Response of the climate system to aerosol direct and indirect forcing: Role of cloud feedbacks



[1] In this study, the response of the climate system to aerosol direct and indirect radiative forcing is investigated. Several multidecadal equilibrium simulations are carried out, using the NCAR CCM3 framework coupled to a separately developed aerosol treatment. The aerosol treatment includes, e.g., a life cycle scheme for particulate sulfate and black carbon (natural and anthropogenic), calculations of aerosol size distributions and CCN activation, as well as computations of direct and indirect forcing (1st and 2nd indirect effect). In all the simulations the full aerosol treatment is run online, hence responding interactively to changes in climate. By far the largest response is caused by the indirect forcing, with a globally averaged cooling of −1.25 K due to anthropogenic aerosols. The largest temperature reduction is found in the Northern Hemisphere because of a larger aerosol burden there. As a result of this cooling pattern, the Intertropical Convergence Zone is displaced southward by a few hundred kilometers. Interestingly, a similar, though less significant displacement is also found in the experiments with the direct effect alone, even though the globally averaged aerosol induced cooling is much weaker in that case, i.e., −0.08 K. The direct radiative forcing is much stronger at the surface than at the top of the atmosphere, and this leads to a slight weakening of the hydrological cycle. Comparing simulations with direct and indirect forcing combined to those with indirect and direct forcing separately, a residual, caused by nonlinear model feedbacks, is manifested through a reduction in precipitation. This reduction amounts to −0.5% in a global average and exceeds −2.5% in the Arctic, highlighting the role of high-latitude climate feedbacks. Globally, cloud feedback is negative in the sense that in the colder climate resulting from anthropogenic aerosol forcing, net cloud forcing is reduced by 15% compared to the original climate state. This is caused by a general cloud thinning, especially at high latitudes, while in the most polluted regions, clouds are thicker through the 2nd indirect effect. The 1st indirect effect, on the other hand, remains intact in the presence of climate feedbacks, yielding a similar signature of cloud droplet reduction as in the pure forcing simulations.

1. Introduction

[2] Over the last 250 years, humans have considerably altered the composition of the atmosphere. There is now little doubt that the increased concentrations of man-made greenhouse gases (CO2, CH4, N2O, CFC gases, etc. [e.g., Houghton, 2002]) will lead to a significant global warming, unless major cooling factors come into play [Houghton et al., 2001]. Much more uncertainty is attributed to the role of anthropogenic aerosols. The aerosols affect climate directly by reflecting and absorbing solar radiation, and to a lesser extent through absorption and emission of longwave radiation. They also affect climate indirectly by altering the amount and size of cloud condensation nuclei (CCN) in the atmosphere. It is very difficult to assess the impact of these two influences by measurements, although improvements have been made lately [e.g., Brenguier et al., 2003]. Consequently, most estimates of the direct and indirect effect have been obtained from simulations with global climate models (GCMs). At present model estimates vary greatly, with typical estimates of the globally averaged direct aerosol radiative forcing ranging from −0.5 to −1.0 W m−2, and a corresponding range of 0 to −2.0 W m−2 for the indirect forcing (ignoring changes in cloud lifetime and precipitation release [Ramaswamy, 2001]). If anything, the range of model estimates has widened in recent years, as more and more detailed aerosol treatments have been introduced, adding new degrees of freedom to the models. Clearly, it is of crucial importance for society at large to investigate what this range of estimates means in terms of climate and climate change. For instance, according to Anderson et al. [2003] and Andreae et al. [2005] the uncertainty regarding the cooling effect of aerosols is so large that it cannot be ruled out that this cooling may so far have almost canceled the anthropogenic greenhouse effect, implying that a very strong warming may be expected in the 21st century, when greenhouse gases are expected to greatly overrun the aerosols in terms of radiative forcing.

[3] The first studies investigating the climate response to aerosol forcing dealt only with the indirect effect, e.g., Rotstayn et al. [2000], Williams et al. [2001], and Rotstayn and Lohmann [2002]. In these studies, using three different GCMs, qualitatively similar results were found, i.e., a substantial cooling effect that was largest at high latitudes of the Northern Hemisphere, and a southward shift of the Intertropical Convergence Zone (ITCZ). The shift of the ITCZ was caused by the stronger cooling of the Northern Hemisphere than the Southern Hemisphere, while the large cooling at high latitudes was explained by ice albedo feedback mechanisms [Williams et al., 2001]. Rotstayn and Lohmann [2002] compared their results with observations and went on to suggest that the southward shift of the ITCZ might partly explain the drought in the Sahel region in Africa in the 1960s and 1970s [Folland and Karl, 2001]. Previously, the drought had been attributed mainly to natural variability, but with a possible influence from overgrazing, which tends to raise the surface albedo of this region at the rim of the Saharan desert. In two recent papers by Feichter et al. [2004] and Takemura et al. [2005], both the direct and indirect forcing were treated simultaneously. In both cases, a southward shift of the ITCZ was found, but the shift was less pronounced in the Feichter et al. [2004] study than in the other studies mentioned above. This may conceivably be related to their more complex treatment of cloud droplet nucleation, which probably tends to give less weight to the effects of sulfate than simpler schemes do.

[4] In the present paper, the climate response of both the direct and indirect effect is studied. Furthermore, by subtracting results from simulations where the indirect effect and the direct effect are simulated separately from simulations in which both effects are treated simultaneously an aerosol effect, caused by nonlinear response of the model to imposed forcings, will be estimated. We term this effect the “residual effect.” We also evaluate the semidirect effect as the change in cloud properties due to direct radiative forcing, similarly to what was done by Lohmann and Feichter [2001]. However, it should be noted that neither Lohmann and Feichter [2001] nor Penner et al. [2003] enabled the oceans to respond to the aerosol forcing in their estimates of the semidirect effect.

[5] The purpose of this study is to investigate interactions between aerosol forcing and the climate system, with a special emphasis on the hydrological cycle. On the other hand, while presenting a plausible control climate, this study does not address the more general issue of attempting to simulate today's climate as well as possible. Likewise, the impact of a doubling of CO2 in a future climate scenario is dealt with in a separate upcoming paper. The next section reviews our aerosol treatment, and describes the setup for the experiments that follow. The main results of these experiments and their interpretation are presented in section 3. This is followed by (section 4) an investigation of the role of cloud feedbacks in modifying the climate response to the aerosol forcing. Finally, section 5 presents a summary and the main conclusions of this study.

2. Model and Experimental Setup

[6] The basic model tool for the experiments is a modified version (CCM-Oslo) of the atmospheric global climate model NCAR Community Climate Model version 3 (CCM3 [Kiehl et al., 1998]), coupled to a slab ocean module. CCM3 is a state-of-the-art global climate model, run at T42 spectral truncation and with 18 levels in the vertical. The modifications to the CCM3 consist of (1) the introduction of a prognostic cloud water scheme, following Rasch and Kristjánsson [1998], and (2) the replacement of the simple aerosol scheme in CCM3 with a detailed aerosol scheme, described by Iversen and Seland [2002, 2003], Kirkevåg and Iversen [2002] and Kristjánsson [2002]. Background aerosols, consisting of sea salt, mineral and water-soluble non-sea-salt particles are prescribed and size distributed. These size distributions are then modified by adding natural and anthropogenic sulfate and black carbon (BC) into an internal mixture, brought about by condensation, coagulation in clear and cloudy air, and wet phase chemical processes in clouds. A normally minor fraction of sulfate and BC is externally mixed, produced by clear-air oxidation followed by nucleation, and by emission of primary particles. Starting from emissions of sulfate precursor gases (SO2, DMS), sulfate particles (SO4) and black carbon, chemical reactions, transport and deposition are computed at every grid point. Emission data, assumed valid for the year 2000, are the same as used by Iversen and Seland [2002, 2003] and in the IPCC TAR [Penner, 2001]. Biogenic emissions, emissions from volcanoes and a minor fraction (10%) of emissions from biomass burning are assumed natural. The remaining emissions are assumed anthropogenic. The largest emission sources in 2000 are fossil fuel combustion, biomass burning and industrial releases. A weakness of the aerosol treatment is the lack of an explicit treatment of organic carbon (OC) aerosols, which are considered to be an important component of the global aerosol [e.g., Kanakidou et al., 2005]. In the latest version of our aerosol scheme [Kirkevåg et al., 2005], OC has been incorporated into the aerosol framework described above. (3) As described in detail by Kristjánsson [2002], cloud droplet number is diagnosed from the aerosol size distributions, after accounting for humidity swelling, via assumptions on the supersaturation (0.05% in stratiform clouds, 0.10% in convective clouds over ocean and 0.15% in convective clouds over land). For a given supersaturation, the number of activated CCN is obtained from Köhler theory, and the droplet concentration is set to this value. The cloud droplet effective size is then obtained from the following relationship:

equation image

where κ is a measure of the dispersion of the cloud droplet number distribution, ql denotes in-cloud liquid water mixing ratio, ρa is air density, ρw water density and N denotes the cloud droplet number concentration. The indirect effect is simulated by (1) 1st indirect effect: feeding the droplet size from equation (1) into the radiation scheme of the model, where it influences cloud optical properties, e.g., smaller droplets due to anthropogenic aerosols will make the clouds more reflective of solar radiation, and (2) 2nd indirect effect: using the cloud droplet size from equation (1) in the calculation of warm cloud autoconversion of cloud water to rain, which according to Rasch and Kristjánsson [1998] is given by:

equation image

In the work by Kristjánsson [2002] the parameters Cl,aut and r3lc were retuned in order to yield more realistic cloud droplet sizes than the original values did, e.g., the autoconversion threshold, r3lc, was set to 10 μm. Anthropogenic aerosols will tend to make cloud droplet number (N) large and cloud droplet size (r3l) small, and this will via equation (2) reduce precipitation release (PWAUT), and hence increase cloud liquid water amounts. It should be noted that only the indirect forcing associated with clouds in liquid phase is considered. Furthermore, since the cloud water scheme of Rasch and Kristjánsson [1998] only treats stratiform clouds and detraining convective clouds, the calculation of indirect forcing is restricted to these clouds only.

[7] In the present study, the aerosol forcing modules are allowed to interact with the dynamics of the climate system, and it is the response to the aerosol forcing that is the focus of the paper. The coupling of the CCM3 to a slab ocean model has been described in detail by Kiehl et al. [1996]. The purpose of using a slab ocean model is to obtain a realistic thermal inertia for the climate system on multidecadal timescales. On the other hand, potential changes in ocean currents due to changes within the climate system are not taken into account. The open (i.e., ice-free) ocean component of the slab ocean model is taken from Hansen et al. [1983]. It consists of a prognostic equation for the ocean mixed layer temperature, subjected to fluxes to and from the atmosphere (F) and horizontal and vertical heat fluxes within the ocean (Q). The ocean mixed layer depth varies according to climatological seasonally varying observational data by Levitus [1982], but is limited to 200 meters to reduce the spin-up time of the model [Kiehl et al., 1996]. A typical range of mixed layer depths in the model is 20–60 m in the tropics and the high-latitude summer hemisphere, while the typical values in the high-latitude winter hemisphere are near 200 m. For the ice-covered ocean, the mixed layer Q flux below sea ice is specified so as to yield approximately a present-day sea ice distribution from observations. Furthermore, to avoid excessive ice growth in experimental simulations, the Q flux is constrained in a globally conserving manner. Sea ice is divided into four layers of uniform thickness, and in each layer a separate heat transfer equation is solved. In the horizontal the sea ice is assumed to completely cover a CCM grid cell. Sea ice is assumed to form at −1.9°C and to melt at 0°C. Snowfall and consequent variations in snow depth on top of the sea ice are taken into account, but changes in snow depth by compaction over time and by sublimation are ignored. A minimum sea ice thickness of 0.25 m is assumed, to avoid numerical difficulties. In the Arctic the sea ice thickness is not allowed to exceed 3 m, while the maximum ice thickness in the Southern Ocean is 0.50 m.

[8] The simulations that follow are of 50-year duration. There is a spin-up period covering the first 5–10 years, during which the climate gradually changes, especially in the runs with present-day aerosol conditions. After this, the model's climate has reached a new equilibrium (Figure 1), and we consequently use the last 40 years of each 50-year simulation in the analysis. In all the simulations the life cycle scheme for sulfate and black carbon is run interactively with the climate evolution. This means that interactions between chemical processes and changes in climate can take place. The importance of these interactions has been studied in detail and is described in a separate study by T. Iversen et al. (manuscript in preparation, 2005). Table 1 gives a schematic overview of all the simulations, and explains the differences in their setup.

Figure 1.

Variation in the globally averaged surface temperature (°C) in the ALLTOT simulation, illustrating the 10 year spin-up and the subsequent new equilibrium.

Table 1. A Schematic Overview of the Experimental Setupa
 Direct ForcingIndirect ForcingAnthropogenic AerosolsNatural AerosolsGreenhouse Gas Concentrations
  • a

    All the experiments are carried out using an interactive slab ocean, as well as online calculations of aerosol chemistry and transport. In experiments DIRTOT and DIRNAT, the prescribed cloud droplet number concentrations of Rasch and Kristjánsson [1998] are used, while in all the other simulations the cloud droplet number concentrations are calculated from activation of CCN, as in the work by Kristjánsson [2002], yielding an interaction between the aerosols and the cloud properties.

DIRTOTyesnoyesyespresent day
DIRNATyesnonoyespresent day
INDTOTnoyesyesyespresent day
INDNATnoyesnoyespresent day
ALLTOTyesyesyesyespresent day
ALLNATyesyesnoyespresent day

[9] The model results have been subjected to a t-test, in order to evaluate their statistical significance. We do not show the results of this test explicitly, but in general, unless explicitly stated otherwise, all the main features, especially at low latitudes are significant at the 95% level, assuming that all the years are statistically independent. Notable exceptions are some of the signals at high latitudes, e.g., the temperature signal near the ice edge off Antarctica, where the statistical significance is often weak. Investigations of the decorrelation time [von Storch and Zwiers, 1999] (not shown) indicate that in the tropics, the annual averages are a suitable sample to use for evaluating statistical significance, while in other parts of the world, the autocorrelations are weaker, especially over land, meaning that an assumption of independence between monthly averages would be appropriate when applying the t-test.

3. Main Results

3.1. Sulfate and Black Carbon Cycle

[10] In Figures 2a and 2b we show the horizontal distribution of the vertically integrated sulfate and black carbon amounts at years 11–50 in the new simulations. The main features are similar to those previously reported by Iversen and Seland [2002, 2003], but in the tropics there is a southward shift for reasons discussed in the next subsection. The largest column burdens of sulfate are found over SE Asia and North America, with high values also over Europe, Mexico, the Middle East and central Africa. Another noticeable feature is the fact that in general the burdens are much higher over the Northern than the Southern Hemisphere. Compared to other studies, the fraction of total sulfate that is anthropogenic is rather large in our simulations. This, together with little vertical transport contributed to a large indirect forcing of −1.83 W m−2 in the study of Kristjánsson [2002]. This estimate was obtained from 5-year simulations in which SST was prescribed and there was no response from the climate system; that is, both the 1st (cloud droplet radius) and 2nd (cloud lifetime) indirect effect were obtained from parallel calls to the radiation and condensation schemes of the model. The rationale for obtaining the 2nd indirect effect as a forcing term was discussed in detail in Kristjánsson [2002]. The indirect forcing was found as the instantaneous change in cloud radiative forcing (shortwave plus longwave) at the top of the atmosphere.

Figure 2.

Simulated column integral of sulfate (mg S m−2) and black carbon (mg C m−2) in the ALLTOT simulation: (a) sulfate and (b) black carbon.

[11] Similarly, Kirkevåg and Iversen [2002] calculated the direct aerosol forcing as the instantaneous change in net radiative flux (positive downward) at TOA due to sulfate and black carbon aerosols by carrying out 5-year integrations in which the aerosol forcing did not interact with the climate system. They obtained a globally averaged direct forcing of −0.11 W m−2, with values ranging from +1.1 W m−2 over the biomass burning regions, to −1.1 W m−2 in air masses dominated by sulfate at midlatitudes. Figure 2b shows the spatial distribution of vertically integrated BC in the response simulation (ALLTOT). In this case, the largest column burdens are found over China and Europe, but the large values associated with the biomass burning areas in central Africa and tropical South America cause a generally smaller difference between the two hemispheres than in the case of sulfate.

[12] We now investigate the changes in the climate state, some of which are caused directly by the aerosol forcing, while others are caused by feedbacks of the climate system to that forcing. Table 2 summarizes the main features of the climate state in the 7 simulations.

Table 2. Global Averages of Key Quantities in the Experiments Defined in Table 1
 Temperature, °CPrecipitation, mm/dayCloud Cover, %Cloud Liquid Water Path, g m−2Cloud Ice Water Path, g m−2Cloud Droplet Effective Radius, μmNet SW Flux at Surface, W m−2
DIRTOT13.82.8957.2   169.6
DIRNAT13.92.9057.1   170.2

3.2. Changes in Climate

[13] Starting with the temperature response due to the direct effect (Figure 3a), as might be expected from the forcing figures, the signal is weak and most of the geographical features are marginally significant. Although the cooling effect dominates, there are also regions of warming, e.g., over high-latitude continents and over Australia. In particular this can be ascribed to BC aerosols above high-albedo surfaces such as ice and snow, low clouds or deserts. In reality, the warming over Australia would tend to be offset by the simultaneous presence of OC aerosols, which have a cooling effect due to their reflection of solar radiation. The globally averaged cooling effect is −0.08 K, which would correspond to an equilibrium climate sensitivity of 0.73 K per W m−2. In the case of the indirect effect, Figure 3b shows the simulated temperature change. The indirect effect causes a globally averaged cooling of −1.25 K, corresponding to a climate sensitivity of about 0.68 K per W m−2. By comparison Meehl et al. [2000] found an equilibrium sensitivity value of 0.55 K per W m−2 for a doubling of CO2 using the NCAR CCM3 with slab ocean and with the same prognostic cloud water scheme [Rasch and Kristjánsson, 1998] as in the present study. In terms of climate efficacy, i.e., the ratio between climate sensitivity to a given forcing over that to a doubling of CO2 [Hansen et al., 2005], this would correspond to a value of 1.24 for indirect forcing and 1.33 for direct forcing. The higher value for sensitivity obtained from the direct effect probably reflects the influence of absorbing aerosols, for which the concept of climate sensitivity is problematic [Hansen et al., 1997]. For the two forcings combined (direct plus indirect), we find a somewhat larger cooling of −1.42 K (Table 3) than by adding the previous two figures together (−1.33 K). Figure 3c shows the vertical distribution of the temperature changes in a cross section. Even though most of the aerosols are found in the lower troposphere, the temperature response has little vertical variation through the troposphere.

Figure 3.

Simulated temperature changes (K) due to aerosol direct and indirect forcing in simulations: (a) DIRTOT minus DIRNAT, (b) INDTOT minus INDNAT, and (c) ALLTOT minus ALLNAT, zonally averaged. Note the different temperature scale in Figure 3a, compared to Figures 3b and 3c.

Table 3. Global Averages of Changes in Key Quantities Due to Different Aerosol Forcings and Associated Climate Feedbacks
 Temperature, KPrecipitation, mm/dayCloud Cover, %LWP, g m−2IWP, g m−2Net SW Flux at Surface, W m−2
DIRECT plus INDIRECT−1.42−0.13−0.001−0.15+0.06−1.82
RESIDUAL−0.10−0.013 (−0.5%)−0.01+0.11+0.06−0.07

[14] Also the horizontal distribution of the temperature response is quite different from the distribution of the forcing pattern. The negative direct plus indirect forcing is most pronounced in the low to midlatitudes of the Northern Hemisphere and is quite weak in the Arctic [Kristjánsson, 2002; Kirkevåg and Iversen, 2002], but the cooling effect is largest in the Arctic (Figures 3a and 3b). This is a consequence of climate feedbacks, in particular associated with the influence of ice and snow on surface albedo, but is also partly due to the influence of ice cover on ocean-atmosphere heat fluxes. In addition to the ice albedo feedback, there is a cloud feedback that influences the Arctic signal. This will be studied in more detail in the next section. The high-latitude amplification is similar, apart from the sign, to what has been found in simulations of global warming due to increased greenhouse gas concentrations [Houghton et al., 2001]. We note that this result is similar to those obtained in the Hadley Centre GCM [Williams et al., 2001] and the Australian CSIRO model [Rotstayn and Lohmann, 2002].

[15] A set of sensitivity experiments in which the ice albedo feedback was artificially suppressed by setting the albedo over sea ice equal to that over ocean, gave a different geographical distribution of the cooling effect, with enhanced cooling in regions of large negative radiative forcing and reduced cooling over the Arctic (not shown). Nevertheless, the globally averaged cooling was almost as large as in the simulations with the ice albedo feedback included. Hence, with the current experimental setup the Arctic ice albedo feedback does not influence the global climate sensitivity much, but mainly causes a geographical redistribution of the regional response patterns.

[16] In order to check the realism in the temperature signals in Figures 3a–3c, a simulation was carried out in which emissions and greenhouse gases were set to preindustrial values (PREIND), while other features of this simulation are identical to the simulation INDNAT (see Table 1). The globally averaged near-surface temperature difference between the INDTOT run and PREIND is +0.26 K, with large positive values mainly in the Southern Hemisphere and in the Arctic (not shown). At midlatitudes in the Northern Hemisphere, there is a cooling, and the same applies to the region underneath the stratus cloud deck off the coast of Peru. This indicates that the cooling signal of aerosols over land may be exaggerated in our model. This problem is being dealt with by introducing various major improvements to the aerosol schemes that will be reported elsewhere. Another feature of the PREIND simulation worth mentioning is that globally averaged precipitation is slightly higher than in the corresponding run with today's greenhouse gases (INDTOT in Table 3). This means that the suppression of the hydrological cycle due to the indirect effect of aerosols is larger than the enhancement of it due to greenhouse gas warming of the oceans, as was also found by Feichter et al. [2004]. From Table 2 we see that the net solar flux reaching the surface is higher in the PREIND run than in any of the other simulations. Conversely it is lower in the runs with direct plus indirect forcing than in the simulations where only one of these effects is treated. This illustrates the role of aerosols for suppressing solar heating of the surface which in turn drives the hydrological cycle being responsible for precipitation.

[17] The so-called semidirect effect [e.g., Ackerman et al., 2000] measures the modulation of clouds due to absorbing aerosols. It appears mainly because although absorbing aerosols warm the atmosphere, they cool the surface, hence suppressing evaporation [e.g., Ramanathan et al., 2001]. As mentioned above, Kirkevåg and Iversen found a globally averaged direct radiative forcing of −0.11 W m−2 in their simulations, but at the surface the global average was −0.60 W m−2, and this may cause a suppression of the hydrological cycle. In our simulations the semidirect effect is manifested through a reduction in cloud water path in the simulations with direct aerosol forcing (top row of Table 3). However, as in the study of Lohmann and Feichter [2001], the reduction is quite small.

[18] As the climate system responds to the aerosol forcing, there are significant changes in sea level pressure, especially due to the indirect effect. The main change is a 1–2 hPa increase in sea level pressure over the Arctic (Figure 4a), most pronounced in the autumn and early winter (Figure 4b). At the surface, the cooling in this region is largest in the winter, even though the radiative forcing is strongest in the summer [Kristjánsson, 2002]. What happens is that the reduction in summer insolation leads to increased sea ice extent, hence reducing the heat flux from the ocean to the atmosphere. As a result, a shallow anticyclone tends to form, blocking the intrusion of low-pressure systems that would normally have brought warm air masses toward the Arctic in the fall and early winter. A similar result, but with the opposite sign, has recently been found in observations of sea level pressure in the Arctic [Zhang et al., 2004], probably related to greenhouse gas warming.

Figure 4.

Simulated changes in sea level pressure (hPa) due to aerosol direct and indirect forcing (ALLTOT minus ALLNAT): (a) annual average and (b) SON season.

[19] In Figure 5 we show the influence of aerosol forcing on precipitation. First in Figure 5a the change in precipitation from the direct radiative forcing is shown. Considering that the forcing and its interhemispheric differences are significantly smaller than for the indirect effect, it is interesting to see the southward shift of the ITCZ, even though its statistical significance is marginal. Elsewhere, the precipitation signal from the direct effect is weak and largely insignificant. The southward shift is caused by interhemispheric differences in radiative forcing, the average value in the Northern Hemisphere being −0.19 W m−2, as compared to −0.04 W m−2 in the Southern Hemisphere [Kirkevåg and Iversen, 2002]. By comparison, Chung and Seinfeld [2005] found a northward shift of the ITCZ due to direct forcing, which can be understood by the fact that the direct forcing in their simulations was positive, since only BC aerosols were considered. The warming effect of BC was largest in the Northern Hemisphere, and this led to the northward shift of the ITCZ.

Figure 5.

Simulated changes in precipitation (%), due to aerosol forcing from (a) simulations with direct forcing (DIRTOT minus DIRNAT), (b) simulations with indirect forcing (INDTOT minus INDNAT), and (c) simulations with direct plus indirect forcing (ALLTOT minus ALLNAT).

[20] The precipitation response due to indirect forcing (Figure 5b) gives a strong and coherent signal, its main features being a southward displacement of the ITCZ by a few degrees of latitude and some regional changes, e.g., a drying over the Sahel region in Africa. The southward shift is expected, since Kristjánsson [2002] found the indirect forcing in the Northern Hemisphere to be about −2.61 W m−2, as compared to −1.06 W m−2 in the Southern Hemisphere. Furthermore, the result is consistent with other studies referred to in the introduction. There is also a general reduction in precipitation (Tables 2 and 3), mainly because of less humidity being present in the atmosphere in the colder climate, but also due the second indirect effect; that is, the more numerous, smaller droplets are less likely to exceed the autoconversion threshold of 10 μm radius (equation (2)). The southward shift of the ITCZ can be verified by viewing the changes in vertical velocity between the simulations with and without anthropogenic aerosols, see Figures 6a and 6b. Note how the rising branch of the Hadley circulation is split in the simulation with aerosol forcing, with a secondary rising branch occurring south of the equator, in addition to the branch near 5°N.

Figure 6.

Zonally averaged vertical velocity, ω, in units of hPa s−1 from (a) simulation with direct plus indirect aerosol forcing (ALLTOT) and (b) simulation without direct plus indirect forcing (ALLNAT).

[21] In Table 3 the aforementioned “residual effect” (see end of section 1) is evaluated by comparing the results of the runs ALLTOT – ALLNAT on the one hand and INDTOT – INDNAT plus DIRTOT – DIRNAT on the other. The difference is not large (compare Figure 5c with Figures 5a and 5b), but we note that the precipitation is reduced by about 0.5%, which is consistent with a further weakening of the hydrological cycle due to absorbing aerosols. The residual effect shows no signal in cloud cover, but this is related to an inherent feature of the host model, largely because of the random overlap assumption. In fact, the global cloud cover is found to be virtually insensitive to almost any model change that we have performed. Also, cloud water path is quite insensitive to the residual effect (Table 3). The residual effect is probably a manifestation of the nonlinearities in the model's response to different forcings, causing the response to the direct and indirect effect to be nonadditive. A thorough investigation of such nonlinear feedbacks has been carried out by Colman et al. [1997], who found that the main contributions to such nonlinearities in their model were associated with high clouds in the tropics.

4. Cloud Feedback

4.1. Main Features

[22] We here investigate the role played by changes in cloud radiative forcing in response to the climate changes that take place as the aerosols are allowed to cool the climate system. In particular we ask the question of what the sign and horizontal distribution of the cloud feedback is. First in Figure 7 we show the time evolution of globally averaged cloud radiative forcing estimated from the INDTOT and INDNAT simulations. Keeping in mind that the imposed indirect forcing is −1.83 W m−2 and that practically all of this is coming from the shortwave part of the spectrum [Kristjánsson, 2002], it is interesting to see how this quantity changes with time during the simulation (Figure 7). During the transition time of the first 10 years, the shortwave component of the indirect forcing is gradually reduced (in absolute value) to almost half the original value, but this is partly compensated by a negative longwave forcing that arises. Both these changes can be attributed to a slight thinning of the clouds in a colder climate, but the influence of that thinning is larger in the short wave than in the long wave because of a larger degree of saturation for cloud emissivity than for cloud albedo. The net effect is a radiative perturbation of about −1.55 W m−2, i.e., some 15% weaker than at the outset, representing a significant negative cloud feedback. Another interesting feature of Figure 7 is the large interannual variability in cloud forcing, meaning that fairly long simulations are needed to evaluate this quantity accurately.

Figure 7.

Time evolution of the globally averaged difference in cloud radiative forcing between runs with (INDTOT) and without (INDNAT) anthropogenic aerosols. Short wave is red, long wave is blue, and net is green. Unit is W m−2.

[23] In order to understand the geographical distribution of these changes in cloud radiative forcing, we now investigate changes in cloud water path, both liquid and ice. First we note by comparing the results from INDTOT and INDNAT in Table 3 that liquid water path (LWP), which in simulations without climate response was increased by 5% because of the 2nd indirect effect [Kristjánsson, 2002], is now virtually unchanged in a global average. What causes the thinning of the clouds, and hence offsets the 2nd indirect effect, is the reduction of atmospheric moisture in the colder climate. Consistently with this assertion, we note that the lowest total cloud water path (LWP plus IWP) values of all the runs are found in the PREIND simulation (Table 2), which has the coldest climate. Figure 8a shows the geographical distribution of the changes in LWP from the combination of indirect forcing and climate response. As in the work by Kristjánsson [2002], an increase is found in a zone of large sulfate amounts stretching from North America toward Europe, as well as over China, southern Africa and parts of S America. The main novelty here is that there are also large areas with decreased LWP amounts, especially over ocean regions poleward of 55° latitude in both hemispheres, but also over parts of the subtropical oceans. The latter areas show up with an opposite sign in the signature for ice water path (IWP, Figure 8b), since these are high clouds that are now cold enough to consist of ice. No such compensation is found in the Arctic where IWP is reduced at the same time as LWP is reduced. Hence the thinning of clouds in this region must be related to climate feedbacks. It is of interest to find out what these feedbacks are, as well as what sign they have, and we will deal with this in the next subsection. Figure 8b also shows signatures of the southward displacement of the ITCZ and its associated anvil (ice) clouds, as well as a general increase in ice water path over the Northern Hemisphere storm tracks, partly caused by the general cooling of the atmosphere. The lack of a signature for the southward shift of the ITCZ in LWP (Figure 8a) is due to the fact that only the detraining, upper part of the convective clouds influences the prognostic cloud water [Rasch and Kristjánsson, 1998], as mentioned in section 2.

Figure 8.

Simulated changes in cloud water path due to aerosol indirect forcing (INDTOT minus INDNAT): (a) cloud liquid water path (g m−2) and (b) cloud ice water path (g m−2).

[24] Figure 9 shows the reduction in effective radius between the simulations INDTOT and INDNAT, which is a measure of the 1st indirect effect. As opposed to the 2nd indirect effect (Figure 8a) this result is quite similar to that of the pure forcing simulations of Kristjánsson [2002], including the global averages (Tables 2 and 3). Exceptions are slightly enhanced cloud droplet radii in some tropical ocean areas, the Arctic and to a lesser extent near 60°S. The larger droplets in the tropical areas are mainly due to a combination of reduced precipitation caused by the southward shift of the ITCZ (Figure 5b) and larger LWP (Figure 8a), hence leading to an increase in cloud droplet size through equation (1). At high latitudes, water clouds occur less frequently in a cooler climate. This will tend to reduce the in-cloud production of sulfate, but because of very slow clear air oxidation rates at high latitudes combined with reduced scavenging in the colder climate, the fraction of sulfate produced in clouds may increase, leading to fewer new CCN and hence larger droplets. Such effects will be further investigated in the upcoming paper by T. Iversen et al. (manuscript in preparation, 2005).

Figure 9.

Difference in cloud droplet effective radius, as seen by satellite, between the simulations INDTOT and INDNAT (μm).

4.2. Clouds in the Arctic

[25] The simultaneous reduction of LWP and IWP in the Arctic that we found in Figures 8a and 8b can be partly understood by looking at changes in cloud cover. First, in Figure 10a we see that effective cloud cover, given as the product of cloud cover and cloud emissivity is reduced in the polar regions. This reduction is related to the changes in sea level pressure that were shown in Figure 4a, indicating a general buildup of high pressure over the polar regions in the colder climate, hence suppressing the supply of moisture to these regions by extratropical cyclone activity. At the same time effective cloud amount is increasing over the midlatitude storm track regions. This is partly due to the 2nd indirect effect discussed in connection with Figure 8a, but since the signal is very strong also at upper levels where only ice clouds are present, and hence the indirect effect is not calculated, this cannot be the only explanation. The geographical features of Figure 10b, showing changes in high clouds due to aerosol indirect forcing, suggest that the midlatitude storm tracks are shifted in position, as also evidenced by the change in rainfall patterns in Figure 5b. The tropical signature in Figures 10a and 10b, on the other hand, is once again a consequence of the southward displacement of the ITCZ, discussed in the previous section.

Figure 10.

Simulated changes in cloud cover due to aerosol indirect forcing (INDTOT minus INDNAT): (a) zonally averaged effective cloud cover (%) and (b) high cloud cover (%).

[26] In Table 4, we display the changes due to aerosol forcing in the same basic variables as in Table 3, but now only for the area north of 60°N. We note how the signals of Table 3 are generally enhanced, including the residual effect, which gives a precipitation reduction of 2.5%. Even though the Arctic has significant BC concentrations (Figure 2b), it seems extremely unlikely that the semidirect effect as such is acting in this environment of relatively high static stability. Rather, this suppression of precipitation, which comes in addition to the suppression by the indirect forcing alone, must be related to a reduced hydrological cycle. To demonstrate this, we have computed the vertically integrated water vapor amounts, and found that the globally averaged suppression in the case of the direct, indirect and direct plus indirect simulations has values of −4.5 g m−2, −91 g m−2 and −104 g m−2, respectively, confirming the role of moisture suppression for explaining the reduction in precipitation.

Table 4. As for Table 3, Except Only for the Area North of 60°N
 Temperature, KPrecipitation, mm/dayCloud Cover, %LWP, g m−2IWP, g m−2Net SW Flux at Surface, W m−2
DIRECT plus INDIRECT−2.77−0.19−0.038−10.04−1.72−0.26
RESIDUAL−0.13−0.03 (−2.5%)+0.014−0.25−0.29+0.14

[27] Field experiments in the Arctic have shown this to be a unique area in the sense that clouds have an overall warming effect, at least at the surface [Intrieri et al., 2002]. This is due to a combination of a low sun angle and a high surface albedo, making the shortwave cloud forcing small, while the longwave cloud forcing of the surface is large because of persistent rather dense clouds throughout much of the year. Consistently with this, in our simulations the net cloud forcing at the surface poleward of 60°N is about +20 W m−2, as compared to about −10 W m−2 at the top of the atmosphere. However, despite this dominance of the longwave over shortwave forcing at the surface, the change in net cloud forcing from INDNAT to INDTOT is dominated by the short wave and has the value +1.5 W m−2; that is, there is a negative cloud feedback in this region. This is in contrast to a recent study by Vavrus [2004], who found the positive cloud feedback to contribute 40% to the Arctic warming resulting from a doubling of CO2 in simulations with the GENESIS2 model. We believe the negative Arctic cloud feedback in our simulations may be an artifact of the oversimplified treatment of ice cloud radiative properties, these clouds having a too large negative net cloud forcing, as well as to biases in Arctic cloud fraction in CCM3, discussed by Rasch and Kristjánsson [1998].

5. Summary and Conclusions

[28] Results from multidecadal equilibrium simulations of the response to indirect and direct aerosol forcing have been presented. The simulations were carried out using an atmospheric GCM (CCM-Oslo) coupled to a slab ocean. The life cycles of sulfate and black carbon aerosols were run interactively enabling the aerosol fields to change according to changes in circulation patterns resulting from climate response.

[29] The main features in the response in temperature and precipitation to indirect forcing are similar to those previously reported by Rotstayn and Lohmann [2002] and Williams et al. [2001], e.g., a southward displacement of the ITCZ, and a maximum cooling in the Arctic, despite the forcing being most negative at midlatitudes of the Northern Hemisphere. The direct forcing causes a much weaker Northern Hemisphere cooling than the indirect forcing does. Nevertheless, a distinct, but marginally statistically significant southward shift of the ITCZ is found. In simulations with direct and indirect effect together, the direct and indirect effects do not simply add up. Rather a nonnegligible “residual effect” occurs, characterized by a weakening of the hydrological cycle presumably due to nonlinear interactions between absorbing aerosols and the indirect forcing. All the above signals are enhanced in the Arctic. This is partly due to a chain reaction between cooling from ice albedo feedback, via higher surface pressure, to a general reduction in cloudiness at all levels of the troposphere. In the midlatitude storm tracks, however, clouds are thicker when the (mainly indirect) aerosol forcing is present because of a combination of the 2nd indirect effect and a reduced insurgence of moisture into the polar regions. In a global average the 2nd indirect effect is swamped by the associated climate change, since the colder climate carries less moisture, leading to thinner clouds. Conversely, the 1st indirect effect appears in a similar manner as in previous simulations that did not allow climate response to aerosol forcing.

[30] A simulation using preindustrial aerosol precursor emissions and preindustrial greenhouse gas concentrations gave a globally averaged temperature which was 0.26 K lower than in simulations with today's emissions. Cloud water paths were smaller than in the present-day simulations, because of less atmospheric moisture in the colder climate, but, interestingly, precipitation amounts were enhanced. This means that in the present-day climate simulations relative to simulations of preindustrial climate the suppression of precipitation by aerosols is stronger than the enhancement of precipitation that inevitably occurs in a warmer climate because of higher CO2 concentrations.


[31] This study was supported by the Norwegian Research Council through the RegClim project. Furthermore, the work has received support of the Norwegian Research Council's Programme for Supercomputing through a grant of computer time. Many thanks to the reviewers whose insightful and constructive comments led to significant improvements of the paper.