meant that aerosol-inhibited regimes over Europe and North America had mostly disappeared, leaving behind some relatively polluted city regions as VOC limited, but with most people living in NOx-limited regions (80% in Europe and North America, 66% in Japan). However, industrialization in Asia moved many cities to be VOC limited (19% in China and 11% in India), but much larger populations in those regions live in aerosol-inhibited regions (36% and 39%, respectively), notably, in the Indo-Gangetic Plain and eastern China. The aerosol-inhibited regime plays a more important role in Asia in 2014 than it did in Europe and North America in 1970.

Understanding of the chemistry in this new third photochemical regime is still poor. The reactive uptake coefficient for HO2 on aerosol (γHO2) can vary by orders of magnitude depending on relative humidity, aerosol composition and the concentration of transition metals12,18,19,20,21. The work here assumes a uniform γHO2 of 0.2, consistent with previous modelling work13. Simulations run with a γHO2 of 0.1 still lead to regions in the aerosol-inhibited regime (Extended Data Fig. 4), but the fraction of the population living in this regime reduces (9% of China and 5% of India). A reduction in the γHO2 to 0.05 results in some areas still being within the regime, but these are now restricted to coastal Chinese regions. A γHO2 of less than 0.05 as suggested by some early studies20 would probably result in very few locations being in the aerosol-inhibited regime. Thus with our model, this new photochemical regime exists for γHO2 greater than 0.05 and is probably important for values greater than 0.1. Field observations of radicals in the aerosol-inhibited regions are rare. Observations and box modelling made in central Beijing22 show a VOC-limited regime consistent with the modelling here, with similar conclusions for central New Delhi23. Observations from a more remote site in the North China Plain24 suggest an NOx-limited regime, whereas the modelling here would suggest an aerosol-limited one for that location. Reducing the γHO2 to 0.1 would give a consistent response here. Another area little explored is the ability to simulate aerosol surface-area density in models (Supplementary Fig. 7). Mass of particulate matter less than 2.5 μm in diameter (PM2.5) is often evaluated in models since it is the primary health metric and more readily observable25, and sometimes aerosol number density is evaluated because of its important role in light scattering and cloud–aerosol interaction26, but systematic evaluations of modelled aerosol surface area are rare.

Impact on O3

The additional radical sink offered by the uptake of HO2 onto aerosol has a profound effect on regional O3 concentrations. Figure 3 shows the impact on O3 concentrations of switching off the uptake of HO2 onto aerosol (γHO2 = 0). In the pre-industrial 1750 simulation, the reactive uptake of HO2 onto aerosols has a small impact on O3 concentrations (<2%). The only places where increases in O3 can be seen is over the biomass-burning regions of Canada and Russia.

Fig. 3: Map of the percentage increase in ozone concentration when uptake of HO2 is switched off for 1750, 1970 and 2014.
figure 3

Percentage increase in monthly mean solar 15:00 lt O3 concentration for July when heterogeneous uptake of HO2 on aerosols is switched off for 1750, 1970 and 2014.

In 1970, industrialized regions of North America, Europe and Japan show larger impacts. Substantial increases in O3 concentrations are seen in northern Germany (28%), Tokyo (18%) and the American Great Lakes (20%). Los Angeles displays the largest impact, resulting in a 35% increase in O3 concentrations when HO2 uptake is switched off. The O3 changes over India and China in 1970 were less than 5%. Thus, for many parts of North America, Europe and Japan in the 1970s, particulate pollution enhanced the uptake of HO2 onto aerosols and reduced O3 pollution.

By 2014, O3 sensitivity to HO2 uptake had reduced in Europe, North America and Japan as policy interventions had reduced particulate-matter concentrations. Removing HO2 uptake in those simulations leads to less than a 5% increase in O3 concentrations. Over China, however, removing the uptake of HO2 onto aerosol increases O3 concentrations in the North China plain by between 20% and 30%. Even higher values (>35%) can be seen over the Bohai and Yellow seas. An additional hot spot can be seen inland over Chengdu. The Indo-Gangetic Plain has some of the highest observed sensitivity globally, with O3 increases of greater than 30%. Uptake of HO2 within the model thus has a large influence on surface O3 concentration, notably in present-day China and India.


Although we have identified regions that are aerosol inhibited from a fast photochemical perspective, aerosols can have an impact on O3 production more widely through a number of processes: other heterogeneous reactions27,28, changes to photolysis rates29 and changes to the boundary layer30. Thus, although the area might be aerosol inhibited from a radical perspective, the overall impact of a change of aerosol concentration may be more complicated.

Future work is necessary to better evaluate modelled aerosol surface area and to create a robust parameterization for γHO2. It would also be useful to have observational data in regions outside of large cities but subject to high particulate pollution as these are the areas influenced by the new regime. The existence, however, of a third aerosol-inhibited photochemical ozone regime, which currently covers large regions and populations in China and India, highlights the complexity of implementing policies to improve air quality. Strategies that focus on reducing particulate matter alone create a trade-off with increases in surface ozone, a phenomenon observed in China13,31. A simulation (Methods) over India, where the 2014 emissions of anthropogenic particulate precursors (SO2, black carbon, organic carbon and dust) and biomass burning are halved (representing an aggressive air-quality intervention) leads to reduced particle surface-area densities (Extended Data Fig. 5) and increased O3 concentrations (Extended Data Fig. 6). To counter this adverse outcome, policy development must account for the aerosol-inhibited regime and include contemporaneous reductions in other anthropogenic emissions. Figure 4 shows the fractional reductions in anthropogenic NOx emissions that are necessary to mitigate the additional O3 that would form from this aerosol emissions reduction. For a large part of the Indo-Gangetic plain, large (40%) parallel reductions in NOx emissions would be needed to stop O3 concentrations from increasing. Although the direct health gains of reducing particulate-matter emissions will probably outweigh the disbenefits on health from increasing O3 concentrations, there would probably be additional negative impacts on ecosystems, crop yield reductions and regional climate warming that would arise if aerosol-inhibition effects were not managed.

Fig. 4: Map of fractional NOx emissions decrease required to offset a 50% reduction in aerosol emissions.
figure 4

Fractional decrease in 2014 NOx emissions that is necessary to offset the increase in surface, monthly mean and solar 15:00 lt O3 caused by a 50% reduction in emissions of aerosol precursors (anthropogenic SO2, black carbon, organic carbon, dust and all biomass-burning emissions).



GEOS-Chem10 version 12.8 ( was run in the tropchem configuration, which consists of 536 gas-phase, 76 heterogeneous and 133 photolytic reactions. Within the stratosphere, a linearized chemistry was used32.

The same meteorology was used for each simulation, but anthropogenic emissions were varied on the basis of the Community Emissions Data System17 emissions for 1750, 1970 and 2014 (the last available year). For each emissions year, the model was run globally at 4° × 5° for two years with the first year considered as spin-up. Boundary conditions and initial conditions were generated from the second year of simulation. These were then used in a one-month (1 June to 31 June) 1.00° × 1.25° nested simulation over a domain of 10° to 60° N latitude and 170° W to 170° E longitude to provide initial conditions for a final 0.500° × 0.625° simulation (1 July to 31s July) run over the same domain. The concentrations of all species within the chemistry scheme were archived each model time step from the high-resolution simulation, and these were then used to calculate chemical fluxes.

The GEOS-FP (forward processing) meteorological fields from the NASA Global Modelling and Assimilation Office ( drove model transport. All simulations used 47 vertical levels extending from the surface to approximately 80 km in altitude.

Other emissions were kept the same in all simulations: the MEGAN33 inventory provided the biogenic emissions, GFED416 provided the biomass-burning emissions and DEAD34 provided the dust emissions.

Loss rates

Chain-reaction termination rates were calculated offline from the archived species concentrations, physical parameters (temperature, pressure and humidity), aerosol parameters (component surface areas and radii) and the model rate constants.

We do not count the production of radical products in the peroxyl-radical self-reaction as termination steps but do assume that non-radical products are long lived on the timescale of the calculation and so offer a chain termination step.

The heterogeneous radical loss rate of HO2 was calculated by iterating over the radius and surface area of each type of aerosol (organic carbon, black carbon, dust, sea salt, nitrate–sulfate–ammonium and secondary organic aerosol). These parameters were used in conjunction with temperature and air density to calculate the first-order HO2 loss rate.

We calculate the flux through each chemical termination route at local solar noon for each grid box for each day of July and sum. We divide the individual monthly sums for each termination process by the total of the three termination processes over the month to derive the fractions shown.

Following ref. 6, we define a grid box’s O3 control regime by the largest of the three termination route.

HO2 uptake coefficient

We have used a default HO2 reactive uptake coefficient (γHO2) of 0.2 (ref. 8). Laboratory studies using pure synthetic aerosol have found relatively low (γHO2 < 0.2) values of uptake coefficients18,35,36,37; however, studies on real-world aerosol have found greater coefficients ranging from 0.08 to 0.40 (refs. 21,38,39). It is likely that minor components of the aerosol such as transition metals may play an important role in this. Some evidence suggests copper appears to greatly enhance the uptake coefficients40,41,42, and this has been shown to be present in Asian aerosol43,44,45.

We run simulations where the γHO2 is reduced to 0.1, 0.05 and 0.00 and show results in Extended Data Fig. 4. We treat H2O as the sole product formed during the heterogeneous uptake of HO246, but our conclusions do not change if we assume production of H2O2.

The use of a single value of γHO2 is unlikely to be realistic. However, current parameterizations do not reproduce the observed temporal variability in γHO2 (ref. 21). Further study will be needed to provide an appropriate parameterization for γHO2. How the value of γHO2 varies with the changing concentration of NOx and radicals is likely to play an important role in determining the importance of HO2 uptake for controlling O3 concentrations.

Indian particulate emissions reductions

Over a more limited Indian domain (3° to 36° N latitude and 63° to 94° E longitude), we repeat the July 2014 simulation at 0.25° by 0.3125°. We then half the 2014 emissions of anthropogenic particulate precursors (SO2, black carbon, organic carbon and anthropogenic dust) and all biomass-burning emissions to simulate an aggressive local effort to reduce particulate-matter pollution. This reduces particulate surface-area densities substantially (Extended Data Fig. 5) and leads to increased O3 concentrations (Extended Data Fig. 6). With the reduced particulate-matter emissions, we then run five simulations reducing anthropogenic NOx emissions by 10%, 20%, 30%, 40% and 50%. For each surface model grid box, we fit a third-order polynomial through a graph of monthly mean 15:00 lt O3 concentration against the percentage NOx reduction and identify when the value of NOx emissions reduction gives the same mean O3 concentration as in the simulation without any reduction in particulate-matter emissions. This value of NOx reduction is then plotted for each grid box as in Fig. 4.