The semi-direct aerosol effect is examined with a mesoscale meteorological model for a polluted Arctic haze episode at Barrow, Alaska during the Indirect and Semi-Direct Aerosol Campaign (ISDAC). Initialized with chemical and aerosol reanalysis fields from a global chemistry model, the WRF/Chem mesoscale model is used to simulate a polluted event over Alaska and its environs on 18–21 April 2008. It is shown that the atmosphere is sensitive to changes in the black carbon concentration, even though it comprises just a small fraction (less than one percent) of the total aerosol mass. Comparisons with a baseline run (which does not include aerosol radiative effects) show that in regions where black carbon is concentrated, the semi-direct effect heats the lower troposphere by about 0.15 K. Additional sensitivity tests reveal that the heating is more uniform and higher in magnitude by up to 0.1 K when the initial concentration of black carbon is doubled, and a reduction in heating occurs when black carbon is reduced to zero. At Barrow, atmospheric warming is sensitive to variations in the black carbon concentration, and heating generally occurs above 0.5 km altitude where black carbon is located. A more stably stratified lower troposphere due to the warming aloft and surface cooling from the aerosol direct effect leads to a redistribution and reduction in the cloud optical thickness and liquid water content. This cloud reduction decreases the amount of downward surface longwave radiation and further lowers the surface temperature at Barrow.
 Arctic regions such as Alaska are affected by large concentrations of airborne pollutants known as Arctic haze. Consisting of many different chemical and aerosol species, Arctic haze has a significant impact on the local weather and climate of the Arctic [McConnell et al., 2007]. Black carbon (BC) is very effective at absorbing shortwave radiation and can perturb the vertical temperature gradient and cloud distribution - a process known as the aerosol semi-direct effect. While BC has been suggested as being the second most dominant source of global warming after carbon dioxide [Ramanathan and Carmichael, 2008], the semi-direct effect is poorly understood as its impact on the weather and climate is unclear [Koch and Del Genio, 2010] and its effects were not included in the 2007 IPCC Assessment Report [Forster et al., 2007]. Investigating the semi-direct effect in the Arctic is necessary given the region's sensitivity to BC and other air pollutants [Law and Stohl, 2007].
 In this study, we investigate the semi-direct effects of BC during a polluted episode of the Indirect and Semi-Direct Aerosol Campaign (ISDAC) with the WRF/Chem mesoscale meteorological model. Reanalysis fields from the global chemistry model MOZART-4 provide aerosol and chemistry species for the initialization and lateral boundaries of all the simulations. The impact of the aerosols on the atmospheric temperature profiles and cloud properties are examined by comparing the simulation to a corresponding baseline run which does not include aerosol radiative effects. Two additional simulations of varying BC concentrations highlight its importance relative to the other aerosol and chemistry species present in the simulation.
 The semi-direct effect was termed by Hansen et al.  to describe aerosol absorption of sunlight which warms the lower troposphere and reduces large-scale cloud cover. Subsequent studies, however, show a greater complexity in the cloud response to the heated layer. Large eddy simulations by Johnson et al.  show that cloud response varies with respect to the altitude of the absorbing aerosol layer relative to the cloud. A simulated stratocumulus cloud deck within an absorbing layer showed a decrease in cloud cover and lifetime. This decrease is evident but less pronounced when an aerosol layer lies both within and above the cloud deck. Clouds located beneath the aerosol layer can actually be enhanced due to a decreased entrainment rate from the strengthened inversion and a shallower boundary layer with higher levels of moisture. Cumulus clouds, however, located beneath absorbing aerosols can be negatively affected since an increase in vertical stability inhibits convection [Fan et al., 2008].
 The semi-direct effect is also dependent on the underlying surface of the absorbing aerosols. Climate model simulations by Allen and Sherwood  show that heating caused by aerosol absorption over land can reduce the relative humidity in regions where moisture is limited. Over the oceans where moisture is plentiful, an increase in stability may lead to higher moisture levels near the ocean surface and more stratified cloud cover [Johnson et al., 2004]. Over oceanic regions of abundant moisture and large-scale convergence, an increase in stability can increase cloud cover as convergence and deep convection are enhanced. Climate model studies show regional variations in the semi-direct effect with a net increase in low level cloud cover over the oceans and some land areas and/or reduced upper level cloud cover [Koch and Del Genio, 2010].
 Arctic haze consists of stratified layers of air pollution in Arctic regions such as Alaska. This haze is primarily composed of sulfates and other particulate matter, and also contains lesser amounts of ammonium, nitrates, BC, and dust [Law and Stohl, 2007; Quinn et al., 2007]. Arctic haze has been observed in the lower and middle troposphere up to an altitude of around 5 km [Hansen and Rosen, 1984; Warneke et al., 2009]. Aircraft measurements of BC over Barrow, Alaska have shown that the greatest concentrations occur in the free atmosphere [Hansen and Rosen, 1984]. On a seasonal scale, aerosol concentrations peak in the winter and early spring when the removal rate is low [Polissar et al., 1999; Sharma et al., 2006]. Stone et al.  estimate that smoke from summertime local biomass burning can heat a polluted layer on the order of 1 K per day. Atmospheric heating by absorbing aerosols occurs on a short timescale (less than a day).
 Some uncertainty exists as to the origin of Arctic haze [Law and Stohl, 2007] and whether biomass burning or anthropogenic emissions is responsible for a majority of the pollutants [Koch and Hansen, 2005; Hegg et al., 2009]. Since a wintertime cold air mass dome usually insulates the polar regions from the mid latitudes, pollutants are thought to travel along quasi-isentropic surfaces [Carlson, 1981] from regions such as Siberia which have major sources of anthropogenic pollutants and biomass burning. The polar dome has been observed to extend as far south as 40 degrees latitude in Asia [Stohl, 2006; Law and Stohl, 2007], and a recent study of Arctic haze over Alaska in April 2008 by Warneke et al.  shows evidence of pollutants originating from agriculture burning in Kazakhstan and an early forest fire season in Siberia. The transport period of aerosols between continents is on the order of 10 to 15 days [Stohl, 2006].
 The ISDAC field campaign took place on the north slope of Alaska in April 2008, and involved research aircraft and ground-based measurements at Barrow. ISDAC's primary objectives are to study the aerosol semi-direct and indirect effects on the local weather and climate, to better understand interactions between clouds, aerosols, and the arctic climate, and to resolve cloud and aerosol processes in mixed phase clouds [McFarquhar et al., 2011]. An analysis of ISDAC aircraft measurements by Shantz et al.  describes polluted conditions in the vicinity of Barrow and Fairbanks, Alaska during 18–22 April at altitudes of 500 m to 6.5 km. The research presented here aims to examine the influence of BC on the semi-direct effect and its impact on the vertical temperature and moisture profiles in the lower troposphere for a polluted episode on 18–21 April at Barrow, Alaska. In the following section, the set-up of the model and configuration of the experiments are discussed. The results are shown in section 3, and the conclusion is presented in the final section.
2. Description of the Numerical Model and Experiment Setup
 The Weather, Research, and Forecasting Chemistry (WRF/Chem) model (v3.2.1) is a fully compressible mesoscale meteorological model designed for research experiments as well as operational weather forecasts [Skamarock et al., 2008]. The chemistry component [Grell et al., 2005] contains a variety of modules for simulating aerosol direct and indirect effects and chemistry processes.
 We chose the MADE/SORGAM (Modal Aerosol Dynamics Model for Europe / Secondary Organic Aerosol Model) aerosol scheme [Ackermann et al., 1998; Schell et al., 2001] for its ability to simulate direct and indirect effects for a wide variety of aerosol species (sulfates, nitrates, ammonium, sea salt, dust, BC, organics, and others). MADE/SORGAM simulates aerosol effects by dividing each species into up to three modal distributions, depending on the aerosol size, and assumes that the aerosols are internally mixed in each mode. Aerosol microphysical properties are calculated in MADE/SORGAM, which also couples the aerosol, cloud, radiation, and chemistry interactions. Soil-derived and sea salt aerosol emissions are incorporated into WRF/Chem [Shaw et al., 2008], and biogenic emissions are managed by the Model of Emissions of Gases and Aerosols from Nature (MEGAN) [Guenther et al., 2006].
 The WRF/Chem model is discretized with fifth/third-order finite differences for the horizontal/vertical advection and third-order Runge-Kutta for the time stepping. To minimize reflections of upward-propagating gravity waves from the model top, an implicit sponge layer is included in the upper 5 km of the model domain. The WRF/Chem simulations use a microphysics scheme from Lin et al.  which contains hydrometeor classes for water vapor, cloud water, rain, cloud ice, snow, and graupel, as well as the Grell cumulus scheme, the Meller-Yamada-Janjic TKE planetary boundary layer scheme, NOAH land surface model, the Momin-Obukhov surface clay physics, Goddard shortwave radiation, and RRTM longwave radiation schemes (see Skamarock et al.  for detailed explanations of these parameterization schemes).
 The model domain, shown in Figure 1, has a 10 km horizontal grid resolution on a polar stereographic grid and a distance of 4500 km in both horizontal directions. There are 40 vertical levels, and the lowest vertical grid spacing is about 25 m. The model top is at the 50 hPa pressure level. A time step of 60 s is used, and the simulation begins on 00 UTC 18 April 2008 and ends on 00 UTC 21 April 2008. NCEP Global Forecast System (GFS) Final Analysis 1 degree resolution data, which has 26 pressure levels and is updated every 6 hours, provides the initialization and lateral boundaries of the meteorological fields. The simulations are also initialized with NCEP-archived 0.5 degree sea-surface temperatures fields (updated daily).
 The chemistry and aerosol species are initialized with reanalysis data from the offline global chemical transport model MOZART-4 [Emmons et al., 2010], as are their lateral boundaries. MOZART-4 (the Model for Ozone and Related Chemical Tracers) reanalysis data has a 2.5 degree horizontal resolution and is updated every 6 hours, and uses regional and global anthropogenic and natural emissions inventories for chemistry and aerosol species. The MOZART-4 reanalysis fields, provided by NCAR/NESL, contain 28 vertical levels with a maximum height at the 2 hPa pressure level.
 A total of four simulations are conducted: a baseline WRF/Chem simulation and three sensitivity tests on BC concentrations. All four simulations use MOZART-4 reanalysis fields for the chemical and aerosol species in the initial conditions and lateral boundaries. The baseline run (referred to as BASE) has no feedback from the aerosols to the radiation schemes. The three BC sensitivity tests, described in Table 1, are characterized by the amount of BC in the initial conditions and lateral boundaries. The BC concentration in the first sensitivity test (referred to as CHEM) is directly obtained from the MOZART-4 fields. In the second sensitivity test (known as CHEM2) the BC concentration is twice of that in CHEM, and BC is reduced to zero in the third sensitivity test (CHEM0).
Table 1. Configuration of the Baseline and WRF/Chem Simulations
baseline WRF/Chem, no aerosol radiative effects
WRF/Chem, initialized with MOZART-4 data
same as CHEM, but with twice the amount of BC
same as CHEM, but with no BC
 Semi-direct aerosol effects are examined by comparing the three sensitivity tests to the corresponding baseline run. Note that a majority of the figures examine differences between the sensitivity tests and baseline run, where the results from the baseline run are subtracted from CHEM, CHEM2, or CHEM0. The responsiveness of the temperature and cloud distribution to the BC concentration is studied by comparing the differences between the three sensitivity tests. The first 24 hours of simulation time is used for model spin-up (in this study a longer spin-up time for the aerosol species is not required as they are present in the model initialization and are not only dependent on emissions). We focus on simulation results from 00 UTC 19 April to 00 UTC 21 April.
 This section begins with the discussion of the prominent synoptic-scale atmospheric features during the time frame of this study. Differences between the three sensitivity tests and the baseline run are examined with potential temperature fields on the 850 hPa pressure surface and at 2 m above ground level. The remainder of this section focuses on meteorological conditions locally at Barrow. Model results are compared to ISDAC observations, and then we compare meteograms showing differences between the sensitivity tests and baseline run for potential temperature and moisture. Lastly, surface meteorological conditions are considered, and shortwave and downward longwave radiation differences on the surface at Barrow are examined.
3.1. Synoptic-Scale Analysis
Figure 2 shows the CHEM results for column-integrated particulate matter of 2.5 micrometers or less (PM2.5) (Figure 2, left) and BC (Figure 2, right) for 00 UTC on 19, 20, and 21 April 2008 (Figures 2 (top), 2 (middle), and 2 (bottom), respectively). Also displayed on the plots are the mean sea level pressure contours (4 hPa contour intervals) and wind barbs at 2 m above ground level. On 00 UTC 19 April, a southeast flow is situated over western Alaska due to a low pressure center in northeast Siberia and high pressure located south of Alaska. High pressure also extends as a ridge northeastward over the Canadian Northwest Territories and into parts of the Arctic Ocean.
 The PM2.5 field on 00 UTC 19 April has a maximum concentration extending in elongated bands on a northeasterly axis from the Pacific ocean to western Alaska and the Bering Sea. Moderate amounts of PM2.5 are concentrated over the Pacific Ocean and Alaska, while lesser amounts are located over northwest Canada and the Arctic Ocean. The maximum PM2.5 concentration is over 5 × 105μg m−2. BC is concentrated over the Pacific Ocean, much of southern and central Alaska, and over Siberia and the Arctic Ocean. Maximum values of BC, on the order of 4.5 to 9 × 102μg m−2, are less than 1 percent of the PM2.5 concentration. It can be seen, thus, that BC comprises only a tiny fraction of the total aerosol mass.
 On 20 April, the low pressure shifts eastward to the north of Alaska. Two additional low pressure centers enter the domain south of the Aleutian Islands and over the Pacific ocean near Siberia. High pressure dominates eastern Alaska, and another high pressure center is located over northern Siberia and adjacent parts of the Arctic Ocean. The pressure gradient over southwest Alaska results in a southeasterly flow over the region. This flow then veers to a more westerly direction over northern Alaska. The aerosol distribution is roughly aligned with this flow over western Alaska where PM2.5 is concentrated. BC continues to be situated over the southern two thirds of Alaska, the Arctic Ocean, and Siberia, while BC entering the model domain is being advected eastward over western parts of the Bering Sea and the Pacific Ocean.
 By 21 April, an elongated trough of low pressure extends over the Bering Sea while high pressure continues to dominate over southeast Alaska and the Canadian Northwest Territories. A high pressure center is located over the Arctic Ocean in the same region as the low pressure center 24 hours earlier, which has progressed eastward toward Victoria Island. The southeasterly surface flow continues to extend over western Alaska. PM2.5 is concentrated over western Alaska, and also extends over the remainder of Alaska, northern Canada, the Pacific Ocean, and the Bering Sea. BC is distributed over much of Alaska and extends over the Pacific Ocean to the south of the Aleutian islands. BC is also present over the Arctic Ocean and Siberia.
 For brevity, plots of the meteorological fields and aerosol species for the CHEM2, CHEM0, and BASE simulations are not shown. The meteorological fields are virtually identical to those shown in Figure 2, and all of the simulations have very similar aerosol species distributions and concentrations. BC is the one exception, of course, where the BC distribution field in CHEM2 is spatially similar to CHEM but twice the magnitude. BC is, of course, absent in CHEM0.
 The potential temperature difference fields on the 850 hPa pressure level are shown in Figure 3 for CHEM-BASE (Figure 3, top), CHEM2-BASE (Figure 3, middle), and CHEM0-BASE (Figure 3, bottom) for 00 UTC 20 April (Figure 3, left) and 00 UTC 21 April (Figure 3, right). The pressure surface at 850 hPa is equivalent to 1.3 km to 1.4 km height, which is a representative altitude for the lower troposphere (above the planetary boundary layer) for all parts of the model domain except the mountainous regions (see Figure 1). The corresponding local time for 00 UTC at Barrow is 4 PM, and the entire domain is in daylight at this time. It is assumed that sufficient daylight time has elapsed for BC absorption of shortwave radiation to heat the atmosphere.
 Potential temperature differences for CHEM-BASE on both days are largely positive in regions where BC is present. The largest positive potential temperature differences extend over and to the south of Alaska, the Bering Sea, and parts of the Arctic Ocean adjacent to Siberia. The positive differences generally range from 0.1 K to 0.15 K. Negative potential temperature differences located over northwest Canada and parts of Siberia are also in areas of high terrain. These potential temperature differences are more reflective of surface cooling from the aerosol direct effect, as the 850 hPa pressure level is near the ground surface in these regions.
 Warming on the 850 hPa pressure surface is even more apparent in the CHEM2-BASE potential temperature difference for 00 UTC April 20 and 21 (Figure 3, middle). Warming is more uniform and has a greater magnitude of 0.15 K to 0.25 K. Warming at 850 hPa is much reduced in the CHEM0-BASE potential temperature difference due to the elimination of BC (Figure 3, bottom). These plots show that the 850 hPa potential temperature surface is sensitive to the BC concentration, despite the fact that BC accounts for less than 1 percent of the total aerosol mass. The overall potential temperature difference in CHEM2-BASE is about 0.1 K higher than in CHEM-BASE.
 The aerosol direct effect, where reflection and absorption of shortwave radiation in the atmosphere lead to surface cooling, is evident in Figure 4 which shows the surface shortwave radiation difference (Figure 4, left) and 2 m above ground-level potential temperature difference (Figure 4, right) on 00 UTC 21 April. Larger quantities of shortwave radiation reach the surface in the baseline run than in any of the other simulations. Shortwave radiation differences for CHEM-BASE are about −15 W m−2 to −40 W m−2 over much of North America, Siberia, and the Arctic Ocean. The CHEM2-BASE difference is about 5 W m−2 greater than the CHEM0-BASE difference, which is probably due to BC absorption of sunlight. Large positive and negative shortwave differences in excess of 100 W m−2 are located in the Bering Sea and southwards over the Pacific Ocean. These regions also contain cloud cover and a large amount of BC and PM2.5. The large shortwave differences may at least be partially explained by the high aerosol concentration and a redistribution of the cloud cover in those regions, which is apparent by examining plots of the liquid water path and ice water path differences for CHEM-BASE on 00 UTC 21 April (Figure 5). Differences in excess of 0.04 kg m−2 for the liquid water path and 0.18 kg m−2 for the ice water path is evident in roughly the same regions with the large variations in shortwave differences. The cloud redistribution could also be a result of direct and semi-direct aerosol effects (as in the work of Gu et al. ).
 The 2 m potential temperature difference (Figure 4, right) is primarily negative over the land and ice-covered Arctic Ocean, and close to zero over the Pacific Ocean and Bering Sea (which are not frozen). There are no significant variations between the difference plots of CHEM-BASE, CHEM2-BASE, and CHEM0-BASE. Negative differences are greatest over the Arctic Ocean, where they range from −0.5 K to −1.0 K. Over land the difference is closer to 0 K to −0.5 K. The variation in the temperature differences between land, sea, and ice are primarily a result of their different heat capacities. The ocean's large heat capacity is reflected by its insensitivity to variations in the shortwave radiation, while the land and ice have smaller heat capacities and are more affected by the reduction in shortwave radiation.
3.2. Comparison With ISDAC Observations
 We now compare the simulation results to surface observations taken at Barrow during the ISDAC campaign. The ISDAC instrument data used in this study consist of surface meteorological conditions, solar and terrestrial radiation, visibility, and the aerosol concentration. The ISDAC surface meteorological observations and corresponding simulation results at Barrow for 19–21 April are shown in Figure 6 for the 2m temperature (first panel), u and v surface wind components (second and third panels), and the surface pressure (fourth panel). The black line denotes the ISDAC observations, the BASE run is blue, and CHEM is red (the simulations' results mostly overlap). The black vertically-dashed lines denote sunrise (14 UTC) and sunset (07 UTC). The simulation results largely agree with the ISDAC observations, though the 2m temperature variations differ somewhat and the observed surface pressure reaches a minimum about 12 hours later (at 20/00 UTC) than the simulated results.
Figure 7 shows the observed and simulated comparisons between the shortwave (SW) and downward longwave (LW) radiation fluxes at Barrow, and the CHEM and observed PM2.5 concentrations. The observed total shortwave radiation (first panel, black line) is significantly greater than the corresponding amount in CHEM (first panel, blue line). In the second panel, the observed direct sunlight (red line) is shown with the diffuse sunlight (blue line). It can be seen that much of the sunlight is diffuse before 20 April, as clouds are generally present during that time. Direct sunlight is more apparent on 20 April, and the total incoming shortwave radiation (first panel) is correspondingly higher.
 The downward LW radiation is shown in Figure 7, third panel, for the ISDAC observations (black line) and CHEM (blue line). The downward LW radiation observations remain mostly constant at about 300 Wm−2 from 19/00 UTC to 19/18 UTC, reaching a maximum of 320 Wm−2 between 19/18 UTC and 20/03 UTC, and then decreases to about 280 Wm−2 thereafter. The relatively high amount before 19/18 UTC is most likely due to cloud cover, and the maximum area of observed downward LW radiation between 19/18 UTC and 20/03 UTC is in part due to the corresponding maximum in observed PM2.5 concentration (fifth panel). There is considerably more variation in the downward LW radiation in CHEM than in the observations. The CHEM results for the PM2.5 (red line) and BC (black line) concentrations are shown on the fourth panel (the values are taken at the lowest grid point). Although only a qualitative comparison can be made between the CHEM PM2.5 and observed PM2.5 (which is in the unit of number per m3), it can be seen that a concentration maximum is evident in both CHEM and the observations between 19/18 UTC and 20/03 UTC. The CHEM maximum, however, exists for a longer period of time than the observed maximum.
3.3. The Semi-Direct Effect at Barrow, Alaska
 It was shown in Figures 2 and 7 that the aerosol concentration over Barrow varies substantially on a daily basis, in part due to the progression of synoptic-scale features in Alaska and the Arctic during that time frame. These aerosol variations and their impact on the temperature and cloud structures over Barrow are now examined further.
 A meteogram of PM2.5 and BC from CHEM is shown in Figure 8 (first and second panels) for the lowest 3 km over Barrow during 00 UTC 19–21 April. CHEM-BASE differences for potential temperature, the heating rate, liquid water content, ice water content, and cloud optical thickness are shown sequentially in the bottom five panels. There are significant variations in the PM2.5 and BC concentrations over Barrow in both altitude and time. PM2.5 is greatest in the lowest kilometer of the atmosphere, and the concentration is at a maximum between the times 19/15 UTC and 20/06 UTC. BC is most concentrated at 0.5 km to 2.5 km altitude, and has localized maximums between 19/12 UTC to 19/18 UTC, and 20/03 UTC to 20/15 UTC. The concentration of BC drops abruptly in the lowest 0.5 km. As was explained in the introduction, much of the BC present over Alaska in April has most likely been transported from distant sources above the boundary layer.
 The BC concentration has an effect on the CHEM-BASE potential temperature and heating rate differences (Figure 8, third and fourth panels, respectively). Above 0.5 km altitude, the potential temperature differences are generally positive, and below that altitude the aerosol direct effect causes the differences to predominantly be negative. At altitudes between 0.5 km to 1 km, the CHEM-BASE potential temperature difference is negative at 19/15 UTC, and is about 0.3 K positive at 19/18 UTC and from 19/21 UTC to 20/02 UTC. Complex variations both in time and altitude are characteristic of the heating rate difference between 19/15 UTC and 19/21 UTC. Between 0.5 km and 1 km, the heating rate switches from negative to positive and back to negative, with magnitudes near 0.00005 Ks−1. The positive maximum occurs on the lower edge of the corresponding BC maximum. The following negative heating rate maximum at 19/19 UTC occurs as the BC concentration decreases over time, and so the heating rate in the CHEM simulation is less positive than in the baseline simulation. The potential temperature difference and the heating rate show the opposite tendencies during the same time period between the ground and 0.5 km altitude, as the negative maximum at 19/17 UTC is located just beneath the corresponding positive counterpart. Between 20/00 UTC and 20/07 UTC both the heating rate and potential temperature differences become more complicated, most likely due to the presence of clouds during that time. There is an overall positive potential temperature difference in a thin layer at about 500 m between 19/21 UTC to 20/12 UTC.
 CHEM-BASE liquid water content and ice water content differences are shown in Figure 8 in the fifth and sixth panels, respectively. Positive liquid water content differences at 19/15 UTC are preceded by a positive ice water content difference at 19/14 UTC. The positive ice water content difference occurs at the same time and altitude as the previously mentioned negative maximum of the heating rate, and the positive liquid water content is associated with the simultaneous negative potential temperature difference. A decrease of 0.15 g m−3 in the liquid water content difference is evident in the lowest 0.5 km altitude between 20/00 UTC and 20/08 UTC. It is evident from the potential temperature difference panel that increased stratification occurring in the lowest kilometer prior to 20/00 UTC leads to a reduction in the liquid water content difference. The ice water content difference fluctuates between positive and negative maximums during this time period, and the differences are about two orders of magnitude smaller than the liquid water content difference. The CHEM-BASE cloud optical thickness difference (Figure 8, seventh panel) is positive at 19/15 UTC, which corresponds with the positive liquid water content difference maximum. Negative values of about 8 (dimensionless units) occur between 20/00 UTC and 20/08 UTC, which occurs alongside the negative maximum in the liquid water content difference. During this time there is a cloud layer in the BASE results, but in the corresponding CHEM simulation evidence of clouds is only apparent for intervals lasting around an hour at 20/03 UTC and 20/06 UTC (not shown). These clouds form at roughly 500 m altitude, which is about the lowest extent of the BC maximum concentration.
 The CHEM2-BASE meteograms are shown in Figure 9. The PM2.5 and BC distribution fields are generally similar to the corresponding CHEM-BASE fields, and BC is twice the magnitude as in the CHEM simulation. The potential temperature difference is also similar, though a more broad and uniform warming of the atmosphere above 0.5 km is evident (third panel). The heating rate difference is also generally similar to the corresponding CHEM-BASE run. The heating rate is overall less negative between 19/15 UTC and 19/21 UTC, which can be attributed to the greater BC concentration during that time period. Drying of the lower atmosphere is more apparent in this simulation, as there are no positive differences in the liquid water content (fifth panel) and the ice water content (sixth panel) is negative at 19/14 UTC. This is likely due to increased stratification from the enhanced absorption of the doubled BC above the cloud layer at 19/14 UTC, and further suggests that the ice water content will generally decrease if BC is present in sufficiently high quantities. The liquid water content difference is around −0.25 g m−3 at 20/06 UTC. The ice water content difference fluctuates between positive and negative between 20/00 UTC and 20/07 UTC, and as in the CHEM-BASE case, the differences are two orders of magnitude less than the liquid water content. The optical thickness differences (seventh panel) are overall negative, with magnitudes of −10 (dimensionless units). In a similar manner to the CHEM simulation, cloud activity is reduced in CHEM2 between 20/00 UTC to 20/08 UTC with about only two hours of clouds appearing around 20/02 UTC.
 Meteograms for CHEM0-BASE are shown in Figure 10. The potential temperature difference (third panel) above 0.5 km does not show the broad positive extent as in the other two meteograms, but does contain some small regions of both positive and negative differences. Below 0.5 km cooling from the direct effect continues to be evident. The heating rate difference (fourth panel) does not exhibit the same positive and negative distribution around 19/18 UTC, as BC forcing is absent. The liquid water content difference (fifth panel) is negative by about −0.12 g m−3 at 20/03 UTC, and the ice water content difference (sixth panel) has the same positive maximum at 19/14 UTC as in the corresponding CHEM-BASE result. The cloud optical thickness resembles the corresponding CHEM-BASE result, though is slightly reduced in magnitude. Both the cloud optical thickness and liquid water content differences are slightly positive at 19/15 UTC. The variations in the liquid water content and cloud optical thickness show a limited sensitivity to the BC concentration, as they are consistently negative between 20/00 UTC and 20/07 UTC in all three sensitivity cases. Increased stratification from the direct effect must therefore play a significant role in reducing the cloud optical thickness, particularly in the lowest 500 m. Cloud activity in CHEM0 is reduced in the 20/00 UTC to 20/08 UTC time frame, and exists for about an hour at 20/02 UTC and for two hours at 20/05 UTC.
 Differences in surface meteorological parameters are now examined. Figure 11 shows the CHEM-BASE surface shortwave radiation difference (first panel), downward longwave radiation difference (second panel), 2 m temperature difference (third panel), 2 m vapor mixing ratio difference (fourth panel), and total column-integrated water vapor and cloud variables difference (fifth panel) at Barrow for the time period of April 19/00 UTC to 21/00 UTC. The surface shortwave radiation difference is primarily negative, and the positive area between 20/01 UTC and 20/06 UTC is related to cloud reduction in CHEM during that time (as is evident in the fifth and seventh panels of Figure 8). The CHEM-BASE downward surface longwave radiation difference is negative at that time due to the decrease in cloud cover in CHEM.
 The CHEM-BASE 2 m temperature difference is primarily negative as a result of both the negative shortwave and downward longwave radiation differences in CHEM. Between 19/16 UTC and 20/00 UTC the 2 m temperature difference is negative due to the negative shortwave radiation difference (the longwave radiation differences are minimal during this time). Between 20/02 UTC and 20/05 UTC the positive shortwave radiation difference is more than offset by the negative downward longwave radiation difference and causes the 2 m temperature difference to generally be negative. The 2 m temperature difference follows the downward longwave difference closely in the nighttime hours, where positive and negative spikes of both differences occur simultaneously.
 CHEM has a lower 2 m vapor mixing ratio than the baseline simulation during both the daylight hours and between the times 20/01 UTC and 20/07 UTC when there is also cloud reduction occurring in CHEM. The difference in the total column-integrated water vapor and cloud variables is generally close to zero except during periods of reduced cloudiness in CHEM when the difference is negative. Overall, aerosol direct and semi-direct effects cool the surface and lower the 2 m vapor mixing ratio.
 The corresponding time series for CHEM2-BASE is shown in Figure 12. As in Figure 11, the 2 m temperature difference is negative in the daylight hours. The large negative differences at 19/15 UTC and 20/06 UTC are due to the decrease in surface longwave radiation in CHEM2 at those times, while the decrease in CHEM2 shortwave radiation is responsible for the negative temperature differences between 19/16 UTC and 20/01 UTC. During nighttime hours, the 2 m temperature difference follows the downward surface longwave radiation difference. The 2 m vapor mixing ratio difference is negative during the daylight hours and when cloud cover is reduced (at 19/15 UTC and 20/06 UTC). The total column-integrated water vapor and cloud variables difference is also negative when cloud cover is reduced.
 The CHEM0-BASE time series is shown in Figure 13. As in the other two time series comparison figures, the reduction in shortwave radiation and downward surface longwave radiation in CHEM0 contribute to the negative 2 m temperature differences, and the 2 m temperature difference at night closely follows the longwave radiation difference. The 2 m vapor mixing ratio and total column-integrated water vapor and cloud variables are less negative than in CHEM-BASE and CHEM2-BASE. Overall, these results show that a reduction in cloud cover results in less downward surface longwave radiation and a decrease in the 2 m temperature.
 In this study, we examined the aerosol semi-direct effect with the mesoscale meteorological chemistry model, WRF/Chem, which contains modules for representing aerosol direct and indirect effects. A total of four simulations were conducted which included three BC sensitivity tests and a corresponding baseline run that excluded aerosol radiative effects. All simulations were initialized with aerosol and chemistry fields from the MOZART-4 global chemistry transport model, and in the second sensitivity test BC was doubled and in the third sensitivity test BC was zeroed-out. The semi-direct effect was quantified by comparing differences between the sensitivity tests and the baseline run.
 The simulations, representing polluted conditions during the ISDAC field experiment on 18–21 April 2008, show that temperature and moisture profiles over Barrow, Alaska are sensitive to the BC concentration even though it comprises just a small fraction (less than one percent) of the total aerosol mass. The simulations also roughly agree with ISDAC surface observations, which indicate that maximum PM2.5 concentrations occur between 19/18 UTC and 20/03 UTC. An examination of the potential temperature difference between the CHEM and baseline run (CHEM-BASE) show that at the 850 hPa pressure level, heating occurs by around 0.15 K in regions where BC is present. This heating was more uniform and generally higher in magnitude by about 0.1 K in the simulation where BC was doubled (CHEM2), and much reduced in the CHEM0 simulation which contained no BC. Direct aerosol effects were evident by the reduction in surface shortwave radiation and the potential temperature at 2 m. The direct effect was shown to be insensitive to variations in the BC concentration, as all three of the BC sensitivity tests displayed similar reductions when compared to the baseline run.
 The BC concentration in the atmosphere above Barrow, Alaska varied considerably in both altitude and time. Heating associated with BC resulted in a general warming of the atmosphere above 500 m where BC was located, as well as localized areas of heating of up to 0.3 K. Near-surface cooling associated with the direct effect was about −0.3 K in the lowest 500 m. The readjustment of the vertical potential temperature profile in the sensitivity tests resulted in increased stability in the lowest kilometer of the atmosphere and a reduction in the liquid water content and cloud optical thickness in the lowest 500 m. This reduction was somewhat enhanced in the CHEM2-BASE results, though was also evident in the other two comparisons. This suggests that while increasing amounts of BC does lead to a decrease in the liquid water content and cloud optical thickness, the increased stability associated with the aerosol direct effect in the lowest 500 m also contributes significantly to these reductions. Cloud lifetime was also reduced in the BC sensitivity tests when compared to BASE. There was generally no discernible relationship between variations in BC and the ice water content difference, though the results do suggest that ice water content may decrease with a sufficient BC concentration.
 At Barrow, the surface shortwave radiation and downward surface longwave radiation were reduced in all three of the simulations when compared to the baseline run. The reduction in shortwave radiation was primarily attributed to the aerosol direct effect. The decrease in cloud cover caused a reduction in the downward longwave radiation. Both the shortwave and downward surface longwave radiation reductions contributed to the lower surface temperatures. During the nighttime hours, the 2 m temperature difference closely followed variations in the downward surface longwave radiation. The 2 m vapor mixing ratio values were generally lower, and reductions in the integrated water vapor and cloud variables occurred with the decrease in the liquid water content and cloud optical thickness.
 These results exemplify the localized nature of the semi-direct effect, as it varies substantially temporally, spatially, and in altitude. As Barrow is not located within close proximity to any major open water sources (as long as the Arctic ocean is ice-covered), we might expect that the semi-direct effect will often contribute to a decrease in cloud cover in the lower troposphere. This of course may change in the summer as the days become longer and the snow and ice melts. These spatial and temporal variations in the semi-direct effect complicate predictions of moisture and low-level cloud cover in general circulation and climate models, as factors such as the season, altitude, geographic location, and BC concentration need to be considered.
 We would like to thank the reviewers for their helpful comments, the WRF/Chem development team, and the Atmospheric Chemistry Division at NESL for their assistance with the MOZART-4 reanalysis fields. This work was supported by the ARM program of the US Department of Energy grant DE-PS02-09ER09-15.