ACRE Downscaling: 20C Reanalysis Application for

Paleoclimate tracer simulations1. Dyn. Downscaling with 20C-Rean

2. 20C Isotope Reanalysis

Kei Yoshimura

20th century Reanalysis (Compo et al., 2011)

• Using only surface pressure data historically recorded since 1870’s

• Ensemble Kalman Filter for data assimilation (56 member)

• T62L28 GFS with NOAH LSM

• Reanalysis skill is comparable to current Day-3 forecast skill (Whitaker et al., 2009)

• Ensemble Mean (EM) fields are publically available.

Whitaker et al. (2009)

1 December 1918

SLP 500 hPa GPH

Question for Dynamical Downscaling

Are the ensemble mean fields appropriate as lateral

boundary conditions for dynamical downscaling?

ORIG EM

19C

20C

Diff

NO!Global Dynamical Downscaling

Transient component of moisture divergence is smoothed out in Ensemble mean

Tota

l div

erge

nce

Mea

nTr

ansi

ent

19C 20C

No di/convergence

uq

uq

qu′

′+

⋅=

Straight forward remedy: Downscale a single member many times

and making an average of them.

Is there any other way to reduce computation?

Modification of single member by ensemble mean increment

• where F is full field of physical variable, n is an ensemble member, bar indicates ensemble mean, and <> indicates running mean (e.g. one-month).

• The downscaling will be performed using Fnnew as a lateral boundary forcing.

Time

Single member forecast Fn

Other members

Monthly running mean of a single member <Fn>

Monthly running mean of all members <Fbar>

incrementCorrected member

nnnew

n FFFF −+=

Specification of experiments• Atmospheric Forcing: 20thC Reanalysis (Compo et al., 2011)

– Also regarded as “truth”.• Experiments: Different by the atmospheric forcings.

– EM: Ensemble mean is used.– S1: Arbitrary chosen single member (run01) is used. – S3: Mean of the runs in which arbitrary chosen three single

members (run01, run11, & run21) are used.– S6: Similar to S3, but 6 single members (S3 + run31, run41, &

run51) are used. – MS: Modified single member is used.

• Periods: – 1871-2008 for EM and MS.– 1871-1873 and 1981-1983 for S1-S6.

• Model: IsoGSM with global spectral nudging (Yoshimura et al., 2008)

Seasonal mean precipitation with MS fieldORIG MS

19C

20C

Diff

DJF mean of moisture divergence in MSTo

tal d

iver

genc

eM

ean

Tran

sien

t19C 20C

Appropriate di/convergence

RMSD in 500Z against “truth”19C 20C

Glo

beN

HSH

RMSD in Precipitation against “truth”19C 20C

Glo

beN

HSH

RMSD in Wind against “truth”19C 20C

Glo

beN

HSH

Figure 10: Global mean precipitation by each experiments from 1871 to 2008. Original 20th

century Reanalysis (green), EM (black), and MS (blue) are shown for all period. Those runs with the direct use of single members (which consists S6) are shown only for 1871-73 and 1981-83.

Long term global precipitation

EM

MS

Summary of Part 1

• Use of ensemble mean field as atmospheric forcings for downscaling study makes big shortcomings, particularly too small precipitation, when the spread of ensemble members is large.

• Downscaling of each single ensemble member is straightforward, but requires lots of resources and time.

• To avoid these problems, we propose a new method which modifies a single member field to have the same monthly skills as ensemble mean field (MS method) .

• Use of the MS method clearly improves skill than direct usage of a single member. About the same skill as when 3 members are directly used.

2. 20C Isotope Reanalysis

Ｖ Surface vapor d18O anomaly at 116W 34.5N

-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

1979 1984 1989 1994 1999 2004

anom

aly

d18O

s1f23yr running mean (f2)3yr running mean (s1)

Bristlecone pine tree at SE CA

nudged

free

Yoshimura et al., 2008

Oceanic sediment δ18O (millions yBP)Icesheet cores δ18O・δD (~800 kyBP)Icecap cores δ18O・δD (~20 kyBP)Speleothem δ18O (~2000 yBP)Treering δ18O (~1000 yBP)Coral δ18O (~400 yBP)

Impact Analysis(a.k.a., forward proxy modeling)

Model WorldReal World

Ice Coreδ18O

Temperature

Precipitation

Circulation

Source

Snow/Glacier

Etc.Etc.

The impact from each component is not quantifiable.

Ice Coreδ18O

Temperature

Precipitation

Circulation

Source

Snow/Glacier

Etc.Etc.

The impact from each component is explicitly quantifiable.

Uncertainty of the impacts is quantifiable by reproducibility.

Why 20C Isotope Reanalysis?

Isotopes in GCM/RCM

• Incorporate water isotopes as passive tracers in GCMs/RCMs. Whenever water phase change takes place, isotopic water (HDO, H2

18O) behave differently to ordinary water (H2O).

Risi et al. 2008Courtesy of JMA

H18OH

HD16O

Typical convective precipitation process

Number of GNIP sites where correlation is significant for 1980-1999

ECHAM GISS-E IsoGSM-R2 IsoGSM-20C

CorrelationNH (210) 147 171 (81%) 174 (83%) 171 (81%)

Tropics (142) 68 82 (58%) 96 (68%) 105 (73%)SH (37) 22 (60%) 18 25 (68%) 25 (67%)

Anomaly Correlation

NH (146) 13 (9%) 12 114 (78%) 93 (63%)Tropics (67) 9 12 (18%) 32 (48%) 34 (50%)

SH (29) 1 3 (10%) 12 (41%) 12 (41%)

GoodBad GoodBad

Comparison with GNIP data

Treering δ18O in Cambodia

Larger amplitude?

Wrong phase?

@Kirirom national park

Courtesy of M. Zhu and L. Stott

Courtesy of L. Stott

Treering δ18O in West US

Seawater δ18O from Coral and Model near Philippines

R= 0.54

R= 0.55

δ18Osw (Red: Model Blue: Coral)

SSS (Red: Model Blue: SODA)

Kojima et al., submitted

Bunaken (2N 125E)

Cora

lδ1

8 Osw

(‰)

Mod

eled

δ18 O

sw(‰

)

Cora

lδ1

8 Osw

(‰)

Mod

eled

δ18 O

sw(‰

)Philippine (13N 124E)

Seawater δ18O from Coral and Model

Courtesy of K. Kojima

-0.4

-0.2

0

0.2

0.4

0.6

0.8

0 2000 4000 6000

Tropical pacific

South Pacific

SE-Asia

Indian sea

Atlantic

相関係数

年平均降水量のばらつき(mm/year)

-0.6-0.4-0.2

00.20.40.60.8

1

0 500 1000

Reproducibility of Interannual δ18Osw and Precip Amount

Large Precip Variability→ Local precipitation is recorded

Small Precip Variability→ Other factors (current, river flow, etc) may play big roles.

Std Dev of Annual Precip (mm/year)

Corr

elat

ion

Kojima et al., in prep.

Icecore δ18O at Eclipse Icefieldlon=-139.47 ; lat=60.51

Yalcin et al., 2003; 2006

Icecap δ18O at Mt Huascaranm, Peru lon=-77.6 ; lat=-9.1

Thompson et al., 1995

Summary of Part 2

• First 20th century Quasi Reanalysis for Isotope is now available.

• First comparisons with paleoclimate proxy data are now underway.

• This effort helps to develop the “forward proxy modeling” approach to more comprehensively understand the proxy data.

Thank you!

kei@aori.u-tokyo.ac.jp

TypeDataset Name

Source Variable Since Time Resolution Location (resolution) Ref.

Direct obs.CRUTEM3 T T 1850 monthly global land (5x5) Brohan et al., 2006

GPCP V4 P P 1901 monthly global land (0.5x0.5) Beck et al., 2005

Proxy grid data

tree ring T 1600 annual W. N. America Briffa et al., 1992

tree ring T and P 1602 annual N. America Fritts 1991

documentary and others

SLP 1500 seasonalN. Atlantic and

EuropeLuterbacher et al.,

2004

Isotope data for regional simulation

GNIP precipitation d18O and dD 1960 monthly global land (600sites) IAEA/WMO, 2006

stream water d18O and dD 1980 monthly USAKendall and Coplen,

2001

tree ring d18O, dD, d13C 1800 1y-5y N. AmericaVarious; reviewed by

McCarroll and Loader, 2004

Isotope data for global simulation

shallow ice core d18O and dD 1860 1y-5yAntarctica,

Greenland, high mountains

Various; sorted by Schneider and Noone,

2007

snow d18O and dD 1960 occasional AntarcticaVarious; collected by Masson-Delmotte et

al., 2008

tree ring d18O 1500 1y-5y VariousVarious; e.g., Tsuji et

al., 2006

ground water d18O 1960 occasional Various Various

speleothem d18O, d13C 1200 annual southern Oman Fleitmann et al., 2004

speleothemd18O, Ca/Mg,

d13C, etc1800 monthly China Johnson et al., 2006

Pseudo Global Warming (Kawase et al., 2009)

との違い

Kawase et al., 2009

低周波：モデル高周波：観測（再解析）

本手法

低周波：観測（アンサンブル平均）高周波：モデル

Time

Single member forecast Fn

Other members

Monthly running mean of a single member <Fn>

Monthly running mean of all members <Fbar>

incrementCorrected member

Method (c.f. Roden et al., 2000)

• Data– NPP weighted annual temperature for esat

– NPP weighted annual RH for eair

– NPP weighted annual d18O in vapor for Rair

– NPP weighted annual surface pressure for eleaf

– NPP&amount weighted annual rain water d18O for Rsw

• Rleaf= α*[1.032*Rleaf*(esat-eleaf)/esat +1.021*Rleaf*(eleaf-eair)/esat + Rair*eair/esat]

Cora

lδ1

8 Osw

(‰)

Mod

eled

δ18 O

sw(‰

)

Fiji (18S 179E)

Cora

lδ1

8 Osw

(‰)

Mod

eled

δ18 O

sw(‰

)Nauru (0.5S 166E)

サンゴ同位体比から得られる海水同位体比の再現性

Courtesy of K. Kojima

Spectral Nudging + Isotope GSM– Poor man’s data assimilation for isotopes –

IsoGSMSIO/UT Isotopic Global Spectral Model

H18OH

HD16O

( )

≥+×−<−

=

−−=

)m/s(71)m/s(71

1

_

_

VCVBVAh

hRRR

mvk

aseaemvkE

L

Lα

αα

( )( )CTTS

SDDS

ejmk

ejmkeff

°−<−=+−′

=

=

20003.0111/_

_

L

αα

ααα

( )

( ) ( )hDDh

RRdmdRm

ne

e

vrr

−′=−

=−=

−=

1/,1

,1 αμμ

αγμ

μβ

γβ

( )( )[ ] ( )( )[ ] qmRmqR

RRmmRRR

rv

rvvrr

/1/

00

00000

−′+′=−++−= εγγε β

Surface evaporation (MJ79)

Ice crystal formation (JM84)

Rain drop evaporation (S75)

Fourier series

Nudge scale Nudging

Forecast

λ

f

Use large scale (>1000km) winds from R2 to constrain dynamical field, so that the isotopic field is also constrained and reproduced in daily to inter-annual time scales.

Yoshimura and Kanamitsu, 2008; Yoshimura et al., 2008

http://meteora.ucsd.edu/~kyoshimura/IsoGSM1

Previous works