23
Y. Fujii 1 , K. Ogawa 1 , K. Ando 2 , and M. Kamachi 1 1: JMA/MRI, 2: JAMSTEC Evaluation of Argo float impacts on the ocean data assimilation system in JMA/MRI 4 th Argo Science Workshop, Sep. 28th, 2012, Venice, Italy 1. Introduction of the current operational Ocean DA system in JMA 2. Impacts of Argo on the assimilation results 3. Impacts of Argo on seasonal and ENSO forecasts Outline

Evaluation of Argo float impacts on the ocean data

  • Upload
    others

  • View
    5

  • Download
    0

Embed Size (px)

Citation preview

Y. Fujii1, K. Ogawa1, K. Ando2, and M. Kamachi1

1: JMA/MRI, 2: JAMSTEC

Evaluation of Argo float impacts

on the ocean data assimilation

system in JMA/MRI

4th Argo Science Workshop, Sep. 28th, 2012, Venice, Italy

1. Introduction of the current operational Ocean DA system in JMA

2. Impacts of Argo on the assimilation results

3. Impacts of Argo on seasonal and ENSO forecasts

Outline

1. Introduction of the current

operational ocean data

assimilation systems in JMA

MOVE System (MOVE/MRI.COM)

Multivariate Ocean Variational Estimation (MOVE) System

→ Ocean Data Assimilation System JMA/MRI. Model: MRI.COM (Tsujino et al. 2011, JO)

Analysis Scheme:

3DVAR with Coupled

T-S EOF

(Fujii and Kamachi

2003, JGR)

Model Insertion:

Incremental Analysis

Updates (IAU)

Observation:

・TS profiles

・Gridded SST

・Altimetry

lll

l

lf w yUSxyx )(

Constraint for SSH observation

)())(())((2

1

)()(2

1

2

1

010

010

,

1

,

yhyxhRhyxh

xyHxRxyHxyBy

h

T

T

m l

lml

T

lmJ

Background Constraint Constraint for T, S observation

Additional Constraints

Seek the amplitudes of EOF modes

y minimizing the cost function J.

→Analysis increment of T and S

will be correlated.

See. Fujii and Kamachi, JGR, 2003

Analysis Increment is represented by the linear combination of the

EOF modes.

Obs.

T

S

Analysis

T

S

3DVAR Analysis Scheme in MOVE

Amplitudes of

EOFs

Effect of the use of Coupled T-S EOF modes

Validation of salinity variation with the independent TRITON buoys

Assimilating T alone

w/o coupled EOF

Assimilating T alone

with coupled EOF

Assimilating T and S

with coupled EOF

Variation of S profiles

observed by TRITON

5ºN

- 1

56

ºE

Eq. -

15

6ºE

5

ºS -

15

6ºE

JMA Seasonal Forecasting System

Atmos. Data Assim. System. (JRA-25, JCDAS)

JMA/MRI-CGCM Ensemble Forecasts

Ocean Data Assim. System (MOVE-G)

Ocean Obs. Data

Atmos. Obs. Data

Initia

l Va

lue

s

NINO34 SST Index Forecasts by the JMA Seasonal Forecasting System

ENSO Forecast : Since Mar. 2008

Seasonal Forecast : Since Mar. 2010

2. Impacts of Argo on the

assimilation result

Experimental Configuration

0 1 2 3 4 5 6 7 8 9

Argo Floats (The last digit of WMO No.) Other Data (CTD,

Buoy, Satellite

SSH, etc.)

Reference Data

(Independent from runs

other than Argo100%)

Argo100%

Argo80%

Argo60%

Argo40%

Argo20%

Argo0%

Assimilated Data

Name of

Assim. Runs

We implemented 6 assimilation runs (2000-2010).

We evaluate impacts through the RMSEs from the reference data.

The reference data is independent from the runs other than the 100% run.

→ We can fairly evaluate the accuracy of those runs using the reference data.

NRMSE = RMS(Assim.-obs.)/RMS(clim.-obs.)×100%

Reduction of Normalized RMSE for 0-300m Sal.

80%-0% (the total impact) 20%-0% (impact of the 1st 20%)

40%-20% (impact of the 2nd 20%) 60%-40% (impact of the 3rd 20%)

80%-60% (impact of the 4th 20%) 100%-80% (not fair evaluation)

Reduction of NRMSE = 100%-NRMSE

Tem

p.(

0-3

00m

) S

al.(0

-300m

)

W-EqPac SubTroNPac E-EqPac MidLatNPac

Reduction of Normalized RMSE from Clim.

MidLatPac

SubTroNPac

W-EqPac E-EqPac

The impact on Sal. is generally larger than that on Temp.

Impact of adding Argo data decreases with the increase

of the number of Argo that already assimilated.

Comparison of RMSE profiles of TS for 0-500m

Clim.

0%

20%

40%

60%.

80%

100%

W-EqPac SubTroNPac

Tem

p.

Sal.

Increase of the number of Argo floats improves the accuracies of TS at most

depths (red←orange←yellow←green←right blue←dark blue).

The impact of Argo floats are notable on salinity, especially in the near-surface

layer.

The impact is also notable on temperature around the thermocline.

Summary on the impacts on assimilation results

Cautions

The impacts include those for model biases and variability.

Saturation of the impact is not caused only by the duplication of the

observed info. Inappropriate error statistics are also influential.

Results depend on the system and the resolution. In particular, the

result will be rather different if we use an eddy-resolving system.

multi-system evaluation is essential.

The impacts on salinity is usually more substantial than those on Temp.

(Partly because T is also improved by buoy and satellite SSH data.)

However, Argo data still has impacts on temperature around the

thermocline.

Impact of adding Argo data decreases with the increase of the number of

Argo that already assimilated. (The effect may get to be saturated(?))

Results

3. Impacts of Argo on seasonal

and ENSO forecasts

Experimental Design of SF-OSE in JMA/MRI

Assimilation(MOVE/MRI.COM-G) → Jan. 2000-Dec. 2009

・ALL → Use all available data (equivalent to the Argo100% run)

・XAF → withholding the ARGO float profiles (equivalent to the Argo 0% run)

・XTT → withholding the TAO/TRITON buoy profiles

Forecast(JMA/MRI-CGCM)→ 2004-2008 (20 cases)

・Forecasts from the assimilation results of ALL, XAF , XTT

・Initial date: Jan. 31st, Apr. 26th, Jul. 30th, Oct. 28th

・Forecast length:13 months

・Number of the ensemble members: 11 (Generated by perturbed SST OBS)

・Flux Correction: Same as in the JMA operation.

・Calibration: performed for ALL, XAF, XTT separately.

Differences in NINO34 Forecasts (Examples)

COBE-SST

ALL

XTT

XAF

Ensemble Mean

Single Member Forecasts

It is difficult to get the statistically significant impacts of ocean observations

Impacts on SST Indices

Improvement of ACCs by assimilating Buoys or Floats

LT: Lead Time

Improvement of ACC for 1-7M LT atmospheric forecasts

VP200:

Velocity Potential at

200hPa Surface

(divergence at the top

of the troposphere)

SLP:

Sea Level Pressure

OLR:

Outgoing Longwave

Radiation

(proxy of precipitation)

Initial: 2006/01/31

ALL .vs. XAF

(Impacts of Floats)

ALL

What Causes the difference? (1) Initial Difference (Eq. Pac. Temp.)

XAF XTT

X-T Sections of ensemble mean difference (ALL-XAF)

Initial: 2007/10/28

ALL .vs. XAF

(Impacts of Floats) ALL

What Causes the difference? (2) Initial Difference (Eq. Pac. Temp.)

XAF

XTT

X-T Sections of ensemble mean difference (ALL-XAF)

Initial: 2007/10/28

ALL .vs. XAF

(Impacts of Floats) ALL

What Causes the difference? (2) Initial Difference (Eq. Pac. Temp.)

XAF

XTT

X-T Sections of ensemble mean difference (ALL-XAF)

Summary on the impacts on seasonal Forecasts

Difficulty of evaluating the impacts on the Seasonal forecasts

Targets are very stochastic.

• The evaluation by ensemble forecasts are required

• Impacts of observations are weaken by stochastic processes.

System (model and assimilation scheme) dependence

• Current coupled models still have large model errors and biases.

• Info. is not fully subtracted from obs. by the current DA system.

Argo data improves the SST forecasts in the eastern and central

equatorial Pacific, especially for relatively-long (7-13 month) lead-time, in

the JMA system.

Forecasts of SLP, precipitation, and the global circulation of the

atmosphere are also improved probably because of the better SST

forecasts.

The better SST forecasts probably stems from the improved subsurface

temperature fields in the initial condition.

Results Fujii et al. 2011, in Climate Variability (InTech book)

Activity in GODAE Ocean View OSEval task team

Impact of ocean observation data on the seasonal forecasting is an

important information for sustaining the observation platform, because

the seasonal forecasting is one of the most influential products from the

ocean observation data.

OSEs for the Seasonal Forecasting (SF-OSE) should be performed

using multi-systems, because of its dependence on systems.

SF-OSE is included in the activities of the GODAE Ocean View OSEval

task teams (as a delayed mode OSE).

JMA/MRI will continue SF-OSE according to the recommendation of the

team. Other groups can join SF-OSE.

It would be great if we can exchange the information of SF-OSE, and

can coordinate the plan of a multi-system evaluation.

Thank you!