Development of CFES–LETKF Ensemble Data Assimilation...

Preview:

Citation preview

Development of CFES–LETKF Ensemble Data Assimilation System

2nd Data Assimilation WorkshopJAMSTEC Yokohama Institute for Earth Sciences, January 13, 2012

Nobu Komori,1 T. Enomoto,1,2 T. Miyoshi,3 B. Taguchi 1

and Team OREDA1 Earth Simulator Center (ESC), JAMSTEC, Yokohama, Japan2 Disaster Prevention Research Institute, Kyoto University, Uji, Kyoto, Japan

3 Department of Atmospheric and Oceanic Science, University of Maryland, College Park, Maryland, U.S.A.

E-mail: komori@jamstec.go.jp

Outline of Talk

• AFES-LETKF ensemble DA system

• Atmospheric reanalysis: ALERA & ALERA2

• CFES-LETKF ensemble DA system

• Preliminary results of CLERA-A

• Summary

LETKF: local transform ensemble Kalman filter

Kalman filter for a scalar

: analysis error (co)variance

: forecast error (co)variance

: observation error (co)variance

Kalman gainq: system noiseM: model

xfn = Mxa

n−1

pfn = M2pa

n−1 + qn−1

xan = xf

n + wn

�xo

n − xfn

pan = (1− wn) pf

n

wn =pf

n

pfn + rn

pa ≡ E��

xa − xt�2

pf ≡ E��

xf − xt�2

r ≡ E��

xo − xt�2

Kalman filter for a scalar

wn =pf

n

pfn + rn

=(rn)−1

�pf

n

�−1+ (rn)−1

xan =

�pf

n

�−1

�pf

n

�−1+ (rn)−1

xfn +

(rn)−1

�pf

n

�−1+ (rn)−1

xon

(pan)−1 =

�pf

n

�−1+ (rn)−1

Kalman filter (KF)[Kalman 1960]

ensemble KF (EnKF)[Evensen 1994]

ensemble transform KF (ETKF)[Bishop et al. 2001]

local ensemble KF (LEKF)[Ott et al. 2004]

local ensemble transform KF (LETKF)[Hunt et al. 2007]

Ensemble update

δXai = δX

fiT

Approximation by ensemble

Ensemble Kalman filter

Pf ≈ 1m − 1

δXδX

ETKF (Bishop et al. 2001)

ALERA

ALERA

AFES-LETKF experimental reanalysis

• first application of LETKF to full AGCM

• provides analysis ensemble spread as error estimates

• a product of collaboration among JMA, JAMSTEC and CIS

ALERA: specifications

• Observations used in NWP at JMA

• AFES T159L48M40

• from 18UTC 1 May 2005 to 12UTC 11 Jan 2007

• u, v, T, T–Td, z, slp

• available from the Earth Simulator Centerhttp://www.jamstec.go.jp/esc/afes/alera/

Hunt et al. 2007; Miyoshi and Yamane 2007; Miyoshi et al. 2007a

ALEDAS: AFES-LETKF Ensemble Data Assimilation System

observation

forecast

t=t0 t=t1

analysisanalysis

LETKF

AFES

LETKF: Hunt et al. 2007; Miyoshi and Yamane 2007; Miyoshi et al. 2007

AFES: Numaguti et al. 1997; Ohfuchi et al. 2004; Enomoto et al. 2008

Miyoshi et al. 2007a SOLA

850 (u,v) and standardized U850 spread

mean

spread

Enomoto et al. 2010 GRL

spread leads spread lagsdays

-0.2

0

0.2

0.4

0.6

-10 -8 -6 -4 -2 0 2 4 6 8 10

Lag correlation betweenmean and spread

0.52

Enomoto et al. 2010 GRL

Effects of buoy observation

Q. Moteki Pers. Comm.

ALL the TRITON buoys (blue ∆) observe surface pressure

ONLY the TAO buoys on the equator (red ∆) observe surface pressure

Influence of sondes in Matsuno-Gill pattern

Moteki et al. 2011 QJRMS

ALERA (w/o MISMO sondes)

with MISMO sondes

ALERA – with MISMO

MISMO Oct–Dec 2006in the Indian Ocean

Influence on typhoon genesis

Impact of Arctic buoys

Inoue et al. 2009 GRL

accuracy in the region w/ Arctic buoys (●) is as high as

in the mid-latitudes

error becomes larger over the entire Arctic

ALERA2

ALERA2 features

• Larger ensemble size (from 40 to 63)

• Improved model physics (cloud and land surface etc)

• Covariance localization by distance

• High-res SST and ice (NOAA daily 1/4° OI SST)

• Publicly obtainable observations (prepbufr)

• Forecast values (precip, radiative and surface fluxes)

Low cloud fraction

Kuwano-Yoshida et al. 2010 QJRMS

Cloud water

Kuwano-Yoshida et al. 2010 QJRMS

FT=120 Z500 RMSD

48

49

50

51

52

53

54

55

56

1.22 2.0 2.1 2.2 2.3 2.4 2.7 2.7m 3.6

48.9

49.5

51.351.251.6

52.1

53.1

53.8

55.4

ALERA

T159L48 T119L48

+MATSIRO

+PDF cloud

against JMA analysis, August 2004m

Error covariance localization by distance

Noise due to local patches

ALERA ALERA2

Miyoshi et al. 2007b SOLA

Surface

Sondes Aircraft

Ships & buoys

QuikScatAMV

GPCP ALERA2

JRA25 ALERA2 spread

2003 2004 2005 2006 2007 2008 2009 2010 2011

stream 2010

ALERA2

PALAU 2008

Mirai Arctic Ocean Cruise

Mirai Arctic Ocean Cruise

VPREX 2010

T-PARC

Winter T-PARC

PALAU 2005

MISMO

CINDY 2011

ALERA

stream 2008

ALERA2 streams

CFES-LETKF

Motivations

• Remove underestimation of ensemble spreadnear the sea surface

• Improve SST–precipitation correlation

➡ Replace AFES with CFES

Saha et al. 2010 BAMS

ObsCFSR

R1R2

SST lead Prec Prec lead SSTLag days

Lag Correlation of Prec. and SST over Western Pacific (winter)

CFES mini

• AFES T119L48, mstrn-X, Emanuel, MATSIRO

• OFES 0.5°x0.5° 54 lev., Sea ice, Noh-Kim mixed layer

• 177 years: interannual to decadal variability

• Richter et al. 2010 GRL,Taguchi et al. 2012 JC, ...

コンター: 平均場, 陰影:標準偏差

CFES 中解像度阪

NCEP/NCAR再解析

冬季ストームトラック活動度

B. Taguchi Pers. Comm.

storm-track activity in wintercontour: mean, color: std. deviation

CFES mini

NCEP/NCAR Reanalysis

Precipitation (DJF)

CMAP

CFES

AFES (T239)

CFES mini

Precipitation (JJA)

CMAP

CFES

AFES (T239)

CFES mini

CLERA-A

• Initial conditions from 41-year OFES spin-up with CORE.v2 Corrected Interannual Forcing

• 39 members + control

• 4 CFES per node

• Atmospheric DA only to study influence of the surface boundary uncertainties

• Starting from 1 August 2008

• Preliminary result on 10 September 2008

MPI implementation

CFES

EnCFES

AFES

...

MPI_COMM_WORLD= MPIA_COMM_WORLD

... ...

MPIA_COMM_WORLD MPIO_COMM_WORLD

MPI_COMM_WORLD = MPI_COMM_COUPLE

EnAFES

...

MPIA_... ...

MPI_COMM_WORLD

MPIA_... MPIA_...

......

......

......MPIO_...

MPI_COMM_COUPLE

MPI_COMM_WORLD

MPI_COMM_COUPLE MPI_COMM_COUPLE

MPIA_... MPIA_... MPIA_...MPIO_... MPIO_...

forecast

6h window

analysis

Forecast–Analysis CycleAFES-LETKF Ensemble DA System

LETKFAFES restartIC

t-3t-2t-1tt+1t+2t+3

merge

split

anal

gues

BC obs

forecast

6h window

analysis

Forecast–Analysis CycleCFES-LETKF Ensemble DA System

LETKFAFES restartIC

t-3t-2t-1tt+1t+2t+3

merge

split

anal

gues

IC OFES restart obs

CLERA-A

ALERA2 Surface temperature

spreadmean

mean spread 2008080500

CLERA-A

ALERA2 Surface temperature

spreadmean

mean spread 2008091000

CLERA-A

ALERA2 Latent heat flux

spreadmean

mean spread 2008091000

CLERA-A − ALERA2

Difference in zonal mean spread

Temperature Specific humidity

Eq NPSP Eq NPSP

1000

850

700

2008091000

Summary

• The CFES-LETKF ensemble DA system has been constructed.

• Ocean ensemble creates perturbed surface BC.

• Needs to handle oceanic biases and drift for a longer run.

Thank You!

Recommended