An Assessment of the Parameterization of Subgrid-Scale Cloud Effects on Radiative Transfer. Part I: Vertical Overlap

Abstract

Different approaches for parameterizing the effects of vertical variability of cloudiness on radiative transfer are assessed using a database constructed from observations derived from lidar and millimeter cloud radar data collected from three different locations. Five different methods for dealing with the vertical overlap of clouds were incorporated into a single radiation model that was applied to the lidar/radar data averaged in time. The calculated fluxes and heating rates derived with this model are compared to broadband fluxes and heating rates calculated with the independent column approximation using the time-resolved cloud data. These comparisons provide a way of evaluating the effects of different overlap assumptions on the calculation of domain-mean fluxes. It was demonstrated how two of the most commonly used overlap schemes, the random and maximum-random methods, suffer a severe problem in that the total cloud amount defined by these methods depends on the vertical resolution of the host model thus creating a vertical-resolution-dependent bias in model total cloudiness and radiative fluxes. A new method is introduced to overcome this problem by preserving the total column cloud amount. Despite these problems, the comparisons presented show that most methods introduce a relatively small bias with respect to the single-column data. This is largely a consequence of the nature of the cloud cover statistics associated with the lidar/radar observations used in this study and might not apply in general. Among the three best-performing methods (random, overcast random, and maximum random), the more commonly used maximum-random method does not perform significantly better than the other two methods with regard to both bias and rms error despite its relative high computational cost. The comparisons also reveal the nature and magnitude of the random errors that are introduced by the subgrid-scale parameterizations. These random errors are large and an inevitable consequence of the parameterization process that treats cloud structure statistically. These errors may be thought of as a source of noise to the general circulation model in which the parameterization is embedded.