Aircraft measurements during the Indirect and Semi-Direct Aerosol Campaign (ISDAC) in April 2008 are used to investigate factors influencing the microphysics and radiative properties of springtime Arctic clouds. The analysis is focused on low-level, liquid-dominated clouds in two separate regimes with respect to cloud and aerosol properties: single-layer stratocumulus with below-cloud aerosol concentrations (Na) less than 250 cm−3 (clean cases); and layered stratocumulus with Na > 500 cm−3 below cloud base, associated with a biomass burning aerosol (polluted cases). For each regime, vertical profiles through cloud are used to determine cloud microphysical and radiative properties. The polluted cases were correlated with warmer, geometrically thicker clouds, with higher droplet number concentrations (Nd), liquid water paths (LWP), optical depths (τ), and albedo (A) relative to clean cases. The mean cloud droplet effective radii (reff), however, were similar (5.7 μm) for both aerosol-cloud regimes. This discrepancy resulted mainly from the higher LWP of clouds in polluted cases, which can be explained by both meteorological (temperature, dynamics) and microphysical (precipitation inhibition) factors. Adiabatic parcel model simulations demonstrate that differences in droplet activation between the aerosol-cloud regimes may play a role, as the higher Na in polluted cases limits activation to larger and/or more hygroscopic particles. The observations and analysis presented here demonstrate the complex interactions among environmental conditions, aerosol, and the microphysics and radiative properties of Arctic clouds.
 Simulating the radiative properties and spatial extent of clouds in global climate models (GCMs) remains a challenge given uncertainties in the microphysical processes of cloud droplet formation and growth [e.g., Denman et al., 2007]. Atmospheric aerosol particles acting as cloud condensation nuclei (CCN) influence the microphysical properties of clouds. For a given liquid water content (LWC), increased aerosol particle number concentrations can result in the formation of more numerous cloud droplets of smaller size, increasing solar albedo; this is referred to as the first indirect effect [Twomey, 1977]. The cloud albedo may be enhanced further by a second indirect effect, in which the above changes in droplet number concentration and size by CCN suppress the formation of precipitation by reducing the efficiency of collision-coalescence, extending cloud lifetimes and increasing fractional cloud cover [Albrecht, 1989; Liou and Ou, 1989]. These indirect effects represent key climate system processes, particularly in the Arctic [Inoue et al., 2006; Vavrus, 2004], which has been shown to be warming almost twice as quickly as the rest of the planet [McBean et al., 2005].
 The formation of cloud droplets from CCN is predicated on the physicochemical properties of aerosol particles, their number concentration, and the ambient supersaturation [Pruppacher and Klett, 1997; Seinfeld and Pandis, 1998]. The size distribution and chemical composition of aerosol particles are the preeminent factors dictating CCN ability. The relative importance of these factors is not clear, as previous studies have demonstrated both size and composition as the predominant factors controlling CCN ability [Dusek et al., 2006; Leaitch et al., 2010; Quinn et al., 2008; Zelenyuk et al., 2010]. The mixing state of aerosol can also play a role; however, the distinction between internally and externally mixed particles as CCN becomes less significant if particles are sufficiently hygroscopic [Wang et al., 2010]. Comparisons of below-cloud aerosol particle number concentrations, Na, with in-cloud droplet concentrations, Nd, indicate two disparate regimes for the activation of particles larger than ∼0.1 μm [Leaitch et al., 1986; McFarquhar et al., 2011; Ramanathan et al., 2001]: in the first, Na ≈ Nd, and essentially all aerosol particles are activated; in the second, Na > Nd, and the competition for vapor limits activation to larger and/or more hygroscopic particles. The vapor supply for droplet activation and growth is increased by higher rates of cooling of air parcels and decreased by the rate at which water condenses on growing droplets; these processes govern the development of supersaturation in air parcels. In terms of the nucleation process, dynamics are most often represented by the vertical (updraft) velocity. Dynamics can also influence the cloud thickness Hc [Peng et al., 2002; Suzuki et al., 2010], which has implications for the cloud radiative properties, as discussed below.
 The effects of aerosol particles on cloud microphysical and radiative properties can be examined in terms of the cloud droplet effective radius (reff), cloud extinction coefficient (bext), optical thickness (τ), and albedo (A) for a particular region or aerosol concentration regime. The effective radius is defined as the ratio of the third moment of the droplet size distribution to the second moment [Hansen and Travis, 1974]:
where r is droplet radius and n(r) is the number distribution function. The effective radius can be used to compute the extinction coefficient as follows:
where ρw is the water density, LWC is the cloud liquid water content in g m−3, and Qext is the extinction efficiency, which is sufficiently close to 2 for wavelengths in the solar spectrum and particles the size of cloud droplets. The vertical integral of bext over the cloud depth Hc gives the optical thickness τ, from which cloud albedo A can be approximated by [Meador and Weaver, 1980]:
which assumes a scattering asymmetry factor (g) of 0.85.
 Analysis of aircraft and ground-based observations in the Arctic [Garrett et al., 2002, 2004], and aircraft observations in both the Arctic and midlatitudes [Peng et al., 2002], have shown that polluted conditions (higher Na) are correlated with higher Nd and smaller reff. Consistent with the first indirect effect, the latter study found larger τ and A for cases with higher Na; however, the resulting solar forcing is expected to be small in the Arctic, given the seasonal variability of aerosol loading, daylight hours, and solar zenith angle, as well as the high reflectivity of the surface [Garrett et al., 2002]. Instead, it is expected that the enhanced longwave emissivity of thin, low-level clouds with smaller reff represents a more significant aerosol indirect effect in the Arctic, resulting in net surface warming [Garrett and Zhao, 2006; Lubin and Vogelmann, 2006; Zhang et al., 1996]. Enhanced longwave emission can drive condensation via radiative cooling, which can, in turn, further increase the efficiency of thermal emission. This positive feedback loop provides a mechanism by which the vertical development of clouds can be accelerated under polluted aerosol conditions [Garrett et al., 2009].
 More vertically developed, or geometrically thicker, clouds have also been attributed to the precipitation inhibition resulting from more numerous droplets of smaller size under conditions of enhanced aerosol loading [L'Ecuyer et al., 2009; Pincus and Baker, 1994; Tietze et al., 2011]. These thicker clouds have higher liquid water paths (LWP: the cloud LWC integrated over the full cloud depth Hc). Conversely, under cleaner aerosol conditions, with a limited number of CCN, cloud droplets have been shown to grow readily to drizzle sizes and precipitate out, thereby reducing cloud water [Mauritsen et al., 2011]. While these studies offer evidence for the second aerosol indirect effect, additional studies have shown that the cloud effects of aerosols are modulated by a variety of microphysical and meteorological (environmental conditions, dynamics) factors [Ackerman et al., 2004; Durkee et al., 2000; Lu and Seinfeld, 2005; Stevens and Feingold, 2009]. Moreover, the aerosol properties and meteorological conditions typically co-vary [Ackerman et al., 2004], obscuring their relative importance in determining cloud microphysical and radiative properties.
 Understanding of aerosol indirect effects in Arctic clouds has been limited by this inherent complexity, as well as the number and scope of field observation campaigns in the region. The U.S. Department of Energy (DOE) Indirect and Semi-Direct Aerosol Campaign (ISDAC) was conducted in Alaska in April 2008 to provide a comprehensive data set for the assessment of aerosol-cloud relationships in Arctic spring [McFarquhar et al., 2011]. Here we present a specific subset of the cloud and aerosol data from ISDAC for two contrasting scenarios: first, single-layer stratocumulus on April 8, 26 and 27 with relatively low aerosol number concentrations (clean cases); and second, layered stratocumulus cloud impacted by aged biomass burning aerosol on April 19 and 20 with significantly higher particle loading (polluted cases). Cloud microphysical and radiative properties are taken from vertical profiles through liquid-dominated cloud in each scenario. Detailed descriptions of aerosol physicochemical properties pertinent to CCN activity, including the size-resolved chemical composition of individual particles, are also provided from observations below cloud base. Adiabatic parcel model simulations are used to investigate the relative roles of aerosol particle concentration, physicochemical properties, and updraft velocity in droplet activation for each scenario. The observations and model simulations are used to investigate aerosol effects on springtime Arctic clouds, including potential buffering effects associated with changing microphysics and meteorology.
 The measurements considered in this analysis were taken in the vicinity of Barrow, Alaska, during ISDAC. A comprehensive overview of this airborne field campaign, including a full description of the measurement capability and instrument payload, is provided elsewhere [McFarquhar et al., 2011]. Briefly, the National Research Council of Canada (NRC) Convair-580 aircraft was equipped with a suite of aerosol, cloud microphysics, and atmospheric state probes (Section 2.1) for a series of sorties in the Arctic troposphere. Flight maneuvers included vertical profiles, porpoising legs, and constant-altitude (horizontal) legs for characterizing the spatial and temporal variability of aerosol and cloud properties. In the present analysis, vertical profiles and porpoising legs through cloud are used for the characterization of cloud microphysical properties, primarily the droplet size distribution and LWC, from which the droplet effective radius and cloud radiative properties are calculated. Measurements in horizontal flight legs below-cloud are used for aerosol characterization. These measurements provide the observational basis for assessing the influence of aerosol concentration and physicochemical properties on droplet activation and growth in an adiabatic cloud parcel model (Section 2.2).
2.1. Aircraft Instrumentation
 Temperature and relative humidity (RH) were measured using a Rosemount 102 probe [Lawson and Cooper, 1990] and EG&G chilled mirror hygrometer, respectively. Vertical gust velocities were determined from a Rosemount 858 gust probe (see Section 4 for discussion of associated uncertainty), and cloud LWC was measured using a Particle Measurement Systems (PMS) King hot-wire probe [King et al., 1978] considered to be accurate to within approximately ±15% for the relatively low values of reff observed in this study [Biter et al., 1987; King et al., 1985; Strapp et al., 2003]. Cloud droplet number concentration (Nd) and size were measured using a Droplet Measurement Technologies (DMT) Cloud Droplet Probe (CDP) [Lance et al., 2010; McFarquhar et al., 2007] or PMS Forward-Scattering Spectrometer Probe (FSSP-100) [Knollenberg, 1976, 1981], which have size ranges of ∼2–50 μm and ∼3–45 μm (diameter), respectively. Uncertainties for Nd measured by these probes vary with airspeed and the size and concentration of droplets, but are estimated to be within about 20% [Dye and Baumgardner, 1984; Isaac and Schmidt, 2009; Lance et al., 2010]. When present, larger particles (droplets and/or ice crystals) were measured using PMS two-dimensional cloud (2D-C) and precipitation (2D-P) probes [Knollenberg, 1976], which cover diameters between 25–800 μm and 200–6400 μm, respectively. Although the accuracy of these probes is not well established (particularly for particles smaller than about 100 μm), due to a variety of issues such as time response, digitization, out-of-focus images, and artifacts due to ice particle shattering on probe forward surfaces, the measurements are used primarily to indicate the relative presence of ice particles in significant numbers, and thus, the analysis is not highly sensitive to these probe uncertainties.
 Aerosol particle number concentration and size distributions were provided by two complementary optical probes: a PMS Passive Cavity Aerosol Spectrometer Probe (PCASP-100X) [Liu et al., 1992] with a size (diameter) range of ∼0.12–3 μm and a PMS FSSP-300 that sizes particles from ∼0.3–20 μm in diameter [Dye et al., 1990]. Based on uncertainties in the sample flow, the uncertainty in Na from the PCASP is estimated to be within about 7%. Uncertainties for the FSSP-300 are associated primarily with the probe sample area; these uncertainties are within about 10–20% for 1 μm particles and increase with increasing particle size [Strapp et al., 1992].
 The size-distributed chemical composition and mixing state of aerosol particles were determined using a single particle mass spectrometer, SPLAT II [Zelenyuk et al., 2009, 2010]. Throughout the entire ISDAC campaign, SPLAT II was operated in two parallel data acquisition modes [Vaden et al., 2010a; Zelenyuk et al., 2010]. One mode was used to characterize the vacuum aerodynamic diameter (Dva) and composition of individual particles in the size range between 50 and 1000 nm. The second mode was used to measure number concentrations of particles larger than 100 nm, Dva size distributions, densities, and particle asphericity, which were used to obtain quantitative information on the size and composition of individual aerosol particles. Vacuum aerodynamic size distributions of aerosols with different compositions were converted to geometric diameter size distributions using their measured densities [Vaden et al., 2010b] for implementation in the adiabatic parcel model described in the following section.
2.2. Adiabatic Parcel Model
 The adiabatic cloud parcel model described by Leaitch et al.  was used to simulate the activation of aerosol particles measured below-cloud and subsequent growth of the nascent cloud droplets during adiabatic lifting. This model has been used previously in various iterations [Hayden et al., 2008; Peng et al., 2005; Shantz et al., 2003, 2008], with aqueous activity considered within the equilibrium framework of Köhler theory [e.g., Pruppacher and Klett, 1997]. The current version [Shantz et al., 2010] uses the κ-Köhler theory of Petters and Kreidenweis , which simplifies the description of solute properties (i.e., particle densities, molecular weights, activity coefficients) for multicomponent and unspeciated compounds. The chemical composition analysis from SPLAT II guides the selection of the hygroscopicity parameter, κ, for model simulations (Section 4). The surface tension of particles in the model is represented by the value for water (0.072 J m−2); it is assumed that κ captures sufficiently any reductions in surface tension by surfactant species that may influence the CCN ability of particles [Shantz et al., 2010].
 All model simulations are initialized at a relative humidity of 98% with respect to water, which has been shown to provide sufficient time for particles to reach their equilibrium sizes before the onset of supersaturation [Peng et al., 2005]. Temperature and pressure are initialized using observations below cloud base. A constant updraft velocity is used to describe the adiabatic ascent of parcels in simulations, with values estimated based on vertical gust velocity measurements in-cloud (Section 4). Based on the above studies employing this model, a water vapor condensation coefficient αc = 1.0 is used in all simulations.
 The size-resolved, individual particle composition measurements from SPLAT II provide particle size distributions for each internally mixed component in the below-cloud aerosol, which are represented by one or more lognormal modes in the model. Alternatively, the bulk chemical composition sampled over the below-cloud flight legs by SPLAT II can be input along with the average aerosol size distribution to simplify the analysis for complex multicomponent mixtures. The implementation of these different treatments of particle composition and size in the parcel model is discussed in Section 4.
3. Observations and Analysis
 A subset of six ISDAC flights near Barrow was selected for analysis: two flights from April 8 (flights 15 and 16), two from April 26 and 27 (flights 30 and 31), and two from April 19 and 20 (flights 25 and 26). The aerosol during the latter two flights was dominated by biomass burning particles, with particle number concentrations considerably higher than those for the other four flights. For each flight, cloud microphysical and radiative properties were determined from the vertical profiles through cloud, as described in Section 3.1. Aerosol physicochemical properties were taken from horizontal flight legs below cloud base on April 27 and 20 (Section 3.2); these properties are assumed to be representative of the cloud precursor aerosol for all four flights with lower Na (clean cases) and the two flights with higher Na (polluted cases), respectively.
3.1. Vertical Profiles
 A total of 19 vertical profiles through single-layer stratocumulus cloud in clean cases, and 12 profiles through layered stratocumulus cloud in polluted cases are used here. The locations of all vertical profiles and corresponding flight tracks in the vicinity of Barrow are shown in Figure 1. Cloud droplet data from the CDP were used for all profiles except those during flight 16, where CDP data were not available and FSSP-100 measurements were substituted. In general, the CDP data were preferred, due to some performance and calibration issues noted for the FSSP-100 during ISDAC (see Appendix A1 for details). The vertical extents of profiles considered in the analysis correspond to the range of altitudes for which Nd > 1 cm−3. All profiles were conducted below the 800 hPa level; hence, the measurements and analysis considered here are applicable to low-level Arctic stratocumulus clouds.
 Vertical profiles of temperature (as well as equivalent potential temperature, θe), LWC, and Nd are shown in Figure 2 for a clean case on April 27 and a polluted case on April 19, which are characteristic of lower and higher below-cloud aerosol concentration cases, respectively. The cloud in the polluted case is geometrically thicker and warmer (Figure 2a), with a mean temperature of 263.8 K compared to 260.6 K for the clean case. To some extent, these differences in temperature can explain the differences in LWC in Figure 2b, for which the mean values are 0.11 g m−3 and 0.07 g m−3 for the polluted and clean case, respectively. The lower mean LWC for the clean case can also be attributed to the increasing sub-adiabaticity observed toward cloud top, which suggests the depletion of cloud water by mixing or precipitation. Still, the LWC profile for the clean case is nearly adiabatic in the lower half of the cloud and the Nd (Figure 2c) are relatively uniform through most of the cloud depth. These observations are consistent with the adiabatic lifting of an air parcel with the maximum in the supersaturation occurring just above cloud base, where the droplet number concentration is defined. As the parcel continues to rise above the peak in the supersaturation, the droplets increase in size and LWC increases, but Nd remains effectively constant. The profiles of LWC and Nd for the polluted case (Figures 2b and 2c, respectively) exhibit similar adiabatic features through most of the cloud depth, with two exceptions near 50 m and 200 m. These exceptions are likely the result of the aircraft encountering horizontal inhomogeneities during its ascent.
 For all profiles considered in this analysis, reff was computed from the droplet size distribution at each 1-s point along the profile using equation (1). These reff values and measured LWC values along the profile were used to compute the extinction coefficient bext using equation (2), which was then integrated over the full cloud depth to determine the optical thickness τ. Using these values, the albedo A for each profile was calculated using equation (3). For further analysis, we assume that the mean reff represents the mean droplet size distribution for each profile. These values do not necessarily correspond to those from satellite retrievals (unless the cloud is vertically homogeneous), for which the uppermost cloud layers (which are more visible to the instrument) contribute most to reff [McFarquhar and Heymsfield, 1998]; however, comparison of mean reff over the full cloud depth with values averaged over the top one-third of cloud show agreement within 20%. Values of Nd were averaged over the full cloud depth, while characteristic values of Na were taken from PCASP measurements just below cloud base. The cloud liquid water path, LWP, was determined by integrating the measured LWC over the full cloud depth for each profile.
 Analysis of 2D-C and 2D-P measurements in- and below-cloud showed that light ice-phase precipitation was pervasive throughout the flights and profiles considered here, particularly the clean cases, with lower Na below-cloud; however, the ice crystal number concentrations were low (less than 0.001% of the average Nd for all cases), indicating that the clouds were composed predominantly of liquid droplets. Since homogeneous freezing of water or aqueous solution droplets is inefficient at temperatures above about −37°C [Pruppacher and Klett, 1997], these crystals were likely formed heterogeneously by interactions of aerosol particles (ice-forming nuclei, or IN) with droplets and/or water vapor. Vertical profiles of Nd and ice crystal concentrations (Ni) from the 2D-C and 2D-P are plotted in Figure 3a for a characteristic clean case on April 27. The ice crystals were predominantly dendrites, with some contribution from larger aggregates (Figure 3b).
 To assess the contribution of large, precipitating ice crystals to the total cloud water content (TWC), which represents the sum of LWC and ice water content (IWC), the latter was estimated from the 2D-C and 2D-P size spectra using the size-to-mass conversion of Mason . All profiles for which the estimated IWC exceeded 5% of TWC are indicated by black circles within the markers in Figure 1. Any additional profiles for which IWC exceeds 20% of TWC are not considered in this analysis. As noted above, ice-phase precipitation was more significant for clean cases, providing an additional mechanism for the depletion of liquid water and vapor [Lohmann, 2002; Prenni et al., 2007], as reflected by the sub-adiabatic LWC for the clean case in Figure 2b.
3.2. Below-Cloud Aerosol Properties
3.2.1. Aerosol Size Distributions
 The average aerosol particle size distributions from the PCASP and FSSP-300 are plotted in Figure 4 for clean (Figures 4a and 4c) and polluted (Figures 4b and 4d) cases from measurements below-cloud on April 27 and 20, respectively. The comparison and analysis of size data from these probes, which have different means of particle sampling, are discussed in Appendix A2. The aerosol number size distribution for the clean case (Figure 4a) shows the majority of particles in the accumulation mode with a number mean diameter of 0.2 μm, consistent with previous measurements of aged aerosol in Arctic spring [e.g., Radke et al., 1989; Shaw, 1984]. Similarly, the accumulation mode aerosol dominates the number size distribution for the polluted case (Figure 4b), but has a slightly larger number mode diameter of 0.24 μm. The larger aerosol for the biomass burning (haze) episode likely results from processing/coating of particles during transport from source regions in Kazakhstan and southern Russia [Warneke et al., 2009]. The lognormal distribution parameters for the bimodal fits to each case are given in Table 1.
Table 1. Lognormal Fitting Parameters for Representative Below-Cloud Aerosol Size Distributions for Clean and Polluted Cases
Aerosol Concentration Regime
Clean, Na < 250 cm−3
Polluted, Na > 500 cm−3
 Volume size distributions for the clean and polluted cases are plotted in Figures 4c and 4d, respectively. The accumulation mode for the clean case (Figure 4c) has a volume mean diameter of ∼0.30 μm, in excellent agreement with the size measurements of Warneke et al.  in aged boundary layer air during the same time period (April 2008) in Alaska. The measurements of Warneke et al.  from the biomass burning episode of April 20, 2008 also compare well with the 0.4 μm volume mode diameter for the polluted case (Figure 4d). In Figure 4d, the relatively large variations in the particle volume concentrations at sizes >1 μm result from the low numbers of particles counted. Although the accumulation mode dominates the number distributions, the coarse mode is most significant in the volume distribution of the clean case and comparable in the volume distribution of the polluted case.
 There is some discrepancy between the PCASP and FSSP-300 measurements in the accumulation mode (from ∼0.3–1 μm) of the number and volume size distributions for the polluted case in Figures 4b and 4d, respectively. This likely results from PCASP sampling losses for particles >0.5 μm, and perhaps also an instrumental anomaly that causes particle concentrations to decrease rapidly between the seventh and eighth PCASP size bins. As well, any errors in calculating the dehumidified FSSP-300 size distribution (Appendix A2) could contribute to the observed discrepancy. The implications of the associated uncertainty in the compiled size distributions and their lognormal fits for parcel model simulations of cloud droplet formation are discussed in Section 5.2.
3.2.2. Aerosol Composition Analysis
 The bulk and size-resolved aerosol particle composition from SPLAT II are shown in Figures 5 and 6 for the same clean and polluted distributions from Figure 4 (April 27 and 20, respectively). The bulk composition is illustrated by pie charts in Figures 5a and 6a that indicate the number fractions of the various components. Each component is classified according to its constituent species, which include sea salt, sulphate, organics, soot, and biomass burning (BB) products. When components have more than one constituent species, these constituents are internally mixed. Components with the same constituent classification (e.g., the four different internally mixed sulphate/BB components in Figure 6), are differentiated by the relative mass fractions of the constituent species. Size distributions for each component in the aerosol are represented by lognormal fits, both for clarity in Figures 5b and 6b, and for representation in the adiabatic parcel model (Section 4). The size-resolved composition analysis is limited to components with sufficient numbers of particles sampled. To account for differences in the numbers of particles sampled for each component, the size distributions in Figures 5b and 6b are normalized to 100 cm−3.
 The bulk composition analysis of below-cloud aerosol for the clean, lower Na case from April 27 (Figure 5a) has been presented previously [Zelenyuk et al., 2010], but is repeated here for the purposes of analysis and discussion. For this case, the aerosol is composed predominantly of organics and internal mixtures of sulphates and organics, with small contributions from BB components and sea salt, suggesting an aged aerosol with sulphate coating organics and BB particles during transport. The size-resolved composition for this case (Figure 5b) indicates that the organic and BB components have mean diameters in the vicinity of 0.2 μm, consistent with an aged aerosol, while the sulphate-organic and sea salt components have slightly smaller mode diameters. Overall, the SPLAT II analysis shows an accumulation mode aerosol comprised largely of organics and internal mixtures of sulphates and organics, in general agreement with complementary single-particle mass spectral measurements in the Alaskan Arctic boundary layer in spring 2008 [Warneke et al., 2010].
 Bulk composition analysis for the polluted case from April 20 (Figure 6a) shows that the aerosol is composed primarily of BB components, in accordance with forest fire and agricultural burning sources in Russia and eastern Eurasia [Warneke et al., 2009]. The individual BB components are internally mixed, with varying proportions of organics, soot, and sulphate, suggesting the coating of BB particles with other organics and sulphate during transport. The SPLAT II analysis also shows a significant contribution from an internal mixture of sulphate and non-BB organics, and a smaller contribution from sea salt. The size-resolved analysis for the polluted case (Figure 6b) indicates that, like the clean cases, the sulphate-containing components have slightly smaller mode diameters than the other components. This accumulation mode aerosol dominated by BB components is in excellent qualitative agreement with other single-particle mass spectral measurements in the same biomass plume presented by Warneke et al. .
4. Adiabatic Parcel Model Simulations
 Adiabatic cloud parcel model simulations were initiated as outlined in Section 2.2 for one clean case (lower Na) on April 27 and one polluted case (higher Na) on April 20. The size distribution and chemical composition of precursor aerosol below-cloud for these cases were presented in Sections 3.2.1 and 3.2.2, respectively. To verify the adiabaticity of clouds in these cases, the measured LWC was compared against adiabatic values determined by the parcel model, LWCadi, at given heights above cloud base. Clouds were considered to be adiabatic or near-adiabatic if the measured LWC was within 20% of the simulated LWCadi [Leaitch et al., 1986; Peng et al., 2005]. Larger deviations were observed in some instances, as LWC measurements in vertical profiles or porpoising legs are complicated by spatial variability resulting from the horizontal component of the aircraft passage through clouds [Peng et al., 2002], and LWCadi values from the parcel model do not account for the entrainment of dry air.
 The aerosol particle size distributions and composition were represented in two ways in the model simulations: first, using lognormal distributions fit to optical probe data (Section 3.2.1) and assuming an internally mixed aerosol; and second, using the size-resolved composition data from SPLAT II for an externally mixed aerosol (Section 3.2.2). For the first approach, the lognormal distribution parameters in Table 1 were used to represent the aerosol size distributions for each case. These distributions were limited to sizes smaller than 1 μm, to match the upper size limit of the aircraft aerosol inlet transmission efficiency, and hence, the upper size limit of the composition analysis from SPLAT II.
 The hygroscopicity of the internally mixed aerosol in the first approach is estimated based on the predominant components in Figures 5a and 6a. For both the clean and polluted cases, the aerosol composition was dominated by organic compounds, for which a value of κ ≈ 0.2 is a reasonable estimate [Chang et al., 2010; Rose et al., 2010]. Mixing of sulphate with the organics will increase the κ values, and based on the relative fractions of sulphate to organics, a possible value of 0.3 is estimated here, consistent with broader estimates of the hygroscopicity of continental aerosol [e.g., Pringle et al., 2010; Rose et al., 2010]. Accordingly, model runs in the present analysis were conducted with both κ = 0.2 and 0.3 for clean and polluted cases.
 For parcel model simulations using the size-resolved chemical composition from SPLAT II (Figures 5b and 6b), each lognormally distributed component was input separately, with an associated κ value determined from its mass-fractional composition and literature κ values for its constituent species [Chang et al., 2010; Petters and Kreidenweis, 2007; Rose et al., 2010]. All components were represented by one or more of the following constituent species: organics (κ = 0.2 for all organic components, including those from BB, following the discussion above), soot (κ = 0.001), sulphate (κ = 0.61), and sea salt (κ = 1.28). For instance, the ‘Sulfate/Organics’ component in Figure 5 is composed of 60% sulphate and 40% organics, for an overall κ ≈ 0.45. It is assumed that soot comprises 10% of the mass of the BB component, with organics comprising the remaining 90%. Aerosol particle size and composition are represented by the four lognormally distributed components for the clean case (Figure 5b), and nine such components for the polluted case (Figure 6b). Variations in κ are typically within about 30–35% [Dusek et al., 2010; Petters and Kreidenweis, 2007], depending on the measurements from which they were derived.
 The parcel model simulations also incorporated measurements obtained in-cloud during horizontal flight legs, which were selected over time intervals for which altitude, Nd, and the ambient conditions showed minimal variation. The updraft velocity used in simulations was estimated based on the standard deviation of Gaussian fits to probability distribution functions (PDFs) of measured vertical gust velocities (ω) in-cloud, σω [e.g., Chuang et al., 1997; Ghan et al., 1997]. From horizontal flight legs in-cloud on April 27 and April 20, σω was determined to be ∼38 cm s−1 and 40 cm s−1, respectively; the uncertainty in σω is estimated to be within ±5 cm s−1 [Leaitch et al., 1996]. Previous studies have shown that updraft velocities involved in droplet nucleation are represented well by fractions of σω; for example, 0.8 σω for midlatitude stratiform clouds [Fountoukis et al., 2007; Peng et al., 2005]. In the present analysis, sensitivity tests were conducted using updraft velocities of 20, 30, and 40 cm s−1, corresponding to values of approximately 0.5σω, 0.75σω, and σω. For simulations assuming an internally mixed aerosol, these tests were extended to the overall hygroscopicity, testing κ values of 0.2 and 0.3. The best estimates for updraft velocity and aerosol hygroscopicity were assessed by comparing simulated Nd from the model with values measured during the horizontal flight legs in-cloud, as discussed in Section 5.2.
5. Results and Discussion
5.1. Effects of Aerosol and Environmental Conditions on Cloud Microphysical and Radiative Properties
 Average values and standard deviations of cloud properties from the 19 vertical profiles in clean cases and 12 profiles in polluted cases are provided in Table 2, along with corresponding values for Na measured by the PCASP below cloud base (particles larger than ∼0.12 μm in diameter, the minimum size threshold of the PCASP; Section 2.1). As exemplified by the vertical profiles in Figure 2, the mean temperature is higher, and the geometric cloud thickness and LWC are larger, for the polluted cases relative to the clean cases. Accordingly, the mean LWP is larger for polluted cases. As well, the below cloud Na is approximately five times that determined for the clean cases. The separation between clean and polluted regimes is evident in the plots of Na and Nd as a function LWP in Figure 7. All clean cases are characterized by mean values of Na < 250 cm−3, while all polluted cases have mean Na > 500 cm−3. These findings are consistent with the Na = 300 cm−3 threshold between clean and polluted cases used by Peng et al. , which was based, in part, on PCASP measurements in Arctic stratocumulus. In Figure 7b, Nd ∼ 150–200 cm−3 marks the transition range between clouds in clean (lower Nd) and polluted (higher Nd) cases. The polluted points with lowest Nd in Figure 7b likely represent profiles near the edge of cloud, where mixing with dry air can act to reduce Nd by evaporation.
Table 2. Average Properties and Standard Deviations for All Clean and Polluted Cloud Cases
Aerosol number concentration (Na), cm−3
142 ± 42
756 ± 132
Droplet number concentration (Nd), cm−3
135 ± 34
304 ± 81
Temperature (T), °C
−12.9 ± 1.3
−7.5 ± 1.1
Liquid water content (LWC), g m−3
0.08 ± 0.02
0.16 ± 0.11
Cloud thickness (Hc), m
190 ± 43
296 ± 164
Liquid water path (LWP), g m−2
15.4 ± 6.4
63.5 ± 66.0
Droplet effective radius (reff), μm
5.7 ± 0.7
5.7 ± 1.2
Cloud optical thickness (τ)
3.91 ± 1.31
14.0 ± 12.6
Cloud albedo (A)
0.36 ± 0.07
0.56 ± 0.25
 For the clean cases, the average activated fraction of particles >0.12 μm, Nd/Na, is 0.99 ± 0.24, indicating that the major fraction of the below-cloud aerosol particles >0.12 μm were activated. For the polluted cases, on the other hand, the average activated fraction of particles >0.12 μm is 0.41 ± 0.13, owing to the competition for vapor among the more numerous aerosol particles, which limits activation to larger and/or more hygroscopic particles. (Note that the updraft velocities were similar for clean and polluted cases, as exemplified in Section 4.) The findings above are consistent with previous analysis of average Nd in-cloud against average Na below-cloud from the PCASP during ISDAC [McFarquhar et al., 2011], which shows a near 1:1 relationship between these variables for Na below about 250 cm−3 (almost all aerosol particles >0.12 μm activated), and a sub-linear relationship for higher Na (fewer particles activated due to vapor competition).
 For an invariant LWP, increased Nd is associated with a reduction in droplet size, represented here by reff; this is the premise of the first indirect effect [Twomey, 1977]. From Table 2 and Figure 7, however, LWP is larger for polluted cases, and despite significantly higher Na, the mean reff of 5.7 μm is the same as that for the clean cases. Previous analyses based on aircraft and ground-based measurements [Garrett et al., 2002, 2004; Peng et al., 2002] have shown higher LWP or LWC and higher Nd for larger aerosol particle loadings, but smaller reff. The similarity of reff for polluted and clean cases in the present work is due to the higher and lower LWP in these cases, respectively. The mean reff from each profile is plotted as a function of the associated LWP in Figure 8. There is no clear size threshold between the two aerosol concentration regimes, such as that observed at reff ≈ 6 μm by Peng et al. . There is, however, a clear divergence of the points for polluted cases when reff is larger than about 6 μm, corresponding to the largest LWP values (greater than about 50 g m−2). This suggests that the larger reff for these points is related to the significantly higher LWP. Over the common LWP range below 50 g m−2, the mean reff for polluted cases falls to 4.8 μm (the value for clean cases is still 5.7 μm, as all clean points have LWP < 50 g m−2), implying a first indirect effect; however, the number of polluted cases in this LWP range is limited.
 The larger LWP of polluted cases relative to clean cases can be attributed to both meteorological and microphysical factors. The correlation of polluted and clean aerosol conditions with warmer and colder clouds, respectively, suggests that the environmental conditions play a role. Similarly, the larger geometric thicknesses of clouds in polluted cases may be due, in part, to differences in local meteorology (temperature inversion at higher altitude, stronger updrafts). These thicker, more vertically developed clouds may also be explained in terms of the positive feedback loop proposed by Garrett et al. , in which the increased emissivity of low-level clouds in the presence of higher Na drives the condensational formation and growth of more droplets in the vertical domain. Another viable explanation is that, despite the significantly larger LWP for polluted cases, the higher Nd for these cases may keep droplet sizes sufficiently small to inhibit the formation of precipitation by collision-coalescence. This is supported by the vertical profiles of reff for polluted cases, such as that for April 20 2008 in Figure 9b, which indicate that reff does not exceed the 10–14 μm threshold size above which drizzle formation becomes significant [e.g., Gerber, 1996; Twohy et al., 2005; vanZanten et al., 2005; Yum and Hudson, 2002]. Using the relationship between Nd and reff [e.g., Durkee et al., 2000], it is estimated that the latter would exceed the 10 μm drizzle threshold if Nd was on the order of 150 cm−3 (the concentration observed for the clean case in Figure 9a), consistent with the suppression of precipitation under more polluted aerosol conditions.
 Correlations between regions with higher Na and clouds with higher LWP and/or larger geometric thicknesses have been observed previously [L'Ecuyer et al., 2009; Pincus and Baker, 1994; Tietze et al., 2011], and attributed primarily to the inhibition of precipitation associated with the second indirect effect. Also consistent with the findings of the present study, analysis by Tietze et al.  showed that LWP in Arctic low-level liquid clouds was relatively more sensitive to pollution plumes than reff, and suggested that the changes in LWP predominate the indirect effect. It should be noted, however, that other studies have shown no conclusive evidence for increased LWP in polluted environments [Avey et al., 2007].
 Precipitation may be a factor in the lower LWP for clean cases by reducing cloud liquid water and serving as a removal mechanism for larger droplets, thereby reducing reff. Vertical profiles of reff for clean cases, such as that for April 27 2008 in Figure 9a, show values smaller than the ∼10–14 μm drizzle threshold; however, this does not preclude the possibility that larger droplets may have already drizzled out, as observed for cases in clean aerosol conditions by Mauritsen et al. . The formation and growth of ice-phase precipitation, which was relatively more abundant for clean cases, would also deplete the cloud liquid water. Heterogeneous freezing of, or on, ice-forming nuclei and subsequent crystal growth and riming would reduce the available water vapor supply for condensation on droplets, limiting the LWC. Ice crystal growth may also occur at the expense of the droplets present; at a given subfreezing temperature, the saturation vapor pressure above ice crystals is lower than that above water droplets, favoring diffusion from the latter to the former through the Wegener-Bergeron-Findeisen process [e.g., Korolev, 2007], and thereby reducing the LWC. In addition, the higher Na in polluted cases may inhibit riming and associated ice crystal growth, limiting ice-phase precipitation and providing an additional avenue for increasing cloud LWP [Morrison et al., 2008].
 The role of precipitation in the observed LWP differences between clean and polluted cases can be estimated using characteristic precipitation rates in the Arctic. Analysis of U.S. National Weather Service data from 2000 to 2004 showed precipitation rates ranging from about 0.04–0.4 mm/day at Barrow in April [Zhao and Garrett, 2008]. Considering the average LWP values in Table 2, there is a difference of ∼48 g/m2 between clouds in clean and polluted cases. For an area of 1 m2, the precipitation rates above correspond to decreases of 1.67–16.7 g of liquid water per hour. At the higher limit, this would deplete 48 g of cloud liquid water in about three hours; hence, precipitation rates of 0.4 mm/day could explain the observed differences in LWP. At the lower limit, on the other hand, it would take over 28 h to deplete an equivalent water mass, and so precipitation would account for a much smaller fraction of the observed LWP differences.
 Supercooled liquid clouds precipitating ice, such as those considered in the present study, are frequently observed in the Arctic [Curry et al., 1996; Intrieri et al., 2002; McFarquhar and Cober, 2004; Shupe et al., 2006], with lifetimes predicated on the hydrometeor properties and environmental conditions [Korolev and Field, 2008; Korolev, 2008; Rauber and Tokay, 1991], and with radiative properties controlled largely by liquid water [McFarquhar and Cober, 2004; Shupe and Intrieri, 2004]. Considering the average cloud radiative properties for clean and polluted cases, the optical thickness τ and albedo A are plotted as functions of LWP in Figures 10a and 10b, respectively (note that the LWP range in Figure 10a is limited to values below 100 g m−2 to facilitate comparison between the clean and polluted data points). There is a general linear correlation between τ and LWP in Figure 10a. Given the narrow range of reff values for clean and polluted cases (exemplified by mean values in Figure 8 and representative vertical profiles in Figure 9), the bext from equation (2) is effectively a linear function of LWC. This relationship extends to τ and LWP, which are the vertical integrals of bext and LWC, respectively. In Figure 10b, the cloud albedo A, which is computed from the integrated τ (equation (3)), increases sharply for LWP < 10 g m−2 and gradually levels off at higher LWP, as more highly attenuating clouds become less sensitive to changes in LWP. Over the common LWP range <50 g m−2, both cloud optical thickness and albedo appear to be slightly higher for the polluted cases with higher Na and Nd, suggesting the presence of more numerous droplets of smaller size associated with the first indirect effect. This is supported by the lower average reff for polluted cases over this LWP range noted above; however, a student's t-test indicates that the differences between the clean and polluted points are not statistically significant at the 80% confidence interval. It should be noted that additional statistical tests indicated that over the full LWP range, the differences in cloud radiative properties were significant at the 95% confidence interval.
5.2. Droplet Activation in Clean and Polluted Aerosol Regimes
 The relative roles of particle size, composition, mixing state, and updraft velocity in droplet activation for the clean case from April 27 and polluted case from April 20 were assessed through sensitivity analysis using the adiabatic parcel model, as outlined in Section 4. The maximum Nd for each model run was compared against the 95th percentile of Nd measured in the horizontal flight legs in-cloud, which were 215 cm−3 and 458 cm−3 for the clean and polluted cases, respectively. The maximum values are compared in each case (the 95th percentile is used for measured values to avoid any outliers) because the model simulations do not consider the influence of entrainment on Nd. The results of the sensitivity analysis are presented in Tables 3a and 3b. The best representations of aerosol physicochemical properties and updraft velocity are assessed in terms of the minimum percentage difference between the measured and simulated Nd values.
Table 3a. Results From Parcel Model Sensitivity Analysis for Clean, Lower Na Case on April 27 2008a
ω, cm s−1
Representations of aerosol particle mixing state, updraft velocity, and for internal mixtures, overall hygroscopicity, in the parcel model are assessed in terms of the percentage difference, % Diff, between simulated Nd and measured values, which is 215 cm−3 for the clean case.
Table 3b. Results From Parcel Model Sensitivity Analysis for Polluted, Higher Na Case on April 20 2008a
ω, cm s−1
Representations of aerosol particle mixing state, updraft velocity, and for internal mixtures, overall hygroscopicity, in the parcel model are assessed in terms of the percentage difference, % Diff, between simulated Nd and measured values, which is 458 cm−3 for the polluted case.
 The analysis for the clean, lower Na case in Table 3a indicates that when an internally mixed aerosol is assumed, its hygroscopicity can be represented equally well by κ = 0.2 or 0.3 in model simulations, and that simulations with updrafts between 30–40 cm s−1 give the best agreement with the measured Nd. The same range of updraft velocities minimizes the difference between measured and modeled Nd when the aerosol composition is externally mixed. The similarity of model results for the internally and externally mixed aerosol suggest that the representation of mixing state is not so important in this case, provided the aerosol hygroscopicity and size distribution are represented appropriately in the model. Similar results are observed for the polluted, higher Na case (Table 3b); however, these results are more sensitive to the updraft velocity in both internally and externally mixed scenarios, on account of the lower activated fraction. In each of these scenarios, updrafts of 20 cm s−1 provide the best agreement between the simulations and measurements.
 Based on the sensitivity analysis, differences in composition and/or mixing state between the aerosol-cloud regimes cannot account for the differences in activation described in Section 5.1. The roles of aerosol particle concentration, size distribution, and updraft velocity in droplet activation are considered further with respect to the time series of supersaturation and Nd from the model simulations producing the best agreement with measurements (minimum percentage difference in Tables 3a and 3b). In Figure 11a, the peak supersaturation in simulations for the clean case decreases in magnitude as the updraft velocity decreases. It is evident that the maximum Nd for each model run in Figure 11b is attained shortly after the peak supersaturation is reached in Figure 11a, indicating that all aerosols that are going to be activated have been activated. Beyond these peaks in the Nd time series, growth of the nascent droplets dominates the vapor supply at the expense of further activation.
 The higher peak supersaturation in simulations for the clean case allows for the activation of smaller particles in the size distribution, consistent with the near-unity activated fraction of particles for clean cases (Section 5.1). Hence, for clean cases, most of the aerosol particle size distribution is activated. For polluted cases, on the other hand, the peak supersaturation is lower, due to the slower updraft velocity and the competition for vapor among the more numerous aerosol particles. The particle size distribution and composition may also play a role, as larger and/or more hygroscopic particles are activated preferentially, with ramifications for the available vapor supply. Accordingly, the activated fraction is lower for polluted cases (Section 5.1), and smaller and/or less hygroscopic particles are not activated. The slightly larger mode diameters in polluted cases (Section 3.2.1 and Figure 4) and preferential activation of larger particles may shift the initial droplet size distributions to larger sizes relative to those for cleaner cases. Subsequent droplet growth, and hence, any precipitation formation preceding the measurements in-cloud, will be determined by a combination of cloud microphysics, environmental conditions, and dynamics. Model simulations in this regard are beyond the scope of the present study. Here we can only point out that the differences in droplet activation between the clean and polluted regimes may play a role in the complex aerosol-cloud-precipitation interactions in low-level Arctic stratocumulus clouds.
 The above discussion is applicable to aerosol size distributions represented by the bimodal lognormal fits in Figure 4 and associated parameters in Table 1. The implications of uncertainty in these size distributions, notably in the accumulation mode for the polluted case (Section 3.2.1), was investigated through a series of parcel model simulations using a lognormal fit to the PCASP data only. The accumulation mode from this fit was represented by a smaller standard deviation (σ = 1.50) relative to the compiled fit from the PCASP and FSSP-300 in Table 1. The results (not shown) indicated that for both κ = 0.2 and 0.3, updrafts of 20 cm s−1 gave simulated Nd within 10% of the measured values. Hence, both representations of the particle size distribution for the polluted case gave similar model results, with computed Nd within the measurement uncertainty.
 Aircraft measurements during ISDAC were used to investigate factors influencing the properties of springtime, liquid-dominated Arctic clouds. Vertical profiles through single-layer stratocumulus cloud on April 8, 26, and 27 (clean cases), and through layered stratocumulus cloud during a biomass burning episode on April 19 and 20 (polluted cases) were analyzed to determine average cloud microphysical and radiative properties. The 19 clean cases were characterized by Na < 250 cm−3 below cloud base and an average activated fraction of 0.99 ± 0.24, indicating that almost all aerosol particles below cloud larger than 0.12 μm were activated to form droplets. For the 12 polluted cases, Na > 500 cm−3 below cloud, and the average activated fraction of 0.41 ± 0.13 was lower due to the competition for vapor among the more numerous particles. The polluted cases were correlated with higher Nd, LWP, τ, and A compared to the clean cases; however, average reff values of 5.7 μm were observed for both aerosol-cloud regimes.
 The similarity of reff for cases with significantly different Na and Nd can be attributed to both meteorological and microphysical factors, with associated differences in LWP playing a dominant role. The polluted cases were correlated with warmer, geometrically thicker clouds, resulting in higher LWP relative to clean cases. Differences in the environmental conditions (dynamics, temperature inversion height) could explain the correlation between polluted cases with higher Na and thicker clouds with higher Nd and LWP; however, these correlations are also consistent with the inhibition of precipitation associated with the second indirect effect. The latter possibility is supported by vertical profiles of reff, which indicate that droplet sizes do not exceed the drizzle threshold under polluted aerosol conditions. More abundant ice-phase precipitation would also factor into the lower LWP observed for clean cases. As well, differences in droplet activation may play a role, as adiabatic parcel model simulations show preferential activation of larger and/or more hygroscopic particles in polluted cases, with implications for the cloud droplet size distributions and precipitation.
 These findings illustrate the complex correlations among aerosol, cloud microphysics, cloud radiative properties, and environmental conditions in low-level, liquid-dominated Arctic stratocumulus. The observed correlations, based on six project flights during ISDAC, cannot be used to infer causal relationships, as it is difficult to disentangle the relative contributions of aerosol and meteorology. Additional observational studies are required to extend these findings beyond the cases examined here, and to further investigate aerosol indirect effects in Arctic clouds.
Appendix A:: Technical Details
A1. Cloud Probe Intercomparison and Calibration
 Both a CDP and FSSP-100 were used to measure cloud droplet concentration and size during ISDAC; however, CDP measurements were preferred in general, as noted in Section 3.1. FSSP-100 bead calibrations and investigations of probe diagnostic parameters revealed that this probe was under-sizing particles. This was corrected by calculating new channel diameters for this probe based on the calibrations, with a maximum correction of about 20% at 10 μm. There was also a larger-than-usual broadening of the FSSP-100 spectra during ISDAC due to poor performance of the velocity averaging module, which cannot be corrected. However, the FSSP-100 measurements still provide reasonable Nd and bulk parameters such as LWC for all flights. Comparison of Nd values from the CDP and FSSP-100 showed good general agreement, with typical differences <20% attributed to uncertainties in the probe sample volumes. In addition, comparison of LWC derived from the CDP and diameter-corrected FSSP-100 data with the bulk values from the King probe also showed agreement, on average, to within ∼20%.
A2. Compiling Aerosol Size Distributions From Optical Probe Data
 Particles entering the PCASP are subject to multiple stages of heating prior to sizing, which have been shown to effectively dry aqueous particles smaller than about 10 μm [Strapp et al., 1992]. The FSSP-300, on the other hand, measures ambient (humidified) particles, and is subject to increasing sizing errors with diameter because of changes in the refractive index due to hydration and the associated uncertainties in the Mie scattering response [Strapp et al., 1992]. To reconcile the sampling of dry versus humidified particles, the dry size distribution was computed from the humidified FSSP-300 distribution using κ-Köhler theory [Petters and Kreidenweis, 2007]:
where Dd is the dry bin diameter, D is the humidified bin diameter, R is the ideal gas constant, Mw is the molecular weight of water, σs/a is the surface tension of the solution/air interface (assumed to be that of water, 0.072 J m−2), and T is the measured temperature. The saturation ratio S is computed from the measured RH with respect to water as S = RH/100 (not valid for very high RH or supersaturated conditions). The hygroscopicity parameter κ is estimated based on the SPLAT II composition analysis (see Section 3.2.2 for analysis and Section 4 for a discussion of κ estimates and uncertainties). In addition, any small modes in the FSSP-300 size distribution between ∼1–3 μm are averaged to obtain a single point, in order to mitigate the influence of apparent Mie scattering errors. Bimodal lognormal distributions are then fit to the composite PCASP and dry-computed FSSP-300 size data.
 This work was supported by the Office of Science (BER), U.S. Department of Energy, grant DE-FG02-09ER64768. The authors thank instrument PIs and support staff from Environment Canada, the U.S. Department of Energy, and the National Research Council of Canada Flight Research Laboratory for their invaluable contributions before, during, and after ISDAC. Special thanks go to Mohammed Wasey, Rob Reed, and Ka Sung for technical support, and to Mark Couture for numerous helpful discussions regarding data representation and analysis. SPLAT II was developed with the support of the Office of Basic Energy Sciences, Division of Chemical Sciences, Geosciences, and Biosciences and the Environmental Molecular Sciences Laboratory, a national scientific user facility sponsored by the DOE's OBER at Pacific Northwest National Laboratory (PNNL). PNNL is operated by the U.S. DOE by Battelle Memorial Institute under contract DE-AC06-76RL0 1830. The authors would also like to thank Xue Zheng, Bruce Albrecht, and two anonymous reviewers for comments that helped to improve this manuscript.