Last updated: December 2008
This project aims at consolidating, within a two-year project, the ensemble forecasting systems for the mesoscale built within COSMO in the past years. The operational ensemble COSMO-LEPS and the experimental COSMO-SREPS system, both running at present at 10 km horizontal mesh-size, have been designed for different forecast ranges (day 3-5 and 1-3, respectively) and with different perturbation strategies.
The aim of the project is to end up with a unique "multi-perturbation-strategy" COSMO ensemble system, benefiting of perturbations which can produce appropriate spread for the entire forecast range (i.e., day 2-5), including a set of model perturbations which should guarantee a good description of the COSMO model error. Furthermore, a calibration strategy should be developed and applied to the ensemble output.
COSMO is currently involved in several ensemble projects. Since November 2002, the COSMO-LEPS system has been running on a daily basis; in the present configuration, COSMO-LEPS is a 16 member ensemble based on the COSMO model with a 10 km horizontal mesh-size and 40 vertical levels. The system is running at ECMWF using the Billing Units provided by the COSMO Countries which are also ECMWF member states (Germany, Greece, Italy, and Switzerland).
COSMO-LEPS is a dynamical downscaling of the ECMWF EPS, taking initial and boundary conditions from 16 selected members of the EPS. COSMO-LEPS is mainly designed for the early medium range (day 3-5) and the forecast range is 5.5 days. Some perturbations to the model are included, namely the used convection scheme and two turbulence parameters take values randomly selected between two options.
As for 2008 and part of 2009, the following modifications to the operational suite were agreed upon and planned:
In addition to the above modifications, the following implementation could be also considered from the second half of 2009 onwards:
In September 2006 the COSMO-SREPS Priority Project started, aiming at the development of an ensemble
system for the short range (until day 2-3) and for data assimilation purposes. Two different kind of
perturbations are applied:
initial and boundary conditions benefit of a multi-model approach, being provided by 4 different
operational global models (currently provided by AEMET) and the COSMO model itself is perturbed by
changing the values of a set of physics parameters. The project is ending in September 2008, delivering
an ensemble system for the short-range based on the COSMO model with a 10 km horizontal mesh-size and
40 levels in the vertical.
The system is running quasi-regularly at ECMWF, using the Billing Units provided by some COSMO Countries. COSMO-SREPS provides initial and boundary conditions to the convection-resolving ensemble under development at DWD (COSMO-DE-EPS) and it is used to compute a flow-dependent B-matrix for a test version of the 1d-Var assimilation of satellite data under development at ARPA-SIM.
The developments of the system planned for the next year, which are part of the PP CONSENS, are:
In addition to these two ensemble systems, DWD is developing a convection-resolving ensemble (COSMO-DE-EPS), nested in COSMO-SREPS, which will be used for the ensemble data assimilation system under development in the Consortium (KENDA Priority Project). In 2011 COSMO-DE-EPS will be no more nested in COSMO-SREPS as it is now, but in the ensemble system based on ICON, currently in its development phase.
The two mesoscale ensembles currently run by COSMO provide probabilistic forecasts at the short and medium range with high spatial resolution (10 km mesh-size), but they are not operating at the convection resolving scale. Though nowadays operational deterministic modelling is approaching the convection-resolving scale (1 km), operational probabilistic systems are usually run at a lower spatial resolution and only COSMO is running operationally an ensemble with a mesh-size as high as 10 km, thus providing probabilistic products at a resolution comparable to that typical of the deterministic run.
As a first need an up-to-date model version has to be used within the ensemble systems and a reduction of the mesh-size, i.e. from 10 to 7 km, is desirable.
As for the scientific development of the mesoscale ensembles, a satisfactory representation of the uncertainty affecting the mesoscale is still lacking. We need to improve the quantification of the errors which are made by the model in the description of these phenomena. The development should then include 1) the perturbation of the lower boundary forcings and 2) the perturbation of the parametrised physical processes. The importance of further perturbing the physics parameters has been underlined by the outcome of the COSMO-SREPS project. Within COSMO it has also been suggested that adding perturbations in the soil parameters can be crucial in order to get a better representation of the model error in terms of surface variables (e.g. Matthias Raschendorfer, personal communication).
A proper development of these types of perturbations implies the necessity of having an experimental system running continuously, covering different seasons and different areas, in order to have a robust statistical assessment of the impact of the perturbations, which will permit to choose the best strategy to be applied operationally. This work is not aiming at substituting the need for improvements in the description of the soil and of the soil-atmosphere exchanges within the model. This important task will be carried out in the next years within the "Consolidation of Lower Boundary Conditions" Priority Project. The ensemble system will automatically benefit of any update in the model brought about by this project.
Furthermore, there is a need of concentrating the efforts made by the COSMO partners for the aim of ensemble development and maintenance. As it is now, the situation would foresee for 2011 the co-existence of two 7/10-km ensembles, one for the short range and one for the medium range, besides the existence of a convection-resolving ensemble. In order to avoid duplications, we should aim at a confluence of the two 7/10-km ensemble systems into a unique COSMO ensemble by 2011, covering both the short and the medium range and based on the most appropriate perturbation strategy for the entire forecast range.
This implies the development of a "multi-perturbation-strategy" ensemble system, with perturbations more appropriate for the short range in the beginning (e.g. day 2), and with perturbations more appropriate for the medium range thereafter.
At present, apart from COSMO model perturbations, which seem to be useful for both forecast ranges to permit a description of the mesoscale model errors, two different strategies are adopted in the two ensembles. For the short-range it has been chosen to perturb initial and boundary conditions by using a multi-model approach, based on 4 state-of-the-art operational high-resolution modelling chains.
This guarantees to have initial conditions which are the best estimates of the atmospheric initial state as obtained by the 4 centres independently and to have boundary conditions provided by global model runs with quite high spatial mesh-size (25-40 km). For the medium-range, instead, the diversity of initial and boundary conditions is obtained by nesting COSMO into some members of the ECMWF global ensemble, targeted for the medium-range and run at a coarser mesh-size (50 km). An ensemble which should cover the entire forecast range having the appropriate spread and skill for all forecast ranges has to be built with an appropriate perturbation strategy, which at present is not defined.
In order to define this "multi-perturbation-strategy" initial and boundary condition perturbation technique, a dedicated study is needed. The impact of the multi-model approach as compared to the global ensemble downscaling approach has to be assessed for different forecast ranges. In addition, the possibility of having a downscaling applied to more than one global ensemble will be considered. The possibility of nesting the COSMO model into different global deterministic systems and even into different global ensemble systems has recently become more and more feasible due to the work done by AEMET and already exploited within COSMO-SREPS and by the starting of the Interoperability Project within the SRNWP framework, which should open the possibility to nest any model in any other model in Europe.
Furthermore, in the framework of the TIGGE-LAM project, it will become possible in the future to have access to initial and boundary conditions for all the global ensembles, simplifying the extension of the downscaling approach to new ensembles.
Within this work on the consolidation of the COSMO ensemble systems, attention should also be paid to the post-processing. Calibration of ensemble forecasts in terms of precipitation and surface temperature has been studied and applied widely in recent years (e.g. Hamill and Whitaker, 2006; Hagedorn et al., 2008; Hamill et al., 2008; Santos-Muñoz et al., 2007). As for the COSMO-LEPS ensemble, it has recently been recognised that a calibration for 24 h precipitation would be desirable to improve the forecast skill (Marsigli et al., 2008).
Last year, work on COSMO-LEPS calibration has been carried out at MeteoSwiss (Felix Fundel et al., 2008). Thirty year of reforecast of one member of COSMO-LEPS have been computed and stored at ECMWF, and they have been used for calibrating the COSMO-LEPS output over Switzerland. This work has shown the potential of using reforecast to improve the forecast skill: for precipitation over Switzerland a calibrated forecast for day 4 is as skilful as a raw forecast for day 1; furthermore, forecasts of rare events having no skill could be turned into skilful forecasts by calibration. Yet, this work has to be extended to achieve a properly calibrated ensemble.
Finally, it should be mentioned that a link between ensemble and data assimilation is established by the KENDA project. The COSMO-DE-EPS system, which is used within KENDA for the development of the ensemble data assimilation, receives initial and boundary conditions by the COSMO-SREPS system.
The actions proposed within the project are:
To guarantee the support to the COSMO-DE-EPS system by continuing to provide, in the next two years, initial and boundary conditions from the COSMO-SREPS ensemble. COSMO-SREPS will be maintained as a system running regularly.
To carry on testing the perturbations of both physics parameters and lower boundary conditions in the experimental COSMO-SREPS system. A system running regularly is needed for this purpose in order to permit a robust statistical assessment. As for the physics parameter perturbations, more parameters will be considered, depending on the development of the model physics.
Since it is no more needed that the COSMO-SREPS ensemble is symmetric with respect to perturbations as it was up to know (i.e. all the possible model perturbations where applied to each ic-bc set, limiting the number of implemented perturbed parameters), more parameters will be considered. Furthermore, the possibility of combining perturbations will be explored, by testing some combination of parameters in order to assess their joint impact on the forecast.
As for the lower boundary perturbations, a methodology has to be defined for this purpose. The impact of applying random perturbations to the lower boundary fields will be explored, investigating which degree of spatial correlation should be retained.
As an alternative methodology, where more that one algorithm for some field determination would be available (e.g. independent soil moisture analyses), the impact of switching between two or more fields will be analysed. After a 2-year testing period, the so-defined model perturbations will be then implemented into the unique mesoscale COSMO ensemble.
To establish a "multi-perturbation-strategy" COSMO ensemble system, using perturbations more appropriate for the short range in the beginning (e.g. day 2), and perturbations more appropriate for the medium range (day 3-5). The action focuses on the definition of which kind of perturbations are required for the two forecast ranges.
The impact of nesting into a pure multi-model system made up of operational global deterministic runs will be compared with using a single-model ensemble or even a multi-model ensemble (e.g. a combination of more than one global ensemble, possibly made available in the framework of the TIGGE-LAM cooperation). This work should be carried out in parallel with the study on the model perturbations.
To develop and test a calibration strategy for the mesoscale ensemble. The reforecasts run by MeteoSwiss for the COSMO-LEPS system will be used for this action. Different calibration techniques will be considered:
CDF calibration using model climatology (following Fundel et al., 2008), which is the technique already
applied by MeteoSwiss to COSMO-LEPS precipitation forecasts, standard bias correction as described in
Hamill and Whitaker (2006),
logistic regression, analog technique (Hamill and Whitaker, 2006), where some different implementation
of the technique should be tested to decide which is the most suitable for high-resolution forecasts.
The impact of the application of these techniques to the ensemble precipitation forecasts will be verified
on a testing period, in order to assess their impact. For this purpose, collection of precipitation data
on the whole domain would be a very important issue to be able to perform a complete ensemble calibration.
The rich dataset made available by the MAP D-PHASE Project could be very useful to carry out some of the analyses. The modular structure of the project implies that the different tasks can easily be carried out in parallel and by scientists from different institutions. Furthermore, actions 2-4 can be carried out almost independently from each other, hence the temporal order of the deriving tasks could even be changed during the project without affecting the outcome.
These are the project tasks:
| Fundel F, M. A. Liniger, C. Appenzeller, A. Walser | 2008 | Calibrated Warnings of Precipitation Events over Switzerland Using COSMO-LEPS Reforecasts, in preparation. |
| Hagedorn R, Hamill TM and Whitaker JS | 2008 | Probabilistic Forecast Calibration Using ECMWF and GFS Ensemble Reforecasts. Part I: 2-meter Temperatures, Monthly Weather Review, in press | Hamill TM and Whitaker JS | 2006 | Probabilistic quantitative precipitation forecasts based on reforecast analogs: theory and application, Monthly Weather Review, 134, 3209-3229. |
| Marsigli C, Montani A and Paccagnella T | 2008 | A spatial verification method applied to the evaluation of high-resolution ensemble forecasts, Meteorological Applications, 15, 125-143 |
| Santos-Muñoz D, Callado A, Garcia-Moya JA, Santos C and Simarro J | 2007 | Bayesian Model Averaging of INM-SREPS, EMS7/ECAM8 Abstracts, Vol. 4, EMS2007-A-00362, 7th EMS Annual Meeting / 8th ECAM, 1-5 October 2007, San Lorenzo de El Escorial, Madrid. |