Stochastic Tendency Perturbations
Uncertainties within the IFS ensemble system are currently represented by a stochastic perturbation technique (SPPT). This represents uncertainties due to the model integration, and is used:
- during the execution of the forecast.
- during creation of the Ensemble of Data Assimilations (EDA) as part of the representation of uncertainties in the forecast initial conditions.
Stochastically Perturbed Parameterisation Tendencies (SPPT)
SPPT randomly perturbs the tendencies from the physical parameterisation schemes. This is done to represent uncertainties in the effects of under-resolved processes that the atmospheric physics parametrisation schemes aim to describe. These uncertainties arise from either or both:
- parametrisation scheme assumptions which incorporate bulk descriptions of sub-grid scale processes active within an individual grid box or column.
- approximations that are necessary to describe poorly constrained processes.
The physics schemes operate through an entire grid box column. The perturbations act to preserve the shape of the unperturbed vertical profile of tendencies by multiplying a column of tendencies with a single random number.
SPPT perturbs tendencies but not fluxes and this may cause certain inconsistencies and different behaviour between perturbed and unperturbed forecasts.
Some Limitations of SPPT
The current SPPT scheme:
- tends to focus its perturbations above the boundary layer. Physics-related uncertainties in near-surface weather parameters (e.g. temperature, visibility) may be under-represented. The ensemble spread for such parameters may be too small.
- perturbations do not explicitly depend on the current synoptic pattern. Occasionally the ensemble may show a very small risk of extreme weather beyond what is synoptically reasonable. For example: convective heating from the seas in winter-time cold NW'ly outbreaks may be damped so that air-masses look unrealistically cold in one or two members).
It should be stressed that overall stochastic perturbations undoubtedly do deliver clear improvements in the ensemble performance.
Additional Sources of Information
(Note: In older material there may be references to issues that have subsequently been addressed)
- Read more on stochastic representation of model uncertainties.
- Read more on stochastic parameterisation and model uncertainty.
- Read more on stochastic methods for representing model uncertainties in the IFS.
Overview
Community Forums
Content Tools