Complexities in the First Aerosol Indirect Effect over the 1 Southern Great Plains 2 3

8 The aerosol first indirect effect (FIE) is typically characterized by a reduction in cloud droplet 9 size and an increase in cloud optical thickness in the presence of high concentrations of 10 condensation nuclei. Past studies have derived observational evidence of the FIE in specific 11 locations and conditions, yet critical uncertainties in the validity of this conceptual model as it 12 applies to a range of cloud types and meteorological settings remain unaddressed. We utilize five 13 years of surface aerosol measurements and Moderate Resolution Imaging Spectroradiometer 14 (MODIS) observations of cloud properties to discern the FIE in springtime cloud statistics over 15 the Southern Great Plains region of the United States. We extend this analysis to explore the role 16 of three confounding factors: cloud phase, observational uncertainty and the role of regional 17 meridional flow. While high aerosol days are dominated by smaller average droplet size in liquid 18 clouds, the response of cloud optical thickness is variable and is dominantly a function of cloud 19 water path. Ice clouds experience more variability in their response to high aerosol loading and 20 satellite retrieval uncertainty thresholds. Finally, the direction of meridional flow does not play a 21 large role in stratifying the cloud response to different aerosol loading. Overall, these 22 observations show that much of the classical theory for liquid clouds is supported. Higher 23 aerosol loadings are correlated with a reduction in effective radius and generally higher cloud 24 optical thickness, and this relationship dominates over any driving influence from the low-level 25 Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-289, 2016 Manuscript under review for journal Atmos. Chem. Phys. Published: 18 April 2016 c © Author(s) 2016. CC-BY 3.0 License.


Introduction
Under the classic paradigm, aerosols can both brighten clouds (Twomey, 1977) and lengthen cloud lifetime (Albrecht, 1989) by reducing droplet size and increasing droplet number.
Along with the aerosol direct and semi-direct effects, understanding these so-called indirect effects is critical for studying Earth's energy budget and reducing uncertainty in global climate projections (Rosenfeld et al., 2014a).However, the effect of aerosols on clouds is sensitive to a variety of complicating factors, and a complete diagnosis of the physical system remains elusive (Stevens and Feingold, 2009).Critical uncertainties in the impact of aerosols on the lifetime of shallow clouds (Small et al., 2009) and the ability of aerosols to invigorate deep convection (Fan et al., 2013) remain.Competing effects of reduced precipitation formation efficiency and enhanced cloud-base evaporation make it difficult to fully determine aerosol-driven changes in liquid cloud water path (Han et al., 2002;Tao et al., 2012).Furthermore, aerosol effects on ice cloud properties (Lee and Penner, 2010;Storelvmo et al., 2011), the impact of measurement technique and uncertainty (Platnick et al., 2004) and the role of large-scale atmospheric conditions in modulating cloud microphysics (Jones et al., 2008;Muhlbauer et al., 2014) remain open questions.The first goal of this study is to observe which traditional FIE elements, including reduced droplet size and increased optical thickness, can be detected over the Southern Great Plains (SGP) by using liquid cloud field statistics as derived from satellite data.We then move beyond these variables to address whether cloud phase, measurement uncertainty and regional meteorology introduce substantial deviations from the standard FIE as determined from observations.
Observations of cloud microphysics under different aerosol conditions are abundant.Data on aerosol optical thickness (AOT), cloud effective radius (r eff ), cloud water path (CWP) and cloud optical thickness (COT) from a variety of satellite platforms have been used to develop correlative relationships between aerosol loading and cloud optical properties that show the first indirect effect at work (Chen et al., 2014;Jones et al., 2008;Kaufman and Koren, 2006;Koren et al., 2004Koren et al., , 2009;;Rosenfeld, 2003).However, most of these studies reserve their analysis for warm clouds.Ice clouds play an important role in Earth's radiation budget (Liou, 2005), and it is likely their r eff , CWP and COT respond to aerosols differently than liquid clouds (Gettelman et al., 2012).Both modeling (Storelvmo et al., 2011) and observational (Jiang et al., 2009) studies have suggested that an increase in available ice nucleation particles will decrease r eff in ice clouds, in a similar fashion to their liquid counterparts.More work is needed to fully develop the relationship between aerosol composition, the balance of heterogeneous and homogeneous freezing and the net climatic effect (Gettelman et al., 2012;Zhou and Penner, 2014).It has been suggested that the heterogeneous activation of certain hydrophilic particles (e.g.mineral dust, organics) can reduce water concentrations and supersaturations necessary for homogenous activation (Chylek et al., 2006).This curtails the further formation of ice particles and limits the efficacy of a Twomey-like mechanism.Further observational statistics of ice cloud r eff , CWP and COT are needed to clarify large-scale patterns of aerosol-ice cloud interactions.
Additionally, satellite retrievals are subject to some degree of uncertainty.For MODIS cloud products, both instrumental and retrieval algorithm error contributes to uncertainty (King Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-289, 2016 Manuscript under review for journal Atmos.Chem.Phys.Published: 18 April 2016 c Author(s) 2016.CC-BY 3.0 License.et al., 1997;Platnick et al., 2004).The impact of this uncertainty depends on the product and the proposed application of the data.Assimilating MODIS observations into forecast products requires a low tolerance for uncertainty because of high model sensitivity to noisy initial conditions (Shi et al., 2011), and uncertainty thresholds for MODIS data products have developed over the instrument history (Chu, 2002;Remer et al., 2002;Tan et al., 2005).These thresholds motivate independent validation of the satellite retrievals and provide insight into the effect of product uncertainty on scientific conclusions.With this in mind, we can use thresholds to understand the role of retrieval uncertainty on observational understanding of the FIE.
Over the SGP, aerosol-cloud interactions occur against the backdrop of rapidly changing meteorology.The SGP region is known for its severe springtime convective weather, largely driven by synoptic-level disturbances moving with the jet stream (Doswell, 1980;Maddox, 1983).As spring transitions into summer, a major dynamical feature of the SGP region, the lowlevel jet (LLJ), evolves and brings warm, moist air off of the Gulf of Mexico (Weaver and Nigam, 2008) and defines the regional warm-season climate (Balling, 1985;Lee et al., 2008;Weaver et al., 2012).Large-scale dynamics set the stage for mesoscale activity by controlling factors critical to aerosol and cloud microphysics in varying degrees, including tropospheric stability (Chen et al., 2014), humidity (Altaratz et al., 2013) and vertical motion (Muhlbauer et al., 2014).The specific role of the largely meridional LLJ flow during the warm season (roughly May-September) in modulating aerosol-cloud interactions over this region has not been fully explored, however.Determining whether the LLJ influences r eff , CWP and COT is a necessary prerequisite before attributing any bulk cloud response to changes in aerosol concentration.
In this manuscript, we use a unique combination of data to improve our understanding of the FIE in the SGP.By utilizing ground-based observations of aerosol concentrations and Atmos.Chem. Phys. Discuss., doi:10.5194/acp-2016-289, 2016 Manuscript under review for journal Atmos.Chem.Phys.Published: 18 April 2016 c Author(s) 2016.CC-BY 3.0 License.composition, and satellite-derived r eff , LWP or IWP, we identify the cloud-aerosol interactions in the SGP and develop an understanding of the role of meteorological conditions on the bulk cloud response to changes in aerosol concentration.

Aerosol Surface Concentrations and Event Identification
To identify the role of aerosols in clouds, we use aerosol measurements from five Interagency Monitoring of Protected Visual Environments (IMPROVE) sites in the Southern Great Plains (Fig. 1; Table 1) to define climatological high and low aerosol events.All sites report reconstructed PM 2.5 mass every three days, including speciated ammonium nitrate, elemental and organic carbon, sulfate and soil (Malm et al., 1994).Springtime (April-May-June; AMJ) averaged concentrations by composition are shown for each site in Fig. 2. By using surface concentration data to determine an aerosol threshold, we assume that the IMPROVE surface concentrations are indicative of concentrations aloft, and we discuss the validity of this assumption in Sect.4.1.
to calculate the cumulative sum of deviations (S) and determine change points in data (Fealy and Sweeney, 2005) (Figure 3) . (1) Positive slopes indicate that the individual aerosol concentration x k for time k is above the average aerosol concentration (x), while negative slopes indicate the data is below average.
Extrema in the slope reveal change points in the data, and Fig. 3 illustrates that 2008 was a change point in the time series of PM 2.5 data that is likely related to economic recession and changes in air quality regulations (Russell et al., 2012).As a result we focus on the 2008-2013 data for statistical consistency.From 2008-2013, aerosol maximum events for each site are determined by concentrations greater than one standard deviation above the 2008-2013 AMJ average, and aerosol minimum events are marked by concentrations less than one standard deviation below the same average.Aerosol maximum events have surface concentrations of about 10 µg m -3 (individual sites ranging from 8.8-13.6 µg m -3 ) and aerosol minimum events have surface concentrations of about 2-4 µg m -3 .At each site, there are about 20-25 events of each type during the five-year period, with the exception of the Tallgrass site, which has very few days that meet the aerosol minimum threshold.Table 1 provides the details of each IMPROVE site, including the number of identified minimum and maximum events and the aerosol minimum and maximum thresholds used.

MODIS Cloud Data
Cloud data was retrieved from the Terra MODIS instrument (Platnick et al., 2015), with retrieval algorithms discussed in King et al. (1997).In this study, we utilize cloud optical and microphysical properties retrieved using the 0.645, 2.13 and 3.75 µm bands at 1 km resolution, including effective radius and cloud optical thickness (King et al., 1997).Cloud water path and phase is also derived at 1 km resolution using r eff and CWP retrievals.Each retrieval value is tagged with a relative uncertainty value determined by contributions from both instrument and algorithm errors.Cloud top temperature is also derived at 5 km resolution.
On each aerosol maximum and minimum day, MODIS cloud properties are selected from 0.1° regions around each IMPROVE site, and both liquid and ice cloud field statistics are calculated.We compare all MODIS retrievals in the study window to data with less than 20% uncertainty in r eff and COT, allowing an examination of the role of MODIS uncertainty on the observed FIE.Because CWP is derived from r eff and COT, we did not constrain the data with CWP uncertainty.Given the three-day surface aerosol sampling time interval, it is more useful to compare distributions of cloud properties from a relatively large number of aerosol events over the five-year period rather than attempt to isolate effects from aerosols on individual events.We use a Kolmogorov-Smirnov analysis (Massey, 1951;Miller, 1956) to determine whether these distributions are statistically distinct.By defining F 1 and F 2 in Eq. ( 2) as the cumulative distribution functions for the aerosol minimum and maximum cloud property distributions, the D test statistic is compared to critical value tables to accept or reject the null hypothesis that the samples are statistically the same, The null hypothesis that the minimum and maximum distributions are statistically the same is tested with a significance value (α) of 0.05.

Meteorological Data and Microphysical Clustering
To determine the role of meteorology, we use daily average output from the North American Regional Reanalysis (NARR; Mesinger et al., 2006).The horizontal spatial resolution is 32 km in the horizontal with 29 vertical pressure levels.While much of the most important and potentially resolvable physics involved in cloud development and aerosol-cloud interactions is beyond the resolution of this reanalysis product, NARR accurately represents the larger scale atmospheric conditions such as wind speed and direction.In our analysis, we examine vertical profiles of temperature, specific humidity, and vertical, meridional and zonal winds in the SGP region.
To identify the potential role of meridional wind direction on cloud microphysics, we use k-means cluster analysis (Arthur and Vassilvitskii, 2007)

Liquid Cloud Distributions
For each site, we evaluate the probability distribution of MODIS observations of r eff , CWP and COT for aerosol minimum and maximum days with the 20% uncertainty filter at the five IMPROVE sites (Fig. 4).Kolmogorov-Smirnov tests indicate statistically distinct cloud properties between aerosol minimum and maximum days, both with and without (not shown) the uncertainty filter.One site (TALL) does not observe any liquid clouds for aerosol minimum events, so we discuss the four sites that have both aerosol minimum and maximum events.The MODIS domain-averaged filtered and unfiltered r eff decreases at all sites where liquid clouds are detected on aerosol maximum days (Fig. 4a, Table 2), with greater site-to-site variation in the response of CWP and COT (Fig. 4b,c, Table 2).There is a decrease of average r eff of 20-39% (filtered) and 10-19% (unfiltered) at the four sites with liquid clouds reported.At three of the four sites, the filtered average CWP decreases between 25-47% at three sites (CACR, UPBU and CEBL).A drop in average CWP generally corresponds to lower average COT on aerosol maximum days, with the unfiltered average CWP decreasing by 31-71% with a corresponding 15-62% decrease in COT at two of the four sites.However, at the CACR site, the decrease in CWP is accompanied by an increase in COT, which is inconsistent with the other sites.
Additionally, a the WIMO site, the unfiltered average CWP increases 71% with a 118% increase in COT on aerosol maxima days.Therefore, while r eff changes are as predicted for the FIE at all sites, the response in COT and CWP appear to be affected by other factors such as synoptic conditions as discussed in Section 3.3 below.

Ice Cloud Distributions
In contrast to the liquid phase retrievals, the sign of change in mean ice cloud effective radius is more sensitive to the uncertainty threshold, likely to due to the known uncertainties in ice cloud retrievals (Kahn et al., 2015;Platnick et al., 2004).When using an uncertainty threshold of 20%, three of the five sites (CACR, TALL and CEBL) show a 4-45% decrease in r eff under higher aerosol loading and two sites (UPBU and WIMO) show a 6-31% increase in r eff  3).However, when all data is considered, the r eff consistently decreases between 10-43% at all sites.KS tests determined that all of the distributions, both filtered and unfiltered for uncertainty threshold, are statistically distinguishable.Compared to the liquid phase, there is a consistent decrease in the average ice cloud water path to higher aerosol loading, with the filtered average decreasing 17-63% at all sites under higher aerosol conditions and the unfiltered data showing a 19-90% decrease at four of the sites (CACR, UPBU, WIMO, TALL).With the filtered data, average COT drops between 12-62% on from aerosol minimum to maximum days at four sites (UPBU, WIMO, TALL and CEBL) yet increases at one site (CACR; 66%).In contrast, unfiltered COT increases between 1.8-50% at WIMO and CEBL but decreases between 79-85% at CACR, UPBU and TALL.Like the liquid phase, there are two sites (CACR, CEBL) where an increase in COT is observed despite a decrease in CWP.Both are sites that are affected by the uncertainty range, and we suspect that this may be due to retrieval error.

Meteorological Conditions
Because the CWP can play an important role in the COT response, this suggests that synoptic conditions may also be affecting cloud properties in addition to aerosol loading.To examine this question, we use the NARR to determine the effects of some synoptic indicators such as zonal, meridional and vertical wind speeds (Figure 6).Generally, the differences in our NARR-derived composite profiles between aerosol minimum and maximum days are most pronounced at approximately 900 hPa and below.Above 900 hPa, the error bars (representing ± one standard deviation) on the average profiles show greater overlap, suggesting little difference in the conditions for these two event types.Profiles of specific humidity (not shown) suggest a moister lower atmosphere when flow comes from the South, consistent with our expectations for  4).At the same altitude, aerosol maximum days show an area of downward vertical motion of 0.05-0.2m s -1 at all sites (Figure 6).This downward motion at around 900 hPa is not evident for aerosol minimum days with the exception of the CEBL site, which shows a slight (0.05 m s -1 ) downward motion closer to 800 hPa.

Microphysical Regimes
Satellite observations have suggested that convective clouds develop through distinct microphysical regimes as described through the relationship between cloud top temperature (CTT) and r eff , with regimes defined as diffusion, collision-coalescence, warm rainout, mixed phase and glaciation (Martins et al., 2011).It is important to note that not all five regimes are necessarily present in each cloud system, depending on aerosol conditions (Rosenfeld and Lensky, 1998).Other observations have supported Rosenfeld and Lensky's initial conclusions (Suzuki et al., 2011) and Woodley, 2003;Rosenfeld et al., 2014b).Adopting this framework has two main benefits; we can compare average cloud top temperature and effective radius profiles between aerosol minimum and maximum days and examine whether the direction of the LLJ clusters the response of these clouds on these plots.
The observed r eff vs. CTT for aerosol minimum (Fig. 7a) and aerosol maximum (Fig. 7b) during our time period of analysis across all sites is shown with approximate microphysical regimes as suggested by Martins et al. (2011).We see a similar relationship between the two variables as in prior studies, yet no clear distinction between our aerosol minimum and maximum events.We had hypothesized that the aerosol minimum and maximum events would separate as the theoretical moderately and heavily polluted microphysical regime curves (Rosenfeld et al., 2014b) reproduced in Fig. 7a,b, yet there is no clear distinction between the aerosol minimum and maximum events with respect to this framework.The lack of clear differences could be explained with several factors.The first is that the separation between aerosol minimum and maximum conditions as defined by the thresholds in Table 1 does  will be needed to test the strength of this assumption, especially for ice clouds whose formation is more detached from boundary layer aerosols.
Applying k-means analysis (for up to five statistically distinct clusters) does not reveal any clustering of mean cloud r eff by meridional wind direction.This is despite the fact that aerosol maximum days are dominated by southerly flow, while aerosol minimum days are dominated by northerly flow.Given the lack of any clear influence of the meridional flow, it is likely that the aerosols are playing a role in bulk cloud field variability than the LLJ during this time of the year.This does not rule out the influence of other large-scale dynamical features as sources of variability in cloud microphysics (Muhlbauer et al., 2014), but it does implicate aerosols as a significant factor over the SGP.While we may be operating over a relatively narrow range of aerosol conditions from a global perspective, differences in cloud property distributions remain meaningful and statistically distinct by our local definition of aerosol minima and maxima.

Liquid and Ice Cloud Responses
Liquid clouds observed in this study exhibit the expected drop in effective radius under aerosol maximum conditions.One out of the four sites with the liquid phase registered has optically thicker clouds on these high aerosol events as well.The response of the COT is conditional on whether CWP increases or not, and we observe decreases in CWP at three out four sites with registered liquid clouds.The site (WIMO) that sees an increase in CWP does not have noteworthy differences in aerosol minimum and maximum concentration thresholds from the other sites (see is often attributed to a decrease in precipitation efficiency caused by smaller droplets within the cloud (Lee et al., 2009;Tao et al., 2012).However, there are known mechanisms for decreasing CWP under high aerosol conditions, including the influence of semi-direct effects (Koch and Del Genio, 2010) and enhanced cloud-base evaporation of smaller droplets (Tao et al., 2012).
The response of the ice clouds to aerosol loading is inconsistent between the sites, with only three out of five sites showing a decrease in effective radius under higher aerosol conditions and considering only data with less than 20% uncertainty.Prior work suggests that ice cloud r eff does often exhibit similar responses to its liquid counterpart (Chylek et al., 2006;Jiang et al., 2009;Storelvmo et al., 2011), yet our data show there may be more variability.This may be due to the uncoupling between surface aerosols and cold cloud processes, as ice nuclei may be transported from other locations aloft.At two of the five sites we observe an increase in effective radius under high aerosol loading, raising the possibility of an "anti-Twomey" effect connected to the dominant ice nucleation pathway and ice nucleation particle composition (Gettelman et al., 2012;Panicker et al., 2010).At all sites and regardless of uncertainty threshold, we observe a decrease in cloud water path in the ice phase on aerosol maximum events.At four out of fives sites (UPBU, WIMO, TALL and CEBL) surface concentrations of combined soil and organics, the particles typically associated with heterogeneous activation, tend to occur in higher proportions than sulfate on aerosol maximum days, indicating it is possible that heterogeneous freezing may be dominant.Heterogeneous freezing has been shown to decrease ice particle concentration and increase ice particle size, leading to increased particle aggregation efficiency and settling rate in climate models; this leads to reductions in ice cloud water path (Hendricks et al., 2011).It is unclear whether the fraction of heterogeneous to homogeneous nucleation has a definitive radiative effect in climate model simulations (Gettelman et al., 2012), though some ice-aerosol processes are clearly more sensitive to background activation processes (Zhou and Penner, 2014).However, k-means clustering analysis does not reveal any significant stratification of the the r eff vs. CTT data in Fig. 7 if we use dominant chemical composition instead of meridional wind direction as the basis for the clustering.This study is limited in its ability to fully diagnose all the potential microphysical mechanisms at work, however, and more investigation is needed before attribution can be made.
From the given MODIS retrievals, it is clear that the level of tolerance for data with high uncertainty can alter both the magnitude and direction of changes in cloud properties over certain sites.For example, average ice r eff increases using data with less than 20% uncertainty but decreases when considering the entire data set.Using observations to determine the feasibility of a physical process (e.g. an anti-Twomey effect for ice clouds) requires careful consideration of uncertainty thresholds, and it is clear more work is needed to better understand the impact of measurement uncertainty on scientific conclusions.

Effect of Meridional Flow
The importance of the jet in the convective meteorology of this region is well-established (Balling, 1985;Lee et al., 2008;Weaver and Nigam, 2008), and it is thus important to consider any indirect roles the LLJ could be playing in setting the stage for observations of aerosol-cloud interactions.Northward flow across this region is known to transport particulate matter over the SGP, and it has been suggested the transport of biomass burning particles can impact tornado formation (Saide et al., 2015).Furthermore, shear at the top of the planetary boundary layer caused by low-level jets can induce downward vertical mixing and could allow for the build-up of particulate matter and other pollutants in the lower atmosphere (Hu et al., 2013).This is consistent with the region of downward vertical motion we observe in the NARR wind speed profiles on aerosol maximum days (Fig. 6) at the same level as the highest LLJ shear.This interpretation requires some degree of confidence in the convective and turbulence parameterizations used in the development of NARR.The exact role of each of these mechanisms on the time scales we observe will require further work with back-trajectory analysis and new developments in LLJ dynamics (Klein et al., 2015).

Conclusions
We present MODIS observations of cloud effective radius, water path and optical thickness under surface aerosol minimum and maximum conditions over the Southern Great Plains region of the United States in the context of region meteorology derived from the North American Regional Reanalysis product.These data show a decrease in effective radius for liquid clouds with higher aerosol concentrations as expected from the traditional first indirect effect mechanism, with changes in cloud optical thickness tied to changes in cloud water path.The response of ice cloud properties is more variable, though measured surface aerosol composition coupled with observations of large reductions in ice cloud water path on aerosol maximum days suggests a heterogeneous freezing mechanism may be a driver of these inconsistencies.K-means cluster analysis does not reveal any direct relationship between meridional flow and mean cloud properties, indicating aerosols are likely playing an important role in springtime cloud variability.We also find that MODIS uncertainty threshold implementation has a strong potential to change scientific conclusions.Our results reemphasize the need for continuous improvement and testing of remote sensing products used for verifying models and drawing conclusions about aerosol-cloud-climate interactions.A great deal of progress has already been made in this regard, particularly in the detection of cirrus clouds (Meyer and Platnick, 2010) and understanding the causes of MODIS retrieval failures (Cho et al., 2015).Determining causal links with relatively uncontrollable observational experiments is difficult, though this kind of work can reinforce and guide modeling efforts.We have shown that characterizing complexities in the traditional first indirect effect mechanism is possible through observations of aerosol concentrations and cloud properties combined with a reanalysis product.More work is needed to better understand the spatial and temporal variability of the impact of cloud phase, measurement uncertainty and regional meteorology on the traditional first indirect effect.
on plots of average CTT vs. average r eff .The clustering algorithm partitions data into a fixed number of clusters by minimizing the sum of squares between the data points and the cluster mean.Cloud top temperature-effective radius space allows us to visualize any potential interactions of the larger scale meteorology (e.g.meridional wind direction) with mean cloud properties and examine the continuum from liquid to ice clouds in a single framework.Any clusters based on the wind direction will help indicate the effect of NARR-determined meteorological conditions on aerosol-cloud interactions.
Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2016-289,2016 Manuscript under review for journal Atmos.Chem.Phys.Published: 18 April 2016 c Author(s) 2016.CC-BY 3.0 License.(Fig 5; Table Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2016-289,2016   Manuscript under review for journal Atmos.Chem.Phys.Published: 18 April 2016 c Author(s) 2016.CC-BY 3.0 License.theLLJ(Berg et al., 2015;Weaver and Nigam, 2008).However, above about 800 hPa, the composite profiles are virtually indistinguishable between aerosol minimum and maximum days at all sites.Two of the most pronounced features in the meteorological data are the average meridional and vertical wind profiles at approximately 900 hPa.All sites exhibit strong southerly flow on aerosol maximum days (59-91% of all days) consistent with the presence of the low level jet in the region and peak windspeeds near the top of the planetary boundary layer (a maximum of 7-10 m s -1 at approximately 900 hPa).In contrast, aerosol minimum days show northerly flow with a weak jet (2-5 m s -1 ) at 900 hPa at two sites, with three sites lacking any distinct low-level jet, though 70-100% of the aerosol minimum days have at least weak northerly flow (see Table not produce dramatic changes to the microphysical evolution of clouds.This indicates that the selection of minimum and maximum concentration thresholds may not follow the same definitions for clean, moderate pollution and heavy pollution, originally developed by Rosenfeld and Lensky (1998).They originally defined their pollution regimes based on a spectrum bounded by maritime clouds (clean) and clouds forming in the presence of smoke over land-based biomass burning (heavily polluted).Even on aerosol minimum days for the SGP sites, average total PM 2.5 concentration is 3.3 µg m -3 .Secondly, we assume that surface concentrations of aerosols are an indicator of CCN availability near clouds as they develop with time.Further work Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2016-289,2016 Manuscript under review for journal Atmos.Chem.Phys.Published: 18 April 2016 c Author(s) 2016.CC-BY 3.0 License.

Figure 1 -Figure 2 -Figure 4 -Figure 5 -Figure 6 -
Figure 1-Southern Great Plains (SGP) region of study with the location of five IMPROVE sites(Table1) utilized for ground-based aerosol concentrations.