Date: 13/12/2018 00:32, Last saved by Marco ArpagausPierre Eckert
Development of short range ensemble (SREPS)Tackle deficiencies in precipitation forecasts
Version 1.0
Motivation
There are various indications, both from verification (WG5) and from customers (e.g., forecasters), that the LM has serious deficiencies in forecasting precipitation. Some of these problems are longstanding (and not necessarily unique to the LM), some others are fairly recent.Short range ensembles have been developed in several countries like the USA, Canada, Japan, the UK, Norway, France…. Improvements, especially in situations of high impact weather have been noticed.
This project aims at highlighting and investigating some of these deficiencies, and putting forward possible improvements. As such, it will need resources from members of many working groups, most notably WGs 1, 2, 3, and 5COSMO already has experience in ensemble forecasting, but mainly as a downscaling of the ECMWF ensemble in the (early) medium range. Perturbations have for the moment been included though the lateral b
oundaries, they are not present in the initial state. Perturbation of the physics has “only” been introduced by running two different convection schemes.
TGeneral axes of investigationhe tasks foreseen are:
Consolidate verification findings (area, season, synoptic situation, vertical stability, etc; cf. project ‘conditional verification’)Development of proper initial perturbations: nudging of perturbed observations, variational assimilation, Kalman filtering, use interpolated state from different coarser analyses.
Select cases/episodes with largest/cleanest signalExplore techniques in perturbation of the trajectories in order to produce realistic spread in the short range.
Optimise use of perturbed lateral boundaries.
Tasks
Phase 1: definition of methodologies and model system set-up
a)Formal agreement between COSMO and INM (use of LM by INM, provision of the different forecasts to COSMO)
b)Organisation of transfer of data between INM and ECMWF
c)Development of methodologies (on LM and LMK) to generate perturbations of model trajectories: tendencies, schemes, parameters, surface forcing. Need expertise of WG 3. Developments realised mainly at ECMWF.
d)Adaptation of model source code to give the possibility to use tuneable or random parameters. Support from WG6.
Phase 2: implementation and testing of the prototype on a regular basis
a)Implement and run an ensemble on the COSMO LEPS area or smaller, 7 or 10 km resolution, 72h forecast, up to 20 members.
b)Further tuning and development of the system according to SIR requirements among others.
c)If required, implementation of an own assimilation procedure.
d)Verification of the quality of the model and comparison with other existing predictions: deterministic LMs, EFI, COSMO LEPS, PEPS…
Depending on verification results, make sensitivity studies on a set of selected cases/episodes:
Deliverables:
Short range small scale (not convective) ensemble forecasts for provision of
- Products to the COSMO community
- boundary conditions for the EELMK forecast system at DWD
- boundary conditions for the SIR filter at DWD
Collaborations:
INM
ECMWF
a.UKMOAnalysis, Data Assimilation:
i.Impact of soil moisture
ii.Impact of atmospheric humidity
iii.…
b.Dynamics and Numerics:
i.Impact of numerical scheme
ii.Impact of horizontal (numerical) diffusion
iii.…
c.Physics:
i.Sensitivity to microphysics scheme (e.g., free parameters)
ii.Sensitivity to convection scheme (e.g., trigger function)
iii.Sensitivity to boundary layer scheme (e.g., parameterization of fluxes)
iv.…
In parallel to the tasks driven by the verification results, idealized tests should be made:
Investigate moist benchmark cases such as Bryan and Fritsch (2002) or the ones suggested by the Bad Orb Meeting 2005.
Project members
ARPA-SIM (Montani, Marsigli, Paccagnella)
DWD (Renner, Theis, NN)
MeteoSwiss (Walser?)
DLR (Keil, Craig)
Resources
Estimated resources (in FTE-years) needed:33
Minimum FTE per year: 1.05
Advanced interpretation of LM outputs
Version 1.0
Motivation
Models usually show various types of systematic errors. These errors can be removed locally by the use of statistical methods like MOS, Kalman filters, non linear training. These errors are expected to decrease in the future due to the improvement of the models. The diagnostics of high impact weather is however often left to the appreciation of the forecasters. Objective guidance in decision making could be helpful to them.
The foreseen increase in resolution of the models will lead to a proliferation of grid points and probably also to an increase of the noise in the forecasts. The effects of the double penalty will increase for events not predicted exactly at the right place at the right time.
The WMO recommends that the National Weather Services work more with so called forecast matrices or forecast databases.
The forecaster has to an actor of basic changes in his/her activities. The multiplication of products and users leads to the fact that forecast shifts are overloaded with production tasks at the detriment of the weather monitoring. The production should thus be automated by the mean of a forecast database. This database is fed with the best possible post-processing issued from numerical models, but the forecaster should keep the last word on the edition of the database. The liberated time can then be devoted to the evaluation of high impact weather situations. The structure of the weather forecast offices and of the storm prediction centers must also be adapted to the new activities. The forecasters must also be involved in the development of the meteorological science and the time devoted to education about the systems, the model and the local climate should be increased (CBS St-Petersburg, February 2005)
The iinitialisation of forecast matrices (weather type, occurrence of various phenomena, probabilities…) should be as good as possible and requests good postprocessing.
Tasks
- Explore more grid point statistics, also on other parameters than precipitation
- New classification and regression algorithms for detection of high impact weather. These include the use of complex predictors like instability indices and non linear classification algorithms like neural networks.
- Generation of weather types (thunderstorm, drizzle, fog,…)
Estimated resources (in FTE-years) needed:1.5
Minimum FTE per year: 0.5
Tentative collaborators:
DWD (Hoffmann, Renner, Theis)
MeteoSwiss (Marchand, Perler, forecasters MétéoGenève)
Others from Poland, Greece, Rumania
Tentative collaborators:
This project will replace the former Work Packages 3.10.1 (‘cure overestimation of low precipitation in winter’) and 3.10.4 (‘understand sharp increase in precipitation overestimation since November 2003).
1