Stochastic generation of subgrid-scale cloudy columns for large-scale models

To use the Monte Carlo Independent Column Approximation method for computing domain-average radiative fluxes in large-scale atmospheric models (LSAMs), a method is needed for generating cloudy subcolumns within LSAM columns. Here, a stochastic cloud generator is introduced to produce the subcolumns. The generator creates a cloud field on a column-by-column basis using information about layer cloud fraction, vertical overlap of cloud fraction and cloud condensate for adjacent layers, and density functions describing horizontal variations in cloud water content. The performance of the generator is assessed using a single day's worth of data from an LSAM simulation that employed a low-resolution two-dimensional cloud-resolving model (CRM) within each LSAM column (a total of similar to59000 cloudy domains). Statistical characteristics of generated cloud fields are compared against original CRM data, and radiative-transfer biases associated with the generator are evaluated. When the generator is initialized to the greatest extent possible with information obtained from the CRM fields. overall biases are small. For example, global-mean total cloud fraction exhibits a bias of -0.004, as compared with -0.024 for maximum-random overlap (MRO) and 0.047 for random overlap. Biases in radiative fluxes and heating rates are in general 1/4 to 1/2 those for MRO with horizontally homogeneous clouds.