Revision of the Stochastically Perturbed Parametrisations model uncertainty scheme in the Integrated Forecasting System

Abstract

The Stochastically Perturbed Parametrisations scheme (SPP) represents model uncertainty in numerical weather prediction by introducing stochastic perturbations into the physical parametrisation schemes. The perturbations are constructed in such a way that the internal consistency of the physical parametrisation schemes is preserved. We developed a revised version of SPP for the Integrated Forecasting System of the European Centre for Medium‐Range Weather Forecasts (ECMWF). The revised version introduces perturbations to additional quantities and modifies the probability distributions sampled by the scheme. Medium‐range ensemble forecasts with the revised SPP are considerably more skilful than ensemble forecasts with the original implementation of SPP. The revised version of SPP is similar, in terms of forecast skill, to the Stochastically Perturbed Parametrisation Tendency scheme (SPPT), which is currently used to represent model uncertainty in the operational ECMWF ensemble forecasts.

More accuracy with less precision

“…We follow the approach of Lang et al . (2021). CRPS is computed every 24 hr, from 24 hr up to 360 hr for a range of variables.…”

Ensemble prediction using a new dataset of ECMWF initial states – OpenEnsemble 1.0

“…Running large numbers of ensemble forecasts requires substantial computational resources. Leutbecher (2019) demonstrated that the number of ensemble start dates is much more important than the size of the ensemble for extracting the mean probabilistic skill of the system. Thus, we keep the number of start dates high but decrease the ensemble size for the higher resolutions in order to save computational resources: only the T L 159 experiments and the basic set-up for T L 399 were run with the full 50 ensemble members.…”

“…Fair versions of probabilistic skill scores indicate how the system would have scored if it had had an infinite number of ensemble members 1 . Leutbecher (2019) illustrated how a fair version of the continuous ranked probability score (CRPS) can be constructed and also explored how many ensemble members are required to calculate a representative fair CRPS. The recommended ensemble size was set to be four to eight members for scientific testing.…”

“…The smaller ensembles are constructed from the 50-member ensemble. A mathematical prerequisite for calculating the fair CRPS is that the ensemble members need to be exchangeable, which is not fulfilled in the ECMWF ensemble under this initial state construction style, where the SV and EDA perturbations are used with plus-minus symmetry (see Leutbecher, 2019). Note that this is no longer the case from 2019 onwards in the ECMWF operational ensemble configuration.…”

“…Lock et al, 2019) as well as exploring completely new methods (see e.g. Ollinaho et al, 2017;Leutbecher et al, 2017;Lang et al, 2021), (2) ensemble modelling studies of tropical storms (see e.g. Lang et al, 2012) offer a unique way to study forecast impacts of both initial state uncertainty and model uncertainty representations, and (3) EPSs provide a potent way of automating the model tuning process (Ollinaho et al, 2013(Ollinaho et al, , 2014Tuppi et al, 2020).…”