Notes on the appropriateness of ‘‘bred modes’’ for generating initial perturbations used in ensemble prediction

Papers by Szunyogh and co-workers and Iyengar and co-workers, among others, claim that “bred growing vectors” (BGVs) are appropriate for generating initial perturbations for application to ensemble weather prediction. The theoretical bases for this claim are purported relationships between the bred vectors and what are called local Lyapunov vectors (LLVs). Several statements in these papers are inaccurate, however, regarding: the properties of LLVs in contrast to those of singular vectors; the relationship between LLVs and BGVs; the relationship between BGVs and analysis errors; the relationship between singular vectors and dynamic imbalances; and the adequacy of very low-resolution models to characterize singular vectors of models used for numerical weather prediction. These inaccuracies and their implications for the use of BGVs in ensemble weather prediction are discussed here.