A stochastic convective backscatter scheme for use in ensemble prediction systems

Stochastic kinetic energy backscatter schemes are amongst several different types of stochastic parametrization used in operational ensemble weather prediction to account for model uncertainty. They aim to introduce structured wind perturbations randomly into the forecast model, targeting regions of high numerical dissipation, mountain-wave drag and deep convection. Substantially different implementations of the backscatter scheme in different weather centres confirm the weak scientific underpinning of such stochastic parametrizations and the lack of knowledge of where the true model uncertainty lies. The stochastic scheme presented here makes the reasonable assumption that random model error lies primarily with the parametrization of deep convection and, in particular, with its interaction with model dynamics close to the grid scale. This scheme introduces well-defined, circularly symmetric patterns of wind forcing centred on grid columns containing parametrized convective mass transfer. Tests of the scheme in ensemble mode are carried out in the European Centre for Medium-Range Weather Forecasts (ECMWF) forecast model, the spectral basis of which enables a clean implementation in terms of spatially filtered vorticity and divergence forcing fields. Its effectiveness as a model error representation is assessed relative to ECMWF's operational backscatter and perturbed parametrization tendency schemes. Bearing in mind the simplicity of the scheme, it is surprisingly effective in increasing spread, reducing the error of the ensemble mean and increasing probability skill scores, particularly in the Tropics. It is proposed that this representation of backscatter addresses model uncertainty more realistically than current operational stochastic backscatter schemes and may remove the need for a perturbed parametrization tendency approach.