Breaking the cloud-parameterization deadlock

A key factor limiting the reliability of simulations of anthropogenic climate change is the inability to accurately represent the various effects of clouds on climate. Despite the best efforts of the community, the problem has resisted solution for several decades. The reasons for this are briefly reviewed and it is argued that it will be many more decades before the problem can be solved through the approaches to cloud parameterization that have been used up to now. An alternative approach, called superparameterization, is then outlined, in which high-resolution cloud system–resolving models (CSRMs) are used in place of the conventional cloud parameterizations. Tests performed with the Community Atmosphere Model show that superparameterizations can give more realistic simulations of the current climate, including greatly improved simulations of the Madden–Julian oscillation and other tropical wave disturbances. Superparameterizations increase the cost of climate simulation by a factor of several hundred dollars, but can make efficient use of massively parallel computers. In addition, superparameterizations make it possible for a climate model to converge to a global CSRM as the horizontal grid spacing of the climate model decreases to a few kilometers. No existing global atmospheric model has this convergence property. Superparameterizations have the potential to greatly increase the reliability of climate change simulations.