Volume 123, Issue 22 p. 12,994-13,015
Research Article
Free Access

ClNO2 Yields From Aircraft Measurements During the 2015 WINTER Campaign and Critical Evaluation of the Current Parameterization

Erin E. McDuffie

Erin E. McDuffie

Chemical Sciences Division, Earth System Research Laboratory, NOAA, Boulder, CO, USA

Cooperative Institute for Research in Environmental Sciences, University of Colorado Boulder, Boulder, CO, USA

Department of Chemistry, University of Colorado Boulder, Boulder, CO, USA

Search for more papers by this author
Dorothy L. Fibiger

Dorothy L. Fibiger

Chemical Sciences Division, Earth System Research Laboratory, NOAA, Boulder, CO, USA

Cooperative Institute for Research in Environmental Sciences, University of Colorado Boulder, Boulder, CO, USA

Now at California Air Resources Board, Sacramento, CA, USA

Search for more papers by this author
William P. Dubé

William P. Dubé

Chemical Sciences Division, Earth System Research Laboratory, NOAA, Boulder, CO, USA

Cooperative Institute for Research in Environmental Sciences, University of Colorado Boulder, Boulder, CO, USA

Search for more papers by this author
Felipe Lopez Hilfiker

Felipe Lopez Hilfiker

Department of Atmospheric Sciences, University of Washington, Seattle, WA, USA

Now at TOFWERK, Thun, Switzerland

Search for more papers by this author
Ben H. Lee

Ben H. Lee

Department of Atmospheric Sciences, University of Washington, Seattle, WA, USA

Search for more papers by this author
Lyatt Jaeglé

Lyatt Jaeglé

Department of Atmospheric Sciences, University of Washington, Seattle, WA, USA

Search for more papers by this author
Hongyu Guo

Hongyu Guo

Earth and Atmospheric Sciences Department, Georgia Institute of Technology, Atlanta, GA, USA

Search for more papers by this author
Rodney J. Weber

Rodney J. Weber

Earth and Atmospheric Sciences Department, Georgia Institute of Technology, Atlanta, GA, USA

Search for more papers by this author
J. Michael Reeves

J. Michael Reeves

Earth Observing Laboratory, NCAR, Boulder, CO, USA

Search for more papers by this author
Andrew J. Weinheimer

Andrew J. Weinheimer

Atmospheric Chemistry Observations & Modeling Laboratory, NCAR, Boulder, CO, USA

Search for more papers by this author
Jason C. Schroder

Jason C. Schroder

Cooperative Institute for Research in Environmental Sciences, University of Colorado Boulder, Boulder, CO, USA

Department of Chemistry, University of Colorado Boulder, Boulder, CO, USA

Search for more papers by this author
Pedro Campuzano-Jost

Pedro Campuzano-Jost

Cooperative Institute for Research in Environmental Sciences, University of Colorado Boulder, Boulder, CO, USA

Department of Chemistry, University of Colorado Boulder, Boulder, CO, USA

Search for more papers by this author
Jose L. Jimenez

Jose L. Jimenez

Cooperative Institute for Research in Environmental Sciences, University of Colorado Boulder, Boulder, CO, USA

Department of Chemistry, University of Colorado Boulder, Boulder, CO, USA

Search for more papers by this author
Jack E. Dibb

Jack E. Dibb

Institute for the Study of Earth, Oceans, and Space, University of New Hampshire, Durham, NH, USA

Search for more papers by this author
Patrick Veres

Patrick Veres

Chemical Sciences Division, Earth System Research Laboratory, NOAA, Boulder, CO, USA

Cooperative Institute for Research in Environmental Sciences, University of Colorado Boulder, Boulder, CO, USA

Search for more papers by this author
Carlena Ebben

Carlena Ebben

Department of Chemistry, University of California Berkeley, Berkeley, CA, USA

Search for more papers by this author
Tamara L. Sparks

Tamara L. Sparks

Department of Chemistry, University of California Berkeley, Berkeley, CA, USA

Search for more papers by this author
Paul J. Wooldridge

Paul J. Wooldridge

Department of Chemistry, University of California Berkeley, Berkeley, CA, USA

Search for more papers by this author
Ronald C. Cohen

Ronald C. Cohen

Department of Chemistry, University of California Berkeley, Berkeley, CA, USA

Search for more papers by this author
Teresa Campos

Teresa Campos

Atmospheric Chemistry Observations & Modeling Laboratory, NCAR, Boulder, CO, USA

Search for more papers by this author
Samuel R. Hall

Samuel R. Hall

Earth and Atmospheric Sciences Department, Georgia Institute of Technology, Atlanta, GA, USA

Search for more papers by this author
Kirk Ullmann

Kirk Ullmann

Atmospheric Chemistry Observations & Modeling Laboratory, NCAR, Boulder, CO, USA

Search for more papers by this author
James M. Roberts

James M. Roberts

Chemical Sciences Division, Earth System Research Laboratory, NOAA, Boulder, CO, USA

Search for more papers by this author
Joel A. Thornton

Corresponding Author

Joel A. Thornton

Department of Atmospheric Sciences, University of Washington, Seattle, WA, USA

Correspondence to: S. S. Brown and J. A. Thornton,

[email protected];

[email protected]

Search for more papers by this author
Steven S. Brown

Corresponding Author

Steven S. Brown

Chemical Sciences Division, Earth System Research Laboratory, NOAA, Boulder, CO, USA

Department of Chemistry, University of Colorado Boulder, Boulder, CO, USA

Correspondence to: S. S. Brown and J. A. Thornton,

[email protected];

[email protected]

Search for more papers by this author
First published: 05 November 2018
Citations: 31

Abstract

Nitryl chloride (ClNO2) plays an important role in the budget and distribution of tropospheric oxidants, halogens, and reactive nitrogen species. ClNO2 is formed from the heterogeneous uptake and reaction of dinitrogen pentoxide (N2O5) on chloride-containing aerosol, with a production yield, ϕ(ClNO2), defined as the moles of ClNO2 produced relative to N2O5 lost. The ϕ(ClNO2) has been increasingly incorporated into 3-D chemical models where it is parameterized based on laboratory-derived kinetics and currently accepted aqueous-phase formation mechanism. This parameterization models ϕ(ClNO2) as a function of the aerosol chloride to water molar ratio. Box model simulations of night flights during the 2015 Wintertime INvestigation of Transport, Emissions, and Reactivity (WINTER) aircraft campaign derived 3,425 individual ϕ(ClNO2) values with a median of 0.138 and range of 0.003 to 1. Comparison of the box model median to those predicted by two other field-based ϕ(ClNO2) derivation methods agreed within a factor of 1.3, within the uncertainties of each method. In contrast, the box model median was 75–84% lower than predictions from the laboratory-based parameterization (i.e., [parameterization − box model]/parameterization). An evaluation of factors influencing this difference reveals a positive dependence of ϕ(ClNO2) on aerosol water, opposite to the currently parameterized trend. Additional factors may include aqueous-phase competition reactions for the nitronium ion intermediate and/or direct ClNO2 loss mechanisms. Further laboratory studies of ClNO2 formation and the impacts of aerosol water, sulfate, organics, and ClNO2 aqueous-phase reactions are required to elucidate and quantify these processes on ambient aerosol, critical for the development of a robust ϕ(ClNO2) parameterization.

Key Points

  • Box modeling of wintertime measurements over the eastern United States provides the first determinations of ClNO2 yields (ϕ(ClNO2)) from aircraft
  • The median ϕ(ClNO2) value derived from the box model is overpredicted by 75-84% by the current laboratory-based ϕ(ClNO2) parameterization
  • Modeled and parameterized ϕ(ClNO2) values show opposite trends with aerosol water, suggesting an incorrectly parameterized dependence

1 Introduction

Atmospheric reactions of nitryl chloride (ClNO2) contribute to tropospheric halogen activation and impact the distribution of oxidants and reactive nitrogen species in polluted regions (Simpson et al., 2015, and references therein). ClNO2 can be formed in up to a 1:1 stoichiometric ratio with soluble nitrate (particulate nitrate, pNO3, or nitric acid, HNO3) from the heterogeneous uptake (defined as γ) and subsequent reaction of dinitrogen pentoxide (N2O5) (R1R5). ClNO2 will photolyze at sunrise R6 but can build up at night in the residual layer (RL) where the ozone (O3) oxidation of NOx (NO + NO2) emissions forms persistent levels of N2O5 (e.g., Brown et al., 2007; Riedel et al., 2013). The production of ClNO2 from NOx and O3 is therefore expected to be most efficient under wintertime conditions where longer nights and cold temperatures stabilize and favor the formation of N2O5 in its equilibrium with the nitrate radical (NO3) R3 and minimize direct loss reactions of NO3 with biogenic volatile organic compounds (VOCs).
urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0001(R1)
urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0002(R2)
urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0003(R3)
urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0004(R4)
urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0005(R5)
urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0006(R6)

Ambient ClNO2 was first observed off the coast of Texas in 2006 (Osthoff et al., 2008). It has since been measured from ship-, ground-, and aircraft-based platforms in both continental and coastal/marine environments throughout North America (Edwards et al., 2013; Faxon et al., 2015; Kercher et al., 2009; Kim et al., 2014; Mielke et al., 2011, 2016; Osthoff et al., 2008; Riedel et al., 2012, 2013; Thornton et al., 2010; Wild et al., 2016; Young et al., 2012), Europe and the United Kingdom (Bannan et al., 2015, 2017; Phillips et al., 2012, 2016; Reyes-Villegas et al., 2018), and Asia (Liu et al., 2017; Tham et al., 2018, 2016, 2014; X. Wang, Wang, Xue, et al., 2017; T. Wang et al., 2016; Z. Wang, Wang, Tham, et al., 2017; H. Wang et al., 2018; X. Wang et al., 2014; Yun et al., 2018). Reported mixing ratios range from a few parts per trillion (pptv) to a maximum of 4,700 pptv (1-min average), measured in December 2013 in Southern China (T. Wang et al., 2016). While the absolute production of ClNO2 will depend on the rate of N2O5 formation R1R3, the uptake efficiency of N2O5 (γ(N2O5)) R4, and the presence of aerosol phase chloride (expected to vary with geographical differences in chlorine emission sources) R5, the yield of ClNO2 relative to reacted N2O5 (ϕ(ClNO2)), is defined as a value between 0 and 1 and is thought to depend only on aerosol-phase chloride and water. A parameterization for ϕ(ClNO2) based on these expected dependences has been derived in previous laboratory-based studies (Behnke et al., 1997; Bertram & Thornton, 2009; Roberts et al., 2009; Ryder et al., 2015), the details of which are discussed further below. This parameterized ClNO2 production yield has been increasingly incorporated into 3-D chemical transport models in order to simulate ClNO2 formation and evaluate its tropospheric implications (e.g., Sarwar et al., 2014; Sherwen et al., 2017). Photodissociation of ClNO2 upon sunrise will release NO2 and atomic chlorine that can lead to O3 formation during the morning hours, while HNO3, in contrast, primarily acts as a net NOx sink in the lower atmosphere. Following this trend, a previous study with the Community Multiscale Air Quality (CMAQ) Model found up to 10% increases in 8-hr-averaged tropospheric O3 in January over the United States when including a nonzero value for ϕ(ClNO2) in the model chemical mechanism (Sarwar et al., 2014). The branching between HNO3 and ClNO2 (i.e., ϕ(ClNO2)) is therefore important to parameterize accurately and evaluate against field-derived results, as it has direct implications for the predicted distributions of tropospheric oxidants and NOx.

Relatively few studies, and none from aircraft, have reported field-derived ClNO2 yields (ϕ(ClNO2)), which require simultaneous observations of ClNO2 and additional measurements such as N2O5 and/or total (particle + gas-phase) nitrate. Existing ground-based determinations of ϕ(ClNO2) have shown no strong seasonal or geographical dependences, and have values that vary within the entire possible range of 0 to 1 (Mielke et al., 2016; Osthoff et al., 2008; Phillips et al., 2016; Riedel et al., 2013; Tham et al., 2016, 2018; Thornton et al., 2010; Wagner et al., 2013, 2012; X. Wang, Wang, Xue, et al., 2017; H. Wang et al., 2018; Young et al., 2013; Yun et al., 2018). In addition, these field-derived ClNO2 yields are lower than those predicted by the laboratory-derived parameterization, which are based on aerosol chloride and water concentrations alone. This disagreement is found in every study that has made the comparison (Riedel et al., 2013; Ryder et al., 2015; Tham et al., 2018; Thornton et al., 2010; Wagner et al., 2013; Z. Wang, Wang, Tham, et al., 2017; X. Wang, Wang, Xue, et al., 2017), which suggests that the current mechanistic understanding of ClNO2 production may be complicated by the presence of additional aerosol-phase components or an undefined loss process that consumes ClNO2 (e.g., Roberts et al., 2008). As ClNO2 formation continues to be incorporated into 3-D models (e.g., Sherwen et al., 2017), further investigation into the source(s) of these field-model discrepancies is required to better understand and improve the predictive capabilities of ClNO2 formation in the wintertime RL.

Here we present the first aircraft determinations of ϕ(ClNO2), derived from a box model analysis of data from the Wintertime INvestigation of Transport, Emissions, and Reactivity (WINTER) campaign, conducted over the eastern United States during 3 February to 13 March 2015. Box model ϕ(ClNO2) results are compared to other observation-based derivation methods, including the ratio of ClNO2 to total soluble nitrate, and the laboratory-based parameterization, in order to evaluate similarities and differences between methods used in previous studies. The large WINTER data set, regional coverage of WINTER flights, and multiple measurements of gas-phase species and aerosol composition additionally allow for discussion and evaluation of factors not captured by the current laboratory-based ϕ(ClNO2) parameterization. These results can help direct future laboratory studies aimed at developing a robust ϕ(ClNO2) parameterization for ambient aerosol, and in combination with our earlier work on γ(N2O5) parameterizations (McDuffie et al., 2018), can help improve model predictions of ClNO2 formation from N2O5 heterogeneous uptake and its impact on tropospheric chemistry.

2 Methods

2.1 Measurement Campaign and Box Model

Chemical measurements, including ClNO2, were collected aboard the National Center for Atmospheric Research/National Science Foundation (NCAR/NSF) C-130 aircraft as part of the WINTER campaign during February–March 2015 (Fibiger et al., 2018; Guo et al., 2016; Kenagy et al., 2018; Lee et al., 2018; McDuffie et al., 2018; Schroder et al., 2018). A series of 13 research flights (RFs) sampled both continental and marine environments as shown in Figure 1a, with 9 flights that included some nighttime hours. A box model, previously described by McDuffie et al. (2018), was used to simultaneously derive the production rate constant of ClNO2 ( urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0007 [s−1] = k5) and the total heterogeneous loss rate constant of N2O5 ( urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0008 [s−1] = k4 + k5) to calculate ϕ(ClNO2) following the urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0009term in equation 1. Assuming that ClNO2 is exclusively formed from reaction on aerosol particles and that it has no nighttime losses, this definition is equivalent to the rightmost term of 1 where ϕ(ClNO2) is defined as the moles of ClNO2 formed relative to the integrated moles of N2O5 lost to aerosol uptake.
urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0010(1)
Details are in the caption following the image
WINTER ϕ(ClNO2) box model results. (a) WINTER flight tracks colored by night (SZA > 90°) and daytime (SZA < 90°) flights. (b) WINTER flight tracks colored by 3,425 box-model-derived values of ϕ(ClNO2). (c) Histogram of WINTER ϕ(ClNO2) results. Both are defined in the text.

Extensive model details are presented in McDuffie et al. (2018) and are only briefly described here and in sections S1 and S2 of the supporting information. The 14-reaction chemical mechanism was initialized at 1.3 hrs prior to sunset (as determined in McDuffie et al., 2018) and integrated forward in time to simulate the nocturnal evolution of an air parcel from the onset of nocturnal chemistry (near sunset), until the time of aircraft measurement. All simulations assumed constant temperature and relative humidity (RH). Initial concentrations of O3 and NO2 were first derived by iteratively fitting the model output to reproduce 10-s-averaged observations of O3 and NO2. Holding these initial concentrations constant, urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0011 and urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0012 were then adjusted to simultaneously reproduce 10-s-averaged observations of N2O5 and ClNO2. Finally, values of ϕ(ClNO2) (values of γ(N2O5) discussed in McDuffie et al., 2018) were calculated from the derived urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0013 and urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0014 products using the middle term of 1. This entire process was repeated for all points in each RF during times when the aircraft was within the RL (defined by flight based on aircraft vertical profiles of potential temperature) and the solar zenith angle (SZA) was >90°. The data were averaged over 10 s to reduce variability in the data products, while maintaining the spatial resolution of ~ 1 km. Minor model updates from McDuffie et al. (2018) are detailed in section S1 and minimally (<3%) impact both γ(N2O5) and ϕ(ClNO2) results.

Chemical measurements are briefly described here with a complete list of chemical measurements presented in Table 1 of McDuffie et al. (2018). During WINTER, multiple instruments reported 1-Hz measurements of NO2 and O3, including the National Oceanic and Atmospheric Administration (NOAA) cavity ring down spectrometer (CRD) (Fuchs et al., 2009; Washenfelder et al., 2011), University of California Berkeley Thermal Dissociation Laser Induced Fluorescence (TD-LIF) instrument (Day et al., 2002), and the NCAR Chemiluminescence (CL) detector (Weinheimer et al., 1994). Measurement accuracies for all instruments were better than 10% for NO2 and 5% for O3. N2O5 was measured at 1 Hz by both the NOAA CRD (Dubé et al., 2006; Wagner et al., 2011) and the University of Washington (UW) Iodide Time-of-Flight Chemical Ionization Mass Spectrometer (I-ToF-CIMS; Lee et al., 2014). Both N2O5 measurements agreed to within their combined measurement uncertainty of 32% on all but one flight (discussed further below). Specific instruments used for WINTER simulations varied by flight, as given in the supporting information of McDuffie et al. (2018). ClNO2 was exclusively measured with the UW I-ToF-CIMS with an accuracy better than 30% and a lower limit of detection (LOD) of 0.6 pptv. Aerosol components for particles <1 μm in diameter were measured by the University of Colorado Boulder High-Resolution Time-of-Flight Aerosol Mass Spectrometer (HR-ToF-AMS, hereafter referred to as AMS) (DeCarlo et al., 2006; Schroder et al., 2018) and the Georgia Institute of Technology Particle Into Liquid Sampler-coupled to Ion Chromatography (PILS-IC, hereafter referred to as PILS) (Guo et al., 2016). Detection limits (at 1 Hz) for the AMS were flight and compound dependent, typically between 0.012 and 0.474 μg/sm3 (sm3 refers to m3 under standard conditions [1 atm and 273.15 K]), and always <1.2 μg/sm3 with measurement accuracies (2σ) of 35% for sulfate, nitrate, and chloride. The PILS measurements of these same compounds had accuracies of 20% with compound specific detection limits of 0.06, 0.12, 0.05 μg/sm3 for sulfate, chloride, and nitrate, respectively. In addition, PILS data, collected for ~90 s every 5 min, were interpolated to match the 10-s interval of the box model results. As discussed further in section 8, particulate chloride was measured by both the AMS and PILS, both of which are used in the analyses below. In contrast to the PILS, the AMS does not efficiently sample refractory species such as NaCl (Hayes et al., 2013), and the reported chloride values from the AMS are therefore expected to be lower than the PILS. Both measurements, however, reported particulate chloride concentrations during WINTER that were lower than the instrument detection limits (PILS: 0.12 μg/sm3, AMS: typically ≤0.03 μg/sm3 for 10-s-averaged data and up to 0.05 μg/sm3 for data points with box model results). For completeness, data reported both above and below the detection limits of each instrument are presented in the analyses below.

Aerosol water concentrations were derived as described previously in the supporting information of McDuffie et al. (2018), with an accuracy of ~ 25%. Briefly, inorganic-associated aerosol water (<1-μm diameter) was calculated using the ISORROPIA thermodynamic model (Fountoukis & Nenes, 2007) as described in Guo et al. (2016), while the organic-associated water was estimated using a constant hygroscopicity factor of 0.1 and organic mass measured by the AMS. Data were filtered to only include points with ambient RH < 95% due to an increased uncertainty in the hygroscopic growth factor at high RH. Data were not filtered for low ambient RH, though points with <40% RH may have an increased uncertainty greater than 25% (Guo et al., 2016). The box model calculation of ϕ(ClNO2) is independent of particulate phase chloride and water and is, therefore, not subject to increased uncertainty associated with these low values. In contrast, the ϕ(ClNO2) parameterization (section 8) is sensitive to uncertainties in both particulate chloride measurements and aerosol water calculations, which are addressed in section 9.

2.2 Box Model Limitations, Uncertainties, and Sensitivity Studies

Box model results for ϕ(ClNO2) are dependent on urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0015 and therefore subject to many of the same model limitations discussed previously in McDuffie et al. (2018). These include the assumption of constant RH and temperature during the course of an air parcel trajectory, and uncertainties in NO3 reactivity ( urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0016) (e.g., VOC measurements, direct NO3 loss, and reaction with radicals, discussed in section S2.2.5). These uncertainties can increase variability in urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0017 and therefore in urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0018. Additional uncertainties specific to urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0019 and ϕ(ClNO2) are discussed below and in detail in supporting information sections S2.1 and S2.2.

First, large measured mixing ratios of ClNO2 relative to N2O5 can result in model non-convergence, similar to the γ(N2O5) measurement sensitivities discussed in McDuffie et al. (2018). Non-convergence occurs when values of urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0020 equal to urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0021 (i.e., ϕ(ClNO2) = 1, its upper limit) are not sufficient to reproduce the observed ClNO2 mixing ratios. A total of 12.6% of WINTER points did not converge, and were removed from this analysis. The majority of these points (389 of 486 total points) occurred during RF03 in a plume of urban outflow off the coast of New York City. Model non-convergence for this and simulations of other flights could arise from multiple sources of box model uncertainties including air age (estimated from observed NOx/NOy ratio, as described in McDuffie et al. (2018)), simulation start time, air parcel dilution/mixing, and disagreement between the CRD and I-ToF-CIMS measurements of N2O5, used as a model fit parameter. Additional sensitivity tests for RF03 were performed to assess the possible sources of this model non-convergence (described in section S2.1 and Figure S1). Based on 19 sensitivity tests (section S2.1), the cause of non-convergence could not be identified. Non-convergent points, however, are not further considered in this analysis as setting each corresponding ϕ(ClNO2) to a value of 1 only increases the WINTER median by 22.8%, to a value of 0.169.

Second, model simulations were conducted assuming no interaction with the surface through dry deposition and/or surface emission. While this is a reasonable assumption for isolated air in the continental RL, a well-mixed marine boundary layer is expected at depths of at least 500 m during wintertime off the U.S. East Coast (Seidel et al., 2012). If air sampled during WINTER was in contact with the ocean surface, deposition should be included in the model. Due to uncertainties in depositional fluxes, deposition was not included in base case simulations and was instead tested through two sensitivity studies. For these tests, the depositional flux for N2O5 was estimated using the exchange velocity derived from an observational analysis by Kim et al. (2014) (further details in section S2.2.1). Including N2O5 deposition increased the median ϕ(ClNO2) over the ocean by 28%, from a value of 0.145 to 0.186 (ocean data only, see section S2.2.1 and Figure S2). The second test included estimates for both N2O5 and ClNO2 deposition. While N2O5 uptake to chloride-rich seawater is expected to result in a positive ClNO2 flux from the ocean surface (provided ClNO2 re-volatilizes to the gas phase), Kim et al. (2014) observed a slight negative ClNO2 flux from eddy covariance measurements at night at a coastal location in Southern California. Including a ClNO2 dry deposition velocity approximately one third the magnitude of that for N2O5 (based on Kim et al., 2014) further increased the median box model ϕ(ClNO2) value over the ocean by an additional 35%, to a value of 0.251 (section S2.2.1). Both values, however, remained lower than those predicted by the laboratory-based parameterization (Figure S3), indicating that the model assumptions related to ocean exchange do not change the main conclusions presented below.

Finally, to test the overall model sensitivity to uncertainties in model fit parameters and assumptions, a series of 18 sensitivity studies were conducted for each flight, with the results presented in section S2.2 (Figures S2–S11) and summarized in Table S1. Of the parameters tested, ϕ(ClNO2) was most sensitive to uncertainties in ClNO2 and N2O5 deposition, which increased the median ϕ(ClNO2) value over the ocean by 73%, to a value of 0.251 (discussed above). The second largest sensitivity was to assumptions in air age, with a 43.7% increase and 25.3% decrease in the median ϕ(ClNO2) under assumptions of younger and older air, respectively (Table S1 and Figure S4). Uncertainties in chemical measurements used as model fit parameters resulted in a range of −29.8% to +34.5% for changes in median ϕ(ClNO2) (absolute values of 0.092 to 0.164; Figures S5–S8). Data over the ocean were additionally tested for sensitivities to air parcel dilution with simultaneous entrainment of background O3, which increased the median of this subset of points by 21.3%, from a value of 0.188 to 0.228 (Figure S2). Dilution/mixing was modeled in this test as a first order loss process for all species with a dilution rate constant of 3.1 × 10−5 s−1, estimated from multiple encounters of the same air parcel on RF03. An additional modeling analysis of WINTER data by Kenagy et al. (2018) derived a similar lifetime for loss associated with both mixing (1/τmix = 1.9 × 10−5 s−1) and deposition (1/ urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0022 = 1.4 × 10−5 s−1). The median ϕ(ClNO2) had less than 7% sensitivities in all remaining tests, including uncertainties in the time elapsed before sunset (Figure S9), NO3 reactivity (Figure S10), and photolysis rates (section S2.2.6 and Figure S11; Madronich et al., 1998; Shetter & Muller, 1999). Despite the relatively large sensitivities observed in some tests, median ϕ(ClNO2) values always remained less than 0.251, within 0.113 of the base case median and lower than the median predicted by the laboratory-based parameterization, discussed in section 8.

3 Results

3.1 Box Model Analysis

Box model simulations resulted in 3,425 individual determinations of ϕ(ClNO2), encompassing nearly the entire possible range, with values from 0.003 to 1. The number of ϕ(ClNO2) determinations reported here (N = 3,425) is larger than the number of γ(N2O5) determinations reported by McDuffie et al. (2018; N = 2,876) due to the dependence of γ(N2O5) on aerosol surface area measurements, which were not required for ϕ(ClNO2) and not always available during WINTER flights. WINTER flight tracks are colored by ϕ(ClNO2) determinations in Figure 1b, with the campaign distribution shown in Figure 1c. The ϕ(ClNO2) distribution had a median and mode of 0.138 (1σ: +0.051/−0.045, described below) and 0.030, respectively. Data in Figure 1b show several areas of larger ϕ(ClNO2) associated with specific flights and generally higher values downwind of New York City, the largest regional NOx source. The ϕ(ClNO2) values otherwise do not show a strong geographical distribution. Data sampled over both ocean (N = 1,896) and land (N = 1,529) encompassed the same range in ϕ(ClNO2) (Figure S13), but with medians of 0.203 and 0.075, respectively. While larger yields may be expected in chloride-rich oceanic air, the two populations may be similar as many WINTER flights over the ocean sampled continental urban outflow. Box model uncertainties (1σ) (time series in Figure S12) were calculated for each individual ϕ(ClNO2) value from the quadrature addition of measurement uncertainties (O3, NO2, N2O5, and ClNO2) and model sensitivities to air age, simulation start time, photolysis rates, dilution, and 50% changes in total urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0023 (section S2). Sensitivity to deposition was not included in the total error calculation due to deposition rate uncertainties, but is discussed further in section 12.

WINTER values are compared in Figure 2 (and Table S2) to all previously reported field determinations of ϕ(ClNO2). Figure 2 shows that ϕ(ClNO2) values are variable and do not show a consistent dependence on geographical location, although the current database may be too sparse to illustrate such differences on continental or regional scales. The WINTER distribution and median appear similar to those reported from both continental and coastal locations across North America (Mielke et al., 2011, 2013, 2016; Osthoff et al., 2008; Riedel et al., 2013; Thornton et al., 2010; Wagner et al., 2012, 2013; Table S2). The reported average values in Europe (Phillips et al., 2016) and polluted regions in China (Tham et al., 2016, 2018; X. Wang, Wang, Xue, et al., 2017; Z. Wang, Wang, Tham, et al., 2017; H. Wang et al., 2018; Yun et al., 2018), however, are larger than the median (Figure 2) and mean (Table S2) during WINTER.

Details are in the caption following the image
Map of all reported field determinations of ϕ(ClNO2). The geographic location of each field study is represented by a diamond, colored by the reported (or calculated) average or median value of ϕ(ClNO2), at each location. All 3,425 determinations from WINTER are shown and colored by the box-model calculated ϕ(ClNO2) values. Additional graph inserts show the maximum range and median or average value reported by each study at each location.

Additional, real geographical differences in ϕ(ClNO2) may be obscured by varying ϕ(ClNO2) derivation methods used in past literature. For example, the method employed by Mielke et al. (2016), Mielke et al. (2013), and Osthoff et al. (2008) (applied to WINTER data as Method 1 in section 6) defines ϕ(ClNO2) as the amount of ClNO2 produced relative to the integrated amount of NO3 radical formed, not N2O5 lost, which may be considered a lower limit to ϕ(ClNO2). Methods relating the amount of observed ClNO2 to total nitrate, as employed by Riedel et al. (2013), Wagner et al. (2012), Phillips et al. (2016), and Tham et al. (2018) (Method #2 in section 6) have additional uncertainties described in the following section. Studies by Tham et al. (2014), X. Wang, Wang, Xue, et al. (2017), and Z. Wang, Wang, Tham, et al. (2017) defined ϕ(ClNO2) following the right hand side of 1, equivalent to the box model calculation for WINTER, but calculated urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0024 from the steady state approximation, which may lead to an over-prediction of urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0025 (underprediction of ϕ(ClNO2)) in cold and/or high-NOx environments (Brown et al., 2003). The studies most directly comparable to WINTER are Yun et al. (2018) and Wagner et al. (2013), who deployed similar versions of the WINTER box model. In contrast to the WINTER model, Wagner et al. (2013) used the right side of 1 to derive ϕ(ClNO2) rather than iteratively fitting the model to ClNO2 observations. Further comparisons of these methods applied to WINTER data are presented next.

3.2 Comparison to Multiple Definitions of ϕ(ClNO2)

In this section, four methods are applied to WINTER data in an attempt to provide a direct comparison and evaluation of methods commonly used in past literature. Due to the difference in aircraft and ground-based observational data, ϕ(ClNO2) derivations using steady state-derived urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0026 in 1 (used by Tham et al., 2016; X. Wang, Wang, Xue, et al., 2017; Z. Wang, Wang, Tham, et al., 2017) and the ratio of ClNO2 to pNO3 production rates (used by Wang et al., 2018) could not be compared to box model results. Each additional method tested here is described below and assumes that heterogeneous production is the only source of ClNO2, and that ClNO2 is stable overnight.

In Method 1, ϕ(ClNO2) is defined in 2 as the amount of observed ClNO2 per amount of NO3 radical produced. urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0027 is defined in 3 as the instantaneous rate of nitrate radical production from the oxidation of NO2 with O3 R2. The term dtSunset is the amount of time elapsed between the onset of nocturnal chemistry (approximately sunset) and the time of aircraft measurement. Previously used by Osthoff et al. (2008), Mielke et al. (2013), and Mielke et al. (2016), this definition of ϕ(ClNO2) may be a lower limit to ϕ(ClNO2) as NO3 production does not always lead to N2O5 and subsequent HNO3/ClNO2 formation. Instantaneous urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0028, however, also decreases overnight as NO2 and O3 are consumed, which could alternatively lead to an overprediction of ϕ(ClNO2) that would increase with simulation duration.
urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0029(2)
urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0030(3)
Method 2 defines ϕ(ClNO2) in 4, calculated from the slope of the linear regression between observed ClNO2 and total soluble nitrate (HNO3 + particulate nitrate). This method has been used by Riedel et al. (2013), Wagner et al. (2012), Phillips et al. (2016), and Tham et al. (2018). Here ClNO2 yields were calculated every 10 s from linear fits of 1-Hz ClNO2 observations against the sum of gas-phase HNO3 and submicron particulate NO3 (i.e., total soluble nitrate), as measured with the I-ToF-CIMS and AMS, respectively. In these fits, intercepts were not forced to zero. Example individual correlations derived from five different flights are highlighted in Figure S14. The total number of derived fits was additionally filtered to only include individual fits with at least eight data points and statistically significant (p < 0.05) correlation coefficients. This correlation filter is expected to bias the campaign median ϕ(ClNO2) value high due to low correlations associated with many of the low ClNO2 yields. Filtering all four methods for the same points, however (described below), can provide a direct methods comparison for this subset of points. Method 2 also excludes particulate NO3 from super micron aerosol (1–4 μm) due to the low measurement frequency (~7 min between samples), which could bias the ϕ(ClNO2) values high if these large particles serve as a reservoir for nitrate formed overnight. This method also assumes no pNO3 contribution from reaction of NO3 with hydrocarbons, though these reactions are expected to be small due to low total NO3 reactivity during winter. To test the sensitivity of Method 2 to time-dependent processes (e.g., deposition), yields were additionally calculated from ClNO2 and total nitrate correlations over increased time intervals of 30 and 100 s. At these lower time resolutions, however, the median ϕ(ClNO2) changed by less than 0.06 for all three calculated intervals (10, 30, and 100 s) and number of simultaneous determinations from Methods 1–3 was reduced from 320 (described below) to fewer than 200. In comparison, the box model (Method 3) is largely independent of observed total NO3 and is not highly sensitive to assumptions of NO3 loss or previous day production.
urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0031(4)
Method 3 is the previously described box model, while the fourth calculates ϕ(ClNO2) using the laboratory-based parameterization provided in 5, using aerosol water and chloride concentrations. This particular calculation uses rate coefficient ratios from Bertram and Thornton (2009), with additional rate constant ratios discussed in the following section. Here parameterized ϕ(ClNO2) values are calculated separately using measurements of particle-phase chloride from both the AMS (nonrefractory submicron chloride only) and PILS (total submicron soluble chloride) instruments, as discussed further below.
urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0032(5)
Details are in the caption following the image
(left) Box and whisker plots comparing four derivation methods for ϕ(ClNO2) during WINTER, illustrating the agreement in ϕ(ClNO2) variability predicted by each method. Details of each method are described in the text and shown in the legend above. Method 4 calculates ϕ(ClNO2) from the laboratory-based parameterization using both (4a) AMS and (4b) PILS particle chloride measurements. Bars represent the 10th, 50th, and 90th percentiles. Boxes show the 25th to 75th percentiles, and stars represent the averages. (right) Median values for each method are shown by red diamonds with red bars representing the uncertainty in the median for each method. Data in both panels are filtered to include points (N = 320) where ϕ(ClNO2) values were simultaneously derived for all methods.
Details are in the caption following the image
Box model results compared to parameterized ϕ(ClNO2) values. (a and b) ϕ(ClNO2) as a function of WINTER aerosol Cl:H2O molar ratio, calculated from (a) AMS and (b) PILS particulate chloride measurements. Laboratory-based parameterizations are shown by gray lines and WINTER box model results shown by blue circles. Red dashed lines represent the total absolute upper and lower error limits of the Bertram and Thornton (2009) parameterization. (c and d) Parameterized ϕ(ClNO2) values against box model results, using (c) AMS and (d) PILS chloride measurements. Black lines are the 1:1 line. In all panels, data with particulate chloride measurements above reported detections limits for AMS and PILS instruments are shown by dark circles; all remaining data are in light blue or gray. Medians are shown in (c) and (d) for data with particulate chloride above instrument detection limits.

The WINTER ϕ(ClNO2) values from Methods 1–4 are compared in Figure 3. Figure 3 is not representative of the entire WINTER campaign distribution. Data have been filtered to only include points with simultaneous ϕ(ClNO2) determinations from all four methods, which reduced the total number from 3,425 to 320, mostly as a result of Method 2 (ratio method). Method 2 produced the lowest number of ϕ(ClNO2) determinations due to the r2 filter that required a statistically significant correlation between total nitrate and ClNO2 over each 10-s interval (described above). The left panel of Figure 3 shows the 10th, 25th, 50th, 75th, and 90th percentiles, while the right shows the medians and error bars that represent the absolute 1σ uncertainty in the median of each method (detailed in section S3.1). Briefly, the total errors associated with Methods 1 and 2 are calculated from the quadrature addition of measurement uncertainties (i.e., ClNO2, O3, NO2, HNO3, and pNO3) and the absolute box model (Method 3) error is calculated as described in the previous section. Error in the parameterization (Method 4) is calculated from uncertainties in AMS and PILS chloride measurements and the aerosol water calculation.

Results in Figure 3 show that for the 320 points compared, Method 1 predicts the lowest average and percentile values (except for the 50th) for ϕ(ClNO2). The median (0.19 ± (1σ) 0.06), however, is within the uncertainties of the medians calculated using both Methods 2 (0.24 ± 0.10) and 3 (0.19 ± 0.06) (right panel). Method 2 derived larger ϕ(ClNO2) values than Methods 1 and 3, but a median (0.24 ± 0.10), again, within the uncertainties of those calculated here for those methods (0.19 ± 0.06 for both) (Figure 3, right). A previous methods comparison of data from winter 2011 in Colorado also showed similarity between Methods 2 and 3, with an average ϕ(ClNO2) value of 0.05 ± 0.15 using Method 2 (Riedel et al., 2013) and a mode of ~0.06 derived from a box model similar to the one used here (Wagner et al., 2013). An additional comparison of Method 2 and the right-hand side of 1 (calculated from the steady state approach and observed ClNO2 production rates) by Z. Wang, Wang, Tham, et al. (2017) found absolute agreement within 0.03 for the campaign average ϕ(ClNO2) value at a ground site in northern China during summer 2017.

Values derived here from the parameterization in 5, using both AMS (a) and PILS (b) particle chloride measurements, were generally larger than those predicted by all other methods, with medians of 0.52 and 0.82, respectively. Larger ϕ(ClNO2) values calculated from PILS chloride data are consistent with the PILS sampling refractory chloride species. Regardless of particle chloride differences, both predicted median values that were factors of 2 to 4.3 larger than other methods, and outside the range of uncertainties associated with Methods 1 and 3 (Figure 3, right). All previous field studies to have made this comparison between data-based methods and the laboratory-based parameterization have shown an overprediction in ϕ(ClNO2) by the parameterization (Riedel et al., 2013; Tham et al., 2018; Thornton et al., 2010; Wagner et al., 2013; X. Wang, Wang, Xue, et al., 2017; Z. Wang, Wang, Tham, et al., 2017). ClNO2 yields derived from ambient seawater samples by Ryder et al. (2015) have also resulted in values lower than the parameterized equivalents. Possible factors associated with this observed difference between field-derived and parameterized ϕ(ClNO2) values during WINTER are discussed in the remaining section.

4 Discussion—Evaluation of the Current ϕ(ClNO2) Parameterization

4.1 Parameterization Background

Behnke et al. (1997) first proposed the chemical mechanism for the bulk-phase reaction of aqueous N2O5 and subsequent formation and evaporation of ClNO2, shown in R7R13. Based on this currently accepted mechanism, an expression for ϕ(ClNO2) has been previously derived from the ratio of ClNO2 production relative to N2O5 loss, assuming the hydrated nitronium ion intermediate (H2ONO2+) is in steady state (e.g., Bertram & Thornton, 2009). This expression, given in 5, simplifies to describe the ClNO2 yield as a competition reaction between Cl and H2O for the H2ONO2+ intermediate (derivation reproduced in section S4). While this ϕ(ClNO2) expression has been consistent across multiple laboratory studies, kinetic laboratory experiments have reported a range of values for the k12/k11 rate constant ratio. Based on observed ClNO2 formation from N2O5 uptake onto aqueous NaCl particles in wetted flow tube experiments, Behnke et al. (1997) derived a value of 836 ± 32 for the term k12/k11. More recent laboratory studies on chloride-containing aerosol have derived values in the range of 450 to 505 (Bertram & Thornton, 2009; Roberts et al., 2009; Ryder et al., 2015). These variations of the ϕ(ClNO2) parameterization are compared in Figure 4, which is described further in the following section.
urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0033(R7)
urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0034(R8)
urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0035(R9)
urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0036(R10)
urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0037(R11)
urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0038(R12)
urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0039(R13)

4.2 Box Model and Chemical ϕ(ClNO2) Parameterization Comparison

Parameterized predictions of the WINTER ϕ(ClNO2) values are shown with the base case box model results in Figures 4a–4d. In Figures 4a and 4b, the parameterized ϕ(ClNO2) values (gray/black lines) and the box model results (blue circles) are plotted as a function of the aerosol Cl:H2O molar ratio, calculated with aerosol water estimates and particulate chloride measurements from both AMS (a) and PILS (b) instruments. Aerosol water and chloride concentrations from 1–4 μm particles were not included in Figure 4 due the small fractional contribution of this size range to aerosol surface area (0–2%) (required for N2O5 uptake), relative to the total surface area contribution from smaller particles (<1 μm). The presence of chloride in these larger particles more likely contributes to the formation of gas-phase HCl through acid displacement, which can serve as a pool of chloride that equilibrates with submicron particles (Osthoff et al., 2008). Dark blue circles in Figures 4a and 4b indicate points where reported aerosol chloride concentrations were above the instrument detection limits. Figures 4c and 4d further show the correlations between the box model results (x axis) and parameterized ϕ(ClNO2) values (y axis) using the Bertram and Thornton (2009) k12/k11 ratio and AMS (c) and PILS (d) particulate chloride. Similarly to Figures 4a and 4b, dark gray circles in Figures 4c and 4d indicate where particulate chloride concentrations were measured above instrument detection limits, with the median ϕ(ClNO2) values for these subsets shown by the red squares.

Results in Figure 4 demonstrate a ϕ(ClNO2) overprediction by the parameterization (regardless of k12/k11 ratio), which was also shown for a subset of WINTER data in the previous section. In Figure 4, over 90% of the individual box model ϕ(ClNO2) values are overpredicted by the Bertram and Thornton (2009) parameterization, when calculated using both AMS and PILS aerosol chloride. In addition, the ratio of the box model median to the medians calculated using the Bertram and Thornton (2009) parameterization range from 0.25 to 0.16, using PILS and AMS chloride, respectively. In other words, the box model median ϕ(ClNO2) value is 75–84% lower than the medians calculated from the parameterization (i.e., (parameterization − box model)/parameterization).

Differences between the box model and parameterized ϕ(ClNO2) values may result from uncertainties in either derivation method. To assess the role of uncertainty in the parameterization, estimates of the upper and lower limits of the parameterized values are shown by the dashed red lines in Figures 4a and 4b, calculated from the Bertram and Thornton (2009) k12/k11 ratio and uncertainties in aerosol water (~25%) and chloride (35% AMS, 20% PILS). Both sets of chloride measurements, above and below instrument detection limits, are also included in Figure 4. Though the majority (≥ 50%) of reported chloride observations were below instrument detection limits (light blue [Figures 4a and 4b] and gray [Figures 4c and 4d]), over 73% of the box model ϕ(ClNO2) values that corresponded to above-LOD chloride measurements, fell below the lower-limit estimate of the ϕ(ClNO2) parameterization (lower red line in Figures 4a and 4b). This trend was consistent between parameterizations calculated using both AMS and PILS measurements, suggesting that uncertainties in the parameterization from the combined uncertainty in aerosol chloride and water are not responsible for the majority of overprediction by the simple chemical ϕ(ClNO2) parameterization.

To further assess the contribution of box model error to the observed differences in Figure 4, upper- and lower-limit box model values (calculated from the analysis of total model error) are plotted together with parameterized ϕ(ClNO2) values in Figure 5, as a function of Cl:H2O, using the Bertram and Thornton (2009) k12/k11 ratio. As discussed in section 5, error in each box model-derived ϕ(ClNO2) value was individually calculated from the quadrature addition of measurement uncertainties (O3, NO2, N2O5, and ClNO2) and model sensitivities to air age, simulation start time, dilution, photolysis rates, and 50% changes in total urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0040. Of these parameters, air age (discussed in section S2.2.1) was the largest contributor to the total model error shown in Figure S12 and Table S1. In Figure 5, points of model non-convergence during sensitivity studies (i.e., urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0041 > urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0042) were conservatively set to ϕ(ClNO2) values of 1. Results show that the median box model values (black squares in Figure 5, calculated from data with particulate chloride > LOD) remain lower than their parameterized equivalents in all comparisons, regardless of chloride measurement. This comparison indicates that while the WINTER ϕ(ClNO2) values are sensitive to model assumptions (air age in particular), box model uncertainties are not the main source of difference between the model and laboratory-based ϕ(ClNO2) parameterization.

Details are in the caption following the image
Highest (red) and lowest (blue) estimated box model ϕ(ClNO2) values, plotted against WINTER Cl:H2O molar ratios. Black curves represent the ϕ(ClNO2) values predicted by the Bertram and Thornton (2009) parameterization. Gray curves are the upper and lower limits of the parameterization error. Squares represent the median of each set of modeled WINTER ϕ(ClNO2) values (pCl > LOD points only), plotted at the WINTER median Cl:H2O molar ratio.

When considering uncertainties in each derivation method, results in Figures 4 and 5 suggest that field-derived ϕ(ClNO2) values are overpredicted by the laboratory-based parameterization. These results are qualitatively consistent with all other reported field-parameterization comparisons (Riedel et al., 2013; Ryder et al., 2015; Tham et al., 2018; Thornton et al., 2010; Wagner et al., 2013; Z. Wang, Wang, Tham, et al., 2017; X. Wang, Wang, Xue, et al., 2017), suggesting the presence of at least one physiochemical process suppressing ϕ(ClNO2) relative to production yields predicted on pure NaCl/inorganic aqueous solutions. In the following sections we use box model ϕ(ClNO2) results and observed WINTER variables to examine possible sources of the difference between the box model and the laboratory-based ϕ(ClNO2) parameterization. In the first section we discuss trends in this difference with measured aerosol components, particularly aerosol water. The last two sections assess two possible mechanistic sources of ϕ(ClNO2) suppression that have been discussed previously in field (Mielke et al., 2013; Phillips et al., 2016; Tham et al., 2018; Z. Wang, Wang, Tham, et al., 2017) and laboratory-based (e.g., Roberts et al., 2008; Ryder et al., 2015) studies of ClNO2 yield. These include: (1) the presence of additional competition reactions for the H2ONO2+ intermediate and (2) direct loss of gas- or aqueous-phase ClNO2 via surface deposition/aerosol uptake and aqueous-phase reaction.

4.3 Sources of Parameterization-Box Model Differences

4.3.1 Observed Trends/Water Dependence

Of the aerosol components calculated or measured during WINTER, the difference between parameterized and box model-derived ϕ(ClNO2) values was most strongly correlated with aerosol water. Table S3 shows that the largest correlation coefficients (for both PILS and AMS calculated parameterizations) were associated with aerosol water molarity (r2 = 0.54 [AMS], 0.22 [PILS]), ambient RH (r2 = 0.53 [AMS], 0.27 [PILS]), and aerosol liquid water content (water mass fraction) (r2 = 0.51 [AMS], 0.21 [PILS]). The only other parameters with correlation coefficients above 0.1 were with wet (including aerosol water) mass fractions of aerosol organics, sulfate, and ammonium. When eliminating the role of water, the dry (excluding water) mass fractions produced lower correlation coefficients (r2 ≤ 0.05) for each of these species. Similarly, correlations with aerosol molar ratios of Cl/NO3, pH (from Guo et al. (2016)), and O:C ratio produced correlation coefficients less than 0.09. Two previous field studies observed a negative correlation between absolute ϕ(ClNO2) values (derived using the steady state of N2O5) and aerosol-phase nitrate mass (Z. Wang, Wang, Tham, et al., 2017), as well as low ClNO2/N2O5 gas-phase ratios corresponding to aerosol with low Cl/organic mass ratios (Mielke et al., 2013). Neither of these studies quantitatively evaluated the role of aerosol composition in the difference between parameterized and field-derived values. For comparison, WINTER box model ϕ(ClNO2) values were only weakly correlated with Cl/organic mass ratio (r2 ≤ 0.027 for both chloride measurements) and showed an even weaker, positive correlation with aerosol nitrate mass (r2 = 0.024).

The difference between the ϕ(ClNO2) parameterization and box model results for each individual point is plotted against aerosol water molarity in Figure 6 (for both AMS [a] and PILS [b] chloride measurements). Trends in Figure 6 show negative correlations between this difference and aerosol water for points with aerosol chloride both above and below the instrument detection limits (black and gray points, respectively). While quantitatively different slopes are derived from each fit, all trends (with AMS and PILS chloride, above and below detection limits) are qualitatively consistent, suggesting that either aerosol water or an associated factor is an important predictor of the difference between the field-derived ϕ(ClNO2) values and those predicted by the current parameterization. Based on the aqueous formation mechanism in R7R13, the role of water in the yield of ClNO2 is to act in competition with aqueous-phase chloride for the H2ONO2+ intermediate. This competition results in a decrease in the parameterized ϕ(ClNO2) as water increases (Figures S15a and S15b). The opposite trend is generally observed for WINTER box model results, which show a positive correlation with water (Figure S15c). Two RFs with the largest water concentrations (exceeding 40 [M]), however, showed no observable trends. Combined, these opposite trends with water lead to the negative slopes in Figure 6.

Details are in the caption following the image
Difference between the ϕ(ClNO2) parameterization and WINTER box model results, calculated using (a) AMS and (b) PILS particulate chloride. Data points above chloride detection limits (LOD) are shown in black with the fit line slopes (solid line) provided in each panel. All data points are shown in gray with fit line slopes (dashed lines) provided in each panel.

Aerosol water is a calculated rather than observed quantity, and could therefore be assumed as a potential source of the disagreement between predicted and observed trends in ϕ(ClNO2). Nevertheless, the trends in Figure 6 are unlikely the result of uncertainty in calculated water molarity. For example, points below 20-M H2O in Figure 6 would require H2O concentrations in 5 to be more than 100 times larger, on average, to bring the parameterization into agreement with the box model, well outside the ~25% uncertainty in [H2O]. Disagreement with the box model at low aerosol water (and RH) may, therefore, suggest that laboratory-based parameterizations are not largely applicable to environments with limited aerosol water since they have been derived from studies conducted at either high RH (> 55%) or on aqueous solutions (Behnke et al., 1997; Bertram & Thornton, 2009; Roberts et al., 2009; Ryder et al., 2015). Correlations between aerosol water and box model ϕ(ClNO2) values in Figure S15c, however, also show quantitatively different trends for each flight, suggesting that multiple factors may be contributing to the discrepancy between box model and parameterized ϕ(ClNO2) values. Only two previous field studies have examined the ϕ(ClNO2) relationship with water. While ϕ(ClNO2) values at a ground site in NW Germany showed no trend between RH values of ~65 and 90% (corresponding to WINTER water concentrations of ~20–45 M) (Phillips et al., 2016), a weak positive correlation of ϕ(ClNO2) with aerosol water (~10–60 M) was also observed at a ground site in Northern China (Tham et al., 2018). Further studies of ClNO2 production under a range of aerosol water conditions will be required to confirm this result.

The physical mechanism for the observed box model trend with water is uncertain but may be related to the physical availability of chloride, as discussed in previous studies as a possible cause of ϕ(ClNO2) suppression in field-derived results (Mielke et al., 2013; Phillips et al., 2016; Z. Wang, Wang, Tham, et al., 2017). The current ϕ(ClNO2) parameterization assumes internally mixed aerosol where all Cl is available for reaction, which may not be the case for ambient aerosol. For example, measured particulate chloride may not be present equally throughout the particle size distribution, an effect that would increase the parameterized ϕ(ClNO2) values if the largest chloride concentrations were present in a different size range than the particles contributing most to the surface area density (i.e., participating in N2O5 uptake). Based on the WINTER aerosol size distributions measured by the UHSAS (0.06–1 μm) and PCASP (1–3 μm), the median of the aerosol surface area distribution (dS/dlogDp) corresponded to particle diameters between 0.12 and 0.3 μm, for the data shown in Figure 4. The size distribution of total particulate chloride, however, was not reported during WINTER and cannot be further evaluated as a possible source of observed difference between the ϕ(ClNO2) parameterization and box model results.

Additionally, even if particulate chloride is present evenly throughout the size distribution, it may not be accessible within the aerosol itself, which may depend on RH and the physical and chemical properties of the aerosol. For example, previous studies have found that aqueous Cl has a propensity to partition away from the surface (e.g., Cummings & Wick, 2013) and that submicron sea salt aerosol may form organic coatings (Ault et al., 2013), especially when aged (Laskin et al., 2012). The increased presence of aerosol organics relative to water has also been shown to change the rate of diffusion and solubility of aqueous N2O5 (e.g., Anttila et al., 2006; Gaston et al., 2014) and, therefore, may also impact the mobility of Cl ions by inducing changes in the aerosol phase or viscosity (e.g., Gržinić et al., 2015; Shiraiwa et al., 2017) and/or the formation of liquid-liquid phase separations (e.g., Bertram et al., 2011). Each of these processes is dependent on RH and the presence of organics and may serve to limit the availability of Cl at the aerosol surface under low aerosol water conditions. This would reduce field-derived ϕ(ClNO2) values relative to the parameterization if N2O5 dissociation and reaction occurs near the surface, physically removed from Cl residing in the bulk.

Changes in aerosol phase and morphology (i.e., core-shell) associated with RH and organic content were not measured during WINTER, but a parameterization by Bertram et al. (2011) was used with RH and O:C ratio measurements to predict the presence of liquid-liquid phase separations, as described in section S3.2. Results in Figure S16, however, show that predictions of phase-separated aerosol do not consistently correspond with the largest differences between the box model results and ϕ(ClNO2) parameterization. In contrast, many of the largest ϕ(ClNO2) differences were associated with the longest organic aerosol mixing times (i.e., largest aerosol diffusion coefficients), predicted by two RH-dependent parameterizations for 200-nm diameter α-pinene secondary organic aerosol (SOA), presented by Maclean et al. (2017) (shown in Figure S17). Aerosol mixing time parameterizations, however, have only been developed for α-pinene SOA and still have significant uncertainty (Maclean et al., 2017) and therefore require more work to determine their applicability to low biogenic WINTER aerosol (see McDuffie et al., 2018) and the extent to which diffusion may impact ϕ(ClNO2). As described in section S3.2, a similar parameterization for aerosol viscosity by Shiraiwa et al. (2017) could not be calculated from WINTER data. Lastly, previous studies have additionally used estimates of the N2O5 diffusion distance prior to reaction (reacto-diffusive length), together with aerosol size and composition to predict N2O5 uptake onto organic and inorganic aerosol (e.g., Anttila et al., 2006; Gaston et al., 2014; Gaston & Thornton, 2016). The utility of this parameter to predict changes in ϕ(ClNO2), however, has not been previously examined and remains uncertain here as the ϕ(ClNO2) difference does not strongly correlate with relevant variables other than aerosol water, such as O:C ratio and organic content (Table S3). Therefore, while the observed water trend may be related to chloride availability, which could be impacted by organic-induced changes in aerosol morphology and viscosity, the cause of this trend remains inconclusive.

4.3.2 Additional Aqueous Competition Reactions

Reaction between the H2ONO2+ intermediate and species other than Cl and H2O could additionally contribute to the observed suppression of ϕ(ClNO2) on ambient aerosol. Such a process would add an additional competition reaction in the form of R14 to the mechanism in R7R13. In order for a reaction of this form to compete with aqueous Cl and cause a reduction in ClNO2 production relative to N2O5 uptake, the product of k14 and the concentration of additional reactive compounds would have to be comparable to k12[Cl]. In addition, agreement between the box model and two nitrate-dependent observational ϕ(ClNO2) methods (Figure 3), suggests that this reaction would also have to produce particle-phase nitrate or gas-phase HNO3 to maintain consistency between the observational methods.
urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0043(R14)

Previous studies have reported evidence of a competition between particle-phase chloride and halogens. For example, enhanced Br2 formation relative to ClNO2 has been observed on ice at Cl:Br ratios <30 (Lopez-Hilfiker et al., 2012). In addition, reaction of N2O5 with dilute NaI and NaBr solutions has shown production of BrNO2, Br2, and I2 (e.g., Behnke et al., 1994; Schweitzer et al., 1998). While the latter studies do not show direct competition with Cl, the stronger nucleophilic character of Br and I relative to chloride may allow for efficient competition. Ambient Br and I concentrations in sea water (expected Cl:Br:I ratios of ~1:1 × 10−3:1 × 10−6) may be too small, however, to compete with Cl via R14 during WINTER. These species may alternatively reduce ϕ(ClNO2) via direct reaction with ClNO2, further discussed in the following section. In addition, these reactions may not lead to the production of NO3 or HNO3 (e.g., BrNO2 formation), making the presence of these reactions potentially inconsistent with the previous observation-based methods (Figure 3), which incorporate nitrate mass balance between N2O5, particulate nitrate, HNO3, and ClNO2.

Additional studies have also found efficient reaction between the nitronium ion and aqueous-phase aromatics (Hoggett et al., 1971; Lüttke et al., 1997; Schofield, 1980; Taylor, 1990). Experiments focused specifically on reactions with a subset of phenols (Heal et al., 2007) derived urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0044 ratios ( urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0045 = k11[H2O]) that correspond to k14/k11 ratios (for average WINTER aerosol water concentrations of 20 M) over an order of magnitude larger than the k12/k11 ratios reported by Behnke et al., 1997; Bertram & Thornton, 2009; Roberts et al., 2009; and Ryder et al., 2015 (further details in section S5). Additionally, flow tube reactions of N2O5 uptake onto seawater mimics (Ryder et al., 2015) showed that both phenol and humic acid at low concentrations (<10 mM) could cause significant reductions in ϕ(ClNO2) relative to pure NaCl solutions, which may result from both a large k14 reaction rate constant and enhanced surface concentration of organics relative to chloride (Ryder et al., 2015). Combined, these past results suggest that even at low organic concentrations, additional competition reactions, generalized by R14, could effectively compete with R12 and decrease the ClNO2 production yield relative to that expected from Cl and water alone. These reactions may also lead to aerosol-phase NO3, organic nitrates, or HNO3, maintaining consistency with observational derivations. While AMS measurements of total nitrate during WINTER did show evidence for the presence of organic nitrates, the calculated inorganic-only nitrate (scaled to PILS-IC measurements; see Schroder et al., 2018) was consistently the largest fraction of total aerosol nitrate measured.

To examine whether there is evidence in the WINTER data to support competition reactions, an additional expression for ϕ(ClNO2) was derived from R9R12 and R14, shown in 6, assuming the H2ONO2+ intermediate is in steady state (see derivation in section S4). Rearranging this expression in 7, a plot of (ϕ(ClNO2)−1 – 1)*[Cl]/[H2O] against [Y]:[H2O] should yield a linear correlation with a slope of k14/k12 and intercept of k11/k12. The identity of Y is unknown, but it could include aqueous-phase species such as organics, halogens, and/or anions such as SO42− that could plausibly react with H2ONO2+. Previous studies on dilute (NH4)2SO4, (NH4)HSO4, and chloride containing solutions, however, have not shown a suppression in ϕ(ClNO2) relative to the parameterization (Roberts et al., 2009) and have indicated that at similar concentrations, Cl is more reactive toward H2ONO2+ than SO42− (Gaston & Thornton, 2016). To maintain consistency between the box model and other observational methods, possible competition reactions would also need to produce either particle or gas-phase nitrate through, for example, hydrolysis of the initial anion-nitronium product in R14.
urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0046(6)
urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0047(7)

Figure 7 shows the correlation between the left term in 7 (rearranged from 6) and molar ratios of (a) SO42−:H2O and (b) Org:H2O (assuming a constant organic molecular weight of 250 g/mol). An additional correlation with the Br reaction product, Br2 (e.g., Behnke et al., 1994; Schweitzer et al., 1998) (BrNO2 not present above instrument LOD), measured by the I-ToF-CIMS, was not statistically significant (not shown). The fit results in Figure 7 provide mixed evidence for the presence of a competition between organics, SO42−, and Cl via R14 during WINTER. The positive correlations are consistent with competition with Cl, with a rate constant (k14) 1.5–50 times larger than k12 (Figure 7). The negative fit intercepts, however, also indicate that this model for ϕ(ClNO2) is incorrect when using SO42− and total aerosol organics as species Y. It is possible, however, that the negative intercepts could result from multiple competition reactions of different rates (i.e., with various organic components) and/or additional processes that cause suppression.

Details are in the caption following the image
Correlation of the (ϕ(ClNO2)−1–1)*[Cl]/[H2O] product from 7 (using box model ϕ(ClNO2)) against aerosol the SO42−:H2O molar ratio in panels (a) and (b) and Org:H2O molar ratio in panels (c) and (d), calculated using AMS (a and c) and PILS (b and d) chloride measurements. Points with corresponding particulate chloride above instrument detection limits are shown in dark gray. Red lines are the linear fits for each correlation with fit equation provided in each figure. Slopes represent the k14/k12 ratio and intercepts represent the k11/k12 ratio. Dashed lines represent the same fits, holding the intercept constant at 0.002 (from Bertram & Thornton, 2009). Organic molarity was calculated by applying a constant molecular weight of 250 g/mol to AMS organic mass concentration measurements. Larger y values correspond to smaller values of box model ϕ(ClNO2).

Alternatively, the hypothesis of a competition reaction can be tested by fitting the [Y]:[Cl] ratio in 7 to WINTER box model ϕ(ClNO2) values. This method does not require knowledge of Y and estimates the [Y]:[Cl] ratio that would be required to explain the observed ϕ(ClNO2) values via R14 by using the k11/k12 ratio of 0.002 from Bertram and Thornton (2009) and a k14/k12 ratio of 1 (i.e., both k12 and k14 are near the diffusion limit). This method largely follows the work of Ryder et al. (2015) who required a molar ratio of at least 2 to explain their observed ϕ(ClNO2) values on ambient sea water samples, assuming k14/k12 = 1. The ratio required here to reproduce WINTER data ranged from 0 to >100, with a median of 6.0 and 4.2 for calculations with PILS and AMS chloride, respectively. For comparison, the median molar ratios of SO42−:Cl and Org:Cl during WINTER were between 7–25 and 2–11, respectively, but also with values exceeding 100.

Results in this section provide mixed evidence for the presence of a competition reaction between Cl and an additional reactive compound. The positive correlations between (ϕ(ClNO2)−1–1)*[Cl]/[H2O] and molar ratios of SO42−:H2O and Org:H2O are consistent with such reactions, but the intercepts that do not reproduce k11/k12 in 7 suggest either that the model is incorrect for sulfate and organics or that there are multiple reactions and/or additional processes contributing to the observed ϕ(ClNO2) suppression. Taking the k14/k12 ratio in 7 as 1, the nucleophile in question would require molar ratios in excess of 100 relative to Cl to explain the lowest ϕ(ClNO2) values. Many of the box model ϕ(ClNO2) values, however, could be reproduced with much more moderate molar ratios of ~6. Further laboratory studies focused on the aqueous kinetics of H2ONO2+ will be required to assess the extent to which a process such as this explains the difference between observed and parameterized ϕ(ClNO2) values.

4.3.3 Direct ClNO2 Loss

Lastly, direct loss of gas-phase ClNO2 could additionally reduce modeled ϕ(ClNO2) values relative to the parameterization. In the box model calculation of ϕ(ClNO2), values were derived by iteratively fitting the model output to gas-phase observations of N2O5 and ClNO2. This method is based on the assumption that ClNO2 formed from reaction R12 will efficiently evaporate to the gas-phase based on the low solubility of ClNO2 in water (KH = 4 × 10−2 M/atm; e.g., Frenzel et al., 1998 ; Roberts et al., 2008), where it is stable throughout the night. Additional direct loss mechanisms of ClNO2, independent from N2O5, would therefore serve to reduce the net ϕ(ClNO2) derived by the model. Possible direct loss mechanisms could include: (1) gas-phase ClNO2 loss through surface deposition and/or aerosol uptake and (2) direct aqueous-phase reaction of ClNO2 prior to evaporation.

Surface deposition and/or aerosol uptake of ClNO2 would serve to reduce the box model calculated ϕ(ClNO2) by reducing ambient gas-phase ClNO2 and the subsequently-derived ClNO2 production rate constant ( urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0048). The effect of ClNO2 loss from aerosol uptake is expected to be small as uptake coefficients have been measured on the order of 1 × 10−5 for dilute salt solutions (e.g., Frenzel et al., 1998; Schweitzer et al., 1998). Adjusting the box model-derived urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0049 values in 1 for loss associated with an uptake coefficient of this magnitude increased the median box model ϕ(ClNO2) value by 1%. The potential loss of ClNO2 through ocean surface deposition has been discussed previously in sections 3 and S2.2.1. Though box model simulations were limited to the RL, increased mixed-layer depths over the ocean allow for possible air-sea exchange of N2O5 and ClNO2 on WINTER flights over marine environments (Figure 1). While ocean emission of ClNO2 may be expected based on the positive water dependence of N2O5 uptake (McDuffie et al., 2018, and references therein) and typical ocean salinity (~0.55 M [Cl]), previous observations of N2O5 and ClNO2 from the Scripps Institution of Oceanography (SIO) pier by Kim et al. (2014) found a net depositional flux of both N2O5 and ClNO2 to the ocean surface. As previously discussed in section 3, adjusting the box model ( urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0050and urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0051) results for deposition of both N2O5 and ClNO2 could reduce, but not entirely eliminate the difference between the ϕ(ClNO2) parameterization and box model results (Figure S3). Combined, these results suggest that possible gas-phase ClNO2 loss through aerosol uptake and/or ocean surface deposition may contribute to the low box model ϕ(ClNO2) values, but are not the only cause.

Direct loss of aqueous-phase ClNO2 could also reduce the box model ϕ(ClNO2) values relative to the simple parameterization. This could occur through direct aqueous-phase reaction of ClNO2 (aq) with species X, as generalized in reaction R15. Though difficult to directly probe with WINTER field data, the possibility of direct ClNO2 reaction can be evaluated using the potential reaction products from R15 and associated variables. For example, previous laboratory studies have identified reaction mechanisms for R15 that form halogenated products such as Br2, BrNO2 (Fickert et al., 1998; Frenzel et al., 1998; Schweitzer et al., 1998, 1999), or Cl2, the latter of which is facilitated by particle acidity (Roberts et al., 2008). Three previous field studies with co-located ClNO2 and Cl2 observations from Colorado, California, and Calgary, however, did not consider heterogeneous ClNO2 chemistry a significant source of Cl2 due to weak ambient particle acidity (Mielke et al., 2011; Riedel et al., 2012, 2013). I-ToF-CIMS observations of BrNO2 did not exceed the instrument detection limit during WINTER (1 s, 1σ of 1 pptv) and no statistically significant correlations were found between WINTER ϕ(ClNO2) values and I-ToF-CIMS observations of Br2 or Cl2 (above their 1 s, 1σ detection limits of 0.5 and 0.4 pptv, respectively). In addition, a negative correlation (p < 0.05) was observed between particle acidity and Cl2, opposite of the expected trend from Roberts et al. (2008), despite high acidity calculated for aerosol during WINTER (pH ~ −2 to 3; Guo et al., 2016).
urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0052(R15)
Without further knowledge of the identity of species X and/or possible reaction products, the possibility of R15 can be evaluated using the ϕ(ClNO2) expression in 8, derived from reactions R9R12 and R15, assuming H2ONO2+ is in steady state, and that aqueous-phase ClNO2 is lost via R15 before it can partition to the gas-phase via R13 (derivation in section S4). Using 8, the k15[X] (s−1) product required to reproduce box model ϕ(ClNO2) values was calculated for each point using values of urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0053(0.002), urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0054 (29), and urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0055 expression (defined in section S4) from Bertram and Thornton (2009), along with estimates of aqueous-phase concentrations of N2O5 and ClNO2 from measured gas-phase mixing ratios and Henry's Law constants of 51 (unitless) (Fried et al., 1994) and 4 × 10−2 M/atm (Frenzel et al., 1998; Roberts et al., 2008), respectively. Derived k15[X] values suggest that reproduction of the box model values by invoking direct ClNO2 loss in 8 would require k15[X] products between 1 × 105 and 8 × 109 s−1 for both AMS and PILS chloride. Assuming a larger solubility for N2O5 of 5 M/atm (e.g., Griffiths et al., 2009; Mentel et al., 1999) would require even larger values of k15[X]. Based on these results, the largest difference between the ϕ(ClNO2) parameterization and box model (requiring the largest k15[X] value) would, therefore, require a reaction rate constant near the diffusion controlled limit (~1 × 109 M−1 s−1), or aqueous concentrations of [X] greater than 1 M. Median differences, however, could be reproduced with more moderate k15[X] values of 9 × 107 and 1 × 108 s−1 for calculations with AMS and PILS chloride, respectively.
urn:x-wiley:2169897X:media:jgrd55075:jgrd55075-math-0056(8)

Results in this section are consistent with the possibility that direct loss of gas- and/or aqueous-phase ClNO2 could contribute to some of the smaller differences found in WINTER data between the ϕ(ClNO2) parameterization and box model. Agreement between these ϕ(ClNO2) values improved when considering the possibility of surface deposition of both gas-phase ClNO2 and N2O5 (Figure S3), though box model median values remained lower than the parameterized equivalents. The addition of gas-phase loss through ClNO2 aerosol uptake only increased the median ϕ(ClNO2) by 1% when considering an uptake coefficient of 1 × 10−5. The possible identity of species X in R15 remains unknown. Low di-halogen concentrations do not provide evidence of direct loss through reactions with halogens, despite the highly acidic WINTER aerosol. Results from calculating k15[X] in 8 suggest that direct aqueous loss of ClNO2 via R15 would require concentrations of species [X] > 1 M and/or reaction rate constants near the diffusion-limited rate to reproduce the lowest box model ϕ(ClNO2) values. Alternatively, in the event that ClNO2 were to remain unreacted and/or trapped in the aerosol due to an organic coating or changes in solubility, the box model ϕ(ClNO2) values would also appear lower than the parameterization. The cause of a physical trapping mechanism is uncertain, however, and if not water dependent, was not elucidated by correlations between ϕ(ClNO2) differences and aerosol composition (Table S3). In addition, any trapped ClNO2 would be reported as particle chloride by the AMS and would require 88% of the measured chloride to be from trapped aqueous-phase ClNO2 in order to account for the difference between the median box model and parameterized ϕ(ClNO2) values. Without additional information about WINTER aerosol composition, morphology/viscosity, and/or other possible aqueous-phase ClNO2 reactions, the possibility of direct aqueous-phase ClNO2 loss during WINTER cannot be further evaluated.

5 Conclusions

A box model analysis of 9 night flights during the 2015 WINTER aircraft campaign derived 3,425 individual determinations of ϕ(ClNO2) with a median value of 0.138 (1σ: +0.050/−0.045) and a range from 0.003 to 1. Comparison of a subset of WINTER box model ϕ(ClNO2) values to those calculated with two other commonly used, data-based methods, showed agreement between their predicted median values, within the uncertainty of each method. In contrast, ϕ(ClNO2) values calculated from a laboratory-based parameterization predicted a median value over a factor of two larger than all other methods and outside the bounds of the combined uncertainties for two. When compared to all WINTER data, the ϕ(ClNO2) parameterization overpredicted ≥90% of the box model values for points both above and below instrument detection limits for particulate chloride. In addition, the box model median ϕ(ClNO2) value was 75–84% lower than the median calculated with the Bertram and Thornton (2009) parameterization, using both AMS and PILS particulate chloride measurements. When considering the combined uncertainties associated with aerosol chloride and water concentrations, the lower-limit estimates of the ϕ(ClNO2) parameterization remained larger than 73% of the box model results. Similarly, upper-limit estimates of the box model results could not reconcile the differences between the box model and current ϕ(ClNO2) parameterization. These results are qualitatively consistent with all previous studies that have compared field-derived and parameterization-predicted ϕ(ClNO2) values.

Physiochemical processes related to this observed difference were assessed using ambient observations of aerosol composition and mechanistic processes that have been discussed in previous laboratory and field-based literature. The observed difference between parameterized and box-modeled ϕ(ClNO2) values was most strongly correlated with calculated aerosol water, with differences decreasing with increases in aerosol water molarity, liquid water content, and RH. This trend was caused by the opposite water dependences predicted by the ϕ(ClNO2) parameterization (negative) and the box model results (positive) and was not driven by uncertainties in the aerosol water calculation. A positive correlation between aerosol water and ϕ(ClNO2) has only been reported in one other field study and may be related to the physical availability of chloride, though this hypothesis could not be confirmed with WINTER data. In addition, the relatively low correlation coefficients between ϕ(ClNO2) differences and aerosol water (≤ 0.53) indicate that multiple factors may cause the low box model values relative to parameterized results.

WINTER results are consistent with additional mechanistic processes contributing to the field-parameterized ϕ(ClNO2) differences, except for those associated with the lowest water concentrations where the greatest differences were observed. These mechanistic processes include aqueous-/gas-phase ClNO2 loss and/or competition reactions with additional reactive aerosol components and were tested by deriving updated expressions for ϕ(ClNO2) after appending additional reactions to the original aqueous-phase formation mechanism. By invoking an additional competition reaction between Cl and an aqueous-phase compound [Y], the Y/Cl molar ratio would need to be ~ 6 to explain many of the differences between parameterized and box model-derived ϕ(ClNO2) values, with the largest differences requiring ratios >100. Tests setting Y to equal either SO42− or total organics suggested that these particular species were not in direct competition with Cl via the applied model, or that there were multiple, overlapping processes leading to the observed ϕ(ClNO2) differences during WINTER. Similarly, loss of aqueous-phase ClNO2 by direct reaction with compound X could only reconcile the largest differences with a reaction rate constant near the diffusion limit or concentration of X greater than 1 M. In addition, loss of gas-phase ClNO2 through surface deposition or aerosol uptake could not explain the largest ϕ(ClNO2) differences. While WINTER data and box modeling results have provided valuable insights, further identification of mechanistic factors influencing ClNO2 formation will be required to develop a robust parameterization that can help improve model predictions of ClNO2 formation from N2O5 heterogeneous uptake and lead to a better understanding of the halogen influence on tropospheric chemistry.

Acknowledgments

The authors would like to thank the NSF-NCAR Research Aircraft Facility staff. We also thank Rebecca S. Hornbrook, Eric C. Apel, and Alan J. Hills for TOGA VOC data from the WINTER campaign and comments during the manuscript preparation process. We finally thank Viral Shah for insightful discussions that informed this analysis. E. E. M. and S. S. B. acknowledge support from the NOAA Atmospheric Chemistry, Climate and Carbon Cycle (AC4) Program. Funding for D.L.F. was supported by NSF award 1433358. J.A.T. and the University of Washington group was funded by NSF AGS award 1360745. J. C. S., P. C. J., and J. L. J. acknowledge grants NSF AGS award 1360834 and NASA NNX15AT96G. The National Center from Atmospheric Research (NCAR) is sponsored by the National Science Foundation (NSF). J. D. was supported by NSF AGS award 1456249. All data from the WINTER campaign are available at http://data.eol.ucar.edu/master_list/?project=WINTER. All referenced supplemental text, figures, and tables can be found in the supporting information. Code for the iterative box model can be found at https://esrl.noaa.gov/csd/groups/csd7/measurements/2015winter/pubs/.