Effective sample size for precipitation estimation in atmospheric general circulation model ensemble experiments: dependence on temporal and spatial averaging scales

The accuracy of climate projections is improved by increasing the number of samples from ensemble experiments, leading to a decrease in the confidence interval of a target climatological variable. The improvement in the accuracy depends on the degree of independence of each ensemble member in the experiments. When the members of ensemble experiments are dependent on each other, the introduction of an effective sample size (ESS) is necessary to correctly estimate the confidence interval. This study is the first attempt to estimate the ESS for precipitation as a function of the number of ensemble members, although some previous studies have investigated another type of ESS in terms of the length of simulation period. The ESS in the present study is intrinsic to the atmospheric general circulation models (AGCM) forced by the ocean boundary condition because the outputs of AGCM ensemble members are similar or dependent on each other due to the commonly used boundary condition, i.e., the distribution of sea surface temperature, sea ice concentration, and sea ice thickness. Looking at the values of ESS as a function of geographical location, those in the tropics and over the ocean are smaller than those at higher latitudes and over continents; precipitation events in areas with smaller (larger) ESS are strongly (weakly) constrained by the ocean boundary condition. The increase in temporal and spatial averaging scales for precipitation estimation leads to the decrease in the ESS, of whose trend is attributed to the spatio-temporal characteristics of the precipitation events as represented by the power spectrum and co-spectrum.