Slide 1 © ECMWF

Diagnostics of the ensemble system for polar regions

Linus Magnusson

Slide 2 © ECMWF

RMSE error (against ERA Interim reanalysis)

2-da

y er

ror

6-da

y er

ror

Antarctic Arctic

(Cf. Magnusson and Källén, 2013, MWR, in press)

Slide 3 © ECMWF

Daily errors (2012, z500, 6-day forecasts)

Arctic

Antarctic

Slide 4 © ECMWF

August 2012 over Antarctica

Slide 5 © ECMWF

Temperature anomaly in the troposphere and stratosphere

3 Aug 13 Aug 23 Aug 2 Sept

500

200

100

50

20

10

5 hPa

Slide 6 © ECMWF

Temperature at 50 hPa

2012-08-14 2012-08-19 2012-08-21 2012-08-29

Slide 7 © ECMWF

Best and worst ensemble members

Z500 (70-90S)

T10 (70-90S) 14 19 29 24

Slide 8 © ECMWF

Concept behind ensemble forecasts

Initial Perturbations

Perturbed forecast Optimal analysis = + +

Model Perturbations

Aim: Simulate the uncertainties in the forecast

Needs all components of uncertainty to simulate the forecast uncertainty

Standard deviation of the error = Standard deviation of the ensemble

Slide 9 © ECMWF

Ensemble of data assimilations – effect of polar orbiting satellites

Slide 10 © ECMWF

EDA standard deviation – T ~500 hPa (15 Oct – 1 Nov, twice a day)

0.15 0.30 0.45 0.6 0.75 0.9

All observations No polar orbiting sat.

(Thanks to M. Bonavita)

Slide 11 © ECMWF

Ensemble forecasts – Antarctic (8 cases = small sample)

RMSE Ens. Stdev.

Slide 12 © ECMWF

Ensemble forecasts - Arctic

RMSE Ens. Stdev.

Slide 13 © ECMWF

Are the uncertainties from the ensemble system reliable in the Arctic?

Slide 14 © ECMWF

RMSE and Ens.Std for temperature at 500 hPa Arctic

Good match!

RMSE (solid) Ens.Std. (dotted)

Slide 15 © ECMWF

RMSE and Ens.Std for temperature at 850 hPa Arctic

Reasonable match.

RMSE (solid) Ens.Std. (dotted)

Slide 16 © ECMWF

RMSE and Ens.Std for 2-metre temperature, Arctic (against SYNOP)

Bad match…

RMSE (solid) Ens.Std. (dotted)

2 6 4 2 8 0

Slide 17 © ECMWF

SYNOP stations north of 65N (6-day error of each station)

Slide 18 © ECMWF

Tarfala Nikkaloukta

17 km between the stations, ENS resolution 32 km..

Slide 19 © ECMWF

Observed and forecasted temperature 2013-02-07

Tarfala

Nikkaloukta

-40

-20

0

-40

-20

0

7 8 9

7 8 9

Analysis (shade) Synop (diamonds)

North-western Scandianavia

How to represent this difference in ENS?

Slide 20 © ECMWF

Stochastic perturbed physical tendencies (SPPT)

dX/dt=Dynamical tend.+(1+r)Physical tend.(convection, radiation, cloud, diffusion, dissipation)

No perturbations in the boundary layer and the stratosphere

Palmer et al. (2009)

Slide 21 © ECMWF

Mean tendencies for temperature (Dec-Jan)

Convection scheme Radiation scheme

Cloud scheme Dynamics (advection)

1000

200

500

1000

200

500

1000

200

500

1000

200

500

90N 90S Eq.

90N 90S Eq. 90N 90S Eq. Thanks to N. Wedi and S. Lang

Slide 22 © ECMWF

Summary

• The forecasting system has improved over the years

• The ensemble system have to simulate the remaining errors

• Still many types of errors to solve or simulate (stratosphere-troposphere, surface, sea-ice, boundary layer)

• How to include sub-grid variability?

Slide 23 © ECMWF

Diagnostics of the ensemble system for polar regionsRMSE error (against ERA Interim reanalysis)Daily errors (2012, z500, 6-day forecasts)August 2012 over AntarcticaTemperature anomaly in the troposphere and stratosphereTemperature at 50 hPaBest and worst ensemble membersConcept behind ensemble forecastsEnsemble of data assimilations – effect of polar orbiting satellitesEDA standard deviation – T ~500 hPa�(15 Oct – 1 Nov, twice a day)Ensemble forecasts – Antarctic�(8 cases = small sample)Ensemble forecasts - ArcticAre the uncertainties from the ensemble system reliable in the Arctic?RMSE and Ens.Std for temperature at 500 hPa ArcticRMSE and Ens.Std for temperature at 850 hPa ArcticRMSE and Ens.Std for 2-metre temperature, Arctic�(against SYNOP)SYNOP stations north of 65N (6-day error of each station)Slide Number 18Observed and forecasted temperature 2013-02-07Stochastic perturbed physical tendencies (SPPT)Mean tendencies for temperature (Dec-Jan)SummarySlide Number 23