Ways to Reduce Critical Uncertainties in the Prediction of Clouds in a Changing Climate

Comparison of climate simulations with observations reveals a large number of systematic biases in current models. Faced with the long-standing biases of climate models and uncertainties in climate change projections, the optimal way to improve models is still open to question.

Resolution

The increase of the (horizontal and vertical) resolution of large-scale models is often cited as a way to improve model simulations. The experience of many modeling centers indicates that increasing the resolution does reduce some biases, such as the occurrence and strength of midlatitude storms, the simulation of extreme precipitation, or of orographic precipitation. However, it is far from solving all the problems. In particular, the simulation of continental precipitation, of the diurnal cycle, or of the Madden-Julian Oscillations does not improve substantially with resolution. This is the case for many other errors of large-scale models, including the difficulties of representing the cloud processes themselves, such as condensation on aerosol particles, glaciation, the size distribution of the cloud particles and their interaction with radiation, the degree of cloud overlap in the vertical, and the conversion of cloud water into precipitation. We know these are inadequately parameterized and lead to modeled clouds with different characteristics from those indicated by our limited database of cloud observations, but it is unclear if increased resolution will ameliorate the situation. Current models have a higher vertical resolution in the boundary layer and thus can resolve some of the vertical structure of stratocumulus and, to a lesser extent, fair weather cumulus. We know that mid-level clouds are underestimated in nearly all models. Why is this? Is it because we cannot resolve the position of cloud top and base? Is it because the radiation scheme is not called often enough? Is it because the diagnosed phase is incorrect? Is it because there is no turbulent mixing scheme outside the boundary layer? It is also becoming clear that supercooled clouds commonly form and are widespread and persistent and have potentially important radiative effects, but are scarcely represented in the models. Is this also a resolution problem?

Complexity

Another avenue of model development is the increase of complexity. Coupled ocean-atmosphere models are now coupled to complex land-surface schemes, aerosol modules, chemistry, carbon cycle, etc. to form so-called Earth System Models. This allows us to investigate new climate feedbacks (such as carbon-climate feedbacks) but does not reduce the uncertainty in climate change projections. On the contrary, intermodel differences in regional precipitation changes and in climate sensitivity are often amplified by carbon-cycle feedbacks (which are very sensitive to precipitation and climate sensitivity changes) or aerosol feedbacks.

Physical Parameterizations

Improving the physical parameterizations used in large-scale models (in particular, the representation of turbulent, convective cloud processes, and the interaction with radiation) seems to be the most efficient way to reduce uncertainties in model projections of the future climate. However, improving parameterizations is diff cult, the number of people actively involved in this work is fairly small at the present, and the progress is slow. National and international funding agencies might play a role in encouraging these activities. Nevertheless, with the arrival of new cloud observations and with the increasing interactions and collaborations between meso- and large-scale modelers, one may expect more progress over the next few years than there has been in the past. From the observations, can we demonstrate that we really need dual moment schemes to represent ice and liquid water, and can we show that such schemes are adequately constrained to lead to improvements? Can observations reliably confirm the existence of large cloud-free regions (see Karcher and Spichtinger, this volume, and references therein), which are very highly supersaturated with respect to ice, and do we need to adjust our parameterization schemes to take this into account? What level of sophistication is needed in the treatment of aerosols?

Using Cloud-resolving Models instead of Cloud Parameterizations in Climate Models

Now that "super-parameterizations" and "global CRMs" have become available (cf. Grabowski and Petch; Collins and Satoh, both this volume), using CRMs instead of cloud parameterizations might constitute an option to reduce the uncertainty in cloud feedbacks associated with cloud parameterizations. Although these new approaches are promising, they are unlikely, however, to solve the cloud-climate problem issue in the near future for at least for two reasons. First, these approaches are computationally very expensive. Thus, it seems unlikely that ensembles of century-scale simulations can be performed with such models to study changes in the global climate or in climate extremes. Second, the resolution of CRMs is insufficient to resolve boundary-layer turbulence or cloud microphysics and, therefore, parameterizations are still required. The results obtained with these models are likely to depend on these parameterizations and at least on some poorly constrained parameters. This dependence should be explored and quantified before the cloud feedbacks produced by these models can be considered less uncertain than those derived from large-scale models.

On the other hand, sensitivity experiments performed with global CRMs or super-parameterizations can be very instructive to explore the physics of cloud feedbacks and climate sensitivity. It would be very valuable, for example, to understand why an aquaplanet global CRM (Miura et al. 2005) and a GCM embedding a two-dimensional CRM within each grid box instead of a cloud parameterization (Wyant, Khairoutdinov et al. 2006) both predict a climate sensitivity weaker than estimated by most global climate models. It would also be valuable (and computationally cheaper) to perform climate simulations by embedding a CRM or a LES over a limited domain of the Earth instead of globally (e.g., a LES in subtropical regions predominantly covered by boundary-layer clouds). A complementary and constructive (rather than competitive) interaction between large- and mesoscale modeling approaches to study the cloud-climate problem is strongly required.

0 0

Post a comment