Last updated: 2008
The final report on SREPS was published as the 13th Tech. Report
The project deals with the development and implementation of a short-range ensemble with the COSMO model, the COSMO SREPS (Short-Range Ensemble Prediction System). This system is built to fulfil some needs that have recently arisen in the COSMO community:
| to have a short-range mesoscale ensemble to improve the support especially in situations of high impact weather |
| to have a very short-range ensemble for data assimilation purposes |
| to provide boundary conditions for the convection-resolving ensemble under development at DWD |
The strategy to generate the mesoscale ensemble members proposed by this project try to take into account as many sources of uncertainty as possible and therefore to model many of the possible causes of forecast error. The proposed system would benefit of: perturbations in the boundary conditions, perturbations of the model and perturbations of the initial conditions.
It is not possible for COSMO, being a limited-area model, to neglect the contribution of the boundary conditions. In order to take into account the error of the global model that comes from the boundaries, a multi-model approach is used. A MUlti-Model MUlti-Boundaries (MUMMUB) ensemble system is currently run by INM, where four different limited-area models (UM, HIRLAM, HRM, MM5) are driven by the four global models (IFS, GME, UM, NCEP) which provides initial and boundary conditions.
Following an informal agreement between COSMO and INM, that is being formalised as task 0 of this project, COSMO will be added to this set of limited-area models and COSMO will benefit by the four runs of COSMO on a very big domain at 25 km of horizontal resolution, nested on the four different global models.
The four INM-COSMO runs are used in the SREPS project to drive a number of COSMO at higher resolution (7 or 10 km). Each of the 25 km COSMO runs provides initial and boundary conditions to a few higher resolution COSMO runs, in which other perturbations are applied to the system.
The SREPS strategy would include the modelling of different sources of forecast error.
With regards to the model error, perturbations to the mesoscale model will be added
following different techniques: by using different parameterisation schemes, by perturbing
the parameters of the schemes, by perturbing the physical tendencies. Furthermore, in a
later stage of the project, mesoscale perturbations to the initial conditions will also be
added.
The best strategy to be followed to this aim will be defined during the project (possible
cooperation with UK Met Office).
SREPS ensemble size is devised to be around 20 members. A simple number could be 16, due to the fact that the system is based on 4 global models (or better on 4 25 km COSMO runs driven by these 4 global models) and 4 higher resolution perturbed COSMO runs could be nested on each of the lower resolution runs.
The SREPS area will be designed in order to include all the COSMO countries; for the sake of simplicity we could choose the COSMO-LEPS area or a similar but smaller domain. A resolution of 10 km will be adopted in the development stage, also allowing a simple comparison with COSMO-LEPS, with the possibility to move to a few kilometres of horizontal resolution in the operational system. The forecast range will be 72 hours at maximum.
The SREPS project is divided into six tasks:
It consisted of three jobs: use of COSMO by INM, provision of the forecasts to COSMO and provision of the interpolation codes
| Task 1.1 | Tendencies |
| Task 1.2 | Schemes (physics and numerics) |
| Task 1.3 | Parameters |
| Task 1.4 | Surface forcings |
This task is not divided into separate subtasks
| Task 3.1 | Organisation of data transfer between INM and ECMWF |
| Task 3.2 | Adaptation of source code to give the possibility to use tuneable or random parameters (external), support from WG6 |
| Task 3.3 | Extracting LM frames and nesting of LM on LM frames (WG6) |
| Task 3.4 | Discussion on computer resources |
| Task 3.5 | Joint meeting SIR-COSMO SREPS |
| Task 3.6 | Possibility to have a COSMO originating centre in the GRIB coding |
| Task 4.1 | Preliminary test of the methodologies under development based on "model perturbed COSMO-LEPS" on case studies |
| Task 4.2 | Test of SREPS prototype on real-time case studies and comparison with COSMO-LEPS and "model perturbed COSMO-LEPS" |
| Task 5.1 | Selection of the methodology to generate the perturbations |
| Task 5.2 | Selection of the testing period |
| Task 5.3 | Investigation on computer resources |
| Task 5.4 | Implementation on a regular basis |
| Task 5.5 | Test of different parameter perturbations |
| Task 6.1 | Selection of the verification methodologies (with WG5) |
| Task 6.2 | Selection of the reference data-set |
| Task 6.3 | Computation of the score |
The project will be structured in an investigation phase followed by an 'operational' phase. The investigation will cover:
This ensemble has to be designed keeping in mind that it could be used for the new COSMO data assimilation scheme. Much attention will be kept in having a correct amount of spread both in the very short range and throughout all the short-range.
SREPS project has strong links with other COSMO activities: WG2 and WG3 activities (establishing the values of the tuneable parameters), WG6 activities (need to keep the tuneable parameters external to the COSMO code). A link with the development of a convection-resolving ensemble at 2.8 km by DWD will also be established, to devise some strategies for the model perturbations which could be shared between the two projects, in order to exchange experience and results and also to obtain two systems not in conflict with each other.
The use of a multi-model approach with a limited-area model permits to have a strong link also with the global project TIGGE, where all the available global ensemble will be collected together with the possibility to be used as providers of boundary conditions for limited-area model runs.
As for the relationship with COSMO-LEPS, the perturbations are ingested into the COSMO-LEPS system mainly through the boundary conditions, which are provided by some selected members of the operational ECMWF EPS. These perturbations comes from different sources:
Since the COSMO-LEPS perturbations derived form the EPS ones, the system is useful especially in the early medium range (day 3-5). In order to assure a smaller scale variability to the mesoscale ensemble, another kind of perturbations is added to the system, by allowing a random choice of the scheme to be used for the parameterisation of the deep convection (Tiedtke or Kain-Fritsch) in the COSMO runs.
A preliminary study suggested that the perturbations applied in this way play a minor role with respect to the perturbations at the boundaries, which explain the major amount of the spread. Then, it appears that the COSMO-LEPS system is not adapt as it is now for the short-range, mainly becauset he main perturbations are designed to grow in the medium range and the sources of small scale error are not well described
The necessity of using a probabilistic approach also for the short-range and for the smaller scales can be understood in the light of the recent model developments. Despite the possibility to explore the scale of the order of few kilometres, the capability of numerical models to correctly forecast local and intense precipitation is still nowadays limited.
This is true even at short time-range, up to 48 hours, due to the loss of atmospheric predictability going down to small spatial and temporal scales. At this scales (1 to 10 km), it is not completely satisfactory just to reproduce precipitating structures, but a correct localisation in space and time is required together with realistic peak values.
This considerations introduce the need of a quantification of the predictability associated with a forecast which has been until now considered better provided by a deterministic very high resolution model. The scales at which the uncertainty needs to be quantified are influenced by phenomena which are only marginally considered in the generation of the perturbations for the global ensemble.
In order to design an ensemble system for these purposes, the use of perturbations that are more local in scale and which can generate a reasonable spread in the short-range is explored.