June 16th, 2009

15
June 16th, 2009 Christian Pagé, CERFACS Laurent Terray, CERFACS - URA 1875 Julien Boé, U California Christophe Cassou, CERFACS - URA 1875 Weather typing approach for seasonal forecasts? HEPEX09 - COST731 Workshop - Toulouse - 15-19 June 2009

description

Weather typing approach for seasonal forecasts?. June 16th, 2009. Christian Pagé, CERFACS Laurent Terray, CERFACS - URA 1875 Julien Boé, U California Christophe Cassou, CERFACS - URA 1875. HEPEX09 - COST731 Workshop - Toulouse - 15-19 June 2009. 1. Motivations. - PowerPoint PPT Presentation

Transcript of June 16th, 2009

Page 1: June 16th, 2009

June 16th, 2009

Christian Pagé, CERFACS

Laurent Terray, CERFACS - URA 1875

Julien Boé, U California

Christophe Cassou, CERFACS - URA 1875

Weather typing approachfor seasonal forecasts?

HEPEX09 - COST731 Workshop - Toulouse - 15-19 June 2009

Page 2: June 16th, 2009

1. Motivations

• Difficult to forecast precipitation adequately at long range and at monthly/seasonal timescales

• Even more at higher spatial resolution (hydrological applications)

• Numerical Models and Ensemble Forecast Systems have more abilities to forecast Large-Scale Circulation than fine-scale local variables at these timescales

• Downscaling techniques based on statistical relationships between the Large-Scale Circulation and local scale fields have proven significant abilities in climate sciences (Boe and Terray, 2007)

• Weather-typing approach

• A sort of extended analog methodology with dynamical and local variable constraints

• Can process a large number of simulations, such as large ensemble forecasts systems of atmospheric and/or hydrological models (low CPU cost)

Monthly/Seasonal forecasts applications?

Page 3: June 16th, 2009

3

Downscaling2. Background

Local fields(precipitations, temperature)

Local geographiccharacteristics

(topography, rugosity)Large-ScaleCirculation

Statistical downscaling

Build a statistical model linking the large-scale circulation and local

precipitation

Statistical Downscaling

From Global OR Regional

Models! (e.g. ARPEGE)

Page 4: June 16th, 2009

4

Classification3. Methodology

Daily Mean Sea-Level Pressure

Clusters group #1

Clusters group #2

Cluster composite:

Average of the variable which is

classified withina group

Each cluster is defined by:- its composite- the days’ distribution within the cluster

Classification: main concepts as inBoe and Terray (2007) statistical downscaling methodology

Composite

Composite

Based on Michelangeli et al, 1995

• Precipitation observations are used in the classification learning phase (multi-variate): discriminant

• Temperature (model AND observations) is also used when selecting analog day

• Distances to all clusters (inter-types) are also consideredPictures by Julien Najac, Cerfacs

Page 5: June 16th, 2009

5

Weather types3. Methodology

NCEP MSLP anomalies (hPa) Weather types examples Winter

Methodology produces Weather types discriminant for precipitation

Related precipitation anomaliesfrom Météo-France 8-kmmesoscale analysis SAFRAN (%)

Page 6: June 16th, 2009

6

3. Methodology Validation

Weather types occurrence validation 1950-1999

Page 7: June 16th, 2009

7

3. Methodology Validation

Downscaled NCEP reanalysisvs SAFRAN analysis

Downscaled ARPEGE V4 vsNCEP reanalysis

1981-2005 Validation Period

Annual total mean precipitation 1981-2005Differences in %

Page 8: June 16th, 2009

8

3. Methodology Validation

Precipitation Time Tendencies Validation

=> Seasonal Cumulated Precipitation (NDJFM) reconstructed by multiple regression using weather types occurrence and clusters’ distances

Correlation observation

/reconstruction1900/2000

1 point=1 station, color: latitude=> blue=south, red=north

Time Tendencies Pr

1951-2000 observation

vs reconstruction

Page 9: June 16th, 2009

9

The Météo-France SIM model for hydrological simulations(Habets et al., 2008)

SAFRAN : meteorological parameters: mesoscale analysis at 8-km resolution

ISBA : water flux andground surface energy fluxes

(evaporation, snow,runoff, water infiltration)

MODCOU : hydrological model(river flows)

Dailyriver flows

Latent

Sensible

Snow

Atmosphere

Source: Météo-France

3. Methodology Validation

Habets, F., et al. (2008), The SAFRAN-ISBA-MODCOU hydrometeorological model applied over France, J. Geophys. Res., 113, D06113, doi:10.1029/2007JD008548.

Page 10: June 16th, 2009

10

3. Methodology Validation

River flow Validation using the SIM hydrometeorological model

Winter MeanOBSNCEP (0.85)SAFRAN (0.97)20101960

500

0

• Precipitation and other meteorological variables reconstructed at 8-km using:

• NCEP reanalysis data (Large-Scale Circulation and Temperature) • Statistical downscaling methodology (SAFRAN analysis used for analog daily data)

• Good agreement of downscaled NCEP data vs SAFRAN and observations

SIM simulations by Eric Martin, Météo-France

Page 11: June 16th, 2009

Could this kind of statistical downscaling weather typing methodology be used for Monthly/Seasonal forecasts?

• Predictability of Weather Regimes at Monthly/Seasonal scales

• Very preliminary and exploratory studies have already been done (Chabot et al., 2008, 2009)

• 4 Standard weather regimes, large North Atlantic Domain

• Many questions still to be addressed !

• Weather types

• Are some weather types more predictable than others at monthly/seasonal scale ? Increase in predictability ?

• If yes, what would be the forcings responsible for the most predictable weather types ?

• Which region and large-scale variable(s) to use ? How many weather types to use ?

• Some questions should be explored by doing a hindcast experiment 11

4. Perspectives

Page 12: June 16th, 2009

12

Thanks for your attention!

Christian Pagé, CERFACSChristian Pagé, [email protected]

Laurent Terray, CERFACS - URA1875Laurent Terray, CERFACS - URA1875Julien BoJulien Boé, U Californiaé, U California

Christophe Cassou, CERFACS - URA1875Christophe Cassou, CERFACS - URA1875

HEPEX09 - COST731 Workshop - Toulouse - 15-19 June 2009

Page 13: June 16th, 2009

13

4. Monthly/Seasonal Methodology Facts

BUT! Numerical models have forecasts performances at monthly timescales which are much better than at seasonal timescales(4 weeks lead time)

Ridge

• A previous preliminary and exploratory study (Chabot et al., 2008) showed that:

• Weather regimes predictability at seasonal timescales is low

• Except when strong oceanic forcing (ENSO, Tropical Atlantic)

• This study used:

• Geopotential Height at 500 hPa (Z500) for Large-Scale Circulation classification (tendencies problems)

• A Large North Atlantic Domain

• Four Standard Weather Types Blocking

Page 14: June 16th, 2009

14

4. Monthly/Seasonal Methodology Facts

• A monthly extension to the Chabot et al., 2008 study shows (Chabot et al., 2009) :

• Good predictability for weather types anomaly sign (60 to 80 % of correct forecasts)

Percentage of correct forecasts for the most probable weather type

Percentage of correct forecasts for the least probable weather type

days days30 30

Page 15: June 16th, 2009

15

3. Methodology Validation

Flow Validation

Winter MeanOBSNCEP (0.85)SAFRAN (0.97)

Annual CycleOBSNCEP ARPEGE-VR

CDFOBSNCEP ARPEGE-VR

Jan Dec Jan Dec Jan Dec

0 1 0 1 0 1

ARIEGE (Foix)

ARIEGE (Foix)

LOIRE(Blois)

LOIRE (Blois)

SEINE (Poses)

SEINE (Poses)

VIENNE (Ingrandes)

0

2500

000

0 0

1200

2500250

150 800

20101960

500

0