Recent advances in the Mercator sea-ice reanalysis system · 2020. 1. 27. · Univariate Sea Ice...

mercator-ocean.eu/marine.copernicus.eu

Recent advances in the Mercator sea-ice

reanalysis system: Improving sea-ice thickness estimates by assimilating

Cryosat-2 and SMOS sea ice thickness data

C.E Testut, G. Ruggiero, C. Bricaud, J.M. Lellouche,

O. Legalloudec , J. Chanut, L. Parent and G. Garric

Mercator-Ocean, Ramonville Saint Agne, France

Univariate Sea Ice Analysis in the operational applications

The Mercator ocean and sea-ice reanalysis system PSY4V3

Development of a multivariate Sea Ice Analysis

Reanalysis assimilating SIC only

Reanalysis assimilating SIC and SIVOLU

=> Preliminary results from hindcast experiment assimilating CS2-SMOS products

Summary and Plans

Outline

OceanPredict DA-TT meeting Cerfacs, Toulouse, France, January, 2020

Model - Nemo 3.1, LIM2-EVP(mono-category) - Global 1/12°, 50 levels, IFS - Oct. 2006 to near real-time

Assimilation

- Analysis based on a 2D local multivariate SEEK/LETKF filter

- 7-days cycle, 3D model update, IAU

- Weakly-coupled DA system using 2 separate analyses :

- Ocean Analysis (SLA, InSitu Data from CORA3.2, SST) , IAU on (h,T,S,U,V)

- Unidata/univariate Sea Ice Analysis

- Sea Ice Concentration (OSI-SAF)

- SIC Error: 1% open ocean, linear from 25% to 5% for SIC values between 0.01 and 1

- Forecast error covariances are built from a prior ensemble of Sea Ice Concentration anomalies => Fixed basis background error

The Mercator ocean and sea-ice system PSY4V3 Main Characteristics

Sea Ice Concentration (OSI-SAF)


Mercator Data Assimilation System (SAM2): Pf: Fixed Basis Background error covariances

We use these anomalies to compute Pf in the analysis

Model trajectory

2007 2008

2009 2010

2011 2012

We generate a pseudo-ensemble from a forced simulation

Model trajectory

anomaly

Running mean of the model trajectory

Temporal window

Analysis date 2 Analysis date 1

Representation by a prior ensemble of anomalies


Restart Strategy: IAU on SIC + …

- … nothing using an implicit concept of thickness conservation

DSICi, Dhi = 0, hi = cte DVi = hi . DSICi

hi

SICi

+

DSICi

=

The Mercator ocean and sea-ice system PSY4V3 Model restarting using the sea ice model update

Innovation OSISAF – PSY4V3/LIM2noEVP

Model update Residual Innovation – Model update

Analysis of Sea Ice Concentration (y2011m10d12, 7 days cycle)


The Mercator ocean and sea-ice system PSY4V3 2006-2018 hindcast experiment assimilating OSI-SAF SIC

CERSAT SIC Misfits

CERSAT SIC Residu

OSI-SAF SIC Innovation

OSI-SAF SIC Residu

Residual vs innovation OSI-SAF (Feb 2012)

Oct.2006 Jan..2015

CERSAT OSI-SAF


OSI-SAF SIC Innovation (PSY4V3)

OSI-SAF SIC Residu (PSY4V3)

OSI-SAF SIC Misfits (Run using Ocean Bias only)

Sep

t 2

01

1

Mar

ch 2

01

2



Oper. system Old system



AMSR Ice AMSR Water

Forecast Ice Hit ice(a) False Alarm (b)

Forecast Water Miss (c) Hit water (d)

Proportion Correct Total: PCT=(a+d)/n

Oper. system

Old system

Oper. system

Old system









Summary and Plans

Outline


Toward a Multivariate Sea Ice Analysis assimilating multi-observations and updating multi-variables

Objectives of a multivariate Sea Ice State Vector

-To assimilate various sea ice data sets (Sea Ice Drift, Thickness, SI Temperature, …)

-To update unobserved variables (Thickness or Volume, …) in order to increase the

control of the sea ice state using more consistent sea ice model update

We need

-To identify and activate an appropriate sea ice state vector according to the used

data sets and the used sea ice model (LIM2-EVP_cat1, LIM3_multi-cat)

-To use relevant background error covariance based on the physics of the days in

order to obtain good extrapolation from observed to unobserved space

=> Ensemble approach using stochastic perturbations

Development Framework:

- NEMO3.6/LIM3 multi-category


Multivariate Sea Ice Model update (y2006m12d28) (after 2 months of hindcast experiment)

Sensitivity hindcast experiment using CREG4 with NEMO3.6/LIM3_5categories

Multivariate state vector for multivariate sea ice analysis: [SIC,SIC_CAT1, SIC_CAT2,SIC_CAT3,SIC_CAT4,SIC_CAT5] with OSI-SAF SIC observations

OSI-SAF SIC Innovation SIC Model update



CAT1 CAT2

CAT3 CAT4 CAT5

Sensitivity hindcast experiment using CREG4 with NEMO3.6/LIM3_5categories

Multivariate state vector for multivariate sea ice analysis: [SIC,SIC_CAT1, SIC_CAT2,SIC_CAT3,SIC_CAT4,SIC_CAT5] with OSI-SAF SIC observations

SIC Model update SIC_CAT1 Model update SIC_CAT2 Model update

SIC_CAT3 Model update SIC_CAT4 Model update SIC_CAT5 Model update



SAM2V1 Multivariate Sea Ice Analysis CREG4 Analysis assimilating OSI-SAF SIC Observations

Assimilation


- 3-days cycle, 4D increment, IAU



- Sea Ice Analysis


- Forecast error covariances are built from a prior ensemble of Sea Ice Model anomalies

- Unidata/Multivariate Sea Ice analysis

- CREG4 reanalysis using multivariate state vector [SIC, SICONCAT(1:15), SIVOLUCAT(1:15)]

- null innovations on SICONCAT(15) and SIVOLUCAT(15)

Model - Nemo 3.6, LIM3/Multi-categories(1:15)

- CREG ¼, 75 levels

- Oct. 2006 to 2014

Sea Ice Concentration (OSI-SAF)


CREG4 Reanalysis starting at 20070102

Model update from analysis : [siconc,siconcat,sivolucat]

(Dsiconcat(1:15), Dsivolucat(1:15)) estimated from analysis based on OSI-SAF SI and null innovations on

SICONCAT(15) and SIVOLUCAT(15)

- Control run

- Forecast

- Analysis

- Control run

- Forecast

- Analysis

Sea Ice Concentration RMS misfits to OSISAF

Sea Ice Concentration misfits average to OSISAF



SIC(%) 20070915

OSI-SAF CREG4 reanalysis CREG4 Control run







- PIOMAS

- Control Run

- Reanalysis

Sea Ice Volume (Arctic)







Control run CREG4 reanalysis






Sea Ice Thickness difference (from ICESat) at y2007m03


SIC(%) 20070915

Control Run

Sea Ice Concentration RMS misfits to OSISAF

Various sea ice state vector are

tested

SIVOLU(m) 20070915

REA100

Control run

REA120

[SIC]

a|v cat from

current shape

REA121

[SIC]

a|v cat from

fixed distrib

REA130

[SICONCAT]

v cat from

fixed distrib

REA150

[SICONCAT,SIVOLUCAT]









Summary and Plans

Outline


SAM2V1 Multivariate Sea Ice Analysis ORCA025 Reanalysis assimilating OSI-SAF SIC and CS2-SMOS data

Assimilation


- 7-days cycle, 4D increment, IAU



- Sea Ice Analysis


- Thickness Error : 0.10 m SIVOLU=SITHI*SIC is in (m3/m2)

- Forecast error covariances are built from a prior ensemble of Sea Ice Model anomalies

- Unidata vs Multidata / Multivariate Sea Ice analysis

- ORCA025 reanalysis using multivariate state vector [SIC, SIVOLU, SICONCAT(1:11), SIVOLUCAT(1:11)]

- null innovations on SICONCAT(11) and SIVOLUCAT(11)

Model - Nemo 3.6, LIM3/Multi-categories(1:11)

- ORCA025 at ¼, 75 levels

- January to April 2011

Sea Ice Concentration (OSI-SAF) Sea Ice thickness (CS2-SMOS)


SAM2V1 Multivariate Sea Ice Analysis ORCA025 Reanalysis assimilating OSI-SAF SIC and CS2/SMOS data

CS2SMOS

Merged

products

from OI

every 7 days

(ESA/AWI)

CS2 data SMOS data

=>

Thickness data From CS2SMOS

merged products on

initial CS2-SMOS grid



15/03/2011

after 2 months

of simulation

SIC misfits

Control RUN SIC innovation from reanalysis

assimilating SIC only

Simulations

starting at

03/01/2011

from GLORYS4V2

OSI-SAF SIC



15/03/2011

after 2 months

of simulation

Thickness misfits

Control RUN Thickness innovation from

reanalysis assimilating SIC only

Simulations

starting at

03/01/2011

from GLORYS4V2

CS2/SMOS

thickness



- Control Run

- Reanalysis SIC only

- Reanalysis SIC and CS2SMOS

Sea Ice Thickness RMS misfits to CS2-SMOS data

- Control Run

- Reanalysis SIC only

- Reanalysis SIC and CS2SMOS

Sea Ice Concentration RMS misfits to OSI-SAF SIC data



15/03/2011

after 2 months

of simulation

Thickness misfits

Control RUN

Thickness innovation from

reanalysis assimilating SIC only

Simulations

starting at

03/01/2011

from GLORYS4V2

Thickness innovation from reanalysis

assimilating SIC and CS2/SMOS



15/03/2011

after 2 months

of simulation

OSI-SAF SIC misfits

Control RUN

SIC innovation from reanalysis

assimilating SIC only

Simulations

starting at

03/01/2011

from GLORYS4V2

SIC innovation from reanalysis

assimilating SIC and CS2/SMOS


Summary and Plans

:

The Univariate Sea Ice Analysis in the Mercator operational applications is

able to well represent the SIC but fails to constraint the volume

A multivariate Sea Ice Analysis is in development for the next Mercator

operational system launched in 2020. Different sea ice state vector are

explored depending of the assimilated observations. We have probably to

work and improve the tuning of the error.

A background error based on Ensemble modelling is probably necessary

to well extrapolate the information. An Ensemble Kalman filter is in

preparation and is tested since few months


Thank You

Recent advances in the Mercator sea-ice reanalysis system · 2020. 1. 27. · Univariate Sea Ice...

Documents

Transcript of Recent advances in the Mercator sea-ice reanalysis system · 2020. 1. 27. · Univariate Sea Ice...