Observational Strategies that Address These Uncertainties Cloud Schemes in Largescale Models

Clouds often form at scales much smaller than the typical size of a grid box in general circulation models (GCMs) and cannot therefore be explicitly predicted in these models. To predict the cloud fraction (together with other cloud properties), many large-scale models diagnose the cloud fraction by using statistical cloud parameterizations. In this approach, subgrid-scale fluctuations of variables (e.g., total water, potential temperature, or vertical velocity) are described by a probability distribution function (PDF) whose statistical moments

(mean, variance, and skewness) must be diagnosed based on large-scale prognostic variables plus eventually some subgrid-scale variables predicted by turbulence or convection schemes (e.g., Bony and Emanuel 2001; Tompkins 2002). Here, the cloud fraction and water content are related to the fraction of the PDF above saturation and its first moment, respectively. This PDF can also be used to predict cloud overlap and consequent radiative properties as well as to provide a better description of the development of precipitation. Such an approach is promising to fill the gap between the different cloud scales and to strengthen the physical coupling between the different cloud processes.

A better documentation and understanding of the influence of subgrid-scale processes (e.g., turbulence, convection, gravity waves) on the PDF of large-scale variables for different cloud regimes (e.g., deep convective clouds, shallow clouds, cirrus) is required to guide development or improve statistical cloud parameterizations. For this purpose, modelers often use simulations from cloud-resolving models (CRMs) to get some guidance. This approach might now be developed with the arrival of global CRM simulations and super-pa-rameterizations. This would allow, for instance, the examination and better understanding of how the PDF of different variables relates to small-scale physical processes and interacts with the large-scale environment. One might then investigate why large-scale models fail to simulate middle-level clouds while some CRMs do a better job (e.g., Liu et al. 2001). An important prerequisite for the success of this approach, however, is that CRMs (or large eddy simulations) simulate the subgrid-scale fluctuations accurately. To assess whether it is actually the case, comparisons between observed and simulated fluctuations and cloud distributions produced by high-resolution models on the 100 m to 2 km scale are required. This emphasizes the need to observe the humidity structure of the atmosphere in three dimensions with a high enough resolution.

0 0

Post a comment