Comment on wcd-2022-6

Abstract

Global numerical weather prediction is often limited by Lorenz–type “butterflies” in the flow. Here we think of these as local flow configurations associated with pronounced uncertainty growth–rates, as demonstrated in short–range ensemble predictions. Some of these configurations correspond to potential instabilities of the flow. Often, forecasts for severe downstream weather only show an improvement in skill when these butterflies have passed. Here we focus on the “cyclogenesis butterfly” – associated with baroclinic and convective instabilities in the extratropics. One question addressed is how do different operational ensemble forecast systems (within the “TIGGE” archive) represent the associated uncertainty growth–rates? To test which might be “better”, an extended spread–error equation is used to investigate how well these non–linear models maintain short–range statistical reliability during cyclogenesis. For the European Centre for Medium–Range Weather Forecasts (ECMWF) ensemble, considerable short–range over–spread is found in the North Atlantic stormtrack during winter 2020/21 – representing a source of untapped predictability (this result appears to generalise to other stormtracks and at least one other model). Flow–type clustering demonstrates that this over–spread is directly associated with the representation of cyclogenesis. We attempt to quantify the contributions to the total spread in cyclogenesis cases from initial uncertainty (as derived from ensemble data assimilation), singular vector perturbations, and model uncertainty. At day 2, we find that up to 25 % can be associated with the singular vectors and up to 25 % with model uncertainty. The over–spread suggests that reductions in forecast error over recent years would permit further development of these uncertainty aspects. The sensitivities of spread to resolution, the explicit representation of convection, and the assimilation of local observations provide additional insight for future development.