2.1. Atmospheric Model
 The atmospheric model used in this study is a low-resolution (spectral R21) version of the Mk3 CSIRO atmospheric GCM. The R21 model has 18 hybrid vertical levels and a horizontal resolution of approximately 5.6° in longitude and 3.2° in latitude. A standard high-resolution version of the Mk3 model has been described in detail by Gordon et al. . However, the standard version did not include an interactive aerosol scheme and used an older radiation scheme that did not treat aerosol scattering or absorption. Since then, a comprehensive treatment of the tropospheric sulfur cycle was included in the CSIRO GCM [Rotstayn and Lohmann, 2002b] and treatments of other aerosol components, based on established schemes from other models, were subsequently added. Here, we summarize the aerosol schemes, the treatment of aerosol-cloud interactions, the features of an updated radiation code, and the treatment of aerosol optical properties in the new (Mk3A) model.
 Transport of aerosols and other trace quantities occurs by advection, vertical turbulent mixing, and vertical transport inside deep convective clouds [Rotstayn and Lohmann, 2002b]. Vertical advection is handled using a flux-corrected transport scheme [Van Leer, 1977], and horizontal advection is handled via a semi-Lagrangian scheme [McGregor, 1993]. The treatment of vertical turbulent mixing is based on stability-dependent K-theory [Louis, 1979]. Under convective conditions, an additional nonlocal counter-gradient flux is added [Holtslag and Boville, 1993]. Convective transport is based on the vertical profiles of the updraft mass flux and compensating subsidence generated by the convection scheme [Gregory and Rowntree, 1990].
 Prognostic variables in the sulfur-cycle model are dimethyl sulfide (DMS), sulfur dioxide (SO2), and sulfate. The treatment of the sulfur chemistry is based on that in ECHAM4 [Feichter et al., 1996]. The carbonaceous aerosol module [Cooke et al., 1999] assumes an e-folding time of 1.15 days for the conversion of black carbon (BC) and particulate organic matter (POM) from their hydrophobic to hydrophilic forms. The treatment of mineral dust emission [Ginoux et al., 2004] is based on satellite analyses that identified major dust sources as topographic depressions in which a sufficiently deep layer of alluvium was able to accumulate [Prospero et al., 2002]. Four size bins are used for prognostic dust, with radii ranging from 0.1 to 1, 1 to 2, 2 to 3, and 3 to 6 μm, respectively. Two modes of sea salt aerosol (film-drop and jet-drop) are diagnosed at each time step as a function of 10-m wind speed above the ocean surface [O'Dowd et al., 1997], but they are not prognostic variables, in the sense that they are not transported by the model. Sea salt aerosol is assumed to be well mixed in the marine boundary layer and is set to zero above the top of the boundary layer. This simple approach is similar to that used in the Met Office's Unified Model [Jones et al., 2001].
 Large-scale wet scavenging processes are linked to the warm rain and frozen precipitation processes in the stratiform cloud microphysical scheme [Rotstayn, 1997; Rotstayn et al., 2000a] and the convection scheme [Gregory and Rowntree, 1990]. Below-cloud scavenging is assumed proportional to the area swept out by precipitation, based on the negative-exponential raindrop or snowflake size distribution, with a constant of proportionality (collection efficiency) as defined in Table 1. In-cloud scavenging is proportional to the amount of precipitation removed, divided by the liquid water (or ice water) content, with a constant of proportionality (scavenging efficiency) as defined in Table 1. Many of the numbers in Table 1 are highly uncertain, especially where frozen precipitation is concerned. Also included in the scheme is reevaporation of aerosol due to evaporation of rain (or snow), as described previously [Rotstayn and Lohmann, 2002b]. These treatments apply to sulfate, POM, BC, and dust (but not sea salt, since it is not a prognostic variable).
Table 1. Efficiencies Assumed for In-Cloud and Below-Cloud Scavenging of Aerosol Mass by Liquid and Frozen Precipitation
| ||Liquid In-Cloud||Liquid Below-Cloud||Frozen In-Cloud||Frozen Below-Cloud|
|Dust 0.1–1 μm||0.1||0.05||0.1||0.01|
|Dust 1–2 μm||0.1||0.1||0.1||0.02|
|Dust 2–3 μm||0.1||0.2||0.1||0.04|
|Dust 3–6 μm||0.1||0.5||0.1||0.1|
 The shortwave radiation scheme is a two-stream code with 12 bands [Grant and Grossman, 1998; Grant et al., 1999]. The aerosol species treated are tropospheric sulfate, BC, POM, dust, sea salt, and stratospheric aerosol from volcanic eruptions. Except for carbonaceous aerosol (which is assumed to be an internal mixture of BC and POM), all the aerosol components are treated as external mixtures. The optical properties of sulfate, sea salt, and hydrophilic POM and BC account for hygroscopic aerosol growth, based on Kohler theory. Details of the assumed size distributions, optical properties and hygroscopic growth of tropospheric aerosol are given in Table 2. Owing to the computational expense of the calculation of the dependence of aerosol single-scattering properties on relative humidity (and BC fraction for carbonaceous aerosol) in the aerosol optical property routine [Grant et al., 1999], we implemented these via lookup tables. Mie calculations were performed to generate tables of aerosol specific extinction, single-scattering albedo (SSA) and asymmetry parameter in each shortwave band, at 21 relative humidities ranging from 0 to 99% (in increments of 5% from 0 to 95%). For carbonaceous aerosol, these tables were generated for 16 equally spaced BC volume fractions ranging from 0 to 30%. In the GCM, linear interpolation in relative humidity (and BC volume fraction) was used to determine the single-scattering properties of each aerosol species. We enhanced the treatment of cloud-radiative effects in the shortwave scheme to include ice clouds [Warren, 1984; Francis et al., 1994], which were approximated as water clouds in the original shortwave code. Also included is a simple treatment of the effect of BC on snow albedo [Hansen and Nazarenko, 2004]. The longwave scheme [Chou et al., 2001; Chou and Lee, 2005] has 10 bands. Unlike the longwave code in the standard Mk3 model, the scheme treats non-CO2 greenhouse gases (methane, nitrous oxide, and halocarbons) and aerosols. At present, the only aerosol included in the longwave scheme (as well as the shortwave scheme) is stratospheric aerosol from volcanic eruptions, which is assumed to have the properties of ammonium sulfate [Sato et al., 1993].
Table 2. Size Distributions and Radiative Properties of Dry Tropospheric Aerosol
| ||Mode Radius, μm||Geometric Standard Deviation||Density, g cm−3||Refractive Index at 550 nm|
|Sulfatea||0.05||1.9||1.77||1.53 − 1.0 × 10−7i [Toon et al., 1976]|
|POM + BCb||0.08||1.65||1.25 (mixture)||POM: 1.53 − 1.0 × 10−7i [Toon et al., 1976]|
| || || ||1.5 (BC)||BC: 1.80 − 0.50i [Twitty and Weinman, 1971]|
|Small dustc||0.01||1.4||2.4||1.53 − 5.5 × 10−3i [d'Almeida et al., 1991]d|
| ||0.045||1.6||2.4||As above|
|Large dustc||0.275||2.5||2.4||As above|
|Small sea salte||0.035||1.92||2.165||1.50 − 1.0 × 10−8i [Shettle and Fenn, 1979]|
|Large sea salte||0.35||1.7||2.165||As above|
 Cloud droplet number concentrations (cm−3) over oceans (Nocean) and land (Nland) were estimated empirically by Menon et al. [2002a] as
where SO4, OM, and SS are the mass concentrations of sulfate, particulate organic matter, and sea salt, respectively, in μg m−3. We use equation (1) over oceans but reduce the coefficient that multiplies SO4 in equation (2) from 0.50 to 0.26, which was the value obtained from extensive observations in an earlier study [Boucher and Lohmann, 1995]. The physical justification for this “tuning” is the higher level of background aerosol over land, which results in lower supersaturations and a weaker expected dependence of N on anthropogenic sulfate, as seen in the observational data [Boucher and Lohmann, 1995]. Equation (1) was based on data from only two field experiments, both in the eastern North Atlantic, and it is uncertain whether it is valid to extend it to large continental areas, so we reverted to the smaller coefficient from Boucher and Lohmann  over land. Although there is evidence that dust particles can act as efficient cloud condensation nuclei when coated with a layer of sulfate or other soluble material [e.g., Yin et al., 2002], we have not included mineral dust in the parameterization of cloud droplet number concentration because of the large uncertainties and because our model does not yet include interactions among different aerosol species.
 In stratiform clouds, the droplet number concentration determines both the first [Twomey, 1977] and second [Albrecht, 1989] indirect effects. The first indirect effect enters the model via the parameterization of droplet effective radius in the radiation scheme [Rotstayn and Liu, 2003]. The second indirect effect enters via the parameterization of autoconversion (coalescence of cloud droplets) in the cloud microphysical scheme [Rotstayn and Liu, 2005]. The treatments of both these effects account for the observed increase of droplet spectral dispersion with increasing droplet concentration [Liu and Daum, 2002]. The convection scheme only includes very simple microphysics [Gregory and Rowntree, 1990], so the second indirect effect of aerosols on convective clouds is omitted. However, the first indirect effect is included, following the same scheme as used for stratiform clouds.
2.3. Experimental Setup
 Each run covers the period 1871 to 2000 and uses an initial condition taken from a preindustrial control run that had been integrated for several hundred years to reach a state of approximate equilibrium. The initial conditions for the individual runs are separated by 20 years to ensure independence of the runs. Each of the eight runs in the “all forcing” (ALL) ensemble is forced by historical changes in long-lived greenhouse gases [Hansen et al., 2002], ozone [Kiehl et al., 1999], solar variations [Lean and Rind, 1998], volcanic sulfate (updated from Sato et al. ), and anthropogenic emissions of aerosols and aerosol precursors (described below). Changes in land cover are not included. The setup of the eight runs in the “all except aerosols” (AXA) ensemble is identical to that of the ALL runs, except that anthropogenic emissions of aerosols and their precursors are held at 1870 levels throughout each run.
 Regarding aerosols, historical anthropogenic emissions are included for sulfur [Smith et al., 2001, 2004] and carbonaceous aerosols [Ito and Penner, 2005]. As discussed by these authors, such historical estimates are highly uncertain. Ninety seven percent of the sulfur emissions are assumed to occur as sulfur dioxide, and the remaining 3% occur as primary sulfate aerosol. The carbonaceous aerosol emissions include primary sources of BC and POM from the burning of fossil fuel, open vegetation, and biofuel. Emissions of BC and POM from open vegetation and biofuel are assumed to be hydrophilic, while emissions from fossil fuel are assumed to be 50% hydrophilic and 50% hydrophobic. Since secondary sources of POM are not included in the inventory, we multiplied the fossil fuel POM source for each year by a constant factor of 11.2 so that the global emission for 1985 matched that from an earlier inventory for the mid-1980s [Penner et al., 1993; Liousse et al., 1996], which gave reasonable agreement with observations when used as the basis for a major model intercomparison [Penner et al., 2001], as well as another recent modeling study [Liu et al., 2005]. Scaling up the fossil fuel POM emissions is a simple way to allow for secondary sources of POM (see discussion below). Anthropogenic sources of sulfur, POM, and BC for the years 1870 and 2000 in our model are compared in Table 3, assuming a scale factor of 1.3 for conversion of organic carbon to POM. Note that even with the large scale-factor we applied to the fossil fuel POM emissions, our fossil fuel POM emissions for the year 2000 (24.3 Tg C) are within the range used in recent models (7.5 Tg C to 28.1 Tg C, according to Liu et al. , who emphasized the uncertainty of the emissions of carbonaceous aerosols). Also, our total POM emissions for the year 2000 (75.3 Tg C, including natural sources) are close to the average (74.3 Tg C) from the AeroCom models considered by Textor et al. .
Table 3. Global Anthropogenic Emissions of Aerosols and Aerosol Precursors for 1870 and 2000
| || ||1870||2000|
|Sulfur (Tg S)||anthropogenica||3.2||61.7|
|POM (Tg C)||total||14.3||58.9|
| ||biomass burningb||13.7||34.6|
| ||fossil fuel||0.1||2.2|
| ||fossil fuel SOAc||0.5||22.1|
|BC (Tg C)||total||2.1||8.2|
| ||biomass burningb||2.1||5.4|
| ||fossil fuel||0.1||2.8|
 Global anthropogenic emissions of sulfur, BC and POM from 1871 to 2000 are shown in Figure 1a. Each species shows a relatively gradual rise until about 1950, followed by a more rapid increase as industrial development occurred after World War II. Global sulfur emissions peaked in the late 1970s and then started to fall, due to the introduction of emission controls. Global emissions of carbonaceous aerosols continued to increase until the late 1990s. There is also considerable geographical variation in the emission histories for these species [Smith et al., 2001; Ito and Penner, 2005]. For example, since about 1980, there has been a strong shift of sulfur emissions away from Europe and North America and toward Asia. Figure 1b shows the variation with time of emissions from “Asia,” which we defined for the purpose of this study as a rectangular region from the equator to 45°N and from 70°E to 160°E (based on the hypothesis that NH aerosol sources from outside this region are too far away from Australia to substantially affect Australian climate). In contrast to the global emissions of sulfur, Asian emissions of sulfur, POM, and BC all continued to increase at least until the mid-1990s.
Figure 1. Time variation of (a) global and (b) Asian anthropogenic emissions of sulfur, POM, and BC. A factor of 1.3 is used in the model to convert from organic carbon to POM. See text for description of the region defined as “Asia.”
Download figure to PowerPoint
 The model also includes natural sources of sulfur [Rotstayn and Lohmann, 2002b]. These comprise SO2 from noneruptive volcanoes, amounting to 8.0 Tg S yr−1 [Spiro et al., 1992; Graf et al., 1997], and biogenic emissions of DMS from oceans. The oceanic DMS source is calculated using the flux parameterization of Nightingale et al.  and a global database of DMS measurements [Kettle et al., 1999; Kettle and Andreae, 2000]. It amounts to 22.5 Tg S yr−1 in 1870 and 22.7 Tg S yr−1 in 2000. The small increase in the DMS source between 1870 and 2000 is mainly due to an increase in wind speed south of 50°S during the simulation (since temperature-related changes in oceanic DMS concentration are not included in the model). For natural organic carbon from terpenes [Guenther et al., 1995], a yield of 13% is assumed for rapid conversion to POM, giving an annual source of 16.4 Tg C (21.3 Tg of POM using a conversion factor of 1.3).
 With this aerosol treatment, we estimate a change in top-of-atmosphere net downward irradiance of −1.1 W m−2 between 1870 and 1990 due to the combined direct and indirect aerosol effects. This figure was obtained from the difference of two 20-year runs with prescribed climatological sea surface temperatures (SSTs), with aerosol and aerosol-precursor emissions set to 1870 or 1990 levels, respectively, and other forcing factors held constant at late 20th century levels. This method (using the difference of two runs) is the usual approach when the second indirect effect is included, since it is difficult to estimate the second indirect effect without allowing the meteorology to evolve. The method was found to be satisfactory by Rotstayn and Penner , even though feedbacks (such as changes in land-surface temperature) are allowed to occur. They referred to it as a “quasi forcing.” A refinement of the method, designed to account for these feedbacks and hence provide a better predictor of the equilibrium global-mean temperature response to a given radiative perturbation, was suggested by Hansen et al. , namely
where Fo is the “quasi forcing” from above, δTo is the global-mean surface temperature change when the radiative perturbation is introduced but the SSTs are held fixed, and λ is an estimate of the model's equilibrium climate sensitivity parameter (in K per W m−2). Hansen et al.  called Fs the “fixed SST forcing.” From the last 10 years of our two 20-year runs, δTo = −0.072 K, and for the CSIRO GCM λ ≈ 0.8 K per W m−2 [Rotstayn and Penner, 2001]. Thus Fs = −1.2 W m−2 is a more accurate estimate of the net anthropogenic aerosol forcing in our model between 1870 and 1990. Of this, the direct aerosol forcing is −0.39 W m−2 (calculated by making a second call to the shortwave radiation scheme, with aerosols turned off, in both runs) and the remainder (−0.8 W m−2) can be attributed to the indirect aerosol effect.
 Figure 2 shows the time evolution of global-mean near-surface temperature (Ts) changes from the HadCRUT2 observations [Jones and Moberg, 2003; Rayner et al., 2003] and from the model runs. The ALL ensemble clearly gives a better simulation of the observed global-mean Ts changes, since the AXA ensemble overestimates the warming after about 1950. This result (that inclusion of aerosol forcing improves the simulation of global-mean Ts changes) has been seen before in other GCMs [e.g., Mitchell et al. 1995].
Figure 2. Time variation of global-mean Ts (relative to 1871–1900 mean) from the HadCRUT2 observations and from the ALL and AXA ensembles. Solid lines show the ensemble mean, and shading shows the range of individual runs within each ensemble. Each run uses its own 1871–1900 mean in the calculation of temperature change.
Download figure to PowerPoint
2.5. Simulation of Australian Rainfall and Cloudiness
 The global distribution of annual rainfall from an earlier version of the Mk3A CSIRO GCM was shown by Rotstayn and Lohmann [2002a], and it was found that the model was broadly successful at simulating the main features of the observed global climatological rainfall pattern. This is also true of the current version (not shown), but since the present study focuses primarily on Australian rainfall trends, it is important to check whether the model can simulate the regional rainfall patterns over Australia. Figure 5 shows observed and modeled seasonal precipitation for JJA and DJF, averaged over the period 1970–1999. The observations are high-resolution (0.5°) gridded data from the Climatic Research Unit (CRU), known as CRU TS 2.1 [Mitchell and Jones, 2005], and the modeled values are the ALL ensemble mean. In both seasons the model is broadly successful at capturing the spatial pattern of precipitation, although it tends to be too dry over southern Australia, especially over the southwestern corner in DJF. Also, the monsoonal rainfall over the northeastern corner is underestimated in DJF. Overall, this is an encouraging result for a low-resolution GCM.
Figure 5. Observed and modeled Australian seasonal precipitation (in mm) for JJA and DJF, averaged over the period 1970–1999.
Download figure to PowerPoint
 The distribution of global cloudiness in an earlier version of the CSIRO GCM was compared with observations by Rotstayn  and was found to be generally satisfactory. Here, we continue our focus on the Australian region and compare observed and modeled cloudiness for JJA and DJF, averaged over the period 1970–1999 (Figure 6). The observed values are from CRU TS 2.1 and are based on surface observations. In both seasons the model captures the broad pattern of cloudiness over Australia, but there are problems in the detail. In JJA the model somewhat underestimates cloudiness over southwestern Australia and overestimates it over northern Australia. In DJF the modeled cloudiness is fairly realistic over northern Australia but too low over southern Australia. The low bias over southern Australia in DJF is similar to the result for rainfall, suggesting a common underlying dynamical bias. We have found that the SH midlatitude storm tracks in the model tend to lie too far to the south; the effects of this bias are noticeable in the next section, where observed and modeled trends are compared.
Figure 6. Satellite-retrieved and modeled Australian cloudiness (in %) for JJA and DJF, averaged over the period 1984–2000.
Download figure to PowerPoint