Modeling aerosols and their interactions with shallow cumuli during the 2007 CHAPS field study


Corresponding author: M. Shrivastava, Atmospheric Sciences and Global Change Division, Pacific Northwest National Laboratory, Richland, WA 99352, USA. (


[1] The Weather Research and Forecasting model coupled with chemistry, using 2 km horizontal grid spacing, is used to simulate two important relationships between aerosols and clouds in the vicinity of Oklahoma City during the June 2007 Cumulus Humilis Aerosol Processing Study (CHAPS). First, the model reproduces the trends of higher nitrate volume fractions in cloud droplet residuals compared to interstitial nonactivated aerosols, as measured using an aerosol mass spectrometer. Comparing simulations with cloud chemistry turned on and off, we show that nitric acid vapor uptake by cloud droplets explains the higher nitrate content of cloud droplet residuals. Second, the model reasonably represents the observations of the first aerosol indirect effect where pollutants in the vicinity of Oklahoma City increase cloud droplet number concentrations and decrease the droplet effective radius. In addition, as documented using an offline optical code, simulated aerosol optical properties depend on several compensating effects including aerosol water content, size-resolved chemical composition, and refractory index of various particle chemical species. All of our four sensitivity test cases clearly show an increase in simulated absorption and a decrease in single scattering albedo within the Oklahoma City plume relative to conditions outside the plume. While previous studies have often focused on cloud–aerosol interactions in stratiform and deep convective clouds, this study highlights the ability of regional-scale models to represent some of the important aspects of cloud–aerosol interactions associated with fields of short-lived shallow cumuli.

1 Introduction

[2] The effect of anthropogenic aerosols on clouds is one of the largest uncertainties in predicting human impact on climate change [Forster et al., 2007]. Anthropogenic particles may act as cloud condensation nuclei (CCN), influencing the microphysical and dynamical properties of clouds. Clouds are multiphase systems in which gaseous compounds, interstitial aerosols, and liquid droplets or solid crystals undergo complex physical and chemical processes. To examine the interactions between clouds and aerosols, coincident measurements of chemical composition of both activated (liquid droplets) and interstitial aerosols, their optical properties, and cloud microphysical properties are needed.

[3] Airborne platforms are one way to obtain the necessary observations. A Counterflow Virtual Impactor (CVI) inlet, which inertially separates large cloud droplets from smaller interstitial aerosols, can be used to selectively sample only cloud droplets. Applications of a CVI have been reported in a number of previous field studies [Hayden et al., 2008; Noone et al., 1993; Sellegri et al., 2003a, 2003b; Twohy and Poellot, 2005]. For example, Sellegri et al. [2003a] found that about 90% of inorganic species were efficiently scavenged in clouds at a mountainous location in France regardless of particle size. In contrast, carbonaceous species contributed 93% of the interstitial particle composition and just 21% of the cloud droplet composition. However, numerous other studies have shown evidence of organic aerosols (OA) acting as CCN in marine stratocumulus clouds [De Bock et al., 2000; Novakov and Penner, 1993; Russell et al., 2000; Twohy et al., 2005]. These studies suggest that interactions of clouds and aerosols could vary spatially and temporally, requiring detailed investigation of the various processes in different locations.

[4] Several studies have looked at relationships between clouds and aerosols near large or very dirty cities [e.g., Alkezweeny et al., 1993; Lu et al., 2008; Sorooshian et al., 2007], but few studies have investigated cloud–aerosol interactions near moderate size cities that are common in North America. In addition, previous studies have focused on studying aerosol cloud relationships within stratiform or deep convective clouds [e.g., Russell et al., 2000; Yang et al., 2011], but conditions within short-lived shallow cumuli may differ due to different updraft velocities, supersaturations, and spatial dimensions. During the U.S. Department of Energy (DOE) supported Cumulus Humilis Aerosol Processing Study (CHAPS), a suite of airborne instruments measured aerosol and shallow cumuli cloud properties near Oklahoma City, Oklahoma [Berg et al., 2009]. To the best of our knowledge, this was the first study in which an Aerodyne aerosol mass spectrometer [AMS; Jayne et al., 2000] was deployed on board the DOE Gulfstream-1 (G1) research aircraft configured to sample behind either the CVI inlet allowing in-situ investigation of the chemical composition of dried cloud droplet residuals, or an isokinetic inlet allowing analysis of the interstitial nonactivated particles.

[5] The CHAPS aerosol and cloud measurements provide a unique opportunity to evaluate the ability of coupled meteorology–chemistry models to simulate aerosol, clouds, and their interactions. The goals of this study are: (1) evaluate the ability of the Weather Research and Forecasting model coupled with chemistry (WRF-Chem) community model to predict the chemical composition and optical properties of aerosols measured in the vicinity of Oklahoma City during CHAPS; (2) examine field observations and model predictions to investigate how cloud processing of aerosols affects the differences in chemical composition of nonactivated interstitial and activated aerosols; (3) analyze the model simulations to determine how aerosols emitted in the vicinity of Oklahoma City affect cloud droplet number concentrations (CDNC) and cloud droplet effective radius (reff) within shallow cumuli; (4) establish a baseline modeling framework for future work parameterizing subgrid-scale cloud–aerosol interactions.

2 Model Description and Simulation Design

[6] Version 3.2.1 of the WRF-Chem community model [Fast et al., 2006; Grell et al., 2005] is used to simulate the atmospheric conditions between 18 and 25 June 2007 for a region that includes the area of CHAPS. Trace gases, aerosols, and clouds are simulated simultaneously with meteorology [Grell et al., 2005]. Table 1 summarizes the WRF-Chem model configuration, and Figure 1a shows the model domain used in this study. A nested grid configuration is used with an outer 10 km grid spacing domain covering the central U.S. and the inner 2 km grid spacing domain, shown by the green rectangle in Figure 1a, covering a 242 × 242 km area surrounding Oklahoma City. The Global Forecast System (GFS) model provides the meteorological initial and boundary conditions. For the coupled cloud–aerosol–meteorology simulations, the WRF meteorology is reinitialized every 24 h at 00 UTC, and the model is run for 36 h. The first 12 h of each simulation are not analyzed to allow sufficient time for model spin up of the meteorology, yielding continuous 24 h long simulation blocks used for analysis. Running WRF-Chem with reinitialized meteorology is found to result in better cloud simulations in terms of both cloud base height and their spatial extent when compared to satellite observations of cloud cover as discussed below. While boundary conditions over the outer domain are obtained from the Model for Ozone and Related chemical Tracers (MOZART) global simulation of trace gases and aerosols [Emmons et al., 2010], the initial conditions of trace gases and aerosols are taken from the corresponding hour in the previous WRF simulation. CDNC are reinitialized to zero at each restart, and the cloud-borne aerosol species in each size bin are resuspended to interstitial aerosol species at the beginning of each meteorological restart. Below, we provide a brief summary of main WRF-Chem modules relevant to this study.

Table 1. WRF-Chem Model Configuration for Clouds, Gas-Phase Photochemistry, and Aerosols
WRF domains10 km resolution outer domain covering the central U.S.
2 km resolution nested domain covering 242 km around Oklahoma City: Feedback On
Simulation period18–27 June 2007
Boundary layerMellor-Yamada-Janjic TKE [Janjic, 1990]
SurfaceUnified Noah land surface model [Chen et al., 1996]
Cloud microphysicsMorrison (two moments) [Morrison et al., 2005; Morrison et al., 2009]
(Both outer and nested domains)
CumulusKain–Fritsch (new Eta) scheme
(Only outer 10 km domain) [Kain, 2004; Kain and Fritsch, 1990]
RadiationRRTMG scheme (both longwave and shortwave)
Gas-phase photochemistrySAPRC-99
Aerosol chemistryMOSAIC for inorganic aerosols
Simplified volatility basis set (VBS) for organic aerosols [Shrivastava et al., 2011]
Boundary conditionsGFS and MOZART model
Figure 1.

(a) WRF-Chem modeling domain showing model predicted sum of column burdens of cloud water, snow, graupel, ice, and rain on 25 June 2007 at 1400 UTC. The green rectangle shows the inner nested grid. (b) WRF predicted cloud fraction averaged over boxes 0.5° by 0.5° (colors) over the inner domain and the GOES 12 reflectivity (grayscale) as described in the text.

2.1 Gas-Phase Chemistry

[7] Gas-phase chemistry in this study is based on the Statewide Air Pollution Research Center (SAPRC-99) mechanism [Carter, 2010], which includes 211 reactions of 56 gases and 18 free radicals. This mechanism is updated to include gas-phase photochemical oxidation of gas-phase organic species to form secondary OA (SOA), which is often the dominant component of fine particulate matter as suggested by Zhang et al. [2007]. We include SOA formed due to oxidation of semivolatile and intermediate volatility organic compounds (S/IVOC) emitted from anthropogenic and biomass burning sources (SI-SOA) and traditional SOA (V-SOA) formed due to oxidation of volatile organic compounds (VOC) precursors from anthropogenic and biogenic emissions [Shrivastava et al., 2011]. The gas-phase chemistry equations are calculated using the Kinetic Pre-Processor within WRF-Chem [Damian et al., 2002].

2.2 Aerosol Processes

[8] The condensation of low-volatility gases (H2SO4 and CH3SO3H) and the dynamic partitioning of semivolatile inorganic gases (HNO3, HCl, and NH3) to size-distributed liquid, mixed-phase, and solid atmospheric aerosols are represented by the Model For Simulating Aerosol Interaction and Chemistry (MOSAIC) aerosol module [Zaveri et al., 2008]. Aerosol species in MOSAIC include sulfate, nitrate, ammonium, sodium, chloride, calcium, carbonate, other inorganics (OIN), elemental carbon (EC), organic matter, and aerosol water. Aerosols are assumed to be internally mixed and are represented by eight-size sections with dry particle diameter ranges of 0.039–0.078, 0.078–0.156, 0.156–0.312, 0.312–0.624, 0.624–1.25, 1.25–2.5, 2.5–5.0, and 5–10 µm. Both particle number and mass are simulated in each bin. The MOSAIC aerosol module includes treatments of nucleation, coagulation, and condensation as described in previous studies [Chapman et al., 2009; Fast et al., 2009]. The size-dependent dry deposition of particles (both number and mass) is based on the approach of Zhang et al. [2001]. In addition, both in-cloud and below-cloud wet removal of trace gases and aerosols are simulated following Easter et al. [2004]. The primary OA (POA) directly emitted by combustion sources were previously represented in WRF-Chem, but SOA formed by photochemical reactions of organic vapors in the atmosphere were not included in MOSAIC [Fast et al., 2009], until recently [Shrivastava et al., 2011]. In this work, we represent both POA and SOA in WRF-Chem using the simplified and computationally efficient two-species volatility basis set (VBS) treatment developed by Shrivastava et al. [2011] to account for the role of organics affecting aerosols and clouds. Shrivastava et al. [2011] comprehensively evaluated this approach using both ground and aircraft measurements from the Megacity Initiative: Local and Global Research Observations (MILAGRO) field study [Molina et al., 2010]. Although much more modeling work is required to develop an accurate representation of SOA [Hallquist et al., 2009], the OA evolution from the present two-species VBS treatment is more realistic than the simplified treatment of all organics as either POA, or using traditional SOA approaches with fixed yields that have severely underpredicted SOA compared to measurements [de Gouw et al., 2005; Dzepina et al., 2011a; Heald et al., 2005; Volkamer et al., 2006].

2.2.1 SOA Formation

[9] Several recent studies have shown that oxidation of S/IVOC precursors represents potentially large and previously unaccounted SOA (SI-SOA) from anthropogenic and biomass burning sources [Dzepina et al., 2011b; Hodzic et al., 2010; Robinson et al., 2007; Shrivastava et al., 2011, 2008; Tsimpidi et al., 2010]. We represent the S/IVOC precursors by high-volatility gas-phase species in the two-species VBS mechanism. Shrivastava et al. [2011] described the two-species VBS mechanism in detail. Here we describe it briefly. SI-SOA is formed due to reaction of S/IVOC precursors with OH radical (assumed reaction rate of 0.57 × 10−11 molecule cm−3 s−1). The mass of the parent S/IVOC is assumed to increase by 50% for each generation of oxidation to account for added oxygen and the increase in O:C ratio during formation of SI-SOA. Note that SI-SOA is assumed to have extremely low volatility (C* of 10−3µg m−3) and would not evaporate under most atmospheric conditions. In addition, SOA produced by oxidation of biogenic and traditional anthropogenic VOC, referred as V-SOA, are included using a one-species treatment. V-SOA C* is assumed to be equal to 1 µg m−3 consistent with atmospherically relevant smog chamber SOA yield measurements (as summarized in Table S1 in the supporting information). In addition to oxidation reactions with OH radicals described by Shrivastava et al. [2011], reactions for biogenic VOC precursors including isoprene (ISOP), terpenes (TERP), and sesquiterpene (SESQ) with O3 and NO3 forming B-V-SOA are included in this study [Hallquist et al., 1999; Kleindienst et al., 2007; Ng et al., 2008; Presto et al., 2005]. The mass yields and reaction rate constants of V-SOA precursors with OH, O3, and NO3 radicals are shown in Table S1 in the supporting online information. With the exception of the SESQ reaction rate with NO3 radical, which is taken from Baek et al. [2011], reaction rates of all other V-SOA precursors are from the SAPRC-99 mechanism. In this study, WRF-Chem predicted that SI-SOA was often much larger than V-SOA (not shown). Mass yields for V-SOA precursors shown in Table S1 are converted into molar yields and included in the SAPRC-99 gas-phase chemical mechanism implemented in WRF-Chem. Note that aqueous formation of SOA is not included in the model, but as discussed later, the effect of aqueous processes on OA mass was likely small during CHAPS.

2.3 Emissions

[10] Hourly aerosol and trace gas emissions from sources other than biomass burning and biogenic emissions are derived from the United States Environmental Protection Agency's 2005 National Emissions Inventory (NEI05) ( adjusted to the CHAPS period (2007) assuming linear variations of emissions between NEI 2005 and 2008. Biomass burning emissions including both gases and aerosols are from the 2007 Fire Inventory from NCAR (FINN07) [Wiedinmyer et al., 2011], but were negligible during CHAPS. Primary particulate emissions include both particulate matter with diameters less than or equal to 2.5 µm (PM2.5) and particulate matter with diameters less than or equal to 10 µm (PM10). NEI05 PM2.5 emissions are speciated into categories of sulfate, nitrate, OA, EC, and other unspecified primary PM2.5 (or PMFINE) referred to as OIN, based on the simplified PM2.5 speciation profiles described by Hsu et al. [2006]. In this work, we assigned all unspeciated PMFINE emissions to the OIN category without further speciation. Surveying bulk speciation profiles of PM2.5 from the literature, Reff et al. [2009] showed that most of the OIN emissions are comprised of crustal elements, potassium, sodium, chlorine, and metal bound oxygen. Among these, potassium and chlorine are primarily emitted from biomass burning, but biomass burning was negligible during CHAPS. Thus, OIN was most likely associated with crustal or other unspeciated sources as described by Reff et al. [2009]. On average OIN contributes ~77% of PM2.5 mass emissions in the NEI05 inventory during the CHAPS study. In addition to emissions, the MOZART model [Emmons et al., 2010] is used to produce the boundary conditions for dust, which are also assigned to the OIN category. FINN07 particulate emissions include organic carbon (converted to OA using an OA/OC ratio of 1.4), black carbon, PM2.5, and PM10. VOC emissions are speciated according to the SAPRC-99 mechanism. Particulate emissions in the inventory are distributed based on MOSAIC code similar to Fast et al. [2006], and PM10 emissions are assigned a speciation profile equivalent to that of PM2.5. Chemical species in each particulate size bin are assumed to be internally mixed, and the same size distribution is assigned to all PM2.5 species including sulfate, nitrate, OA, EC, and OIN. The PM2.5 emissions are split into the first six size bins as 2.6%, 14.7%, 35%, 33.2%, 12.5%, and 1.9% respectively. This distribution is consistent with the measured average size distribution of particles sampled below-cloud layers within plumes of elevated CO mixing ratios on 20, 23, 24, and 25 June. Anthropogenic area emissions are emitted at the surface while anthropogenic point emissions are vertically placed within the model based on NEI05 stack parameters. Wildfire emissions are vertically injected up to an altitude of 300 m above ground level, with local diurnal factors equivalent to those used in MILAGRO simulations [Fast et al., 2009] in which maximum and minimum emission rates are in the afternoon and morning, respectively. However, biomass burning fires in the modeling domain were negligible during the period of CHAPS study.

[11] Model of Emissions of Gases and Aerosols from Nature (MEGAN [Guenther et al., 2006] is used to generate biogenic VOC emissions in the CHAPS modeling domain for the central U.S. The 138 biogenic species from MEGAN are lumped into three biogenic VOC classes: ISOP, TERP, and SESQ. In addition, anthropogenic VOC emissions including lumped classes corresponding to alkanes (ALK4 and ALK5), olefins (OLE1 and OLE2), and aromatics (ARO1 and ARO2) are included in the NEI05 inventory corresponding to the SAPRC-99 mechanism, as described by Tsimpidi et al. [2010].

2.3.1 S/IVOC and POA Emissions

[12] The S/IVOC emissions are coemitted with POA, but their emissions are not reported in NEI05 and FINN07 inventories. Also, POA emission factors reported in the inventories are based on quartz filter measurements of diluted exhaust, which capture just a portion of semivolatile organic vapors (SVOC) [Shrivastava et al., 2008]. In this study, we assume that total SVOC emissions are two times POA for both anthropogenic and biomass burning sources [Hodzic et al., 2010; Shrivastava et al., 2011; Tsimpidi et al., 2010]. Note that similar to SI-SOA, POA is represented by species with extremely low volatility (C* of 10−3µg m−3) in the two-species VBS mechanism [Shrivastava et al., 2011]. While this may lead to overestimating POA contribution as it is expected to be semivolatile, as we discuss later, POA contributed to less than 15% of predicted OA during CHAPS. Emissions of IVOC are estimated as 1.5 times total of SVOC and POA emissions [Robinson et al., 2007; Shrivastava et al., 2008]. The sum of SVOC and IVOC emissions is 6.5 times total POA emissions as described in Shrivastava et al. [2011]. Note that emissions of SVOC and IVOC are poorly constrained due to lack of source-specific measurements. However, as we show later, WRF-Chem predicts that OA (mostly SI-SOA) shows reasonable agreement with AMS OA.

2.4 Aerosol–Cloud–Radiation Interactions

[13] WRF-Chem includes interactions between aerosols, radiation, and clouds for the direct, semidirect, and first and second indirect effects as described by Fast et al. [2006], Gustafson et al. [2007], and Chapman et al. [2009]. Aerosol–cloud interactions occur in two directions: aerosols affect clouds by acting as CCN, and clouds (and precipitation) affect aerosols through aqueous-phase chemistry in hydrometeors (following by evaporation) and wet removal. For grid-resolved clouds in both the outer and inner nested domains in this study, the double-moment Morrison microphysics scheme [Morrison et al., 2005; Morrison et al., 2009] is used to predict number and mass mixing ratios of cloud water, cloud ice, snow, rain, and graupel/hail associated with several microphysical processes. In the current study, the first indirect effect is represented by explicitly predicting cloud droplet sizes and number concentrations that are used by the updated Rapid Radiative Transfer Model (RRTMG) radiation schemes [Iacono et al., 2008]. The second indirect effect is handled by the microphysics scheme, where the aerosol properties (number, size, and hygroscopicity) affect cloud droplet nucleation and, subsequently, precipitation and cloud lifetime. Using the new aerosol–cloud couplings for the Morrison scheme, Yang et al. [2011] recently demonstrated the ability of WRF-Chem to predict the cloud macrostructure (such as cloud depth and cloud base height) and optical and microphysical properties (such as cloud top reff and cloud water path) of marine stratocumulus clouds over the Southeast Pacific ocean.

2.4.1 Aerosol Activation and Cloud-Borne Aerosol

[14] The activation of aerosols to form cloud drops in WRF-Chem is described in detail by Chapman et al. [2009]. Briefly, the activation of aerosols from interstitial to cloud-borne state is based on maximum supersaturation determined from a Gaussian spectrum of updraft velocities and internally mixed aerosol properties within each size bin [Abdul-Razzak and Ghan, 2002; Ghan and Easter, 2006]. The number and mass mixing ratios of both interstitial and cloud-borne aerosol species are tracked separately in WRF-Chem. Cloud-borne aerosols that are transported into noncloudy grid cells are resuspended to interstitial aerosols. Other processes affecting cloud-borne aerosols are cloud chemistry (described below), in-cloud wet removal, and dry deposition (when cloud exists in the lowest model level).

[15] The mechanism of Fahey and Pandis [2001] is implemented to model the aqueous-phase processes leading to interaction of water soluble constituents of aerosols and dissolved trace gases, in a bulk approach. This mechanism includes 50 aqueous-phase species, 17 aqueous-phase ionic equilibria, 21 gas-phase/aqueous-phase reversible reactions, and 109 aqueous-phase chemical reactions. The oxidation of dissolved S(IV) by hydrogen peroxide, ozone, trace metals, and radical species is included. Although cloud droplet size-dependent mass transfer for species such as SO2 and NH3 is included, the model assumes “instant and complete uptake” of more soluble species (high effective Henry's coefficient) including HNO3, H2SO4, and HCl by the cloud water. After the bulk cloud water uptake and aqueous reaction calculations, the resulting changes to aqueous species are divided among the different cloud-borne aerosol size bins. Note that these simplifications in the model were designed for regional-scale studies (such as the current study), as opposed to cloud-scale studies where the timescales are shorter, and size-dependent mass transfer for species such as HNO3 and H2SO4 is more critical [e.g., Romakkaniemi et al., 2006; Xue and Feingold, 2004]. In this study, the terms “cloud chemistry” or “bulk aqueous chemistry” include physical uptake of gaseous species such as HNO3 into cloud droplets and their subsequent dissociation in the aqueous phase.

2.4.2 Aerosol Hygroscopicity

[16] The hygroscopicity parameter κ relates the chemical composition of aerosols to their activation into cloud droplets, as described by the “κ-Kohler” theory [Petters and Kreidenweis, 2007]. κ is used in calculations of activation of aerosols to cloud droplets in WRF-Chem. Model aerosol hygroscopicity in each size bin is determined as a volume-weighted average hygroscopicity of all interstitial aerosol chemical components contributing to that bin. Thus, κ also reflects the bulk chemical composition of aerosols. The default hygroscopicity values in WRF-Chem for chemical components such as sulfate, nitrate, ammonium, and chloride ions were used to compute a volume-weighted average hygroscopicity for activation. The model uncertainty in OA hygroscopicity is mainly related to representing the sources and aging of organic species forming SOA in the atmosphere. In this study, the hygroscopicity of OA was explicitly calculated accounting for source type and SOA formation. The hygroscopicity of SI-SOA species formed due to oxidation of organic vapors emitted by anthropogenic and biomass burning sources is parameterized as a function of O:C ratios following Jimenez et al. [2009]. The κ value for anthropogenic POA is assigned as 1.0 × 10−6 due to its low O:C ratio, while κ of 0.04 is assigned for fresh biomass burning emissions (O:C ratio around 0.3) since they are more oxidized as compared to fossil emissions [Aiken et al., 2008]. The upper bound of κ for highly processed and oxidized organics (O:C ratios of 0.8 or higher) is assumed to be 0.2. A significant source of uncertainty is introduced by the large contribution of poorly defined OIN fraction of aerosol that is typically relevant to crustal, metal, or other unspeciated sources such as off-road diesel engines, waste incinerators, etc. as described by Reff et al. [2009]. Without knowledge of the size distribution and chemical composition of OIN, it is difficult to assign κ value to this species. For example, κ values for just mineral dust could vary between 0.001 and 0.5 in the atmosphere as a function of its mineralogy, chemical reaction pathways during its transport, and its chemical mixing state [Sullivan et al., 2009, 2010]. In this work, we choose a constant κ value of 0.14 for OIN, acknowledging the high associated uncertainty.

3 Observations and Model Configuration

3.1 Field Study Description

[17] CHAPS was designed to study cloud–aerosol interactions in the vicinity of Oklahoma City, Oklahoma during June 2007 and was conducted in conjunction with the DOE Atmospheric Radiation Measurement (ARM) Cloud and Land Surface Interaction Campaign [Miller et al., 2007]. Two aircraft were deployed during CHAPS, the DOE G1 and the NASA Langley King Air B200. The G1 was equipped for in-situ measurements of the aerosol chemical and optical properties, cloud microphysics, and the atmospheric state. The G1 flight pattern was designed to measure conditions, below, within, and above the layers with shallow cumuli upwind and downwind of Oklahoma City. Two different aerosol inlets were utilized on the G1, an isokinetic inlet for sampling particles in clear air and nonactivated particles in clouds and the CVI inlet for sampling cloud drops. Post campaign analysis of the CVI and cloud drop size distribution data indicated that the droplet cut size of the CVI during CHAPS was about 15 µm diameter (meaning that drops and particles smaller than 15 µm diameter were excluded from the inlet). While sampling in clouds, there were cases of droplet shattering within the isokinetic inlet, leading to large numbers of small particles. In general, the particle mass associated with these events is quite small and should not have a significant impact on the results presented here. The King Air was equipped with a downward looking High Spectral Resolution Lidar and flew above the G1. Berg et al. [2009] provided a comprehensive overview of CHAPS and the suite of measurements for both aerosols and clouds.

[18] Most of the instruments on the G1 sampled at a rate of 1 Hz. Given the G1's air speed of approximately 100 m s−1, this corresponds to distance of 100 m. Thus, to facilitate comparisons with the model, the G1 data were averaged over 10 s intervals. In cases in which the G1 was flying through a broken cloud field, there could be 10 s intervals during which the aircraft was both within and outside of a cloud. To address these cases, two separate time series were developed, one within and the other outside clouds. For comparison with observations, model predictions are spatially and temporally interpolated to the location of the measurement data for all aircraft flights using the Pacific Northwest National Laboratory (PNNL's) Aerosol Modeling Testbed toolkit developed for WRF-Chem [Fast et al., 2011]. For clarity, both observational and model data were then averaged to 1 min intervals.

4 Results and Discussion

4.1 Representing Convective Processes

[19] Figure 1b compares the model simulated cloud fraction and the measured reflectivity data from the visible channel of GOES-12 geostationary satellite [Wagener and Ma, 2012] at 14:00 UTC on 25 June 2007. Model predicted cloud cover over half-degree boxes (defined as the fraction of the total number of model grid columns within each half-degree box which consist of cloud water, snow, rain, grauple, and/or ice) is consistent with the satellite-derived reflectivity within the green rectangle shown in Figure 1a. While these variables are not directly comparable, they permit us to qualitatively evaluate the location and frequency of predicted clouds in the model simulation. For example, as shown in Figure 1b, red dots representing high cloud fraction predicted by the model are colocated with higher satellite reflectivity shown in grayscale. Since coarse grids in numerical models cannot resolve small clouds such as fair-weather cumulus, which are on the order of 100s of meters to a kilometer [e.g., Berg and Kassianov, 2008], convective parameterizations are used to represent these subgrid-scale processes. In this study, the Kain–Fritsch convective scheme [Kain, 2004; Kain and Fritsch, 1990] is used to parameterize the effects of subgrid-scale convection over the coarse-resolution outer domain (10 km grid spacing). No convective parameterization is used on the inner domain that utilized 2 km grid spacing. Even the 2 km fine grid spacing utilized here is insufficient for representing the details of shallow cumuli dynamics. In addition, completing coupled cloud–aerosol–meteorology simulations using WRF-Chem at the 2 km grid spacing utilized here is already computationally expensive, making it difficult to test finer grid spacings. Since prescribed aerosol simulations are more computationally efficient, as a test, two simulations are conducted using just prescribed aerosols and meteorology (MET), i.e., gas/aerosol emissions and chemistry are turned off using: (1) 2 km inner grid spacing and (2) 660 m inner grid spacing. In these tests, the aerosol activation routine uses a prescribed aerosol number mixing ratio of 109 kg −1 (~1000 cm−3) with lognormal size distribution to calculate cloud droplet nucleation tendencies. All other meteorological and physical variables are kept same as runs with gas/aerosol emissions and chemistry turned on (Table 1). We note that one of the objectives of this study is to establish a baseline simulation against which coarser-resolution simulations with a new shallow cumulus parameterization can be compared. Here we compare the cloud microphysical properties at the 2 km and 660 m grid spacings to demonstrate that the cloud microphysical properties with prescribed aerosols are qualitatively similar at the two resolutions. Figure 2 compares the daytime (1400 UTC to 2100 UTC) shallow cumuli cloud fractions for the 2 km and 660 m grid spacing inner domains for four different simulated days in June 2007: 20th, 23rd, 24th, and 25th. We define the shallow cumuli cloud fraction as the ratio of the number of vertical columns where the model predicts shallow clouds to the total number of columns in the inner domain. In this study, we define shallow clouds by cloud base height less than 2.5 km and cloud thickness (defined as distance between top and base of the cloud) less than or equal to 1.5 km. This definition is consistent with properties of fair-weather cumuli observed over the U.S. DOE's ARM Climate Research Facility Southern Great Plains site [Berg and Kassianov, 2008]. In addition, these shallow clouds are chosen because they are generally nonprecipitating, so that wet removal can be ignored. Figure 2 shows that the predicted fraction of shallow cumuli from 2 km grid spacing MET runs are very similar to higher-resolution 660 m grid results for all 4 days.

Figure 2.

Percentage cloud cover in the nested domain based on column counts where we have cloud base height less than 2.5 km and cloud thickness less than 1.5 km.

[20] Figure 3 shows normalized frequency of cloud water mixing ratio (LWC) and grid-resolved vertical velocity (w) within shallow clouds over the inner domain on 25 June 2007. 25 June was the cloudiest day during CHAPS as seen from simulated cloud fractions in Figure 2 as well as the cloud amounts observed along the flight track (not shown). Figure 3a shows that the LWC distributions are similar for both 2 km and 660 m grid spacing MET runs. However, the 2 km grid shows a larger fraction of w values close to zero, while the 660 m grid shows a longer tail of higher vertical velocities (Figure 3b). Similarly, Figure 3c shows that 2 km grid has a higher fraction of lower CDNC values as compared to the 660 m grid. When averaged over all in-cloud grid cells over the inner domain during this period, the 660 m grid shows higher w of 0.29 ± 0.60 m s−1 (average ± 1 standard deviation) compared to the 2 km grid (w of 0.11 ± 0.33 m s−1), but the cloud water mixing ratios and the CDNC are very similar between the 660 m and 2 km grid simulations (LWC values 0.20 ± 0.18 and 0.17 ± 0.16 g g kg−1 and CDNC values of 260.3 ± 190.7 and 236.1 ± 193.6 cm−3 for 660 m and 2 km resolution, respectively). In addition, simulated shallow cumuli cloud fractions in Table 2 are nearly independent of the grid spacing. These results show that in a statistical sense, the 2 km spacing inner domain simulates similar cloud microphysical properties as the 660 m inner domain, and the 2 km grid spacing simulations can be used for studying aerosol processing in shallow clouds. However, we note that these 2 km grid spacing simulations are not designed to capture all of the complexity of the dynamic processes within the shallow cumulus clouds.

Figure 3.

Normalized frequency counts of (a) cloud water mixing ratio and (b) vertical velocity (w) and (c) cloud droplet number concentration in shallow clouds for the 2 km and 660 m grid spacing inner domains on 25 June 2007 using prescribed aerosol simulations.

Table 2. Model Configurations Used by the Offline Optical Code Demonstrating Sensitivity to Various Parameters Including Aerosol Water, OIN Content, and Complex Refractive Index ni for OINa
Model CasesAerosol WaterOIN FactorOIN ni
  • a

    The OIN factor is applied to the simulated OIN in all aerosol size bins.

Case 1Dry (zero water)1.00.006
Case 2Wet (as simulated)1.00.006
Case 3Wet (as simulated)0.50.006
Case 4Wet (as simulated)1.00.0004

4.2 Aerosol Chemical Composition and Optical Properties

[21] Here we evaluate simulated aerosol chemical composition and optical properties using measurements from the G1 aircraft during CHAPS. We analyze the coupled cloud–aerosol–meteorology simulations during 18 to 25 June 2007 as described in section 2. While G1 measurements are available for several days, 25 June had the largest observed and simulated shallow cumuli cloud cover and is most suitable to model-measurement comparisons of aerosols within and outside clouds. Figure 4 compares predictions of WRF-Chem to G1 measurements on 25 June when the aircraft was outside the clouds and intersected several plumes downwind of Oklahoma City. Note that the uneven spacing of the G1 data in Figure 4 is because data between flight legs with constant altitude and transects within clouds have been removed. In this case, the data shown in Figure 4 are for cases in which the G1 was flying in clear sky. As shown in Figure 4a, during the first and the last part of the flight, the aircraft flew above clouds, and CO mixing ratios of approximately 100 ppbv were observed. When the aircraft flew below clouds, it intersected the Oklahoma City plume twice, with CO mixing ratios increasing to 140–150 ppbv. Note that CO is a chemically quasi-inert species on timescales of several days and is frequently used to identify air associated with fresh urban emissions [DeCarlo et al., 2010]. Figure 4a shows that simulated plume locations agree with measurements, but predicted CO values are 30% higher both times that the Oklahoma City plume was intersected. Concurrent with this increase in CO, increases in aerosol mass concentrations are also evident in Figures 4c–4f. AMS measurements show that sulfate and OA are major contributors to fine aerosol. Figure 4c shows that the model underpredicts sulfate aerosol mass by a factor of ~2 (on average) as compared to AMS measurements in both plumes below-cloud layers, indicated by the gray-shaded areas. Part of this underprediction may be related to underprediction in sulfur dioxide (SO2) mixing ratios shown in Figure 4b. The model also underpredicts the nitrate and ammonium aerosol concentrations in both plumes below clouds. This is likely related to the representation of the magnitude and spatial and temporal variation of precursor emissions (e.g., SO2 and NH3). Figure 4f shows that the model predicts total OA concentrations relatively well compared to below-cloud measurements from the AMS made within the Oklahoma City plume. The simulation indicates that SOA from anthropogenic sources is a major contributor to OA (~70%), while biogenic SOA contributes less than 20% (not shown). Biomass burning is negligible in the region surrounding the Oklahoma City during this period. The POA contributions to OA are less than 15% throughout the flight transect (not shown). Figures 4c–4f also show that WRF-Chem overestimates sulfate, nitrate, ammonium, and total OA mass at some other times when the aircraft was flying at higher altitudes or above the clouds.

Figure 4.

WRF-Chem predictions with measurements along the G1 flight transect under clear sky conditions on 25 June 2007: (a) CO mixing ratio, (b) SO2 mixing ratio, (c) sulfate aerosol, (d) nitrate aerosol, (e) ammonium aerosol, and (f) organic aerosol (OA) mass concentrations. For comparison to Aerosol Mass Spectrometer (AMS) measurements, simulated aerosol mass is calculated in the size range 0.039 – 0.625 µm. CO and SO2 mixing ratios and aerosol concentrations are shown on the left hand vertical axes, while the aircraft altitude is shown by green dots corresponding to the right hand vertical axes. The gray-shaded areas correspond to times when the aircraft was flying below clouds and includes periods during which the aircraft intersected fresh emissions as shown by significant increase in CO.

[22] In addition to chemical composition, evaluation of optical properties (such as light scattering and absorption) of aerosols gives valuable insights about how well regional scale models such as WRF-Chem represent climate relevant properties. Optical properties such as scattering and absorption depend on particle chemical composition, size distribution, water content, and refractive index of the various particle chemical species. We note that MOSAIC [Zaveri et al., 2008] does not treat water uptake associated with OIN and OA. However, the aerosol activation routine in WRF-Chem uses the hygroscopicity of all chemical species including OIN and OA as described in section 2.4.2. Thus, the simulated water content of nonactivated aerosol represents the lower bound hygroscopicity of OIN and OA. The refractive indices of various chemical species, except OIN, are assigned as described in Table 1 of Barnard et al. [2010]. For all organic species including both POA and SOA, the real part of the refractive index is assumed as 1.45, while the complex refractive index is assumed as zero. For OIN, we use the refractive index of mineral dust. The complex refractive index (ni) of dust is highly variable with values ranging from 0.0004 to 0.006 at wavelengths around 550 nm, while the real part of the refractive index of dust (OIN) has less variation [e.g., Dubovik and King, 2000; Kandler et al., 2007; Osborne et al., 2008; Otto et al., 2009; Patterson et al., 1977; Petzold et al., 2009; Zhao et al., 2010].

[23] Figures 5a and 5b compare optical scattering and absorption of aerosols at 550 nm wavelength, measured using the TSI model 3563 integrating nephelometer and Radiance Research Particle Soot Absorption Photometer on board the G1, to model the simulation for the times indicated by gray-shaded areas in Figure 4. Aerosol optical property calculations are performed using an offline code that employs the same modules as WRF-Chem and the same hourly aerosol concentrations. The advantage of using the offline code is that sensitivity studies investigating the effects of aerosol properties (such as aerosol water content and refractive index) can be performed much faster. In this work, we perform sensitivity simulations using the offline optical code with both dry and wet aerosol and the upper and lower bound of ni for OIN. Thus, we define four cases as shown in Table 2: Cases 1, 2, and 3 use the upper bound ni of 0.006, while Case 4 uses the lower bound ni of 0.0004 for OIN. Case 1 is the dry case (calculated by setting the aerosol water content in each aerosol size bin to zero), while Cases 2, 3, and 4 are wet cases using simulated aerosol water content from the online code. Case 3 is designed to examine the sensitivity of simulated optical properties to OIN content by reducing OIN in all size bins by 50%. As discussed earlier, the errors in the assumed OIN size distribution may have led to misplacement of OIN mass among the eight WRF-Chem size bins. However, since OIN is not speciated further in the model and the size distribution of OIN is not known, it is difficult to refine this analysis. Calculations for each of these four cases are done using only the first six aerosol size bins because the isokinetic sampling inlet aboard the G1 prevented most particles larger than 2.5 µm from entering the instruments. Note that for the 550 nm wavelength, the aerosols with diameters ranging from 0.1 to 1 µm (size bins 3–5) scatter visible light most efficiently [Seinfeld and Pandis, 1998]. The CHAPS period often had high ambient relative humidity causing significant aerosol water uptake. For example, WRF-Chem predicted that ambient relative humidity was higher than 80% along aircraft flight transects shown by gray-shaded areas in Figures 4 and 5. The relative humidity in the nephelometer was approximately 40% during the research flights, suggesting that aerosols may not have been completely dry. Cases 1 and 2 compare optical properties for dry and wet aerosols, respectively.

Figure 5.

Optical properties of aerosols with measurements along the G1 flight transect under clear sky conditions on 25 June 2007 at wavelength of 550 nm: (a) scattering, (b) absorption, (c) single-scattering albedo, (d) average volume size distribution during this time. The green plus marks (right-hand vertical axis) in Figure 4b show WRF-Chem-simulated BC concentrations. Model results are calculated offline using the hourly online WRF-Chem predictions of aerosol concentrations and are shown for four different cases. The dry and wet cases are with and without aerosol water. The 0.5OIN, wet case represents calculations, after reducing WRF-Chem predicted OIN concentrations by 50% in all size bins. Cases 1, 2, and 3 use the upper bound of complex refractive index (ni) for OIN, while Case 4 represents the wet case with the lower bound complex refractive index for OIN as discussed in the text.

[24] Figure 5a shows that when aerosols are wet (Cases 2 and 4), the model significantly overestimates scattering as compared to observations. Also, aerosol scattering from Case 4 (low ni) is very similar to Case 2 (black star and red squares, respectively). In comparison, the dry aerosol Case 1 is in much closer agreement to the observations. As expected, Figure 5b shows that aerosol absorption is same for both dry and wet aerosol (Cases 1 and 2), since only BC and OIN contribute to aerosol absorption computed in the model, via the complex part of refractive index. Note that we assumed all particles are internally mixed and used the volume mixing approximation for optical calculations. Thus, water lowers the volume-weighted complex refractive index (as water is nonabsorbing at 550 nm wavelength), but the absorption cross-sectional area increases due to water, resulting in very small differences between the dry and wet cases (Cases 1 and Case 2, respectively) in simulated absorption. Figure 5b also shows that the simulated aerosol absorption increases with increasing simulated BC concentrations (shown by green plus marks) both times that the aircraft intersects the Oklahoma City plume. Also, both Cases 1 and 2 overpredict absorption as compared to measurements. The effect of complex refractive index of OIN on simulated absorption is shown by comparing Cases 2 and 4 representing the upper and lower bounds of ni, respectively. The simulated absorption is significantly reduced with the lower ni for OIN, and Case 4 gives much better agreement with measured absorption compared to Case 2. Figure 5c compares observed single scattering albedo (SSA; defined as the ratio of the aerosol scattering to the sum of scattering and absorption) at 550 nm to the simulated SSA. Case 1 significantly underpredicts the SSA compared to measurements, mainly due to high absorption prediction and decreased scattering. Although Case 2 shows better agreement with SSA measurements, the high scattering predicted by this case points to uncertainties in predicting the chemical composition of aerosols. For example, as we discuss below, the amount of OIN is likely overestimated in the simulation. In comparison, Case 4 (low ni) overpredicts SSA as compared to measurements.

[25] Figure 5d compares observed average aerosol volume distributions to WRF-Chem predictions during the times indicated by gray-shaded areas in Figure 4 that includes periods during which the G1 intersected the Oklahoma City plume. The observed aerosol volume distributions are derived by combining Dual Mobility Particle Sizer (DMPS) measurements (in the size range 17–350 nm) and aircraft nose-mounted Passive Cavity Aerosol Spectrometer (PCASP) for particles ranging 350 nm to 2.8 µm. The observations indicate a bimodal volume distribution with peaks at 300 nm and 2.7 µm. WRF-Chem predictions are shown by histograms indicating the chemical composition of aerosols in each of the six size bins. The model also predicts a bimodal distribution with peaks at size bin 3 (156–312 nm) and size bin 6 (1.2 – 2.5 µm). It is important to note that OIN is the dominant component of the simulated aerosols in size bins 2–6. While, OIN constitutes 50–60% of volume in size bins 2–5, it comprises 90% of aerosol volume in size bin 6, and the simulated volume in size bin 6 is a factor of 2 too high.

[26] The likely overprediction of OIN has an important implication in the calculated aerosol optical properties. The optical properties derived for Case 3, in which the OIN is reduced by a factor of 2 for all size bins, are shown in Figures 5a–5c. Case 3 includes aerosol water, as the dry aerosol Case 1 significantly underpredicts the SSA compared to observations and decreasing the OIN decreased the scattering even further (not shown). Figure 5a shows that Case 3 results in good agreement with measured scattering during 17.4–17.6 h UTC, but underpredicts scattering by about 35% during 17.8–18.1 h UTC. However, aerosol absorption shown in Figure 5b is in much better agreement with measurements because OIN contributes ~70% to aerosol absorption in Case 2 even within the plumes (indicated by increasing BC in Figure 5b) and reducing OIN concentrations reduces aerosol absorption. Figure 5b also indicates that Case 3 (lower OIN concentrations) and CASE 4 (lower OIN ni) give similar decreases in simulated absorption. Also, Figure 5c shows that Case 3 results in better agreement with measured SSA as compared to Cases 1, 2, and 4. However, this agreement of Case 3 results does not constrain the uncertainty in OIN emissions. Figure 4 showed that the model underpredicts other aerosol components including sulfate, nitrate, and ammonium within the plumes, which also affect the predicted aerosol scattering and SSA. We also note that all four cases clearly show a consistent increase in simulated absorption and a decrease in SSA within the Oklahoma City plumes. In addition, if the water uptake by OIN and OA were included, we would expect an increase in scattering and SSA in Cases 2–4. Figure 5 shows that there are several compensating effects due to uncertainties in predicting size-resolved particle chemical composition, the complex part of refractive index of OIN and the water content of the particles. In order to further improve aerosol optical predictions and to obtain the right answers for the right reasons, detailed measurements of particle chemical speciation, their size distribution and refractive indices of various aerosol components including OIN need to be accurately simulated. Future modeling studies should provide better chemical speciation of the unassigned portion of PM2.5 mass (OIN), particularly when simulated aerosol optical properties are of interest, by using methods such as the profiles of Reff et al. [2009].

4.3 Particle and CCN Number

[27] In order to reliably simulate aerosol–cloud interactions, models need to predict CCN concentration accurately. The CCN concentration depends on the particle number, size distribution, hygroscopicity, and the ambient supersaturations. Figure 6a compares the simulated to the measured particle number concentrations (combining the DMPS and PCASP measurements) in the 39–2500 nm dry diameter range along the aircraft flight transects flown on 25 June. Figure 6a shows that the model consistently overestimates the particle number for this size range, i.e., by a factor of 1.7 on an average in the plumes indicated by the gray-shaded regions (17.4–18.1 h UTC). Figure 6b compares the measured and simulated average particle number distribution within the plumes. Figure 6b shows that although the model overpredicts the particle number in size bin 2 (particle diameters ranging between 78 to 156 nm), it shows much better agreement with measurements for particles in size bin 3 (particle diameters ranging between 156 and 312 nm). Consistently, Figure 6c shows that the model simulates the particle number concentration in size bin 3 reasonably well within the plumes. As discussed below, particles in size bin 3 are the dominant contributors to simulated CCN number at the supersaturations used by the CCN instrument.

Figure 6.

Measured particle number and CCN concentrations and the WRF-Chem predictions for aircraft flight transect on 25 June: (a) particle number concentrations for dry diameter range 39 nm to 2.5 µm, (b) average particle number distribution between 17.4 and 18.1 h UTC, i.e., within plumes, (c) particle number concentrations in size bin 3 (diameters ranging 156–312 nm), and (d) CCN number concentrations. In addition to total CCN number concentrations (red points), the simulated CCN number corresponding to particle size bin 3 is also indicated (orange points).

[28] Since the online WRF code predicts CCN at simulated supersaturations, we cannot compare them directly to CCN measurements which are at different supersaturations. For direct comparison to CCN measurements, we calculated model predicted CCN concentrations using an offline code (using the same methodology as the online WRF-Chem model) at the supersaturations used by the CCN instrument. The inputs to the offline code include the size-dependent number concentration, chemical composition, and hygroscopicity of particles predicted by the online WRF-Chem model. Figure 6d compares measured and simulated CCN number concentrations and shows that the simulated CCN number reasonably agrees with the measurements within the plumes (indicated by gray-shaded regions). Figure 6d also shows the simulated contribution of particles in size bin 3 (indicated by orange points) to the simulated CCN concentrations. On this flight, the CCN instrument aboard the G1 measured CCN concentrations at a single supersaturation of 0.26% within the plumes (17.4–18.1 h UTC). Based on the average simulated hygroscopicities of size bins 2 and 3 (0.16 and 0.18, respectively), a critical supersaturation of 0.26% corresponds to dry diameters of 99 nm and 105 nm, respectively. Thus, all of the simulated particles in bin 3 (156 – 312 nm) could activate at 0.26% supersaturation, but only a small fraction (17% on average within the plumes) of particles in size bin 2 could activate at this supersaturation. Particles in size bin 3 (particle diameter range 156–312 nm) contribute 61% on average to the total simulated CCN number (at 0.26% supersaturation), while particles in size bin 2 contribute 32%, and particles in larger size bins 4 to 8 (diameters larger than 312 nm) contribute 7% on average. As the model reasonably simulates the particle number concentration in the size bin 3 (Figure 6c) which contributes the most to CCN number, it shows much better agreement between measured and simulated CCN number concentrations (Figure 6d), even though the 39–2500 nm diameter particle number (Figure 6a) is overpredicted by a factor of 1.7 (on an average in the plumes).

[29] Since the hygroscopicity of OIN is uncertain (as discussed in section 2.4.2), we used the offline code to investigate the sensitivity of the predicted CCN number concentrations to the assumed OIN hygroscopicity. The value of OIN hygroscopicity (0.14) used in this study overpredicts CCN number by about 12% within the plumes. In comparison, assuming an OIN hygroscopicity of 0.5 increases simulated CCN number concentration by a factor of 2 on average compared to the base simulation (OIN hygroscopicity of 0.14) and also overpredicts the observed CCN number (Figure S1 in the supporting information). Also, an OIN hygroscopicity of 0.0 decreases predicted CCN number concentration by 32% compared to the base simulation and underpredicts measured CCN number by 23% (on average within the plumes). These results demonstrate the importance of representing the hygroscopicity of OIN accurately in models.

4.4 Chemical Composition of Cloud Droplet Residuals and Interstitial Aerosols

[30] While it is useful to characterize the absolute concentrations of aerosols, their relative volume fraction ξi can be used as intensive metric to compare relative differences in chemical composition of aerosols in cloud systems. ξi is calculated in WRF-Chem with appropriate assumptions of density of each chemical component [Barnard et al., 2010] and is used to determine overall hygroscopicity and CCN activation as discussed in previous section. To illustrate how absolute concentrations compare with volume fractions, Figures 7a and 7b show absolute concentrations and relative volume fractions of AMS aerosol species (sulfate, nitrate, ammonium, and organics) measured within shallow cumuli to corresponding WRF-Chem predictions on 25 June 2007. The sum of mass concentrations and overall volume fractions in four WRF-Chem aerosol size bins (0.039–0.624 µm) are computed to compare with AMS measurements. Figure 7a shows smaller mass concentrations of sulfate and OA in cloud droplet residuals (CVI inlet) as compared to interstitial aerosols (isokinetic inlet), i.e., median mass concentrations values are a factor of 4 smaller for both sulfate and OA in cloud droplet residuals. The low aerosol mass loading observed in cloud droplet residuals is mainly because the CVI inlet only samples cloud droplets larger than approximately 15 µm, in order to ensure a reliable separation between cloud drops and large nonactivated aerosols. An analysis of cloud droplet size distribution measured by Cloud–Aerosol Spectrometer (CAS) on a selected flight transect on 25 June showed that a large fraction of number of cloud droplets are smaller than the CVI cutoff size of 15 µm, leading to a bias in the measured aerosol mass. For this reason, it is better to compare volume fractions to evaluate differences in chemical composition of cloud droplet residuals and nonactivated aerosols.

Figure 7.

(a) Absolute volume concentrations and (b) volume fractions of sulfate, nitrate, ammonium, and organic aerosol (OA) from observations and WRF-Chem simulations within shallow cumuli along a 100 km  × 100 km domain surrounding Oklahoma City on 25 June 2007. ISO: aerosols sampled through isokinetic inlet, CVI: aerosols sampled through Counterflow Virtual Impactor inlet, INT: simulated interstitial (nonactivated) aerosols (analogue to ISO), ACT: simulated activated aerosols (analogue to CVI). The boxes indicate 25th and 75th percentiles, while the whiskers indicate the 5th and 95th percentiles.

[31] Figures 8a and 8b compare relative volume fractions of sulfate, nitrate, ammonium, and OA on 20 and 23 June, respectively. It is important to note that on all three days, i.e., 20, 23, and 25 June (Figures 7b, 7a, and 8b), observed nitrate volume fractions in cloud droplet residuals are significantly higher than in interstitial aerosols. Median nitrate volume fractions are almost negligible (0.003, 0.01, and 0.02) in interstitial aerosols, but are significantly larger in cloud droplet residuals (0.05, 0.1, and 0.08) on 20, 23, and 25 June, respectively. We note that this observed nitrate enhancement does not include the large number of cloud droplets smaller than the CVI cut size of approximately 15 µm observed during CHAPS. Our volume fraction analysis assumes that the relative composition of the larger droplet residuals is the same as the smaller droplets. However, Hayden et al. [2008] showed that more nitrate was found in smaller sizes compared to sulfate. In addition, the modeling study of Wurzler et al. [1995] demonstrated that more HNO3 uptake occurs in smaller cloud droplets close to the cloud base because of higher surface area of the numerous smaller cloud droplets and the relatively high solubility of HNO3. In addition, there is likely some degassing of nitrate behind the CVI as cloud droplets evaporate; however, the nitrate could remain in the particle phase if HNO3 reacts to produce a less volatile form of nitrate, particles retain some water to preserve the acidic nitrate, or nitrate reacts with NH3 or other alkaline compounds [Hayden et al., 2008]. Thus, the CVI measurements most likely under-represent the nitrate enhancement in cloud droplet residuals, and the actual nitrate enhancement may be even larger. Enrichment of nitrate in cloud droplet residuals could be due to a variety of reasons including preferential scavenging of nitrate containing particles, or dissolution of nitric acid (HNO3) vapors into cloud droplets [for example, Fuzzi et al., 1994; Mehlmann and Warneck, 1995] without efficient volatilization during evaporation. NO2 is only slightly soluble in water [Wesely, 1989], and the direct oxidation of NO2 in the liquid phase is slow and was unlikely to have contributed to the observed nitrate enrichment in the short-lived shallow cumulus clouds.

Figure 8.

Volume fractions of sulfate, nitrate, ammonium, and OA from observations and WRF-Chem simulations within shallow cumuli clouds along a 100 km × 100 km domain surrounding Oklahoma City on (a) 20 June and (b) 23 June. ISO: aerosols sampled through isokinetic inlet, CVI: aerosols sampled through Counterflow Virtual Impactor inlet, INT: simulated interstitial (nonactivated) aerosols (analogue to ISO), ACT: simulated activated aerosols (analogue to CVI). The boxes indicate 25th and 75th percentiles, while the whiskers indicate the 5th and 95th percentiles.

[32] To further investigate the role of aqueous processing of aerosols in clouds, we compared model simulations with cloud (aqueous) chemistry turned on and off during 12 daytime hours (1300 UTC on 25 June to 00 UTC on 26 June). Figure 7b shows that when cloud chemistry is turned on, the median nitrate volume fraction in activated aerosols predicted by WRF-Chem is 0.2, which is an order of magnitude higher than predicted in interstitial aerosols (volume fraction of 0.02). In contrast, when cloud chemistry is turned off, median nitrate volume fractions in interstitial and activated aerosols predicted by WRF-Chem are very similar (0.09 and 0.11 respectively). The WRF-Chem simulations suggest the nitrate enhancement in activated aerosols is primarily due to uptake of HNO3 vapor by cloud droplets. The differences in the interstitial aerosol nitrate volume in the two simulations (cloud chemistry on versus off) are due to a combination of two effects. When cloud chemistry is turned on, uptake of HNO3 vapor by cloud droplets reduces HNO3 vapor concentrations to near zero (not shown), which may cause some of the interstitial aerosol nitrate to evaporate. When cloud chemistry is turned off, HNO3 vapor concentrations within cloud are initially similar to those outside cloud, but the water content of interstitial particles within cloud is much higher than those outside the cloud. This reduces the equilibrium HNO3 vapor concentration and causes condensation of HNO3 vapor on interstitial aerosols. Thus, both effects support the higher interstitial aerosol nitrate volume in simulations with cloud chemistry turned off as compared to simulations with cloud chemistry on, as shown in Figure 7. As noted earlier, the bulk aqueous chemistry code does not treat the size-dependent mass transfer of HNO3. Our bulk results for HNO3 uptake on cloud droplets are robust, but we do not have size-dependent results to compare with cloud parcel models that include size-dependent mass transfer rates [e.g., Xue and Feingold, 2004].

[33] Our findings of enhanced nitrate in cloud droplet residuals are consistent with previous studies [Hayden et al., 2008; Sellegri et al., 2003b]. While Sellegri et al. [2003b] sampled oceanic air masses over a clean mountainous location during the winter in France, Hayden et al. [2008] sampled more polluted air masses downwind of regions with large emissions (coal-fired power plants) in Cleveland, Ohio during the summer. Hayden et al. [2008] investigated various cloud types including stratocumulus, cumulus, and towering cumulus which likely have different cloud dynamics compared to the short-lived shallow cumuli sampled during CHAPS, which are presented in this study. Comparison of our investigation with previous studies [Hayden et al., 2008; Sellegri et al., 2003b] suggests that in both clean and more polluted air masses, and among the different cloud types investigated, cloud droplets show enhanced nitrate as compared to nonactivated aerosols.

[34] Figure 7b shows that the observed median volume fraction of sulfate in cloud droplet residuals (0.21) is very similar to that in interstitial aerosols (0.22) on 25 June. Consistently, the sulfate volume fractions in the two simulations (cloud chemistry on/off) are similar to each other, indicating little aqueous sulfate production, which is plausible as SO2 mixing ratios were generally low. On 20 June (Figure 8a), the observed volume fraction of sulfate is very similar in both interstitial aerosols and cloud droplet residuals, and the simulated volume fractions are somewhat greater in activated aerosols. In contrast, the observed median sulfate volume fraction within cloud droplet residuals is a factor of 2.5 lower than nonactivated aerosols on 23 June (Figure 8b), which is not seen in the simulations.

[35] The median observed OA volume fraction in cloud droplet residuals is slightly lower than in interstitial aerosols on 20 June (median OA volume fractions of 0.67 vs. 0.74) and 25 June (median OA volume fractions of 0.56 vs. 0.69) as shown in Figures 8a and 7b, respectively. On 23 June (Figure 8b), even though observed median OA volume fraction in cloud droplet residuals is slightly higher than interstitial aerosols, lower OA volume fractions are frequently observed in cloud droplet residuals, as denoted by the 5th and 25th percentiles on all days. There are several possible explanations for these OA observations. The first possibility is that aerosols with lower organic content are preferentially activated due to lower hygroscopicity of OA. However, the lower OA volume fractions in droplet residuals could also be explained by cloud chemistry: HNO3 uptake and aqueous sulfate production increase nitrate and sulfate volume fractions and thus lower the OA volume fraction. The effect of increasing nitrate fraction on the reduction in OA volume fraction depends on the contribution of nitrate to the total aerosol volume. On all days, the observed nitrate contributed less than 2% and 10% to nonactivated and activated aerosol volumes, respectively; hence, the increase in nitrate may be partly responsible for the decrease in OA volume fractions within cloud droplets. A third possible explanation is that the activated aerosols come predominantly from boundary layer air transported into clouds by updrafts, while the interstitial aerosols may come from the subcloud layer, or could be entrained through the cloud edges or top, which could have different aerosol properties. For example on 25 June, median OA volume fraction in nonactivated aerosols within cloud-level air is 0.6, which is higher than median OA volume fraction of 0.5 measured in boundary layer air below clouds (not shown). The lower OA volume fraction in the below-cloud layer supports the possibility of compositional differences due to the entrainment of air parcels containing compositionally different aerosols within clouds. However, due to very limited measurements of size-resolved chemical composition and no mixing state measurements during CHAPS, it is difficult to explicitly identify the processes affecting aerosol activation. Aqueous chemistry involving organics may also increase aqueous SOA, especially due to interaction of biogenic and anthropogenic pollutants [Ervens et al., 2011]. As shown in Figures 7 and 8, AMS measurements indicate that organic content of activated aerosols is lower than for interstitial aerosols, suggesting the aqueous production of OA was small in comparison to the other mechanisms that lead to reduced organic volume fraction.

[36] WRF-Chem also predicts significantly lower organic volume fractions in activated aerosols as compared to interstitial aerosols, consistent with observations shown in Figures 7b, 8a, and 8b. However, simulations with cloud chemistry turned on result in larger difference between organic content of interstitial and activated aerosols as compared to cases when cloud chemistry is turned off (ratio of 1.6 vs. 1.2 in organic content of interstitial to activated aerosols with cloud chemistry turned on and off on 25 June). This difference between the two simulations (cloud chemistry turned on/off) is mainly due to the increase in nitrate volume fraction within cloud droplets when cloud chemistry is turned on. The model predicts that nitrate contributes 1.6% and 20% on average to the simulated nonactivated and activated particle volumes, respectively, on 25 June (not shown). It is to be noted that the current model simulation does not treat aqueous or cloud processing of organics.

[37] In this section, both field observations and model results showed that chemical composition of cloud droplet residuals differs from interstitial aerosols. Cloud droplet residuals have a higher volume fraction of nitrate and a lower fraction of OA than interstitial aerosols. Comparison of model results with and without cloud chemistry showed that uptake of HNO3 vapor by cloud droplets increases the nitrate content of cloud droplets. In the following section, we examine model results to investigate how aerosols affect cloud microphysics in the atmosphere.

4.5 Effects of Aerosols on Clouds

[38] Analysis of observations during CHAPS showed that cloud microphysical properties such as CDNC and cloud droplet reff were a function of both pollutant loading and cloud vertical velocity [Berg et al., 2011]. Here we investigate whether the model predicts similar relationships between cloud microphysical properties, pollutant loading, and cloud vertical velocity as shown by Berg et al. [2011]. In order to segregate the individual effects of cloud vertical velocity and pollutant loading on cloud properties in the model, we follow the methodology presented by Berg et al. [2011]. In this study, we evaluate sensitivity to pollutant loading in terms of variations in CO' calculated by removing the mean of CO computed every hour for the grid cells classified as shallow cloud. Using CO' accounts for variations of the background CO during the period of interest. It is important to note that model predicted CO and CCN concentrations are strongly correlated (not shown) and CO is a good indicator of pollutant loading in the domain. Also, we examine sensitivity of cloud microphysical properties with respect to the simulated grid-resolved vertical velocity w.

[39] Figure 9 shows contour plot of frequency counts representing the range of CO' and w sampled within the simulated shallow clouds on 25 June 2007 across a 100 × 100 km domain surrounding Oklahoma City. In Figure 9, Box A is defined by CO' values between −20 and 40 ppbv and w between 1.0 and 1.5 m s−1. Points in this box are used to determine the sensitivity of cloud properties to variations in pollutant loading for small variations in cloud updraft velocity w. As we discuss later, the properties of cloud droplets such as CDNC and reff were sensitive to pollutant loading only for updraft velocities larger than 1 m s−1. In comparison, Box B is defined by CO' values between −5 and +5 ppbv and w values between −0.5 and 2.5 m s−1. Points in this box are used to determine the sensitivity of cloud droplets to variations in updraft velocities, for a narrow range of pollutant loadings. WRF-Chem predicts that frequency counts of w greater than 2.5 m s−1 within shallow clouds are negligible (not shown). In contrast, Berg et al. [2011] observed w values as high as 3.5 m s−1 within shallow cumuli clouds during CHAPS. This is due to limitations of 2 km grid spacing in representing higher updraft velocities within simulated clouds, as discussed earlier.

Figure 9.

WRF-Chem simulated frequency counts of CO' and w within shallow cumulus clouds from 9 to 17 h local time on 25 June 2007. Counts are for bins with ∆CO' = 10 ppbv and ∆w = 0.5 m s−1. Points within box A are used to examine the sensitivity of cloud parameters to pollutant loading for small variations in updraft velocities. Points within box B are used to examine the sensitivity of cloud parameters to variations in updraft velocity for small variations in pollutant loading.

[40] Figure 10 show the dependence of CDNC, reff, and cloud water mixing ratio on CO' for box A and on w for box B of Figure 9. Equally spaced bins spanning the range of simulated values were used to examine their relationships. In Figure 10, bins with less than 10 data points were excluded from the analysis. As shown in Figure 10a, CDNC shows an increasing trend with w during updrafts (i.e., positive values of w), increasing by a factor of 1.45 as the mean w increases from 0.72 to 2.24 m s−1. This CDNC trend is not seen at smaller and negative values of w. However, CDNC has much larger variability for this w range, as indicated by large standard deviations shown in Figure 10a. Also, droplet nucleation in WRF-Chem depends on both grid-resolved and subgrid turbulent vertical velocity and CDNC becomes more dependent on the subgrid turbulent w when the grid-resolved w is small (or negative). Note that since we only sample data within shallow clouds and given the relatively coarse resolution of the model simulation, we do not see large occurrences of cloudy downdrafts (negative values of w).

Figure 10.

WRF-Chem simulated variations of (a) cloud droplet number concentrations (CDNC), (b) effective radius reff, and (c) cloud water mixing ratio (g kg−1) with variations of updraft velocity w (CO' ~ 0 ppbv) corresponding to box B in Figure 8; variations of (d) CDNC, (e) reff, and (f) cloud water mixing ratio (g kg−1) with pollutant loading represented by variations in CO' corresponding to box A in Figure 8. Error bars indicate standard deviations across all grid cells in the 100 km × 100 km domain surrounding Oklahoma City for 9 h on 25 June 2007.

[41] Figure 10d shows that CDNC increases with increase in pollutant loading. Increasing CO' from −15 to 35 ppbv as shown in Figure 10d increases CDNC by a factor of 1.5. These trends of increasing CDNC with updraft strengths and pollutant loading simulated by the model are consistent with analysis of measurements during CHAPS presented by Berg et al. [2011]. The observations of Berg et al. [2011] included data from six separate days, but do not include data from 25 June (due to an issue with the measurement of w on that particular day). The results shown in Figure 10 include data from just the 25th, which was the cloudiest day of CHAPS and is the optimal day for looking at the impact of clouds on the aerosol. While the trends of CDNC and reff with changes in CO' and w' can be compared qualitatively to the results from Berg et al. [2011] study, the two are not directly comparable. Note that the simulation indicates that CDNC increases by about 50% with increase in pollutant loading (Figure 10d) at updraft velocity between 1 and 1.5 m s−1 as defined by box A shown in Figure 9. However, CDNC showed no significant increases with pollutant loading at smaller updraft velocities close to zero (not shown). In contrast, Berg et al. [2011] showed significant increase in CDNC corresponding to updraft velocities in the range −0.5 to 0.5 m s−1. This difference is consistent with the fact that grid spacing of 2 km used in this study does not resolve all cloud dynamic processes of shallow cumuli accurately, including important factors like small-scale variation of the vertical velocity, convective downdrafts, and mixing associated with entrainment and detrainment.

[42] Figure 10b shows that reff predicted by Morrison microphysics scheme in WRF-Chem increases with an increase in updraft velocity w. In comparison, Berg et al. [2011] found a decrease of reff up to w' values of 1 m s−1 followed by an increase of reff at higher updraft velocities. They attributed this increase of reff to a larger increase in the number of bigger drops at higher updraft velocities (~3 ms−1). For box B, the simulated cloud water mixing ratio continuously increases with increase in w, as shown in Figure 10c. This increase in cloud water is a stronger influence on reff (Figure 10b) than the increase in CDNC, since reff increases with increase in w consistent with increasing trend of cloud water (Figure 10c).

[43] Most importantly, both analysis of CHAPS field data by Berg et al. [2011] and the model simulation examined in this study (Figures 10d and 10e) clearly show the first aerosol indirect effect or the Twomey effect [Twomey, 1977] where increase in pollutant loading (as defined by CO') causes an increase in CDNC and decrease in reff within shallow clouds. Note that cloud water mixing ratio does not show any trend with increase in pollutant loading as shown in Figure 10f, consistent with our choice of a narrow range of w to isolate the effects of pollutant loading on cloud microphysical properties. Thus, even at this relatively coarse resolution, the model is able to capture some aspects of aerosol indirect effects.

5 Conclusions

[44] The WRF-Chem community model was used to simulate aerosols, clouds, and their interactions during the DOE-sponsored CHAPS field study, which was conducted in June 2007 near Oklahoma City, OK. This study has important implications because there are many moderately sized cities in North America compared to larger, more polluted megacities. The main findings of this study are:

  1. [45] There are significant differences in the chemical composition of cloud droplet residuals and nonactivated interstitial aerosols within cloud layer. AMS observations show large enrichment in the nitrate volume fraction of cloud droplet residuals as compared to nonactivated aerosols. Consistent with AMS observations, the simulations show large enrichment in nitrate content of cloud droplet residuals only when cloud chemistry is turned on, but no enrichment is simulated in the absence of cloud chemistry. The simulation shows that the uptake of HNO3 vapor in cloud droplets is mainly responsible for the large enhancement in nitrate content of cloud droplet residuals. In addition, our findings of enhanced nitrate in cloud droplet residuals are consistent with previous studies [Hayden et al., 2008; Sellegri et al., 2003b].

  2. [46] While the nitrate is enhanced in the cloud residuals, both AMS measurements and the simulation do not indicate an enrichment in sulfate content of cloud droplet residuals, possibly due to low SO2 concentrations. The model also simulates the lower OA fraction in cloud droplet residuals as compared to nonactivated aerosols measured by AMS. While low hygroscopicity of OA relative to other inorganic species could be a playing a role, other factors such as nitrate enhancement in cloud droplets (increasing nitrate fraction lowers OA volume fraction), or entrainment of cloud-level air having different chemical composition from boundary layer air could also be responsible.

  3. [47] In this study, the model did not include an aqueous SOA formation mechanism, but AMS measurements suggest that any aqueous SOA formation was small during CHAPS.

  4. [48] Simulated values of the aerosol optical properties depend on several compensating effects due to uncertainties in predicting size-resolved particle chemical composition (including the OIN contribution), the complex part of refractive index of OIN and the water content of the particles. However, all of our four sensitivity test cases, varying the OIN contribution, water content and the complex refractive index of OIN, clearly show a consistent increase in simulated absorption and a decrease in SSA within the Oklahoma City plumes. As discussed earlier, since MOSAIC does not include water uptake to OIN and OA, the wet cases (Case 2, 3, and 4) effectively assume that the hygroscopicity of OIN and OA is zero. The test case with OIN content reduced by 50% and wet aerosols shows the most consistent agreement with measurements of all three parameters including scattering, absorption, and SSA. However, measurements of size-resolved aerosol chemical composition, detailed chemical speciation and refractory index of all aerosol components including OIN are necessary to correctly represent all processes affecting aerosol optical properties in models.

  5. [49] The WRF-Chem simulation, using 2 km horizontal grid spacing, captures key relationships between aerosol processes and cloud microphysical properties. Analysis of WRF-Chem results within shallow clouds shows clear evidence of the first aerosol indirect effect or the Twomey effect [Twomey, 1977] where an increase in pollutant loading causes an increase in CDNC and decrease in reff. These results are consistent with previously published observations of the Twomey effect during CHAPS [Berg et al., 2011]. The model also shows an increase in CDNC and reff with increasing updraft velocity and liquid water content, at a narrow range of pollutant loading (CO' values between −5 and +5 ppbv).

[50] This study also identifies some criteria that could be used to evaluate aerosol cloud interactions. For example, Figure 10 shows how cloud properties including CDNC, reff and cloud water mixing ratio vary with pollutant loading and cloud vertical velocity. Also, we discussed how our findings of enhanced nitrate levels in cloud droplet residuals are consistent with other studies [Hayden et al., 2008; Sellegri et al., 2003b], which sampled different air masses (clean and polluted) and cloud types. Similar studies on other cloud and air mass types in the future would help to better understand the aerosol–cloud interactions.

[51] Overall, this study demonstrates that even with a horizontal grid spacing of 2 km, WRF-Chem is able to simulate several important aspects of relationships between aerosols and clouds observed during the CHAPS campaign. Since models covering large regional domains use resolutions on the order of several kilometers to reduce associated computational costs, these models most likely underestimate the cloud fractions associated with shallow cumuli and their vertical velocities, and advanced cloud parameterizations are needed to improve the representation of cloud–aerosol interactions. The results presented in this study will be used as a baseline for comparison with future simulations running coupled cloud–aerosol–meteorology with a new shallow cumulus parameterization to represent shallow cumuli cloud dynamics at coarser resolutions.


[52] Brody Bourque, with support of the Global Change Education Program, assisted with the processing of the G1 data collected during CHAPS. Dr. Jason Olfert of the University of Calgary operated the DMPS. Drs. John Ogren (NOAA) and Elisabeth Andrews (CIRES) played critical roles in the deployment of the CVI. Drs. Yi-Nan Lee (BNL) and John Jayne (Aerodyne) operated the AMS. John Hubbe (PNNL) and Dr. Stephen Springston (BNL) assisted with the integration and operation of instruments on board the G1. Dr. Gunnar Senum (BNL) processed data collected using the CAS and PCASP probes. This research was supported by the U.S. DOE's Atmospheric Science Research (ASR) Program under Contract DE-AC06-76RLO 1830 at Pacific Northwest National Laboratory (PNNL). PNNL is operated for the U.S. DOE by Battelle Memorial Institute.