Recent Experiences with the INM Multi-model EPS scheme

Recent Experiences with the INM Multi-model EPS scheme

García-Moya, J.A., Callado, A., Santos, C., Santos, D., Simarro, J., B Orfila.

Modelling Area – Spanish Met Service INMEWGLAN/SREPS meeting

Federal Office of Meteorology and Climatology MeteoSwiss

9-12 October 2006

OCTOBER 2006 EWGLAM/SRNWP Meetings ZURICH

OutlineIntroductionSREPS system at INMMonitoring and postprocessingVerification against observation vs

Verification against analysisFurther Work and Future of SREPSConclusions


Introduction

Triggering and Consolidating INM SREPS

NWP plan, (April 1999)SAMEX. (Summer 2000)Global boundaries; LAMs. 2000-2004Cray X1E. (2001-2005)Gathering the team


Ensemble for Short Range

Surface parameters are the most important ones for weather forecast.Forecast of extreme events (convective precip, gales,…) is probabilistic.Short Range Ensemble prediction can help to forecast these events.Forecast risk (Palmer, ECMWF Seminar 2002) is the goal for both Medium- and, also, Short-Range Prediction.


What do we need?

Enough computer power.Research

Following recommendations of the workshops.

Technical difficulties.Large storage system.Database software (MARS like ECMWF).Post-processing and graphics software.Enough staff for maintenance and monitoring.Verification software.


SREPS at INM

72 hours forecast four times a day (00, 06, 12 y 18 UTC).Characteristics:

5 models.4 boundary conditions.4 last ensembles (HH, HH-6, HH-12, HH-18).

20 member ensemble every 6 hoursTime-lagged Super-Ensemble of 80 members every 6 hours.


Multi-model

Hirlam (http://hirlam.org).

HRM from DWD (German Weather Service).

MM5 (http://box.mmm.ucar.edu/mm5/).

UM from UKMO (Great Britain Weather Service).

LM (Lokal Model) from COSMO consortium.


Multi-Boundaries

From different global deterministic models:ECMWFUM from UKMO (Great Britain Weather Service)

AVN from NCEPGME from DWD (German Weather Service)


The team

José A. García-Moya.Carlos Santos (Hirlam, verification & graphics, web server).Daniel Santos (MM5, Bayesian Model Average).Alfons Callado (UM & grib software).Juan Simarro (HRM, LM and Vertical interpolation software).


Thanks to…

MetOfficeKen Mylne, Jorge Bornemann

DWDDetlev Majewski, Michael Gertz

ECMWFMetview Team

COSMOChiara Marsigli, Ulrich Schättler


Current Ensemble

72 hours forecast twice a day (00 & 12 UTC).Characteristics:

5 models.4 boundary conditions.

20 member ensemble every 12 hours



HP Computer Cray X1e

16 nodes, 8 MSP’s each ( ~2.4 Tf peak perf.)

Deterministic ForecastSREPSClimatic runs


Post-processing

Integration areas 0.25 latxlon, 40 levelsInterpolation to a common area

~ North Atlantic + EuropeGrid 380x184, 0.25º

SoftwareEnhanced PC + LinuxECMWF Metview + Local developments

OutputsDeterministicEnsemble probabilistic


Monitoring in real time

Intranet web serverDeterministic outputs

Models X BCs tables Maps for each couple (model,BCs)

Ensemble probabilistic outputsProbability maps: 6h accumulated precipitation, 10m wind speed, 24h 2m temperature trendEnsemble mean & Spread mapsEPSgrams (work in progress)

Verification: Deterministic & ProbabilisticAgainst ECMWF analysisAgainst observations


Monit 2: all models X bcs


Monit 3: All Prob 24h 2m T trend


Monit 4: Spread - Emean maps


Case Study 2006061000

More than 15 mm/6 hours


VerificationThe 2006 first half (6months) verification results against both references observations and ECMWF analysis are available.

Calibration: with synoptic variables Z500, T500, PmslResponse to binary events: reliability and resolution of surface variables: 10m surface wind, 6h and 24h accumulated precipitation


Verification exerciseInterpolation to a common area

~ North Atlantic + North Africa + EuropeLat-lon Grid 380x184, 0.25º

~180 days (Jan1 to Jun30 2006).Two different references:

Analysis: ECMWF (6h and 24h det fc for Acc. Prec.)Observations: TEMP & SYNOP

Verification software ~ ECMWF Metview + Local developments

Deterministic scoresSynoptic variables: Bias & RMSE for each member & Ens Mean

Probabilistic ensemble scoresSynoptic variables: CalibrationSurface variables: Response to binary events


Probabilistic ensemble scores

Ensemble calibration:Synoptic variables:

Z500, T500, Pmsl

Scores:Rank histogramsSpread-skill


Rank histograms: examples

Large spreadSmall spread

Over predictionUnder prediction

Well calibrated


Z500

Verification exercise:~ North Atlantic + North Africa + Europe, Lat-lon Grid 380x184, 0.25º~180 days (Jan1 to Jun30 2006).Analysis: ECMWF (6h and 24h det fc for Acc. Prec.)Observations: TEMP & SYNOP

Synoptic variables (here Z500) spread-skill & rank histograms against observations, show the ensemble is under-dispersive, a bit under-forecastingThe same against ECMWF analysis is very good


Probabilistic ensemble scores

Response to binary events:Surface variables:

10m surface wind (10,15,20m/s thresholds)6h accumulated precipitation (1,5,10,20mm thresholds)24h accumulated precipitation (1,5,10,20mm thresholds)

Scores:Reliability, sharpness (H+24, H+48)ROC, Relative Value (H+24, H+48)BSS, ROCA with forecast length


24hAccPrec ROC & ROCA

Surface variables against observations show medium/quite good reliability and good resolution, degrading with threshold (clearly) and forecast length Here is shown 24h Accumulated precipitation performance in HH+30 forecasts: reliability with sharpness, ROC and ROCA, Brier Skill Score, Relative economic value.Verification against ECMWF analysis is much better


Further work

The ensemble performance could be improved with some post-processing, today under development (Flattery method):

Bias correctionCalibration using Bayesian Model Averaging (BMA)


ROAD MAP

2003-2004 Research to find best ensemble for the Short Range

Jun 04 – Jun 05 Building Multimodel System

Jun 05-Dec 05 Mummubn/16 members

Daily run non-operational

Mar 06 Mummub 16/16 members

Once a day

Jun 06 Mummub20 members

Twice a day

July 06 Obs verfication

September 06 40 member lagged Super-

ensemble

Twice a day

October 06 BMA Calibration

January 07 Broadcast products Experimental


FUTURE ISSUES

Aladin and WRF as additional forecasting modelsMulti analysis from HIRLAM 3DVAR model and first guess from global model forecastsAlternative methods for multiple initial conditions Verification against observations (high resolution precipitation network over Europe)More post-process software (clustering)Statistical downscaling applied to SREPS outputsConvergence with GLAMEPS and regional THORPEXData policy aspects


Conclusions

A Multi-model-Multi-boundaries Short Range Ensemble Prediction System (MMSREPS), is preoperational at INMVerification results (2006 first half), against both observations and ECMWF analysis have been obtainedThese first results look promising:

Verification against ECMWF analysis shows very good resultsVerification against observations shows quite good results

Ensemble is under-dispersive Good response to binary events

Future of INM SREPS is still open

Documents

Recent Experiences with the INM Multi-model EPS scheme