The Monte Carlo Independent Column Approximation’s Conditional Random Noise: Impact on Simulated Climate

Abstract

The Monte Carlo Independent Column Approximation (McICA) method for computing domain-average radiative fluxes is unbiased with respect to the full ICA, but its flux estimates contain conditional random noise. Results for five experiments are used to assess the impact of McICA-related noise on simulations of global climate made by the NCAR Community Atmosphere Model (CAM). The experiment with the least noise (an order of magnitude below that of basic McICA) is taken as the reference. Two additional experiments help demonstrate how the impact of noise depends on the time interval between calls to the radiation code. Each experiment is an ensemble of seven 15-month simulations. Experiments with very high noise levels feature significant reductions to cloudiness in the lowermost model layer over tropical oceans as well as changes in highly related quantities. This bias appears immediately, stabilizes after a couple of model days, and appears to stem from nonlinear interactions between clouds and radiative heating. Outside the Tropics, insignificant differences prevail. When McICA sampling is confined to cloudy subcolumns and when, on average, 50% more samples, relative to basic McICA, are drawn for selected spectral intervals, McICA noise is much reduced and the results of the simulation are almost statistically indistinguishable from the reference. This is true both for mean fields and for the nature of fluctuations on scales ranging from 1 day to at least 30 days. While calling the radiation code once every 3 h instead of every hour allows the CAM additional time to incorporate McICA-related noise, the impact of noise is enhanced only slightly. In contrast, changing the radiative time step by itself produces effects that generally exceed the impact of McICA?s noise.