JOURNAL OF GEOPHYSICAL RESEARCH, VOL. ???, XXXX, DOI:10.1029/, 1 2 Do Anthropogenic Aerosols Enhance or Suppress the Surface Cloud Forcing in the Arctic? 1 1 K. Alterskjær, J. E. Kristjánsson, C. Hoose 1, 2 K. Alterskjær, Department of Geosciences, Meteorology and Oceanography Section, University of Oslo, P.O.Box 1022, 0315 Oslo, NORWAY. (karialt@geo.uio.no) C. Hoose, Department of Geosciences, Meteorology and Oceanography Section, University of Oslo, P.O.Box 1022, 0315 Oslo, NORWAY. J. E. Kristjánsson, Department of Geosciences, Meteorology and Oceanography Section, University of Oslo, P.O.Box 1022, 0315 Oslo, NORWAY. 1 Department of Geosciences, Meteorology and Oceanography Section, University of Oslo, Oslo, NORWAY. 2 Now at Karlsruhe Institute of Technology, Institute for Meteorology and Climate Research, Karlsruhe, GERMANY. D R A F T July 20, 2010, 6:20am D R A F T X-2 3 4 ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC Abstract. Earlier studies suggest that aerosol-cloud interactions may have contributed 5 to the increase in surface air temperature recently observed in the Arctic. 6 While those studies focused on longwave effects of strong pollution events 7 around Barrow, Alaska, we use a global climate model (CAM-Oslo) to study 8 the annual and seasonal net radiative effect of aerosol-cloud interactions over 9 the entire Arctic region. The model is validated against and adjusted to match 10 observations from the SHEBA campaign along with measuring stations within 11 the Arctic region. Several sensitivity experiments were conducted which in- 12 cluded changes in both cloud properties and aerosol concentrations. Results 13 show that the longwave indirect effect at the surface lies between 0.10 W/m2 14 and 0.85 W/m2 averaged annually north of 71◦ N, while the shortwave indi- 15 rect effect lies between −1.29 W/m2 and −0.52 W/m2 . Due to longwave dom- 16 inance in winter, 6 out of 11 simulations give a positive change in net cloud 17 forcing between October and May (−0.16 W/m2 to 0.29 W/m2 ), while the 18 change in forcing averaged over the summer months is negative for all model 19 simulations (from −2.63 W/m2 to −0.23 W/m2 ). The annually averaged change 20 in net cloud forcing at the surface is negative in 10 out of 11 simulations, ly- 21 ing between −0.98 W/m2 and 0.12 W/m2 . In conclusion our results point 22 to a small decrease in the surface radiative flux due to the aerosol indirect 23 effect in the Arctic, but these estimates are subject to uncertainties in the 24 frequency of thin clouds and biases in the estimated cloud cover. D R A F T July 20, 2010, 6:20am D R A F T ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC X-3 1. Introduction 25 The Arctic region is particularly sensitive to climate change due to the positive feedback 26 between surface temperature and surface albedo [Wang and Key, 2005] and the increase 27 in air temperature in the bottom layers of the atmosphere over the last decades is almost 28 twice as large here as in the rest of the world [e. g. Graversen et al., 2008]. Due to the 29 rapid changes found in this region, there has been an increasing scientific interest in the 30 Arctic in general. This was made evident by the implementation of the International 31 Polar Year in 2007-2008. 32 Several factors contribute to climate change in the Arctic, among these are the increased 33 surface radiative flux resulting from increasing anthropogenic greenhouse gas concentra- 34 tions and reduced surface albedo due to soot deposition on snow. Another possible cause 35 discussed in two empirical studies published in Nature in 2006, is the change in Arctic 36 clouds due to human activities (Garrett and Zhao [2006] (GZ06) and Lubin and Vogel- 37 mann [2006] (LV06)). Of particular interest is the influence of anthropogenic emissions 38 of pollution on the thin, non-opaque clouds common in the Arctic region. 39 Clouds in the Arctic differ from clouds elsewhere in that they have a net warming effect 40 at the surface - there is positive net cloud forcing [Intrieri et al., 2002a]. This happens 41 because the longwave (LW) radiation dominates the radiation regime, due to large solar 42 zenith angles throughout the year, combined with a high surface albedo. Consequently, 43 the LW radiation plays a much more important role in this region than at lower latitudes, 44 and the greenhouse effect of clouds in the LW leads to a net surface warming by clouds in D R A F T July 20, 2010, 6:20am D R A F T X-4 ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC 45 the Arctic. The shortwave (SW) radiation, however, dominates during mid-summer and 46 clouds have a net cooling effect at the surface [Intrieri et al., 2002a]. 47 GZ06 and LV06 focused on what is known as the first aerosol indirect effect or the 48 Twomey effect [Twomey, 1977]. This effect is described as an increase in the cloud optical 49 depth through pollution aerosols leading to more numerous, smaller sized droplets, while 50 the water content of the cloud is assumed to be constant. As is the case for the cloud 51 optical depth, the cloud LW emissivity increases with such a change in cloud properties. 52 Based on this GZ06 and LV06 suggested that anthropogenic emissions of pollution might 53 increase the LW emissivity of the thin, non-opaque clouds common in the Arctic causing 54 their warming effect to increase. Results from their studies suggest that the LW cloud 55 forcing at the surface increases by 3.3 to 8.2 W/m2 in the presence of large anthropogenic 56 emissions of sulfate precursors, due to increases in the LW emissivity of low level clouds. 57 GZ06 and LV06 also suggested a possible contribution from the second indirect effect, 58 through increased liquid water content, and even changes in cloud amount. 59 In this study we will examine how the surface cloud forcing in the Arctic region has 60 changed due to interactions between clouds and anthropogenic aerosols. We will study 61 this with focus on the time of year, and we will compare our results to earlier findings 62 whenever possible. In the following section we describe the model tools and methods used 63 in our study. Section 3 presents basic features of cloud cover, sulfate concentration, cloud 64 water path and cloud radiative forcing as simulated by the model, comparing our results 65 to observations. The influence of anthropogenic emissions on the sulfate concentration, 66 cloud properties and cloud radiative forcing is investigated in section 4. We discuss the 67 results in section 5 and summarize our findings with conclusions in section 6. D R A F T July 20, 2010, 6:20am D R A F T ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC X-5 2. Model Tools and Methods 68 GZ06 and LV06 both based their findings on observational data that have been gathered 69 under specific atmospheric conditions from the area around Barrow, on the north slope of 70 Alaska. In this study we wish to examine the effects of cloud-aerosol interactions in the 71 Arctic region as a whole. In order to do so, our best option as of today is to use numerical 72 modeling. The use of a three dimensional (3D) climate model allows us to study both 73 the spatial variations in cloud-aerosol interactions and the effect of these interactions over 74 time and for different seasons. Long term averages will show the overall importance of 75 the indirect effects. The use of a one dimensional model allows us to study how specific 76 changes in cloud properties affect the surface cloud forcing in the Arctic. 2.1. Model Description: CAM-Oslo 77 The atmospheric general circulation model used here is the CAM-Oslo, extended from 78 NCAR-CAM3 (National Center for Atmospheric Research - Community Atmosphere 79 Model version 3) [Collins et al., 2006a]. The CAM-Oslo includes modules for aerosol life- 80 cycling and interactions with radiation described by Seland et al. [2008]. The model also 81 includes a prognostic calculation of cloud droplet number concentration (CDNC) in which 82 droplet activation is based on chemical composition, size distribution and parametrized 83 subgrid-scale vertical velocity [Storelvmo et al., 2006; Hoose et al., 2009]. It is run as a 84 stand-alone atmospheric model with prescribed climatological sea surface temperatures. 85 The horizontal resolution of the model is approximately 2.8◦ x 2.8◦ (T42 spectral trunca- 86 tion), and there are 26 layers in the vertical. The vertical coordinate is a hybrid coordinate 87 that follows the terrain in the lower troposphere and gradually becomes a pressure coor- D R A F T July 20, 2010, 6:20am D R A F T X-6 ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC 88 dinate when entering the lower stratosphere. A 20 minute time step is used both for the 89 dynamics and the physics. 2.2. CAM-Oslo Model Modifications 90 Several modifications were made to better suit the model to the focus of our study. 91 Cloud-aerosol interactions lead to changes in cloud microphysics and therefore in cloud 92 radiative properties. Before model modification, calculations in the LW part of the spec- 93 trum did not depend on the size of cloud droplets. This was because this dependence is 94 insignificant when SW radiation is dominating the radiation regime and because water 95 clouds at low- and mid-latitudes are often optically thick in the LW. The emissivity of 96 these clouds is therefore not influenced by cloud droplet size. In the Arctic, the LW ra- 97 diation is much more important, and optically thin clouds persist for large parts of the 98 year. We can therefore no longer neglect the LW emissivity dependence on cloud droplet 99 size. We derived an expression for the dependency of the LW absorption coefficient, and 100 therefore the LW emissivity, on cloud droplet size and implemented this expression in the 101 model, enhancing its capability to accurately simulate Arctic conditions. 102 The LW cloud emissivity is given by Collins et al. [2004] (section 4.9.5): ϵ = 1 − e−1.66∗kabs ∗CW P (1) 103 where CWP is the cloud water path i.e. the integrated total cloud water content in a 104 column above a certain surface area, in units of g/m2 , while 1.66 is the diffusivity factor 105 and kabs is the mass absorption coefficient for condensed water. For mixed phased clouds 106 kabs will be a weighted mean of the absorption coefficients for liquid and solid particles, 107 respectively. In this study we consider the aerosol influence on liquid cloud particles only. D R A F T July 20, 2010, 6:20am D R A F T X-7 ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC 108 It is therefore the liquid water mass absorption coefficient (kabs,liquid ) which we express 109 in terms of effective radius (re = 110 emissivity to depend on droplet size. The absorption coefficient for liquid particles is given 111 by: kabs,liquid = ∫ ∫ πr3 n(r)dr/ πr2 n(r)dr), in order for the LW cloud βa LW C (2) 112 where βa is the volume absorption coefficient for liquid droplets and LWC is the liquid 113 water content of the cloud in units of mass per unit volume (LW C = 43 πρL r3 n(r)dr, 114 where ρL is the bulk density of liquid water). The volume absorption coefficient is given 115 by: ∫ ∫ βa = π ∞ 0 n(r)r2 Qa (r)dr (3) 116 where n(r)dr is the cloud droplet size distribution as a function of radius, r, while Qa is 117 the absorption efficiency. Based on Mie calculations following equations (3) and (6) in 118 Chýlek et al. [1992], the model Qa is approximated as follows: For radii greater than a 119 certain rmax , Qa is constant and equal to 1.0, while for r smaller than rmax , Qa increases 120 linearly with r (Qa =a1 r). The parameter rmax varies with wavelength because Qa is 121 wavelength dependent (Paltridge and Platt [1976, Figure 8.4] and Garrett et al. [2002]). 122 However, in some general circulation models, including the CAM-Oslo model, the cloud 123 emissivity is constant over the entire LW spectrum and one representative rmax must be 124 used. According to Paltridge and Platt [1976, p. 200] the value of the mass absorption 125 coefficient for 11 µm is near the average value for the entire window region from 8 µm to 126 14 µm, and from Wien’s Displacement Law [e. g. Liou, 2002] we know that the wavelength 127 for the intensity peak of the earth’s radiation field lies within this window region. Around D R A F T July 20, 2010, 6:20am D R A F T X-8 ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC 128 a wavelength of 11 µm rmax can be approximated by 10 µm and a1 by 0.1 [Garrett et al., 129 2002]. These are the values used over the whole LW spectrum in our calculations. 130 131 The effective radius (re ) is constant in the population of droplets. Therefore Qa (re ) does not vary with r and can be taken outside the integral. βa ≈ πQa { 132 ∫ ∞ n(r)r2 dr, 0 Qa = 0.1 re [µm−1 ] f or re < 10µm Qa = 1.0 f or re ≥ 10µm (4) Solving this integral, using the definition of effective radius and LWC, leaves: 3 Qa , 4 ρ r L e { Qa = 0.1 re [µm−1 ] f or re < 10µm Qa = 1.0 f or re ≥ 10µm kabs,liquid ≈ (5) 133 The mass absorption coefficient and hence the cloud LW emissivity thus depends on the 134 droplet effective radius. Expression 5 is used in all model simulations of this study unless 135 otherwise stated. Approximating Qa in this manner is a simplification. In reality there 136 is a continuum between the two regimes both because clouds have a droplet spectrum 137 and because thermal radiation has a wavelength spectrum [Garrett et al., 2002]. The 138 sensitivity of the LW emissivity to changes in cloud droplet size may therefore be affected 139 by our assumptions. This will be investigated closer in section 5.3. 2.3. CAM-Oslo Model Setup 140 Two different emission fields were used in order to study the effect of increased amounts 141 of pollution on cloud forcing: One field based on present day (year 2000) emissions, hereby 142 referred to as PD, and another field based on pre-industrial emissions, hereby referred to 143 as PI. They are both based on the AeroCom emissions [Dentener et al., 2006]. However, 144 due to uncertainties about pre-industrial forest fires, we modified the AeroCom PI field D R A F T July 20, 2010, 6:20am D R A F T ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC X-9 145 such that the emissions prior to the industrial revolution were nowhere higher than the 146 present day emissions. The total global annual emissions of each species in each of the 147 emission fields are listed in Table 1. 148 The model was run for five years and the results shown are averaged over these years. 149 Such a long integration time diminishes variations due to specific weather events. The 150 summer season here includes the months of June, July, August and September, while 151 the winter season is an average over the remaining months. The model was run off-line, 152 meaning that the meteorological evolution is the same in all model runs. This allows us 153 to study how the clouds and the radiative balance are changed between emission fields 154 without feedbacks due to the aerosol forcing. The simulated change in cloud forcing 155 with pollution is then only a result of aerosols interacting with the clouds and we avoid 156 noise from synoptic variability in our results. This also implies that all feedbacks due 157 to aerosol induced cloud changes such as the semi-indirect effect and changes in cloud 158 cover are precluded from this investigation. However, as explained in Kristjánsson [2002], 159 the contribution to the indirect effect from instantaneous suppression of precipitation 160 release is accounted for. This is not treated as a feedback in our model because a control 161 simulation propagates the model. 2.4. Model Description: One Dimensional (1D) Model 162 A one dimensional column model was used to study the radiative effects of placing 163 specific clouds in preferred environments. The model input includes cloud parameters 164 such as cloud droplet effective radius and liquid water path (LWP) i.e. the integrated 165 liquid water content in a column above a certain surface area (g/m2 ). The model uses the 166 radiation scheme from the NCAR CCM3 model [Kiehl et al., 1998] to give instantaneous D R A F T July 20, 2010, 6:20am D R A F T X - 10 ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC 167 values of radiation fluxes. This scheme is very similar to the one used in NCAR-CAM3 and 168 therefore in CAM-Oslo [Collins et al., 2006a]. In addition to cloud parameters the input 169 includes location given by latitude and time of year, and gas and temperature profiles 170 suited for the chosen environment. The output is averaged over the chosen latitude, equal 171 to a 24 hour mean. In this study the model ran with 26 layers in the vertical corresponding 172 to the vertical layers of CAM-Oslo. 2.5. Calculation of Cloud Radiative Forcing 173 The change in cloud forcing due to aerosol-cloud interactions can be taken as a measure 174 of the aerosol indirect effect. Cloud forcing (CF) is defined as “the radiative impact that 175 clouds have on the atmosphere, surface, or top-of-the-atmosphere (TOA) relative to clear 176 skies” [Shupe and Intrieri , 2004]. In the following we will be mainly concerned with the 177 CF at the surface (CFS), which is given by: LWCFS = N etLWallsky − N etLWclear (6) SWCFS = N etSWallsky − N etSWclear (7) 178 where LWCFS and SWCFS denote the longwave and shortwave cloud forcings at the 179 surface, respectively, NetLW and NetSW denote the net downward flux at the surface, 180 i.e., downward flux minus upward flux, allsky refers to the true atmospheric state includ- 181 ing clouds, while clear refers to a hypothesized atmosphere with all clouds removed but 182 otherwise identical conditions. 183 184 Adding the longwave and shortwave contributions, a net cloud forcing at the surface (Net CFS) is defined as: Net CFS = LWCFS + SWCFS D R A F T (8) July 20, 2010, 6:20am D R A F T ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC X - 11 185 The LWCFS depends mainly on the ability of the clouds to absorb and emit LW radi- 186 ation, and therefore on the LW emissivity (Equation 1) of the cloud. The clouds absorb 187 a fraction of the radiation emitted by the surface and then re-emit energy toward the 188 ground, and because the clouds become optically thick at low LWPs (∼50 g/m2 ) the 189 LWCFS depends mainly on the lowest cloud base. The flux density emitted toward the 190 surface depends on the cloud base temperature, T, and the cloud properties such as cloud 191 particle size and LWP (see the Stefan-Boltzmann law and Equation 1). The higher the 192 emissivity, the larger is the LWCFS. 193 The SWCFS depends mainly on the ability of the clouds to reflect SW radiation, and 194 therefore on the cloud albedo, A. The albedo of a cloud can be approximated by its optical 195 depth, τ , and the asymmetry factor, g, alone [Meador and Weaver , 1980]: (1 − g)τ 1 + (1 − g)τ A = (9) 196 The optical depth depends on cloud droplet effective radius and LWP through [e. g. Liou, 197 2002]: τ = 198 (10) This gives: A = 199 3 LW P 2 ρL re 1 (11) 2 ρL re 1+ 3 (1 − g)LW P The higher the cloud albedo, the more negative is the SWCFS. When the influence of anthropogenic aerosols on clouds and climate is considered, it is the changes in LWCFS and SWCFS from a clean to a polluted case that are of interest. These changes are given as: ∆LWCFS = (N etLWallsky,polluted − N etLWallsky,clean ) D R A F T July 20, 2010, 6:20am D R A F T X - 12 ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC −(N etLWclear,polluted − N etLWclear,clean ) (12) ∆SWCFS = (N etSWallsky,polluted − N etSWallsky,clean ) −(N etSWclear,polluted − N etSWclear,clean ) 200 (13) and the sum of the two is defined as: ∆Net CFS = ∆LWCFS + ∆SWCFS (14) 201 As the focus of this investigation is on the radiative effect of aerosol-cloud interactions, 202 the change in net flux due to aerosols in clear conditions is not contained in our simula- 203 tions. However, it is clear that the anthropogenic aerosols may influence the Arctic also 204 via the direct effect, i.e., by reflection and absorption of solar radiation. In this paper the 205 term “change in cloud forcing” refers to the aerosol indirect effect. 3. Basic Features of Arctic Clouds and Aerosols 206 In this section we will check whether the model output is consistent with observations. 207 The main focus will be on cloud cover, sulfate, liquid water path and cloud forcing, 208 and observations made during the SHEBA (Surface Heat Budget of the Arctic Ocean) 209 campaign will be an important part of this validation. The SHEBA campaign took place 210 in the Beaufort and Chukchi Seas (from 75.3◦ N, 142.7◦ W to 80.5◦ N, 166◦ W) from October 211 1997 to October 1998 [Intrieri et al., 2002b; Maslanik et al., 2001] and its main observables 212 include the sea ice mass balance and the surface energy balance. The advantage of using 213 this data set is that it comprises one continuous year of data, something not matched by 214 any other campaign this far into the Arctic region. D R A F T July 20, 2010, 6:20am D R A F T ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC X - 13 3.1. Cloud Cover 215 Observational data gathered by Warren et al. [1988] show that in general the Arc- 216 tic cloud cover has a minimum in wintertime with values just below 50 percent and a 217 maximum in late summer/early fall that peaks around 85 percent (see Figure 1a, black 218 dash-dot line). The seasonal variation in Arctic cloud cover is well reproduced by the 219 CAM-Oslo model. Nevertheless, the model seems to underestimate the total cloud cover 220 from April until December. The largest errors occur during the last half of the year when 221 the cloud cover is underestimated by around 10 percent. The data presented by Warren 222 et al. [1988] are, however, based on ground-based manual observations and are therefore 223 likely to be somewhat inaccurate, both due to the dark season and due to the sparsity of 224 measurements in the remote Arctic region. 225 We also compare simulated average cloud cover to observations made during the SHEBA 226 campaign (see Figure 1a). The SHEBA data were obtained from manual observations, 227 from ground-based LIDAR/RADAR measurements as well as from satellite, and show a 228 large spread between the different observational methods. This highlights the difficulty 229 in determining cloud fraction. From Figure 1a it seems that the CAM-Oslo cloud frac- 230 tion is lower than most observations - sometimes by up to 30%. This may lead to an 231 underestimation of cloud-aerosol interactions and hence of the cloud radiative forcing. 232 Contrary to this, when comparing the simulated fraction of low level Arctic clouds 233 (below 700 hPa) to satellite observations presented by Kay and Gettelman [2009, Figure 234 4] we find that the fraction of these clouds is overestimated during summer (24% between 235 65◦ N and 82◦ N), while it is underestimated in early fall (16%). A possible overestimation 236 in summer may lead to a negative bias in the aerosol indirect effect due to SW dominance, D R A F T July 20, 2010, 6:20am D R A F T X - 14 ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC 237 while the net effect of an underestimation of cloud cover in September and October is less 238 obvious. 239 The vertical placement of the clouds is also of importance, as the LW cloud forcing is 240 temperature dependent. Observational data from the SHEBA campaign show that the 241 Arctic clouds often lie close to the surface (cloud bases below 1 km) [Intrieri et al., 2002b, 242 Figure 7]. This tendency is well reproduced by the CAM-Oslo model (see Figure 2). 3.2. Particulate Sulfate (SO4 ) 243 In this work the terms sulfate and SO4 both refer to particulate sulfate. 244 The simulated present day sulfate burden over the Arctic region is plotted in Figure 245 3a. The data are annually averaged and column integrated and show that the burden is 246 largest over northern Eurasia, gradually decreasing as we move further into the Arctic. 247 This is consistent with the current understanding of the transport of air pollution into 248 the region [Stohl , 2006], namely that pollution in the Arctic mainly originates in Eurasia. 249 The annually averaged burdens of organic and black carbon have the same spatial pattern 250 as sulfate. 251 The surface concentrations of SO4 are verified against measurements taken at sev- 252 eral monitoring stations (Zeppelin, Spitsbergen, Norway (78.9◦ N, 11.9◦ E); Alert, Canada 253 (82.5◦ N, 62.3◦ W); Janiskoski, Russia (69◦ N, 29◦ E); and Barrow, Alaska (71.3◦ N, 254 156.6◦ W)) [AMAP , 2006, chapter 4]. We find that the seasonal variation in SO4 con- 255 centration at all stations is fairly well reproduced by the CAM-Oslo model (Figure 4) - 256 with large concentrations in winter and early spring and minima occurring during sum- 257 mer. These summer minima are caused by a shift in the east-west pressure gradients 258 across Eurasia so that less pollution is transported into the Arctic, combined with an D R A F T July 20, 2010, 6:20am D R A F T X - 15 ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC 259 increase in precipitation during summer, scavenging the SO4 from the lowest layers of 260 the atmosphere [Barrie, 1986]. Additionally, the Arctic air mass is less stable during 261 the summer than during the winter. This is associated with increased turbulent transfer 262 [Quinn et al., 2008] and thus removal of aerosols through dry and wet deposition. 263 The CAM-Oslo model’s ability to reproduce the mass concentration of SO4 varies over 264 the Arctic region. Figure 4 shows that the surface concentrations at Zeppelin are well 265 reproduced by the model, while simulations for Janiskoski show an overestimation of SO4 266 compared to observations, although the simulated mean SO4 concentrations are seldom 267 larger than the maximum monthly mean observed during the five year period from 1996 268 to 2000. Unlike simulations for stations on the Eurasian side of the Arctic, simulations for 269 the North American sites generally underestimate SO4 compared to observations. There 270 can be several reasons for this underestimation. First of all these sites are further away 271 from the main sources of Arctic SO4 than the Eurasian sites [Stohl , 2006]. This may 272 point to the transport pathways themselves being inaccurate or to important sources 273 of SO4 precursors being ignored. Another reason for the small concentrations in North 274 America may be inaccurate SO4 removal processes. In section 3.3 we will show that the 275 model cloud liquid water path is too high, which may lead to an overestimated in-cloud 276 scavenging of SO4 . For further details on the treatment of scavenging in CAM-Oslo see 277 Seland et al. [2008]. In section 5.4 the sensitivity of our results to the SO4 concentration 278 will be tested both by reducing the in-cloud scavenging and by increasing the emissions 279 of SO4 precursors. 280 We compared the accuracy of our results at Zeppelin and Janiskoski to the accuracy of 281 results from models participating in the AeroCom project (Aerosol Comparisons between D R A F T July 20, 2010, 6:20am D R A F T X - 16 ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC 282 Observations and Models; Textor et al. [2006]). The red lines in Figure 4 show the model 283 median of ten AeroCom A models simulating surface concentrations of SO4 at Zeppelin 284 and Janiskoski for the year 2000 (medians for Alert and Barrow were not available) [http : 285 //nansen.ipsl.jussieu.f r/AEROCOM/]. Note that the median seasonal variation is 286 opposite to what is observed. It is clear that the CAM-Oslo results are in general a better 287 fit to the observations than the AeroCom model median. 288 Comparing the CAM-Oslo simulated vertical profiles of SO4 to observations is chal- 289 lenging. First of all, measurements of SO4 in the Arctic are limited both in number and 290 in geographical distribution. Secondly, there is large variability in the observations, also 291 when taken with short time intervals in the same regions [e. g. Dreiling and Friederich, 292 1997; Scheuer et al., 2003]. These measurements are generally instantaneous aircraft 293 measurements and are limited both in time and space. This makes it difficult to compare 294 observations of vertical profiles to our monthly averaged profiles. 295 Figure 5a shows the concentration of sulfate with height in terms of µg S per unit vol- 296 ume of air. The values which are annually and zonally averaged over the Arctic region, 297 show that north of 70 to 75◦ N the largest concentrations are found at around 800 to 900 298 hPa. Although we have no averaged observed vertical profiles to verify the concentrations 299 of this cross section, 800 hPa is the height found by Dreiling and Friederich [1997] to 300 have the largest concentration of particles of all sizes. A comparison with Scheuer et al. 301 [2003] shows that the simulated near surface concentrations of SO4 , which are the most 302 important for our study, are of the same order of magnitude as measurements taken dur- 303 ing the TOPSE (Tropospheric Ozone Production about the Spring Equinox Experiment) 304 campaign. The measurements show mean SO4 concentrations in the bottom two kilome- D R A F T July 20, 2010, 6:20am D R A F T ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC X - 17 305 ters of the atmosphere of between 50 and 230 pptv during springtime, while the simulated 306 values along 70◦ W range between 60 and 170 pptv for the same altitudes. 3.3. Liquid and Ice Water Path 307 Measurements taken during the SHEBA campaign were used by Zhang et al. [2002] to 308 retrieve monthly averaged liquid water paths for the region covered by the campaign. A 309 maximum of around 100 g/m2 was reached in August, when also the cloud fraction reaches 310 its peak value. This is in accordance with the typical range of Arctic LWP only seldom 311 exceeding 150 g/m2 [Löhnert et al., 2003]. A comparison between the monthly averaged 312 LWP (including its uncertainty) retrieved from measurements and the simulated LWP for 313 the SHEBA region shows that the CAM-Oslo overestimates the LWP by a factor of 3 to 314 5, depending on season (see Figure 1b). 315 Model intercomparison studies by Morrison et al. [2009] and Karlsson and Svensson 316 (manuscript in prep.) have found excessive LWPs over the Arctic region in NCAR-CCSM3 317 (Community Climate System Model version 3) [Collins et al., 2006b] and SCAM3, a single- 318 column version of the NCAR-CAM3. In the case studied by Morrison et al. [2009] the 319 SCAM3 simulated LWP averages to 298 g/m2 while the observed LWP ranges from 55 to 320 121 g/m2 , depending on the retrieval method. The model intercomparison by Karlsson 321 and Svensson (manuscript in prep.) shows that the NCAR-CCSM3 simulated LWP over 322 the Arctic Ocean is from 2 to 3.7 times the ensemble model mean LWP. These results 323 suggest that the overestimation of LWP in CAM-Oslo may be linked to problems in the 324 NCAR-CCSM3, as both the CAM-Oslo and the SCAM3 are developed from this model. 325 The excessive simulated liquid water amounts may be caused by several factors. An 326 underestimated auto-conversion rate will lead to little loss of cloud water through precip- D R A F T July 20, 2010, 6:20am D R A F T X - 18 ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC 327 itation. Another possibility is too little conversion from liquid water to ice particles. The 328 overestimated LWP may also be caused by an overestimated transport of moisture into 329 the region or by stably stratified conditions allowing model clouds to become thicker than 330 what occurs in nature. We return to this problem in section 3.5. 331 Ice water path (IWP) retrievals have very high uncertainties. Nevertheless, it should 332 be mentioned that Shupe et al. [2006] found observed IWPs on the order of 42 g/m2 and 333 Morrison et al. [2003] reported IWPs on the order of 34.6 g/m2 for the SHEBA region. 334 Our model has an average IWP of 24.0 g/m2 . Also, Karlsson and Svensson (manuscript 335 in prep.) found that the NCAR-CCSM3 has among the lowest ice water paths of the 336 models in their study. Combined with the positive bias in LWP this may point to a bias 337 in the conversion between solid and liquid particles or a bias in the distinction between 338 solid and liquid particles in our model. Due to the limited amount of measurements in 339 the Arctic combined with high uncertainties we cannot conclude on the exact reason for 340 the too low ice water paths. 3.4. Cloud Forcing 341 The simulated cloud forcing (CF) is compared to observations of Intrieri et al. [2002a], 342 ignoring the turbulent flux that is part of their study. The simulated cloud forcing in the 343 SHEBA region (74◦ -81◦ N and 144◦ -169◦ W [Zhang et al., 2002]) differs significantly from 344 what was measured at SHEBA, especially during the summer (Table 2 under case names 345 ’SHEBA’ and ’CAM-Oslo std. LWP’). This is mainly caused by a large difference in the 346 SWCFS between the simulations and observations. However, the simulated LWCFS is 347 also larger than the observed forcing. We know from Figure 1a that the cloud cover is 348 reproduced fairly well by the model, or slightly underestimated. Comparing the model D R A F T July 20, 2010, 6:20am D R A F T ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC X - 19 349 surface albedo to measurements taken over the SHEBA region [Curry et al., 2000; Intrieri 350 et al., 2002a] we find that it is within reasonable range. Rough monthly estimates of 351 aircraft measured surface albedos from Curry et al. [2000] are 0.76, 0.67 and 0.50 for May, 352 June and July respectively. CAM-Oslo results from the same region and time period are 353 0.78, 0.66 and 0.49, in excellent agreement with the observations. Without time averaged 354 temperature profiles for this area we cannot exclude that the LWCFS is affected by a bias 355 in cloud base temperature. However, based on the simulated forcing being larger than 356 observations in both wavelength ranges, it is likely that the discrepancy between modeled 357 and observed cloud forcing is caused by the optical depth and the emissivity of the clouds 358 being too large. 3.5. Model Modifications to Improve the Simulated LWP and CFS 359 The simulated effective radius around Barrow, Alaska, is 15 µm when averaged below 360 700 hPa, with larger values in summer than in winter (16 µm vs. 14 µm). While comparing 361 our simulated re to observations is challenging because observed values vary significantly, 362 the seasonal variation simulated around Barrow is consistent with the findings of Dong 363 and Mace [2003] in the same region. Additionally, the simulated effective radius in the 364 SHEBA region is 10 µm when averaged annually over the whole vertical column, which 365 is consistent with the findings of Curry et al. [2000]. Consequently it is unlikely that the 366 overestimation of cloud optical depth and emissivity is caused by an underestimation of re 367 (see Equation 10 for τ and Equations 1 and 5 for ϵ). Instead, it is most likely associated 368 with an excessive LWP in the model (Figure 1b). 369 Several approaches were used in order to reduce the liquid water amount in the model. 370 This is of particular importance for our study as the sensitivity of cloud emissivity and D R A F T July 20, 2010, 6:20am D R A F T X - 20 ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC 371 albedo to changes in LWP depends on the amount of water that the clouds initially hold. 372 The goal was to keep the LWP within the range of the observed values and simultaneously 373 find what simulation had cloud forcing closest to the SHEBA measurements of this quan- 374 tity. Modifying the LWP affects the global net radiation, but as we focus on the Arctic 375 region and on the change in cloud forcing between PI and PD, this is not a concern. 376 Different modifications of the auto-conversion parametrization were tried allowing more 377 water to be lost through precipitation. The auto-conversion threshold radius, r3lc , was 378 successively reduced from 15 µm to 10 µm and 7.5 µm. This radius decides the size that 379 cloud particles must reach before the onset of precipitation, as described in Rasch and 380 Kristjánsson [1998, Equation 21]. In addition, we changed the lower limit for which auto- 381 conversion is fully efficient, here named autlim, as described in Kristjánsson [2002, section 382 2.4], from 5.0 mm day−1 [Kristjánsson, 2002] to 0.5 mm day−1 [Rasch and Kristjánsson, 383 1998] and then to 0.0 mm day−1 . This parameter accounts for the decrease in collection 384 efficiency in a cloud droplet distribution that has been modified by precipitation. Results 385 from the simulations with modified auto-conversion can be seen as the light blue and the 386 red lines in Figure 1b. The water amounts are now much closer to the retrieved values, but 387 are still on the high side. From Table 2 it is clear that these simulations still overestimate 388 the SW and the LW components of the surface cloud forcing (case names ’1st auto-conv.’ 389 (r3lc = 10 µm and autlim = 0.5 mm day −1 ) and ’2nd auto-conv.’ (r3lc = 7.5 µm and 390 autlim = 0.0 mm day −1 )). 391 We also reduced the cloud liquid water path by modifying both the auto-conversion and 392 the cloud particle ice fraction (’fice’). The auto-conversion threshold radius was changed 393 from 15 to 7.5 µm and a lower limit for fully efficient auto-conversion of 0.0 mm day−1 D R A F T July 20, 2010, 6:20am D R A F T ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC X - 21 394 was used instead of 5.0 mm day−1 . In the standard version of CAM-Oslo the fraction of 395 cloud particles that are solid is temperature dependent and increases linearly from 0 to 1 396 as the temperature decreases form 263K to 233K. In reality the ice fraction is expected 397 to be influenced by aerosol properties. We increased ’fice’ by letting it go from 0 to 1 398 between 273K and 243K. Results from this simulation are seen as the purple line in Figure 399 1b. The LWP is now consistent with observations, but the SW and the LW components 400 of the surface cloud forcing are still overestimated (Table 2 under case name ’Auto-conv. 401 + fice’). 402 For simplicity we then conducted several idealized experiments where we forced a re- 403 duction in the LWP, and found that reducing it by a factor of five through reducing the 404 CDNC gave the results closest to the observed values. This can be seen from the green 405 line in Figure 1b and from results under case name ’LWP/5 ∆CDNC’ in Table 2. Note 406 that the model radiation scheme does not depend explicitly on the CDNC, but rather on 407 LWP and re . Physically, however, reducing the LWP while keeping the re constant is the 408 same as reducing the cloud droplet number concentration. A reduction in CDNC is not in 409 itself an improvement, as this concentration is already low. The averaged observed CDNC 410 in single-layer stratus clouds obtained during the Mixed-Phase Arctic Cloud Experiment 411 (M-PACE) during fall 2004 was 43.6 ± 30.5 cm−3 [McFarquhar et al., 2007]. The simu- 412 lated CDNC is around 17 cm−3 in the standard model version for all clouds during this 413 season in the same area (around Barrow and Oliktok, Alaska). This mean gives weight 414 also to glaciated mixed-phased clouds where nearly all the water is in solid form and the 415 CDNC consequently is very low. It is therefore likely that the simulated CDNC of the 416 persistent low level stratus clouds is somewhat higher than 17 cm−3 . D R A F T July 20, 2010, 6:20am D R A F T X - 22 ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC 417 Additionally, we tried reducing the cloud droplet effective radius, keeping the CDNC 418 constant, in order to obtain LWP values consistent with observations. However, this led to 419 surface cloud forcing much stronger than what was observed during the SHEBA campaign 420 (see Table 2 under case name ’LWP/5 ∆re ’). 421 Note that the only adjustment to the LWP that affects the model thermodynamics is 422 the one in which the cloud particle ice fraction is modified. All the other methods used 423 to adjust the LWP are applied after the liquid water is formed and therefore have no 424 effect on the thermodynamic state of the atmosphere. By similar arguments, the only 425 adjustments to the LWP that affect the aerosol concentration are the modifications of the 426 auto-conversion and the cloud particle ice fraction. We are not directly comparing these 427 runs to the standard model, but are consistently calculating the differences between PD 428 and PI for two “auto-conversion” and two “fice” simulations, respectively. 429 The results shown in the next section are from simulations in which the LWP was 430 reduced by a factor of five through reducing the CDNC, as these simulations give the best 431 agreement between observed and simulated LWP and surface cloud forcing. In section 5 432 we will discuss how our results are affected by this choice. 4. Anthropogenic Influence 433 In the following subsections we will investigate the simulated changes in cloud properties 434 due to anthropogenic aerosol emissions and how these changes affect the surface radiative 435 balance in the Arctic region. D R A F T July 20, 2010, 6:20am D R A F T ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC X - 23 4.1. Changes in Sulfate (SO4 ) Concentrations 436 The only change between simulations of the pre-industrial and the present day climate 437 is the anthropogenic emissions of chemical compounds in to the atmosphere. The increase 438 in SO4 and other anthropogenic particles in this period leads to an increase in the concen- 439 tration of cloud condensation nuclei (CCN) and is expected to influence cloud properties. 440 Figure 3b shows the increase in the column burden of sulfate from pre-industrial times 441 until today. The increase is largest over northern Eurasia, where the burden itself is 442 also largest (see Figure 3a). The large increase in this area is not surprising as northern 443 Eurasia is a significant source region of anthropogenic sulfate precursors (see section 3.2). 444 Figure 5b shows the vertical distribution of the change in sulfate concentration between 445 PD and PI, averaged annually over the Arctic region. The largest change in concentration 446 occurs around 800 - 900 hPa. However, there are relatively large signals of change both 447 above and below this level. According to Shindell et al. [2008], the upper part of this 448 signal may be influenced by the increase in emissions of SO4 precursors in Asia as well as 449 Europe. In section 3.1 we showed that both the modeled and the observed clouds in the 450 Arctic in general lie close to the surface (∼ 900 - 950 hPa, see e. g. Figure 2). As seen 451 in Figure 5b, changes in aerosol concentrations occur at the same levels and we therefore 452 expect that this change will affect cloud properties. 4.2. Changes in Cloud Droplet Effective Radius (re ) 453 The simulated effective radius averaged annually over the cloud droplet number con- 454 centration decreases from 11.7 µm in pristine conditions (PI) to 9.8 µm in the polluted 455 present day regime north of 71◦ N. By comparison, observations by GZ06 show an average 456 decrease in effective radius from 12.9 µm to 9.9 µm between clean and polluted conditions. D R A F T July 20, 2010, 6:20am D R A F T X - 24 ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC 457 Thus, the first indirect effect is present in our simulations and the change in re is of the 458 same order of magnitude as observed values. 459 Figure 5c displays the vertical cross section of the annually averaged effective radius over 460 the Arctic region. This figure shows that the layers below 500 to 600 hPa have effective 461 radii above 10 µm, and we note that the LW cloud emissivity is sensitive to changes in 462 cloud droplet size (see Equation 5). 463 Figure 5d shows that the changes in effective radius due to anthropogenic emissions 464 are largest between 500 and 800 hPa. The change in effective radius is large if the re is 465 large initially and there is a large relative increase in CDNC. This is what creates the 466 maximum between 500 and 800 hPa. Above this maximum the re is initially small while 467 close to the surface the CDNC is high pre-industrially and the relative increase in CDNC 468 with increasing aerosol levels is therefore small (not shown here). 469 The largest reductions in effective radius of about 2.5 µm occur well above the height 470 of the highest cloud fraction (Figure 2), and have therefore only limited influence on the 471 surface cloud forcing. On the other hand, the decrease in re of 0.6 to 1.0 µm close to the 472 surface will increase both the SW and the LW surface cloud forcing from pre-industrial 473 times to present day. 4.3. Changes in Liquid Water Path (LWP) 474 As displayed in Figure 6a the spatial pattern of annually averaged change in LWP 475 between the PD and the PI scenario has similarities with the pattern of change in the 476 integrated SO4 concentration (Figure 3b). Note that the LWP increases in the more 477 polluted regime, as expected from the reduced droplet size and therefore reduced loss of 478 water due to precipitation release. Hence, the model simulates a distinct second indirect D R A F T July 20, 2010, 6:20am D R A F T ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC X - 25 479 effect. The simulated ice water path (not shown) does not change between scenarios, as 480 aerosols in this study do not affect ice nucleation. 481 The average increase in liquid water path between the two scenarios is about 2.3 g/m2 482 north of 71◦ N in the ’LWP/5 ∆CDNC’ simulations, going from 27.8 to 30.1 g/m2 . This 483 is within the same range as the change found by GZ06 of 2.4 g/m2 , from 31.1 g/m2 in 484 pristine conditions to 33.5 g/m2 in an atmosphere with high aerosol concentrations. We 485 also note that the clouds in the pre-industrial aerosol regime have liquid water paths in 486 the same range as the average clean clouds observed by GZ06. 487 The vertical distribution of changes in in-cloud liquid water mixing ratio (LWMR) 488 averaged annually over the Arctic region is shown in Figure 6b. Contrary to the changes 489 in effective radius, the LWMR increases the most close to the surface. The reason for 490 this is that the liquid water amount is largest near the surface in pre-industrial times 491 (not shown). This affects the change in LWMR in the following manner: The onset of 492 precipitation is determined by the size of the cloud droplets (see section 3.2), but the 493 amount of water lost through this process increases with the in-cloud liquid water mixing 494 ratio, as well as with re [Rasch and Kristjánsson, 1998, eq. 21]. This means that the 495 change in the precipitation amount and hence in the LWMR is large for a given change 496 in re if the LWMR is large. Although a small change in CDNC leads to small changes 497 in the effective radius at surface levels, this change is large enough to affect the model 498 auto-conversion and hence the amount of water lost through precipitation. 499 The average LWP described so far says nothing about the thickness of each individual 500 cloud simulated by CAM-Oslo. There may be episodes of very high or very low LWPs 501 that affect this average greatly. This is of importance because the cloud optical properties D R A F T July 20, 2010, 6:20am D R A F T X - 26 ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC 502 vary non-linearly with LWP and the LW cloud emissivity reaches saturation for all LWPs 503 above about 50 g/m2 . One might therefore question whether clouds are thin enough to be 504 affected in the LW by a change in cloud droplet size or LWP. Figure 7 shows the fraction 505 of time that has vertically integrated LWPs below 50 g/m2 when clouds are present. It 506 reaches a minimum in August, simultaneously with the maximum in cloud cover and 507 average LWP (see Figures 1a and 1b). The fraction is never below 59 % pre-industrially 508 and never below 55 % in present day. We conclude that a large fraction of clouds is non- 509 opaque in the LW and therefore sensitive to changes in effective radius and liquid water 510 path due to anthropogenic aerosols. 4.4. Changes in Longwave Cloud Forcing at the Surface (LWCFS) 511 512 In this subsection we will study the simulated changes in LW cloud forcing at the surface between pre-industrial times and the year 2000 (’LWP/5 ∆CDNC’ simulations). 513 The annually averaged change in surface LW cloud forcing (LWCFS) north of 71◦ N is 514 0.55 W/m2 , corresponding to a 1.6 % increase from pre-industrial times until today. 515 This increase is caused by an increase in the CCN concentration influencing the cloud 516 effective radius and the liquid water path. 517 There is a significant seasonal variation in the change in LWCFS with anthropogenic 518 aerosol emissions (Figures 9a and 9b). Averaged north of 71◦ N the LW cloud forcing 519 changes by 0.99 W/m2 during summer, while it changes by only 0.33 W/m2 during 520 winter time. The large changes in LW cloud forcing during summer may be highly influ- 521 enced by the fraction of low clouds being larger during the summer season (0.75) than 522 during winter (0.40). According to Shupe and Intrieri [2004], clouds that are important 523 to the LW surface radiation balance in the Arctic typically have bases at low altitudes D R A F T July 20, 2010, 6:20am D R A F T ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC X - 27 524 (below 4 km). A high fraction of these clouds allows changes in cloud radiative properties 525 to occur over large areas and therefore cause larger changes in the LW surface radiation 526 budget during summer than during winter. 527 In addition, results show that the cloud liquid water path changes much more in summer 528 than it does in winter (4.2 g/m2 vs. 1.3 g/m2 ). This is because the high LWPs in the 529 summer make the amount of water lost through precipitation very sensitive to changes 530 in cloud droplet size (see section 4.3). The re at low levels changes by approximately the 531 same amount during the summer and the winter (-1.46 µm vs. -1.51 µm below 700 hPa). 532 In the winter re is smaller than in summer, but the pollution events are stronger, while in 533 the summer re at low levels is large and therefore sensitive to the little pollution that is 534 present at low levels during this season (see section 4.2). The large change in LWP causes 535 a large change in the LW emissivity and therefore in the LWCFS in summer. 4.5. Changes in Shortwave Cloud Forcing at the Surface (SWCFS) 536 Changes in re and LWP due to an increase in the anthropogenic aerosol concentrations 537 will affect the SW cloud forcing, as long as solar radiation is present. We find that on 538 average the simulated SWCFS changes by −0.85 W/m2 north of 71◦ N between the PI 539 and the PD scenarios, representing a 6.5% increase in the magnitude of the SW cloud 540 forcing. The seasonal variation in the change in surface cloud forcing is much stronger 541 in the SW than in the LW, due to the sun being absent or at high solar zenith angles 542 through most of the winter season. Because of the low signals during winter, we will now 543 focus only on the summer season. 544 545 The change in surface SW cloud forcing during the summer has an average of −2.17 W/m2 north of 71◦ N. From Figure 9c it is clear that the changes are larger towards D R A F T July 20, 2010, 6:20am D R A F T X - 28 ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC 546 the lower Arctic latitudes. This happens for two reasons. First, the solar zenith angle 547 is smaller here, causing a larger possible impact of the clouds on the radiation budget. 548 Secondly, areas around the North Pole and over Greenland are covered by snow and ice. 549 The cloud albedo increase will be less important here as the clouds are above highly 550 reflective surfaces. 551 The magnitudes of both the relative and the absolute change in SWCFS during summer 552 are larger than the corresponding magnitudes simulated for the LW case. There are several 553 reasons for this. Figure 8 shows that a given change in re or LWP affects the SW cloud 554 albedo more than the LW cloud emissivity under averaged simulated summer conditions 555 (PI)(re = 11.6 µm and LWP = 43.5 g/m2 ). Depending on surrounding conditions such 556 as the surface albedo and the vertical temperature profile, this behaviour will lead to 557 larger changes in SWCFS than in its LW counterpart for a given change in re or LWP. 558 Additionally, since cloud albedo saturates at much higher LWPs than cloud emissivity, a 559 larger fraction of clouds has radiative properties sensitive to changes in re and LWP in 560 the SW than in the LW. 561 Another reason for the large SW signals is the simulated changes in re and LWMR at 562 levels well above the surface (section 4.2 and 4.3). These changes will affect the SWCFS 563 because they affect the cloud albedo (see Equation 11). Conversely, the effect of these 564 changes on LWCFS is expected to be very small as this forcing is mainly influenced by 565 changes that occur in the bottom cloud layers, which are low in the Arctic (see Figure 2). D R A F T July 20, 2010, 6:20am D R A F T ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC X - 29 4.6. Changes in Net Cloud Forcing at the Surface (Net CFS) 566 The changes in cloud forcing in both the LW and the SW have now been examined. 567 Here, we will study the total influence of increased aerosol levels interacting with Arctic 568 clouds. 569 The annually averaged change in Arctic net CFS between the PD and the PI scenario 570 is −0.30 W/m2 (not shown). This confirms that the increased magnitude of SW cloud 571 forcing with pollution is larger than the increased warming by clouds due to LW effects. If 572 the fraction of low clouds is overestimated in summer as suggested by the comparison to 573 the findings of Kay and Gettelman [2009] (section 3.1), the strong SW effects in summer 574 will be overestimated and it is possible that there is a negative bias in the aerosol indirect 575 effect. 576 During summer the net cloud forcing in present day conditions is positive over ice 577 covered surfaces, while the areas of open water and the southern regions of the Arctic 578 experience negative cloud forcing (not shown). The LW component thus dominates where 579 the surface albedo is high. Despite this, the change in surface net cloud forcing with 580 anthropogenic aerosols is negative over most of the Arctic region during the summer 581 (Figure 9d). The large change in SW cloud forcing dominates the change in net forcing 582 completely, even over areas covered by surface ice. The change in net surface cloud forcing 583 averages to −1.18 W/m2 north of 71◦ N during the summer. 584 In the winter, the SW cloud forcing is of less importance than in the summer and the 585 change in net forcing with pollution is dominated by LW effects. Anthropogenic aerosols 586 interacting with clouds lead to a net increase in the winter surface flux on the order of 587 0.14 W/m2 north of 71◦ N. D R A F T July 20, 2010, 6:20am D R A F T X - 30 ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC 5. Discussion 588 We will now compare the results from the ’LWP/5 ∆CDNC’ simulations to earlier 589 findings and go on to discuss the sensitivity of our results to our assumptions, as well as to 590 the sulfate concentration. The ’LWP/5 ∆CDNC’ simulations are highlighted in this work 591 because the simulated surface cloud forcing and LWP agree well with the measurements 592 taken during the SHEBA campaign. As this campaign was limited both in time and 593 space, focusing only on these simulations may not give an adequate overall view of the 594 Arctic conditions. In this section we will therefore study the sensitivity of our results to 595 the model LWP through several sensitivity experiments not directly linked to the SHEBA 596 campaign. In addition to annual, winter and summer averages the sensitivity experiments 597 include spring averages (January to April) to show that results from the polluted spring 598 months are in agreement with what is presented in section 4. 5.1. Comparison With Earlier Findings 599 5.1.1. Longwave 600 The simulated annual increase in surface LW cloud forcing of 0.55 W/m2 from pre- 601 industrial times until today is one order of magnitude less than the change in this forcing 602 suggested by GZ06 and LV06. In these two articles, increases of between 3.3 and 8.2 603 W/m2 were found when going from pristine to polluted conditions under cloudy skies. 604 We will in the following paragraphs discuss possible reasons for this discrepancy. 605 First of all, the GZ06 and LV06 studies show the radiative effect of increased aerosol 606 levels in a certain area and under certain conditions. We, on the other hand, study the 607 overall change in surface cloud forcing under all conditions and over the entire Arctic 608 region. The different goals of the studies also result in fundamental differences in the D R A F T July 20, 2010, 6:20am D R A F T ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC X - 31 609 approach used. First, GZ06 and LV06 have looked at specific conditions for cloud type 610 and pollution, and it may be that these favorable conditions are met too seldom or over too 611 short time periods to affect our monthly averaged results. If this is the case, instantaneous 612 results should include signals of change that are significantly larger than our seasonal 613 averages. The fraction of time with changes in LWCFS above 3.3 W/m2 in the Arctic 614 region is plotted in Figure 10. The value 3.3 W/m2 is the lower limit for the increase in 615 surface flux found by GZ06 and LV06. The plot shows that changes of this magnitude 616 occur throughout the year and the 3D model thus simulates changes that are consistent 617 with the range observed by GZ06 and LV06. The fraction of time this occurs, however, 618 is very limited, with a peak of approximately 4% in late summer/early fall. Results 619 around Barrow, Alaska, show similar seasonal variation and changes of the same order of 620 magnitude as the Arctic average. 621 Secondly, the clean and the polluted scenarios found in GZ06 and LV06 contain the lower 622 and the upper quartile of present day aerosol concentrations respectively. This means 623 that they compare situations with especially large differences in aerosol conditions, while 624 this study compares all conditions of the present day regime to the clean pre-industrial 625 scenario. This will influence the magnitude of change found in the surface LWCF. It is 626 difficult to say whether the different approaches used in the studies are sufficient to explain 627 why our average results are lower than the earlier findings, as there is no information on 628 the fraction of time studied in the GZ06 and LV06 articles. 629 The average results found in our study may also be influenced by possible model arti- 630 facts. In the following paragraphs, we investigate whether features of the LW radiation 631 scheme and the simulated cloud cover and cloud properties can explain the differences in D R A F T July 20, 2010, 6:20am D R A F T X - 32 ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC 632 results. We used the 1D model to study whether the LW radiation scheme itself was ca- 633 pable of reproducing the findings of GZ06, using their observed changes in re and LWP as 634 input. In doing this we forced the simulated clouds to be similar to the ones observed; the 635 clouds are at low levels (below 1.5 km) and are all liquid. The model was run with cloud 636 parameter input similar to what was observed under both pristine (re = 12.9 µm and 637 LWP = 31.1 g/m2 ) and polluted conditions (re = 9.9 µm and LWP = 33.5 g/m2 ). GZ06 638 found changes in LWCFS of between 3.3 and 5.2 W/m2 , while the 1D model simulates 639 a change of 2.1 to 2.6 W/m2 depending on cloud base height and season. The difference 640 between observed and modeled changes may be due to the temperature profiles of the 1D 641 model, which are averaged north of 65.5◦ N, not being representative for the area studied 642 by GZ06. It may also be due to the fact that in these 1D tests we simulate July and 643 January only and therefore do not get an annual mean as presented in GZ06. Despite 644 the noted difference, the results are of the same order of magnitude as findings by GZ06. 645 This suggests that the LW radiation scheme used both in the 1D and the 3D model reacts 646 to changes in LWP and re in accordance with observations. The radiation scheme itself 647 is therefore not likely to cause the large discrepancy between the 3D model results and 648 observations. 649 The LW indirect effect at the surface will also be affected by the simulated cloud frac- 650 tion and the sensitivity of cloud LW emissivity to changes in cloud parameters. As noted 651 in section 3.1 the CAM-Oslo cloud fraction is lower than most observations, and under- 652 estimation of the radiative effect of cloud-aerosol interactions is likely (see Figure 1a). 653 However, a discrepancy of up to 30 % in cloud cover alone is not enough to explain the 654 difference in results found between this and earlier studies. The sensitivity of the LW D R A F T July 20, 2010, 6:20am D R A F T ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC X - 33 655 emissivity will also influence the LW aerosol indirect effect. It increases with decreasing 656 LWP and becomes especially large for LWP below 20 g/m2 (Figure 8). Curry and Her- 657 man [1985] observed from aircraft measurements that the LWP of the Arctic stratus is 658 frequently below this value. In our simulations this occurs 36 % of the time annually 659 when clouds are present and we cannot rule out that this time fraction may be too low. 660 Based on this and the underestimated cloud cover it is possible that the time fraction of 661 4 % found to have surface indirect effects consistent with the findings of GZ06 and LV06 662 is somewhat underestimated. 663 In addition, the results will be affected by the magnitude of change in cloud parameters 664 with increasing aerosol load. In sections 4.2 and 4.3 we found that the changes in re and 665 LWP averaged in height are consistent with observations. Although the integrated LWP 666 includes changes in water amounts at all altitudes, the liquid water amount changes most 667 close to the surface (Figure 6b) and it is at these levels that we expect the largest influence 668 on the LW cloud forcing. The effective radius, on the other hand, changes much less in 669 surface layers than it does averaged in height. While GZ06 found a decrease in re of 3 µm 670 between pristine and polluted conditions, our annually averaged results show a change 671 of 0.6 to 1.0 µm in layers important to the LWCFS. However, annually around Barrow, 672 Alaska, the model reproduces the observed reductions in low level cloud effective radius 673 10 % of the time. 674 One final aspect that may lead to discrepancies in results is differences in the weather 675 and temperature conditions between the model and the observed cases. The change in 676 LW surface cloud forcing with pollution is influenced by the temperature of the cloud 677 base. If the vertical temperature profiles in our simulations differ from those common at D R A F T July 20, 2010, 6:20am D R A F T X - 34 ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC 678 the measuring sites used by GZ06 and LV06, it will affect the results. However, large sys- 679 tematic biases would be needed for this to greatly influence the results. Additionally, the 680 results shown in this section are averaged over five years and no particular meteorological 681 event or temperature anomaly will affect the average results. 682 In summary, it is clear that there are significant differences between this and the two 683 earlier studies of the LW indirect at the Arctic surface. While our study aims to show the 684 overall importance of the phenomenon, the GZ06 and LV06 studies show its magnitude 685 under specific conditions. In addition to the differences between the studies the low change 686 in surface LWCF may be influenced by an underestimated cloud fraction and by a possible 687 underestimation of the frequency of the most sensitive clouds in our simulations. However, 688 if the simulated indirect effect at the surface is to reach the values found by GZ06 and 689 LV06, large changes are needed in these parameters. 690 5.1.2. Shortwave 691 There has been less emphasis on the SW than on the LW indirect effect in the Arctic, 692 and observational data similar to what were used by GZ06 and LV06 to study the LW effect 693 are not available for the visible through near-infrared wavelengths. Lubin and Vogelmann 694 [2007] have, however, simulated the SW first indirect effect during springtime for water 695 clouds at four different Arctic latitudes. They used a 179-band SW discrete-ordinates 696 method and used changes in re as described by GZ06. In doing this, they found that 697 for March and April the changes in the SW radiative flux at the surface due to the first 698 indirect effect are comparable in magnitude to the increased LW flux found by GZ06 and 699 LV06. For May and June, the decrease in the SW surface flux was larger than the increase D R A F T July 20, 2010, 6:20am D R A F T ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC X - 35 700 in the LW, meaning that the simulated net surface radiation under cloudy skies decreases 701 due to the first indirect effect. These findings are consistent with our results (section 4). 5.2. Sensitivity to Simulated LWP and IWP 702 We now investigate whether our results are sensitive to the manner in which the model 703 LWP is reduced (section 3.5) or to the magnitude of these reductions. Table 3 shows 704 the change in surface cloud forcing with anthropogenic aerosols for several simulations in 705 which different modifications of the LWP and the IWP were used. 706 From Table 3 it is clear that the manner in which the model LWP is reduced does not 707 greatly affect the anthropogenic change in surface cloud forcing. Results from the standard 708 CAM-Oslo are shown under the case name ’std. LWP’ in the second column of the table. 709 Columns 3 to 6 list results from simulations where the LWP was decreased by a factor of 710 five or ten through either reducing the cloud droplet number concentration (’LWP/5(10) 711 ∆CDNC’) or through reducing the cloud droplet effective radius (’LWP/5(10) ∆re ’). As 712 the model re was found to be consistent with observations (section 3.5), reducing this 713 quantity is not something we consider to be physically accurate, but rather a test of the 714 other “extreme” way of changing the LWP besides reducing the CDNC. 715 Reducing the LWP by a factor of ten brings it well below the values observed during the 716 SHEBA campaign (Figure 1b), and we consider these simulations to be tests where the 717 clouds are in general too thin and therefore too sensitive to changes in CCN concentration. 718 The changes in LW cloud forcing simulated with these model versions are larger than they 719 are for simulations in which the LWP was reduced by a factor of five (see Table 3 under 720 case names ’LWP/10(5) ∆CDNC’ and ’LWP/10(5) ∆re ’). One exception is the wintertime 721 change in LWCFS labeled ’LWP/10 ∆re ’, in which reductions in cloud droplet size bring D R A F T July 20, 2010, 6:20am D R A F T X - 36 ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC 722 the effective radii more frequently below 10 µm and therefore make the LW emissivity 723 less sensitive to changes in re (Equation 5). Unlike in the LW, the changes in SW cloud 724 forcing are smaller in magnitude for LWP/10 simulations than for simulations with smaller 725 reductions in LWP. The SWCFS is highly influenced by the absolute changes in re and 726 LWP being smaller in LWP/10 simulations than in LWP/5 simulations, while the LWCFS 727 is influenced by a large increase in the sensitivity of the LW cloud emissivity to changes 728 in re and LWP for low LWPs (see Figure 8). Notice that the case where LWP is reduced 729 by a factor of ten through reducing the CDNC is the only case that simulates an increase 730 in surface cloud forcing with pollution on an annual basis (0.12 W/m2 ). 731 Results from simulations where the LWP is reduced by modifying the auto-conversion 732 are shown in column 7 under the case name ’1st Auto-conv’ (r3lc = 10 µm and autlim = 733 0.5 mm day −1 , see section 3.5). The LWCFS changes less from pre-industrial times to 734 present day in this case than in the case presented in section 4 (’LWP/5 ∆CDNC’), while 735 the opposite is true for the SWCFS. We also ran a test where the ice fraction of the cloud 736 (fice) was increased as in section 3.5 (’Fice + LWP/5 ∆CDNC’). This case shows smaller 737 changes in both LW and SW cloud forcing at the surface than results discussed in section 738 4. The table further contains results from modifications of both the auto-conversion and 739 the cloud particle ice fraction as described in section 3.5 (’Auto-conv. + Fice’). This 740 simulation behaves similarly to the ’1st Auto-conv’ simulation. Neither of these cases 741 point to an increased warming effect of the Arctic clouds due to anthropogenic aerosols, 742 and the magnitude of change is relatively insensitive to how the model liquid water path 743 is reduced. D R A F T July 20, 2010, 6:20am D R A F T X - 37 ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC 744 In summary Table 3 contains results both from simulations where the cloud LWP is 745 reduced in crude manners by simply reducing the size of the cloud particles or the CDNC 746 and from simulations where it is reduced in more physically accurate manners. None of 747 them show signals of change significantly larger than what we found using the version in 748 which the LWP was reduced by a factor of five through reductions in the CDNC. 5.3. Sensitivity to the Treatment of LW Absorption Efficiency 749 In section 2.2 we presented a new parametrization of the LW absorption efficiency, Qa , 750 as part of the LW absorption coefficient, kabs . We will now investigate whether our results 751 are sensitive to this parametrization. 752 As noted in section 2.2, the parametrization of Qa is a simplification because we ignore 753 the effects of the cloud droplet spectrum and the thermal radiation wavelength spectrum. 754 One consequence of this may be that we overestimate the sensitivity of Qa to droplet 755 size for small droplets and therefore underestimate the sensitivity of kabs and the LW 756 emissivity to droplet size for re <10 µm (kabs = 757 simulations in which the Qa was (unrealistically) set to one for all effective radii rendering 758 kabs sensitive to cloud droplet size for all re : kabs = 3 Qa 4 ρ L re = 3 0.1 ). 4 ρL 3 1 , f or all re 4 ρL r e We ran a set of test (15) 759 The results of these simulations show a slightly increased change in LWCFS compared 760 to the standard ’LWP/5 ∆CDNC’ run (see Table 3 under case name ’Qa =1 LWP/5 761 ∆CDNC’), but the results do not contradict our earlier findings. We conclude that the 762 Qa parametrization for re < 10 µm is not what causes the small simulated changes in LW 763 cloud forcing at the surface from pre-industrial times until today. D R A F T July 20, 2010, 6:20am D R A F T X - 38 ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC 5.4. Sensitivity to Aerosol Concentration 764 In this section we examine whether the simulated change in surface cloud forcing is un- 765 derestimated because of low aerosol concentrations (Figure 4). To do this we increase the 766 concentrations by either decreasing the in-cloud scavenging or by increasing the emissions 767 of SO4 precursors. Results shown in this subsection are from simulations in which the 768 LWP is reduced by a factor of five through ∆CDNC. 769 The high liquid water content of the CAM-Oslo model in the Arctic may lead to an 770 overestimated in-cloud scavenging. In order to check whether this is of importance we 771 ran test runs in which the in-cloud scavenging coefficient was reduced from 1 to 0.1, 772 thus reducing the wet deposition on SO4 . This allows us to investigate whether the low 773 simulated change in cloud forcing is due to an underestimation of CCN available to affect 774 cloud properties. Results from these runs show average SO4 concentrations that exceed 775 the observational means by 201% at the Zeppelin station, 228% at Janiskoski, 223% at 776 Alert and 176% at Barrow. 777 The change in cloud forcing simulated with reduced in-cloud scavenging shows that the 778 increased concentration of SO4 does not greatly affect the indirect forcing (see Table 3 779 under case name ’In-cloud scav. + LWP/5 ∆CDNC’). Comparing these results to results 780 under ’LWP/5 ∆CDNC’ we see that the LWCFS changes less than for simulations where 781 the scavenging is left unchanged, while the SWCFS changes more. This can be explained 782 by the reduction in in-cloud scavenging leading to reduced effective radii and increased 783 LWP of clouds pre-industrially (∆re = -0.4 µm, ∆LWP= 2.3 g/m2 ) as well as in present 784 day. D R A F T July 20, 2010, 6:20am D R A F T ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC X - 39 785 The sensitivity of our results to aerosol concentrations was tested further through simu- 786 lations in which the present day emissions of SO2 from fossil fuel combustion were doubled 787 (2SOx) (see Table 1). This led to an increase in sulfate burden of approximately the same 788 magnitude as the increase from pre-industrial times until present day (∆SO4 (2SOx-PD)= 789 1.07 * 10−6 kg S/m2 vs. ∆SO4 (PD-PI)= 0.89 * 10−6 kg S/m2 ). Results from this simu- 790 lation are shown under case name ’2SOx - PI LWP/5 ∆CDNC’ in Table 3. Even though 791 the difference in SO4 concentration between the runs nearly doubles, the forcing only 792 increases by about 50 %. This experiment shows that even with significantly increased 793 concentrations of SO4 , the signals of change in LWCFS are not within the range found by 794 GZ06 and LV06. The SW effects dominate results from this simulation as well as most 795 others. Only during winter is the net radiative effect of increased aerosol levels positive. 6. Summary and Conclusions 796 The observed increase in surface air temperature over the last decades is almost twice 797 as large in the Arctic as in the rest of the world [e. g. Graversen et al., 2008]. Garrett 798 and Zhao [2006] and Lubin and Vogelmann [2006] suggest that aerosol-cloud interactions 799 may contribute to the observed temperature amplification in this region. In this study 800 we have investigated the overall importance of the suggested increase in surface radiative 801 flux due to increased CCN concentrations in clouds. 802 Using the CAM-Oslo global climate model we have studied simulated changes in the 803 radiative balance at the Arctic surface due to aerosol-cloud interactions. The simulated 804 cloud cover, cloud water path and cloud radiative forcing were verified against observations 805 and model modifications were made to better suit the model to the focus of our study. 806 The simulated SO4 concentrations were compared both to observations and to the median D R A F T July 20, 2010, 6:20am D R A F T X - 40 ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC 807 of ten models participating in the AeroCom project (section 3.2). We found that results 808 from the CAM-Oslo model were in better agreement with observations than the AeroCom 809 model median, and conclude that the CAM-Oslo global climate model is well suited for 810 this study. Our results show that the indirect effects of anthropogenic aerosols are close 811 to the same magnitude in the LW and the SW, with a net result of −0.30 W/m2 . We have 812 conducted several sensitivity experiments that show that our findings are robust against 813 model assumptions, changes in cloud properties and aerosol concentrations. 814 Below is a summary of key findings concerning aerosol-indirect effect at the Arctic 815 surface. Numbers in parentheses give the minimum and the maximum results from the 816 sensitivity experiments. 817 • The simulated increase in LW cloud forcing at the surface due to anthropogenic 818 aerosols averages to 0.55 (0.10 to 0.85) W/m2 annually, to 0.99 (0.16 to 1.53) W/m2 819 from June to September and to 0.33 (0.07 to 0.51) W/m2 from October to May. The 820 seasonal variation is caused by larger changes in cloud emissivity in summer than in 821 winter, combined with high fractions of low clouds in summer. 822 • The simulated LW indirect effect is one order of magnitude lower than suggested 823 by Garrett and Zhao [2006] and Lubin and Vogelmann [2006]. This discrepancy may be 824 caused by a combination of effects. GZ06 and LV06 showed the magnitude of change 825 in surface LWCF under specific conditions, whereas this study includes a variety of con- 826 ditions at all times of the year providing results for the average changes. In addition, 827 underestimation of cloud cover and a possible underestimation of the frequency of the 828 most sensitive clouds may influence the results. However, large changes are needed in 829 these parameters for the LW indirect effect to reach the values found by GZ06 and LV06. D R A F T July 20, 2010, 6:20am D R A F T ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC X - 41 830 • The corresponding simulated change in surface SW cloud forcing due to anthro- 831 pogenic aerosols averages to −0.85 (−1.29 to −0.52) W/m2 annually and −2.17 (−3.28 832 to −1.28) W/m2 in summer. 833 • The annual change in surface net cloud forcing averages to −0.30 (−0.98 to 0.12) 834 W/m2 . During the summer, the net surface cloud forcing decreases by 1.18 (2.63 to 0.23) 835 W/m2 , while in the winter LW effects dominate, and changes in cloud properties due to 836 anthropogenic aerosols increase the surface radiative flux by 0.14 (−0.15 to 0.29) W/m2 . 837 The net cloud forcing will be particularly sensitive to overestimation of the summer cloud 838 cover as the negative SW forcing is strong during this season. A possible overestimation 839 of the SWCFS in summer may lead to a negative bias in the net aerosol indirect effect. 840 • The sensitivity experiments show that the annually averaged changes in net CFS are 841 positive only in one of eleven simulations and our results suggest that increased levels of 842 anthropogenic aerosols in Arctic clouds may lead to a small decrease in the radiative flux 843 at the surface. Our general findings depend little on model assumptions, changes in cloud 844 properties and aerosol concentrations. 845 In recent years (after the fall of the Soviet Union) the emissions of the SO4 precursor 846 SO2 have decreased dramatically in Europe and Russia [Karnieli et al., 2009], and Quinn 847 et al. [2007] found that the concentrations of non-seasalt SO4 decreased by 30-70 % from 848 the early 1990s to present in the Canadian, Norwegian and Finnish Arctic. Additionally, 849 Sharma et al. [2006] found a clear downward trend in concentrations of equivalent black 850 carbon (BC) in the high Arctic. From the findings of this study it is likely that a reduction 851 in CCN amount in the Arctic will decrease the magnitude of the negative indirect effect 852 and therefore in sum work to increase the net positive cloud forcing found in this region. D R A F T July 20, 2010, 6:20am D R A F T X - 42 ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC 853 The less pollution that enters the Arctic region, the larger the increase in surface radiative 854 flux compared to present day. 855 Furthermore, the expected reduction in the polar ice cap and therefore in the surface 856 albedo may increase the importance of the negative SW surface cloud forcing. Larger areas 857 of open waters may, on the other hand, affect both the cloud fraction and the thickness 858 of the clouds in the region and it is difficult to predict the net effect of sea ice reductions. 859 Although the simulated changes in surface cloud forcing are smaller than what is found 860 in earlier studies, they are of the same order of magnitude as the BC surface forcing via 861 snow and ice albedos in sea ice areas. Flanner et al. [2007] estimate that the annual 862 mean of the instantaneous surface forcing of BC on snow in these areas are around 0.20 863 W/m2 during a year of average BC emissions. The changes in Arctic surface cloud forcing 864 due to anthropogenic aerosols may therefore be of importance and should be studied 865 further. To enable this, the accuracy of climate models needs to be improved, especially 866 in dealing with cloud water amount and conversion between liquid water and ice particles, 867 and a higher frequency and larger geographical spread in Arctic measuring campaigns are 868 needed. The current lack of comprehensive observations limits the possibility of verifying 869 and improving current climate models. 870 Acknowledgments. This study was partly funded by the Norwegian Research Council 871 through the projects POLARCAT (grant No. 175916) and NorClim (grant No. 178246), 872 and has received support from the Norwegian Research Council’s Programme for Super- 873 computing through a grant of computing time. The authors would like to thank Aero- 874 Com for access to their data, and are also grateful to Alf Kirkevåg, Øyvind Seland, Frode 875 Stordal, Terje Berntsen and Gunnar Myhre for helpful discussions. Finally, we are thank- D R A F T July 20, 2010, 6:20am D R A F T ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC X - 43 876 ful to three anonymous reviewers whose comments led to significant improvements of the 877 paper. References 878 AMAP (2006), Arctic Monitoring and Assessment Programme (AMAP) assessment 879 2006: 880 //www.amap.no. 881 882 883 884 Acidifying pollutants, Arctic haze and acidification in the Arctic, http : Barrie, L. A. (1986), Arctic air pollution: An overview of current knowledge, Atmos. Environ., 20, 643–663. Chýlek, P., P. Damiano, and E. P. Shettle (1992), Infrared emittance of water clouds, J. Atmos. Sci., 49, 1459–1472. 885 Collins, W. D., et al. (2004), Description of the NCAR Community Atmosphere 886 Model (CAM 3.0), National Center For Atmospheric Research (NCAR), NCAR/TN- 887 464+STR. 888 889 890 891 892 893 894 895 Collins, W. D., et al. (2006a), The formulation and atmospheric simulation of the Community Atmosphere Model Version 3 (CAM3), J. Climate, 19, 2144–2161. Collins, W. D., et al. (2006b), The Community Climate System Model Version 3 (CCSM3), J. Climate, 19, 2122–2143. Curry, J. A., and G. F. Herman (1985), Infrared radiative properties of summertime Arctic stratus clouds, J. Clim. Appl. Meteorol., 24, 525–538. Curry, J. A., et al. (2000), FIRE Arctic Clouds Experiment, Bull. Am. Meteorol. Soc., 81, 5–29. D R A F T July 20, 2010, 6:20am D R A F T X - 44 896 897 898 899 900 901 ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC Dentener, F., et al. (2006), Emissions of primary aerosol and precursor gases in the years 2000 and 1750 prescribed data-sets for AeroCom, Atmos. Chem. Phys., 6, 4321–4344. Dong, X., and G. G. Mace (2003), Arctic stratus properties and radiative forcing derived from ground-based data collected at Barrow, Alaska, J. Climate, 16, 445–461. Dreiling, V., and B. Friederich (1997), Spatial distribution of the Arctic haze aerosol size distribution in western and eastern Arctic, Atmos. Res., 44, 133–152. 902 Flanner, M. G., C. S. Zender, J. T. Randerson, and P. J. Rasch (2007), Present-day 903 climate forcing and response from black carbon in snow, J. Geophys. Res., 112, D11202, 904 doi:10.1029/2006JD008003. 905 906 907 908 Garrett, J. T., L. F. Radke, and P. V. Hobbs (2002), Aerosol effects on cloud emissivity and surface longwave heating in the Arctic, J. Atmos. Sci., 59, 769–778. Garrett, T. J., and C. Zhao (2006), Increased Arctic cloud longwave emissivity associated with pollution from mid-latitudes, Nature, 440, 787–789, doi:10.1038/nature04636. 909 Graversen, R. G., T. Mauritsen, M. Tjernström, E. Källen, and G. Svensson (2008), Ver- 910 tical structure of recent Arctic warming, Nature, 451, 53–56, doi:10.1038/nature06502. 911 Hoose, C., J. E. Kristjánsson, T. Iversen, A. Kirkevåg, Ø. Seland, and A. Gettelman 912 (2009), Constraining cloud droplet number concentration in GCMs suppresses the 913 aerosol indirect effect, Geophys. Res. Lett., 36, L12807, doi:10.1029/2009GL038568. 914 Intrieri, J. M., C. W. Fairall, M. D. Shupe, P. O. G. Persson, E. L. Andreas, P. S. Guest, 915 and R. E. Moritz (2002a), An annual cycle of Arctic surface cloud forcing at SHEBA, 916 J. Geophys. Res., 107 (C10), 8039, doi:10.1029/2000JC000439. 917 Intrieri, J. M., M. D. Shupe, T. Uttal, and B. J. McCarty (2002b), An annual cycle of 918 Arctic cloud characteristics observed by radar and lidar at SHEBA, J. Geophys. Res., D R A F T July 20, 2010, 6:20am D R A F T X - 45 ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC 919 107 (C10), 8030, doi:10.1029/2000JC000423. 920 Karnieli, A., Y. Derimian, R. Indoitu, N. Panov, R. C. Levy, L. A. Remer, W. Maen- 921 haut, and B. N. Holben (2009), Temporal trend in anthropogenic sulfur aerosol trans- 922 port from central and eastern Europe to Israel, J. Geophys. Res., 114, D00D19, doi: 923 10.1029/2009JD011870. 924 925 Kay, J. E., and A. Gettelman (2009), Cloud influence on and response to seasonal Arctic sea ice loss, J. Geophys. Res., 114, D18204, doi:10.1029/2009JD011773. 926 Kiehl, J. T., J. J. Hack, G. B. Bonan, B. A. Boville, D. L. Williamson, and P. J. Rasch 927 (1998), The National Center for Atmospheric Research Community Climate Model: 928 CCM3, J. Climate, 11, 1131–1149. 929 930 931 932 Kristjánsson, J. E. (2002), Studies of the aerosol indirect effect from sulfate and black carbon aerosols, J. Geophys. Res., 107 (D15), 4246, doi:10.1029/2001JD000887. Liou, K. N. (2002), An introduction to atmospheric radiation, 2nd ed., Academic Press, p.373. 933 Löhnert, U., G. Feingold, T. Uttal, A. S. Frisch, and M. D. Shupe (2003), Analysis of two 934 independent methods for retrieving liquid water profiles in spring and summer Arctic 935 boundary clouds, J. Geophys. Res., 108 (D7), doi:10.1029/2002JD002861. 936 937 Lubin, D., and A. M. Vogelmann (2006), A climatologically significant aerosol longwave indirect effect in the Arctic, Nature, 439, 453–456, doi:10.1038/nature04449. 938 Lubin, D., and A. M. Vogelmann (2007), Expected magnitude of the aerosol shortwave 939 indirect effect in springtime Arctic liquid water clouds, Geophys. Res. Lett., 34, L11801, 940 doi:10.1029/2006GL028750. D R A F T July 20, 2010, 6:20am D R A F T X - 46 ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC 941 Maslanik, J. A., J. Key, C. W. Fowler, T. Nguyen, and X. Wang (2001), Spatial and 942 temporal variability of satellite-derived cloud and surface characteristics during FIRE- 943 ACE, J. Geophys. Res., 106 (D14), 15,223–15,249. 944 McFarquhar, G. M., G. Zhang, M. R. Poellot, G. L. Kok, R. McCoy, T. Tooman, 945 A. Fridlind, and A. J. Heymsfield (2007), Ice properties of single-layer stratocumulus 946 during the Mixed-Phase Arctic Cloud Experiment: 1. observations, J. Geophys. Res., 947 112, D24201, doi:10.1029/2007JD008633. 948 Meador, W. E., and W. R. Weaver (1980), Two-stream approximations to radiative trans- 949 fer in planetary atmospheres: a unified description of existing methods and a new im- 950 provement, J. Atmos. Sci., 37, 630–643. 951 Morrison, H., M. D. Shupe, and J. A. Curry (2003), Modeling clouds observed at SHEBA 952 using a bulk microphysics parameterization implemented into a single-column model, 953 J. Geophys. Res., 108 (D8), 4255, doi:10.1029/2002JD002229. 954 Morrison, H., et al. (2009), Intercomparison of model simulations of mixed-phase clouds 955 observed during the ARM Mixed-Phase Arctic Cloud Experiment. II: Multilayer cloud, 956 Q. J. R. Meteorol. Soc., 135, 1003–1019. 957 Paltridge, G. W., and C. M. R. Platt (1976), Radiative processes in meteorology and cli- 958 matology, Developments in atmospheric science, Elsevier Scientific publishing company. 959 Quinn, P. K., G. Shaw, E. Andrews, E. G. Dutton, T. Ruoho-Airola, and S. L. Gong 960 961 962 (2007), Arctic haze: current trends and knowledge gaps, Tellus, 59B, 99–114. Quinn, P. K., et al. (2008), Short-lived pollutants in the Arctic: their climate impact and possible mitigation strategies, Atmos. Chem. Phys., 8, 1723–1735. D R A F T July 20, 2010, 6:20am D R A F T ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC X - 47 963 Rasch, P. J., and J. E. Kristjánsson (1998), A comparison of the CCM3 model climate 964 using diagnosed and predicted condensate parameterizations, J. Climate, 11, 1587–1614. 965 Scheuer, E., R. W. Talbot, J. E. Dibb, G. K. Seid, L. DeBell, and B. Lefer (2003), Seasonal 966 distributions of fine aerosol sulfate in the North American Arctic basin during TOPSE, 967 J. Geophys. Res., 108 (D4), 8370, doi:10.1029/2001JD001364. 968 Seland, Ø., T. Iversen, A. Kirkevåg, and T. Storelvmo (2008), Aerosol-climate interactions 969 in the CAM-Oslo atmospheric GCM and investigation of associated basic shortcomings, 970 Tellus, 60A, 459–491. 971 Sharma, S., E. Andrews, L. A. Barrie, J. A. Ogren, and D. Lavoué (2006), Variations 972 and sources of the equivalent black carbon in the high arctic revealed by long-term 973 observations at Alert and Barrow: 1989 - 2003, J. Geophys. Res., 111, D14208, doi: 974 10.1029/2005JD006581. 975 976 Shindell, D. T., et al. (2008), A multi-model assessment of pollution transport to the Arctic, Atmos. Chem. Phys., 8, 5353–5372. 977 Shupe, M. D., and J. M. Intrieri (2004), Cloud Radiative Forcing of the Arctic Surface: 978 The Influence of Cloud Properties, Surface Albedo, and Solar Zenith Angle, J. Climate, 979 17, 616–628. 980 981 982 983 Shupe, M. D., S. Y. Matrosov, and T. Uttal (2006), Arctic mixed-phase cloud properties derived from surface-based sensors, J. Atmos. Sci., 63, 697–711. Stohl, A. (2006), Characteristics of atmospheric transport into the Arctic troposphere, J. Geophys. Res., 111, D11306, doi:10.1029/2005JD006888. 984 Storelvmo, T., J. E. Kristjánsson, S. J. Ghan, A. Kirkevåg, Ø. Seland, and T. Iversen 985 (2006), Predicting cloud droplet number concentration in Community Atmosphere D R A F T July 20, 2010, 6:20am D R A F T X - 48 Figure 1. ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC a) Observed and simulated total cloud fraction. Black: The Arctic region. Color: SHEBA region. b) Monthly variation in average cloud LWP in the SHEBA region. Observed values reproduced from Zhang et al. [2002]. Figure 2. Simulated zonally averaged annual cloud fraction north of 65◦ N. The black line indicates 1 km height above sea level. Figure 3. a) Simulated annually averaged column integrated sulfate concentrations in the Arctic region [kg S/m2 ], present day. b) Simulated changes in column integrated SO4 concentrations from pre-industrial times to present day [kg S/m2 ], annual mean. The characters plotted show the locations of the stations in Figure 4; Z: Zeppelin; J: Janiskoski; A: Alert; B: Barrow. 986 987 988 989 990 Model (CAM)-Oslo, J. Geophys. Res., 111, D24208, doi:10.1029/2005JD006300. Textor, C., et al. (2006), Analysis and quantification of the diversities of aerosol life cycles within AeroCom, Atmos. Chem. Phys., 6, 1777–1813. Twomey, S. (1977), The influence of pollution on the shortwave albedo of clouds, J. Atmos. Sci., 34, 1149–1152. 991 Wang, X., and J. R. Key (2005), Arctic surface, cloud, and radiation properties based on 992 the AVHRR Polar Pathfinder dataset. part I: spatial and temporal characteristics, J. 993 Climate, 18, 2558–2574. 994 Warren, S. G., C. J. Hahn, J. London, R. M. Chervin, and R. L. Jenne (1988), Global dis- 995 tribution of total cloud cover and cloud type amounts over the ocean., NCAR Tech.Note, 996 NCAR/TN317+STR, 1–42. 997 Zhang, J., U. Lohmann, and B. Lin (2002), A new statistically based autoconversion 998 rate parameterization for use in large-scale models, J. Geophys. Res., 107 (D24), 4750, 999 doi:10.1029/2001JD001484. D R A F T July 20, 2010, 6:20am D R A F T ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC Figure 4. X - 49 Observed (green) and simulated (blue) SO4 concentrations at selected Arctic stations [µg S/m3 ]. The locations of the stations are plotted in Figure 3. Observational data from AMAP [2006] have been averaged over five years and error bars show the maximum and minimum observed monthly means. The median of 10 AeroCom A models [Textor et al., 2006] for Spitsbergen (including Zeppelin) and Janiskoski for the year 2000 is shown in red. Simulated SO4 from both CAM-Oslo and the AeroCom A models are non sea salt only. Note that the axes differ. Figure 5. a) Simulated zonally averaged annual sulfate concentrations in the Arctic region [µg S/m3 ], present day. b) Simulated zonally averaged changes in sulfate concentrations from pre-industrial times to present day [µg S/m3 ], annual mean. c) Simulated annual Arctic mean of effective cloud droplet radius [µm] in present day conditions. d) Simulated zonally averaged change in annually averaged effective radius from pre-industrial times to present day [µm]. Note that the color bars differ. Figure 6. a) Simulated annually averaged change in LWP [g/m2 ] from pre-industrial times to present day. b) Simulated zonally and annually averaged change in in-cloud liquid water mixing ratio [kg/kg] from pre-industrial times to present day. Figure 7. Fraction of time when clouds are present with vertically integrated LWP below 50 g/m2 . The presence of clouds is defined as times when LWP > 5 g/m2 . Results are averaged north of 71◦ N. Figure 8. Solid grey lines: Cloud SW albedo (thick) and LW emissivity (thin) as a function of liquid water path (bottom axis, re = 11.6 µm). Dashed black lines: Cloud SW albedo (thick) and LW emissivity (thin) as a function of effective radius (top axis, LWP = 43.5 g/m2 ). Cloud ice fraction = 0. D R A F T July 20, 2010, 6:20am D R A F T X - 50 Figure 9. ALTERSKJÆR ET AL.: AEROSOL INDIRECT EFFECTS IN THE ARCTIC Simulated anthropogenic change in cloud forcing at the surface [W/m2 ] from pre- industrial times to present day. a) ∆LWCFS, winter season (October - May). b) ∆LWCFS, summer season (June - September). c) ∆SWCFS, summer season. d) ∆Net CFS, summer season. Note that the color scales are reversed in c) and d). Figure 10. Fraction of time that the change in LW cloud forcing at the surface from pre- industrial times to present day is greater than 3.3 W/m2 , Arctic average. D R A F T July 20, 2010, 6:20am D R A F T JOURNAL OF GEOPHYSICAL RESEARCH, VOL. ???, XXXX, DOI:10.1029/, Tables Copyright 2010 by the American Geophysical Union. 0148-0227/10/$9.00 1 X-2 : TABLES Global annual emissions of Table 1. DiMethyl Sulphide (DMS), SO2 , particulate SO4 , black carbon (BC), particulate organic matter (OM) including secondary organic aerosols, sea-salt (SS) and mineral dust (DUST) [Tg yr−1 ]. Emissions of DMS, SO2 and SO4 are in Tg S yr−1 . PD: Present day emissions [Dentener et al., 2006], PI: Pre-industrial emissions, 2SOx: Present day emissions except that emissions of SO2 from fossil fuel combustion are doubled compared to present day. PD PI 2SOx DMS 18.2 18.2 18.2 SO2 68.6 14.9 120.9 SO4 1.8 0.4 3.1 BC 7.7 1.3 7.7 OM 65.4 30.0 65.4 SS 7925 7925 7925 DUST 1678 1678 1678 X-3 : TABLES Table 2. Observed and simulated surface cloud forcing in the SHEBA region. Note that a turbulent heat flux of approximately -6 W/m2 has been subtracted from the SHEBA total cloud forcing described in Intrieri et al. [2002a]. SHEBA CAM-Oslo CAM-Oslo CAM-Oslo 1st 2nd Auto std. LWP LWP/5 ∆CDNC LWP/5 ∆re auto-conv. auto-conv. fice Annual average LWCFS 35 to 41 39.1 35.8 38.2 39.0 40.9 37.1 SWCFS -10.5 to -9.5 -23.1 -13.6 -16.8 -20.5 -20.0 -18.6 Net CFS 25 to 30 16.0 22.1 21.4 18.6 20.9 18.5 LWCFS 25 to 30 28.4 27.2 28.7 28.7 29.4 26.2 SWCFS -1 to 0 -4.7 -2.5 -3.0 -4.1 -4.2 -3.4 Net CFS 24 to 30 23.7 24.7 25.7 24.6 25.2 22.9 LWCFS 45 to 50 60.7 52.9 57.3 59.8 64.0 58.9 SWCFS -26 to -24 -60.0 -36.0 -44.5 -53.2 -51.6 -49.0 Net CFS 19 to 21 0.7 16.9 12.8 6.5 12.4 9.9 Winter average Summer average 0.10 -0.84 -0.74 0.07 -0.22 -0.15 0.16 -2.07 -1.90 0.04 -0.13 -0.09 Winter average ∆ LWCFS ∆ SWCFS ∆ Net CFS Summer average ∆ LWCFS ∆ SWCFS ∆ Net CFS Spring average ∆ LWCFS ∆ SWCFS ∆ Net CFS 0.19 -0.11 0.08 0.99 -2.17 -1.18 0.33 -0.19 0.14 0.55 -0.85 -0.30 0.06 -0.10 -0.04 0.53 -2.07 -1.55 0.14 -0.20 -0.06 0.27 -0.83 -0.56 0.28 -0.07 0.21 1.46 -1.69 -0.23 0.44 -0.15 0.29 0.78 -0.66 0.12 0.07 -0.09 -0.02 0.65 -1.96 -1.31 0.13 -0.18 -0.05 0.30 -0.78 -0.47 0.07 -0.16 -0.09 0.38 -3.01 -2.63 0.14 -0.29 -0.16 0.22 -1.20 -0.98 0.09 -0.08 0.01 0.57 -1.28 -0.72 0.15 -0.14 0.02 0.29 -0.52 -0.23 0.07 -0.12 -0.05 0.59 -2.50 -1.91 0.17 -0.21 -0.04 0.31 -0.97 -0.66 0.24 -0.09 0.15 1.02 -2.12 -1.09 0.40 -0.19 0.21 0.61 -0.83 -0.23 Sensitivity to model Qa 10 Qa = 1.0 LWP/5 ∆CDNC fossil fuel combustion are doubled compared to present day emissions. 0.14 -0.07 0.06 0.87 -2.40 -1.53 0.24 -0.16 0.07 0.45 -0.91 -0.46 0.27 -0.16 0.11 1.53 -3.28 -1.74 0.51 -0.29 0.22 0.85 -1.29 -0.44 Sensitivity to model aerosol concentration 11 12 In-cloud scav. + 2SOx - PI LWP/5 ∆CDNC LWP/5 ∆CDNC presented in the last column are from a simulation where the emissions of SO2 from Sensitivity to model LWP 2 3 4 5 6 7 8 9 std. LWP/5 LWP/5 LWP/10 LWP/10 1st Fice + LWP/5 Auto-conv. LWP ∆CDNC ∆re ∆CDNC ∆re Auto-conv. ∆CDNC + Fice Annual average ∆ LWCFS ∆ SWCFS ∆ Net CFS Column number Case name Simulated changes in surface cloud forcing between pre-industrial and present day conditions* [W/m2 ]. See sections 5.2, 5.3 and 5.4 for details. *) Results Table 3. X-4 : TABLES