The effects on low-level cloud microstructures of varying aerosol regimes in the Arctic are examined using ground-based measurements obtained near Barrow, Alaska. Episodic ‘arctic haze’ events produced high cloud droplet concentrations, and small cloud droplet effective radii. By contrast, the fresh nucleation of aerosols within the Arctic produced particles high in number concentration but generally too small to be efficient cloud condensation nuclei. Comparisons with similar analyses done at lower latitudes suggest that the ‘indirect effect’ of haze aerosol on low-cloud effective radii is particularly high in the Arctic.
 From late fall to early summer, patches of reduced visibility approximately 1000 km across are observed in all regions of the Arctic [Barrie, 1986; Radke and Hobbs, 1989]. Called ‘arctic haze’, these events appear to be caused by strong east-west pressure gradients, which generate episodic ‘surges’ of polluted air from either Central and Western Europe or Siberia [Barrie, 1986]. Due to low precipitation rates and slow mixing across the arctic front, the aerosol in the Arctic accumulate over winter and do not dissipate until spring when the arctic front retreats, allowing moist destabilizing air to intrude from mid-latitudes. The pollution is typically most concentrated in the lowest 2 km of the atmosphere, although it is found in filaments up to altitudes of at least 7 km [Radke et al., 1989].
 Arctic hazes have been studied primarily from the viewpoint of their potential direct effects on the polar radiation budget [Shaw et al., 1993]. However, the long-range transport of pollutants to the Arctic might affect local climate indirectly. High aerosol loadings tend to be associated with smaller and more numerous cloud droplets, which make stratiform clouds more reflective to incoming solar radiation [Twomey, 1977]. In this study, interactions between arctic haze and the microstructure of clouds are explored using combined radiation and aerosol measurements obtained at the North Slope of Alaska - Adjacent Arctic Ocean (NSA-AAO) Atmospheric Radiation Measurement (ARM) program and NOAA Climate Monitoring and Diagnostics Laboratory (CMDL) sites just northeast of Barrow, Alaska. Because the data sets are long term, they provide an opportunity to study seasonal transitions in aerosol effects on arctic clouds.
2. Measurements and Retrievals
 Surface data used in this study are obtained entirely from NSA-AAO. Specific data products from the NSA-AAO ARM and CMDL sites to be used in this study are summarized in Table 1.
Table 1. Summary of Surface Data Products at NSA-AAO Used in This Study
Millimeter wave radar
cloud top, cloud fraction
cloud base, cloud fraction
state variable profiles
liquid water path
Solar broadband fluxes
total aerosol concentrations
scattering by aerosol <1 μm
 Condensation nucleus concentrations CN and the aerosol light scattering coefficient σsp were measured year round with a TSI Inc. model 3760 condensation nucleus counter, and the 550 nm channel of a TSI Inc. model 3563 3-wavelength nephelometer. The extent to which hygroscopic aerosols scatter light depends strongly on the ambient relative humidity (RH). This sensitivity was minimized by drying all aerosol to less than 40% RH. Further, before measurement, the aerosol were passed through one of two impactors that removed either particles >1 μm or >10 μm diameter. The goal of this study is to focus on anthropogenic aerosol that are effective CCN, rather than larger, insoluble and less numerous Asian mineral aerosol that are sometimes advected to the Arctic. Therefore, only measurements of scattering from aerosol smaller than 1 μm are considered here. Further, we have excluded local anthropogenic aerosol by limiting the data set to time periods when CMDL was upwind of Barrow. Conservatively, the nephelometer has a lower sensitivity threshold of 0.4 Mm−1.
 Cloud microphysical and radiative properties are derived using a retrieval method designed by Dong and Mace  for application to single-layer, overcast low-level stratus clouds in the Arctic. The algorithm uses δ2-stream radiative transfer model results and surface measurements to obtain cloud droplet effective radius re, LWP and optical depth τ. The algorithm has been extensively validated at mid-latitudes [Dong et al., 2002]. In the Arctic, direct validation was not possible. On a seasonal basis, however, retrievals of cloud microphysical properties were generally similar to those measured in situ during airborne arctic field programs. From the ground based retrievals, cloud droplet number concentration is inferred from
where, ΔZ is the cloud thickness derived from ceilometer and radar, ρw is the bulk density of water, and σd is the assumed width of an log-normal droplet size distribution. Dong and Mace  used a value of 0.38 for σd, based on mid-latitude observations. Here we have adopted σd = 0.27 based on reanalysis of airborne measurements of droplet size distributions <50 μm diameter obtained during four University of Washington field campaigns in the Arctic between 1982 and 1998. Cloud microphysics retrievals are restricted to time periods between April and October, when the cosine of the solar zenith angle μ0 > 0.2 and LWP > 40 g m−2. It is not possible to directly compare ground based measurements of aerosols to retrievals of cloud microstructures aloft since aerosol vertical profiles are typically inhomogeneous. We have tried to limit this uncertainty by restricting cloud retrievals to time periods when cloud tops were below 1.5 km.
 For this study we examined ARM and CMDL data from the months April through October between the years 1999 and 2002, However, successful simultaneous measurements and retrievals of all aerosol and cloud properties listed in Table 1 were particularly sparse during 2001 and 2002. As a result, our analyses presented here are limited to the years 1999 and 2000.
3. Effects of Arctic Haze on Cloud Properties
 Cloud condensation nuclei (CCN), which represent the fraction of all aerosol that may nucleate cloud droplets, were not measured at CMDL. However, the subset of CN that have sizes between 0.1 and 1.0 μm diameter (commonly named accumulation mode aerosol) are typically both efficient CCN and scatterers (per unit mass) of light. Thus it is reasonable to expect that local fluctuations in σsp should correspond with changes in cloud microstructures.
Figure 1 compares half-hourly averaged surface measurements of CN and σsp data from CMDL between April and October for the years 1999 and 2000. Visible haze events were most pronounced in the months of April, May and October, which is consistent with measurements of pollutant mass concentrations in the Arctic [Sirois and Barrie, 1999]. However, the highest concentrations of CN were typically observed in summer when σsp was lowest; extreme concentrations were close to a thousand per cubic centimeter. This is probably because in summer most arctic CN are produced locally, and the time scales associated with their coagulation into size ranges large enough to scatter light are shorter than the time scales associated with their removal by clouds. In fact, high concentrations of small CN are nucleated within thin layers of high humidity, that are often present directly above arctic boundary layer clouds in summer [Garrett et al., 2002a]. Because low-level clouds are more common in summer, and high actinic flux is a precursor to gas-to-particle conversion, it is to be expected that the concentrations of CN will be decoupled from levels of winter and springtime haze.
 Certainly, it is likely there were days when a substantial subset of CN had the potential to act as cloud nuclei. However, to ascertain the effects of Arctic haze on cloud properties, separate from nucleation events, we divide the data set in Figure 1 into two domains. In domain A, nucleation events produce numerous aerosol often too small to be efficient scatterers of light. By contrast, during haze events (domain B) these small aerosol have been sufficiently aged to be depleted by coagulation, and CN and σsp are highly correlated. Accordingly, we have divided the data by requiring that the correlation coefficient for CN and σsp in domain B is maximized, but with the constraint that domain B contains at least 25% of the data. The latter constraint is somewhat arbitrary, but is chosen so that it is approximately consistent with the fraction of months between April and October that are prone to haze events (Figure 1). Since data in domain B do not necessarily have high scattering, only high correlation between σsp and CN, they are distinct from ‘nucleation events’ by being characterized by aerosol that are mostly large enough to fall in the Mie scattering regime.
 For cloudy days characterized as haze events in 1999 and 2000, relationships between LWP, re, N and σsp are shown in Figure 2. The sensitivity of re to haze aerosol can be described by the indirect effect parameter IE
 Expressed this way, the IE emphasizes relative rather than absolute sensitivities, which is useful in remote sensing applications where retrieved quantities may exhibit bias but accurately reproduce trends. Since re is also a function of LWP, theoretically equation (2) should be evaluated at a constant value of LWP [Feingold, 2003]. In Figure 2, we have limited this sensitivity by evaluating IE in pseudo-logarithmically spaced bins in LWP of 40–100 g m−2 and 100–400 g m−2.
 Values of IE range from 0.13 to 0.19. For comparison, values of IE derived from global satellite data outside the Arctic show values of 0.04 over land and 0.085 over oceans [Bréon et al., 2002]. Values of IE derived from ground-based measurements over the Great Plains by Feingold et al.  ranged from 0.02 to 0.16 (average 0.10). The Great Plains data examined non-precipitating stratocumulus with updraft velocities greater than 0.1 m s−1. Low level stratus clouds in the Arctic typically have smaller updraft velocities. Theoretically, higher updraft velocities in stratiform clouds should correspond with higher values of supersaturation, greater droplet activation, and consequently higher IE [Feingold, 2003]. Thus it appears that the microstructures of arctic stratus are particularly sensitive to haze aerosol concentrations.
 We represent the effects of haze on cloud droplet concentrations, or its nucleation efficiency NE, according to:
Since re ∝ N−1/3 (equation (1)), it follows that NE ≃ 3IE. For the years 1999 and 2000, NE is about 0.4, compared to an average value of 0.15 for IE (Figure 2). Significant scatter in the data may be due in part to the fact that retrievals for N have more embedded assumptions than retrievals of LWP or re. For example, the log-normal width of the droplet spectrum σd is assumed here to be constant. Nonetheless, high levels of haze generally correspond with higher droplet concentrations.
 Nucleation events (domain A in Figure 1) had no effect on cloud microstructures. At the 95% confidence level there was no statistically significant relationship between ln N or ln re and the logarithm of CN concentrations (Figure 3). By contrast, within domain B (represented by haze events), ln CN was on average positively correlated with ln N (r2 = 0.27) and negatively correlated with ln re (r2 = 0.28), consistent with results shown in Figure 2. The reason nucleation aerosol in domain A were not effective CCN is probably because they were too small.
 In this paper we have used the collocation of ARM and CMDL laboratories near Barrow, Alaska, to examines the effects of aerosols on the microphysics of low-level clouds in the Arctic. The CMDL measurements were separated into two categories based on the ratio of condensation nucleus concentrations to total light scattering. Simple statistical analyses show that aerosol nucleated in situ do not influence cloud properties, except perhaps to maintain a clean background. Rather, the microphysical properties of low-level arctic clouds are highly sensitive to long-range transport of pollution from lower latitudes. Perhaps because these aerosol are particularly aged - their residence time in the Arctic is weeks rather than days - they appear to be more effective cloud condensation nuclei than accumulation mode aerosol at lower latitudes.
 In the Arctic, the indirect radiative effects of pollution are likely different from those at lower latitudes, largely because an increase in radiative forcing due to higher cloud albedo is mitigated by an already bright arctic surface. Also, in summer when insolation in the Arctic is highest, haze concentrations are lowest. Rather, it may be the effects of haze on cloud emissivity that are most important because in the Arctic it is only thermal radiation that is present year round. During the more polluted winter and spring, low-level arctic clouds tend to be sufficiently thin to be gray-bodies [Hobbs and Rangno, 1998]. Theoretically, such clouds would be expected to have higher values of cloud emissivity if the clouds were polluted, and cloud droplet effective radii small [Garrett et al., 2002b]. An increased cloud emissivity plausibly corresponds to increased downwelling infrared fluxes from the cloud and increased rates of spring-time snow pack melting [Zhang et al., 1996]. For example, Garrett et al. [2002b] showed using a radiative transfer model that in thin water clouds characteristic of the Arctic in April, the sensitivity of longwave cloud radiative forcing at the surface to changes in cloud effective radius d(CRF)/dre is about −1.5 W m−2 μm−1. Assuming the indirect effect parameter IE is about 0.15 and the cloud droplet effective radius is initially 12 μm, a doubling in σsp such as might be seen during a minor arctic haze event would correspond to an increase in downwelling infrared flux at the surface of about 2 W m−2. Assuming the surface were to respond as a blackbody, this added forcing would produce an increase in surface temperatures of about 0.7°C.
 The retrieval techniques used in this study are limited to thicker clouds when the cosine of the solar zenith angle is >0.2. A more detailed study of the effect of long-range aerosol pollution transport on arctic cloud properties and surface infrared fluxes will need to focus on thin clouds during the more frequently polluted dark months.
 CMDL data were provided by J. Ogren. This research was supported by grant ATM-0303962 from the National Science Foundation, and grant NNG04GDG4G from the NASA Radiation Program.