51
Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS) A ten-year experiment of real-time Potential Vorticity modifications and inversions at Météo-France

Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

  • Upload
    tana

  • View
    49

  • Download
    0

Embed Size (px)

DESCRIPTION

A ten-year experiment of real-time Potential Vorticity modifications and inversions at M é t é o-France. Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS). Forecaster Expertise. Senior forecaster expertise at Météo-France:. - PowerPoint PPT Presentation

Citation preview

Page 1: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

Philippe Arbogast, Karine MaynardCNRM-GAME (Météo-France & CNRS)

A ten-year experiment of real-time

Potential Vorticity modifications and

inversions at

Météo-France

Page 2: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

2

Forecaster Expertise

Page 3: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

3

Senior forecaster expertise at Météo-France:

Verify NWP outputs at the very short range against observations in real time

Recognize coherent dynamical features using conceptual modelsChose the “best member “ among several solutions provided by

deterministic forecasts and scenarios from ensemblesMonitor severe weather warning

In particular: Assessment of upper-level dynamics expressed in terms of

PV/dynamical tropopause within NWP using satellite images (WV channels from geostationnary satellites)

And since 2005 : PV modifications of global analyses (or +3h,+6h forecasts) in

real time

Page 4: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

4

Forecaster Expertise

Page 5: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

5

Un état initial incertain conduit à une prévision incertaine

On peut estimer l’incertitude de l’état analysé en chaque point d’observation (radiances, RS, avions commerciaux…) par comparaison entre ébauche, observations et analyse

La prévision d’ensemble transporte l’incertitude dans le temps et l’espace renseigne la confiance dans la prévision

La sensibilité aux conditions initiales indique la position des erreurs initiales qui ont leur maximum d’amplification en 30h dans la zone cible (polygone violet)

Deux méthodes pour propager l’incertitude :

Forecaster Expertise

Page 6: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

6

Un état initial incertain conduit à une prévision incertaine

On peut estimer l’incertitude de l’état analysé en chaque point d’observation (radiances, RS, avions commerciaux…) par comparaison entre ébauche, observations et analyse

La prévision d’ensemble transporte l’incertitude dans le temps et l’espace renseigne la confiance dans la prévision

Finalement…. À une prévision incertaine correspond une erreur de prévision

La sensibilité aux conditions initiales indique la position des erreurs initiales qui ont leur maximum d’amplification en 30h dans la zone cible (polygone violet)

Deux méthodes pour propager l’incertitude :

Forecaster Expertise

Page 7: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

7

Page 8: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

8

Page 9: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

9

Forecaster Expertise

Objective link between WV and dynamics , particlcle filter (Wirth, Michel, Guth)

Page 10: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

10

Improvement of initial state through PV improvement

tropopause 2D modification/correction (surface with potential vorticity=1.5pvu) et MSLP (SYNERGIE)

3D PV correction buiding (using vertical PV covariance errors)PV inversionRerun of the model …

Page 11: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

11

1997 1998 2000 2002 2004 2006 2008 2010 2012 2014

4DVar

global

More and more sat. Data are assimilatedh and v resolution increase(5010km over Europe)

Explicit

microphysics

global LAM NH

2.5km

Global ensemble

35 members

Global ensemble

11 members

3DVar

global

Global EDA

Lothar&Martin

KlausXynthia

1st PV in

version with

Forecast improvement

MF project kick-off

QGPV +simple corre

ctions

(Hello et a

l., 2004 M

et.

Apps)

Ertel P

V graphical

modif+inversion+model ru

n

Suite in

operation

(Arbogast et a

l, 2008 Q

JRMS, 2011

W&F)

Experiment involving senior forecasters ?

Decision taken

Page 12: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

12

Improvement of initial state through PV improvement

Outline of the method

Page 13: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

13

Case study: windstorm Klaus (23-24 January 2009)

Page 14: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

14

Case study: windstorm Klaus (23-24 January 2009)

Page 15: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

15

Page 16: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

16

1st step:

What can be inferred from comparison between model and satellite/surface observations using Global ARPEGE run at 0600UTC and observations between 0600UTC and 1200UTC ? (decision required at 1200UTC)

Page 17: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

17

It appears clearly that the amplitude of the upper-level feature is underestimated by the model.

Page 18: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

18

“model to satellite” approach to reduce the uncertainty

Observation (Meteosat 8) 6h forecast

Valid time :1200UTC 23 January 2009

Page 19: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

19

Iso-PV

PV

PV correction (z) after 1D-var

x

y

z

Methodology PV modifications

Page 20: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

20

PV inversion

Before modification

After modification

Page 21: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

22

PV inversion at the Météo-France’s weather room

Outcomes of the experiments in real-time of not:

1. Particularly efficient when type-B cyclogenesis is present2. Reliable approach in average 3. Marginal computational cost 4. Suitable for surface systems 5. Suitable in cases of mesoscale convections (not only windstorms)

But

1. Difficult to monitor the initial state for explosive cyclogenesis without pre-existing upper PV features (Lothar, December 1999)

2. Less forecast busts to be corrected with time (more observations,  flow-dependant B matrices)

3. Growing importance of ensembles

Page 22: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

23

Resultat Klaus

Page 23: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

24

25 experiments/attempts of model state improvement have been achieved by 4 different senior forecasters and 3 scientists.

A subset of 14 randomly chosen runs has been built (2 runs for each forecaster/scientist)

t

12 UTC23 Jan 2009

12UTC24 Jan 200906 UTC

23 Jan 2009

Operational run

Modified runs

Operational run

obse

rvat

ions

obse

rvat

ions

obse

rvat

ions

Experiments design:

Page 24: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

25

PV inversion at the Météo-France’s weather room

Outcomes of the experiments in real-time of not:

1. Particularly efficient when type-B cyclogenesis is present2. Reliable approach in average 3. Marginal computational cost 4. Suitable for surface systems 5. Suitable in cases of mesoscale convections (not only windstorms)

But

1. Difficult to monitor the initial state for explosive cyclogenesis without pre-existing upper PV features (Lothar, December 1999)

2. Less forecast busts to be corrected with time (more observations,  flow-dependant B matrices)

3. Growing importance of ensembles

Page 25: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

26

Purpose :

Do several forecasters come to the same conclusion in terms of initial conditions errors and modifications (in terms of dynamical tropopause) that could be applied?

Common features among modifications ?

Page 26: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

27

3 first EOFs of the 14x14 covariance matrix of the perturbations set

(resp 50%, 9%,5% of the total variance)

The projection onto the first EOF maximizes the forecast skill

Page 27: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

28

Forecast skill (24h)

better than R6 AND R12 oper

Worst than R6 AND R12 oper

RMS Error for MSLPRMS Error for 10m wind magnitude

Oper 1200UTC MSLP RMSE

Oper 1200UTC wind RMSE

Oper R12 RMSE

Page 28: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

29

AS TEMPERATURE CTPIni v.2007

EQM CTPIni

EQM ARPEGE

+15h(~Tx J)

+27h(~Tn J+1)

Page 29: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

30

PV inversion at the Météo-France’s weather room

Outcomes of the experiments in real-time of not:

1. Particularly efficient when type-B cyclogenesis is present2. Reliable approach in average 3. Marginal computational cost 4. Suitable for surface systems 5. Suitable in cases of mesoscale convections (not only windstorms)

But

1. Difficult to monitor the initial state for explosive cyclogenesis without pre-existing upper PV features (Lothar, December 1999)

2. Less forecast busts to be corrected with time (more observations,  flow-dependant B matrices)

3. Growing importance of ensembles

Page 30: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

31

TSR 9-10h

Page 31: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

32

PV inversion at the Météo-France’s weather room

Outcomes of the experiments in real-time of not:

1. Particularly efficient when type-B cyclogenesis is present2. Reliable approach in average 3. Marginal computational cost 4. Suitable for surface systems 5. Suitable in cases of mesoscale convections (not only windstorms)

But

1. Difficult to monitor the initial state for explosive cyclogenesis without pre-existing upper PV features (Lothar, December 1999)

2. Less forecast busts to be corrected with time (more observations,  flow-dependant B matrices)

3. Growing importance of ensembles

Page 32: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

33

Situation du 28 mars 2008

33

Page 33: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

34

Situation le 28 mars 2008 à 06TU

34

Page 34: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

35 35

Page 35: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

36 36

Page 36: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

37

PV inversion at the Météo-France’s weather room

Outcomes of the experiments in real-time of not:

1. Particularly efficient when type-B cyclogenesis is present2. Reliable approach in average 3. Marginal computational cost 4. Suitable for surface systems 5. Suitable in cases of mesoscale convections (not only windstorms)

But

1. Difficult to monitor the initial state for explosive cyclogenesis without pre-existing upper PV features (Lothar, December 1999)

2. Less forecast busts to be corrected with time (more observations,  flow-dependant B matrices)

3. Growing importance of ensembles

Page 37: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

38

Page 38: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

39 R.d.Hullessen, Le Midi Libre

Page 39: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

40

After corrections

Initial state 1.5 PVU height and WV (M8) picture – Areas where corrections are applied are outlined

Page 40: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

41

PV

18UTC 00UTC

Page 41: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

42

Page 42: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

43

Argence, Vich,

Page 43: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

44

PV inversion at the Météo-France’s weather room

Outcomes of the experiments in real-time of not:

1. Particularly efficient when type-B cyclogenesis is present2. Reliable approach in average 3. Marginal computational cost 4. Suitable for surface systems 5. Suitable in cases of mesoscale convections (not only windstorms)

But

1. Difficult to monitor the initial state for explosive cyclogenesis without pre-existing upper PV features (Lothar, December 1999)

2. Less forecast busts to be corrected with time (more observations,  flow-dependant B matrices)

3. Growing importance of ensembles

Page 44: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

45

Page 45: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

46

PV inversion at the Météo-France’s weather room

Outcomes of the experiments in real-time of not:

1. Particularly efficient when type-B cyclogenesis is present2. Reliable approach in average 3. Marginal computational cost 4. Suitable for surface systems 5. Suitable in cases of mesoscale convections (not only windstorms)

But

1. Difficult to monitor the initial state for explosive cyclogenesis without pre-existing upper PV features (Lothar, December 1999)

2. Less forecast busts to be corrected with time (more observations,  flow-dependant B matrices)

3. Growing importance of ensembles

Page 46: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

47

Avec l’outil CTPini on retrace (en marron) le champ de PVu qu’on souhaiterait avoir (l’original est en bleu).

Page 47: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

48

Sur le réseau de ce 2 mai à 06TU sérieux problèmes de calage sur un retour d’est, que ce soit avec Arpège (en bas) ou avec Arome (en haut).

On a au moins en altitude un noyau de PVu qui n’est pas au bon endroit.

Page 48: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

49

Page 49: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

50

En bleu Pearp éch03 en marron CTpini

Page 50: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

51

Conclusion

Intrinsic uncertainty in human PV modifications Fairly good reliability of corrections provided by different experts (common features) Evidence of model improvement Common expertise better than than individual one.

Future

Within ensemble (Vich et al 2012 in Tellus)Training/tool for sensitivity study (Ricard et al.)

Page 51: Philippe Arbogast, Karine Maynard CNRM-GAME (Météo-France & CNRS)

52

4DVar assimilation instead of inversion