2.4.3 Advances in Modelling the Aerosol Direct Effect

Since the TAR, more complete aerosol modules in a larger number of global atmospheric models now provide estimates of the direct RF. Several models have resolutions better than 2° by 2° in the horizontal and more than 20 to 30 vertical levels; this represents a considerable enhancement over the models used in the TAR. Such models now include the most important anthropogenic and natural species. Tables 2.4, 2.5 and 2.6 summarise studies published since the TAR. Some of the more complex models now account explicitly for the dynamics of the aerosol size distribution throughout the aerosol atmospheric lifetime and also parametrize the internal/external mixing of the various aerosol components in a more physically realistic way than in the TAR (e.g., Adams and Seinfeld, 2002; Easter et al., 2004; Stier et al., 2005). Because the most important aerosol species are now included, a comparison of key model output parameters, such as the total τ_aer, against satellite retrievals and surface-based sun photometer and lidar observations is possible (see Sections 2.4.2 and 2.4.4). Progress with respect to modelling the indirect effects due to aerosol-cloud interactions is detailed in Section 2.4.5 and Section 7.5. Several studies have explored the sensitivity of aerosol direct RF to current parametrization uncertainties. These are assessed in the following sections.

Major progress since the TAR has been made in the documentation of the diversity of current aerosol model simulations. Sixteen groups have participated in the Global Aerosol Model Intercomparison (AeroCom) initiative (Kinne et al., 2006). Extensive model outputs are available via a dedicated website (Schulz et al., 2004). Three model experiments (named A, B, and PRE) were analysed. Experiment A models simulate the years 1996, 1997, 2000 and 2001, or a five-year mean encompassing these years. The model emissions and parametrizations are those determined by each research group, but the models are driven by observed meteorological fields to allow detailed comparisons with observations, including those from MODIS, MISR and the AERONET sun photometer network. Experiment B models use prescribed AeroCom aerosol emissions for the year 2000, and experiment PRE models use prescribed aerosol emissions for the year 1750 (Dentener et al., 2006; Schulz et al., 2006). The model diagnostics included information on emission and deposition fluxes, vertical distribution and sizes, thus enabling a better understanding of the differences in lifetimes of the various aerosol components in the models.

This paragraph discusses AeroCom results from Textor et al. (2006). The model comparison study found a wide range in several of the diagnostic parameters; these, in turn, indicate which aerosol parametrizations are poorly constrained and/or understood. For example, coarse aerosol fractions are responsible for a large range in the natural aerosol emission fluxes (dust: ±49% and sea salt: ±200%, where uncertainty is 1 standard deviation of inter-model range), and consequently in the dry deposition fluxes. The complex dependence of the source strength on wind speed adds to the problem of computing natural aerosol emissions. Dust emissions for the same time period can vary by a factor of two or more depending on details of the dust parametrization (Luo et al., 2003; Timmreck and Schulz, 2004; Balkanski et al., 2004; Zender, 2004), and even depend on the reanalysis meteorological data set used (Luo et al., 2003). With respect to anthropogenic and natural emissions of other aerosol components, modelling groups tended to make use of similar best guess information, for example, recently revised emissions information available via the Global Emissions Inventory Activity (GEIA). The vertical aerosol distribution was shown to vary considerably, which is a consequence of important differences in removal and vertical mixing parametrizations. The inter-model range for the fraction of sulphate mass below 2.5 km to that of total sulphate is 45 ± 23%. Since humidification takes place mainly in the boundary layer, this source of inter-model variability increases the range of modelled direct RF. Additionally, differences in the parametrization of the wet deposition/vertical mixing process become more pronounced above 5 km altitude. Some models have a tendency to accumulate insoluble aerosol mass (dust and carbonaceous aerosols) at higher altitudes, while others have much more efficient wet removal schemes. Tropospheric residence times, defined here as the ratio of burden over sinks established for an equilibrated one-year simulation, vary by 20 to 30% for the fine-mode aerosol species. These variations are of interest, since they express the linearity of modelled emissions to aerosol burden and eventually to RF.

Considerable progress has been made in the systematic evaluation of global model results (see references in Tables 2.4 to 2.6). The simulated global τ_aer at a wavelength of 0.55 µm in models ranges from 0.11 to 0.14. The values compare favourably to those obtained by remote sensing from the ground (AERONET, about 0.135) and space (satellite composite, about 0.15) (Kinne et al., 2003, 2006), but significant differences exist in regional and temporal distributions. Modelled absorption optical thickness has been suggested to be underestimated by a factor of two to four when compared to observations (Sato et al., 2003) and DRE efficiencies have been shown to be lower in models both for the global average and regionally (Yu et al., 2006) (see Section 2.4.4.7). A merging of modelled and observed fields of aerosol parameters through assimilation methods of different degrees of complexity has also been performed since the TAR (e.g., Yu et al., 2003; Chung et al., 2005). Model results are constrained to obtain present-day aerosol fields consistent with observations. Collins et al. (2001) showed that assimilation of satellite-derived fields of τ_aer can reduce the model bias down to 10% with respect to daily mean τ_aer measured with a sun photometer at the Indian Ocean Experiment (INDOEX) station Kaashidhoo. Liu et al. (2005) demonstrated similar efficient reduction of errors in τ_aer. The magnitude of the global dust cycle has been suggested to range between 1,500 and 2,600 Tg yr^–1 by minimising the bias between model and multiple dust observations (Cakmur et al., 2006). Bates et al. (2006) focused on three regions downwind of major urban/population centres and performed radiative transfer calculations constrained by intensive and extensive observational parameters to derive 24-hour average clear-sky DRE of –3.3 ± 0.47, –14 ± 2.6 and –6.4 ± 2.1 W m^–2 for the north Indian Ocean, the northwest Pacific and the northwest Atlantic, respectively. By constraining aerosol models with these observations, the uncertainty associated with the DRE was reduced by approximately a factor of two.