Transcript García-Moya, J.A., Santos, C., Escribŕ, P.A., Santos, D - c

Short-Range Ensemble Prediction System at INM
García-Moya, J.A., Santos, C., Escribà, P.A., Santos, D., Callado, A., Simarro, J. (NWPD, INM, SPAIN)
2nd SRNWP Workshop on Short Range Ensemble, Bologna, April 2005
SREPS at INM
OUTLINE
Meteorological Framework
•Main Weather Forecast issues are related
with Short-Range extreme events.
•Convective precipitation is the most
dangerous weather event in Spain (Some
fast cyclogenesis, several cases of more
than 200 mm/few hours every year).
•Multi-model approach
•Multi-boundaries:
From few global
deterministic models
New computer Cray X1
Hirlam
HRM
MM5
UM
Two main phases (2002-2005) :
ECMWF
GME
AVN
UKMO
BIAS & RMS
1. Cray X1
770 Gf
15 nodes (4 MSPs/node)
Determistic Forecast.
2. Cray X1e
2300 Gf
15 nodes (8 MSPs/node)
Deterministic + SREPS
Stamps View of Multimodel-Multiboundaries
•Deterministic ECMWF as reference up-left
•HRM, MM5, Hirlam models in rows
TEST RUN PERFORMANCE & VERIFICATION
•AVN, ECMWF, GME, UKMO BCs in columns
TALAGRAND (RANK HISTOGRAMS)
SPREAD & EMSD
SPREAD vs EMSD
Test run & validation
Hirlam, HRM and MM5.
36 hours forecast once a day (00 UTC).
MAPS SPREAD & EM
PLUMES
5 days of comparison (20040103-20040107).
Four different initial and boundary conditions (EMCWF, GME from DWD,
AVN from NCEP and UM from UKMO).
Use ECMWF operational analysis as reference.
No control experiment, then “natural” BCs will be “control” for each model
(ECMWF for Hirlam, GME for HRM, AVN for MM5).
MULTIMODEL PROGRESS
EACH MODEL & BCs OUTPUTS
•Test run area (beige) improved to Larger area (blue)
DAYLY PRE-OPERATIONAL RUN
•HRM and UM models in migration process
•GME BCs not yet in large enough area, UM BCs almost
running
•Running daily (Hirlam,MM5) models X (AVN,ECMWF) BCs
INTRANET WEB SERVER
ENSEMBLE OUTPUTS: PROBABILITY MAPS
Monitoring in real
time
Deterministic outputs
for each model and BCs
•Models X BCs tables
Ensemble probabilistic
outputs
•Probability maps: 6h
accumulated precipitation,
10m wind speed, 2m
temperature trends
•Ensemble mean & Spread
maps
•EPSgrams
ENSEMBLE OUTPUTS: ENSEMBLE MEAN & SPREAD MAPS
Ensemble mean & Spread Maps
Verification
•Ensemble mean (isolines) and spread
(coulours).
•Deterministic scores
•Spatial distribution of variability.
•Talagrand, Spread vs
Emsd, ROC, etc.
•Variability comparison with
meteorological pattern.
CONCLUSIONS
Conclusions for Multimodel
Advantages
Future
• Better representation of perturbations (SAMEX results)
•Verification software for multimodel ensemble (precipitations,
ROC curves, …)
• Better results
•UM model ready to use
Disadvantages
• Difficult to implement operationally (four different models should be
maintained operationally)
• Expensive in terms of human resources
• No control experiment in the ensemble, use of “centroid” as control
•Daily run at midday
•Post process software (targeting clustering)
•Bayesian model averaging for improvement in calibration and
better skill for weighted average
References
Palmer, T. et al, 2004: Development of a European Multi-Model Ensemble System for Seasonal to Interannual Prediction (DEMETER). ECMWF, Technical Memorandum nº434.
J.A. Garcia – Moya, [email protected]
Hou, D., E. Kalnay, and K. K. Droegemeier, 2001: Objective verification of the SAMEX'98 ensemble forecasts. Mon. Wea. Rev., 129, 73-91.
Carlos Santos, [email protected]
Raftery A., Balabdaoui F., Gneiting T. and Polakowski M., 2003: Using bayesian model averaging to calibrate forecast ensembles. Technical report nº440. Department of
Statistics. University of Washington.
Numerical Weather Prediction Department. INM.