© Crown copyright 2005 ECMWF User Meeting, 14-16 June 2006 Page 1 Developments in the Use of Short and Medium-Range Ensembles at the Met Office Ken Mylne Ensemble Forecasting Manager
Mar 27, 2015
© Crown copyright 2005 ECMWF User Meeting, 14-16 June 2006 Page 1
Developments in the Use of Short and Medium-Range Ensembles at the Met Office
Ken Mylne
Ensemble Forecasting Manager
© Crown copyright 2005 ECMWF User Meeting, 14-16 June 2006 Page 2
Contents
First Guess Early Warnings – again!
Example of EPS Use for Confidence
New Applications of EPSDrought-planning ForecastsBest-Member ProjectEurorisk PREVIEW Windstorms
MOGREPS – The new Met Office short-range ensemble
First Guess Early Warnings- A confession
© Crown copyright 2005 ECMWF User Meeting, 14-16 June 2006 Page 4
FGEW Verification – an error
Over the last few years we have reported on verification results from our First Guess Early WarningsA key feature was a maximum in skill at 4 days
Much discussion and investigationWith regret and apologies, I have to report that this
“result” was caused by an error in our verification schemeMisalignment of forecast times with verifying
times in the verification databaseDay 4 was correctly aligned, other days wrong
© Crown copyright 2005 ECMWF User Meeting, 14-16 June 2006 Page 5
Corrected Verification - ROC
With the bug corrected the verification shows a near-constant ROC area for days 1-4 of the forecast with a slow decline from day 5 or 6.
© Crown copyright 2005 ECMWF User Meeting, 14-16 June 2006 Page 6
Corrected Verification - Reliability
Reliability diagrams show useful resolution at 2 and 4 days.
© Crown copyright 2005 ECMWF User Meeting, 14-16 June 2006 Page 7
Corrected Verification – Cost-loss
Cost-loss curves show slow reduction in values at longer lead timesLargest effect for large C/L ratios
Increasing chance of capturing event at high probability
Example of EPS Use for Confidence
© Crown copyright 2005 ECMWF User Meeting, 14-16 June 2006 Page 9
Example – low spread, high confidence
Cold period in November ‘06 with snow in SW
ECMWF EPS gave very high confidence of blocking breakdown
Allowed issue of high confidence of return to mild conditions on Wed 30th 5 days aheadAnalysis of 0600 on 30th confirms this was correct
New Applications of EPS - i) Drought Planning Forecasts
© Crown copyright 2005 ECMWF User Meeting, 14-16 June 2006 Page 11
Drought planning forecasts
Probabilities for 10-day rainfall totals
Mod-High probability of below normalOnly low risk of very dry (<5mm in 10 days) in SE England
New Applications of EPS - ii) “Best Member” Project
© Crown copyright 2005 ECMWF User Meeting, 14-16 June 2006 Page 13
Integrating Ensembles into Operations
We would like to make all forecasts probabilisticBUT Met Office customers still demand a best-estimate deterministic forecastCurrently use field modification
Link UK GM at T+36 with preferred solution laterCan we choose an ensemble “Best Member”?
Best fit to preferred solution, or Ensemble Mean?Representative member of largest cluster?
Best Member could give:Dynamically consistent forecast at all timesFull set of consistent model fields
© Crown copyright 2005 ECMWF User Meeting, 14-16 June 2006 Page 14
Best Member project
Project being run with Operations Centre to test:Can we identify a suitable Best Member?How long does a ‘good’ EPS member stay ‘good’? Can we re-calculate probabilities?
Expect BM selected for mesoscale may be different from that for synoptic scale/medium range forecasts.Use error tracking to identify source regions
Consider short and medium ranges separately, using MOGREPS and EC EPS respectively.
New Applications of EPS - iii) Eurorisk PREVIEW Windstorms
© Crown copyright 2005 ECMWF User Meeting, 14-16 June 2006 Page 16
Ensemble Forecasts PEACE (MeteoFrance)
SRNWP-PEPS (DWD)
LAMEPS (Met.no)
ECMWF EPS (ECMWF)
MOGREPS ( Met Office UK)
COSMO-LEPS
Eurorisk PREVIEW Windstorms
Database (Met Office UK)
Kalman Filter
Wind Gust Calculation
(Meteo-France)
Windstorm Forecast Products
Website(Met Office UK)
Risk Mapping Data(SMHI)
Windstorm Risk Mapping Products
Multi-model ensemble forecasts for Windstorm Risk Adapted existing site-specific database tables for additional ensembles 196 sites across 9 countries
© Crown copyright 2005 ECMWF User Meeting, 14-16 June 2006 Page 17
Products Traffic light maps for initial alerts Site-specific details:
Meteograms Wind-roses
Eurorisk PREVIEW Windstorms
© Crown copyright 2005 ECMWF User Meeting, 14-16 June 2006 Page 18
Eurorisk PREVIEW Windstorms
Alert thresholds will be a combination of probability and severity, eg.
Severity (Climate Percentile)90% 95% 99% 99.9%
Prob >80%
50-80%
20-50%
1-20%
Thresholds referenced to site climatesAnalysed from ERA40Calibrated using site observations
MOGREPS – Met Office Global and Regional EPS
© Crown copyright 2005 ECMWF User Meeting, 14-16 June 2006 Page 20
MOGREPS – The Met Office short-range Ensemble
Ensemble designed for short-rangeRegional ensemble over N. Atlantic and Europe (NAE)Nested within global ensembleETKF perturbationsStochastic physicsT+72 global, T+36 regionalAim to assess uncertainty in short-range, eg.:
Rapid cyclogenesis Local details (wind etc) Precipitation Fog and cloud
NAE
MOGREPS is on Operational Trial for 1 year from September 2005
Page 21© Crown copyright 2005 NWP Seminar 3rd Feb 2006
T+12 perturbed forecast
T+12 ensemble mean forecast
( - ) + =
( - ) + =
( - ) + =
( - ) + =
( - ) + =Transform matrix
Control analysis
Perturbed analysis
0.9 Pert 1-0.1 Pert 2-0.1 Pert 3-0.1 Pert 4-0.1 Pert 5
Ensemble Transform Kalman Filter (ETKF)
Page 22© Crown copyright 2005 NWP Seminar 3rd Feb 2006
Stochastic Convective Vorticity (SCV)Unresolved impact of organised convection (MCSs)Not used in the higher resolution regional ensemble
Random Parameters (RP)Structural error due to tuneable parameters
Stochastic Kinetic Energy Backscatter (SKEB)Similar to ECMWF CASBS schemeExcess dissipation of energy at small scalesSKEB implemented in global ensemble in PS11
Stochastic schemes for the UM
Impact is propagated to next cycle through the ETKF!
© Crown copyright 2005 ECMWF User Meeting, 14-16 June 2006 Page 23
Display system
© Crown copyright 2005 ECMWF User Meeting, 14-16 June 2006 Page 24
Products – Postage stamps
© Crown copyright 2005 ECMWF User Meeting, 14-16 June 2006 Page 25
Products – Probability charts
© Crown copyright 2005 ECMWF User Meeting, 14-16 June 2006 Page 26
MOGREPS Site-specific forecasts
EPS Meteograms
MOGREPS Plume
Kalman filter MOS is being implemented for MOGREPS forecasts
Page 27© Crown copyright 2005 NWP Seminar 3rd Feb 2006
Verification
Verification to date is very basicVerification is being implemented within the Area-based Verification system (ABV) and Site-specific Verification system (SBV)
Verification performed over NAE area for forecasts from global ensemble
Performed (except where stated) against analysis
For 111 cycles between 17/10/05 and 9/1/06
Page 28© Crown copyright 2005 NWP Seminar 3rd Feb 2006
500hPa height – spread and RMSE
Spread optimised by variable inflation factor against observations in u, v, T and RH at T+12
Appears too large because verified against analysis (tbc)
Ensemble mean skill does not currently improve on control skill
Spread and RMSE for 500hPa GPH
Page 29© Crown copyright 2005 NWP Seminar 3rd Feb 2006
500hPa height – rank histogram
Rank histogram is encouragingly flatClose to ideal
Suggests that ETKF perturbations are representative of genuine analysis errors
This performance seems much improved on ECMWF ensemble
Rank Histogram at T+72 for 500hPa GPH
Solid – operational, Dotted – PS11 upgrade
© Crown copyright 2005 ECMWF User Meeting, 14-16 June 2006 Page 30
Verification - T+36 6hr total > 0.5mm
Global Ensemble Regional Ensemble
© Crown copyright 2005 ECMWF User Meeting, 14-16 June 2006 Page 31
Verification - T+36 6hr total > 5mm
Global Ensemble Regional Ensemble
© Crown copyright 2005 ECMWF User Meeting, 14-16 June 2006 Page 32
Verification - T+36 6hr total > 0.5mm
Global Ensemble
Global Regional
0.84 0.88
© Crown copyright 2005 ECMWF User Meeting, 14-16 June 2006 Page 33
Conclusions
FGEW verification resolved!
EPS continues to become more integrated in forecasting proceduresFirst-choice for new services (eg drought planning)How to produce best deterministic forecast?Basis of new risk-management services for EU
MOGREPS provides a promising new short-range ensemble capability complements EPS for medium-range
Any questions?