Diapositive 1

1
Some examples of the use of ECMWF products at
Météo-France

• Medium-range forecasts in tropical areas
• Evolution of the medium range forecasts in
  metropolitan France
• Calibration of the EPS-based probabilities

ECMWF users meeting, 14-16 juin 2006
2

• Medium-range forecasts in tropical areas
• Evolution of the medium range forecasts in
  metropolitan France
• Calibration of the EPS-based probabilities

ECMWF users meeting, 14-16 juin 2006
3
Medium range forecasts in tropical areas
ECMWF users meeting, 14-16 juin 2006
4
Medium range forecasts in tropical areas
New-Caledonia Medium-range forecast fully
operationnal since january 2006, On mondays and
fridays, forecast for D4-D7, available on
http//www.meteo.nc As Thierry Lefort
presented here 2 years ago, the forecast is based
on weather patterns (see also ECMWF Newsletter
N102).
ECMWF users meeting, 14-16 juin 2006
5
Medium range forecasts in tropical areas
New-Caledonia Case of tropical cylone JIM
T511 model from 01/23/2006 12h
168h
180h
central pressure 956 hPa
central pressure 965 hPa
ECMWF users meeting, 14-16 juin 2006
6
Medium range forecasts in tropical areas
New-Caledonia Case of tropical cylone JIM
The medium-range bulletin from 23 january
mentioned the risk
 Type de temps le régime d'alizé de vendredi
va vite céder la place à une nouvelle offensive
de l'air tropical venu du nord. Cette situation
est favorable à la formation de dépression
tropicale en mer de Corail. 
 Weather regime The trade winds of Friday will
rapidly let the place to a new offensive of
northern tropical air. This situation is
favourable to the genesis of a tropical
depression in the Coral Sea. 
ECMWF users meeting, 14-16 juin 2006
7
Medium range forecasts in tropical areas
New-Caledonia Case of tropical cylone JIM
The EPS-based products let us anticipate the
arrival of the cyclone
Probability MSLP lt 1000 hPa
Position of sea level pressure minima (blue lt
995 hPa, red lt 990 hPa)
ECMWF users meeting, 14-16 juin 2006
8
Medium range forecasts in tropical areas
New-Caledonia Case of tropical cylone JIM
A gale warning has been issued on 26 and the
system has been upgraded to tropical cyclone JIM
on the 28 january. observed trajectory
ECMWF users meeting, 14-16 juin 2006
9
Medium range forecasts in tropical areas
New-Caledonia Case of tropical cylone JIM
Comparison T511 / T799 from 2006/01/27 0h
T511 T799 OBS
ECMWF users meeting, 14-16 juin 2006
10
Medium range forecasts in tropical areas
New-Caledonia promoting EPS use
ECMWF users meeting, 14-16 juin 2006
11
Medium range forecasts in tropical areas

• Polynesia special assitance in Bora-Bora
• The goal of the operation install a pipeline
  near Bora-Bora
• Weather conditions needed
• no waves (height lt 1,5 m) ,
• no wind (windspeed lt 4 m/s), during 48 hours
• Difficulty a 6-day journey is necessary to the
  special boat managing the installation !

ECMWF users meeting, 14-16 juin 2006
12
Medium range forecasts in tropical areas

• Polynesia special assitance in Bora-Bora
• step 1 study of the climatological frequency of
  such weather conditions
• 4 if looking at the windspeed observed in
  Bora-Bora
• one time during february 2005 if looking at
  Christmas buoy observations
• two times during october 2004 if looking at
  ECMWF wave analysis

22 october 2005 18h
ECMWF users meeting, 14-16 juin 2006
13
Medium range forecasts in tropical areas
Polynesia special assitance in Bora-Bora step 2
development of EPS-based products and
experiencing EPS
wave  rose 
wave height EPSgram
ECMWF users meeting, 14-16 juin 2006
14
Medium range forecasts in tropical areas
Polynesia special assitance in Bora-Bora step 3
production of a daily medium-range forecast
dedicated to this operation
ECMWF users meeting, 14-16 juin 2006
15
Medium range forecasts in tropical areas

• Polynesia special assitance in Bora-Bora
• The operation finally took place on 15 and 16
  march
• despite an increase of the windspeed in the
  afternoon of the 16th, the installation succeeded
  !
• Very positive consequences
• better comprehension of wave model and ensemble
• exchange with ECMWF and Météo-France specialists
• first experimentation of medium-range forecast
• since may 2006, a bulletin is produced once a
  week !

ECMWF users meeting, 14-16 juin 2006
16
Medium range forecasts in tropical areas

• General comments about tropics
• more verification products on tropical areas
  (on web pages)
• more wave observations/altimeter data/analyses
  (on web pages)
• waves can be very dangerous (even with quite low
  heights 3m) floods, destructions of houses
  near the coasts, transport of coral on aerodroms
  and roads, strong streams in the lagoons

ECMWF users meeting, 14-16 juin 2006
17

• Medium-range forecasts in tropical areas
• Evolution of the medium range forecasts in
  metropolitan France
• Calibration of the EPS-based probabilities

ECMWF users meeting, 14-16 juin 2006
18
Evolution of the medium-range forecast in
metropolitan France

• Motivations
• improvements of the models
• request from media service (more probabilistic
  products)
• request from specialized marine forecasters

ECMWF users meeting, 14-16 juin 2006
19
Evolution of the medium-range forecast in
metropolitan France

• Proposition 1
• extend the forecast range to D8/D9 short
  text cursor

Strong signal for dry/sunny weather
Strong signal for precipitations
No signal
In that case, the forecaster indicates a
important risk of precipitation
ECMWF users meeting, 14-16 juin 2006
20
Evolution of the medium-range forecast in
metropolitan France

• Proposition 2
• provide probabilistic products with forecasters
  comments

Example Probability Tmax gt 25C The forecaster
will indicate which plot is pertinent and add a
specific comment
ECMWF users meeting, 14-16 juin 2006
21
Evolution of the medium-range forecast in
metropolitan France

• Proposition 3
• transmit the medium-range forecasters expertise
  to the marine forecaster, with dedicated products

Wave height probabilities with Douglas
thresholds Ex Prob SWHgt2.5m EPS from 20060110
12h 168h
ECMWF users meeting, 14-16 juin 2006
22
Evolution of the medium-range forecast in
metropolitan France

• Perspectives
• extend the forecast range to D15 with the
  VAREPS
• training and linking with monthly forecasts

ECMWF users meeting, 14-16 juin 2006
23
Evolution of the medium-range forecast in
metropolitan France

• Monthly forecasts
• since january 2006, a bulletin made by a
  forecaster, is provided to a specific user.

ECMWF users meeting, 14-16 juin 2006
24
Evolution of the medium-range forecast in
metropolitan France

• Monthly forecasts
• the foracasters have assessed the 2m-temperature
  anomaly, since january 2006 (20 cases)

week 1
week 2
week 3
week 4
ECMWF users meeting, 14-16 juin 2006
25

• Medium-range forecasts in tropical areas
• Evolution of the medium range forecasts in
  metropolitan France
• Calibration of the EPS-based probabilities

ECMWF users meeting, 14-16 juin 2006
26
Calibration of the EPS based probabilities
The final medium-range forecasts on specific
points (cities) are computed with EPS
statistical adaptation. The statistical
adaptations (multi-linear regression, Kalman
filter) correct the local biases. Some problems
of reliability or lack of spread remains after
these computations the calibration tries to
correct them.
ECMWF users meeting, 14-16 juin 2006
27
Calibration of the EPS based probabilities
At this time, the method based on rank histograms
(described by Hamill and Colucci) is used at
Météo-France but the improvement is not important
(all the points are mixed to compute the
histograms). The calibration method called BMA
(Bayesian Model Averaging, described by A. E.
Raftery http//www.stat.washington.edu/raftery/Res
earch/Bayes/bayes_papers.html ) is beeing studied
at Météo-France, in order to realize a local
calibration.
ECMWF users meeting, 14-16 juin 2006
28
Calibration of the EPS based probabilities
Principle 1 - every EPS member is considered
as a PDF (Probability Density Function), The
EPS distribution is then represented with the
combination of the individual PDFs 2 - the BMA
searches for the combination of the individual
PDFs that will maximize a likelihood problem.
Ex 5 members ensemble, parametertemperature
threeshold S Prob T gt S 20
Prob T gt S area A
ECMWF users meeting, 14-16 juin 2006
29
Calibration of the EPS based probabilities

• Concrete example study on 2m temperature
• the individual PDFs are assumed to be gaussian
• the standard deviation of all the individual
  PDFs is the same
• the combination of the individual PDFs is a
  simple sum (ponderation 1/N)
• The only variable to compute is the standard
  deviation of the inidividual PDFs.
• To do it, the EM (Expectation Maximization) is
  used
• maximisation of L values on the sample data

Forecasted distribution
L
obs
ECMWF users meeting, 14-16 juin 2006
30
Calibration of the EPS based probabilities
Results on min/max 2m temperature (26 stations, 6
months)
Percentage of observations inside the EPS range
D D1 D2 D3 D4 D5 D6
D7 D8
ECMWF users meeting, 14-16 juin 2006
31
Calibration of the EPS based probabilities
Results on 2m temperature (26 stations, 6 months)
EPS range (max value min value)
D D1 D2 D3 D4 D5 D6
D7 D8
ECMWF users meeting, 14-16 juin 2006
32
Calibration of the EPS based probabilities

• Current work on precipitations
• non- gaussian parameter
• limited parameter
• large frequency of zero-precipitation cases
• at this time no statistical adaptation is made
  on precipitations

ECMWF users meeting, 14-16 juin 2006
33
Calibration of the EPS based probabilities

• Current work on precipitations method beeing
  tested
• 1- discriminant analysis to separate cases
  precipitations/no-precipitations
• 2- barycentric regression applied on the
  precipitations cases, in order to correct biases
• 3- BMA applied on precipitations cases
• - individual PDFs are gamma distributions
• - the unknown variable is the shape parameter of
  gamma distribution
• - EM method is also used to find this value

ECMWF users meeting, 14-16 juin 2006
34
Calibration of the EPS based probabilities
Current work on precipitations first results
Example of calibrated PDF
Cumulative Distribution Function
ECMWF users meeting, 14-16 juin 2006
35
Calibration of the EPS based probabilities
Current work on precipitations first results
Prob RR60 t48h station Bordeaux, period
2 years
Direct output
calibrated
ECMWF users meeting, 14-16 juin 2006
36
Calibration of the EPS based probabilities
Current work on precipitations first results
Prob RR6gt1 mm t48h station Bordeaux, period
2 years
Direct output
calibrated
ECMWF users meeting, 14-16 juin 2006
37
Calibration of the EPS based probabilities

• Next steps
• implementation of BMA in operations
• Continue the current work on precipitations
• Study on windspeed

ECMWF users meeting, 14-16 juin 2006
38
Acknowledgments
ECMWF Jean Bidlot EDF Clarisse Fil-Tardieu,
Laurent Dubus Metéo-France Serge Lepape,
Thierry Lefort, Serge Farges,
Jean-Michel Lefebvre,
Jean-Marc Jacquin and many
forecasters Students from Toulouse university
Marie Paya, Carole Destruel
ECMWF users meeting, 14-16 juin 2006