Download - Ensemble Forecasting at RSMC La Réunion
Ensemble Forecasting at RSMC La Réunion
WMO/WWRP Training, Nanjing, dec 2011
Matthieu Plu, Thierry Dupont, Sébastien Langlade, Nicole Girardot
METEO-FRANCE
RSMC La Réunion
Laboratoire de l’Atmosphère et des Cyclones
Outline
Presentation of RSMC La Réunion Available products from ensemble forecasts
– Météo-France data
– ECMWF data
– TIGGE data Track forecasts Intensity forecasts Wave forecasts Storm surge SWFDP RSMC La Réunion website
RSMC La Réunion
Since 1993, Météo-France La Réunion: WMO RSMC for tropical cyclones in the South-West Indian Ocean (SWIO)
RSMC La Réunion
Cyclone activity in the SWIO :– 12% of the total worldwide cyclone activity
– Impacts : many human losses every year (particularly in Madagascar), and significant economical impact in the richer countries (example : Dina 2002 95 M€ losses in La Réunion)
– Severe winds and intense rainfall (La Réunion records), waves and storm surge (African coast and atolls)
Role of RSMC La Réunion for tropical cyclones :– Provide appropriate guidance information (analyses, forecasts) in real-
time ; forecast term until 5 days
– Climatology, database of TC
– Training and R&D activities
RSMC La Réunion
RSMC La Réunion
… but individual tracks are not so regular …
RSMC La Réunion
Equatorial waves (Kelvin waves, Rossby waves, Madden-Julian oscillation, …) modulate cyclone activity.
Mid-latitude events also play a significant role on cyclone dynamics (track and intensity)
Density of occurrence of a mid-latitude Rossby-wave breakings at 200hPa during the rapid Intensification of a cyclone
RSMC La Réunion
Track error of deterministic models in the SWIO:
Direct position error(2009-2010 season)
Available products from ensemble forecasts
Météo-France PEARP
Available products from ensemble forecasts
ECMWF ensemble prediction system (EPS)
Available products from ensemble forecasts
Other TIGGE products
– MOGREPS (UKMO) : cxml
Available products from ensemble forecasts
Other TIGGE products (cxml)
– GEFS (NCEP) : cxml + wind and precipitation
Available products from ensemble forecasts
Other TIGGE products (cxml)
– CENS (CMC) : cxml + wind and precipitation
TC track forecasts
General method for track prediction– Step 1 : Build the official RSMC track forecast
– This step relies on deterministic models and on « consensus » of several models
Ensemble tracks may help to eliminate from the consensus the outlying deterministic tracks
TC track forecasts
General method for track prediction
– Step 2 : Writing the bulletin
– Ensemble forecasts are used to :
• Provide some confidence of the forecast
• Give some possible alternative scenarios
TC track forecasts
Illustration on the case of Bingiza
Tropical cyclone Bingiza 13 February 2011 - 07UTC (Terra image – source : http://rapidfire.sci.gsfc.nasa.gov)
TC track forecasts
Illustration on the case of Bingiza: early prediction of impact zone
Bingiza, 2011-02-10 06UTCModerate Tropical Storm
TC track forecasts
Illustration on the case of Bingiza: early prediction of impact zone
Bingiza, 2011-02-10 06UTCModerate Tropical Storm
TC track forecasts
Illustration on the case of Bingiza: early prediction of impact zone
Bingiza, 2011-02-10 06UTCModerate Tropical Storm
PEARP
EPS
TC track forecasts
Illustration on the case of Bingiza: early prediction of impact zone
Bingiza, 2011-02-10 06UTCModerate Tropical Storm
TC track forecasts
Illustration on the case of Bingiza: early prediction of impact zone
– Most of the models and ensembles indicate movement towards the South
– The spread of ensemble tracks is large
– BULLETIN : 2011-02-10 00UTC […] The track forecasts by numerical models are highly spread. There is some agreement that the storm movement may be somehow erratic during the next 48h. Then, some models suggest a track towards the South-West towards Madagascar, some others towards the South. The inhabitants of Mauritius, La Réunion and of Madagascar should follow the evolution of this system.
– Not a precise warning
TC track forecasts
Illustration on the case of Bingiza: short-term prediction of landfall
Bingiza, 2011-02-12 12UTCTropical cyclone
TC track forecasts
Illustration on the case of Bingiza: short-term prediction of landfall
Bingiza, 2011-02-12 12UTCTropical cyclone
TC track forecasts
Illustration on the case of Bingiza: short-term prediction of landfall
TC track forecasts
Illustration on the case of Bingiza: short-term prediction of landfall
– The ensembles confirm the RSMC track scenario
– BULLETIN : 2011/02/12 06UTC […] Models indicate a direction towards the East coast of Madagascar and landfall between Masaola and Sainte Marie Island on Monday. The environmental conditions are favorable to intensification before landfall. The inhabitants of the East coast of Madagascar must follow with great attention the evolution of Bingiza.
– 48h before landfall, the forecast is:• Precise: day of landfall given, location (~150km)• Right: this forecast has been verified
– Ensembles provided confidence in this forecast
TC track forecasts
• Subjective evaluation of ensemble systems to predict landfall:
• Position
• Timing
Black: 0-24h
Purple: 24h-48h
Red: 48h-72h
Orange: 72h-96h
Yellow: 96h-120h
• Existence of the observed track in the ensemble tracks:
• as a majority
• as a minority
• not at allEPS
12/02 – 12UTC
TC track forecasts
LANDFALL AT MASAOLA ?
LANDFALL AT 02-14 AT 02UTC ?
REALISTIC TRACKS AT MEDIUM RANGE ?
REALISTIC TRACKS AT LONG RANGE ?
02-0918UTC
02-1018UTC
02-1118UTC
02-1218UTC
02-1318UTC
02-1006UTC
02-1106UTC
02-1206UTC
02-1306UTC
02-1406UTC
CENS EPS GEFS
MOGREPS PEARP Poor Man Ensemble
TC track forecasts
TC track forecasts
Conclusion on the case of Bingiza:
– A difficult case for track forecast
– At early stage, ensembles do not do better than deterministic models for predicting the impact areas
– 48h before landfall, ensembles provide precise and good information
– At every stage, ensembles are useful to quantify the confidence of the forecast:
• early stage: low confidence• 48h before landfall: high confidence• Such information is conveyed in the RSMC bulletins.
Multi-ensembleAll forecasts
TC track forecasts
Some other exemples :
– Cherono (Tropical storm 2011)
Multi-ensemble2011-03-17
TC track forecasts
Some other exemples :
– Yasi (TC, Australia 2011)
TC track forecasts
Some other exemples :
– Yasi (TC, Australia 2011)
TC intensity forecasts
Intensity forecasts: a difficult task!
Intensity&
Structure changeof the cyclone
Davidson et al. (2008)
Hanley et al. (2001)
Gray (1979), Emanuel (1986)
Montgomery (2008)
Nguyen et al. (2006)
Wang (2002)
Schubert (1999)
Initial structure
External forcing - Atmosphere
- Océan
Internal dynamics
Fujita (1952)
Chan and Williams (1987)
Ma et al. (2011)
Leroux et al. (2012)
TC intensity forecasts
Intensity forecasts: a difficult task! General forecasting method:
– Interpretation of the internal structure and the environment (favorable/unfavorable) from model analyses and satellite images;
– Deterministic models indicate a possible trend for the evolution of intensity;
– Ensembles have not been used for intensity forecasts yet;
– Exemples for « climatological » and « rapid » intensification.
TC intensity forecasts
TC Gelane (2010): a « rapid intensification ».
INT
EN
SIT
Y (
hP
a) Before
Rapid intensification
TC intensity forecasts
TC Gelane (2010) before rapid intensification (02/18 - 00UTC):
– All deterministic models predict no intensification
– RSMC bulletin:• 12h-24h Strong Tropical Storm• 36h-60h Moderate Tropical Storm• 72h Tropical Depression
– Verification :• 12h-24h Tropical Cyclone• 36h Intense Tropical Cyclone !!
TC intensity forecasts
TC Gelane (2010) before rapid intensification (02/18 - 00UTC):
– Available ECMWF ensemble products :
Observation: 930hPa !!980
TC intensity forecasts
TC Gelane (2010)
INT
EN
SIT
Y (
hP
a)
Before Rapid filling
TC intensity forecasts
TC Gelane (2010) before rapid filling (02/19 - 12UTC):
– All the deterministic models predicted Gelane to encounter strong vertical wind shear rapid filling
– The difference between the models when filling will occur, depending on the track forecast
Deterministic models
EPS strike probabilities
STRONG SHEAR
TC intensity forecasts
TC Gelane (2010) before rapid filling (02/19 - 12UTC):
– Available ECMWF ensemble products :
Observation: 930hPa !!
TC intensity forecasts
Hurricane Irene (North Atlantic, 2011) : a « climatological » intensification.
Does the ensemble bring additional information with regard to the deterministic forecast?
TC intensity forecasts
Global models fail to represent correctly intensity evolutions, Lagrangian EPSgrams do not bring useful information yet, Some hope with resolution increase and with new products ?
TC intensity forecasts
Global models fail to represent correctly intensity evolutions, Lagrangian EPSgrams do not bring useful information yet, Some hope with resolution increase and with new products:
Products in test (not operational)
High wave forecasts
Two ECMWF products are used:– EPSgrams– Probability maps
High wave forecasts
Two ECMWF products are used:– EPSgrams
High wave forecasts
Two ECMWF products are used:– Probability maps
Ech 24 h Ech 48 h Ech 72 h
Storm surge forecasts
Storm surge highly depends on :
– Track
– Intensity
– Surface wind structure
R&D activities have begun to forecast storm surge:
– Prediction of wind extensions (radius of maximum wind, other thresholds…),
– Generate a « climatology » of storm surge in the SWIO from a surge model and a « climatology » of cyclones
SWFDP Website of RSMC La Réunion
• Purpose :– Provide guidance in real-time regarding tropical cyclones, and extreme
wind and rainfall– A limited-access to the SWFDP and RA I national meteorological
services– Available in two languages: French & English
Input data :– Aladin-Reunion forecast fields– For each cyclone, the official RSMC track forecast– Deterministic models (Météo-France, ECMWF, UKMO) – Forecast tracks from ensembles (PEArp, ECMWF EPS, NCEP, CMC,
MOGREPS) and multi-ensemble– Fields (mslp, wind, rainfall) from ensembles (PEArp, ECMWF EPS,
NCEP, CMC) and multi-ensemble
SWFDP Website of RSMC La Réunion
Permanent products:– Aladin-Réunion forecasts (00UTC and 12UTC, until 84h)
– Probabilities and quantiles of strong winds and intense rainfall for 4 ensembles + multi-ensemble
– mslp spaghettis forecasts (for cyclogenesis) for 4 ensembles + multi-ensemble
In case of a cyclone in the basin:– Official RSMC track forecast + uncertainty cone
– Track forecasts from deterministic models
– Tracks, live time, Strike probabilities, track for the 5 ensembles + multi-ensemble
SWFDP Website
2. Website of RSMC La Réunion
Early detection of the cyclogenesis of Bingiza
Illustrations sur le cas de Bingiza (février 2011)
Early detection of the cyclogenesis of Bingiza
Illustrations sur le cas de Bingiza (février 2011)
Early detection of the cyclogenesis of Bingiza
Illustrations sur le cas de Bingiza (février 2011)
Early detection of the cyclogenesis of Bingiza
Illustrations sur le cas de Bingiza (février 2011)
Early detection of the cyclogenesis of Bingiza
Illustrations sur le cas de Bingiza (février 2011)
Early detection of the cyclogenesis of Bingiza
Illustrations sur le cas de Bingiza (février 2011)
Early detection of the cyclogenesis of Bingiza
Illustrations sur le cas de Bingiza (février 2011)
Early detection of the cyclogenesis of Bingiza
Website of RSMC La Réunion
Official RSMC forecastUncertainty cone (derived from EPS)
Track forecasts from deterministic models
SWFDP website of RSMC La Réunion
Regions where strong winds and heavy rainfall may occur are indicated in RSMC bulletins before landfall
More diagnoses are available in the SWFDP/RSMC La Réunion website Probabilities of attaining a threshold :
– Wind: 28kts, 48kts, 64kts, 90kts
– Rainfall: 5mm/6h, 10mm/6h, 20mm/6h, 50mm/6h, 5mm/24h, 20mm/24h, 50mm/24h, 100mm/24h
Website of RSMC La Réunion
Official RSMC forecastUncertainty cone (derived from EPS)
Track forecasts from deterministic models
Bingiza, before landfall
Website of RSMC La Réunion
Bingiza, before landfall
Website of RSMC La Réunion
Bingiza, before landfall
Bingiza, before landfall
Website of RSMC La Réunion
Bingiza, before landfall
SWFDP website of RSMC La Réunion
Quantiles 25%, 50%, 75%, 90%:
~ a realistic extreme value
the median value
90%
50%
75%
25%
SWFDP website of RSMC La Réunion
the median value
a realistic extreme value
Conclusion
Ensemble prediction products are now a major input at RSMC La Réunion, especially for:
– Track prediction– Impacts: waves, rainfall, winds
Two main paths for using ensembles:– Confirm the RSMC scenario, or propose alternative scenarios– Convey some information about uncertainty in the bulletins
Future R&D activities:– Comparison of the ensemble track forecasts from different ensembles,
and from multi-ensemble
– Calibration of the ensemble forecasts for wind and precipitation
– A major difficulty: verification and calibration of ensemble forecasts where the observations are sparse