The Multiscale Non‐Gaussian Statistics of Free‐Running 1000‐Member General Circulation Model Ensembles

ABSTRACT

Probabilistic atmospheric forecasts are useful for disaster management and a variety of socioeconomic activities. One avenue to improve these forecasts is through advancing our understanding of forecast distributions. This study advances that understanding through a multiscale investigation of 1000‐member free‐running ensembles of a coarse resolution intermediate complexity general circulation model (GCM). We found direct non‐parametric evidence that a time‐varying equilibrium forecast distribution manifests at long lead times (> 40% longer than the variance saturation time). Interestingly, the forecast distribution exhibits the strongest non‐Gaussian characteristics prior to equilibration. That spike in non‐Gaussianity is likely because the forecast distribution encountered the boundary separating physically realistic and unrealistic atmospheric states. The characteristics of the time‐varying equilibrium distribution are also explored. Our findings thus imply that (1) useful information may be present in probabilistic forecasts beyond the variance saturation time, and (2) models of forecast distributions should account for the complicated structures that arise in multivariate forecast statistics prior to equilibration. These implications and our findings merit further investigation using more realistic GCMs.