Forecasting severe weather events, more than 24 h ahead, at METEO FRANCE: An operational trial 11th...

20
Forecasting severe weather events, more than 24 h ahead, at METEO FRANCE: An operational trial 11th EMS/10th ECAM, B. Gillet-Chaulet, DPREVI/PG January 23 2009, 15 H, Meteosat.

Transcript of Forecasting severe weather events, more than 24 h ahead, at METEO FRANCE: An operational trial 11th...

Page 1: Forecasting severe weather events, more than 24 h ahead, at METEO FRANCE: An operational trial 11th EMS/10th ECAM, B. Gillet-Chaulet, DPREVI/PG January.

Forecasting severe weather events, more than 24 h ahead,

at METEO FRANCE:An operational trial

11th EMS/10th ECAM,

B. Gillet-Chaulet,

DPREVI/PG

January 23 2009, 15 H, Meteosat.

Page 2: Forecasting severe weather events, more than 24 h ahead, at METEO FRANCE: An operational trial 11th EMS/10th ECAM, B. Gillet-Chaulet, DPREVI/PG January.

Context and motivations :

Beyond the short range (24 H) at METEO FRANCE No equivalent oper. warning procedures available (not yet…) NWP improvements during recent years

« Predictability » => Express uncertainty

Xynthia windstorm : 3 days in advance !

Page 3: Forecasting severe weather events, more than 24 h ahead, at METEO FRANCE: An operational trial 11th EMS/10th ECAM, B. Gillet-Chaulet, DPREVI/PG January.

Approach and Product :

occurrence of “severe weather events”– Definition :

orange or red level, “Vigilance” procedure, 24 h ahead.

- Violent winds,

- Heavy rains *, (not floods).

- Violent thunderstorms,

- Snow/Ice.

• (not temperatures).

For one entire Day (00 to 24H), D+2, D+3… Zoning (fixed), administrative regions (significant) Risk Index (usual thunderstorms/fogs…)

=> Quantify the risk

Risk Scale:

0, no risk,

1, low,

2, medium,

3, high.

Page 4: Forecasting severe weather events, more than 24 h ahead, at METEO FRANCE: An operational trial 11th EMS/10th ECAM, B. Gillet-Chaulet, DPREVI/PG January.

In the past, (CNP -National Service- : December. 2004, ¼ de France). Reference/”truth” data : “Vigilance” level . Example : How much is a “low” risk (1),

– Over the north-eastern regions, – Parameter – D+2 ?

Approach and Product :

Page 5: Forecasting severe weather events, more than 24 h ahead, at METEO FRANCE: An operational trial 11th EMS/10th ECAM, B. Gillet-Chaulet, DPREVI/PG January.

In the past, (CNP -National Service- : December. 2004, ¼ de France). Reference/”truth” data : “Vigilance” level . Example : How much is a “low” risk (1),

– Over the north-eastern regions, – Parameter – D+2 ?

Approach and Product :

Page 6: Forecasting severe weather events, more than 24 h ahead, at METEO FRANCE: An operational trial 11th EMS/10th ECAM, B. Gillet-Chaulet, DPREVI/PG January.

In the past, (CNP -National Service- : December. 2004, ¼ de France). Reference/”truth” data : “Vigilance” level . Example : How much is a “low” risk (1),

– Over the north-eastern regions, – Parameter – D+2 ?

Approach and Product :

nothing nothing nothing nothing

20 % 1/5 in this example

Page 7: Forecasting severe weather events, more than 24 h ahead, at METEO FRANCE: An operational trial 11th EMS/10th ECAM, B. Gillet-Chaulet, DPREVI/PG January.

D+2, Regions (in total):

3

36

68

87

8

0

10

20

30

40

50

60

70

80

90

100

quasi nul faible moyen élevé

Risque prévu

Fré

qu

en

ce

ob

se

rv

é

Prévision 2005/2010

Fréq. Phén. 2005/2010

Fréq. Obs/Prév

295 103 103 76 13

3504 3092 286 111 15

N Cases

Period:

1st Nov – 1st April

2004/2005

to

2009/2010

(Winter before last)

Approach and Product :

Forecast

No Risk Low Medium High

Climatic

frequency

Page 8: Forecasting severe weather events, more than 24 h ahead, at METEO FRANCE: An operational trial 11th EMS/10th ECAM, B. Gillet-Chaulet, DPREVI/PG January.

3

36

68

87

8

0

10

20

30

40

50

60

70

80

90

100

quasi nul faible moyen élevé

Risque prévu

Fré

qu

en

ce

ob

se

rv

é

Prévision 2005/2010

95%

95%

Fréq. Phén. 2005/2010

95%

95%

Fréq. Obs/Prév

295 103 103 76 13

3504 3092 286 111 15

Rare phenomena:

Significant ?

Period:

1st Nov – 1st April

2004/2005

to

2009/2010

(Winter before last)

Approach and Product :

D+2, Regions (in total):

N Cases

No Risk Low Medium High

Forecast

Climatic

frequency And Feed-back(Low risk ?)

Page 9: Forecasting severe weather events, more than 24 h ahead, at METEO FRANCE: An operational trial 11th EMS/10th ECAM, B. Gillet-Chaulet, DPREVI/PG January.

1

28

70

100

8

0

10

20

30

40

50

60

70

80

90

100

quasi nul faible moyen élevé

Risque prévu

Fré

qu

en

ce

ob

se

rvé

Prévision 2005/2010

95%

95%

Fréq. Phén. 2005/2010

95%

95%

Prév. 2010/2011

Fréq. 2010/2011

Fréq. Obs/Prév

48 5 18 19 6

608 510 65 27 6

Latest Winter :

1st Nov – 1st April

2010/2011

Reliable !

Approach and Product :

D+2, Regions (in total):

N Cases

HighMediumLowNo Risk

Climatic

frequency

Forecast

Page 10: Forecasting severe weather events, more than 24 h ahead, at METEO FRANCE: An operational trial 11th EMS/10th ECAM, B. Gillet-Chaulet, DPREVI/PG January.

Sample size, season, region, range, discrimination between indexes…

“Calibration” (« Dressing ») :

Approach and Product :

No Risk,

Low,

Medium,

High.

No Risk,

Low,

Medium,

High.

Page 11: Forecasting severe weather events, more than 24 h ahead, at METEO FRANCE: An operational trial 11th EMS/10th ECAM, B. Gillet-Chaulet, DPREVI/PG January.

Examples : XynthiaValid for : Sunday 28/02/2010

3 days ahead 2 days ahead

D+3 D+2

No Risk

Low

Medium

High

Calibrated Scale:

Page 12: Forecasting severe weather events, more than 24 h ahead, at METEO FRANCE: An operational trial 11th EMS/10th ECAM, B. Gillet-Chaulet, DPREVI/PG January.

South-East, 7/09/2010Valid for : Tuesday 7/09/2010

3 days ahead 2 days ahead

D+3 D+2

Lyon, daily rainfall

104,1 mm

Absolute Record

(previous 1935)

No Risk

Low

Medium

High

Calibrated Scale:

Page 13: Forecasting severe weather events, more than 24 h ahead, at METEO FRANCE: An operational trial 11th EMS/10th ECAM, B. Gillet-Chaulet, DPREVI/PG January.

Paris under the snow, 8/12/2010Valid for : Wednesday 8/12/2010

3 days ahead 2 days ahead

D+3 D+2

FRANCE - A la mi-journée, la neige continuait de tomber sur la place de la

Concorde à Paris. L'Île-de-France, comme dix-neuf autres départements, ont été placés par Météo France en vigilance

orange neige et verglas ce mercredi. AFP/ Médina DE MIGUEL

No Risk

Low

Medium

High

Calibrated Scale:

Page 14: Forecasting severe weather events, more than 24 h ahead, at METEO FRANCE: An operational trial 11th EMS/10th ECAM, B. Gillet-Chaulet, DPREVI/PG January.

235

108

0

50

100

150

200

250

300

350

400

450

pod

détectés non détectés

275

3494

0

500

1000

1500

2000

2500

3000

3500

4000

far

prévus à tort non prévus

Events point of view (ROC) :

D+2, Regions (in total), Period 2004-2011 (01/11 – 01/04) :

Probability of detection (POD)( Hit Rate) :

Probability of false detection (POFD)(False alarm rate) :

Scale /10 !

69% 7%

No Risk,

Low,

Medium,

High.

Scale

Observed Not Observed

badgood

Page 15: Forecasting severe weather events, more than 24 h ahead, at METEO FRANCE: An operational trial 11th EMS/10th ECAM, B. Gillet-Chaulet, DPREVI/PG January.

235

108

0

50

100

150

200

250

300

350

400

450

pod

détectés non détectés

275

3494

0

500

1000

1500

2000

2500

3000

3500

4000

far

prévus à tort non prévus

19

324

0

50

100

150

200

250

300

350

400

450

pod

détectés non détectés

2

3767

0

500

1000

1500

2000

2500

3000

3500

4000

far

prévus à tort non prévus

69% 7%

6% 0, %

AROC = 0.81 ( 0.84 , 0.79 )

Events point of view (ROC) :

D+2, Regions (in total), Period 2004-2011 (01/11 – 01/04) :

Probability of detection (POD)( Hit Rate) :

Probability of false detection (POFD)(False alarm rate) :

Observed Not ObservedScale /10 !

Scale

No Risk,

Low,

Medium,

High.

Scale

No Risk,

Low,

Medium,

High.

goodbad

Page 16: Forecasting severe weather events, more than 24 h ahead, at METEO FRANCE: An operational trial 11th EMS/10th ECAM, B. Gillet-Chaulet, DPREVI/PG January.

Conclusion :

Thanks to recent improvements (research, operational…)

Signal at D+2, D+3, and beyond ! (7 Day test…)(Non deterministic)

Proposal of synthesis… -> Since 1st of March 2010, daily issue (14 H 30 local time) :

– COGIC, SCHAPI, CNIR, CRIR, INVS (feed back ?)

CNP production (regions collaboration), Available on the internal Website

http://previ-gene/ Similar products already available on

several Websites…(MET OFFICE)

Page 17: Forecasting severe weather events, more than 24 h ahead, at METEO FRANCE: An operational trial 11th EMS/10th ECAM, B. Gillet-Chaulet, DPREVI/PG January.

Thank you for your attention

Page 18: Forecasting severe weather events, more than 24 h ahead, at METEO FRANCE: An operational trial 11th EMS/10th ECAM, B. Gillet-Chaulet, DPREVI/PG January.

One

Context

back

Page 19: Forecasting severe weather events, more than 24 h ahead, at METEO FRANCE: An operational trial 11th EMS/10th ECAM, B. Gillet-Chaulet, DPREVI/PG January.

Title : Forecasting severe weather events, more than 24 hours ahead, at METEO FRANCE:

An operational trial.

For the short range, typically 24 hours ahead, severe weather forecast procedures have been in effect for a long time. The French “vigilance” watch map celebrates its tenth anniversary in 2011. It puts in concrete form the first mission of METEO FRANCE in the domain of protection of people and goods, and has proved generally successful. Improvements in numerical weather prediction during recent years, now enable a focus on forecasting dangerous weather phenomena at longer ranges (e.g. within the Medium Range) and beyond the requirements for triggering a “vigilance” warning. However, this can not be just a simple extension of the validity of the vigilance watch map, because uncertainty generally quickly increases with the forecast range and makes a deterministic approach difficult.

A systematic production, experienced on a daily basis since December 2004 at the national forecasting service of METEO FRANCE has tried to forecast the risk of occurrence of dangerous phenomena (violent winds, heavy rain, violent thunderstorms, snow/ice) that might reach orange or red warning levels of “vigilance”, several days ahead.

Every morning, a risk index has been assessed by the forecasters, based on their study of deterministic models, numerous EPS products, for a given day from D+2 until D+7, and over given geographical areas of France. The index has been selected from no risk, unlikely, likely, certain. After choosing a reference or “truth” data (the colour of “vigilance” actually issued at the considered scale), accuracy and skill of the forecast has been shown. Then the idea has been to provide, in real time, the probabilities corresponding to the reliability of the chosen index taking into account the phenomenon, the samples representativeness and the discrimination between indexes.

Since March 2010, encouraging results have allowed us to propose institutional partners of METEO FRANCE a forecast of the risk of occurrence of these phenomena in terms of probabilities at the scale of the administrative French regions for D+2 and D+3.

First feed-back from the users is interesting, but shows that probabilities must be correctly interpreted. In fact, one still thinks that a probability of 50% corresponds to the toss of a coin, whereas it is significantly high when the climatological frequency of the phenomenon is low. Furthermore, decision makers have to learn how to make best use of this new type of information.

 

Abstract:

Page 20: Forecasting severe weather events, more than 24 h ahead, at METEO FRANCE: An operational trial 11th EMS/10th ECAM, B. Gillet-Chaulet, DPREVI/PG January.

Résumé :

Ce travail concerne la prévision des phénomènes météorologiques présentant un risque pour la sécurité des personnes et des biens. A vingt-quatre heures d’échéances, des procédures d’avertissement existent depuis longtemps : la « Vigilance » de Météo France fête ses dix ans d’existence en 2011. Les progrès notables réalisés en prévision ces dernières années permettent d’anticiper ces phénomènes à des échéances plus lointaines. Cependant, leur faible prévisibilité empêche une approche déterministe. Une production systématique, quotidienne, expérimentée depuis décembre 2004 à DPREVI/PG, s’attache à prévoir le risque d’occurrence des phénomènes « Vent violent, Pluie-innondation, Orages, Neige-verglas » susceptibles de relever des niveaux de vigilance orange ou rouge, à plusieurs jours d’échéance. En mars 2010, les résultats encourageants ont permis de proposer aux partenaires institutionnels de Météo France une prévision du risque d’occurrence de ces phénomènes, assortie d’une probabilité calibrée, à l’échelle des régions économiques françaises, pour les journées J+2 et J+3.