1. The COSMO-LEPS system at ECMWF
Chiara Marsigli, Andrea Montani, Tiziana Paccagnella
ARPA-SIM, Bologna, Italy

2. Outline The Limited-area Ensemble Prediction System COSMO-LEPS
Research priorities and system development
LAM ensemble for the short-range: COSMO-SREPS
COSMO-LEPS requirements

3. COSMO-LEPS late-short early-medium range (3-5 days)
boundary conditions
“downscaling” ensemble
initial conditions
LAM ensemble
ECMWF EPS
LM run
LM run
LM run

4. COSMO-LEPS suite at ECMWF (running from 5/11/2002, new suite from 5/2/2006)
16 Representative Members driving the 16 Lokal Modell integrations
(weighted according to the cluster population)
Random choice of the convection scheme (Tiedtke or Kain-Fristch)
4 variables
Z U V Q
3 levels
hPa
d-1 d
d+1
d+2
d+3
d+4
d+5
oldest EPS
Cluster Analysis + selection of 16 RMs 00
youngest EPS
2 time steps
European Area
Complete Linkage 12
clustering period
Recent changes
suite running daily at ECMWF managed by ARPA-SIM (ECMWF time-critical application);
Δx = 10 km - 40 ML;
forecast range: 132h;
computer time (4 million BU in 2006) provided by the COSMO partners which are ECMWF member states
archived on MARS
COSMO-LEPS integration domain
COSMO-LEPS
clustering area

5. COSMO–LEPS verification (SON 2004)
Verification grid
COSMO-LEPS (10 km) vs EPS (80 km)
over 4500 stations
box included in scores if at least 10 obs fall in the box
verification of the distributions

6. Average values - boxes 1.5x1.5 deg
tp > 10mm/24h
tp > 10mm/24h
COSMO-LEPS
10-MEMBER EPS
51-MEMBER EPS
tp > 20mm/24h
tp > 20mm/24h

7. Maximum values - boxes 1.5x1.5 deg
tp > 10mm/24h
tp > 10mm/24h
COSMO-LEPS
10-MEMBER EPS
51-MEMBER EPS
tp > 20mm/24h
tp > 20mm/24h

8. COSMO-LEPS box size New verification! 1mm/24h MED 0.5 MED 1.0 MED 1.5

9. Research priorities
member representativeness in terms of precipitation (weighting problem)
addition of model perturbations
short-range ensemble => SREPS

10. 1. Representativness (weighting problem)
no evidence of better scores obtained by the weighted COSMO-LEPS against the not weighted one in terms of precipitation
except in a subjective evaluation (ARPAL, D. Sacchetti)
problem of representativeness in precipitation?
too large clustering area?

11. ARPAL, D. Sacchetti

12. representativness Brute force (51 members) 12/11/2002
different clusters
different clusters
same cluster
same cluster
(mm)
Brute force (51 members) 12/11/2002

13. “islands”
Compute p1 and p2 percentiles of the whole precipitation field
Identify an island starting from a point where p2 is overcome and extending until not going under the p1 value
Compute an index defined as
I=a*bs+(1-a)*(dpd-1)
where bs is the rms distance between the island masks (a field =1 where there is an island and =0 otherwise) and dpd=(p1+p2)mi/(p1+p2)mj

14. ALPINE suite (SPCOLEPS billing units)
Experimental suite running daily from 15 July 2005
Same configuration as the operational COSMO-LEPS (red), but both clustering and integration domain are centred over the Alps (blue).
ope
alp
On single events, differences between the 2 suites can be noticeable.
Prob maps of 3-day rainfall exceeding 50 and 100 mm (fc24-96h); forecast starting at UTC
Objective verification on-going!

15. NW Europe suite (SPCOWIND billing units)
Generation of a new limited-area ensemble forecast system for the prediction of wind and wind gusts over North-Western Europe (EURORISK-PREVIEW project, WINDSTORM)
Aim: provision of site-specific wind forecasts to be post-processed and combined with other products by UK Met. Office.
Same configuration as the operational COSMO-LEPS (red), but both clustering and integration domain are shifted northwards (blue).
Experimental suite will start on June 2006

16. 2. Addition of model perturbations
boundary conditions
initial conditions
LAM ensemble
model (LAM):
parameters, schemes, tendencies, surface forcing

17. Parameter list (COSMO)

18. Test of a perturbed COSMO-LEPS vs operational COSMO-LEPS
Operational system (COSMO-LEPS):
b.c. and i.c. perturbations from 10 Representative Members selected out of 2 ECMWF EPS; use of Tiedtke or Kain-Fritsch random
Perturbed system:
addition of perturbations of the parameters of the model physics -> runtime perturbations
Forecast range: +72h
10 members
09/04/ UTC

19. clope
>10mm/6h
>20mm/6h
>30mm/6h
>40mm/6h

20. clpert
>10mm/6h
>20mm/6h
>30mm/6h
>40mm/6h

21. clope 30-36
>10mm/6h
>20mm/6h
>30mm/6h
>40mm/6h

22. clpert
>10mm/6h
>20mm/6h
>30mm/6h
>40mm/6h

23. Spread-skill

24. 3. Short-range ensemble (SREPS)
test the model perturbations on a short-range limited-area ensemble (same LAM model, same domain) using a multi-model approach for ic and bc
if working, use model perturbations also in COSMO-LEPS

25. ic and bc: multi-model approach
a MUlti-Model MUlti-Boundaries ensemble is operational at INM (Spain): 4 different limited-area models (HIRLAM, HRM, MM5, UM) are nested on 4 different global models (IFS, GME, AVN, UM) => 16 members ensemble (25 km horizontal resolution)
from May 2006 LM has been added
IFS LM 25km
GME LM 25km
AVN LM 25 km
UM LM 25 km
Courtesy of INM

26. IFS – ECMWF global
COSMO-SREPS (Short-Range Ensemble Prediction System) short range (up to 3 days)
LM at 25 km on IFS
GME – DWD global
LM at 25 km on GME P1 P2 P3 P4
4 different parameter perturbations (kept fix during the whole integration)
by INM Spain
UM – UKMO global
LM at 25 km on UM
16 LM runs at 10 km
AVN – NCEP global
LM at 25 km on NCEP

27. ensemble spread
12/05
12UTC
13/05
00UTC
13/05
12UTC

28. Requirements
3-hourly post-processing of ECMWF deterministic run up to fc+144h (like EPS members) so that both the LMDET run and the COSMO-LEPS runs have boundary conditions every 3 h
assistance so as to be able to interpolate on a regular lat-lon grid (using MARS at ECMWF) grib files produced on a rotated grid (like the COSMO-LEPS one)