Volume 46, Issue 9 p. 4980-4990
Research Letter
Free Access

The Role of Clouds in the Tropospheric NOx Cycle: A New Modeling Approach for Cloud Chemistry and Its Global Implications

Christopher D. Holmes

Corresponding Author

Christopher D. Holmes

Department of Earth, Ocean, and Atmospheric Science, Florida State University, Tallahassee, FL, USA

Correspondence to: C. D. Holmes,

[email protected]

Search for more papers by this author
Timothy H. Bertram

Timothy H. Bertram

Department of Chemistry, University of Wisconsin, Madison, WI, USA

Search for more papers by this author
Kaitlyn L. Confer

Kaitlyn L. Confer

Department of Earth, Ocean, and Atmospheric Science, Florida State University, Tallahassee, FL, USA

Search for more papers by this author
Kelly A. Graham

Kelly A. Graham

Department of Earth, Ocean, and Atmospheric Science, Florida State University, Tallahassee, FL, USA

Search for more papers by this author
Allison C. Ronan

Allison C. Ronan

Department of Earth, Ocean, and Atmospheric Science, Florida State University, Tallahassee, FL, USA

Search for more papers by this author
Charles K. Wirks

Charles K. Wirks

Department of Earth, Ocean, and Atmospheric Science, Florida State University, Tallahassee, FL, USA

Search for more papers by this author
Viral Shah

Viral Shah

Harvard John A. Paulson School of Engineering and Applied Sciences, Harvard University, Cambridge, MA, USA

Search for more papers by this author
First published: 22 April 2019
Citations: 47


We present a new method for simulating heterogeneous (surface and multiphase) cloud chemistry in atmospheric models that do not spatially resolve clouds. The method accounts for cloud entrainment within the chemical rate expression, making it more accurate and stable than other approaches. Using this “entrainment-limited uptake,” we evaluate the role of clouds in the tropospheric NOx cycle. Past literature suggests that on large scales, losses of N2O5 and NO3 in clouds are much less important than losses on aerosols. We find, however, that cloud reactions provide 25% of tropospheric NOx loss in high latitudes and 5% of global loss. Homogeneous, gas phase hydrolysis of N2O5 is likely 2% or less of global NOx loss. Both clouds and aerosols have similar impacts on global tropospheric O3 and OH levels, around 2% each. Accounting for cloud uptake reduces the sensitivity of atmospheric chemistry to aerosol surface area and uptake coefficient since clouds and aerosols compete for the same NO3 and N2O5.

Key Points

  • Entrainment-limited uptake provides a new, realistic representation of cloud chemistry for atmospheric models that do not resolve clouds
  • Clouds reduce global tropospheric NOx, O3, and OH through uptake of NO3 and N2O5
  • Accounting for cloud chemistry reduces the sensitivity of tropospheric oxidants to aerosol heterogeneous chemistry

Plain Language Summary

Cloud water droplets and ice crystals enable some aqueous and surface chemical reactions that otherwise would not occur in the gaseous atmosphere. While clouds are widespread and familiar, methods for simulating their multiphase chemical effects in global atmospheric models have been inadequate. We present an efficient mathematical method to represent the combined effects of cloud chemistry and entrainment in large-scale atmospheric chemistry models that do not resolve individual clouds. By applying the approach to nitrogen oxides, we show that clouds have a previously unrecognized impact on tropospheric ozone, an air pollutant and greenhouse gas, and hydroxyl, a key atmospheric oxidant.

1 Introduction

Nitrogen oxides (NOx = NO + NO2) play a critical role in tropospheric chemistry, by catalyzing the chemical production of key oxidants—O3 and OH—during hydrocarbon oxidation. NOx is emitted into the atmosphere by combustion, lightning, and soil microbes and removed mainly as nitric acid (HNO3), which is efficiently scavenged in dry and wet deposition (Logan, 1983). Most HNO3 forms through the reaction of NO2 with OH; however, hydrolysis of N2O5 on aerosol and cloud surfaces is an important secondary pathway (Alexander et al., 2009; Dentener & Crutzen, 1993). While past literature suggests that tropospheric N2O5 loss on clouds are small on regional and global scales compared to loss on aerosol (Dentener & Crutzen, 1993; Jacob, 2000), this work revisits the role of clouds in light of our improved understanding of reactive chemistry on tropospheric aerosol.

N2O5, which is produced through reactions between NO2 and NO3, is a nighttime reservoir of NOx (Atkinson et al., 1986; Brown et al., 2003; Chang et al., 2011; Platt & Janssen, 1995). N2O5 decomposes at warm temperatures and under sunlight, but its main loss in the lower troposphere is hydrolysis in aerosol and cloud water (Brown et al., 2007; Heikes & Thompson, 1983). The hydrolysis product is HNO3, which typically remains particle bound, although gaseous ClNO2 is possible in particles with high salt content (Behnke et al., 1997; Bertram & Thornton, 2009; Finlayson-Pitts et al., 1989; Thornton & Abbatt, 2005; Thornton et al., 2010). Heterogeneous reaction rates, including surface and multiphase reactions, are controlled by mass transfer from the gas phase to liquid surfaces and details of surface or aqueous chemistry can be encapsulated in a reactive uptake coefficient, γ, that represents the reaction probability for a molecule impacting the cloud or aerosol surface (Jacob, 2000; Ravishankara, 1997; Sander, 1999; Schwartz, 1986). For cloud water, urn:x-wiley:00948276:media:grl58903:grl58903-math-0001 at 298 K and rises with colder temperatures (Bertram & Thornton, 2009; Burkholder et al., 2015; van Doren et al., 1990). For aerosols, the N2O5 uptake coefficient strongly depends on composition (Bertram & Thornton, 2009; Brown et al., 2006; McDuffie et al., 2018), but, with the exception of stratospheric sulfuric acid aerosol, the uptake coefficient is generally less than pure water (Ammann et al., 2013; Burkholder et al., 2015; Crowley, Ammann, et al., 2010). While clouds have greater urn:x-wiley:00948276:media:grl58903:grl58903-math-0002 and far greater surface area than tropospheric aerosol (around 1,000 times greater surface area; inferred from aerosol, Heald et al., 2014, and cloud, Pruppacher & Jaenicke, 1995, data), their influence on global N2O5 and tropospheric chemistry is limited by the small volume occupied by clouds and the intervals of hours to days between cloud contacts (Lelieveld & Crutzen, 1990, 1991).

In atmospheric models, N2O5 hydrolysis on aerosols lowers the global tropospheric NOx burden around 15% and lowers O3 and OH around 5% each, relative to models without this process (for urn:x-wiley:00948276:media:grl58903:grl58903-math-0003; Evans & Jacob, 2005; Macintyre & Evans, 2010). This aerosol uptake accounts for 8–40% of global HNO3 production (Alexander et al., 2009; Bauer, 2004; Evans & Jacob, 2005; Hauglustaine et al., 2014; Xu & Penner, 2012). Earlier studies suggested larger impacts but overestimated the uptake coefficient for tropospheric aerosol ( urn:x-wiley:00948276:media:grl58903:grl58903-math-0004 was assumed, based on laboratory data available at the time; Dentener & Crutzen, 1993; Liao et al., 2003; Tie et al., 2001). One of the earliest studies concluded that N2O5 hydrolysis in cloud water has little impact on tropospheric chemistry, except in the remote southern hemisphere, because in their model, ubiquitous aerosol already consumed most N2O5 (Dentener & Crutzen, 1993). As a result, some atmospheric chemistry models neglect N2O5 hydrolysis in clouds (Evans & Jacob, 2005; Folberth et al., 2006; Hauglustaine et al., 2014; Tie et al., 2001; Xu & Penner, 2012). However, since their urn:x-wiley:00948276:media:grl58903:grl58903-math-0005 value was too high in light of more recent laboratory and field studies, Dentener and Crutzen (1993) overestimated the role of aerosols relative to clouds in N2O5 hydrolysis and it appears that the importance of clouds in the global NOx budget has not been revisited since.

In this work, we add N2O5 hydrolysis on clouds to a global atmospheric chemistry model and quantify its impact on tropospheric oxidants. Our treatment of cloud uptake of N2O5 uses a new mathematical modeling approach, developed in section 2, for treating mass exchange between gas and cloud droplets in a partially cloudy model grid cell. We show that this new approach dramatically reduces errors in the chemical solver relative to existing methods used in global atmospheric chemistry models. Sections 3 and 4 describe the improved model and examine the broader implications for tropospheric chemistry.

2 Reactive Chemical Uptake in Partly Cloudy Conditions

The rate at which aqueous reactions consume gas, such as N2O5, within a cloud, fog, or other aerosol, is (Fuchs & Sutugin, 1971; Jacob, 2000; Sander, 1999; Schwartz, 1986)

Here ci is the gas concentration in interstitial air, A is the surface area density of cloud or aerosol, r is the droplet or aerosol radius, Dg is the diffusivity in air, v is the mean molecular speed, and γ is the reactive uptake coefficient. The surface area density of spherical cloud droplets is related to the liquid water content, L, by A = 3L/ρr, where ρ is the density of water. For typical cloud conditions ( urn:x-wiley:00948276:media:grl58903:grl58903-math-0007), the characteristic timescale for heterogeneous loss of N2O5 inside a cloud is urn:x-wiley:00948276:media:grl58903:grl58903-math-0008. This timescale is much shorter than the residence time of air in clouds, which is 15–120 min for many stratus and cumulus clouds (Feingold et al., 1998, 2013; Kogan, 2004; Lelieveld et al., 1989; Stevens et al., 1996) although it can be longer for cirrus (Podglajen et al., 2016). As a result, N2O5 and other gases with fast, irreversible, aqueous loss are depleted in clouds relative to surrounding clear air (Brown et al., 2016; Platt et al., 1981). The loss rate of N2O5 and similar gases therefore strongly depends on the rate at which these gases are entrained into the cloud from the surrounding air.

Global atmospheric chemistry models, and some regional models, do not resolve individual clouds or entrainment, so these processes must be parameterized to simulate cloud heterogeneous chemistry. Several approximations have been previously used in such models, but all have significant shortcomings. Early models appreciated the intermittent nature of clouds and alternated cloudy and cloud-free conditions in each grid cell in proportion to the time that air resides within cloud (Lelieveld & Crutzen, 1990, 1991; Liang & Jacob, 1997). However, this approach is complex to implement, computationally slow, and we are not aware of current 3-D models that use it. Instead, many chemical transport models apply equation 1 using a grid-average surface area density, fcA, where A is the surface area density within cloud and fc is the cloud fraction of the grid cell (Huijnen et al., 2014; Jacob et al., 2004; Parrella et al., 2012; Williams et al., 2009). In this “thin-cloud approximation,” the resulting characteristic loss time is (fcki)−1, which is 20–60 s for N2O5 and cloud fractions in the range 0.1 < fc < 0.5. This loss rate is much too fast (Figure 1; Müller, 2014) because it is equivalent to spreading cloud water throughout the grid cell or assuming that cloudy and cloud-free air are continuously well mixed, making all gas in the grid cell susceptible to cloud uptake. Another approach is to partition the grid cell into portions that are within cloud and subject to uptake and another portion that is not (Tost et al., 2006). If this “cloud partitioning method” is applied at the start of each model time step, it implicitly assumes that mixing between cloudy and clear air portions of the grid cell happens with the same timescale as the model time step. If the time step is changed, as commonly happens when changing model resolution, then the entrainment and gas-aqueous exchange rates are altered, which is unphysical (Figure 1). Furthermore, for fast reactions, the numerical error actually rises as the time step is reduced, which violates expectations for numerical convergence.

Details are in the caption following the image
Uptake of a reactive gas (γ = 0.03) in a partly cloudy environment as simulated with several methods (section 2). The approximate entrainment-limited uptake (equation 4) matches the exact solution (equations 2 and 3) within the accuracy of the red and black lines, while the commonly used thin cloud and cloud repartitioning methods predict much faster loss. Time steps for the cloud partitioning method are in parentheses. All methods use the same cloud and chemical conditions: fc = 0.2, τc = 1 hr, L = 0.3 g/m3r = 10 μm, Dg = 0.2 cm2/s, v = 250 m/s,γ = 0.03.
We derive a more realistic treatment of reactive chemical uptake in partly cloudy conditions for use in global and regional atmospheric models. In this approach, which we call “entrainment-limited uptake,” the gas concentration in the cloudy part of the grid cell, ci, is depleted by heterogeneous surface or multiphase reactions at rate kici, where ki is given by equation 1, and by detrainment to clear air at rate kcci, where τc = 1/kc is the residence time of air in cloud. The in-cloud concentration is also replenished by entrainment from clear air. As shown in the supporting information, the mass balance equations imply that gas is consumed from a partly cloudy grid cell at the rate
where c is the usual grid cell concentration, which averages over cloudy and cloud-free regions. The term x/(1+x) is the fraction of gas in the grid cell that is within cloud, and x is related to cloud properties and chemical rates by
The supporting information provides a detailed derivation of these equations. The loss rate thus depends on the ratio of reactive uptake versus detrainment rates (ki/kc) as well as the cloud fraction (fc). If meteorological and chemical data can provide fc, kc, and ki, then numerical models can easily calculate k, but the controls on heterogeneous reaction rates are more apparent from the approximate solution

Within a cloud or aerosol, equation 1 shows that uptake is limited by gas phase diffusion (Dg term) and reactive uptake (γ term) acting in series. On scales larger than a single cloud, equation 4 shows that entrainment (kc term) also limits uptake in series with the two in-cloud processes, so k ≤ ki. Under completely cloudy conditions, the entrainment term vanishes and equation 4 reduces to equation 1 (i.e., k = ki when fc = 1), meaning that our entrainment-limited uptake is a generalization of the Schwartz (1986) equation for partly cloudy conditions.

Scale analysis of equation 4 highlights the importance of entrainment for limiting the loss of N2O5, and other soluble compounds, in clouds. While the timescale for N2O5 loss inside a cloud is urn:x-wiley:00948276:media:grl58903:grl58903-math-0012, as we showed above, the characteristic time for loss in a partly cloudy region (fc = 0.1 − 0.5,τc = 3,600 s) is k−1 = 1–10 hr. For N2O5, the characteristic time is effectively determined by entrainment alone since (fkc)−1 is 2–3 orders of magnitude larger than (fcki)−1. However, the in-cloud chemistry influences the loss rate for high cloud fractions (fc ≳ 0.95), low uptake coefficients (γ ≲ 10−3), or very low surface area density (A 0.001 cm2/cm3). In the limit of ki/kc ≪ 1, meaning that in-cloud reactions are much slower than entrainment mixing, then k ≈ fcki, which is the thin-cloud approximation. Thus, the thin-cloud approximation is reasonable for reactions with an in-cloud characteristic time, urn:x-wiley:00948276:media:grl58903:grl58903-math-0013, longer than about 10 hr (corresponding to γ ≲ 10−6).

Figure 1 illustrates the loss of N2O5 simulated with entrainment-limited uptake in comparison to the thin-cloud and cloud partitioning methods, all using the same assumptions about cloud properties. In the thin-cloud approximation, all N2O5 is lost within minutes. The cloud partitioning method simulates slower N2O5 loss, but loss rate is inversely proportional to the time step of the model, growing faster as the step size decreases. Only the entrainment-limited approach simulates slow, steady loss over hours, and the results are practically indistinguishable between the exact (equations 2 and 3) and approximate (equation 4) expressions. Because the entrainment-limited uptake is encapsulated in a first-order loss coefficient, it has very minimal computational cost while producing much greater realism compared to the other approaches.

3 Model Description

We assess the impact of clouds on N2O5 hydrolysis and tropospheric chemistry using the GEOS-Chem global chemical transport model (version 11-01, www.geos-chem.org). Simulations here use the tropospheric chemistry mechanism (Parrella et al., 2012). MERRA-2 (Modern-Era Retrospective Reanalysis for Research and Applications, Version 2) reanalysis meteorology (Gelaro et al., 2017) drives transport and provides cloud properties, which we resolve at 4° × 5° and 47 vertical layers. Prior work has found that the sensitivity of global chemical responses to perturbations are generally consistent across model resolutions (Holmes et al., 2013). Emissions follow the model defaults. For NOx, this includes the EDGAR global fossil fuel inventory (version 4.2; EC-JRC, 2011) and Yevich and Logan (2003) biofuel inventory, which are replaced by regional inventories where they are available (NEI2011v1 in United States: United States Environmental Protection Agency, 2015; Travis et al., 2016; BRAVO in Mexico: Kuhns et al., 2005; CAC in Canada: van Donkelaar et al., 2008; EMEP in Europe: European Monitoring and Evaluation Programme, 2014; and MIX in East Asia: Li et al., 2014). GFED4.1s provides biomass burning emissions (van der Werf et al., 2017). Soil and lightning NOx emissions respond to the model's meteorology (Hudman et al., 2012; Murray et al., 2012) and include a bug fix for lightning (Lee Murray, personal communication, 9 August 2018).

The model previously included heterogeneous NOx chemistry on aerosol surfaces (Evans & Jacob, 2005). We update the reactive uptake coefficients for consistency with recent literature (Table S1; Ammann et al., 2013; Atkinson et al., 2016; Bertram & Thornton, 2009; Bröske et al., 2003; Burkholder et al., 2015; Crowley, Ammann, et al., 2010; Escorcia et al., 2010; Fenter & Rossi, 1997; Ryder et al., 2015; Tan et al., 2016). In particular, NO3 uptake on organic aerosol is slower (Atkinson et al., 2016) and N2O5 reactions on urn:x-wiley:00948276:media:grl58903:grl58903-math-0014 aerosol now depend on the H2O and urn:x-wiley:00948276:media:grl58903:grl58903-math-0015 concentrations in aerosol (Bertram & Thornton, 2009; Shah et al., 2018). Past model versions assumed that the product of N2O5 uptake on all surfaces is entirely HNO3. This is reasonable for clouds, but laboratory and field studies have demonstrated that aerosol uptake can produce ClNO2, which escapes to the gas phase and recycles NOx. The yield of ClNO2 rises with the chloride content of the aerosol and reaches 1 for the concentrations found in sea salt aerosol (Behnke et al., 1997; Bertram & Thornton, 2009), so we adopt this yield for sea salt. For other aerosols in our simulations, the product remains HNO3 because GEOS-Chem does not track the chloride content of nonsea salt aerosol. The simulated aerosol burden and optical depth, which are indicators of aerosol surface area, are consistent with other recent global models (Myhre et al., 2013). As recommended by Atkinson et al. (2015), the model does not include homogeneous, gas phase hydrolysis of N2O5: N2O5(g)+H2O(g) → 2HNO3(g). Other models are sensitive to this reaction (Emmerson & Evans, 2009; Williams et al., 2009), however, so we assess its impact in a sensitivity simulation using an upper limit for the rate constant (<1 × 10−22 cm3 molecule–1 s–1; Atkinson et al., 2015).

We add uptake of N2O5, NO3, and NO2 on cloud water and cloud ice to the model using the entrainment-limited approach derived in section 2 (equations 2 and 3) and recent assessments of reactive uptake coefficients (Table S1; Ammann et al., 2013; Burkholder et al., 2015; Crowley, Ammann, et al., 2010). The surface area of liquid water clouds, which is needed for equation 1, is derived from the MERRA-2 liquid water content using A = 3L/ρr, where r is the effective radius and ρ = 1,000 kg/m3 is the density of liquid water. We assume r = 10 μm for marine clouds and r = 6 μm for continental clouds. For ice clouds, we use the empirical, temperature-dependent effective radius, r(T), reported by Heymsfield et al. (2014, their equation 9e). Bearing in mind that ice surface area is about 9 times larger than the cross-sectional area (Schmitt & Heymsfield, 2005), or 2.25 times the surface area of a sphere with the same effective radius, the ice surface area density is A = 6.75I/ρicer(T), where I is the MERRA-2 ice water content and ρice = 910 kg/m3 is the density of ice. When compared to satellite observations, MERRA-2 cloud properties reproduce the zonal mean and frequency distributions of optical depth and radiative effects within 30% (Bosilovich et al., 2015; Hongyu Liu, personal communication, 16 October 2018). Uncertainties in the cloud surface area and uptake coefficient have little effect on the results here because N2O5 and NO3 uptake are primarily limited by entrainment and NO2 uptake is very small. The residence time of air in clouds, which is used in the entrainment-limited uptake equations, varies by cloud type and weather conditions. MERRA-2, however, like many other meteorological reanalyses, provides only total cloud fraction, with no information about cloud types. In this work, we use τc= 1 hr for the cloud residence time based on mean values for stratus and stratocumulus clouds (Feingold et al., 1998, 2013; Kogan, 2004; Stevens et al., 1996), which comprise a large fraction of global cloud cover (Lelieveld et al., 1989; Pruppacher & Jaenicke, 1995). Future work is needed to prescribe spatial and temporal variation in the cloud residence time from global reanalysis data.

The effects of heterogeneous reactions on NOx, O3, and OH chemistry are assessed in four simulations. The control simulation includes uptake on aerosols and clouds, as described above, and is our best representation of this heterogeneous chemistry. Three sensitivity tests neglect uptake on clouds, aerosols, or both. An additional sensitivity test includes homogeneous hydrolysis of N2O5. All simulations begin with 6 months of spin-up, and then results for 2015 are analyzed.

4 Impact of Cloud Heterogeneous Chemistry

Figure 2 shows the contributions of reaction pathways to NOx loss by latitude in the model. Uptake of N2O5, NO3, and NO2 in clouds, the innovation in this work, is far from negligible, as suggested in past literature. Clouds provide 25% of NOx loss at high latitudes and 1%–5% at low latitudes. In the Northern Hemisphere, most of this loss occurs in liquid water clouds because NOx concentrations are highest at low altitudes due to surface emissions. In the Southern Hemisphere, NOx is more evenly distributed vertically, so ice clouds provide most of the uptake. Aerosol uptake provides up to 30% of simulated NOx loss in midlatitudes of the Northern Hemisphere, and homogeneous gas phase reactions provide the majority of NOx loss at all latitudes. When integrated over just the lowest 2 km, aerosol uptake in our model rises to 50% of NOx loss in northern midlatitudes, which is similar to results from field studies focused on this region (Aldener et al., 2006; Jaegle et al., 2018; Wagner et al., 2013).

Details are in the caption following the image
Chemical pathways for NOx conversion to HNO3 (a) in the zonal and annual mean, with black line showing HNO3 production (right vertical axis), (b) by month in the northern hemisphere extratropics (30°N–90°N), and (c) N2O5 loss frequency in the northern extratropics.

Seasonally, heterogeneous NOx loss on aerosols and liquid water clouds is greatest in winter, particularly in the northern extratropics (Figure 2b). This occurs for both aerosols and liquid water clouds. For loss on aerosols, the main driver of this seasonal cycle is the rise in N2O5 concentrations in winter due to its thermal stability at cold temperatures (Dentener & Crutzen, 1993; Platt et al., 1981). In our simulations, the mean N2O5 loss frequency on aerosols, which does not depend on N2O5 concentration, is nearly constant throughout the year in the northern extratropics (Figure 2c). Loss of N2O5 on liquid water clouds has a stronger seasonal cycle than loss on aerosols, meaning that the loss frequency is greater in winter than summer. This happens because cloud coverage increases in winter, particularly in the Northern Hemisphere (Stubenrauch et al., 2006). The uptake coefficient ( urn:x-wiley:00948276:media:grl58903:grl58903-math-0016) on liquid water also increases at cold temperatures (Burkholder et al., 2015), but this effect has little influence on the seasonal cycle because gas phase diffusion and cloud entrainment impose greater limits on uptake rate. The seasonal cycle of N2O5 uptake on ice clouds in the northern extratropics differs from uptake on liquid water clouds. In our simulations, ice clouds take up N2O5 more effectively in summer than winter (Figure 2c), despite the lesser extent of ice clouds in summer. The reason is that summertime convection transports NOx to high altitudes where ice clouds persist. This effect offsets the seasonal cycle of N2O5 concentrations so that the overall loss of N2O5 on ice clouds is nearly constant year-round (Figure 2b).

Table 1 quantifies the global NOx budget and the impact of neglecting heterogeneous reactions on tropospheric chemistry. Uptake on clouds is 5% of the global chemical NOx sink (0.2 Tmol N/year) and aerosols remove another 27% (1.1 Tmol N/year). Most heterogeneous loss on both clouds and aerosols is from N2O5 uptake and, secondarily, from NO3 uptake. NO2 uptake is small on aerosols and negligible on clouds because of the very small reactive uptake coefficients (Table S1). Homogeneous reactions, primarily NO+ OH → HNO3 and NO+ VOC → HNO+ products, provide the remaining 68% of chemical loss. In addition to the chemical sinks, dry deposition of NO2 and organic nitrates provides an additional sink for 0.3 Tmol N/year, which is similar to the cloud heterogeneous losses.

Table 1. Effect of Heterogeneous NOx Chemistry on Tropospheric Composition and NOx Loss
Simulation Control Cloud offa Aerosol offa Both offa
Aerosol reactions Yes Yes No No
Cloud reactions Yes No Yes No
NOx 42.9 Gmol N +0.7% +3.8% +8.9%
O3 6.9 Tmol +2.4% +2.5% +7.0%
OH 16.3 Mmol +2.7% +2.0% +6.0%
urn:x-wiley:00948276:media:grl58903:grl58903-math-0017 8.7 year –1.9% –3.4% –6.9%
NOx sinks, Tmol N/year
Total P(HNO3) 4.11 –0.8% –2.4% –5.7%
Homogeneousb 2.89 +3.2% +21.4% +33.0%
NO2 + aerosol 0.04 –0.6%
NO3 + aerosol 0.13 +2.9%
N2O5 + aerosol 0.80 +10.8%
NO2 + cloud 0.00 +24.1%
NO3 + cloud 0.02 +73.9%
N2O5 + cloud 0.19 +131%
Depositionc 0.29 +0.7% +27.1% +49.1%
  • a Percent changes are (Experimental − Control)/(Control) × 100%.
  • b NO+ OH → HNO3 and NO+ VOC → HNO+ products. Homogeneous hydrolysis of N2O5 could provide an additional 0.1 Tmol N/year.
  • c Dry and wet deposition of NO2, N2O5, and organic nitrates. Of these, dry deposition of NO2 and organic nitrates are each roughly half and others are much smaller.

If the cloud heterogeneous reactions are neglected, the tropospheric NOx burden in the model increases 0.7%, which results in a 2% increase in tropospheric O3 and a 2% decrease in the CH4 lifetime due to tropospheric OH. Neglecting heterogeneous aerosol reactions yields changes of a similar magnitude: 4% increase in NOx, 3% increase in O3, and 3% decrease in CH4 lifetime. Thus, the impacts of aerosol and cloud heterogeneous chemistry on O3 and CH4 lifetime is similar, despite the fivefold greater NOx loss on aerosols. The differences arise from the locations where NOx is lost. Studies of aviation, ship, and industrial emissions show that marginal changes in NOx abundance have the greatest effect on tropospheric O3 in low-NOx environments (Lin et al., 1988), particularly at low latitudes and high altitudes (Fry et al., 2012; Fuglestvedt et al., 2008; Holmes et al., 2014; Köhler et al., 2008, 2013). While aerosols remove NOx mainly from the lower troposphere over industrial regions with high NOx, clouds have a greater impact in remote, high-altitude, low-NOx environments, which gives clouds a disproportionate impact on global O3 and OH.

Cloud uptake modestly shifts the simulated HNO3 deposition, which is relevant for acid rain and nutrient loading (Figure S1). Although absolute changes in deposition are small, the overall effect of clouds is to increase HNO3 deposition in the middle and high latitudes by about 2%, particularly in the storm tracks, where clouds are most prevalent. There is a compensating 2% decrease of HNO3 deposition in the marine subtropical subsidence zones, due to the reduced HNO3 production from OH + NO2. Over the tropical continents, which have both clouds and high OH levels, the two effects compete, but the model suggests that adding the cloud uptake increases HNO3 deposition by 1% in these regions.

Our results suggest that tropospheric NOx, O3, and OH levels are much less sensitive to heterogeneous aerosol chemistry than previous model studies suggested. For example, Macintyre and Evans (2010) reported that omitting N2O5 uptake on aerosols raised tropospheric NOx by about 15% and O3 and OH by 5%, which is 2–3 times the effect that we report (Table 1 and above). Earlier studies reported even greater sensitivity, but subsequent studies found that their urn:x-wiley:00948276:media:grl58903:grl58903-math-0018 values on aerosol were too high (Dentener & Crutzen, 1993; Tie et al., 2001). Part of the explanation is that we treat the product of N2O5 uptake to sea salt aerosol as HNO3 + ClNO2, rather than 2HNO3, as in past models. That change reduces the NOx sink from aerosol heterogeneous reactions by 15% since sea salt accounts for one third of global N2O5 loss on aerosol in our model. However, the major reason for our lesser sensitivity is that clouds and aerosols compete for the same, limited supply of N2O5 and NO3, so any changes in aerosol uptake are largely offset, or buffered, by opposing changes in cloud uptake. When aerosol uptake of NOx is removed from the model, NOx uptake on clouds increases 60%. This competition explains why the results of neglecting both cloud and aerosol heterogeneous reactions simultaneously are much greater than the sum of neglecting them individually (Table 1).

Although the addition of heterogeneous NOx chemistry in clouds has significant impacts on tropospheric oxidant chemistry and budgets, discussed above, adding this chemistry has little meaningful effect on the model's comparison to most aircraft and sonde observations of NOx and O3. This is because the 1%–3% decreases in mean background NOx and O3 due to cloud chemistry are much smaller than the environmental variability or prior model biases for these gases (Jacob et al., 2003; Logan, 1999; Logan et al., 2012). Examples are shown in Figure S2. The mechanism could potentially be tested further with nighttime airborne measurements of N2O5 and NO3 lifetimes (e.g., Brown et al., 2006, 2009) averaged over large, partly cloudy, low-aerosol regions. The chemistry of N2O5 uptake on water and ice is sufficiently well understood, however, that the reaction should be included in atmospheric models regardless of its benefit or detriment to observational comparisons. Nevertheless, this shift should slightly diminish the surface O3 bias found in many models (Young et al., 2018) and push the model closer to the observationally constrained CH4 lifetime (11.2 ± 1.3 year versus 8.7 year simulated; Prather et al., 2012). Considering that doubling the spatial resolution to 2° × 2.5° increases the lifetime about 0.5 year (Holmes et al., 2013), a higher-resolution version of the present model using the same chemistry would likely be at the lower end of the uncertainty envelope.

Recent global model studies disagree on the relative importance of processes contributing to HNO3 and nitrate production. Most have neglected N2O5 and NO3 uptake in clouds (Alexander et al., 2009; Bauer, 2004; Hauglustaine et al., 2014; Xu & Penner, 2012), but one study suggested that this process provides 41% of HNO3 production (Williams et al., 2009). That model used the thin-cloud approximation, however, which significantly overestimates N2O5 uptake, as we showed in section 2, so our estimate of 5% is likely more realistic. Heterogeneous reactions on aerosols are reported to provide 8%–41% of global HNO3 production (Alexander et al., 2009; Bauer, 2004; Hauglustaine et al., 2014; Williams et al., 2009; Xu & Penner, 2012). Trends and variability in aerosol surface area will change the aerosol uptake rate, and these studies examined different years, but that is unlikely to account for the large range. However, the smallest estimate included reactions only on sulfate aerosol (Hauglustaine et al., 2014), and the largest assumed a very large uptake coefficient ( urn:x-wiley:00948276:media:grl58903:grl58903-math-0019; Xu & Penner, 2012). Neglecting those extremes, the literature suggests that aerosol uptake is 18%–35% of HNO3 production. Our result of 27% falls in the middle of this range. Homogeneous hydrolysis of N2O5 is neglected in many models, because its rate and importance remain unclear (Alecu & Marshall, 2014; Atkinson et al., 2015; Brown & Stutz, 2012; Brown et al., 2009; Crowley, Schuster, et al., 2010), but the reaction has a meaningful impact on tropospheric O3 when it is included in box models (Emmerson & Evans, 2009). Contrary to the box model results, tests in our global model using the upper limit rate from IUPAC (Atkinson et al., 2015) find that homogeneous hydrolysis supplies less than 2% of global HNO3 production (0.1 Tmol N/year) and changes mean tropospheric O3 and OH by under 0.5% each. Another global model also reported a similar upper limit on HNO3 production (<5%, 0.1 Tmol N/year; Williams et al., 2009), implying that the homogeneous hydrolysis rate for N2O5 is likely not a major source of uncertainty in tropospheric chemistry modeling.

5 Conclusions

While the global importance of heterogeneous NOx chemistry on aerosol surfaces has been widely recognized for several decades (Dentener & Crutzen, 1993; Heikes & Thompson, 1983; Jacob, 2000), heterogeneous reactions in clouds have been thought to be minor or negligible. Our results, however, show that heterogeneous NOx loss in clouds plays a significant role in the tropospheric NOx cycle and meaningfully impacts tropospheric oxidants. Cloud uptake provides up to 25% of tropospheric NOx loss at high latitudes and averages 5% globally. Homogeneous, gas phase hydrolysis of N2O5 is likely 2% or less of the tropospheric NOx sink. Although aerosols take up more NOx than clouds, neglecting either cloud or aerosol heterogeneous reactions in an atmospheric chemistry model has similar impacts on global tropospheric chemistry: raising the tropospheric O3 burden by about 2% and reducing the CH4 lifetime due to tropospheric OH by 2%. Moreover, accounting for NOx uptake on clouds nearly halves the sensitivity of tropospheric oxidants to heterogeneous aerosol reactions, because both clouds and aerosols compete for the same NO3 and N2O5.

The method of entrainment-limited uptake, which we developed here, has clear advantages over other approaches used to simulate cloud chemistry in atmospheric chemistry models. The method better represents cloud heterogeneous rates than the widely used thin cloud or cloud partitioning approaches while adding little or no computational burden to numerical chemical solvers. Future work to specify spatiotemporal variation in the residence time of air in clouds could further improve the method. The entrainment-limited equations are appropriate for any prognostic atmospheric chemistry model that does not resolve individual clouds, whether global or regional, and applicable to any reactive compound with irreversible surface or multiphase loss. Beyond NOx cycling, which we examined here, the method may advance global modeling and understanding of other cloud chemical reactions, including HO2 and formaldehyde uptake, photolysis, SO2 oxidation, mercury reduction, and aerosol processing.


This work was supported by the NASA New Investigator Program (NNX16AI57G). THB acknowledges support from the NOAA Climate Program Office's Atmospheric Chemistry, Carbon Cycle, and Climate program (NA18OAR4310109). Source code for the model used here is available at https://doi.org/10.5281/zenodo.2587245