Initial-Value vs. Model-Induced Forecast Error: A New Perspective

Numerical models of the atmosphere are based on best theory available. Understandably, theoretical assessment errors induced by use such is confounding. Without clear guidance, experimental separation model-induced part total forecast error also challenging. In this study, and ensemble perturbation variances were decomposed. Smaller- larger-scale components, separated as a function lead time, independent. They associated with features completely vs. only partially lost skill, respectively. For their phenomenological description, variance was further decomposed orthogonally into positional structural components. An analysis various components revealed that chaotically amplifying initial predominantly led to differences in forecasts, while interpreted an indicator error. Model-induced found be relatively small. These results confirmed earlier assumptions limited empirical evidence numerical may near perfect scales they well resolve.