126.96.36.199 Uncertainties Due to Model Biases
One of the large sources of uncertainties is the poor knowledge of the amount and distribution of anthropogenic aerosols used in the model simulations, particularly for pre-industrial conditions. Some studies show a large sensitivity in the RF to the ratio of pre-industrial to present-day aerosol number concentrations.
All climate models discussed above include sulphate particles; some models produce them from gaseous precursors over oceans, where ambient concentrations are low, while some models only condense mass onto pre-existing particles over the continents. Some other climate models also include sea salt and dust particles produced naturally, typically relating particle production to wind speed. Some models include anthropogenic nitrate, BC and organic compounds, which in turn affect activation. Models also have weaknesses in representing convection processes and aerosol distributions, and simulating updraft velocities and convection-cloud interactions. Even without considering the existing biases in the model-generated clouds, differences in the aerosol chemical composition and the subsequent treatment of activation lead to uncertainties that are difficult to quantify and assess. The presence of organic carbon, owing to its distinct hygroscopic and absorption properties, can be particularly important for the cloud albedo effect in the tropics (Ming et al., 2007).
Modelling the cloud albedo effect from first principles has proven difficult because the representation of aerosol-cloud and convection-cloud interactions in climate models are still crude (Lohmann and Feichter, 2005). Clouds often do not cover a complete grid box and are inhomogeneous in terms of droplet concentration, effective radii and LWP, which introduces added complications in the microphysical and radiative transfer calculations. Model intercomparisons (e.g., Lohmann et al., 2001; Menon et al., 2003) suggest that the predicted cloud distributions vary significantly between models, particularly their horizontal and vertical extents; also, the vertical resolution and parametrization of convective and stratiform clouds are quite different between models (Chen and Penner, 2005). Even high-resolution models have difficulty in accurately estimating the amount of cloud liquid and ice water content in a grid box.
It has proven difficult to compare directly the results from the different models, as uncertainties are not well identified and quantified. All models could be suffering from similar biases, and modelling studies do not often quote the statistical significance of the RF estimates that are presented. Ming et al. (2005b) demonstrated that it is only in the mid-latitude NH that their model yields a RF result at the 95% confidence level when compared to the unforced model variability. There are also large differences in the way that the different models treat the appearance and evolution of aerosol particles and the subsequent cloud droplet formation. Differences in the horizontal and vertical resolution introduce uncertainties in their ability to accurately represent the shallow warm cloud layers over the oceans that are most susceptible to the changes due to anthropogenic aerosol particles. A more fundamental problem is that GCMs do not resolve the small scales (order of hundreds of metres) at which aerosol-cloud interactions occur. Chemical composition and size distribution spectrum are also likely insufficiently understood at a microphysical level, although some modelling studies suggest that the albedo effect is more sensitive to the size than to aerosol composition (Feingold, 2003; Ervens et al., 2005; Dusek et al., 2006). Observations indicate that aerosol particles in nature tend to be composed of several compounds and can be internally or externally mixed. The actual conditions are difficult to simulate and possibly lead to differences among climate models. The calculation of the cloud albedo effect is sensitive to the details of particle chemical composition (activation) and state of the mixture (external or internal). The relationship between ambient aerosol particle concentrations and resulting cloud droplet size distribution is important during the activation process; this is a critical parametrization element in the climate models. It is treated in different ways in different models, ranging from simple empirical functions (Menon et al., 2002a) to more complex physical parametrizations that also tend to be more computationally costly (Abdul-Razzak and Ghan, 2002; Nenes and Seinfeld, 2003; Ming et al., 2006). Finally, comparisons with observations have not yet risen to the same degree of verification as, for example, those for the direct RF estimates; this is not merely due to model limitations, since the observational basis also has not yet reached a sound footing.
Further uncertainties may be due to changes in the droplet spectral shape, typically considered invariant in climate models under clean and polluted conditions, but which can be substantially different in typical atmospheric conditions (e.g., Feingold et al., 1997; Ackerman et al., 2000b; Erlick et al., 2001; Liu and Daum, 2002). Liu and Daum (2002) estimated that a 15% increase in the width of the size distribution can lead to a reduction of between 10 and 80% in the estimated RF of the cloud albedo indirect effect. Peng and Lohmann (2003), Rotstayn and Liu (2003) and Chen and Penner (2005) studied the sensitivity of their estimates to this dispersion effect. These studies confirm that their estimates of the cloud albedo RF, without taking the droplet spectra change into account, are overestimated by about 15 to 35%.
The effects of aerosol particles on heterogeneous ice formation are currently insufficiently understood and present another level of challenge for both observations and modelling. Ice crystal concentrations cannot be easily measured with present in situ instrumentation because of the difficulty of detecting small particles (Hirst et al., 2001) and frequent shattering of ice particles on impact with the probes (Korolev and Isaac, 2005). Current GCMs do not have sufficiently rigorous microphysics or sub-grid scale processes to accurately predict cirrus clouds or super-cooled clouds explicitly. Ice particles in clouds are often represented by simple shapes (e.g., spheres), even though it is well known that few ice crystals are like that in reality. The radiative properties of ice particles in GCMs often do not effectively simulate the irregular shapes that are normally found, nor do they simulate the inclusions of crustal material or soot in the crystals.