Treatment of model uncertainty from radiation by the Stochastically Perturbed Parametrization Tendencies (SPPT) scheme and associated revisions in the ECMWF ensembles

The Stochastically Perturbed Parametrization Tendencies (SPPT) scheme is used in ensemble forecast systems to represent uncertainties due to the atmospheric physics parametrizations. The scheme works by introducing random perturbations to the tendencies from the physics schemes. The size of the perturbations is modulated by the size of the net tendencies from the full physics suite. Recent work has identified an inconsistency in this approach to attributing model uncertainty: greater uncertainty is attributed to the night-time sky than to the daytime sky in clear-sky conditions, which is not deemed physically realistic. This paper documents the details of a revised configuration of SPPT, which became active in the recent operational version of the ensemble forecast system ("CY45R1") at the European Centre for Medium-Range Weather Forecasts (ECMWF) - released in June 2018. In this work, a modification to SPPT is outlined such that the clear-sky contribution from the radiation scheme is subtracted from the physics tendencies that are acted on by SPPT. Inspection of individual vertical profiles of physics tendencies provides a greater understanding of the interactions between processes, and helps demonstrate that the revision gives a more realistic description of the associated uncertainties. Forecast experiments with the revised version of SPPT demonstrate neutral to positive impacts on the skill of the ensemble forecasts in the medium-range; and in the extended-range, a significantly improved match between the spread and error of the principal components describing the Madden-Julian Oscillation. In addition, it is shown that including the revisions in the ensemble of data assimilations leads to increases in spread in the boundary layer.