Key Aspects of the Simulation of Clouds in Largescale Models

Part of the reason why progress in the representation of clouds in large-scale models has been so slow is that major aspects of the simulated cloud distribution could not be assessed observationally. For instance, the vertical structure of cloud layers, their overlap, the cloud water content, and the cloud water phase are known to play a key role in the radiation budget at the top of the atmosphere (TOA), at the surface, and in the troposphere. Given the lack of reliable and global observations of these quantities, a good agreement between models and observations of TOA radiative fluxes or of the total cloud cover could be obtained with compensating errors. However, differences in the way the radiative balance is achieved can affect the sensitivity of radiative fluxes to a change in climate.

The vertical structure of clouds constitutes one well-known example of a key factor critical for climate studies for which observations have long been lacking. Another example is the cloud phase. Many papers have drawn attention to the vastly different amounts of cloud ice held in various climate models, all of which satisfy the TOA radiation constraint. In large-scale models, the cloud water phase is still commonly diagnosed as a simple function of temperature. As climate warms, the fraction of cloud water may increase at the expense of cloud ice, and owing to differences in the microphysical and radiative properties of liquid and ice clouds, the change in cloud phase contributes to cloud feedbacks. Uncertainties in the diagnostic of the cloud water phase in the current climate thus translate into uncertainties in cloud feedbacks (e.g., Tsushima et al. 2006).

The simplicity of the cloud phase diagnostic in large-scale models has long been justified by the fact that the factors influencing glaciation processes and the presence of supercooled droplets are still poorly understood and loosely constrained by observations. It is thus essential to use new data from satellite or ground-based measurements to provide a better explanation of the factors that influence the cloud phase, and to develop more reliable parameterizations for large-scale models.

In addition to the lack of key measurements, the representation of clouds in large-scale models is complicated by our poor understanding of the large-scale controls of (measured) cloud properties and of the physical processes through which the different cloud properties interact with each other (cf. Bretherton and Hartmann; Grabowski and Petch, both this volume). Such an understanding would require that measurements of cloud properties are done simultaneously for several variables, over a wide range of meteorological situations. Numerical weather prediction (NWP) models may also be very helpful in that regard. If an NWP model is producing the observed cloud amount in the right place at the right time (which has been shown to be the case in Illingworth et al. 2007), then we can be reasonably sure that the meteorological processes which produce the clouds are being well represented. NWP model outputs can then be used to understand how clouds are controlled by meteorological processes.

With the open availability of model simulations to the international climate science community (Meehl et al. 2007), the number of biases of climate models reported in the literature has dramatically increased. However, compared to the hundreds of scientists involved in the analysis of climate simulations, the number of scientists actually working on the development and continuous improvement of physical parameterizations in climate models is fairly small. This may be caused by the fact that institutional structures do not reward this activity enough, especially since it may take a long time and considerable effort to get real improvements, while it is much easier to demonstrate errors in models. In this situation, how should the effort of parameterization improvement be concentrated on the most critical processes?

The same question arises regarding the reduction of uncertainties in the models' projections of the future climate. The Fourth Assessment Report (AR4) (IPCC 2007) reports a large range of global climate sensitivity estimates among climate models. The largest contribution to this spread arises from intermodel differences in cloud feedbacks (Soden and Held 2006). Given the very large number of factors or processes potentially involved in these differences, which ones should we concentrate on to reduce, as efficiently as possible, the uncertainties in future climate change?

Based on different methodologies, recent studies suggest that the response of marine low-level clouds to climate change was the root cause of a large part of intermodel differences in global cloud feedbacks (Bony and Dufresne 2005; Webb et al. 2006; Wyant, Bretherton et al. 2006; Williams and Tselioudis 2007). By using an analysis method based on the stratification of the large-scale tropical circulation into dynamic regimes (Bony et al. 2004) or by analyzing simulations performed with idealized and simplified (aquaplanet) versions of climate models (Medeiros et al. 2008), it was shown that intermodel differences in the tropical cloud response were dominated by the response of clouds in the trade-wind regions. This suggests that an improvement in the representation of shallow convection and trade cumulus clouds is crucial for climate sensitivity. These studies shed light on the "silent majority" of tropical clouds that has locally a less spectacular impact on radiation than deep convec-tive clouds or stratus clouds, but plays a major role for climate sensitivity. This finding will foster further studies focused on the understanding, the simulation, and the evaluation of shallow clouds, and will thus help to reduce this critical uncertainty.

This example shows that by carrying out idealized studies of climate change and by decomposing the global cloud feedback problem into components related to specific physical processes, the problem becomes more tractable and can suggest targeted diagnostics of model-data comparison or of data analysis. Such approaches have the great potential of identifying the processes that are critical for climate change projections. This provides guidance to establish a hierarchy of necessary model developments, helps to fill the gap between climate studies and process studies, and contributes to a better understanding (and thus a better assessment of our confidence) in the models' results. Therefore, such studies should be considered as a key step in the strategy to reduce critical uncertainties associated with the response of clouds to a changing climate.

One caveat associated with this strategy, however, is that processes or cloud regimes that may be missing in all the models or that may be represented equally badly in all the models may not be identified as a key source of uncertainty, as they might actually play an important role in nature. It is thus important to complement this strategy with comparisons of models with observations, and with idealized studies investigating the potential impact of model weaknesses on the simulation of climate.

Studies of this kind are necessary to assess the extent to which some processes contribute more than others to uncertainties in climate change projections. For instance, the inability of large-scale models (NWP or climate models) to simulate accurately the diurnal cycle of convection over tropical land areas is often cited as a concern for the credibility of climate change projections (whatever they are). This bias, which assuredly reveals some weaknesses in the models' representation of physical processes, is likely to be a concern for representing realistically the interactions between local precipitation and vegetation processes for instance. However, for other issues, such as the magnitude of global climate change or the change in some monsoon characteristics, the extent to which this bias actually affects climate model projections has yet to be demonstrated or assessed.

Recent studies show that climate models still exhibit substantial biases in their simulation of the water vapor and temperature distributions in the current climate. However, John and Soden (2007) found no relationship between the biases exhibited by models in the current climate and the magnitude of the water vapor-lapse rate feedback produced in climate change. They interpret this result by the fact that the water vapor feedback depends on the fractional change of humidity, and that this quantity is insensitive to biases in the mean state. Although we cannot exclude the possibility that biases in humidity are associated with biases in cloudiness (such an association may even be expected), this example illustrates the fact that biases in the mean state do not necessarily affect climate change feedbacks.

Similarly, aerosol effects are known to play a key role in the evolution of the 20th century climate through their direct effect on radiation. They have an influence on the formation and radiative properties of clouds at the small or regional scale. It has been proposed that the radiative effects of aerosols at the surface and in the troposphere affect surface temperature and large-scale atmospheric dynamics, and then affect climate phenomena such as the south Asian monsoon (e.g., Ramanathan et al. 2005). Some studies have indicated that the indirect effect of aerosols on liquid water clouds has the potential to affect global-scale climate change. Thus, evaluating and refining the representation of these processes in climate models certainly constitutes an important area of model development. In addition, we know very little about the ability of a very small fraction of the aerosol particles to act as ice nuclei, thus influencing the glaciation of clouds and so affecting the cloud lifetime and precipitation efficiency. Potentially, such processes are very sensitive to small amounts of anthropogenic aerosols and have led to the suggestion that they might affect the development of deep convective clouds.

However, in terms of the understanding and simulation of the tropical or global cloud response to climate change, it is still unclear whether the interaction between aerosols and clouds is of primary importance, compared to the influence of other small-scale (e.g., boundary-layer turbulence and shallow convection) or large-scale (e.g., changes in the large-scale circulation) processes. In the remote trade-wind regions of the Pacific Ocean, for instance, it is unlikely that the physical and radiative properties of trade wind cumuli will be strongly affected by anthropogenic changes in aerosol properties. We can therefore consider, for the moment, that improving the representation of cloud-aerosol interactions in climate models is of lower priority, to reduce the uncertainty in simulated large-scale climate changes, than the improvement of physical processes (e.g., boundary-layer turbulence, atmospheric convection, and radiative transfer).

0 0

Post a comment