Slide 1 International Typhoon Workshop Tokyo 2009 Slide 1 Impact of increased satellite data density in sensitive areas Carla Cardinali, Peter Bauer, Roberto

International Typhoon Workshop Tokyo 2009

Slide 1

Slide 1

Impact of increased satellite data density

in sensitive areas

Carla Cardinali, Peter Bauer, Roberto Buizza, Jean-Noël ThépautECMWF

Florian Harnisch and Martin WeissmannDLR

Many thanks to Fernando Prates


Slide 2

Slide 2

Background• Thinning of data is applied to:

－ reduce data volume－ avoid the introduction of spatial observation error correlation that is

currently not accounted for in data assimilation algorithm • Thinning is performed statically on a fixed latitude/longitude grid.

Objective Evaluate impact of selective satellite observational data thinning on medium-range

NWP aiming at denser data in sensitive areas and less dense data in other areas trade-off between data impact and data volume.

Approach• Experiments with global data thinning:

－ Change global latitude/longitude thinning grid.• Experiments with data thinning in selected regions:

－ Increased density in sensitive areas and reduced density elsewhere using a Singular Vector based measure to identify areas from which forecast errors are growing fast (ECMWF 2007 QJ papers).

－ Sensitive areas are computed for Southern Hemisphere

Study contents


Slide 3

Slide 3

• More dense satellite data coverage on SV-areas in the Southern Hemisphere

• More dense data coverage in SV-areas in Typhoon areas

• Typhoon Track Forecast impact assessment

• Forecast sensitivity to evaluate the 24-hour forecast impact

Outline


Slide 4

Slide 4

Southern Hemisphere Experiments

Selective data thinning• Thinn_Cntrl : ~ is 1.25o proxy for Thinn_1.25• Thinn_SV : ~ is 1.25o and 0.625o in SV areas.• Thinn_RD : ~ is 1.25o and 0.625o in randomly distributed areas.• Thinn_CSV : ~ is 1.25o and 0.625o in Climatological SV areas.• Thinn_0.625

Additional information• All experiments are run at T511L91 (12-hour 4D-Var) for 01/12/2008-

28/02/2009.• All experiments are verified with T799L91 operational model analyses (without

first 7 days (spin-up) i.e. 83 cases).• All SV/RD/CSV areas occupy same fraction (15%) of the Southern hemisphere.• The SV-based climatology derived from the mean 2007 SV-areas.


Slide 5

Slide 5

Data coverage: Single case01/01/2009 00 UTCdata density AMSU-A channel 9:

Singular Vectors:

Randomly distributed circular areas:

2007 Singular Vector climatology:


Slide 6

Slide 6

Data coverage: Average01-07/01/2009 00 and 12 UTC data density AMSU-A channel 9:

Singular Vectors:

Randomly distributed circular areas:

2007 Singular Vector climatology:


Slide 7

Slide 7

Selective data thinning: DFSDecrease of DFS relative to the Thin_0.625 experiment

0 10 20 30 40 50 60 70 80 90 100

Thin_0.6

SV

RD

C_sv

Thin_1.2

DFS (%)0 10 20 30 40 50 60 70 80 90 100

Thin_0.6

SV

RD

C_sv

Thin_1.2

DFS (%)

Global Southern Hemisphere


Slide 8

Slide 8

Selective data thinning: Forecast impact SV-CNTRLSouthern H. Normalized RMSE 95% confidence 83 cases

1000 hPa

500 hPa

200 hPa

0 1 2 3 4 5 6 7 8 Forecast Day


Slide 9

Slide 9

Selective data thinning: Forecast impact SV-RDSouthern H. Normalized RMSE 95% confidence 83 cases

1000 hPa

500 hPa

200 hPa

0 1 2 3 4 5 6 7 8 Forecast Day


Slide 10

Slide 10

Selective data thinning: Forecast impact SV - CSVSouthern H. Normalized RMSE 95% confidence 83 cases

1000 hPa

500 hPa

200 hPa

0 1 2 3 4 5 6 7 8 Forecast Day


Slide 11

Slide 11

Sinlaku: Track forecast between 00 UTC 09 - 19 Sept.last forecast verification time 12 UTC 20 Sept.(classified as extra-tropical in best track data from 00 UTC 21 Sept)

Hagupit: track forecast between 00 UTC 20 - 24 Sept.last forecast verification time 00 UTC 25 Sept.(dispersing over land, tropical depression from 00 UTC 25 Sept)

Jangmi: track forecast between 00 UTC 25 - 29 Sept.last forecast verification time 12 UTC 30 Sept.(classified as extra-tropical in best track data from 00 UTC 01 Oct)

Typhoons TPARK campaign Summer 2008


Slide 12

Slide 12

Targeting Typhoon season with extra-satellite data

Selective data thinning experiments

• Cntrl : 1.25o Global • SV-Sat: 1.25o Global and 0.625o in SV areas.• Drop : 1.25o Global +Targeted Dropsondes• SV-Sat-Drop: Targeted Dropsondes+ SV areas 0.625o

Additional information

• All experiments are run at T799TL95/159/255 L91 (12-hour 4D-Var) • 06-30 September 2008• Verification and SV-target region 10-50N, 110-180E• 20 Leading T95L62 SV• SVs area cover 20% of the target region


Slide 13

Slide 13

Targeting Typhoon season with extra-satellite data: SV-areas


Slide 14

Slide 14

09 + 1011 Sept

SV-S

at +

Dro

p

cntrl

Sinlaku 09-19 September: mean track error km

Dro

p

cntr

cntrl

SV -S

at


Slide 15

Slide 15

intensification

1

2

3


Slide 16

Slide 16

CNTRL very accurate track forecast : difficult to improve

Hagupit 20-24 September


Slide 17

Slide 17

cntrl

SV-S

at +

Dro

pSV

-Sat

cntrl

Dro

p

cntrl

Hagupit 20-24 September


Slide 18

Slide 18

Difficult to determine TC position over land

Jangmi 25-27 September


Slide 19

Slide 19

cntrl

SV-S

at+D

rop

Dro

p

cntrl

SV-S

at

cntrl

Jangmi 25-27 September


Slide 20

Slide 20

Forecast sensitivity to observation

The tool provides information on the observation type, subtype, variable and level responsible for the forecast error variation

, ,a

a

J JJ

xy y

y y x

Forecast error

KT

Observation departure


Slide 21

Slide 21

Forecast Sensitivity to Obs: SV-Sat+Drop


Slide 22

Slide 22

Forecast Sensitivity to Obs: SV-Sat+Drop

Forecast error andVerifying analysis


Slide 23

Slide 23

ConclusionsSelective data thinning• Forecast scores are best for experiment with increased data density in SV-based

areas that are updated for each analysis.• 40% loss of DFS by increasing the data density over SV areas instead than

globally.

Targeting Typhoon with extra satellite data• Limited statistical sample• Extra-satellite data gave a more consistent impact due to homogeneous coverage

and data diversity (moist, temperature, cloud, precipitation and surface wind)

Forecast Sensitivity To Observation (FSO)• The forecast value per Observation shows that dropsondes are more beneficial

that extra-radiances• Strong impact per dropsonde produces more extreme beneficial/detrimental

impact• Computation of forecast error by using observation instead of analysis field is

likely to shows larger dropsonde impact on typhoon.


Slide 24

Slide 24

Forecast sensitivity to observation: Equations and Solution

a b bx = x + K(y - Hx )Analysis solution

Analysis sensitivity to observation and background

a

a

J J

x

y y x

J is a measure of the forecast error: energy norm Forecast error sensitivity to the analysis

a

Jx

( )bJ JJ

y y Hxy y

1

a

J J

R HA

y x

The tool providesinformation on the observation type, subtype, variable and level responsible for the forecast error variation

Rabier F, et al. 1996.

Solve the linear system:Compute the δJ

1Ta

x K R HAy

Documents

Slide 1 International Typhoon Workshop Tokyo 2009 Slide 1 Impact of increased satellite data density in sensitive areas Carla Cardinali, Peter Bauer, Roberto